VirtualBox

source: vbox/trunk/src/VBox/Additions/linux/sharedfolders/regops.c@ 85569

Last change on this file since 85569 was 83049, checked in by vboxsync, 5 years ago

linux/vobxsf: Corrected kernel-page test in vbsf_iter_unlock_pages, as we should only do set_page_dirty for user pages from what I can tell.
  • Property svn:eol-style set to native
  • Property svn:keywords set to Author Date Id Revision
File size: 148.1 KB
Line 
1/* $Id: regops.c 83049 2020-02-11 15:17:18Z vboxsync $ */
2/** @file
3 * vboxsf - VBox Linux Shared Folders VFS, regular file inode and file operations.
4 */
5
6/*
7 * Copyright (C) 2006-2020 Oracle Corporation
8 *
9 * Permission is hereby granted, free of charge, to any person
10 * obtaining a copy of this software and associated documentation
11 * files (the "Software"), to deal in the Software without
12 * restriction, including without limitation the rights to use,
13 * copy, modify, merge, publish, distribute, sublicense, and/or sell
14 * copies of the Software, and to permit persons to whom the
15 * Software is furnished to do so, subject to the following
16 * conditions:
17 *
18 * The above copyright notice and this permission notice shall be
19 * included in all copies or substantial portions of the Software.
20 *
21 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
22 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
23 * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
24 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
25 * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
26 * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
27 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
28 * OTHER DEALINGS IN THE SOFTWARE.
29 */
30
31
32/*********************************************************************************************************************************
33* Header Files *
34*********************************************************************************************************************************/
35#include "vfsmod.h"
36#include <linux/uio.h>
37#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 32)
38# include <linux/aio.h> /* struct kiocb before 4.1 */
39#endif
40#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 12)
41# include <linux/buffer_head.h>
42#endif
43#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 12) \
44 && LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 31)
45# include <linux/writeback.h>
46#endif
47#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 23) \
48 && LINUX_VERSION_CODE < KERNEL_VERSION(3, 16, 0)
49# include <linux/splice.h>
50#endif
51#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 17) \
52 && LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 23)
53# include <linux/pipe_fs_i.h>
54#endif
55#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 10)
56# include <linux/swap.h> /* for mark_page_accessed */
57#endif
58#include <iprt/err.h>
59
60#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 18)
61# define SEEK_END 2
62#endif
63
64#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 16, 0)
65# define iter_is_iovec(a_pIter) ( !((a_pIter)->type & ITER_KVEC) )
66#elif LINUX_VERSION_CODE < KERNEL_VERSION(3, 19, 0)
67# define iter_is_iovec(a_pIter) ( !((a_pIter)->type & (ITER_KVEC | ITER_BVEC)) )
68#endif
69
70#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 17, 0)
71# define vm_fault_t int
72#endif
73
74#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 5, 20)
75# define pgoff_t unsigned long
76#endif
77
78#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 5, 12)
79# define PageUptodate(a_pPage) Page_Uptodate(a_pPage)
80#endif
81
82
83/*********************************************************************************************************************************
84* Structures and Typedefs *
85*********************************************************************************************************************************/
86#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 16, 0)
87struct vbsf_iov_iter {
88 unsigned int type;
89 unsigned int v_write : 1;
90 size_t iov_offset;
91 size_t nr_segs;
92 struct iovec const *iov;
93# ifdef VBOX_STRICT
94 struct iovec const *iov_org;
95 size_t nr_segs_org;
96# endif
97};
98# ifdef VBOX_STRICT
99# define VBSF_IOV_ITER_INITIALIZER(a_cSegs, a_pIov, a_fWrite) \
100 { vbsf_iov_iter_detect_type(a_pIov, a_cSegs), a_fWrite, 0, a_cSegs, a_pIov, a_pIov, a_cSegs }
101# else
102# define VBSF_IOV_ITER_INITIALIZER(a_cSegs, a_pIov, a_fWrite) \
103 { vbsf_iov_iter_detect_type(a_pIov, a_cSegs), a_fWrite, 0, a_cSegs, a_pIov }
104# endif
105# define ITER_KVEC 1
106# define iov_iter vbsf_iov_iter
107#endif
108
109#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
110/** Used by vbsf_iter_lock_pages() to keep the first page of the next segment. */
111struct vbsf_iter_stash {
112 struct page *pPage;
113 size_t off;
114 size_t cb;
115# if LINUX_VERSION_CODE < KERNEL_VERSION(4, 11, 0)
116 size_t offFromEnd;
117 struct iov_iter Copy;
118# endif
119};
120#endif /* >= 3.16.0 */
121/** Initializer for struct vbsf_iter_stash. */
122#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 11, 0)
123# define VBSF_ITER_STASH_INITIALIZER { NULL, 0 }
124#else
125# define VBSF_ITER_STASH_INITIALIZER { NULL, 0, ~(size_t)0 }
126#endif
127
128
129/*********************************************************************************************************************************
130* Internal Functions *
131*********************************************************************************************************************************/
132DECLINLINE(void) vbsf_put_page(struct page *pPage);
133static void vbsf_unlock_user_pages(struct page **papPages, size_t cPages, bool fSetDirty, bool fLockPgHack);
134static void vbsf_reg_write_sync_page_cache(struct address_space *mapping, loff_t offFile, uint32_t cbRange,
135 uint8_t const *pbSrcBuf, struct page **papSrcPages,
136 uint32_t offSrcPage, size_t cSrcPages);
137
138
139/*********************************************************************************************************************************
140* Provide more recent uio.h functionality to older kernels. *
141*********************************************************************************************************************************/
142#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 16, 0) && LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
143
144/**
145 * Detects the vector type.
146 */
147static int vbsf_iov_iter_detect_type(struct iovec const *paIov, size_t cSegs)
148{
149 /* Check the first segment with a non-zero length. */
150 while (cSegs-- > 0) {
151 if (paIov->iov_len > 0) {
152 if (access_ok(VERIFY_READ, paIov->iov_base, paIov->iov_len))
153 return (uintptr_t)paIov->iov_base >= USER_DS.seg ? ITER_KVEC : 0;
154 AssertMsgFailed(("%p LB %#zx\n", paIov->iov_base, paIov->iov_len));
155 break;
156 }
157 paIov++;
158 }
159 return 0;
160}
161
162
163# undef iov_iter_count
164# define iov_iter_count(a_pIter) vbsf_iov_iter_count(a_pIter)
165static size_t vbsf_iov_iter_count(struct vbsf_iov_iter const *iter)
166{
167 size_t cbRet = 0;
168 size_t cLeft = iter->nr_segs;
169 struct iovec const *iov = iter->iov;
170 while (cLeft-- > 0) {
171 cbRet += iov->iov_len;
172 iov++;
173 }
174 return cbRet - iter->iov_offset;
175}
176
177
178# undef iov_iter_single_seg_count
179# define iov_iter_single_seg_count(a_pIter) vbsf_iov_iter_single_seg_count(a_pIter)
180static size_t vbsf_iov_iter_single_seg_count(struct vbsf_iov_iter const *iter)
181{
182 if (iter->nr_segs > 0)
183 return iter->iov->iov_len - iter->iov_offset;
184 return 0;
185}
186
187
188# undef iov_iter_advance
189# define iov_iter_advance(a_pIter, a_cbSkip) vbsf_iov_iter_advance(a_pIter, a_cbSkip)
190static void vbsf_iov_iter_advance(struct vbsf_iov_iter *iter, size_t cbSkip)
191{
192 SFLOG2(("vbsf_iov_iter_advance: cbSkip=%#zx\n", cbSkip));
193 if (iter->nr_segs > 0) {
194 size_t const cbLeftCur = iter->iov->iov_len - iter->iov_offset;
195 Assert(iter->iov_offset <= iter->iov->iov_len);
196 if (cbLeftCur > cbSkip) {
197 iter->iov_offset += cbSkip;
198 } else {
199 cbSkip -= cbLeftCur;
200 iter->iov_offset = 0;
201 iter->iov++;
202 iter->nr_segs--;
203 while (iter->nr_segs > 0) {
204 size_t const cbSeg = iter->iov->iov_len;
205 if (cbSeg > cbSkip) {
206 iter->iov_offset = cbSkip;
207 break;
208 }
209 cbSkip -= cbSeg;
210 iter->iov++;
211 iter->nr_segs--;
212 }
213 }
214 }
215}
216
217
218# undef iov_iter_get_pages
219# define iov_iter_get_pages(a_pIter, a_papPages, a_cbMax, a_cMaxPages, a_poffPg0) \
220 vbsf_iov_iter_get_pages(a_pIter, a_papPages, a_cbMax, a_cMaxPages, a_poffPg0)
221static ssize_t vbsf_iov_iter_get_pages(struct vbsf_iov_iter *iter, struct page **papPages,
222 size_t cbMax, unsigned cMaxPages, size_t *poffPg0)
223{
224 while (iter->nr_segs > 0) {
225 size_t const cbLeft = iter->iov->iov_len - iter->iov_offset;
226 Assert(iter->iov->iov_len >= iter->iov_offset);
227 if (cbLeft > 0) {
228 uintptr_t uPtrFrom = (uintptr_t)iter->iov->iov_base + iter->iov_offset;
229 size_t offPg0 = *poffPg0 = uPtrFrom & PAGE_OFFSET_MASK;
230 size_t cPagesLeft = RT_ALIGN_Z(offPg0 + cbLeft, PAGE_SIZE) >> PAGE_SHIFT;
231 size_t cPages = RT_MIN(cPagesLeft, cMaxPages);
232 struct task_struct *pTask = current;
233 size_t cPagesLocked;
234
235 down_read(&pTask->mm->mmap_sem);
236 cPagesLocked = get_user_pages(pTask, pTask->mm, uPtrFrom, cPages, iter->v_write, 1 /*force*/, papPages, NULL);
237 up_read(&pTask->mm->mmap_sem);
238 if (cPagesLocked == cPages) {
239 size_t cbRet = (cPages << PAGE_SHIFT) - offPg0;
240 if (cPages == cPagesLeft) {
241 size_t offLastPg = (uPtrFrom + cbLeft) & PAGE_OFFSET_MASK;
242 if (offLastPg)
243 cbRet -= PAGE_SIZE - offLastPg;
244 }
245 Assert(cbRet <= cbLeft);
246 return cbRet;
247 }
248 if (cPagesLocked > 0)
249 vbsf_unlock_user_pages(papPages, cPagesLocked, false /*fSetDirty*/, false /*fLockPgHack*/);
250 return -EFAULT;
251 }
252 iter->iov_offset = 0;
253 iter->iov++;
254 iter->nr_segs--;
255 }
256 AssertFailed();
257 return 0;
258}
259
260
261# undef iov_iter_truncate
262# define iov_iter_truncate(iter, cbNew) vbsf_iov_iter_truncate(iter, cbNew)
263static void vbsf_iov_iter_truncate(struct vbsf_iov_iter *iter, size_t cbNew)
264{
265 /* we have no counter or stuff, so it's a no-op. */
266 RT_NOREF(iter, cbNew);
267}
268
269
270# undef iov_iter_revert
271# define iov_iter_revert(a_pIter, a_cbRewind) vbsf_iov_iter_revert(a_pIter, a_cbRewind)
272void vbsf_iov_iter_revert(struct vbsf_iov_iter *iter, size_t cbRewind)
273{
274 SFLOG2(("vbsf_iov_iter_revert: cbRewind=%#zx\n", cbRewind));
275 if (iter->iov_offset > 0) {
276 if (cbRewind <= iter->iov_offset) {
277 iter->iov_offset -= cbRewind;
278 return;
279 }
280 cbRewind -= iter->iov_offset;
281 iter->iov_offset = 0;
282 }
283
284 while (cbRewind > 0) {
285 struct iovec const *pIov = --iter->iov;
286 size_t const cbSeg = pIov->iov_len;
287 iter->nr_segs++;
288
289 Assert((uintptr_t)pIov >= (uintptr_t)iter->iov_org);
290 Assert(iter->nr_segs <= iter->nr_segs_org);
291
292 if (cbRewind <= cbSeg) {
293 iter->iov_offset = cbSeg - cbRewind;
294 break;
295 }
296 cbRewind -= cbSeg;
297 }
298}
299
300#endif /* 2.6.19 <= linux < 3.16.0 */
301#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 16, 0) && LINUX_VERSION_CODE < KERNEL_VERSION(3, 16, 35)
302
303/** This is for implementing cMaxPage on 3.16 which doesn't have it. */
304static ssize_t vbsf_iov_iter_get_pages_3_16(struct iov_iter *iter, struct page **papPages,
305 size_t cbMax, unsigned cMaxPages, size_t *poffPg0)
306{
307 if (!(iter->type & ITER_BVEC)) {
308 size_t const offPg0 = iter->iov_offset & PAGE_OFFSET_MASK;
309 size_t const cbMaxPages = ((size_t)cMaxPages << PAGE_SHIFT) - offPg0;
310 if (cbMax > cbMaxPages)
311 cbMax = cbMaxPages;
312 }
313 /* else: BVEC works a page at a time and shouldn't have much of a problem here. */
314 return iov_iter_get_pages(iter, papPages, cbMax, poffPg0);
315}
316# undef iov_iter_get_pages
317# define iov_iter_get_pages(a_pIter, a_papPages, a_cbMax, a_cMaxPages, a_poffPg0) \
318 vbsf_iov_iter_get_pages_3_16(a_pIter, a_papPages, a_cbMax, a_cMaxPages, a_poffPg0)
319
320#endif /* 3.16.0-3.16.34 */
321#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19) && LINUX_VERSION_CODE < KERNEL_VERSION(3, 18, 0)
322
323static size_t copy_from_iter(uint8_t *pbDst, size_t cbToCopy, struct iov_iter *pSrcIter)
324{
325 size_t const cbTotal = cbToCopy;
326 Assert(iov_iter_count(pSrcIter) >= cbToCopy);
327# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 16, 0)
328 if (pSrcIter->type & ITER_BVEC) {
329 while (cbToCopy > 0) {
330 size_t const offPage = (uintptr_t)pbDst & PAGE_OFFSET_MASK;
331 size_t const cbThisCopy = RT_MIN(PAGE_SIZE - offPage, cbToCopy);
332 struct page *pPage = rtR0MemObjLinuxVirtToPage(pbDst);
333 size_t cbCopied = copy_page_from_iter(pPage, offPage, cbThisCopy, pSrcIter);
334 AssertStmt(cbCopied <= cbThisCopy, cbCopied = cbThisCopy);
335 pbDst += cbCopied;
336 cbToCopy -= cbCopied;
337 if (cbCopied != cbToCopy)
338 break;
339 }
340 } else
341# endif
342 {
343 while (cbToCopy > 0) {
344 size_t cbThisCopy = iov_iter_single_seg_count(pSrcIter);
345 if (cbThisCopy > 0) {
346 if (cbThisCopy > cbToCopy)
347 cbThisCopy = cbToCopy;
348 if (pSrcIter->type & ITER_KVEC)
349 memcpy(pbDst, (void *)pSrcIter->iov->iov_base + pSrcIter->iov_offset, cbThisCopy);
350 else if (copy_from_user(pbDst, pSrcIter->iov->iov_base + pSrcIter->iov_offset, cbThisCopy) != 0)
351 break;
352 pbDst += cbThisCopy;
353 cbToCopy -= cbThisCopy;
354 }
355 iov_iter_advance(pSrcIter, cbThisCopy);
356 }
357 }
358 return cbTotal - cbToCopy;
359}
360
361
362static size_t copy_to_iter(uint8_t const *pbSrc, size_t cbToCopy, struct iov_iter *pDstIter)
363{
364 size_t const cbTotal = cbToCopy;
365 Assert(iov_iter_count(pDstIter) >= cbToCopy);
366# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 16, 0)
367 if (pDstIter->type & ITER_BVEC) {
368 while (cbToCopy > 0) {
369 size_t const offPage = (uintptr_t)pbSrc & PAGE_OFFSET_MASK;
370 size_t const cbThisCopy = RT_MIN(PAGE_SIZE - offPage, cbToCopy);
371 struct page *pPage = rtR0MemObjLinuxVirtToPage((void *)pbSrc);
372 size_t cbCopied = copy_page_to_iter(pPage, offPage, cbThisCopy, pDstIter);
373 AssertStmt(cbCopied <= cbThisCopy, cbCopied = cbThisCopy);
374 pbSrc += cbCopied;
375 cbToCopy -= cbCopied;
376 if (cbCopied != cbToCopy)
377 break;
378 }
379 } else
380# endif
381 {
382 while (cbToCopy > 0) {
383 size_t cbThisCopy = iov_iter_single_seg_count(pDstIter);
384 if (cbThisCopy > 0) {
385 if (cbThisCopy > cbToCopy)
386 cbThisCopy = cbToCopy;
387 if (pDstIter->type & ITER_KVEC)
388 memcpy((void *)pDstIter->iov->iov_base + pDstIter->iov_offset, pbSrc, cbThisCopy);
389 else if (copy_to_user(pDstIter->iov->iov_base + pDstIter->iov_offset, pbSrc, cbThisCopy) != 0) {
390 break;
391 }
392 pbSrc += cbThisCopy;
393 cbToCopy -= cbThisCopy;
394 }
395 iov_iter_advance(pDstIter, cbThisCopy);
396 }
397 }
398 return cbTotal - cbToCopy;
399}
400
401#endif /* 3.16.0 <= linux < 3.18.0 */
402
403
404
405/*********************************************************************************************************************************
406* Handle management *
407*********************************************************************************************************************************/
408
409/**
410 * Called when an inode is released to unlink all handles that might impossibly
411 * still be associated with it.
412 *
413 * @param pInodeInfo The inode which handles to drop.
414 */
415void vbsf_handle_drop_chain(struct vbsf_inode_info *pInodeInfo)
416{
417 struct vbsf_handle *pCur, *pNext;
418 unsigned long fSavedFlags;
419 SFLOGFLOW(("vbsf_handle_drop_chain: %p\n", pInodeInfo));
420 spin_lock_irqsave(&g_SfHandleLock, fSavedFlags);
421
422 RTListForEachSafe(&pInodeInfo->HandleList, pCur, pNext, struct vbsf_handle, Entry) {
423 AssertMsg( (pCur->fFlags & (VBSF_HANDLE_F_MAGIC_MASK | VBSF_HANDLE_F_ON_LIST))
424 == (VBSF_HANDLE_F_MAGIC | VBSF_HANDLE_F_ON_LIST), ("%p %#x\n", pCur, pCur->fFlags));
425 pCur->fFlags |= VBSF_HANDLE_F_ON_LIST;
426 RTListNodeRemove(&pCur->Entry);
427 }
428
429 spin_unlock_irqrestore(&g_SfHandleLock, fSavedFlags);
430}
431
432
433/**
434 * Locates a handle that matches all the flags in @a fFlags.
435 *
436 * @returns Pointer to handle on success (retained), use vbsf_handle_release() to
437 * release it. NULL if no suitable handle was found.
438 * @param pInodeInfo The inode info to search.
439 * @param fFlagsSet The flags that must be set.
440 * @param fFlagsClear The flags that must be clear.
441 */
442struct vbsf_handle *vbsf_handle_find(struct vbsf_inode_info *pInodeInfo, uint32_t fFlagsSet, uint32_t fFlagsClear)
443{
444 struct vbsf_handle *pCur;
445 unsigned long fSavedFlags;
446 spin_lock_irqsave(&g_SfHandleLock, fSavedFlags);
447
448 RTListForEach(&pInodeInfo->HandleList, pCur, struct vbsf_handle, Entry) {
449 AssertMsg( (pCur->fFlags & (VBSF_HANDLE_F_MAGIC_MASK | VBSF_HANDLE_F_ON_LIST))
450 == (VBSF_HANDLE_F_MAGIC | VBSF_HANDLE_F_ON_LIST), ("%p %#x\n", pCur, pCur->fFlags));
451 if ((pCur->fFlags & (fFlagsSet | fFlagsClear)) == fFlagsSet) {
452 uint32_t cRefs = ASMAtomicIncU32(&pCur->cRefs);
453 if (cRefs > 1) {
454 spin_unlock_irqrestore(&g_SfHandleLock, fSavedFlags);
455 SFLOGFLOW(("vbsf_handle_find: returns %p\n", pCur));
456 return pCur;
457 }
458 /* Oops, already being closed (safe as it's only ever increased here). */
459 ASMAtomicDecU32(&pCur->cRefs);
460 }
461 }
462
463 spin_unlock_irqrestore(&g_SfHandleLock, fSavedFlags);
464 SFLOGFLOW(("vbsf_handle_find: returns NULL!\n"));
465 return NULL;
466}
467
468
469/**
470 * Slow worker for vbsf_handle_release() that does the freeing.
471 *
472 * @returns 0 (ref count).
473 * @param pHandle The handle to release.
474 * @param pSuperInfo The info structure for the shared folder associated with
475 * the handle.
476 * @param pszCaller The caller name (for logging failures).
477 */
478uint32_t vbsf_handle_release_slow(struct vbsf_handle *pHandle, struct vbsf_super_info *pSuperInfo, const char *pszCaller)
479{
480 int rc;
481 unsigned long fSavedFlags;
482
483 SFLOGFLOW(("vbsf_handle_release_slow: %p (%s)\n", pHandle, pszCaller));
484
485 /*
486 * Remove from the list.
487 */
488 spin_lock_irqsave(&g_SfHandleLock, fSavedFlags);
489
490 AssertMsg((pHandle->fFlags & VBSF_HANDLE_F_MAGIC_MASK) == VBSF_HANDLE_F_MAGIC, ("%p %#x\n", pHandle, pHandle->fFlags));
491 Assert(pHandle->pInodeInfo);
492 Assert(pHandle->pInodeInfo && pHandle->pInodeInfo->u32Magic == SF_INODE_INFO_MAGIC);
493
494 if (pHandle->fFlags & VBSF_HANDLE_F_ON_LIST) {
495 pHandle->fFlags &= ~VBSF_HANDLE_F_ON_LIST;
496 RTListNodeRemove(&pHandle->Entry);
497 }
498
499 spin_unlock_irqrestore(&g_SfHandleLock, fSavedFlags);
500
501 /*
502 * Actually destroy it.
