VirtualBox

source: vbox/trunk/src/VBox/Additions/linux/sharedfolders/regops.c@ 88571

Last change on this file since 88571 was 88571, checked in by vboxsync, 3 years ago

Additions: Linux: vboxsf: adapt to 5.10.x+ kernels, bugref:9994.

This patch fixed the issue when executables in VBox share
could no be executed even if share was mounted with '-o exec'
option.
  • Property svn:eol-style set to native
  • Property svn:keywords set to Author Date Id Revision
File size: 146.4 KB
Line 
1/* $Id: regops.c 88571 2021-04-16 17:59:50Z 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 RTLNX_VER_MIN(2,5,32)
38# include <linux/aio.h> /* struct kiocb before 4.1 */
39#endif
40#if RTLNX_VER_MIN(2,5,12)
41# include <linux/buffer_head.h>
42#endif
43#if RTLNX_VER_RANGE(2,5,12, 2,6,31)
44# include <linux/writeback.h>
45#endif
46#if RTLNX_VER_RANGE(2,6,23, 3,16,0)
47# include <linux/splice.h>
48#endif
49#if RTLNX_VER_RANGE(2,6,17, 2,6,23)
50# include <linux/pipe_fs_i.h>
51#endif
52#if RTLNX_VER_MIN(2,4,10)
53# include <linux/swap.h> /* for mark_page_accessed */
54#endif
55#include <iprt/err.h>
56
57#if RTLNX_VER_MAX(2,6,18)
58# define SEEK_END 2
59#endif
60
61#if RTLNX_VER_MAX(3,16,0)
62# define iter_is_iovec(a_pIter) ( !((a_pIter)->type & ITER_KVEC) )
63#elif RTLNX_VER_MAX(3,19,0)
64# define iter_is_iovec(a_pIter) ( !((a_pIter)->type & (ITER_KVEC | ITER_BVEC)) )
65#endif
66
67#if RTLNX_VER_MAX(4,17,0)
68# define vm_fault_t int
69#endif
70
71#if RTLNX_VER_MAX(2,5,20)
72# define pgoff_t unsigned long
73#endif
74
75#if RTLNX_VER_MAX(2,5,12)
76# define PageUptodate(a_pPage) Page_Uptodate(a_pPage)
77#endif
78
79
80/*********************************************************************************************************************************
81* Structures and Typedefs *
82*********************************************************************************************************************************/
83#if RTLNX_VER_MAX(3,16,0)
84struct vbsf_iov_iter {
85 unsigned int type;
86 unsigned int v_write : 1;
87 size_t iov_offset;
88 size_t nr_segs;
89 struct iovec const *iov;
90# ifdef VBOX_STRICT
91 struct iovec const *iov_org;
92 size_t nr_segs_org;
93# endif
94};
95# ifdef VBOX_STRICT
96# define VBSF_IOV_ITER_INITIALIZER(a_cSegs, a_pIov, a_fWrite) \
97 { vbsf_iov_iter_detect_type(a_pIov, a_cSegs), a_fWrite, 0, a_cSegs, a_pIov, a_pIov, a_cSegs }
98# else
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 }
101# endif
102# define ITER_KVEC 1
103# define iov_iter vbsf_iov_iter
104#endif
105
106#if RTLNX_VER_MIN(2,6,19)
107/** Used by vbsf_iter_lock_pages() to keep the first page of the next segment. */
108struct vbsf_iter_stash {
109 struct page *pPage;
110 size_t off;
111 size_t cb;
112# if RTLNX_VER_MAX(4,11,0)
113 size_t offFromEnd;
114 struct iov_iter Copy;
115# endif
116};
117#endif /* >= 3.16.0 */
118/** Initializer for struct vbsf_iter_stash. */
119#if RTLNX_VER_MIN(4,11,0)
120# define VBSF_ITER_STASH_INITIALIZER { NULL, 0 }
121#else
122# define VBSF_ITER_STASH_INITIALIZER { NULL, 0, ~(size_t)0 }
123#endif
124
125
126/*********************************************************************************************************************************
127* Internal Functions *
128*********************************************************************************************************************************/
129DECLINLINE(void) vbsf_put_page(struct page *pPage);
130static void vbsf_unlock_user_pages(struct page **papPages, size_t cPages, bool fSetDirty, bool fLockPgHack);
131static void vbsf_reg_write_sync_page_cache(struct address_space *mapping, loff_t offFile, uint32_t cbRange,
132 uint8_t const *pbSrcBuf, struct page **papSrcPages,
133 uint32_t offSrcPage, size_t cSrcPages);
134
135
136/*********************************************************************************************************************************
137* Provide more recent uio.h functionality to older kernels. *
138*********************************************************************************************************************************/
139#if RTLNX_VER_RANGE(2,6,19, 3,16,0)
140
141/**
142 * Detects the vector type.
143 */
144static int vbsf_iov_iter_detect_type(struct iovec const *paIov, size_t cSegs)
145{
146 /* Check the first segment with a non-zero length. */
147 while (cSegs-- > 0) {
148 if (paIov->iov_len > 0) {
149 if (access_ok(VERIFY_READ, paIov->iov_base, paIov->iov_len))
150#if RTLNX_VER_MIN(5,10,0)
151 return (uintptr_t)paIov->iov_base >= TASK_SIZE_MAX ? ITER_KVEC : 0;
152#else
153 return (uintptr_t)paIov->iov_base >= USER_DS.seg ? ITER_KVEC : 0;
154#endif
155 AssertMsgFailed(("%p LB %#zx\n", paIov->iov_base, paIov->iov_len));
156 break;
157 }
158 paIov++;
159 }
160 return 0;
161}
162
163
164# undef iov_iter_count
165# define iov_iter_count(a_pIter) vbsf_iov_iter_count(a_pIter)
166static size_t vbsf_iov_iter_count(struct vbsf_iov_iter const *iter)
167{
168 size_t cbRet = 0;
169 size_t cLeft = iter->nr_segs;
170 struct iovec const *iov = iter->iov;
171 while (cLeft-- > 0) {
172 cbRet += iov->iov_len;
173 iov++;
174 }
175 return cbRet - iter->iov_offset;
176}
177
178
179# undef iov_iter_single_seg_count
180# define iov_iter_single_seg_count(a_pIter) vbsf_iov_iter_single_seg_count(a_pIter)
181static size_t vbsf_iov_iter_single_seg_count(struct vbsf_iov_iter const *iter)
182{
183 if (iter->nr_segs > 0)
184 return iter->iov->iov_len - iter->iov_offset;
185 return 0;
186}
187
188
189# undef iov_iter_advance
190# define iov_iter_advance(a_pIter, a_cbSkip) vbsf_iov_iter_advance(a_pIter, a_cbSkip)
191static void vbsf_iov_iter_advance(struct vbsf_iov_iter *iter, size_t cbSkip)
192{
193 SFLOG2(("vbsf_iov_iter_advance: cbSkip=%#zx\n", cbSkip));
194 if (iter->nr_segs > 0) {
195 size_t const cbLeftCur = iter->iov->iov_len - iter->iov_offset;
196 Assert(iter->iov_offset <= iter->iov->iov_len);
197 if (cbLeftCur > cbSkip) {
198 iter->iov_offset += cbSkip;
199 } else {
200 cbSkip -= cbLeftCur;
201 iter->iov_offset = 0;
202 iter->iov++;
203 iter->nr_segs--;
204 while (iter->nr_segs > 0) {
205 size_t const cbSeg = iter->iov->iov_len;
206 if (cbSeg > cbSkip) {
207 iter->iov_offset = cbSkip;
208 break;
209 }
210 cbSkip -= cbSeg;
211 iter->iov++;
212 iter->nr_segs--;
213 }
214 }
215 }
216}
217
218
219# undef iov_iter_get_pages
220# define iov_iter_get_pages(a_pIter, a_papPages, a_cbMax, a_cMaxPages, a_poffPg0) \
221 vbsf_iov_iter_get_pages(a_pIter, a_papPages, a_cbMax, a_cMaxPages, a_poffPg0)
222static ssize_t vbsf_iov_iter_get_pages(struct vbsf_iov_iter *iter, struct page **papPages,
223 size_t cbMax, unsigned cMaxPages, size_t *poffPg0)
224{
225 while (iter->nr_segs > 0) {
226 size_t const cbLeft = iter->iov->iov_len - iter->iov_offset;
227 Assert(iter->iov->iov_len >= iter->iov_offset);
228 if (cbLeft > 0) {
229 uintptr_t uPtrFrom = (uintptr_t)iter->iov->iov_base + iter->iov_offset;
230 size_t offPg0 = *poffPg0 = uPtrFrom & PAGE_OFFSET_MASK;
231 size_t cPagesLeft = RT_ALIGN_Z(offPg0 + cbLeft, PAGE_SIZE) >> PAGE_SHIFT;
232 size_t cPages = RT_MIN(cPagesLeft, cMaxPages);
233 struct task_struct *pTask = current;
234 size_t cPagesLocked;
235
236 down_read(&pTask->mm->mmap_sem);
237 cPagesLocked = get_user_pages(pTask, pTask->mm, uPtrFrom, cPages, iter->v_write, 1 /*force*/, papPages, NULL);
238 up_read(&pTask->mm->mmap_sem);
239 if (cPagesLocked == cPages) {
240 size_t cbRet = (cPages << PAGE_SHIFT) - offPg0;