503 */
504 rc = VbglR0SfHostReqCloseSimple(pSuperInfo->map.root, pHandle->hHost);
505 if (RT_FAILURE(rc))
506 LogFunc(("Caller %s: VbglR0SfHostReqCloseSimple %#RX64 failed with rc=%Rrc\n", pszCaller, pHandle->hHost, rc));
507 pHandle->hHost = SHFL_HANDLE_NIL;
508 pHandle->fFlags = VBSF_HANDLE_F_MAGIC_DEAD;
509 kfree(pHandle);
510 return 0;
511}
512
513
514/**
515 * Appends a handle to a handle list.
516 *
517 * @param pInodeInfo The inode to add it to.
518 * @param pHandle The handle to add.
519 */
520void vbsf_handle_append(struct vbsf_inode_info *pInodeInfo, struct vbsf_handle *pHandle)
521{
522#ifdef VBOX_STRICT
523 struct vbsf_handle *pCur;
524#endif
525 unsigned long fSavedFlags;
526
527 SFLOGFLOW(("vbsf_handle_append: %p (to %p)\n", pHandle, pInodeInfo));
528 AssertMsg((pHandle->fFlags & (VBSF_HANDLE_F_MAGIC_MASK | VBSF_HANDLE_F_ON_LIST)) == VBSF_HANDLE_F_MAGIC,
529 ("%p %#x\n", pHandle, pHandle->fFlags));
530 Assert(pInodeInfo->u32Magic == SF_INODE_INFO_MAGIC);
531
532 spin_lock_irqsave(&g_SfHandleLock, fSavedFlags);
533
534 AssertMsg((pHandle->fFlags & (VBSF_HANDLE_F_MAGIC_MASK | VBSF_HANDLE_F_ON_LIST)) == VBSF_HANDLE_F_MAGIC,
535 ("%p %#x\n", pHandle, pHandle->fFlags));
536#ifdef VBOX_STRICT
537 RTListForEach(&pInodeInfo->HandleList, pCur, struct vbsf_handle, Entry) {
538 Assert(pCur != pHandle);
539 AssertMsg( (pCur->fFlags & (VBSF_HANDLE_F_MAGIC_MASK | VBSF_HANDLE_F_ON_LIST))
540 == (VBSF_HANDLE_F_MAGIC | VBSF_HANDLE_F_ON_LIST), ("%p %#x\n", pCur, pCur->fFlags));
541 }
542 pHandle->pInodeInfo = pInodeInfo;
543#endif
544
545 pHandle->fFlags |= VBSF_HANDLE_F_ON_LIST;
546 RTListAppend(&pInodeInfo->HandleList, &pHandle->Entry);
547
548 spin_unlock_irqrestore(&g_SfHandleLock, fSavedFlags);
549}
550
551
552
553/*********************************************************************************************************************************
554* Misc *
555*********************************************************************************************************************************/
556
557#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 6)
558/** Any writable mappings? */
559DECLINLINE(bool) mapping_writably_mapped(struct address_space const *mapping)
560{
561# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 6)
562 return !list_empty(&mapping->i_mmap_shared);
563# else
564 return mapping->i_mmap_shared != NULL;
565# endif
566}
567#endif
568
569
570#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 5, 12)
571/** Missing in 2.4.x, so just stub it for now. */
572DECLINLINE(bool) PageWriteback(struct page const *page)
573{
574 return false;
575}
576#endif
577
578
579/**
580 * Helper for deciding wheter we should do a read via the page cache or not.
581 *
582 * By default we will only use the page cache if there is a writable memory
583 * mapping of the file with a chance that it may have modified any of the pages
584 * already.
585 */
586DECLINLINE(bool) vbsf_should_use_cached_read(struct file *file, struct address_space *mapping, struct vbsf_super_info *pSuperInfo)
587{
588 if ( (file->f_flags & O_DIRECT)
589 || pSuperInfo->enmCacheMode == kVbsfCacheMode_None)
590 return false;
591 if ( pSuperInfo->enmCacheMode == kVbsfCacheMode_Read
592 || pSuperInfo->enmCacheMode == kVbsfCacheMode_ReadWrite)
593 return true;
594 Assert(pSuperInfo->enmCacheMode == kVbsfCacheMode_Strict);
595 return mapping
596 && mapping->nrpages > 0
597 && mapping_writably_mapped(mapping);
598}
599
600
601
602/*********************************************************************************************************************************
603* Pipe / splice stuff mainly for 2.6.17 >= linux < 2.6.31 (where no fallbacks were available) *
604*********************************************************************************************************************************/
605
606#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 17) \
607 && LINUX_VERSION_CODE < KERNEL_VERSION(3, 16, 0)
608
609# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 30)
610# define LOCK_PIPE(a_pPipe) do { if ((a_pPipe)->inode) mutex_lock(&(a_pPipe)->inode->i_mutex); } while (0)
611# define UNLOCK_PIPE(a_pPipe) do { if ((a_pPipe)->inode) mutex_unlock(&(a_pPipe)->inode->i_mutex); } while (0)
612# else
613# define LOCK_PIPE(a_pPipe) pipe_lock(a_pPipe)
614# define UNLOCK_PIPE(a_pPipe) pipe_unlock(a_pPipe)
615# endif
616
617
618/** Waits for the pipe buffer status to change. */
619static void vbsf_wait_pipe(struct pipe_inode_info *pPipe)
620{
621 DEFINE_WAIT(WaitStuff);
622# ifdef TASK_NONINTERACTIVE
623 prepare_to_wait(&pPipe->wait, &WaitStuff, TASK_INTERRUPTIBLE | TASK_NONINTERACTIVE);
624# else
625 prepare_to_wait(&pPipe->wait, &WaitStuff, TASK_INTERRUPTIBLE);
626# endif
627 UNLOCK_PIPE(pPipe);
628
629 schedule();
630
631 finish_wait(&pPipe->wait, &WaitStuff);
632 LOCK_PIPE(pPipe);
633}
634
635
636/** Worker for vbsf_feed_pages_to_pipe that wakes up readers. */
637static void vbsf_wake_up_pipe(struct pipe_inode_info *pPipe, bool fReaders)
638{
639 smp_mb();
640 if (waitqueue_active(&pPipe->wait))
641 wake_up_interruptible_sync(&pPipe->wait);
642 if (fReaders)
643 kill_fasync(&pPipe->fasync_readers, SIGIO, POLL_IN);
644 else
645 kill_fasync(&pPipe->fasync_writers, SIGIO, POLL_OUT);
646}
647
648#endif
649#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 17) \
650 && LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 31)
651
652/** Verify pipe buffer content (needed for page-cache to ensure idle page). */
653static int vbsf_pipe_buf_confirm(struct pipe_inode_info *pPipe, struct pipe_buffer *pPipeBuf)
654{
655 /*SFLOG3(("vbsf_pipe_buf_confirm: %p\n", pPipeBuf));*/
656 return 0;
657}
658
659
660/** Maps the buffer page. */
661static void *vbsf_pipe_buf_map(struct pipe_inode_info *pPipe, struct pipe_buffer *pPipeBuf, int atomic)
662{
663 void *pvRet;
664 if (!atomic)
665 pvRet = kmap(pPipeBuf->page);
666 else {
667 pPipeBuf->flags |= PIPE_BUF_FLAG_ATOMIC;
668 pvRet = kmap_atomic(pPipeBuf->page, KM_USER0);
669 }
670 /*SFLOG3(("vbsf_pipe_buf_map: %p -> %p\n", pPipeBuf, pvRet));*/
671 return pvRet;
672}
673
674
675/** Unmaps the buffer page. */
676static void vbsf_pipe_buf_unmap(struct pipe_inode_info *pPipe, struct pipe_buffer *pPipeBuf, void *pvMapping)
677{
678 /*SFLOG3(("vbsf_pipe_buf_unmap: %p/%p\n", pPipeBuf, pvMapping)); */
679 if (!(pPipeBuf->flags & PIPE_BUF_FLAG_ATOMIC))
680 kunmap(pPipeBuf->page);
681 else {
682 pPipeBuf->flags &= ~PIPE_BUF_FLAG_ATOMIC;
683 kunmap_atomic(pvMapping, KM_USER0);
684 }
685}
686
687
688/** Gets a reference to the page. */
689static void vbsf_pipe_buf_get(struct pipe_inode_info *pPipe, struct pipe_buffer *pPipeBuf)
690{
691 page_cache_get(pPipeBuf->page);
692 /*SFLOG3(("vbsf_pipe_buf_get: %p (return count=%d)\n", pPipeBuf, page_count(pPipeBuf->page)));*/
693}
694
695
696/** Release the buffer page (counter to vbsf_pipe_buf_get). */
697static void vbsf_pipe_buf_release(struct pipe_inode_info *pPipe, struct pipe_buffer *pPipeBuf)
698{
699 /*SFLOG3(("vbsf_pipe_buf_release: %p (incoming count=%d)\n", pPipeBuf, page_count(pPipeBuf->page)));*/
700 page_cache_release(pPipeBuf->page);
701}
702
703
704/** Attempt to steal the page.
705 * @returns 0 success, 1 on failure. */
706static int vbsf_pipe_buf_steal(struct pipe_inode_info *pPipe, struct pipe_buffer *pPipeBuf)
707{
708 if (page_count(pPipeBuf->page) == 1) {
709 lock_page(pPipeBuf->page);
710 SFLOG3(("vbsf_pipe_buf_steal: %p -> 0\n", pPipeBuf));
711 return 0;
712 }
713 SFLOG3(("vbsf_pipe_buf_steal: %p -> 1\n", pPipeBuf));
714 return 1;
715}
716
717
718/**
719 * Pipe buffer operations for used by vbsf_feed_pages_to_pipe.
720 */
721static struct pipe_buf_operations vbsf_pipe_buf_ops = {
722 .can_merge = 0,
723# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 23)
724 .confirm = vbsf_pipe_buf_confirm,
725# else
726 .pin = vbsf_pipe_buf_confirm,
727# endif
728 .map = vbsf_pipe_buf_map,
729 .unmap = vbsf_pipe_buf_unmap,
730 .get = vbsf_pipe_buf_get,
731 .release = vbsf_pipe_buf_release,
732 .steal = vbsf_pipe_buf_steal,
733};
734
735
736/**
737 * Feeds the pages to the pipe.
738 *
739 * Pages given to the pipe are set to NULL in papPages.
740 */
741static ssize_t vbsf_feed_pages_to_pipe(struct pipe_inode_info *pPipe, struct page **papPages, size_t cPages, uint32_t offPg0,
742 uint32_t cbActual, unsigned fFlags)
743{
744 ssize_t cbRet = 0;
745 size_t iPage = 0;
746 bool fNeedWakeUp = false;
747
748 LOCK_PIPE(pPipe);
749 for (;;) {
750 if ( pPipe->readers > 0
751 && pPipe->nrbufs < PIPE_BUFFERS) {
752 struct pipe_buffer *pPipeBuf = &pPipe->bufs[(pPipe->curbuf + pPipe->nrbufs) % PIPE_BUFFERS];
753 uint32_t const cbThisPage = RT_MIN(cbActual, PAGE_SIZE - offPg0);
754 pPipeBuf->len = cbThisPage;
755 pPipeBuf->offset = offPg0;
756# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 23)
757 pPipeBuf->private = 0;
758# endif
759 pPipeBuf->ops = &vbsf_pipe_buf_ops;
760 pPipeBuf->flags = fFlags & SPLICE_F_GIFT ? PIPE_BUF_FLAG_GIFT : 0;
761 pPipeBuf->page = papPages[iPage];
762
763 papPages[iPage++] = NULL;
764 pPipe->nrbufs++;
765 fNeedWakeUp |= pPipe->inode != NULL;
766 offPg0 = 0;
767 cbRet += cbThisPage;
768
769 /* done? */
770 cbActual -= cbThisPage;
771 if (!cbActual)
772 break;
773 } else if (pPipe->readers == 0) {
774 SFLOGFLOW(("vbsf_feed_pages_to_pipe: no readers!\n"));
775 send_sig(SIGPIPE, current, 0);
776 if (cbRet == 0)
777 cbRet = -EPIPE;
778 break;
779 } else if (fFlags & SPLICE_F_NONBLOCK) {
780 if (cbRet == 0)
781 cbRet = -EAGAIN;
782 break;
783 } else if (signal_pending(current)) {
784 if (cbRet == 0)
785 cbRet = -ERESTARTSYS;
786 SFLOGFLOW(("vbsf_feed_pages_to_pipe: pending signal! (%zd)\n", cbRet));
787 break;
788 } else {
789 if (fNeedWakeUp) {
790 vbsf_wake_up_pipe(pPipe, true /*fReaders*/);
791 fNeedWakeUp = 0;
792 }
793 pPipe->waiting_writers++;
794 vbsf_wait_pipe(pPipe);
795 pPipe->waiting_writers--;
796 }
797 }
798 UNLOCK_PIPE(pPipe);
799
800 if (fNeedWakeUp)
801 vbsf_wake_up_pipe(pPipe, true /*fReaders*/);
802
803 return cbRet;
804}
805
806
807/**
808 * For splicing from a file to a pipe.
809 */
810static ssize_t vbsf_splice_read(struct file *file, loff_t *poffset, struct pipe_inode_info *pipe, size_t len, unsigned int flags)
811{
812 struct inode *inode = VBSF_GET_F_DENTRY(file)->d_inode;
813 struct vbsf_super_info *pSuperInfo = VBSF_GET_SUPER_INFO(inode->i_sb);
814 ssize_t cbRet;
815
816 SFLOGFLOW(("vbsf_splice_read: file=%p poffset=%p{%#RX64} pipe=%p len=%#zx flags=%#x\n", file, poffset, *poffset, pipe, len, flags));
817 if (vbsf_should_use_cached_read(file, inode->i_mapping, pSuperInfo)) {
818 cbRet = generic_file_splice_read(file, poffset, pipe, len, flags);
819 } else {
820 /*
821 * Create a read request.
822 */
823 loff_t offFile = *poffset;
824 size_t cPages = RT_MIN(RT_ALIGN_Z((offFile & ~PAGE_CACHE_MASK) + len, PAGE_CACHE_SIZE) >> PAGE_CACHE_SHIFT,
825 PIPE_BUFFERS);
826 VBOXSFREADPGLSTREQ *pReq = (VBOXSFREADPGLSTREQ *)VbglR0PhysHeapAlloc(RT_UOFFSETOF_DYN(VBOXSFREADPGLSTREQ,
827 PgLst.aPages[cPages]));
828 if (pReq) {
829 /*
830 * Allocate pages.
831 */
832 struct page *apPages[PIPE_BUFFERS];
833 size_t i;
834 pReq->PgLst.offFirstPage = (uint16_t)offFile & (uint16_t)PAGE_OFFSET_MASK;
835 cbRet = 0;
836 for (i = 0; i < cPages; i++) {
837 struct page *pPage;
838 apPages[i] = pPage = alloc_page(GFP_USER);
839 if (pPage) {
840 pReq->PgLst.aPages[i] = page_to_phys(pPage);
841# ifdef VBOX_STRICT
842 ASMMemFill32(kmap(pPage), PAGE_SIZE, UINT32_C(0xdeadbeef));
843 kunmap(pPage);
844# endif
845 } else {
846 cbRet = -ENOMEM;
847 break;
848 }
849 }
850 if (cbRet == 0) {
851 /*
852 * Do the reading.
853 */
854 uint32_t const cbToRead = RT_MIN((cPages << PAGE_SHIFT) - (offFile & PAGE_OFFSET_MASK), len);
855 struct vbsf_reg_info *sf_r = (struct vbsf_reg_info *)file->private_data;
856 int vrc = VbglR0SfHostReqReadPgLst(pSuperInfo->map.root, pReq, sf_r->Handle.hHost, offFile, cbToRead, cPages);
857 if (RT_SUCCESS(vrc)) {
858 /*
859 * Get the number of bytes read, jettison the request
860 * and, in case of EOF, any unnecessary pages.
861 */
862 uint32_t cbActual = pReq->Parms.cb32Read.u.value32;
863 AssertStmt(cbActual <= cbToRead, cbActual = cbToRead);
864 SFLOG2(("vbsf_splice_read: read -> %#x bytes @ %#RX64\n", cbActual, offFile));
865
866 VbglR0PhysHeapFree(pReq);
867 pReq = NULL;
868
869 /*
870 * Now, feed it to the pipe thingy.
871 * This will take ownership of the all pages no matter what happens.
872 */
873 cbRet = vbsf_feed_pages_to_pipe(pipe, apPages, cPages, offFile & PAGE_OFFSET_MASK, cbActual, flags);
874 if (cbRet > 0)
875 *poffset = offFile + cbRet;
876 } else {
877 cbRet = -RTErrConvertToErrno(vrc);
878 SFLOGFLOW(("vbsf_splice_read: Read failed: %Rrc -> %zd\n", vrc, cbRet));
879 }
880 i = cPages;
881 }
882
883 while (i-- > 0)
884 if (apPages[i])
885 __free_pages(apPages[i], 0);
886 if (pReq)
887 VbglR0PhysHeapFree(pReq);
888 } else {
889 cbRet = -ENOMEM;
890 }
891 }
892 SFLOGFLOW(("vbsf_splice_read: returns %zd (%#zx), *poffset=%#RX64\n", cbRet, cbRet, *poffset));
893 return cbRet;
894}
895
896#endif /* 2.6.17 <= LINUX_VERSION_CODE < 2.6.31 */
897#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 17) \
898 && LINUX_VERSION_CODE < KERNEL_VERSION(3, 16, 0)
899
900/**
901 * For splicing from a pipe to a file.
902 *
903 * Since we can combine buffers and request allocations, this should be faster
904 * than the default implementation.
905 */
906static ssize_t vbsf_splice_write(struct pipe_inode_info *pPipe, struct file *file, loff_t *poffset, size_t len, unsigned int flags)
907{
908 struct inode *inode = VBSF_GET_F_DENTRY(file)->d_inode;
909 struct vbsf_super_info *pSuperInfo = VBSF_GET_SUPER_INFO(inode->i_sb);
910 ssize_t cbRet;
911
912 SFLOGFLOW(("vbsf_splice_write: pPipe=%p file=%p poffset=%p{%#RX64} len=%#zx flags=%#x\n", pPipe, file, poffset, *poffset, len, flags));
913 /** @todo later if (false) {
914 cbRet = generic_file_splice_write(pPipe, file, poffset, len, flags);
915 } else */ {
916 /*
917 * Prepare a write request.
918 */
919# ifdef PIPE_BUFFERS
920 uint32_t const cMaxPages = RT_MIN(PIPE_BUFFERS, RT_ALIGN_Z(len, PAGE_SIZE) >> PAGE_SHIFT);
921# else
922 uint32_t const cMaxPages = RT_MIN(RT_MAX(RT_MIN(pPipe->buffers, 256), PIPE_DEF_BUFFERS),
923 RT_ALIGN_Z(len, PAGE_SIZE) >> PAGE_SHIFT);
924# endif
925 VBOXSFWRITEPGLSTREQ *pReq = (VBOXSFWRITEPGLSTREQ *)VbglR0PhysHeapAlloc(RT_UOFFSETOF_DYN(VBOXSFREADPGLSTREQ,
926 PgLst.aPages[cMaxPages]));
927 if (pReq) {
928 /*
929 * Feed from the pipe.
930 */
931 struct vbsf_reg_info *sf_r = (struct vbsf_reg_info *)file->private_data;
932 struct address_space *mapping = inode->i_mapping;
933 loff_t offFile = *poffset;
934 bool fNeedWakeUp = false;
935 cbRet = 0;
936
937 LOCK_PIPE(pPipe);
938
939 for (;;) {
940 unsigned cBufs = pPipe->nrbufs;
941 /*SFLOG2(("vbsf_splice_write: nrbufs=%#x curbuf=%#x\n", cBufs, pPipe->curbuf));*/
942 if (cBufs) {
943 /*
944 * There is data available. Write it to the file.
945 */
946 int vrc;
947 struct pipe_buffer *pPipeBuf = &pPipe->bufs[pPipe->curbuf];
948 uint32_t cPagesToWrite = 1;
949 uint32_t cbToWrite = pPipeBuf->len;
950
951 Assert(pPipeBuf->offset < PAGE_SIZE);
952 Assert(pPipeBuf->offset + pPipeBuf->len <= PAGE_SIZE);
953
954 pReq->PgLst.offFirstPage = pPipeBuf->offset & PAGE_OFFSET;
955 pReq->PgLst.aPages[0] = page_to_phys(pPipeBuf->page);
956
957 /* Add any adjacent page buffers: */
958 while ( cPagesToWrite < cBufs
959 && cPagesToWrite < cMaxPages
960 && ((pReq->PgLst.offFirstPage + cbToWrite) & PAGE_OFFSET_MASK) == 0) {
961# ifdef PIPE_BUFFERS
962 struct pipe_buffer *pPipeBuf2 = &pPipe->bufs[(pPipe->curbuf + cPagesToWrite) % PIPE_BUFFERS];
963# else
964 struct pipe_buffer *pPipeBuf2 = &pPipe->bufs[(pPipe->curbuf + cPagesToWrite) % pPipe->buffers];
965# endif
966 Assert(pPipeBuf2->len <= PAGE_SIZE);
967 Assert(pPipeBuf2->offset < PAGE_SIZE);
968 if (pPipeBuf2->offset != 0)
969 break;
970 pReq->PgLst.aPages[cPagesToWrite] = page_to_phys(pPipeBuf2->page);
971 cbToWrite += pPipeBuf2->len;
972 cPagesToWrite += 1;
973 }
974
975 /* Check that we don't have signals pending before we issue the write, as
976 we'll only end up having to cancel the HGCM request 99% of the time: */
977 if (!signal_pending(current)) {
978 struct vbsf_inode_info *sf_i = VBSF_GET_INODE_INFO(inode);
979 vrc = VbglR0SfHostReqWritePgLst(pSuperInfo->map.root, pReq, sf_r->Handle.hHost, offFile,
980 cbToWrite, cPagesToWrite);
981 sf_i->ModificationTimeAtOurLastWrite = sf_i->ModificationTime;
982 } else
983 vrc = VERR_INTERRUPTED;
984 if (RT_SUCCESS(vrc)) {
985 /*
986 * Get the number of bytes actually written, update file position
987 * and return value, and advance the pipe buffer.