241 if (cPages == cPagesLeft) {
242 size_t offLastPg = (uPtrFrom + cbLeft) & PAGE_OFFSET_MASK;
243 if (offLastPg)
244 cbRet -= PAGE_SIZE - offLastPg;
245 }
246 Assert(cbRet <= cbLeft);
247 return cbRet;
248 }
249 if (cPagesLocked > 0)
250 vbsf_unlock_user_pages(papPages, cPagesLocked, false /*fSetDirty*/, false /*fLockPgHack*/);
251 return -EFAULT;
252 }
253 iter->iov_offset = 0;
254 iter->iov++;
255 iter->nr_segs--;
256 }
257 AssertFailed();
258 return 0;
259}
260
261
262# undef iov_iter_truncate
263# define iov_iter_truncate(iter, cbNew) vbsf_iov_iter_truncate(iter, cbNew)
264static void vbsf_iov_iter_truncate(struct vbsf_iov_iter *iter, size_t cbNew)
265{
266 /* we have no counter or stuff, so it's a no-op. */
267 RT_NOREF(iter, cbNew);
268}
269
270
271# undef iov_iter_revert
272# define iov_iter_revert(a_pIter, a_cbRewind) vbsf_iov_iter_revert(a_pIter, a_cbRewind)
273void vbsf_iov_iter_revert(struct vbsf_iov_iter *iter, size_t cbRewind)
274{
275 SFLOG2(("vbsf_iov_iter_revert: cbRewind=%#zx\n", cbRewind));
276 if (iter->iov_offset > 0) {
277 if (cbRewind <= iter->iov_offset) {
278 iter->iov_offset -= cbRewind;
279 return;
280 }
281 cbRewind -= iter->iov_offset;
282 iter->iov_offset = 0;
283 }
284
285 while (cbRewind > 0) {
286 struct iovec const *pIov = --iter->iov;
287 size_t const cbSeg = pIov->iov_len;
288 iter->nr_segs++;
289
290 Assert((uintptr_t)pIov >= (uintptr_t)iter->iov_org);
291 Assert(iter->nr_segs <= iter->nr_segs_org);
292
293 if (cbRewind <= cbSeg) {
294 iter->iov_offset = cbSeg - cbRewind;
295 break;
296 }
297 cbRewind -= cbSeg;
298 }
299}
300
301#endif /* 2.6.19 <= linux < 3.16.0 */
302#if RTLNX_VER_RANGE(3,16,0, 3,16,35)
303
304/** This is for implementing cMaxPage on 3.16 which doesn't have it. */
305static ssize_t vbsf_iov_iter_get_pages_3_16(struct iov_iter *iter, struct page **papPages,
306 size_t cbMax, unsigned cMaxPages, size_t *poffPg0)
307{
308 if (!(iter->type & ITER_BVEC)) {
309 size_t const offPg0 = iter->iov_offset & PAGE_OFFSET_MASK;
310 size_t const cbMaxPages = ((size_t)cMaxPages << PAGE_SHIFT) - offPg0;
311 if (cbMax > cbMaxPages)
312 cbMax = cbMaxPages;
313 }
314 /* else: BVEC works a page at a time and shouldn't have much of a problem here. */
315 return iov_iter_get_pages(iter, papPages, cbMax, poffPg0);
316}
317# undef iov_iter_get_pages
318# define iov_iter_get_pages(a_pIter, a_papPages, a_cbMax, a_cMaxPages, a_poffPg0) \
319 vbsf_iov_iter_get_pages_3_16(a_pIter, a_papPages, a_cbMax, a_cMaxPages, a_poffPg0)
320
321#endif /* 3.16.0-3.16.34 */
322#if RTLNX_VER_RANGE(2,6,19, 3,18,0)
323
324static size_t copy_from_iter(uint8_t *pbDst, size_t cbToCopy, struct iov_iter *pSrcIter)
325{
326 size_t const cbTotal = cbToCopy;
327 Assert(iov_iter_count(pSrcIter) >= cbToCopy);
328# if RTLNX_VER_MIN(3,16,0)
329 if (pSrcIter->type & ITER_BVEC) {
330 while (cbToCopy > 0) {
331 size_t const offPage = (uintptr_t)pbDst & PAGE_OFFSET_MASK;
332 size_t const cbThisCopy = RT_MIN(PAGE_SIZE - offPage, cbToCopy);
333 struct page *pPage = rtR0MemObjLinuxVirtToPage(pbDst);
334 size_t cbCopied = copy_page_from_iter(pPage, offPage, cbThisCopy, pSrcIter);
335 AssertStmt(cbCopied <= cbThisCopy, cbCopied = cbThisCopy);
336 pbDst += cbCopied;
337 cbToCopy -= cbCopied;
338 if (cbCopied != cbToCopy)
339 break;
340 }
341 } else
342# endif
343 {
344 while (cbToCopy > 0) {
345 size_t cbThisCopy = iov_iter_single_seg_count(pSrcIter);
346 if (cbThisCopy > 0) {
347 if (cbThisCopy > cbToCopy)
348 cbThisCopy = cbToCopy;
349 if (pSrcIter->type & ITER_KVEC)
350 memcpy(pbDst, (void *)pSrcIter->iov->iov_base + pSrcIter->iov_offset, cbThisCopy);
351 else if (copy_from_user(pbDst, pSrcIter->iov->iov_base + pSrcIter->iov_offset, cbThisCopy) != 0)
352 break;
353 pbDst += cbThisCopy;
354 cbToCopy -= cbThisCopy;
355 }
356 iov_iter_advance(pSrcIter, cbThisCopy);
357 }
358 }
359 return cbTotal - cbToCopy;
360}
361
362
363static size_t copy_to_iter(uint8_t const *pbSrc, size_t cbToCopy, struct iov_iter *pDstIter)
364{
365 size_t const cbTotal = cbToCopy;
366 Assert(iov_iter_count(pDstIter) >= cbToCopy);
367# if RTLNX_VER_MIN(3,16,0)
368 if (pDstIter->type & ITER_BVEC) {
369 while (cbToCopy > 0) {
370 size_t const offPage = (uintptr_t)pbSrc & PAGE_OFFSET_MASK;
371 size_t const cbThisCopy = RT_MIN(PAGE_SIZE - offPage, cbToCopy);
372 struct page *pPage = rtR0MemObjLinuxVirtToPage((void *)pbSrc);
373 size_t cbCopied = copy_page_to_iter(pPage, offPage, cbThisCopy, pDstIter);
374 AssertStmt(cbCopied <= cbThisCopy, cbCopied = cbThisCopy);
375 pbSrc += cbCopied;
376 cbToCopy -= cbCopied;
377 if (cbCopied != cbToCopy)
378 break;
379 }
380 } else
381# endif
382 {
383 while (cbToCopy > 0) {
384 size_t cbThisCopy = iov_iter_single_seg_count(pDstIter);
385 if (cbThisCopy > 0) {
386 if (cbThisCopy > cbToCopy)
387 cbThisCopy = cbToCopy;
388 if (pDstIter->type & ITER_KVEC)
389 memcpy((void *)pDstIter->iov->iov_base + pDstIter->iov_offset, pbSrc, cbThisCopy);
390 else if (copy_to_user(pDstIter->iov->iov_base + pDstIter->iov_offset, pbSrc, cbThisCopy) != 0) {
391 break;
392 }
393 pbSrc += cbThisCopy;
394 cbToCopy -= cbThisCopy;
395 }
396 iov_iter_advance(pDstIter, cbThisCopy);
397 }
398 }
399 return cbTotal - cbToCopy;
400}
401
402#endif /* 3.16.0 <= linux < 3.18.0 */
403
404
405
406/*********************************************************************************************************************************
407* Handle management *
408*********************************************************************************************************************************/
409
410/**
411 * Called when an inode is released to unlink all handles that might impossibly
412 * still be associated with it.
413 *
414 * @param pInodeInfo The inode which handles to drop.
415 */
416void vbsf_handle_drop_chain(struct vbsf_inode_info *pInodeInfo)
417{
418 struct vbsf_handle *pCur, *pNext;
419 unsigned long fSavedFlags;
420 SFLOGFLOW(("vbsf_handle_drop_chain: %p\n", pInodeInfo));
421 spin_lock_irqsave(&g_SfHandleLock, fSavedFlags);
422
423 RTListForEachSafe(&pInodeInfo->HandleList, pCur, pNext, struct vbsf_handle, Entry) {
424 AssertMsg( (pCur->fFlags & (VBSF_HANDLE_F_MAGIC_MASK | VBSF_HANDLE_F_ON_LIST))
425 == (VBSF_HANDLE_F_MAGIC | VBSF_HANDLE_F_ON_LIST), ("%p %#x\n", pCur, pCur->fFlags));
426 pCur->fFlags |= VBSF_HANDLE_F_ON_LIST;
427 RTListNodeRemove(&pCur->Entry);
428 }
429
430 spin_unlock_irqrestore(&g_SfHandleLock, fSavedFlags);
431}
432
433
434/**
435 * Locates a handle that matches all the flags in @a fFlags.