988 */
989 uint32_t cbActual = pReq->Parms.cb32Write.u.value32;
990 AssertStmt(cbActual <= cbToWrite, cbActual = cbToWrite);
991 SFLOG2(("vbsf_splice_write: write -> %#x bytes @ %#RX64\n", cbActual, offFile));
992
993 cbRet += cbActual;
994
995 while (cbActual > 0) {
996 uint32_t cbAdvance = RT_MIN(pPipeBuf->len, cbActual);
997
998 vbsf_reg_write_sync_page_cache(mapping, offFile, cbAdvance, NULL,
999 &pPipeBuf->page, pPipeBuf->offset, 1);
1000
1001 offFile += cbAdvance;
1002 cbActual -= cbAdvance;
1003 pPipeBuf->offset += cbAdvance;
1004 pPipeBuf->len -= cbAdvance;
1005
1006 if (!pPipeBuf->len) {
1007 struct pipe_buf_operations const *pOps = pPipeBuf->ops;
1008 pPipeBuf->ops = NULL;
1009 pOps->release(pPipe, pPipeBuf);
1010
1011# ifdef PIPE_BUFFERS
1012 pPipe->curbuf = (pPipe->curbuf + 1) % PIPE_BUFFERS;
1013# else
1014 pPipe->curbuf = (pPipe->curbuf + 1) % pPipe->buffers;
1015# endif
1016 pPipe->nrbufs -= 1;
1017 pPipeBuf = &pPipe->bufs[pPipe->curbuf];
1018
1019# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 30)
1020 fNeedWakeUp |= pPipe->inode != NULL;
1021# else
1022 fNeedWakeUp = true;
1023# endif
1024 } else {
1025 Assert(cbActual == 0);
1026 break;
1027 }
1028 }
1029
1030 *poffset = offFile;
1031 } else {
1032 if (cbRet == 0)
1033 cbRet = vrc == VERR_INTERRUPTED ? -ERESTARTSYS : -RTErrConvertToErrno(vrc);
1034 SFLOGFLOW(("vbsf_splice_write: Write failed: %Rrc -> %zd (cbRet=%#zx)\n",
1035 vrc, -RTErrConvertToErrno(vrc), cbRet));
1036 break;
1037 }
1038 } else {
1039 /*
1040 * Wait for data to become available, if there is chance that'll happen.
1041 */
1042 /* Quit if there are no writers (think EOF): */
1043 if (pPipe->writers == 0) {
1044 SFLOGFLOW(("vbsf_splice_write: No buffers. No writers. The show is done!\n"));
1045 break;
1046 }
1047
1048 /* Quit if if we've written some and no writers waiting on the lock: */
1049 if (cbRet > 0 && pPipe->waiting_writers == 0) {
1050 SFLOGFLOW(("vbsf_splice_write: No waiting writers, returning what we've got.\n"));
1051 break;
1052 }
1053
1054 /* Quit with EAGAIN if non-blocking: */
1055 if (flags & SPLICE_F_NONBLOCK) {
1056 if (cbRet == 0)
1057 cbRet = -EAGAIN;
1058 break;
1059 }
1060
1061 /* Quit if we've got pending signals: */
1062 if (signal_pending(current)) {
1063 if (cbRet == 0)
1064 cbRet = -ERESTARTSYS;
1065 SFLOGFLOW(("vbsf_splice_write: pending signal! (%zd)\n", cbRet));
1066 break;
1067 }
1068
1069 /* Wake up writers before we start waiting: */
1070 if (fNeedWakeUp) {
1071 vbsf_wake_up_pipe(pPipe, false /*fReaders*/);
1072 fNeedWakeUp = false;
1073 }
1074 vbsf_wait_pipe(pPipe);
1075 }
1076 } /* feed loop */
1077
1078 if (fNeedWakeUp)
1079 vbsf_wake_up_pipe(pPipe, false /*fReaders*/);
1080
1081 UNLOCK_PIPE(pPipe);
1082
1083 VbglR0PhysHeapFree(pReq);
1084 } else {
1085 cbRet = -ENOMEM;
1086 }
1087 }
1088 SFLOGFLOW(("vbsf_splice_write: returns %zd (%#zx), *poffset=%#RX64\n", cbRet, cbRet, *poffset));
1089 return cbRet;
1090}
1091
1092#endif /* 2.6.17 <= LINUX_VERSION_CODE < 3.16.0 */
1093
1094#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 30) \
1095 && LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 23)
1096/**
1097 * Our own senfile implementation that does not go via the page cache like
1098 * generic_file_sendfile() does.
1099 */
1100static ssize_t vbsf_reg_sendfile(struct file *pFile, loff_t *poffFile, size_t cbToSend, read_actor_t pfnActor,
1101# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 8)
1102 void *pvUser
1103# else
1104 void __user *pvUser
1105# endif
1106 )
1107{
1108 struct inode *inode = VBSF_GET_F_DENTRY(pFile)->d_inode;
1109 struct vbsf_super_info *pSuperInfo = VBSF_GET_SUPER_INFO(inode->i_sb);
1110 ssize_t cbRet;
1111 SFLOGFLOW(("vbsf_reg_sendfile: pFile=%p poffFile=%p{%#RX64} cbToSend=%#zx pfnActor=%p pvUser=%p\n",
1112 pFile, poffFile, poffFile ? *poffFile : 0, cbToSend, pfnActor, pvUser));
1113 Assert(pSuperInfo);
1114
1115 /*
1116 * Return immediately if asked to send nothing.
1117 */
1118 if (cbToSend == 0)
1119 return 0;
1120
1121 /*
1122 * Like for vbsf_reg_read() and vbsf_reg_read_iter(), we allow going via
1123 * the page cache in some cases or configs.
1124 */
1125 if (vbsf_should_use_cached_read(pFile, inode->i_mapping, pSuperInfo)) {
1126 cbRet = generic_file_sendfile(pFile, poffFile, cbToSend, pfnActor, pvUser);
1127 SFLOGFLOW(("vbsf_reg_sendfile: returns %#zx *poffFile=%#RX64 [generic_file_sendfile]\n", cbRet, poffFile ? *poffFile : UINT64_MAX));
1128 } else {
1129 /*
1130 * Allocate a request and a bunch of pages for reading from the file.
1131 */
1132 struct page *apPages[16];
1133 loff_t offFile = poffFile ? *poffFile : 0;
1134 size_t const cPages = cbToSend + ((size_t)offFile & PAGE_OFFSET_MASK) >= RT_ELEMENTS(apPages) * PAGE_SIZE
1135 ? RT_ELEMENTS(apPages)
1136 : RT_ALIGN_Z(cbToSend + ((size_t)offFile & PAGE_OFFSET_MASK), PAGE_SIZE) >> PAGE_SHIFT;
1137 size_t iPage;
1138 VBOXSFREADPGLSTREQ *pReq = (VBOXSFREADPGLSTREQ *)VbglR0PhysHeapAlloc(RT_UOFFSETOF_DYN(VBOXSFREADPGLSTREQ,
1139 PgLst.aPages[cPages]));
1140 if (pReq) {
1141 Assert(cPages > 0);
1142 cbRet = 0;
1143 for (iPage = 0; iPage < cPages; iPage++) {
1144 struct page *pPage;
1145 apPages[iPage] = pPage = alloc_page(GFP_USER);
1146 if (pPage) {
1147 Assert(page_count(pPage) == 1);
1148 pReq->PgLst.aPages[iPage] = page_to_phys(pPage);
1149 } else {
1150 while (iPage-- > 0)
1151 vbsf_put_page(apPages[iPage]);
1152 cbRet = -ENOMEM;
1153 break;
1154 }
1155 }
1156 if (cbRet == 0) {
1157 /*
1158 * Do the job.
1159 */
1160 struct vbsf_reg_info *sf_r = (struct vbsf_reg_info *)pFile->private_data;
1161 read_descriptor_t RdDesc;
1162 RdDesc.count = cbToSend;
1163# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 8)
1164 RdDesc.arg.data = pvUser;
1165# else
1166 RdDesc.buf = pvUser;
1167# endif
1168 RdDesc.written = 0;
1169 RdDesc.error = 0;
1170
1171 Assert(sf_r);
1172 Assert((sf_r->Handle.fFlags & VBSF_HANDLE_F_MAGIC_MASK) == VBSF_HANDLE_F_MAGIC);
1173
1174 while (cbToSend > 0) {
1175 /*
1176 * Read another chunk. For paranoid reasons, we keep data where the page cache
1177 * would keep it, i.e. page offset bits corresponds to the file offset bits.
1178 */
1179 uint32_t const offPg0 = (uint32_t)offFile & (uint32_t)PAGE_OFFSET_MASK;
1180 uint32_t const cbToRead = RT_MIN((cPages << PAGE_SHIFT) - offPg0, cbToSend);
1181 uint32_t const cPagesToRead = RT_ALIGN_Z(cbToRead + offPg0, PAGE_SIZE) >> PAGE_SHIFT;
1182 int vrc;
1183 pReq->PgLst.offFirstPage = (uint16_t)offPg0;
1184 if (!signal_pending(current))
1185 vrc = VbglR0SfHostReqReadPgLst(pSuperInfo->map.root, pReq, sf_r->Handle.hHost, offFile,
1186 cbToRead, cPagesToRead);
1187 else
1188 vrc = VERR_INTERRUPTED;
1189 if (RT_SUCCESS(vrc)) {
1190 /*
1191 * Pass what we read to the actor.
1192 */
1193 uint32_t off = offPg0;
1194 uint32_t cbActual = pReq->Parms.cb32Read.u.value32;
1195 bool const fIsEof = cbActual < cbToRead;
1196 AssertStmt(cbActual <= cbToRead, cbActual = cbToRead);
1197 SFLOG3(("vbsf_reg_sendfile: Read %#x bytes (offPg0=%#x), wanted %#x ...\n", cbActual, offPg0, cbToRead));
1198
1199 iPage = 0;
1200 while (cbActual > 0) {
1201 uint32_t const cbPage = RT_MIN(cbActual, PAGE_SIZE - off);
1202 int const cbRetActor = pfnActor(&RdDesc, apPages[iPage], off, cbPage);
1203 Assert(cbRetActor >= 0); /* Returns zero on failure, with RdDesc.error holding the status code. */
1204
1205 AssertMsg(iPage < cPages && iPage < cPagesToRead, ("iPage=%#x cPages=%#x cPagesToRead=%#x\n", iPage, cPages, cPagesToRead));
1206
1207 offFile += cbRetActor;
1208 if ((uint32_t)cbRetActor == cbPage && RdDesc.count > 0) {
1209 cbActual -= cbPage;
1210 cbToSend -= cbPage;
1211 iPage++;
1212 } else {
1213 SFLOG3(("vbsf_reg_sendfile: cbRetActor=%#x (%d) cbPage=%#x RdDesc{count=%#lx error=%d} iPage=%#x/%#x/%#x cbToSend=%#zx\n",
1214 cbRetActor, cbRetActor, cbPage, RdDesc.count, RdDesc.error, iPage, cPagesToRead, cPages, cbToSend));
1215 vrc = VERR_CALLBACK_RETURN;
1216 break;
1217 }
1218 off = 0;
1219 }
1220
1221 /*
1222 * Are we done yet?
1223 */
1224 if (RT_FAILURE_NP(vrc) || cbToSend == 0 || RdDesc.error != 0 || fIsEof) {
1225 break;
1226 }
1227
1228 /*
1229 * Replace pages held by the actor.
1230 */
1231 vrc = VINF_SUCCESS;
1232 for (iPage = 0; iPage < cPages; iPage++) {
1233 struct page *pPage = apPages[iPage];
1234 if (page_count(pPage) != 1) {
1235 struct page *pNewPage = alloc_page(GFP_USER);
1236 if (pNewPage) {
1237 SFLOGFLOW(("vbsf_reg_sendfile: Replacing page #%x: %p -> %p\n", iPage, pPage, pNewPage));
1238 vbsf_put_page(pPage);
1239 apPages[iPage] = pNewPage;
1240 } else {
1241 SFLOGFLOW(("vbsf_reg_sendfile: Failed to allocate a replacement page.\n"));
1242 vrc = VERR_NO_MEMORY;
1243 break;
1244 }
1245 }
1246 }
1247 if (RT_FAILURE(vrc))
1248 break; /* RdDesc.written should be non-zero, so don't bother with setting error. */
1249 } else {
1250 RdDesc.error = vrc == VERR_INTERRUPTED ? -ERESTARTSYS : -RTErrConvertToErrno(vrc);
1251 SFLOGFLOW(("vbsf_reg_sendfile: Read failed: %Rrc -> %zd (RdDesc.error=%#d)\n",
1252 vrc, -RTErrConvertToErrno(vrc), RdDesc.error));
1253 break;
1254 }
1255 }
1256
1257 /*
1258 * Free memory.
1259 */
1260 for (iPage = 0; iPage < cPages; iPage++)
1261 vbsf_put_page(apPages[iPage]);
1262
1263 /*
1264 * Set the return values.
1265 */
1266 if (RdDesc.written) {
1267 cbRet = RdDesc.written;
1268 if (poffFile)
1269 *poffFile = offFile;
1270 } else {
1271 cbRet = RdDesc.error;
1272 }
1273 }
1274 VbglR0PhysHeapFree(pReq);
1275 } else {
1276 cbRet = -ENOMEM;
1277 }
1278 SFLOGFLOW(("vbsf_reg_sendfile: returns %#zx offFile=%#RX64\n", cbRet, offFile));
1279 }
1280 return cbRet;
1281}
1282#endif /* 2.5.30 <= LINUX_VERSION_CODE < 2.6.23 */
1283
1284
1285/*********************************************************************************************************************************
1286* File operations on regular files *
1287*********************************************************************************************************************************/
1288
1289/** Wrapper around put_page / page_cache_release. */
1290DECLINLINE(void) vbsf_put_page(struct page *pPage)
1291{
1292#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 6, 0)
1293 put_page(pPage);
1294#else
1295 page_cache_release(pPage);
1296#endif
1297}
1298
1299
1300/** Wrapper around get_page / page_cache_get. */
1301DECLINLINE(void) vbsf_get_page(struct page *pPage)
1302{
1303#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 6, 0)
1304 get_page(pPage);
1305#else
1306 page_cache_get(pPage);
1307#endif
1308}
1309
1310
1311/** Companion to vbsf_lock_user_pages(). */
1312static void vbsf_unlock_user_pages(struct page **papPages, size_t cPages, bool fSetDirty, bool fLockPgHack)
1313{
1314 /* We don't mark kernel pages dirty: */
1315 if (fLockPgHack)
1316 fSetDirty = false;
1317
1318 while (cPages-- > 0)
1319 {
1320 struct page *pPage = papPages[cPages];
1321 Assert((ssize_t)cPages >= 0);
1322 if (fSetDirty && !PageReserved(pPage))
1323 set_page_dirty(pPage);
1324 vbsf_put_page(pPage);
1325 }
1326}
1327
1328
1329/**
1330 * Worker for vbsf_lock_user_pages_failed_check_kernel() and
1331 * vbsf_iter_lock_pages().
1332 */
1333static int vbsf_lock_kernel_pages(uint8_t *pbStart, bool fWrite, size_t cPages, struct page **papPages)
1334{
1335 uintptr_t const uPtrFrom = (uintptr_t)pbStart;
1336 uintptr_t const uPtrLast = (uPtrFrom & ~(uintptr_t)PAGE_OFFSET_MASK) + (cPages << PAGE_SHIFT) - 1;
1337 uint8_t *pbPage = (uint8_t *)uPtrLast;
1338 size_t iPage = cPages;
1339
1340 /*
1341 * Touch the pages first (paranoia^2).
1342 */
1343 if (fWrite) {
1344 uint8_t volatile *pbProbe = (uint8_t volatile *)uPtrFrom;
1345 while (iPage-- > 0) {
1346 *pbProbe = *pbProbe;
1347 pbProbe += PAGE_SIZE;
1348 }
1349 } else {
1350 uint8_t const *pbProbe = (uint8_t const *)uPtrFrom;
1351 while (iPage-- > 0) {
1352 ASMProbeReadByte(pbProbe);
1353 pbProbe += PAGE_SIZE;
1354 }
1355 }
1356
1357 /*
1358 * Get the pages.
1359 * Note! Fixes here probably applies to rtR0MemObjNativeLockKernel as well.
1360 */
1361 iPage = cPages;
1362 if ( uPtrFrom >= (unsigned long)__va(0)
1363 && uPtrLast < (unsigned long)high_memory) {
1364 /* The physical page mapping area: */
1365 while (iPage-- > 0) {
1366 struct page *pPage = papPages[iPage] = virt_to_page(pbPage);
1367 vbsf_get_page(pPage);
1368 pbPage -= PAGE_SIZE;
1369 }
1370 } else {
1371 /* This is vmalloc or some such thing, so go thru page tables: */
1372 while (iPage-- > 0) {
1373 struct page *pPage = rtR0MemObjLinuxVirtToPage(pbPage);
1374 if (pPage) {
1375 papPages[iPage] = pPage;
1376 vbsf_get_page(pPage);
1377 pbPage -= PAGE_SIZE;
1378 } else {
1379 while (++iPage < cPages) {
1380 pPage = papPages[iPage];
1381 vbsf_put_page(pPage);
1382 }
1383 return -EFAULT;
1384 }
1385 }
1386 }
1387 return 0;
1388}
1389
1390
1391/**
1392 * Catches kernel_read() and kernel_write() calls and works around them.
1393 *
1394 * The file_operations::read and file_operations::write callbacks supposedly
1395 * hands us the user buffers to read into and write out of. To allow the kernel
1396 * to read and write without allocating buffers in userland, they kernel_read()
1397 * and kernel_write() increases the user space address limit before calling us
1398 * so that copyin/copyout won't reject it. Our problem is that get_user_pages()
1399 * works on the userspace address space structures and will not be fooled by an
1400 * increased addr_limit.
1401 *
1402 * This code tries to detect this situation and fake get_user_lock() for the
1403 * kernel buffer.
1404 */
1405static int vbsf_lock_user_pages_failed_check_kernel(uintptr_t uPtrFrom, size_t cPages, bool fWrite, int rcFailed,
1406 struct page **papPages, bool *pfLockPgHack)
1407{
1408 /*
1409 * Check that this is valid user memory that is actually in the kernel range.
1410 */
1411#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 0, 0) || defined(RHEL_81)
1412 if ( access_ok((void *)uPtrFrom, cPages << PAGE_SHIFT)
1413 && uPtrFrom >= USER_DS.seg)
1414#else
1415 if ( access_ok(fWrite ? VERIFY_WRITE : VERIFY_READ, (void *)uPtrFrom, cPages << PAGE_SHIFT)
1416 && uPtrFrom >= USER_DS.seg)
1417#endif
1418 {
1419 int rc = vbsf_lock_kernel_pages((uint8_t *)uPtrFrom, fWrite, cPages, papPages);
1420 if (rc == 0) {
1421 *pfLockPgHack = true;
1422 return 0;
1423 }
1424 }
1425
1426 return rcFailed;
1427}
1428
1429
1430/** Wrapper around get_user_pages. */
1431DECLINLINE(int) vbsf_lock_user_pages(uintptr_t uPtrFrom, size_t cPages, bool fWrite, struct page **papPages, bool *pfLockPgHack)
1432{
1433# if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 9, 0)
1434 ssize_t cPagesLocked = get_user_pages_unlocked(uPtrFrom, cPages, papPages,
1435 fWrite ? FOLL_WRITE | FOLL_FORCE : FOLL_FORCE);
1436# elif LINUX_VERSION_CODE >= KERNEL_VERSION(4, 6, 0)
1437 ssize_t cPagesLocked = get_user_pages_unlocked(uPtrFrom, cPages, fWrite, 1 /*force*/, papPages);
1438# elif LINUX_VERSION_CODE >= KERNEL_VERSION(4, 4, 168) && LINUX_VERSION_CODE < KERNEL_VERSION(4, 5, 0)
1439 ssize_t cPagesLocked = get_user_pages_unlocked(current, current->mm, uPtrFrom, cPages, papPages,
1440 fWrite ? FOLL_WRITE | FOLL_FORCE : FOLL_FORCE);
1441# elif LINUX_VERSION_CODE >= KERNEL_VERSION(4, 0, 0)
1442 ssize_t cPagesLocked = get_user_pages_unlocked(current, current->mm, uPtrFrom, cPages, fWrite, 1 /*force*/, papPages);
1443# else
1444 struct task_struct *pTask = current;
1445 ssize_t cPagesLocked;
1446 down_read(&pTask->mm->mmap_sem);
1447 cPagesLocked = get_user_pages(pTask, pTask->mm, uPtrFrom, cPages, fWrite, 1 /*force*/, papPages, NULL);
1448 up_read(&pTask->mm->mmap_sem);
1449# endif
1450 *pfLockPgHack = false;
1451 if (cPagesLocked == cPages)
1452 return 0;
1453
1454 /*
1455 * It failed.
1456 */
1457 if (cPagesLocked < 0)
1458 return vbsf_lock_user_pages_failed_check_kernel(uPtrFrom, cPages, fWrite, (int)cPagesLocked, papPages, pfLockPgHack);
1459
1460 vbsf_unlock_user_pages(papPages, cPagesLocked, false /*fSetDirty*/, false /*fLockPgHack*/);
1461
1462 /* We could use uPtrFrom + cPagesLocked to get the correct status here... */
1463 return -EFAULT;
1464}
1465
1466
1467/**
1468 * Read function used when accessing files that are memory mapped.