436 *
437 * @returns Pointer to handle on success (retained), use vbsf_handle_release() to
438 * release it. NULL if no suitable handle was found.
439 * @param pInodeInfo The inode info to search.
440 * @param fFlagsSet The flags that must be set.
441 * @param fFlagsClear The flags that must be clear.
442 */
443struct vbsf_handle *vbsf_handle_find(struct vbsf_inode_info *pInodeInfo, uint32_t fFlagsSet, uint32_t fFlagsClear)
444{
445 struct vbsf_handle *pCur;
446 unsigned long fSavedFlags;
447 spin_lock_irqsave(&g_SfHandleLock, fSavedFlags);
448
449 RTListForEach(&pInodeInfo->HandleList, pCur, struct vbsf_handle, Entry) {
450 AssertMsg( (pCur->fFlags & (VBSF_HANDLE_F_MAGIC_MASK | VBSF_HANDLE_F_ON_LIST))
451 == (VBSF_HANDLE_F_MAGIC | VBSF_HANDLE_F_ON_LIST), ("%p %#x\n", pCur, pCur->fFlags));
452 if ((pCur->fFlags & (fFlagsSet | fFlagsClear)) == fFlagsSet) {
453 uint32_t cRefs = ASMAtomicIncU32(&pCur->cRefs);
454 if (cRefs > 1) {
455 spin_unlock_irqrestore(&g_SfHandleLock, fSavedFlags);
456 SFLOGFLOW(("vbsf_handle_find: returns %p\n", pCur));
457 return pCur;
458 }
459 /* Oops, already being closed (safe as it's only ever increased here). */
460 ASMAtomicDecU32(&pCur->cRefs);
461 }
462 }
463
464 spin_unlock_irqrestore(&g_SfHandleLock, fSavedFlags);
465 SFLOGFLOW(("vbsf_handle_find: returns NULL!\n"));
466 return NULL;
467}
468
469
470/**
471 * Slow worker for vbsf_handle_release() that does the freeing.
472 *
473 * @returns 0 (ref count).
474 * @param pHandle The handle to release.
475 * @param pSuperInfo The info structure for the shared folder associated with
476 * the handle.
477 * @param pszCaller The caller name (for logging failures).
478 */
479uint32_t vbsf_handle_release_slow(struct vbsf_handle *pHandle, struct vbsf_super_info *pSuperInfo, const char *pszCaller)
480{
481 int rc;
482 unsigned long fSavedFlags;
483
484 SFLOGFLOW(("vbsf_handle_release_slow: %p (%s)\n", pHandle, pszCaller));
485
486 /*
487 * Remove from the list.
488 */
489 spin_lock_irqsave(&g_SfHandleLock, fSavedFlags);
490
491 AssertMsg((pHandle->fFlags & VBSF_HANDLE_F_MAGIC_MASK) == VBSF_HANDLE_F_MAGIC, ("%p %#x\n", pHandle, pHandle->fFlags));
492 Assert(pHandle->pInodeInfo);
493 Assert(pHandle->pInodeInfo && pHandle->pInodeInfo->u32Magic == SF_INODE_INFO_MAGIC);
494
495 if (pHandle->fFlags & VBSF_HANDLE_F_ON_LIST) {
496 pHandle->fFlags &= ~VBSF_HANDLE_F_ON_LIST;
497 RTListNodeRemove(&pHandle->Entry);
498 }
499
500 spin_unlock_irqrestore(&g_SfHandleLock, fSavedFlags);
501
502 /*
503 * Actually destroy it.
504 */
505 rc = VbglR0SfHostReqCloseSimple(pSuperInfo->map.root, pHandle->hHost);
506 if (RT_FAILURE(rc))
507 LogFunc(("Caller %s: VbglR0SfHostReqCloseSimple %#RX64 failed with rc=%Rrc\n", pszCaller, pHandle->hHost, rc));
508 pHandle->hHost = SHFL_HANDLE_NIL;
509 pHandle->fFlags = VBSF_HANDLE_F_MAGIC_DEAD;
510 kfree(pHandle);
511 return 0;
512}
513
514
515/**
516 * Appends a handle to a handle list.