1469 *
1470 * We read from the page cache here to present the a cohertent picture of the
1471 * the file content.
1472 */
1473static ssize_t vbsf_reg_read_mapped(struct file *file, char /*__user*/ *buf, size_t size, loff_t *off)
1474{
1475#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 16, 0)
1476 struct iovec iov = { .iov_base = buf, .iov_len = size };
1477 struct iov_iter iter;
1478 struct kiocb kiocb;
1479 ssize_t cbRet;
1480
1481 init_sync_kiocb(&kiocb, file);
1482 kiocb.ki_pos = *off;
1483 iov_iter_init(&iter, READ, &iov, 1, size);
1484
1485 cbRet = generic_file_read_iter(&kiocb, &iter);
1486
1487 *off = kiocb.ki_pos;
1488 return cbRet;
1489
1490#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
1491 struct iovec iov = { .iov_base = buf, .iov_len = size };
1492 struct kiocb kiocb;
1493 ssize_t cbRet;
1494
1495 init_sync_kiocb(&kiocb, file);
1496 kiocb.ki_pos = *off;
1497
1498 cbRet = generic_file_aio_read(&kiocb, &iov, 1, *off);
1499 if (cbRet == -EIOCBQUEUED)
1500 cbRet = wait_on_sync_kiocb(&kiocb);
1501
1502 *off = kiocb.ki_pos;
1503 return cbRet;
1504
1505#else /* 2.6.18 or earlier: */
1506 return generic_file_read(file, buf, size, off);
1507#endif
1508}
1509
1510
1511/**
1512 * Fallback case of vbsf_reg_read() that locks the user buffers and let the host
1513 * write directly to them.
1514 */
1515static ssize_t vbsf_reg_read_locking(struct file *file, char /*__user*/ *buf, size_t size, loff_t *off,
1516 struct vbsf_super_info *pSuperInfo, struct vbsf_reg_info *sf_r)
1517{
1518 /*
1519 * Lock pages and execute the read, taking care not to pass the host
1520 * more than it can handle in one go or more than we care to allocate
1521 * page arrays for. The latter limit is set at just short of 32KB due
1522 * to how the physical heap works.
1523 */
1524 struct page *apPagesStack[16];
1525 struct page **papPages = &apPagesStack[0];
1526 struct page **papPagesFree = NULL;
1527 VBOXSFREADPGLSTREQ *pReq;
1528 loff_t offFile = *off;
1529 ssize_t cbRet = -ENOMEM;
1530 size_t cPages = (((uintptr_t)buf & PAGE_OFFSET_MASK) + size + PAGE_OFFSET_MASK) >> PAGE_SHIFT;
1531 size_t cMaxPages = RT_MIN(RT_MAX(pSuperInfo->cMaxIoPages, 1), cPages);
1532 bool fLockPgHack;
1533
1534 pReq = (VBOXSFREADPGLSTREQ *)VbglR0PhysHeapAlloc(RT_UOFFSETOF_DYN(VBOXSFREADPGLSTREQ, PgLst.aPages[cMaxPages]));
1535 while (!pReq && cMaxPages > 4) {
1536 cMaxPages /= 2;
1537 pReq = (VBOXSFREADPGLSTREQ *)VbglR0PhysHeapAlloc(RT_UOFFSETOF_DYN(VBOXSFREADPGLSTREQ, PgLst.aPages[cMaxPages]));
1538 }
1539 if (pReq && cMaxPages > RT_ELEMENTS(apPagesStack))
1540 papPagesFree = papPages = kmalloc(cMaxPages * sizeof(sizeof(papPages[0])), GFP_KERNEL);
1541 if (pReq && papPages) {
1542 cbRet = 0;
1543 for (;;) {
1544 /*
1545 * Figure out how much to process now and lock the user pages.
1546 */
1547 int rc;
1548 size_t cbChunk = (uintptr_t)buf & PAGE_OFFSET_MASK;
1549 pReq->PgLst.offFirstPage = (uint16_t)cbChunk;
1550 cPages = RT_ALIGN_Z(cbChunk + size, PAGE_SIZE) >> PAGE_SHIFT;
1551 if (cPages <= cMaxPages)
1552 cbChunk = size;
1553 else {
1554 cPages = cMaxPages;
1555 cbChunk = (cMaxPages << PAGE_SHIFT) - cbChunk;
1556 }
1557
1558 rc = vbsf_lock_user_pages((uintptr_t)buf, cPages, true /*fWrite*/, papPages, &fLockPgHack);
1559 if (rc == 0) {
1560 size_t iPage = cPages;
1561 while (iPage-- > 0)
1562 pReq->PgLst.aPages[iPage] = page_to_phys(papPages[iPage]);
1563 } else {
1564 cbRet = rc;
1565 break;
1566 }
1567
1568 /*
1569 * Issue the request and unlock the pages.
1570 */
1571 rc = VbglR0SfHostReqReadPgLst(pSuperInfo->map.root, pReq, sf_r->Handle.hHost, offFile, cbChunk, cPages);
1572
1573 Assert(cPages <= cMaxPages);
1574 vbsf_unlock_user_pages(papPages, cPages, true /*fSetDirty*/, fLockPgHack);
1575
1576 if (RT_SUCCESS(rc)) {
1577 /*
1578 * Success, advance position and buffer.
1579 */
1580 uint32_t cbActual = pReq->Parms.cb32Read.u.value32;
1581 AssertStmt(cbActual <= cbChunk, cbActual = cbChunk);
1582 cbRet += cbActual;
1583 offFile += cbActual;
1584 buf = (uint8_t *)buf + cbActual;
1585 size -= cbActual;
1586
1587 /*
1588 * Are we done already? If so commit the new file offset.
1589 */
1590 if (!size || cbActual < cbChunk) {
1591 *off = offFile;
1592 break;
1593 }
1594 } else if (rc == VERR_NO_MEMORY && cMaxPages > 4) {
1595 /*
1596 * The host probably doesn't have enough heap to handle the
1597 * request, reduce the page count and retry.
1598 */
1599 cMaxPages /= 4;
1600 Assert(cMaxPages > 0);
1601 } else {
1602 /*
1603 * If we've successfully read stuff, return it rather than
1604 * the error. (Not sure if this is such a great idea...)
1605 */
1606 if (cbRet > 0) {
1607 SFLOGFLOW(("vbsf_reg_read: read at %#RX64 -> %Rrc; got cbRet=%#zx already\n", offFile, rc, cbRet));
1608 *off = offFile;
1609 } else {
1610 SFLOGFLOW(("vbsf_reg_read: read at %#RX64 -> %Rrc\n", offFile, rc));
1611 cbRet = -EPROTO;
1612 }
1613 break;
1614 }
1615 }
1616 }
1617 if (papPagesFree)
1618 kfree(papPages);
1619 if (pReq)
1620 VbglR0PhysHeapFree(pReq);
1621 SFLOGFLOW(("vbsf_reg_read: returns %zd (%#zx), *off=%RX64 [lock]\n", cbRet, cbRet, *off));
1622 return cbRet;
1623}
1624
1625
1626/**
1627 * Read from a regular file.
1628 *
1629 * @param file the file
1630 * @param buf the buffer
1631 * @param size length of the buffer
1632 * @param off offset within the file (in/out).
1633 * @returns the number of read bytes on success, Linux error code otherwise
1634 */
1635static ssize_t vbsf_reg_read(struct file *file, char /*__user*/ *buf, size_t size, loff_t *off)
1636{
1637 struct inode *inode = VBSF_GET_F_DENTRY(file)->d_inode;
1638 struct vbsf_super_info *pSuperInfo = VBSF_GET_SUPER_INFO(inode->i_sb);
1639 struct vbsf_reg_info *sf_r = file->private_data;
1640 struct address_space *mapping = inode->i_mapping;
1641
1642 SFLOGFLOW(("vbsf_reg_read: inode=%p file=%p buf=%p size=%#zx off=%#llx\n", inode, file, buf, size, *off));
1643
1644 if (!S_ISREG(inode->i_mode)) {
1645 LogFunc(("read from non regular file %d\n", inode->i_mode));
1646 return -EINVAL;
1647 }
1648
1649 /** @todo XXX Check read permission according to inode->i_mode! */
1650
1651 if (!size)
1652 return 0;
1653
1654 /*
1655 * If there is a mapping and O_DIRECT isn't in effect, we must at a
1656 * heed dirty pages in the mapping and read from them. For simplicity
1657 * though, we just do page cache reading when there are writable
1658 * mappings around with any kind of pages loaded.
1659 */
1660 if (vbsf_should_use_cached_read(file, mapping, pSuperInfo))
1661 return vbsf_reg_read_mapped(file, buf, size, off);
1662
1663 /*
1664 * For small requests, try use an embedded buffer provided we get a heap block
1665 * that does not cross page boundraries (see host code).
1666 */
1667 if (size <= PAGE_SIZE / 4 * 3 - RT_UOFFSETOF(VBOXSFREADEMBEDDEDREQ, abData[0]) /* see allocator */) {
1668 uint32_t const cbReq = RT_UOFFSETOF(VBOXSFREADEMBEDDEDREQ, abData[0]) + size;
1669 VBOXSFREADEMBEDDEDREQ *pReq = (VBOXSFREADEMBEDDEDREQ *)VbglR0PhysHeapAlloc(cbReq);
1670 if (pReq) {
1671 if ((PAGE_SIZE - ((uintptr_t)pReq & PAGE_OFFSET_MASK)) >= cbReq) {
1672 ssize_t cbRet;
1673 int vrc = VbglR0SfHostReqReadEmbedded(pSuperInfo->map.root, pReq, sf_r->Handle.hHost, *off, (uint32_t)size);
1674 if (RT_SUCCESS(vrc)) {
1675 cbRet = pReq->Parms.cb32Read.u.value32;
1676 AssertStmt(cbRet <= (ssize_t)size, cbRet = size);
1677 if (copy_to_user(buf, pReq->abData, cbRet) == 0)
1678 *off += cbRet;
1679 else
1680 cbRet = -EFAULT;
1681 } else
1682 cbRet = -EPROTO;
1683 VbglR0PhysHeapFree(pReq);
1684 SFLOGFLOW(("vbsf_reg_read: returns %zd (%#zx), *off=%RX64 [embed]\n", cbRet, cbRet, *off));
1685 return cbRet;
1686 }
1687 VbglR0PhysHeapFree(pReq);
1688 }
1689 }
1690
1691#if 0 /* Turns out this is slightly slower than locking the pages even for 4KB reads (4.19/amd64). */
1692 /*
1693 * For medium sized requests try use a bounce buffer.
1694 */
1695 if (size <= _64K /** @todo make this configurable? */) {
1696 void *pvBounce = kmalloc(size, GFP_KERNEL);
1697 if (pvBounce) {
1698 VBOXSFREADPGLSTREQ *pReq = (VBOXSFREADPGLSTREQ *)VbglR0PhysHeapAlloc(sizeof(*pReq));
1699 if (pReq) {
1700 ssize_t cbRet;
1701 int vrc = VbglR0SfHostReqReadContig(pSuperInfo->map.root, pReq, sf_r->Handle.hHost, *off,
1702 (uint32_t)size, pvBounce, virt_to_phys(pvBounce));
1703 if (RT_SUCCESS(vrc)) {
1704 cbRet = pReq->Parms.cb32Read.u.value32;
1705 AssertStmt(cbRet <= (ssize_t)size, cbRet = size);
1706 if (copy_to_user(buf, pvBounce, cbRet) == 0)
1707 *off += cbRet;
1708 else
1709 cbRet = -EFAULT;
1710 } else
1711 cbRet = -EPROTO;
1712 VbglR0PhysHeapFree(pReq);
1713 kfree(pvBounce);
1714 SFLOGFLOW(("vbsf_reg_read: returns %zd (%#zx), *off=%RX64 [bounce]\n", cbRet, cbRet, *off));
1715 return cbRet;
1716 }
1717 kfree(pvBounce);
1718 }
1719 }
1720#endif
1721
1722 return vbsf_reg_read_locking(file, buf, size, off, pSuperInfo, sf_r);
1723}
1724
1725
1726/**
1727 * Helper the synchronizes the page cache content with something we just wrote
1728 * to the host.
1729 */
1730static void vbsf_reg_write_sync_page_cache(struct address_space *mapping, loff_t offFile, uint32_t cbRange,
1731 uint8_t const *pbSrcBuf, struct page **papSrcPages,
1732 uint32_t offSrcPage, size_t cSrcPages)
1733{
1734 Assert(offSrcPage < PAGE_SIZE);
1735 if (mapping && mapping->nrpages > 0) {
1736 /*
1737 * Work the pages in the write range.
1738 */
1739 while (cbRange > 0) {
1740 /*
1741 * Lookup the page at offFile. We're fine if there aren't
1742 * any there. We're skip if it's dirty or is being written
1743 * back, at least for now.
1744 */
1745 size_t const offDstPage = offFile & PAGE_OFFSET_MASK;
1746 size_t const cbToCopy = RT_MIN(PAGE_SIZE - offDstPage, cbRange);
1747 pgoff_t const idxPage = offFile >> PAGE_SHIFT;
1748 struct page *pDstPage = find_lock_page(mapping, idxPage);
1749 if (pDstPage) {
1750 if ( pDstPage->mapping == mapping /* ignore if re-purposed (paranoia) */
1751 && pDstPage->index == idxPage
1752 && !PageDirty(pDstPage) /* ignore if dirty */
1753 && !PageWriteback(pDstPage) /* ignore if being written back */ ) {
1754 /*
1755 * Map the page and do the copying.
1756 */
1757 uint8_t *pbDst = (uint8_t *)kmap(pDstPage);
1758 if (pbSrcBuf)
1759 memcpy(&pbDst[offDstPage], pbSrcBuf, cbToCopy);
1760 else {
1761 uint32_t const cbSrc0 = PAGE_SIZE - offSrcPage;
1762 uint8_t const *pbSrc = (uint8_t const *)kmap(papSrcPages[0]);
1763 AssertMsg(cSrcPages >= 1, ("offFile=%#llx cbRange=%#zx cbToCopy=%#zx\n", offFile, cbRange, cbToCopy));
1764 memcpy(&pbDst[offDstPage], &pbSrc[offSrcPage], RT_MIN(cbToCopy, cbSrc0));
1765 kunmap(papSrcPages[0]);
1766 if (cbToCopy > cbSrc0) {
1767 AssertMsg(cSrcPages >= 2, ("offFile=%#llx cbRange=%#zx cbToCopy=%#zx\n", offFile, cbRange, cbToCopy));
1768 pbSrc = (uint8_t const *)kmap(papSrcPages[1]);
1769 memcpy(&pbDst[offDstPage + cbSrc0], pbSrc, cbToCopy - cbSrc0);
1770 kunmap(papSrcPages[1]);
1771 }
1772 }
1773 kunmap(pDstPage);
1774 flush_dcache_page(pDstPage);
1775 if (cbToCopy == PAGE_SIZE)
1776 SetPageUptodate(pDstPage);
1777# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 10)
1778 mark_page_accessed(pDstPage);
1779# endif
1780 } else
1781 SFLOGFLOW(("vbsf_reg_write_sync_page_cache: Skipping page %p: mapping=%p (vs %p) writeback=%d offset=%#lx (vs%#lx)\n",
1782 pDstPage, pDstPage->mapping, mapping, PageWriteback(pDstPage), pDstPage->index, idxPage));
1783 unlock_page(pDstPage);
1784 vbsf_put_page(pDstPage);
1785 }
1786
1787 /*
1788 * Advance.
1789 */
1790 if (pbSrcBuf)
1791 pbSrcBuf += cbToCopy;
1792 else
1793 {
1794 offSrcPage += cbToCopy;
1795 Assert(offSrcPage < PAGE_SIZE * 2);
1796 if (offSrcPage >= PAGE_SIZE) {
1797 offSrcPage &= PAGE_OFFSET_MASK;
1798 papSrcPages++;
1799# ifdef VBOX_STRICT
1800 Assert(cSrcPages > 0);
1801 cSrcPages--;
1802# endif
1803 }
1804 }
1805 offFile += cbToCopy;
1806 cbRange -= cbToCopy;
1807 }
1808 }
1809 RT_NOREF(cSrcPages);
1810}
1811
1812
1813/**
1814 * Fallback case of vbsf_reg_write() that locks the user buffers and let the host
1815 * write directly to them.
1816 */
1817static ssize_t vbsf_reg_write_locking(struct file *file, const char /*__user*/ *buf, size_t size, loff_t *off, loff_t offFile,
1818 struct inode *inode, struct vbsf_inode_info *sf_i,
1819 struct vbsf_super_info *pSuperInfo, struct vbsf_reg_info *sf_r)
1820{
1821 /*
1822 * Lock pages and execute the write, taking care not to pass the host
1823 * more than it can handle in one go or more than we care to allocate
1824 * page arrays for. The latter limit is set at just short of 32KB due
1825 * to how the physical heap works.
1826 */
1827 struct page *apPagesStack[16];
1828 struct page **papPages = &apPagesStack[0];
1829 struct page **papPagesFree = NULL;
1830 VBOXSFWRITEPGLSTREQ *pReq;
1831 ssize_t cbRet = -ENOMEM;
1832 size_t cPages = (((uintptr_t)buf & PAGE_OFFSET_MASK) + size + PAGE_OFFSET_MASK) >> PAGE_SHIFT;
1833 size_t cMaxPages = RT_MIN(RT_MAX(pSuperInfo->cMaxIoPages, 1), cPages);
1834 bool fLockPgHack;
1835
1836 pReq = (VBOXSFWRITEPGLSTREQ *)VbglR0PhysHeapAlloc(RT_UOFFSETOF_DYN(VBOXSFWRITEPGLSTREQ, PgLst.aPages[cMaxPages]));
1837 while (!pReq && cMaxPages > 4) {
1838 cMaxPages /= 2;
1839 pReq = (VBOXSFWRITEPGLSTREQ *)VbglR0PhysHeapAlloc(RT_UOFFSETOF_DYN(VBOXSFWRITEPGLSTREQ, PgLst.aPages[cMaxPages]));
1840 }
1841 if (pReq && cMaxPages > RT_ELEMENTS(apPagesStack))
1842 papPagesFree = papPages = kmalloc(cMaxPages * sizeof(sizeof(papPages[0])), GFP_KERNEL);
1843 if (pReq && papPages) {
1844 cbRet = 0;
1845 for (;;) {
1846 /*
1847 * Figure out how much to process now and lock the user pages.
1848 */
1849 int rc;
1850 size_t cbChunk = (uintptr_t)buf & PAGE_OFFSET_MASK;
1851 pReq->PgLst.offFirstPage = (uint16_t)cbChunk;
1852 cPages = RT_ALIGN_Z(cbChunk + size, PAGE_SIZE) >> PAGE_SHIFT;
1853 if (cPages <= cMaxPages)
1854 cbChunk = size;
1855 else {
1856 cPages = cMaxPages;
1857 cbChunk = (cMaxPages << PAGE_SHIFT) - cbChunk;
1858 }
1859
1860 rc = vbsf_lock_user_pages((uintptr_t)buf, cPages, false /*fWrite*/, papPages, &fLockPgHack);
1861 if (rc == 0) {
1862 size_t iPage = cPages;
1863 while (iPage-- > 0)
1864 pReq->PgLst.aPages[iPage] = page_to_phys(papPages[iPage]);
1865 } else {
1866 cbRet = rc;
1867 break;
1868 }
1869
1870 /*
1871 * Issue the request and unlock the pages.
1872 */
1873 rc = VbglR0SfHostReqWritePgLst(pSuperInfo->map.root, pReq, sf_r->Handle.hHost, offFile, cbChunk, cPages);
1874 sf_i->ModificationTimeAtOurLastWrite = sf_i->ModificationTime;
1875 if (RT_SUCCESS(rc)) {
1876 /*
1877 * Success, advance position and buffer.
1878 */
1879 uint32_t cbActual = pReq->Parms.cb32Write.u.value32;
1880 AssertStmt(cbActual <= cbChunk, cbActual = cbChunk);
1881
1882 vbsf_reg_write_sync_page_cache(inode->i_mapping, offFile, cbActual, NULL /*pbKrnlBuf*/,
1883 papPages, (uintptr_t)buf & PAGE_OFFSET_MASK, cPages);
1884 Assert(cPages <= cMaxPages);
1885 vbsf_unlock_user_pages(papPages, cPages, false /*fSetDirty*/, fLockPgHack);
1886
1887 cbRet += cbActual;
1888 buf = (uint8_t *)buf + cbActual;
1889 size -= cbActual;
1890
1891 offFile += cbActual;
1892 if ((file->f_flags & O_APPEND) && (g_fSfFeatures & SHFL_FEATURE_WRITE_UPDATES_OFFSET))
1893 offFile = pReq->Parms.off64Write.u.value64;
1894 if (offFile > i_size_read(inode))
1895 i_size_write(inode, offFile);
1896
1897 sf_i->force_restat = 1; /* mtime (and size) may have changed */
1898
1899 /*
1900 * Are we done already? If so commit the new file offset.
1901 */
1902 if (!size || cbActual < cbChunk) {
1903 *off = offFile;
1904 break;
1905 }
1906 } else {
1907 vbsf_unlock_user_pages(papPages, cPages, false /*fSetDirty*/, fLockPgHack);
1908 if (rc == VERR_NO_MEMORY && cMaxPages > 4) {
1909 /*
1910 * The host probably doesn't have enough heap to handle the
1911 * request, reduce the page count and retry.
1912 */
1913 cMaxPages /= 4;
1914 Assert(cMaxPages > 0);
1915 } else {
1916 /*
1917 * If we've successfully written stuff, return it rather than
1918 * the error. (Not sure if this is such a great idea...)