517 *
518 * @param pInodeInfo The inode to add it to.
519 * @param pHandle The handle to add.
520 */
521void vbsf_handle_append(struct vbsf_inode_info *pInodeInfo, struct vbsf_handle *pHandle)
522{
523#ifdef VBOX_STRICT
524 struct vbsf_handle *pCur;
525#endif
526 unsigned long fSavedFlags;
527
528 SFLOGFLOW(("vbsf_handle_append: %p (to %p)\n", pHandle, pInodeInfo));
529 AssertMsg((pHandle->fFlags & (VBSF_HANDLE_F_MAGIC_MASK | VBSF_HANDLE_F_ON_LIST)) == VBSF_HANDLE_F_MAGIC,
530 ("%p %#x\n", pHandle, pHandle->fFlags));
531 Assert(pInodeInfo->u32Magic == SF_INODE_INFO_MAGIC);
532
533 spin_lock_irqsave(&g_SfHandleLock, fSavedFlags);
534
535 AssertMsg((pHandle->fFlags & (VBSF_HANDLE_F_MAGIC_MASK | VBSF_HANDLE_F_ON_LIST)) == VBSF_HANDLE_F_MAGIC,
536 ("%p %#x\n", pHandle, pHandle->fFlags));
537#ifdef VBOX_STRICT
538 RTListForEach(&pInodeInfo->HandleList, pCur, struct vbsf_handle, Entry) {
539 Assert(pCur != pHandle);
540 AssertMsg( (pCur->fFlags & (VBSF_HANDLE_F_MAGIC_MASK | VBSF_HANDLE_F_ON_LIST))
541 == (VBSF_HANDLE_F_MAGIC | VBSF_HANDLE_F_ON_LIST), ("%p %#x\n", pCur, pCur->fFlags));
542 }
543 pHandle->pInodeInfo = pInodeInfo;
544#endif
545
546 pHandle->fFlags |= VBSF_HANDLE_F_ON_LIST;
547 RTListAppend(&pInodeInfo->HandleList, &pHandle->Entry);
548
549 spin_unlock_irqrestore(&g_SfHandleLock, fSavedFlags);
550}
551
552
553
554/*********************************************************************************************************************************
555* Misc *
556*********************************************************************************************************************************/
557
558#if RTLNX_VER_MAX(2,6,6)
559/** Any writable mappings? */
560DECLINLINE(bool) mapping_writably_mapped(struct address_space const *mapping)
561{
562# if RTLNX_VER_MIN(2,5,6)
563 return !list_empty(&mapping->i_mmap_shared);
564# else
565 return mapping->i_mmap_shared != NULL;
566# endif
567}
568#endif
569
570
571#if RTLNX_VER_MAX(2,5,12)
572/** Missing in 2.4.x, so just stub it for now. */
573DECLINLINE(bool) PageWriteback(struct page const *page)
574{
575 return false;
576}
577#endif
578
579
580/**
581 * Helper for deciding wheter we should do a read via the page cache or not.
582 *
583 * By default we will only use the page cache if there is a writable memory
584 * mapping of the file with a chance that it may have modified any of the pages
585 * already.