1919 */
1920 if (cbRet > 0) {
1921 SFLOGFLOW(("vbsf_reg_write: write at %#RX64 -> %Rrc; got cbRet=%#zx already\n", offFile, rc, cbRet));
1922 *off = offFile;
1923 } else {
1924 SFLOGFLOW(("vbsf_reg_write: write at %#RX64 -> %Rrc\n", offFile, rc));
1925 cbRet = -EPROTO;
1926 }
1927 break;
1928 }
1929 }
1930 }
1931 }
1932 if (papPagesFree)
1933 kfree(papPages);
1934 if (pReq)
1935 VbglR0PhysHeapFree(pReq);
1936 SFLOGFLOW(("vbsf_reg_write: returns %zd (%#zx), *off=%RX64 [lock]\n", cbRet, cbRet, *off));
1937 return cbRet;
1938}
1939
1940
1941/**
1942 * Write to a regular file.
1943 *
1944 * @param file the file
1945 * @param buf the buffer
1946 * @param size length of the buffer
1947 * @param off offset within the file
1948 * @returns the number of written bytes on success, Linux error code otherwise
1949 */
1950static ssize_t vbsf_reg_write(struct file *file, const char *buf, size_t size, loff_t * off)
1951{
1952 struct inode *inode = VBSF_GET_F_DENTRY(file)->d_inode;
1953 struct vbsf_inode_info *sf_i = VBSF_GET_INODE_INFO(inode);
1954 struct vbsf_super_info *pSuperInfo = VBSF_GET_SUPER_INFO(inode->i_sb);
1955 struct vbsf_reg_info *sf_r = file->private_data;
1956 struct address_space *mapping = inode->i_mapping;
1957 loff_t pos;
1958
1959 SFLOGFLOW(("vbsf_reg_write: inode=%p file=%p buf=%p size=%#zx off=%#llx\n", inode, file, buf, size, *off));
1960 Assert(sf_i);
1961 Assert(pSuperInfo);
1962 Assert(sf_r);
1963 AssertReturn(S_ISREG(inode->i_mode), -EINVAL);
1964
1965 pos = *off;
1966 if (file->f_flags & O_APPEND)
1967 pos = i_size_read(inode);
1968
1969 /** @todo XXX Check write permission according to inode->i_mode! */
1970
1971 if (!size) {
1972 if (file->f_flags & O_APPEND) /** @todo check if this is the consensus behavior... */
1973 *off = pos;
1974 return 0;
1975 }
1976
1977 /** @todo Implement the read-write caching mode. */
1978
1979 /*
1980 * If there are active writable mappings, coordinate with any
1981 * pending writes via those.
1982 */
1983 if ( mapping
1984 && mapping->nrpages > 0
1985 && mapping_writably_mapped(mapping)) {
1986#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32)
1987 int err = filemap_fdatawait_range(mapping, pos, pos + size - 1);
1988 if (err)
1989 return err;
1990#else
1991 /** @todo ... */
1992#endif
1993 }
1994
1995 /*
1996 * For small requests, try use an embedded buffer provided we get a heap block
1997 * that does not cross page boundraries (see host code).
1998 */
1999 if (size <= PAGE_SIZE / 4 * 3 - RT_UOFFSETOF(VBOXSFWRITEEMBEDDEDREQ, abData[0]) /* see allocator */) {
2000 uint32_t const cbReq = RT_UOFFSETOF(VBOXSFWRITEEMBEDDEDREQ, abData[0]) + size;
2001 VBOXSFWRITEEMBEDDEDREQ *pReq = (VBOXSFWRITEEMBEDDEDREQ *)VbglR0PhysHeapAlloc(cbReq);
2002 if ( pReq
2003 && (PAGE_SIZE - ((uintptr_t)pReq & PAGE_OFFSET_MASK)) >= cbReq) {
2004 ssize_t cbRet;
2005 if (copy_from_user(pReq->abData, buf, size) == 0) {
2006 int vrc = VbglR0SfHostReqWriteEmbedded(pSuperInfo->map.root, pReq, sf_r->Handle.hHost,
2007 pos, (uint32_t)size);
2008 sf_i->ModificationTimeAtOurLastWrite = sf_i->ModificationTime;
2009 if (RT_SUCCESS(vrc)) {
2010 cbRet = pReq->Parms.cb32Write.u.value32;
2011 AssertStmt(cbRet <= (ssize_t)size, cbRet = size);
2012 vbsf_reg_write_sync_page_cache(mapping, pos, (uint32_t)cbRet, pReq->abData,
2013 NULL /*papSrcPages*/, 0 /*offSrcPage0*/, 0 /*cSrcPages*/);
2014 pos += cbRet;
2015 if ((file->f_flags & O_APPEND) && (g_fSfFeatures & SHFL_FEATURE_WRITE_UPDATES_OFFSET))
2016 pos = pReq->Parms.off64Write.u.value64;
2017 *off = pos;
2018 if (pos > i_size_read(inode))
2019 i_size_write(inode, pos);
2020 } else
2021 cbRet = -EPROTO;
2022 sf_i->force_restat = 1; /* mtime (and size) may have changed */
2023 } else
2024 cbRet = -EFAULT;
2025
2026 VbglR0PhysHeapFree(pReq);
2027 SFLOGFLOW(("vbsf_reg_write: returns %zd (%#zx), *off=%RX64 [embed]\n", cbRet, cbRet, *off));
2028 return cbRet;
2029 }
2030 if (pReq)
2031 VbglR0PhysHeapFree(pReq);
2032 }
2033
2034#if 0 /* Turns out this is slightly slower than locking the pages even for 4KB reads (4.19/amd64). */
2035 /*
2036 * For medium sized requests try use a bounce buffer.
2037 */
2038 if (size <= _64K /** @todo make this configurable? */) {
2039 void *pvBounce = kmalloc(size, GFP_KERNEL);
2040 if (pvBounce) {
2041 if (copy_from_user(pvBounce, buf, size) == 0) {
2042 VBOXSFWRITEPGLSTREQ *pReq = (VBOXSFWRITEPGLSTREQ *)VbglR0PhysHeapAlloc(sizeof(*pReq));
2043 if (pReq) {
2044 ssize_t cbRet;
2045 int vrc = VbglR0SfHostReqWriteContig(pSuperInfo->map.root, pReq, sf_r->handle, pos,
2046 (uint32_t)size, pvBounce, virt_to_phys(pvBounce));
2047 sf_i->ModificationTimeAtOurLastWrite = sf_i->ModificationTime;
2048 if (RT_SUCCESS(vrc)) {
2049 cbRet = pReq->Parms.cb32Write.u.value32;
2050 AssertStmt(cbRet <= (ssize_t)size, cbRet = size);
2051 vbsf_reg_write_sync_page_cache(mapping, pos, (uint32_t)cbRet, (uint8_t const *)pvBounce,
2052 NULL /*papSrcPages*/, 0 /*offSrcPage0*/, 0 /*cSrcPages*/);
2053 pos += cbRet;
2054 *off = pos;
2055 if (pos > i_size_read(inode))
2056 i_size_write(inode, pos);
2057 } else
2058 cbRet = -EPROTO;
2059 sf_i->force_restat = 1; /* mtime (and size) may have changed */
2060 VbglR0PhysHeapFree(pReq);
2061 kfree(pvBounce);
2062 SFLOGFLOW(("vbsf_reg_write: returns %zd (%#zx), *off=%RX64 [bounce]\n", cbRet, cbRet, *off));
2063 return cbRet;
2064 }
2065 kfree(pvBounce);
2066 } else {
2067 kfree(pvBounce);
2068 SFLOGFLOW(("vbsf_reg_write: returns -EFAULT, *off=%RX64 [bounce]\n", *off));
2069 return -EFAULT;
2070 }
2071 }
2072 }
2073#endif
2074
2075 return vbsf_reg_write_locking(file, buf, size, off, pos, inode, sf_i, pSuperInfo, sf_r);
2076}
2077
2078#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
2079
2080/**
2081 * Companion to vbsf_iter_lock_pages().
2082 */
2083DECLINLINE(void) vbsf_iter_unlock_pages(struct iov_iter *iter, struct page **papPages, size_t cPages, bool fSetDirty)
2084{
2085 /* We don't mark kernel pages dirty (KVECs, BVECs, PIPEs): */
2086 if (!iter_is_iovec(iter))
2087 fSetDirty = false;
2088
2089 while (cPages-- > 0)
2090 {
2091 struct page *pPage = papPages[cPages];
2092 if (fSetDirty && !PageReserved(pPage))
2093 set_page_dirty(pPage);
2094 vbsf_put_page(pPage);
2095 }
2096}
2097
2098
2099/**
2100 * Locks up to @a cMaxPages from the I/O vector iterator, advancing the
2101 * iterator.
2102 *
2103 * @returns 0 on success, negative errno value on failure.
2104 * @param iter The iterator to lock pages from.
2105 * @param fWrite Whether to write (true) or read (false) lock the pages.
2106 * @param pStash Where we stash peek results.
2107 * @param cMaxPages The maximum number of pages to get.
2108 * @param papPages Where to return the locked pages.
2109 * @param pcPages Where to return the number of pages.
2110 * @param poffPage0 Where to return the offset into the first page.
2111 * @param pcbChunk Where to return the number of bytes covered.
2112 */
2113static int vbsf_iter_lock_pages(struct iov_iter *iter, bool fWrite, struct vbsf_iter_stash *pStash, size_t cMaxPages,
2114 struct page **papPages, size_t *pcPages, size_t *poffPage0, size_t *pcbChunk)
2115{
2116 size_t cbChunk = 0;
2117 size_t cPages = 0;
2118 size_t offPage0 = 0;
2119 int rc = 0;
2120
2121 Assert(iov_iter_count(iter) + pStash->cb > 0);
2122 if (!(iter->type & ITER_KVEC)) {
2123 /*
2124 * Do we have a stashed page?
2125 */
2126 if (pStash->pPage) {
2127 papPages[0] = pStash->pPage;
2128 offPage0 = pStash->off;
2129 cbChunk = pStash->cb;
2130 cPages = 1;
2131 pStash->pPage = NULL;
2132 pStash->off = 0;
2133 pStash->cb = 0;
2134 if ( offPage0 + cbChunk < PAGE_SIZE
2135 || iov_iter_count(iter) == 0) {
2136 *poffPage0 = offPage0;
2137 *pcbChunk = cbChunk;
2138 *pcPages = cPages;
2139 SFLOGFLOW(("vbsf_iter_lock_pages: returns %d - cPages=%#zx offPage0=%#zx cbChunk=%zx (stashed)\n",
2140 rc, cPages, offPage0, cbChunk));
2141 return 0;
2142 }
2143 cMaxPages -= 1;
2144 SFLOG3(("vbsf_iter_lock_pages: Picked up stashed page: %#zx LB %#zx\n", offPage0, cbChunk));
2145 } else {
2146# if LINUX_VERSION_CODE < KERNEL_VERSION(4, 11, 0)
2147 /*
2148 * Copy out our starting point to assist rewinding.
2149 */
2150 pStash->offFromEnd = iov_iter_count(iter);
2151 pStash->Copy = *iter;
2152# endif
2153 }
2154
2155 /*
2156 * Get pages segment by segment.
2157 */
2158 do {
2159 /*
2160 * Make a special case of the first time thru here, since that's
2161 * the most typical scenario.
2162 */
2163 ssize_t cbSegRet;
2164 if (cPages == 0) {
2165# if LINUX_VERSION_CODE < KERNEL_VERSION(3, 19, 0)
2166 while (!iov_iter_single_seg_count(iter)) /* Old code didn't skip empty segments which caused EFAULTs. */
2167 iov_iter_advance(iter, 0);
2168# endif
2169 cbSegRet = iov_iter_get_pages(iter, papPages, iov_iter_count(iter), cMaxPages, &offPage0);
2170 if (cbSegRet > 0) {
2171 iov_iter_advance(iter, cbSegRet);
2172 cbChunk = (size_t)cbSegRet;
2173 cPages = RT_ALIGN_Z(offPage0 + cbSegRet, PAGE_SIZE) >> PAGE_SHIFT;
2174 cMaxPages -= cPages;
2175 SFLOG3(("vbsf_iter_lock_pages: iov_iter_get_pages -> %#zx @ %#zx; %#zx pages [first]\n", cbSegRet, offPage0, cPages));
2176 if ( cMaxPages == 0
2177 || ((offPage0 + (size_t)cbSegRet) & PAGE_OFFSET_MASK))
2178 break;
2179 } else {
2180 AssertStmt(cbSegRet < 0, cbSegRet = -EFAULT);
2181 rc = (int)cbSegRet;
2182 break;
2183 }
2184 } else {
2185 /*
2186 * Probe first page of new segment to check that we've got a zero offset and
2187 * can continue on the current chunk. Stash the page if the offset isn't zero.
2188 */
2189 size_t offPgProbe;
2190 size_t cbSeg = iov_iter_single_seg_count(iter);
2191 while (!cbSeg) {
2192 iov_iter_advance(iter, 0);
2193 cbSeg = iov_iter_single_seg_count(iter);
2194 }
2195 cbSegRet = iov_iter_get_pages(iter, &papPages[cPages], iov_iter_count(iter), 1, &offPgProbe);
2196 if (cbSegRet > 0) {
2197 iov_iter_advance(iter, cbSegRet); /** @todo maybe not do this if we stash the page? */
2198 Assert(offPgProbe + cbSegRet <= PAGE_SIZE);
2199 if (offPgProbe == 0) {
2200 cbChunk += cbSegRet;
2201 cPages += 1;
2202 cMaxPages -= 1;
2203 SFLOG3(("vbsf_iter_lock_pages: iov_iter_get_pages(1) -> %#zx @ %#zx\n", cbSegRet, offPgProbe));
2204 if ( cMaxPages == 0
2205 || cbSegRet != PAGE_SIZE)
2206 break;
2207
2208 /*
2209 * Get the rest of the segment (if anything remaining).
2210 */
2211 cbSeg -= cbSegRet;
2212 if (cbSeg > 0) {
2213 cbSegRet = iov_iter_get_pages(iter, &papPages[cPages], iov_iter_count(iter), cMaxPages, &offPgProbe);
2214 if (cbSegRet > 0) {
2215 size_t const cPgRet = RT_ALIGN_Z((size_t)cbSegRet, PAGE_SIZE) >> PAGE_SHIFT;
2216 Assert(offPgProbe == 0);
2217 iov_iter_advance(iter, cbSegRet);
2218 SFLOG3(("vbsf_iter_lock_pages: iov_iter_get_pages() -> %#zx; %#zx pages\n", cbSegRet, cPgRet));
2219 cPages += cPgRet;
2220 cMaxPages -= cPgRet;
2221 cbChunk += cbSegRet;
2222 if ( cMaxPages == 0
2223 || ((size_t)cbSegRet & PAGE_OFFSET_MASK))
2224 break;
2225 } else {
2226 AssertStmt(cbSegRet < 0, cbSegRet = -EFAULT);
2227 rc = (int)cbSegRet;
2228 break;
2229 }
2230 }
2231 } else {
2232 /* The segment didn't start at a page boundrary, so stash it for
2233 the next round: */
2234 SFLOGFLOW(("vbsf_iter_lock_pages: iov_iter_get_pages(1) -> %#zx @ %#zx; stashed\n", cbSegRet, offPgProbe));
2235 Assert(papPages[cPages]);
2236 pStash->pPage = papPages[cPages];
2237 pStash->off = offPgProbe;
2238 pStash->cb = cbSegRet;
2239 break;
2240 }
2241 } else {
2242 AssertStmt(cbSegRet < 0, cbSegRet = -EFAULT);
2243 rc = (int)cbSegRet;
2244 break;
2245 }
2246 }
2247 Assert(cMaxPages > 0);
2248 } while (iov_iter_count(iter) > 0);
2249
2250 } else {
2251 /*
2252 * The silly iov_iter_get_pages_alloc() function doesn't handle KVECs,
2253 * so everyone needs to do that by themselves.
2254 *
2255 * Note! Fixes here may apply to rtR0MemObjNativeLockKernel()
2256 * and vbsf_lock_user_pages_failed_check_kernel() as well.
2257 */
2258# if LINUX_VERSION_CODE < KERNEL_VERSION(4, 11, 0)
2259 pStash->offFromEnd = iov_iter_count(iter);
2260 pStash->Copy = *iter;
2261# endif
2262 do {
2263 uint8_t *pbBuf;
2264 size_t offStart;
2265 size_t cPgSeg;
2266
2267 size_t cbSeg = iov_iter_single_seg_count(iter);
2268 while (!cbSeg) {
2269 iov_iter_advance(iter, 0);
2270 cbSeg = iov_iter_single_seg_count(iter);
2271 }
2272
2273# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 19, 0)
2274 pbBuf = iter->kvec->iov_base + iter->iov_offset;
2275# else
2276 pbBuf = iter->iov->iov_base + iter->iov_offset;
2277# endif
2278 offStart = (uintptr_t)pbBuf & PAGE_OFFSET_MASK;
2279 if (!cPages)
2280 offPage0 = offStart;
2281 else if (offStart)
2282 break;
2283
2284 cPgSeg = RT_ALIGN_Z(cbSeg, PAGE_SIZE) >> PAGE_SHIFT;
2285 if (cPgSeg > cMaxPages) {
2286 cPgSeg = cMaxPages;
2287 cbSeg = (cPgSeg << PAGE_SHIFT) - offStart;
2288 }
2289
2290 rc = vbsf_lock_kernel_pages(pbBuf, fWrite, cPgSeg, &papPages[cPages]);
2291 if (rc == 0) {
2292 iov_iter_advance(iter, cbSeg);
2293 cbChunk += cbSeg;
2294 cPages += cPgSeg;
2295 cMaxPages -= cPgSeg;
2296 if ( cMaxPages == 0
2297 || ((offStart + cbSeg) & PAGE_OFFSET_MASK) != 0)
2298 break;
2299 } else
2300 break;
2301 } while (iov_iter_count(iter) > 0);
2302 }
2303
2304 /*
2305 * Clean up if we failed; set return values.
2306 */
2307 if (rc == 0) {
2308 /* likely */
2309 } else {
2310 if (cPages > 0)
2311 vbsf_iter_unlock_pages(iter, papPages, cPages, false /*fSetDirty*/);
2312 offPage0 = cbChunk = cPages = 0;
2313 }
2314 *poffPage0 = offPage0;
2315 *pcbChunk = cbChunk;
2316 *pcPages = cPages;
2317 SFLOGFLOW(("vbsf_iter_lock_pages: returns %d - cPages=%#zx offPage0=%#zx cbChunk=%zx\n", rc, cPages, offPage0, cbChunk));
2318 return rc;
2319}
2320
2321
2322/**
2323 * Rewinds the I/O vector.
2324 */
2325static bool vbsf_iter_rewind(struct iov_iter *iter, struct vbsf_iter_stash *pStash, size_t cbToRewind, size_t cbChunk)
2326{
2327 size_t cbExtra;
2328 if (!pStash->pPage) {
2329 cbExtra = 0;
2330 } else {
2331 cbExtra = pStash->cb;
2332 vbsf_put_page(pStash->pPage);
2333 pStash->pPage = NULL;
2334 pStash->cb = 0;
2335 pStash->off = 0;
2336 }
2337
2338# if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 11, 0) || LINUX_VERSION_CODE < KERNEL_VERSION(3, 16, 0)
2339 iov_iter_revert(iter, cbToRewind + cbExtra);
2340 return true;
2341# else
2342 /** @todo impl this */
2343 return false;
2344# endif
2345}
2346
2347
2348/**
2349 * Cleans up the page locking stash.
2350 */
2351DECLINLINE(void) vbsf_iter_cleanup_stash(struct iov_iter *iter, struct vbsf_iter_stash *pStash)
2352{
2353 if (pStash->pPage)
2354 vbsf_iter_rewind(iter, pStash, 0, 0);
2355}
2356
2357
2358/**
2359 * Calculates the longest span of pages we could transfer to the host in a
2360 * single request.
2361 *
2362 * @returns Page count, non-zero.
2363 * @param iter The I/O vector iterator to inspect.
2364 */
2365static size_t vbsf_iter_max_span_of_pages(struct iov_iter *iter)
2366{
2367 size_t cPages;
2368# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 16, 0)
2369 if (iter_is_iovec(iter) || (iter->type & ITER_KVEC)) {
2370#endif
2371 const struct iovec *pCurIov = iter->iov;
2372 size_t cLeft = iter->nr_segs;
2373 size_t cPagesSpan = 0;
2374
2375 /* iovect and kvec are identical, except for the __user tagging of iov_base. */
2376 AssertCompileMembersSameSizeAndOffset(struct iovec, iov_base, struct kvec, iov_base);
2377 AssertCompileMembersSameSizeAndOffset(struct iovec, iov_len, struct kvec, iov_len);
2378 AssertCompile(sizeof(struct iovec) == sizeof(struct kvec));
2379
2380 cPages = 1;
2381 AssertReturn(cLeft > 0, cPages);
2382
2383 /* Special case: segment offset. */
2384 if (iter->iov_offset > 0) {
2385 if (iter->iov_offset < pCurIov->iov_len) {
2386 size_t const cbSegLeft = pCurIov->iov_len - iter->iov_offset;
2387 size_t const offPage0 = ((uintptr_t)pCurIov->iov_base + iter->iov_offset) & PAGE_OFFSET_MASK;
2388 cPages = cPagesSpan = RT_ALIGN_Z(offPage0 + cbSegLeft, PAGE_SIZE) >> PAGE_SHIFT;
2389 if ((offPage0 + cbSegLeft) & PAGE_OFFSET_MASK)
2390 cPagesSpan = 0;
2391 }
2392 SFLOGFLOW(("vbsf_iter: seg[0]= %p LB %#zx\n", pCurIov->iov_base, pCurIov->iov_len));
2393 pCurIov++;
2394 cLeft--;
2395 }
2396
2397 /* Full segments. */
2398 while (cLeft-- > 0) {
2399 if (pCurIov->iov_len > 0) {
2400 size_t const offPage0 = (uintptr_t)pCurIov->iov_base & PAGE_OFFSET_MASK;
2401 if (offPage0 == 0) {
2402 if (!(pCurIov->iov_len & PAGE_OFFSET_MASK)) {
2403 cPagesSpan += pCurIov->iov_len >> PAGE_SHIFT;
2404 } else {
2405 cPagesSpan += RT_ALIGN_Z(pCurIov->iov_len, PAGE_SIZE) >> PAGE_SHIFT;
2406 if (cPagesSpan > cPages)
2407 cPages = cPagesSpan;
2408 cPagesSpan = 0;
2409 }
2410 } else {
2411 if (cPagesSpan > cPages)
2412 cPages = cPagesSpan;
2413 if (!((offPage0 + pCurIov->iov_len) & PAGE_OFFSET_MASK)) {
2414 cPagesSpan = pCurIov->iov_len >> PAGE_SHIFT;
2415 } else {
2416 cPagesSpan += RT_ALIGN_Z(offPage0 + pCurIov->iov_len, PAGE_SIZE) >> PAGE_SHIFT;
2417 if (cPagesSpan > cPages)
2418 cPages = cPagesSpan;
2419 cPagesSpan = 0;
2420 }
2421 }
2422 }
2423 SFLOGFLOW(("vbsf_iter: seg[%u]= %p LB %#zx\n", iter->nr_segs - cLeft, pCurIov->iov_base, pCurIov->iov_len));
2424 pCurIov++;
2425 }
2426 if (cPagesSpan > cPages)
2427 cPages = cPagesSpan;
2428# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 16, 0)
2429 } else {
2430 /* Won't bother with accurate counts for the next two types, just make
2431 some rough estimates (does pipes have segments?): */
2432 size_t cSegs = iter->type & ITER_BVEC ? RT_MAX(1, iter->nr_segs) : 1;
2433 cPages = (iov_iter_count(iter) + (PAGE_SIZE * 2 - 2) * cSegs) >> PAGE_SHIFT;
2434 }
2435# endif
2436 SFLOGFLOW(("vbsf_iter_max_span_of_pages: returns %#zx\n", cPages));
2437 return cPages;
2438}
2439
2440
2441/**
2442 * Worker for vbsf_reg_read_iter() that deals with larger reads using page
2443 * locking.