586 */
587DECLINLINE(bool) vbsf_should_use_cached_read(struct file *file, struct address_space *mapping, struct vbsf_super_info *pSuperInfo)
588{
589 if ( (file->f_flags & O_DIRECT)
590 || pSuperInfo->enmCacheMode == kVbsfCacheMode_None)
591 return false;
592 if ( pSuperInfo->enmCacheMode == kVbsfCacheMode_Read
593 || pSuperInfo->enmCacheMode == kVbsfCacheMode_ReadWrite)
594 return true;
595 Assert(pSuperInfo->enmCacheMode == kVbsfCacheMode_Strict);
596 return mapping
597 && mapping->nrpages > 0
598 && mapping_writably_mapped(mapping);
599}
600
601
602
603/*********************************************************************************************************************************
604* Pipe / splice stuff mainly for 2.6.17 >= linux < 2.6.31 (where no fallbacks were available) *
605*********************************************************************************************************************************/
606
607#if RTLNX_VER_RANGE(2,6,17, 3,16,0)
608
609# if RTLNX_VER_MAX(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 RTLNX_VER_RANGE(2,6,17, 2,6,31)
650
651/** Verify pipe buffer content (needed for page-cache to ensure idle page). */
652static int vbsf_pipe_buf_confirm(struct pipe_inode_info *pPipe, struct pipe_buffer *pPipeBuf)
653{
654 /*SFLOG3(("vbsf_pipe_buf_confirm: %p\n", pPipeBuf));*/
655 return 0;
656}
657
658
659/** Maps the buffer page. */
660static void *vbsf_pipe_buf_map(struct pipe_inode_info *pPipe, struct pipe_buffer *pPipeBuf, int atomic)
661{
662 void *pvRet;
663 if (!atomic)
664 pvRet = kmap(pPipeBuf->page);
665 else {
666 pPipeBuf->flags |= PIPE_BUF_FLAG_ATOMIC;
667 pvRet = kmap_atomic(pPipeBuf->page, KM_USER0);
668 }
669 /*SFLOG3(("vbsf_pipe_buf_map: %p -> %p\n", pPipeBuf, pvRet));*/
670 return pvRet;
671}
672
673
674/** Unmaps the buffer page. */
675static void vbsf_pipe_buf_unmap(struct pipe_inode_info *pPipe, struct pipe_buffer *pPipeBuf, void *pvMapping)
676{
677 /*SFLOG3(("vbsf_pipe_buf_unmap: %p/%p\n", pPipeBuf, pvMapping)); */
678 if (!(pPipeBuf->flags & PIPE_BUF_FLAG_ATOMIC))
679 kunmap(pPipeBuf->page);
680 else {
681 pPipeBuf->flags &= ~PIPE_BUF_FLAG_ATOMIC;
682 kunmap_atomic(pvMapping, KM_USER0);
683 }
684}
685
686
687/** Gets a reference to the page. */
688static void vbsf_pipe_buf_get(struct pipe_inode_info *pPipe, struct pipe_buffer *pPipeBuf)
689{
690 page_cache_get(pPipeBuf->page);
691 /*SFLOG3(("vbsf_pipe_buf_get: %p (return count=%d)\n", pPipeBuf, page_count(pPipeBuf->page)));*/
692}
693
694
695/** Release the buffer page (counter to vbsf_pipe_buf_get). */
696static void vbsf_pipe_buf_release(struct pipe_inode_info *pPipe, struct pipe_buffer *pPipeBuf)
697{
698 /*SFLOG3(("vbsf_pipe_buf_release: %p (incoming count=%d)\n", pPipeBuf, page_count(pPipeBuf->page)));*/
699 page_cache_release(pPipeBuf->page);
700}
701
702
703/** Attempt to steal the page.
704 * @returns 0 success, 1 on failure. */
705static int vbsf_pipe_buf_steal(struct pipe_inode_info *pPipe, struct pipe_buffer *pPipeBuf)
706{
707 if (page_count(pPipeBuf->page) == 1) {
708 lock_page(pPipeBuf->page);
709 SFLOG3(("vbsf_pipe_buf_steal: %p -> 0\n", pPipeBuf));
710 return 0;
711 }
712 SFLOG3(("vbsf_pipe_buf_steal: %p -> 1\n", pPipeBuf));
713 return 1;
714}
715
716
717/**
718 * Pipe buffer operations for used by vbsf_feed_pages_to_pipe.
719 */
720static struct pipe_buf_operations vbsf_pipe_buf_ops = {
721 .can_merge = 0,
722# if RTLNX_VER_MIN(2,6,23)
723 .confirm = vbsf_pipe_buf_confirm,
724# else
725 .pin = vbsf_pipe_buf_confirm,
726# endif
727 .map = vbsf_pipe_buf_map,
728 .unmap = vbsf_pipe_buf_unmap,
729 .get = vbsf_pipe_buf_get,
730 .release = vbsf_pipe_buf_release,
731 .steal = vbsf_pipe_buf_steal,
732};
733
734
735/**
736 * Feeds the pages to the pipe.
737 *
738 * Pages given to the pipe are set to NULL in papPages.