2444 */
2445static ssize_t vbsf_reg_read_iter_locking(struct kiocb *kio, struct iov_iter *iter, size_t cbToRead,
2446 struct vbsf_super_info *pSuperInfo, struct vbsf_reg_info *sf_r)
2447{
2448 /*
2449 * Estimate how many pages we may possible submit in a single request so
2450 * that we can allocate matching request buffer and page array.
2451 */
2452 struct page *apPagesStack[16];
2453 struct page **papPages = &apPagesStack[0];
2454 struct page **papPagesFree = NULL;
2455 VBOXSFREADPGLSTREQ *pReq;
2456 ssize_t cbRet = 0;
2457 size_t cMaxPages = vbsf_iter_max_span_of_pages(iter);
2458 cMaxPages = RT_MIN(RT_MAX(pSuperInfo->cMaxIoPages, 2), cMaxPages);
2459
2460 pReq = (VBOXSFREADPGLSTREQ *)VbglR0PhysHeapAlloc(RT_UOFFSETOF_DYN(VBOXSFREADPGLSTREQ, PgLst.aPages[cMaxPages]));
2461 while (!pReq && cMaxPages > 4) {
2462 cMaxPages /= 2;
2463 pReq = (VBOXSFREADPGLSTREQ *)VbglR0PhysHeapAlloc(RT_UOFFSETOF_DYN(VBOXSFREADPGLSTREQ, PgLst.aPages[cMaxPages]));
2464 }
2465 if (pReq && cMaxPages > RT_ELEMENTS(apPagesStack))
2466 papPagesFree = papPages = kmalloc(cMaxPages * sizeof(sizeof(papPages[0])), GFP_KERNEL);
2467 if (pReq && papPages) {
2468
2469 /*
2470 * The read loop.
2471 */
2472 struct vbsf_iter_stash Stash = VBSF_ITER_STASH_INITIALIZER;
2473 do {
2474 /*
2475 * Grab as many pages as we can. This means that if adjacent
2476 * segments both starts and ends at a page boundrary, we can
2477 * do them both in the same transfer from the host.
2478 */
2479 size_t cPages = 0;
2480 size_t cbChunk = 0;
2481 size_t offPage0 = 0;
2482 int rc = vbsf_iter_lock_pages(iter, true /*fWrite*/, &Stash, cMaxPages, papPages, &cPages, &offPage0, &cbChunk);
2483 if (rc == 0) {
2484 size_t iPage = cPages;
2485 while (iPage-- > 0)
2486 pReq->PgLst.aPages[iPage] = page_to_phys(papPages[iPage]);
2487 pReq->PgLst.offFirstPage = (uint16_t)offPage0;
2488 AssertStmt(cbChunk <= cbToRead, cbChunk = cbToRead);
2489 } else {
2490 cbRet = rc;
2491 break;
2492 }
2493
2494 /*
2495 * Issue the request and unlock the pages.
2496 */
2497 rc = VbglR0SfHostReqReadPgLst(pSuperInfo->map.root, pReq, sf_r->Handle.hHost, kio->ki_pos, cbChunk, cPages);
2498 SFLOGFLOW(("vbsf_reg_read_iter_locking: VbglR0SfHostReqReadPgLst -> %d (cbActual=%#x cbChunk=%#zx of %#zx cPages=%#zx offPage0=%#x\n",
2499 rc, pReq->Parms.cb32Read.u.value32, cbChunk, cbToRead, cPages, offPage0));
2500
2501 vbsf_iter_unlock_pages(iter, papPages, cPages, true /*fSetDirty*/);
2502
2503 if (RT_SUCCESS(rc)) {
2504 /*
2505 * Success, advance position and buffer.
2506 */
2507 uint32_t cbActual = pReq->Parms.cb32Read.u.value32;
2508 AssertStmt(cbActual <= cbChunk, cbActual = cbChunk);
2509 cbRet += cbActual;
2510 kio->ki_pos += cbActual;
2511 cbToRead -= cbActual;
2512
2513 /*
2514 * Are we done already?
2515 */
2516 if (!cbToRead)
2517 break;
2518 if (cbActual < cbChunk) { /* We ASSUME end-of-file here. */
2519 if (vbsf_iter_rewind(iter, &Stash, cbChunk - cbActual, cbActual))
2520 iov_iter_truncate(iter, 0);
2521 break;
2522 }
2523 } else {
2524 /*
2525 * Try rewind the iter structure.
2526 */
2527 bool const fRewindOkay = vbsf_iter_rewind(iter, &Stash, cbChunk, cbChunk);
2528 if (rc == VERR_NO_MEMORY && cMaxPages > 4 && fRewindOkay) {
2529 /*
2530 * The host probably doesn't have enough heap to handle the
2531 * request, reduce the page count and retry.
2532 */
2533 cMaxPages /= 4;
2534 Assert(cMaxPages > 0);
2535 } else {
2536 /*
2537 * If we've successfully read stuff, return it rather than
2538 * the error. (Not sure if this is such a great idea...)
2539 */
2540 if (cbRet <= 0)
2541 cbRet = -EPROTO;
2542 break;
2543 }
2544 }
2545 } while (cbToRead > 0);
2546
2547 vbsf_iter_cleanup_stash(iter, &Stash);
2548 }
2549 else
2550 cbRet = -ENOMEM;
2551 if (papPagesFree)
2552 kfree(papPages);
2553 if (pReq)
2554 VbglR0PhysHeapFree(pReq);
2555 SFLOGFLOW(("vbsf_reg_read_iter_locking: returns %#zx (%zd)\n", cbRet, cbRet));
2556 return cbRet;
2557}
2558
2559
2560/**
2561 * Read into I/O vector iterator.
2562 *
2563 * @returns Number of bytes read on success, negative errno on error.
2564 * @param kio The kernel I/O control block (or something like that).
2565 * @param iter The I/O vector iterator describing the buffer.
2566 */
2567# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 16, 0)
2568static ssize_t vbsf_reg_read_iter(struct kiocb *kio, struct iov_iter *iter)
2569# else
2570static ssize_t vbsf_reg_aio_read(struct kiocb *kio, const struct iovec *iov, unsigned long cSegs, loff_t offFile)
2571# endif
2572{
2573# if LINUX_VERSION_CODE < KERNEL_VERSION(3, 16, 0)
2574 struct vbsf_iov_iter fake_iter = VBSF_IOV_ITER_INITIALIZER(cSegs, iov, 0 /*write*/);
2575 struct vbsf_iov_iter *iter = &fake_iter;
2576# endif
2577 size_t cbToRead = iov_iter_count(iter);
2578 struct inode *inode = VBSF_GET_F_DENTRY(kio->ki_filp)->d_inode;
2579 struct address_space *mapping = inode->i_mapping;
2580
2581 struct vbsf_reg_info *sf_r = kio->ki_filp->private_data;
2582 struct vbsf_super_info *pSuperInfo = VBSF_GET_SUPER_INFO(inode->i_sb);
2583
2584 SFLOGFLOW(("vbsf_reg_read_iter: inode=%p file=%p size=%#zx off=%#llx type=%#x\n",
2585 inode, kio->ki_filp, cbToRead, kio->ki_pos, iter->type));
2586 AssertReturn(S_ISREG(inode->i_mode), -EINVAL);
2587
2588 /*
2589 * Do we have anything at all to do here?
2590 */
2591 if (!cbToRead)
2592 return 0;
2593
2594 /*
2595 * If there is a mapping and O_DIRECT isn't in effect, we must at a
2596 * heed dirty pages in the mapping and read from them. For simplicity
2597 * though, we just do page cache reading when there are writable
2598 * mappings around with any kind of pages loaded.
2599 */
2600 if (vbsf_should_use_cached_read(kio->ki_filp, mapping, pSuperInfo)) {
2601# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 16, 0)
2602 return generic_file_read_iter(kio, iter);
2603# else
2604 return generic_file_aio_read(kio, iov, cSegs, offFile);
2605# endif
2606 }
2607
2608 /*
2609 * Now now we reject async I/O requests.
2610 */
2611 if (!is_sync_kiocb(kio)) {
2612 SFLOGFLOW(("vbsf_reg_read_iter: async I/O not yet supported\n")); /** @todo extend FsPerf with AIO tests. */
2613 return -EOPNOTSUPP;
2614 }
2615
2616 /*
2617 * For small requests, try use an embedded buffer provided we get a heap block
2618 * that does not cross page boundraries (see host code).
2619 */
2620 if (cbToRead <= PAGE_SIZE / 4 * 3 - RT_UOFFSETOF(VBOXSFREADEMBEDDEDREQ, abData[0]) /* see allocator */) {
2621 uint32_t const cbReq = RT_UOFFSETOF(VBOXSFREADEMBEDDEDREQ, abData[0]) + cbToRead;
2622 VBOXSFREADEMBEDDEDREQ *pReq = (VBOXSFREADEMBEDDEDREQ *)VbglR0PhysHeapAlloc(cbReq);
2623 if (pReq) {
2624 if ((PAGE_SIZE - ((uintptr_t)pReq & PAGE_OFFSET_MASK)) >= cbReq) {
2625 ssize_t cbRet;
2626 int vrc = VbglR0SfHostReqReadEmbedded(pSuperInfo->map.root, pReq, sf_r->Handle.hHost,
2627 kio->ki_pos, (uint32_t)cbToRead);
2628 if (RT_SUCCESS(vrc)) {
2629 cbRet = pReq->Parms.cb32Read.u.value32;
2630 AssertStmt(cbRet <= (ssize_t)cbToRead, cbRet = cbToRead);
2631 if (copy_to_iter(pReq->abData, cbRet, iter) == cbRet) {
2632 kio->ki_pos += cbRet;
2633 if (cbRet < cbToRead)
2634 iov_iter_truncate(iter, 0);
2635 } else
2636 cbRet = -EFAULT;
2637 } else
2638 cbRet = -EPROTO;
2639 VbglR0PhysHeapFree(pReq);
2640 SFLOGFLOW(("vbsf_reg_read_iter: returns %#zx (%zd)\n", cbRet, cbRet));
2641 return cbRet;
2642 }
2643 VbglR0PhysHeapFree(pReq);
2644 }
2645 }
2646
2647 /*
2648 * Otherwise do the page locking thing.
2649 */
2650 return vbsf_reg_read_iter_locking(kio, iter, cbToRead, pSuperInfo, sf_r);
2651}
2652
2653
2654/**
2655 * Worker for vbsf_reg_write_iter() that deals with larger writes using page
2656 * locking.
2657 */
2658static ssize_t vbsf_reg_write_iter_locking(struct kiocb *kio, struct iov_iter *iter, size_t cbToWrite, loff_t offFile,
2659 struct vbsf_super_info *pSuperInfo, struct vbsf_reg_info *sf_r, struct inode *inode,
2660 struct vbsf_inode_info *sf_i, struct address_space *mapping, bool fAppend)
2661{
2662 /*
2663 * Estimate how many pages we may possible submit in a single request so
2664 * that we can allocate matching request buffer and page array.
2665 */
2666 struct page *apPagesStack[16];
2667 struct page **papPages = &apPagesStack[0];
2668 struct page **papPagesFree = NULL;
2669 VBOXSFWRITEPGLSTREQ *pReq;
2670 ssize_t cbRet = 0;
2671 size_t cMaxPages = vbsf_iter_max_span_of_pages(iter);
2672 cMaxPages = RT_MIN(RT_MAX(pSuperInfo->cMaxIoPages, 2), cMaxPages);
2673
2674 pReq = (VBOXSFWRITEPGLSTREQ *)VbglR0PhysHeapAlloc(RT_UOFFSETOF_DYN(VBOXSFWRITEPGLSTREQ, PgLst.aPages[cMaxPages]));
2675 while (!pReq && cMaxPages > 4) {
2676 cMaxPages /= 2;
2677 pReq = (VBOXSFWRITEPGLSTREQ *)VbglR0PhysHeapAlloc(RT_UOFFSETOF_DYN(VBOXSFWRITEPGLSTREQ, PgLst.aPages[cMaxPages]));
2678 }
2679 if (pReq && cMaxPages > RT_ELEMENTS(apPagesStack))
2680 papPagesFree = papPages = kmalloc(cMaxPages * sizeof(sizeof(papPages[0])), GFP_KERNEL);
2681 if (pReq && papPages) {
2682
2683 /*
2684 * The write loop.
2685 */
2686 struct vbsf_iter_stash Stash = VBSF_ITER_STASH_INITIALIZER;
2687 do {
2688 /*
2689 * Grab as many pages as we can. This means that if adjacent
2690 * segments both starts and ends at a page boundrary, we can
2691 * do them both in the same transfer from the host.
2692 */
2693 size_t cPages = 0;
2694 size_t cbChunk = 0;
2695 size_t offPage0 = 0;
2696 int rc = vbsf_iter_lock_pages(iter, false /*fWrite*/, &Stash, cMaxPages, papPages, &cPages, &offPage0, &cbChunk);
2697 if (rc == 0) {
2698 size_t iPage = cPages;
2699 while (iPage-- > 0)
2700 pReq->PgLst.aPages[iPage] = page_to_phys(papPages[iPage]);
2701 pReq->PgLst.offFirstPage = (uint16_t)offPage0;
2702 AssertStmt(cbChunk <= cbToWrite, cbChunk = cbToWrite);
2703 } else {
2704 cbRet = rc;
2705 break;
2706 }
2707
2708 /*
2709 * Issue the request and unlock the pages.
2710 */
2711 rc = VbglR0SfHostReqWritePgLst(pSuperInfo->map.root, pReq, sf_r->Handle.hHost, offFile, cbChunk, cPages);
2712 sf_i->ModificationTimeAtOurLastWrite = sf_i->ModificationTime;
2713 SFLOGFLOW(("vbsf_reg_write_iter_locking: VbglR0SfHostReqWritePgLst -> %d (cbActual=%#x cbChunk=%#zx of %#zx cPages=%#zx offPage0=%#x\n",
2714 rc, pReq->Parms.cb32Write.u.value32, cbChunk, cbToWrite, cPages, offPage0));
2715 if (RT_SUCCESS(rc)) {
2716 /*
2717 * Success, advance position and buffer.
2718 */
2719 uint32_t cbActual = pReq->Parms.cb32Write.u.value32;
2720 AssertStmt(cbActual <= cbChunk, cbActual = cbChunk);
2721
2722 vbsf_reg_write_sync_page_cache(mapping, offFile, cbActual, NULL /*pbSrcBuf*/, papPages, offPage0, cPages);
2723 vbsf_iter_unlock_pages(iter, papPages, cPages, false /*fSetDirty*/);
2724
2725 cbRet += cbActual;
2726 cbToWrite -= cbActual;
2727
2728 offFile += cbActual;
2729 if (fAppend && (g_fSfFeatures & SHFL_FEATURE_WRITE_UPDATES_OFFSET))
2730 offFile = pReq->Parms.off64Write.u.value64;
2731 kio->ki_pos = offFile;
2732 if (offFile > i_size_read(inode))
2733 i_size_write(inode, offFile);
2734
2735 sf_i->force_restat = 1; /* mtime (and size) may have changed */
2736
2737 /*
2738 * Are we done already?
2739 */
2740 if (!cbToWrite)
2741 break;
2742 if (cbActual < cbChunk) { /* We ASSUME end-of-file here. */
2743 if (vbsf_iter_rewind(iter, &Stash, cbChunk - cbActual, cbActual))
2744 iov_iter_truncate(iter, 0);
2745 break;
2746 }
2747 } else {
2748 /*
2749 * Try rewind the iter structure.
2750 */
2751 bool fRewindOkay;
2752 vbsf_iter_unlock_pages(iter, papPages, cPages, false /*fSetDirty*/);
2753 fRewindOkay = vbsf_iter_rewind(iter, &Stash, cbChunk, cbChunk);
2754 if (rc == VERR_NO_MEMORY && cMaxPages > 4 && fRewindOkay) {
2755 /*
2756 * The host probably doesn't have enough heap to handle the
2757 * request, reduce the page count and retry.
2758 */
2759 cMaxPages /= 4;
2760 Assert(cMaxPages > 0);
2761 } else {
2762 /*
2763 * If we've successfully written stuff, return it rather than
2764 * the error. (Not sure if this is such a great idea...)
2765 */
2766 if (cbRet <= 0)
2767 cbRet = -EPROTO;
2768 break;
2769 }
2770 }
2771 } while (cbToWrite > 0);
2772
2773 vbsf_iter_cleanup_stash(iter, &Stash);
2774 }
2775 else
2776 cbRet = -ENOMEM;
2777 if (papPagesFree)
2778 kfree(papPages);
2779 if (pReq)
2780 VbglR0PhysHeapFree(pReq);
2781 SFLOGFLOW(("vbsf_reg_write_iter_locking: returns %#zx (%zd)\n", cbRet, cbRet));
2782 return cbRet;
2783}
2784
2785
2786/**
2787 * Write from I/O vector iterator.
2788 *
2789 * @returns Number of bytes written on success, negative errno on error.
2790 * @param kio The kernel I/O control block (or something like that).
2791 * @param iter The I/O vector iterator describing the buffer.
2792 */
2793# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 16, 0)
2794static ssize_t vbsf_reg_write_iter(struct kiocb *kio, struct iov_iter *iter)
2795# else
2796static ssize_t vbsf_reg_aio_write(struct kiocb *kio, const struct iovec *iov, unsigned long cSegs, loff_t offFile)
2797# endif
2798{
2799# if LINUX_VERSION_CODE < KERNEL_VERSION(3, 16, 0)
2800 struct vbsf_iov_iter fake_iter = VBSF_IOV_ITER_INITIALIZER(cSegs, iov, 1 /*write*/);
2801 struct vbsf_iov_iter *iter = &fake_iter;
2802# endif
2803 size_t cbToWrite = iov_iter_count(iter);
2804 struct inode *inode = VBSF_GET_F_DENTRY(kio->ki_filp)->d_inode;
2805 struct vbsf_inode_info *sf_i = VBSF_GET_INODE_INFO(inode);
2806 struct address_space *mapping = inode->i_mapping;
2807
2808 struct vbsf_reg_info *sf_r = kio->ki_filp->private_data;
2809 struct vbsf_super_info *pSuperInfo = VBSF_GET_SUPER_INFO(inode->i_sb);
2810# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 16, 0)
2811 loff_t offFile = kio->ki_pos;
2812# endif
2813# if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 1, 0)
2814 bool const fAppend = RT_BOOL(kio->ki_flags & IOCB_APPEND);
2815# else
2816 bool const fAppend = RT_BOOL(kio->ki_filp->f_flags & O_APPEND);
2817# endif
2818
2819
2820 SFLOGFLOW(("vbsf_reg_write_iter: inode=%p file=%p size=%#zx off=%#llx type=%#x\n",
2821 inode, kio->ki_filp, cbToWrite, offFile, iter->type));
2822 AssertReturn(S_ISREG(inode->i_mode), -EINVAL);
2823
2824 /*
2825 * Enforce APPEND flag (more later).
2826 */
2827 if (fAppend)
2828 kio->ki_pos = offFile = i_size_read(inode);
2829
2830 /*
2831 * Do we have anything at all to do here?
2832 */
2833 if (!cbToWrite)
2834 return 0;
2835
2836 /** @todo Implement the read-write caching mode. */
2837
2838 /*
2839 * Now now we reject async I/O requests.
2840 */
2841 if (!is_sync_kiocb(kio)) {
2842 SFLOGFLOW(("vbsf_reg_write_iter: async I/O not yet supported\n")); /** @todo extend FsPerf with AIO tests. */
2843 return -EOPNOTSUPP;
2844 }
2845
2846 /*
2847 * If there are active writable mappings, coordinate with any
2848 * pending writes via those.
2849 */
2850 if ( mapping
2851 && mapping->nrpages > 0
2852 && mapping_writably_mapped(mapping)) {
2853# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32)
2854 int err = filemap_fdatawait_range(mapping, offFile, offFile + cbToWrite - 1);
2855 if (err)
2856 return err;
2857# else
2858 /** @todo ... */
2859# endif
2860 }
2861
2862 /*
2863 * For small requests, try use an embedded buffer provided we get a heap block
2864 * that does not cross page boundraries (see host code).
2865 */
2866 if (cbToWrite <= PAGE_SIZE / 4 * 3 - RT_UOFFSETOF(VBOXSFWRITEEMBEDDEDREQ, abData[0]) /* see allocator */) {
2867 uint32_t const cbReq = RT_UOFFSETOF(VBOXSFWRITEEMBEDDEDREQ, abData[0]) + cbToWrite;
2868 VBOXSFWRITEEMBEDDEDREQ *pReq = (VBOXSFWRITEEMBEDDEDREQ *)VbglR0PhysHeapAlloc(cbReq);
2869 if (pReq) {
2870 if ((PAGE_SIZE - ((uintptr_t)pReq & PAGE_OFFSET_MASK)) >= cbReq) {
2871 ssize_t cbRet;
2872 if (copy_from_iter(pReq->abData, cbToWrite, iter) == cbToWrite) {
2873 int vrc = VbglR0SfHostReqWriteEmbedded(pSuperInfo->map.root, pReq, sf_r->Handle.hHost,
2874 offFile, (uint32_t)cbToWrite);
2875 sf_i->ModificationTimeAtOurLastWrite = sf_i->ModificationTime;
2876 if (RT_SUCCESS(vrc)) {
2877 cbRet = pReq->Parms.cb32Write.u.value32;
2878 AssertStmt(cbRet <= (ssize_t)cbToWrite, cbRet = cbToWrite);
2879 vbsf_reg_write_sync_page_cache(mapping, offFile, (uint32_t)cbRet, pReq->abData,
2880 NULL /*papSrcPages*/, 0 /*offSrcPage0*/, 0 /*cSrcPages*/);
2881
2882 offFile += cbRet;
2883 if (fAppend && (g_fSfFeatures & SHFL_FEATURE_WRITE_UPDATES_OFFSET))
2884 offFile = pReq->Parms.off64Write.u.value64;
2885 kio->ki_pos = offFile;
2886 if (offFile > i_size_read(inode))
2887 i_size_write(inode, offFile);
2888
2889# if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 11, 0)
2890 if ((size_t)cbRet < cbToWrite)
2891 iov_iter_revert(iter, cbToWrite - cbRet);
2892# endif
2893 } else
2894 cbRet = -EPROTO;
2895 sf_i->force_restat = 1; /* mtime (and size) may have changed */
2896 } else
2897 cbRet = -EFAULT;
2898 VbglR0PhysHeapFree(pReq);
2899 SFLOGFLOW(("vbsf_reg_write_iter: returns %#zx (%zd)\n", cbRet, cbRet));
2900 return cbRet;
2901 }
2902 VbglR0PhysHeapFree(pReq);
2903 }
2904 }
2905
2906 /*
2907 * Otherwise do the page locking thing.
2908 */
2909 return vbsf_reg_write_iter_locking(kio, iter, cbToWrite, offFile, pSuperInfo, sf_r, inode, sf_i, mapping, fAppend);
2910}
2911
2912#endif /* >= 2.6.19 */
2913
2914/**
2915 * Used by vbsf_reg_open() and vbsf_inode_atomic_open() to
2916 *
2917 * @returns shared folders create flags.
2918 * @param fLnxOpen The linux O_XXX flags to convert.
2919 * @param pfHandle Pointer to vbsf_handle::fFlags.
2920 * @param pszCaller Caller, for logging purposes.
2921 */
2922uint32_t vbsf_linux_oflags_to_vbox(unsigned fLnxOpen, uint32_t *pfHandle, const char *pszCaller)
2923{
2924 uint32_t fVBoxFlags = SHFL_CF_ACCESS_DENYNONE;
2925
2926 /*
2927 * Disposition.
2928 */
2929 if (fLnxOpen & O_CREAT) {
2930 Log(("%s: O_CREAT set\n", pszCaller));
2931 fVBoxFlags |= SHFL_CF_ACT_CREATE_IF_NEW;
2932 if (fLnxOpen & O_EXCL) {
2933 Log(("%s: O_EXCL set\n", pszCaller));
2934 fVBoxFlags |= SHFL_CF_ACT_FAIL_IF_EXISTS;
2935 } else if (fLnxOpen & O_TRUNC) {
2936 Log(("%s: O_TRUNC set\n", pszCaller));
2937 fVBoxFlags |= SHFL_CF_ACT_OVERWRITE_IF_EXISTS;
2938 } else
2939 fVBoxFlags |= SHFL_CF_ACT_OPEN_IF_EXISTS;
2940 } else {
2941 fVBoxFlags |= SHFL_CF_ACT_FAIL_IF_NEW;
2942 if (fLnxOpen & O_TRUNC) {
2943 Log(("%s: O_TRUNC set\n", pszCaller));
2944 fVBoxFlags |= SHFL_CF_ACT_OVERWRITE_IF_EXISTS;
2945 }
2946 }
2947
2948 /*
2949 * Access.
2950 */
2951 switch (fLnxOpen & O_ACCMODE) {
2952 case O_RDONLY:
2953 fVBoxFlags |= SHFL_CF_ACCESS_READ;
2954 *pfHandle |= VBSF_HANDLE_F_READ;
2955 break;
2956
2957 case O_WRONLY:
2958 fVBoxFlags |= SHFL_CF_ACCESS_WRITE;
2959 *pfHandle |= VBSF_HANDLE_F_WRITE;
2960 break;
2961
2962 case O_RDWR:
2963 fVBoxFlags |= SHFL_CF_ACCESS_READWRITE;
2964 *pfHandle |= VBSF_HANDLE_F_READ | VBSF_HANDLE_F_WRITE;
2965 break;
2966
2967 default:
2968 BUG();
2969 }
2970
2971 if (fLnxOpen & O_APPEND) {
2972 Log(("%s: O_APPEND set\n", pszCaller));
2973 fVBoxFlags |= SHFL_CF_ACCESS_APPEND;
2974 *pfHandle |= VBSF_HANDLE_F_APPEND;
2975 }
2976
2977 /*
2978 * Only directories?
2979 */
2980 if (fLnxOpen & O_DIRECTORY) {
2981 Log(("%s: O_DIRECTORY set\n", pszCaller));
2982 fVBoxFlags |= SHFL_CF_DIRECTORY;
2983 }
2984
2985 return fVBoxFlags;
2986}
2987
2988
2989/**
2990 * Open a regular file.
2991 *
2992 * @param inode the inode
2993 * @param file the file
2994 * @returns 0 on success, Linux error code otherwise
2995 */
2996static int vbsf_reg_open(struct inode *inode, struct file *file)
2997{
2998 int rc, rc_linux = 0;
2999 struct vbsf_super_info *pSuperInfo = VBSF_GET_SUPER_INFO(inode->i_sb);
3000 struct vbsf_inode_info *sf_i = VBSF_GET_INODE_INFO(inode);
3001 struct dentry *dentry = VBSF_GET_F_DENTRY(file);
3002 struct vbsf_reg_info *sf_r;
3003 VBOXSFCREATEREQ *pReq;
3004
3005 SFLOGFLOW(("vbsf_reg_open: inode=%p file=%p flags=%#x %s\n", inode, file, file->f_flags, sf_i ? sf_i->path->String.ach : NULL));
3006 Assert(pSuperInfo);
3007 Assert(sf_i);
3008
3009 sf_r = kmalloc(sizeof(*sf_r), GFP_KERNEL);
3010 if (!sf_r) {
3011 LogRelFunc(("could not allocate reg info\n"));
3012 return -ENOMEM;
3013 }
3014
3015 RTListInit(&sf_r->Handle.Entry);
3016 sf_r->Handle.cRefs = 1;
3017 sf_r->Handle.fFlags = VBSF_HANDLE_F_FILE | VBSF_HANDLE_F_MAGIC;
3018 sf_r->Handle.hHost = SHFL_HANDLE_NIL;
3019
3020 /* Already open? */
3021 if (sf_i->handle != SHFL_HANDLE_NIL) {
3022 /*
3023 * This inode was created with vbsf_create_worker(). Check the CreateFlags:
3024 * O_CREAT, O_TRUNC: inherent true (file was just created). Not sure
3025 * about the access flags (SHFL_CF_ACCESS_*).
3026 */
3027 sf_i->force_restat = 1;
3028 sf_r->Handle.hHost = sf_i->handle;
3029 sf_i->handle = SHFL_HANDLE_NIL;
3030 file->private_data = sf_r;
3031
3032 sf_r->Handle.fFlags |= VBSF_HANDLE_F_READ | VBSF_HANDLE_F_WRITE; /** @todo fix */
3033 vbsf_handle_append(sf_i, &sf_r->Handle);
3034 SFLOGFLOW(("vbsf_reg_open: returns 0 (#1) - sf_i=%p hHost=%#llx\n", sf_i, sf_r->Handle.hHost));
3035 return 0;
3036 }
3037
3038 pReq = (VBOXSFCREATEREQ *)VbglR0PhysHeapAlloc(sizeof(*pReq) + sf_i->path->u16Size);
3039 if (!pReq) {
3040 kfree(sf_r);
3041 LogRelFunc(("Failed to allocate a VBOXSFCREATEREQ buffer!\n"));
3042 return -ENOMEM;
3043 }
3044 memcpy(&pReq->StrPath, sf_i->path, SHFLSTRING_HEADER_SIZE + sf_i->path->u16Size);
3045 RT_ZERO(pReq->CreateParms);
3046 pReq->CreateParms.Handle = SHFL_HANDLE_NIL;
3047
3048 /* We check the value of pReq->CreateParms.Handle afterwards to
3049 * find out if the call succeeded or failed, as the API does not seem
3050 * to cleanly distinguish error and informational messages.
3051 *
3052 * Furthermore, we must set pReq->CreateParms.Handle to SHFL_HANDLE_NIL
3053 * to make the shared folders host service use our fMode parameter */
3054
3055 /* We ignore O_EXCL, as the Linux kernel seems to call create
3056 beforehand itself, so O_EXCL should always fail. */
3057 pReq->CreateParms.CreateFlags = vbsf_linux_oflags_to_vbox(file->f_flags & ~O_EXCL, &sf_r->Handle.fFlags, __FUNCTION__);
3058 pReq->CreateParms.Info.Attr.fMode = inode->i_mode;
3059 LogFunc(("vbsf_reg_open: calling VbglR0SfHostReqCreate, file %s, flags=%#x, %#x\n",
3060 sf_i->path->String.utf8, file->f_flags, pReq->CreateParms.CreateFlags));
3061 rc = VbglR0SfHostReqCreate(pSuperInfo->map.root, pReq);
3062 if (RT_FAILURE(rc)) {
3063 LogFunc(("VbglR0SfHostReqCreate failed flags=%d,%#x rc=%Rrc\n", file->f_flags, pReq->CreateParms.CreateFlags, rc));
3064 kfree(sf_r);
3065 VbglR0PhysHeapFree(pReq);
3066 return -RTErrConvertToErrno(rc);
3067 }
3068
3069 if (pReq->CreateParms.Handle != SHFL_HANDLE_NIL) {
3070 vbsf_dentry_chain_increase_ttl(dentry);
3071 vbsf_update_inode(inode, sf_i, &pReq->CreateParms.Info, pSuperInfo, false /*fInodeLocked*/, 0 /*fSetAttrs*/);
3072 rc_linux = 0;
3073 } else {
3074 switch (pReq->CreateParms.Result) {
3075 case SHFL_PATH_NOT_FOUND:
3076 vbsf_dentry_invalidate_ttl(dentry);
3077 rc_linux = -ENOENT;
3078 break;
3079 case SHFL_FILE_NOT_FOUND:
3080 vbsf_dentry_invalidate_ttl(dentry);
3081 /** @todo sf_dentry_increase_parent_ttl(file->f_dentry); if we can trust it. */
3082 rc_linux = -ENOENT;
3083 break;
3084 case SHFL_FILE_EXISTS:
3085 vbsf_dentry_chain_increase_ttl(dentry);
3086 vbsf_update_inode(inode, sf_i, &pReq->CreateParms.Info, pSuperInfo, false /*fInodeLocked*/, 0 /*fSetAttrs*/);
3087 rc_linux = -EEXIST;
3088 break;
3089 default:
3090 vbsf_dentry_chain_increase_parent_ttl(dentry);
3091 rc_linux = 0;
3092 break;
3093 }
3094 }
3095
3096 sf_r->Handle.hHost = pReq->CreateParms.Handle;
3097 file->private_data = sf_r;
3098 vbsf_handle_append(sf_i, &sf_r->Handle);
3099 VbglR0PhysHeapFree(pReq);
3100 SFLOGFLOW(("vbsf_reg_open: returns 0 (#2) - sf_i=%p hHost=%#llx\n", sf_i, sf_r->Handle.hHost));
3101 return rc_linux;
3102}
3103
3104
3105/**
3106 * Close a regular file.
3107 *
3108 * @param inode the inode
3109 * @param file the file
3110 * @returns 0 on success, Linux error code otherwise
3111 */
3112static int vbsf_reg_release(struct inode *inode, struct file *file)
3113{
3114 struct vbsf_inode_info *sf_i = VBSF_GET_INODE_INFO(inode);
3115 struct vbsf_reg_info *sf_r = file->private_data;
3116
3117 SFLOGFLOW(("vbsf_reg_release: inode=%p file=%p\n", inode, file));
3118 if (sf_r) {
3119 struct vbsf_super_info *pSuperInfo = VBSF_GET_SUPER_INFO(inode->i_sb);
3120 struct address_space *mapping = inode->i_mapping;
3121 Assert(pSuperInfo);
3122
3123 /* If we're closing the last handle for this inode, make sure the flush
3124 the mapping or we'll end up in vbsf_writepage without a handle. */
3125 if ( mapping
3126 && mapping->nrpages > 0
3127 /** @todo && last writable handle */ ) {
3128#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 25)
3129 if (filemap_fdatawrite(mapping) != -EIO)
3130#else
3131 if ( filemap_fdatasync(mapping) == 0
3132 && fsync_inode_data_buffers(inode) == 0)
3133#endif
3134 filemap_fdatawait(inode->i_mapping);
3135 }
3136
3137 /* Release sf_r, closing the handle if we're the last user. */
3138 file->private_data = NULL;
3139 vbsf_handle_release(&sf_r->Handle, pSuperInfo, "vbsf_reg_release");
3140
3141 sf_i->handle = SHFL_HANDLE_NIL;
3142 }
3143 return 0;
3144}
3145
3146
3147/**
3148 * Wrapper around generic/default seek function that ensures that we've got
3149 * the up-to-date file size when doing anything relative to EOF.
3150 *
3151 * The issue is that the host may extend the file while we weren't looking and
3152 * if the caller wishes to append data, it may end up overwriting existing data
3153 * if we operate with a stale size. So, we always retrieve the file size on EOF
3154 * relative seeks.
3155 */
3156static loff_t vbsf_reg_llseek(struct file *file, loff_t off, int whence)
3157{
3158 SFLOGFLOW(("vbsf_reg_llseek: file=%p off=%lld whence=%d\n", file, off, whence));
3159
3160 switch (whence) {
3161#ifdef SEEK_HOLE
3162 case SEEK_HOLE:
3163 case SEEK_DATA:
3164#endif
3165 case SEEK_END: {
3166 struct vbsf_reg_info *sf_r = file->private_data;
3167 int rc = vbsf_inode_revalidate_with_handle(VBSF_GET_F_DENTRY(file), sf_r->Handle.hHost,
3168 true /*fForce*/, false /*fInodeLocked*/);
3169 if (rc == 0)
3170 break;
3171 return rc;
3172 }
3173 }
3174
3175#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 8)
3176 return generic_file_llseek(file, off, whence);
3177#else
3178 return default_llseek(file, off, whence);
3179#endif
3180}
3181
3182
3183/**
3184 * Flush region of file - chiefly mmap/msync.
3185 *
3186 * We cannot use the noop_fsync / simple_sync_file here as that means
3187 * msync(,,MS_SYNC) will return before the data hits the host, thereby
3188 * causing coherency issues with O_DIRECT access to the same file as
3189 * well as any host interaction with the file.
3190 */
3191#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 1, 0)
3192static int vbsf_reg_fsync(struct file *file, loff_t start, loff_t end, int datasync)
3193{
3194# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 16, 0)
3195 return __generic_file_fsync(file, start, end, datasync);
3196# else
3197 return generic_file_fsync(file, start, end, datasync);
3198# endif
3199}
3200#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 35)
3201static int vbsf_reg_fsync(struct file *file, int datasync)
3202{
3203 return generic_file_fsync(file, datasync);
3204}
3205#else /* < 2.6.35 */
3206static int vbsf_reg_fsync(struct file *file, struct dentry *dentry, int datasync)
3207{
3208# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 31)
3209 return simple_fsync(file, dentry, datasync);
3210# else
3211 int rc;
3212 struct inode *inode = dentry->d_inode;
3213 AssertReturn(inode, -EINVAL);
3214
3215 /** @todo What about file_fsync()? (<= 2.5.11) */
3216
3217# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 12)
3218 rc = sync_mapping_buffers(inode->i_mapping);
3219 if ( rc == 0
3220 && (inode->i_state & I_DIRTY)
3221 && ((inode->i_state & I_DIRTY_DATASYNC) || !datasync)
3222 ) {
3223 struct writeback_control wbc = {
3224 .sync_mode = WB_SYNC_ALL,
3225 .nr_to_write = 0
3226 };
3227 rc = sync_inode(inode, &wbc);
3228 }
3229# else /* < 2.5.12 */
3230 /** @todo
3231 * Somethings is buggy here or in the 2.4.21-27.EL kernel I'm testing on.
3232 *
3233 * In theory we shouldn't need to do anything here, since msync will call
3234 * writepage() on each dirty page and we write them out synchronously. So, the
3235 * problem is elsewhere... Doesn't happen all the time either. Sigh.
3236 */
3237 rc = fsync_inode_buffers(inode);
3238# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 10)
3239 if (rc == 0 && datasync)
3240 rc = fsync_inode_data_buffers(inode);
3241# endif
3242
3243# endif /* < 2.5.12 */
3244 return rc;
3245# endif
3246}
3247#endif /* < 2.6.35 */
3248
3249
3250#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 5, 0)
3251/**
3252 * Copy a datablock from one file to another on the host side.
3253 */
3254static ssize_t vbsf_reg_copy_file_range(struct file *pFileSrc, loff_t offSrc, struct file *pFileDst, loff_t offDst,
3255 size_t cbRange, unsigned int fFlags)
3256{
3257 ssize_t cbRet;
3258 if (g_uSfLastFunction >= SHFL_FN_COPY_FILE_PART) {
3259 struct inode *pInodeSrc = pFileSrc->f_inode;
3260 struct vbsf_inode_info *pInodeInfoSrc = VBSF_GET_INODE_INFO(pInodeSrc);
3261 struct vbsf_super_info *pSuperInfoSrc = VBSF_GET_SUPER_INFO(pInodeSrc->i_sb);
3262 struct vbsf_reg_info *pFileInfoSrc = (struct vbsf_reg_info *)pFileSrc->private_data;
3263 struct inode *pInodeDst = pInodeSrc;
3264 struct vbsf_inode_info *pInodeInfoDst = VBSF_GET_INODE_INFO(pInodeDst);
3265 struct vbsf_super_info *pSuperInfoDst = VBSF_GET_SUPER_INFO(pInodeDst->i_sb);
3266 struct vbsf_reg_info *pFileInfoDst = (struct vbsf_reg_info *)pFileDst->private_data;
3267 VBOXSFCOPYFILEPARTREQ *pReq;
3268
3269 /*
3270 * Some extra validation.
3271 */
3272 AssertPtrReturn(pInodeInfoSrc, -EOPNOTSUPP);
3273 Assert(pInodeInfoSrc->u32Magic == SF_INODE_INFO_MAGIC);
3274 AssertPtrReturn(pInodeInfoDst, -EOPNOTSUPP);
3275 Assert(pInodeInfoDst->u32Magic == SF_INODE_INFO_MAGIC);
3276
3277# if LINUX_VERSION_CODE < KERNEL_VERSION(4, 11, 0)
3278 if (!S_ISREG(pInodeSrc->i_mode) || !S_ISREG(pInodeDst->i_mode))
3279 return S_ISDIR(pInodeSrc->i_mode) || S_ISDIR(pInodeDst->i_mode) ? -EISDIR : -EINVAL;
3280# endif
3281
3282 /*
3283 * Allocate the request and issue it.
3284 */
3285 pReq = (VBOXSFCOPYFILEPARTREQ *)VbglR0PhysHeapAlloc(sizeof(*pReq));
3286 if (pReq) {
3287 int vrc = VbglR0SfHostReqCopyFilePart(pSuperInfoSrc->map.root, pFileInfoSrc->Handle.hHost, offSrc,
3288 pSuperInfoDst->map.root, pFileInfoDst->Handle.hHost, offDst,
3289 cbRange, 0 /*fFlags*/, pReq);
3290 if (RT_SUCCESS(vrc))
3291 cbRet = pReq->Parms.cb64ToCopy.u.value64;
3292 else if (vrc == VERR_NOT_IMPLEMENTED)
3293 cbRet = -EOPNOTSUPP;
3294 else
3295 cbRet = -RTErrConvertToErrno(vrc);
3296
3297 VbglR0PhysHeapFree(pReq);
3298 } else
3299 cbRet = -ENOMEM;
3300 } else {
3301 cbRet = -EOPNOTSUPP;
3302 }
3303 SFLOGFLOW(("vbsf_reg_copy_file_range: returns %zd\n", cbRet));
3304 return cbRet;
3305}
3306#endif /* > 4.5 */
3307
3308
3309#ifdef SFLOG_ENABLED
3310/*
3311 * This is just for logging page faults and such.
3312 */
3313
3314/** Pointer to the ops generic_file_mmap returns the first time it's called. */
3315static struct vm_operations_struct const *g_pGenericFileVmOps = NULL;
3316/** Merge of g_LoggingVmOpsTemplate and g_pGenericFileVmOps. */
3317static struct vm_operations_struct g_LoggingVmOps;
3318
3319
3320/* Generic page fault callback: */
3321# if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 11, 0)
3322static vm_fault_t vbsf_vmlog_fault(struct vm_fault *vmf)
3323{
3324 vm_fault_t rc;
3325 SFLOGFLOW(("vbsf_vmlog_fault: vmf=%p flags=%#x addr=%p\n", vmf, vmf->flags, vmf->address));
3326 rc = g_pGenericFileVmOps->fault(vmf);
3327 SFLOGFLOW(("vbsf_vmlog_fault: returns %d\n", rc));
3328 return rc;
3329}
3330# elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 23)
3331static int vbsf_vmlog_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
3332{
3333 int rc;
3334# if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 10, 0)
3335 SFLOGFLOW(("vbsf_vmlog_fault: vma=%p vmf=%p flags=%#x addr=%p\n", vma, vmf, vmf->flags, vmf->address));
3336# else
3337 SFLOGFLOW(("vbsf_vmlog_fault: vma=%p vmf=%p flags=%#x addr=%p\n", vma, vmf, vmf->flags, vmf->virtual_address));
3338# endif
3339 rc = g_pGenericFileVmOps->fault(vma, vmf);
3340 SFLOGFLOW(("vbsf_vmlog_fault: returns %d\n", rc));
3341 return rc;
3342}
3343# endif
3344
3345
3346/* Special/generic page fault handler: */
3347# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 26)
3348# elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 1)
3349static struct page *vbsf_vmlog_nopage(struct vm_area_struct *vma, unsigned long address, int *type)
3350{
3351 struct page *page;
3352 SFLOGFLOW(("vbsf_vmlog_nopage: vma=%p address=%p type=%p:{%#x}\n", vma, address, type, type ? *type : 0));
3353 page = g_pGenericFileVmOps->nopage(vma, address, type);
3354 SFLOGFLOW(("vbsf_vmlog_nopage: returns %p\n", page));
3355 return page;
3356}
3357# else
3358static struct page *vbsf_vmlog_nopage(struct vm_area_struct *vma, unsigned long address, int write_access_or_unused)
3359{
3360 struct page *page;
3361 SFLOGFLOW(("vbsf_vmlog_nopage: vma=%p address=%p wau=%d\n", vma, address, write_access_or_unused));
3362 page = g_pGenericFileVmOps->nopage(vma, address, write_access_or_unused);
3363 SFLOGFLOW(("vbsf_vmlog_nopage: returns %p\n", page));
3364 return page;
3365}
3366# endif /* < 2.6.26 */
3367
3368
3369/* Special page fault callback for making something writable: */
3370# if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 11, 0)
3371static vm_fault_t vbsf_vmlog_page_mkwrite(struct vm_fault *vmf)
3372{
3373 vm_fault_t rc;
3374 SFLOGFLOW(("vbsf_vmlog_page_mkwrite: vmf=%p flags=%#x addr=%p\n", vmf, vmf->flags, vmf->address));
3375 rc = g_pGenericFileVmOps->page_mkwrite(vmf);
3376 SFLOGFLOW(("vbsf_vmlog_page_mkwrite: returns %d\n", rc));
3377 return rc;
3378}
3379# elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 30)
3380static int vbsf_vmlog_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
3381{
3382 int rc;
3383# if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 10, 0)
3384 SFLOGFLOW(("vbsf_vmlog_page_mkwrite: vma=%p vmf=%p flags=%#x addr=%p\n", vma, vmf, vmf->flags, vmf->address));
3385# else
3386 SFLOGFLOW(("vbsf_vmlog_page_mkwrite: vma=%p vmf=%p flags=%#x addr=%p\n", vma, vmf, vmf->flags, vmf->virtual_address));
3387# endif
3388 rc = g_pGenericFileVmOps->page_mkwrite(vma, vmf);
3389 SFLOGFLOW(("vbsf_vmlog_page_mkwrite: returns %d\n", rc));
3390 return rc;
3391}
3392# elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
3393static int vbsf_vmlog_page_mkwrite(struct vm_area_struct *vma, struct page *page)
3394{
3395 int rc;
3396 SFLOGFLOW(("vbsf_vmlog_page_mkwrite: vma=%p page=%p\n", vma, page));
3397 rc = g_pGenericFileVmOps->page_mkwrite(vma, page);
3398 SFLOGFLOW(("vbsf_vmlog_page_mkwrite: returns %d\n", rc));
3399 return rc;
3400}
3401# endif
3402
3403
3404/* Special page fault callback for mapping pages: */
3405# if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 10, 0)
3406static void vbsf_vmlog_map_pages(struct vm_fault *vmf, pgoff_t start, pgoff_t end)
3407{
3408 SFLOGFLOW(("vbsf_vmlog_map_pages: vmf=%p (flags=%#x addr=%p) start=%p end=%p\n", vmf, vmf->flags, vmf->address, start, end));
3409 g_pGenericFileVmOps->map_pages(vmf, start, end);
3410 SFLOGFLOW(("vbsf_vmlog_map_pages: returns\n"));
3411}
3412# elif LINUX_VERSION_CODE >= KERNEL_VERSION(4, 8, 0)
3413static void vbsf_vmlog_map_pages(struct fault_env *fenv, pgoff_t start, pgoff_t end)
3414{
3415 SFLOGFLOW(("vbsf_vmlog_map_pages: fenv=%p (flags=%#x addr=%p) start=%p end=%p\n", fenv, fenv->flags, fenv->address, start, end));
3416 g_pGenericFileVmOps->map_pages(fenv, start, end);
3417 SFLOGFLOW(("vbsf_vmlog_map_pages: returns\n"));
3418}
3419# elif LINUX_VERSION_CODE >= KERNEL_VERSION(3, 15, 0)
3420static void vbsf_vmlog_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
3421{
3422 SFLOGFLOW(("vbsf_vmlog_map_pages: vma=%p vmf=%p (flags=%#x addr=%p)\n", vma, vmf, vmf->flags, vmf->virtual_address));
3423 g_pGenericFileVmOps->map_pages(vma, vmf);
3424 SFLOGFLOW(("vbsf_vmlog_map_pages: returns\n"));
3425}
3426# endif
3427
3428
3429/** Overload template. */
3430static struct vm_operations_struct const g_LoggingVmOpsTemplate = {
3431# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 23)
3432 .fault = vbsf_vmlog_fault,
3433# endif
3434# if LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 25)
3435 .nopage = vbsf_vmlog_nopage,
3436# endif
3437# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
3438 .page_mkwrite = vbsf_vmlog_page_mkwrite,
3439# endif
3440# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 15, 0)
3441 .map_pages = vbsf_vmlog_map_pages,
3442# endif
3443};
3444
3445/** file_operations::mmap wrapper for logging purposes. */
3446extern int vbsf_reg_mmap(struct file *file, struct vm_area_struct *vma)
3447{
3448 int rc;
3449 SFLOGFLOW(("vbsf_reg_mmap: file=%p vma=%p\n", file, vma));
3450 rc = generic_file_mmap(file, vma);
3451 if (rc == 0) {
3452 /* Merge the ops and template the first time thru (there's a race here). */
3453 if (g_pGenericFileVmOps == NULL) {
3454 uintptr_t const *puSrc1 = (uintptr_t *)vma->vm_ops;
3455 uintptr_t const *puSrc2 = (uintptr_t *)&g_LoggingVmOpsTemplate;
3456 uintptr_t volatile *puDst = (uintptr_t *)&g_LoggingVmOps;
3457 size_t cbLeft = sizeof(g_LoggingVmOps) / sizeof(*puDst);
3458 while (cbLeft-- > 0) {
3459 *puDst = *puSrc2 && *puSrc1 ? *puSrc2 : *puSrc1;
3460 puSrc1++;
3461 puSrc2++;
3462 puDst++;
3463 }
3464 g_pGenericFileVmOps = vma->vm_ops;
3465 vma->vm_ops = &g_LoggingVmOps;
3466 } else if (g_pGenericFileVmOps == vma->vm_ops)
3467 vma->vm_ops = &g_LoggingVmOps;
3468 else
3469 SFLOGFLOW(("vbsf_reg_mmap: Warning: vm_ops=%p, expected %p!\n", vma->vm_ops, g_pGenericFileVmOps));
3470 }
3471 SFLOGFLOW(("vbsf_reg_mmap: returns %d\n", rc));
3472 return rc;
3473}
3474
3475#endif /* SFLOG_ENABLED */
3476
3477
3478/**
3479 * File operations for regular files.
3480 *
3481 * Note on splice_read/splice_write/sendfile:
3482 * - Splice was introduced in 2.6.17. The generic_file_splice_read/write
3483 * methods go thru the page cache, which is undesirable and is why we
3484 * need to cook our own versions of the code as long as we cannot track
3485 * host-side writes and correctly invalidate the guest page-cache.
3486 * - Sendfile reimplemented using splice in 2.6.23.
3487 * - The default_file_splice_read/write no-page-cache fallback functions,
3488 * were introduced in 2.6.31. The write one work in page units.
3489 * - Since linux 3.16 there is iter_file_splice_write that uses iter_write.
3490 * - Since linux 4.9 the generic_file_splice_read function started using
3491 * read_iter.
3492 */
3493struct file_operations vbsf_reg_fops = {
3494 .open = vbsf_reg_open,
3495 .read = vbsf_reg_read,
3496 .write = vbsf_reg_write,
3497#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 16, 0)
3498 .read_iter = vbsf_reg_read_iter,
3499 .write_iter = vbsf_reg_write_iter,
3500#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
3501 .aio_read = vbsf_reg_aio_read,
3502 .aio_write = vbsf_reg_aio_write,
3503#endif
3504 .release = vbsf_reg_release,
3505#ifdef SFLOG_ENABLED
3506 .mmap = vbsf_reg_mmap,
3507#else
3508 .mmap = generic_file_mmap,
3509#endif
3510#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 17) && LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 31)
3511 .splice_read = vbsf_splice_read,
3512#endif
3513#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 16, 0)
3514 .splice_write = iter_file_splice_write,
3515#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 17)
3516 .splice_write = vbsf_splice_write,
3517#endif
3518#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 30) && LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 23)
3519 .sendfile = vbsf_reg_sendfile,
3520#endif
3521 .llseek = vbsf_reg_llseek,
3522 .fsync = vbsf_reg_fsync,
3523#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 5, 0)
3524 .copy_file_range = vbsf_reg_copy_file_range,
3525#endif
3526};
3527
3528
3529/**
3530 * Inodes operations for regular files.
3531 */
3532struct inode_operations vbsf_reg_iops = {
3533#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 18)
3534 .getattr = vbsf_inode_getattr,
3535#else
3536 .revalidate = vbsf_inode_revalidate,
3537#endif
3538 .setattr = vbsf_inode_setattr,
3539};
3540
3541
3542
3543/*********************************************************************************************************************************
3544* Address Space Operations on Regular Files (for mmap, sendfile, direct I/O) *
3545*********************************************************************************************************************************/
3546
3547/**
3548 * Used to read the content of a page into the page cache.
3549 *
3550 * Needed for mmap and reads+writes when the file is mmapped in a
3551 * shared+writeable fashion.
3552 */
3553static int vbsf_readpage(struct file *file, struct page *page)
3554{
3555 struct inode *inode = VBSF_GET_F_DENTRY(file)->d_inode;
3556 int err;
3557
3558 SFLOGFLOW(("vbsf_readpage: inode=%p file=%p page=%p off=%#llx\n", inode, file, page, (uint64_t)page->index << PAGE_SHIFT));
3559 Assert(PageLocked(page));
3560
3561 if (PageUptodate(page)) {
3562 unlock_page(page);
3563 return 0;
3564 }
3565
3566 if (!is_bad_inode(inode)) {
3567 VBOXSFREADPGLSTREQ *pReq = (VBOXSFREADPGLSTREQ *)VbglR0PhysHeapAlloc(sizeof(*pReq));
3568 if (pReq) {
3569 struct vbsf_super_info *pSuperInfo = VBSF_GET_SUPER_INFO(inode->i_sb);
3570 struct vbsf_reg_info *sf_r = file->private_data;
3571 uint32_t cbRead;
3572 int vrc;
3573
3574 pReq->PgLst.offFirstPage = 0;
3575 pReq->PgLst.aPages[0] = page_to_phys(page);
3576 vrc = VbglR0SfHostReqReadPgLst(pSuperInfo->map.root,
3577 pReq,
3578 sf_r->Handle.hHost,
3579 (uint64_t)page->index << PAGE_SHIFT,
3580 PAGE_SIZE,
3581 1 /*cPages*/);
3582
3583 cbRead = pReq->Parms.cb32Read.u.value32;
3584 AssertStmt(cbRead <= PAGE_SIZE, cbRead = PAGE_SIZE);
3585 VbglR0PhysHeapFree(pReq);
3586
3587 if (RT_SUCCESS(vrc)) {
3588 if (cbRead == PAGE_SIZE) {
3589 /* likely */
3590 } else {
3591 uint8_t *pbMapped = (uint8_t *)kmap(page);
3592 RT_BZERO(&pbMapped[cbRead], PAGE_SIZE - cbRead);
3593 kunmap(page);
3594 /** @todo truncate the inode file size? */
3595 }
3596
3597 flush_dcache_page(page);
3598 SetPageUptodate(page);
3599 unlock_page(page);
3600 return 0;
3601 }
3602 err = -RTErrConvertToErrno(vrc);
3603 } else
3604 err = -ENOMEM;
3605 } else
3606 err = -EIO;
3607 SetPageError(page);
3608 unlock_page(page);
3609 return err;
3610}
3611
3612
3613/**
3614 * Used to write out the content of a dirty page cache page to the host file.
3615 *
3616 * Needed for mmap and writes when the file is mmapped in a shared+writeable
3617 * fashion.
3618 */
3619#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 52)
3620static int vbsf_writepage(struct page *page, struct writeback_control *wbc)
3621#else
3622static int vbsf_writepage(struct page *page)
3623#endif
3624{
3625 struct address_space *mapping = page->mapping;
3626 struct inode *inode = mapping->host;
3627 struct vbsf_inode_info *sf_i = VBSF_GET_INODE_INFO(inode);
3628 struct vbsf_handle *pHandle = vbsf_handle_find(sf_i, VBSF_HANDLE_F_WRITE, VBSF_HANDLE_F_APPEND);
3629 int err;
3630
3631 SFLOGFLOW(("vbsf_writepage: inode=%p page=%p off=%#llx pHandle=%p (%#llx)\n",
3632 inode, page, (uint64_t)page->index << PAGE_SHIFT, pHandle, pHandle ? pHandle->hHost : 0));
3633
3634 if (pHandle) {
3635 struct vbsf_super_info *pSuperInfo = VBSF_GET_SUPER_INFO(inode->i_sb);
3636 VBOXSFWRITEPGLSTREQ *pReq = (VBOXSFWRITEPGLSTREQ *)VbglR0PhysHeapAlloc(sizeof(*pReq));
3637 if (pReq) {
3638 uint64_t const cbFile = i_size_read(inode);
3639 uint64_t const offInFile = (uint64_t)page->index << PAGE_SHIFT;
3640 uint32_t const cbToWrite = page->index != (cbFile >> PAGE_SHIFT) ? PAGE_SIZE
3641 : (uint32_t)cbFile & (uint32_t)PAGE_OFFSET_MASK;
3642 int vrc;
3643
3644 pReq->PgLst.offFirstPage = 0;
3645 pReq->PgLst.aPages[0] = page_to_phys(page);
3646 vrc = VbglR0SfHostReqWritePgLst(pSuperInfo->map.root,
3647 pReq,
3648 pHandle->hHost,
3649 offInFile,
3650 cbToWrite,
3651 1 /*cPages*/);
3652 sf_i->ModificationTimeAtOurLastWrite = sf_i->ModificationTime;
3653 AssertMsgStmt(pReq->Parms.cb32Write.u.value32 == cbToWrite || RT_FAILURE(vrc), /* lazy bird */
3654 ("%#x vs %#x\n", pReq->Parms.cb32Write, cbToWrite),
3655 vrc = VERR_WRITE_ERROR);
3656 VbglR0PhysHeapFree(pReq);
3657
3658 if (RT_SUCCESS(vrc)) {
3659 /* Update the inode if we've extended the file. */
3660 /** @todo is this necessary given the cbToWrite calc above? */
3661 uint64_t const offEndOfWrite = offInFile + cbToWrite;
3662 if ( offEndOfWrite > cbFile
3663 && offEndOfWrite > i_size_read(inode))
3664 i_size_write(inode, offEndOfWrite);
3665
3666 /* Update and unlock the page. */
3667 if (PageError(page))
3668 ClearPageError(page);
3669 SetPageUptodate(page);
3670 unlock_page(page);
3671
3672 vbsf_handle_release(pHandle, pSuperInfo, "vbsf_writepage");
3673 return 0;
3674 }
3675
3676 /*
3677 * We failed.
3678 */
3679 err = -EIO;
3680 } else
3681 err = -ENOMEM;
3682 vbsf_handle_release(pHandle, pSuperInfo, "vbsf_writepage");
3683 } else {
3684 /** @todo we could re-open the file here and deal with this... */
3685 static uint64_t volatile s_cCalls = 0;
3686 if (s_cCalls++ < 16)
3687 printk("vbsf_writepage: no writable handle for %s..\n", sf_i->path->String.ach);
3688 err = -EIO;
3689 }
3690 SetPageError(page);
3691 unlock_page(page);
3692 return err;
3693}
3694
3695
3696#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 24)
3697/**
3698 * Called when writing thru the page cache (which we shouldn't be doing).
3699 */
3700int vbsf_write_begin(struct file *file, struct address_space *mapping, loff_t pos,
3701 unsigned len, unsigned flags, struct page **pagep, void **fsdata)
3702{
3703 /** @todo r=bird: We shouldn't ever get here, should we? Because we don't use
3704 * the page cache for any writes AFAIK. We could just as well use
3705 * simple_write_begin & simple_write_end here if we think we really
3706 * need to have non-NULL function pointers in the table... */
3707 static uint64_t volatile s_cCalls = 0;
3708 if (s_cCalls++ < 16) {
3709 printk("vboxsf: Unexpected call to vbsf_write_begin(pos=%#llx len=%#x flags=%#x)! Please report.\n",
3710 (unsigned long long)pos, len, flags);
3711 RTLogBackdoorPrintf("vboxsf: Unexpected call to vbsf_write_begin(pos=%#llx len=%#x flags=%#x)! Please report.\n",
3712 (unsigned long long)pos, len, flags);
3713# ifdef WARN_ON
3714 WARN_ON(1);
3715# endif
3716 }
3717 return simple_write_begin(file, mapping, pos, len, flags, pagep, fsdata);
3718}
3719#endif /* KERNEL_VERSION >= 2.6.24 */
3720
3721
3722#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 10)
3723
3724# ifdef VBOX_UEK
3725# undef iov_iter /* HACK ALERT! Don't put anything needing vbsf_iov_iter after this fun! */
3726# endif
3727
3728/**
3729 * This is needed to make open accept O_DIRECT as well as dealing with direct
3730 * I/O requests if we don't intercept them earlier.
3731 */
3732# if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 7, 0)
3733static ssize_t vbsf_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3734# elif LINUX_VERSION_CODE >= KERNEL_VERSION(4, 1, 0)
3735static ssize_t vbsf_direct_IO(struct kiocb *iocb, struct iov_iter *iter, loff_t offset)
3736# elif LINUX_VERSION_CODE >= KERNEL_VERSION(3, 16, 0) || defined(VBOX_UEK)
3737static ssize_t vbsf_direct_IO(int rw, struct kiocb *iocb, struct iov_iter *iter, loff_t offset)
3738# elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 6)
3739static ssize_t vbsf_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, loff_t offset, unsigned long nr_segs)
3740# elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 55)
3741static int vbsf_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, loff_t offset, unsigned long nr_segs)
3742# elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 41)
3743static int vbsf_direct_IO(int rw, struct file *file, const struct iovec *iov, loff_t offset, unsigned long nr_segs)
3744# elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 35)
3745static int vbsf_direct_IO(int rw, struct inode *inode, const struct iovec *iov, loff_t offset, unsigned long nr_segs)
3746# elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 26)
3747static int vbsf_direct_IO(int rw, struct inode *inode, char *buf, loff_t offset, size_t count)
3748# elif LINUX_VERSION_CODE == KERNEL_VERSION(2, 4, 21) && defined(I_NEW) /* RHEL3 Frankenkernel. */
3749static int vbsf_direct_IO(int rw, struct file *file, struct kiobuf *buf, unsigned long whatever1, int whatever2)
3750# else
3751static int vbsf_direct_IO(int rw, struct inode *inode, struct kiobuf *buf, unsigned long whatever1, int whatever2)
3752# endif
3753{
3754 TRACE();
3755 return -EINVAL;
3756}
3757
3758#endif
3759
3760/**
3761 * Address space (for the page cache) operations for regular files.
3762 *
3763 * @todo the FsPerf touch/flush (mmap) test fails on 4.4.0 (ubuntu 16.04 lts).
3764 */
3765struct address_space_operations vbsf_reg_aops = {
3766 .readpage = vbsf_readpage,
3767 .writepage = vbsf_writepage,
3768 /** @todo Need .writepages if we want msync performance... */
3769#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 12)
3770 .set_page_dirty = __set_page_dirty_buffers,
3771#endif
3772#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 24)
3773 .write_begin = vbsf_write_begin,
3774 .write_end = simple_write_end,
3775#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 45)
3776 .prepare_write = simple_prepare_write,
3777 .commit_write = simple_commit_write,
3778#endif
3779#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 10)
3780 .direct_IO = vbsf_direct_IO,
3781#endif
3782};
3783
Note: See TracBrowser for help on using the repository browser.

© 2024 Oracle Support Privacy / Do Not Sell My Info Terms of Use Trademark Policy Automated Access Etiquette