739 */
740static ssize_t vbsf_feed_pages_to_pipe(struct pipe_inode_info *pPipe, struct page **papPages, size_t cPages, uint32_t offPg0,
741 uint32_t cbActual, unsigned fFlags)
742{
743 ssize_t cbRet = 0;
744 size_t iPage = 0;
745 bool fNeedWakeUp = false;
746
747 LOCK_PIPE(pPipe);
748 for (;;) {
749 if ( pPipe->readers > 0
750 && pPipe->nrbufs < PIPE_BUFFERS) {
751 struct pipe_buffer *pPipeBuf = &pPipe->bufs[(pPipe->curbuf + pPipe->nrbufs) % PIPE_BUFFERS];
752 uint32_t const cbThisPage = RT_MIN(cbActual, PAGE_SIZE - offPg0);
753 pPipeBuf->len = cbThisPage;
754 pPipeBuf->offset = offPg0;
755# if RTLNX_VER_MIN(2,6,23)
756 pPipeBuf->private = 0;
757# endif
758 pPipeBuf->ops = &vbsf_pipe_buf_ops;
759 pPipeBuf->flags = fFlags & SPLICE_F_GIFT ? PIPE_BUF_FLAG_GIFT : 0;
760 pPipeBuf->page = papPages[iPage];
761
762 papPages[iPage++] = NULL;
763 pPipe->nrbufs++;
764 fNeedWakeUp |= pPipe->inode != NULL;
765 offPg0 = 0;
766 cbRet += cbThisPage;
767
768 /* done? */
769 cbActual -= cbThisPage;
770 if (!cbActual)
771 break;
772 } else if (pPipe->readers == 0) {
773 SFLOGFLOW(("vbsf_feed_pages_to_pipe: no readers!\n"));
774 send_sig(SIGPIPE, current, 0);
775 if (cbRet == 0)
776 cbRet = -EPIPE;
777 break;
778 } else if (fFlags & SPLICE_F_NONBLOCK) {
779 if (cbRet == 0)
780 cbRet = -EAGAIN;
781 break;
782 } else if (signal_pending(current)) {
783 if (cbRet == 0)
784 cbRet = -ERESTARTSYS;
785 SFLOGFLOW(("vbsf_feed_pages_to_pipe: pending signal! (%zd)\n", cbRet));
786 break;
787 } else {
788 if (fNeedWakeUp) {
789 vbsf_wake_up_pipe(pPipe, true /*fReaders*/);
790 fNeedWakeUp = 0;
791 }
792 pPipe->waiting_writers++;
793 vbsf_wait_pipe(pPipe);
794 pPipe->waiting_writers--;
795 }
796 }
797 UNLOCK_PIPE(pPipe);
798
799 if (fNeedWakeUp)
800 vbsf_wake_up_pipe(pPipe, true /*fReaders*/);
801
802 return cbRet;
803}
804
805
806/**
807 * For splicing from a file to a pipe.
808 */
809static ssize_t vbsf_splice_read(struct file *file, loff_t *poffset, struct pipe_inode_info *pipe, size_t len, unsigned int flags)
810{
811 struct inode *inode = VBSF_GET_F_DENTRY(file)->d_inode;
812 struct vbsf_super_info *pSuperInfo = VBSF_GET_SUPER_INFO(inode->i_sb);
813 ssize_t cbRet;
814
815 SFLOGFLOW(("vbsf_splice_read: file=%p poffset=%p{%#RX64} pipe=%p len=%#zx flags=%#x\n", file, poffset, *poffset, pipe, len, flags));
816 if (vbsf_should_use_cached_read(file, inode->i_mapping, pSuperInfo)) {
817 cbRet = generic_file_splice_read(file, poffset, pipe, len, flags);
818 } else {
819 /*
820 * Create a read request.
821 */
822 loff_t offFile = *poffset;
823 size_t cPages = RT_MIN(RT_ALIGN_Z((offFile & ~PAGE_CACHE_MASK) + len, PAGE_CACHE_SIZE) >> PAGE_CACHE_SHIFT,
824 PIPE_BUFFERS);
825 VBOXSFREADPGLSTREQ *pReq = (VBOXSFREADPGLSTREQ *)VbglR0PhysHeapAlloc(RT_UOFFSETOF_DYN(VBOXSFREADPGLSTREQ,
826 PgLst.aPages[cPages]));
827 if (pReq) {
828 /*
829 * Allocate pages.
830 */
831 struct page *apPages[PIPE_BUFFERS];
832 size_t i;
833 pReq->PgLst.offFirstPage = (uint16_t)offFile & (uint16_t)PAGE_OFFSET_MASK;
834 cbRet = 0;
835 for (i = 0; i < cPages; i++) {
836 struct page *pPage;
837 apPages[i] = pPage = alloc_page(GFP_USER);
838 if (pPage) {
839 pReq->PgLst.aPages[i] = page_to_phys(pPage);
840# ifdef VBOX_STRICT
841 ASMMemFill32(kmap(pPage), PAGE_SIZE, UINT32_C(0xdeadbeef));
842 kunmap(pPage);
843# endif
844 } else {
845 cbRet = -ENOMEM;
846 break;
847 }
848 }
849 if (cbRet == 0) {
850 /*
851 * Do the reading.
852 */
853 uint32_t const cbToRead = RT_MIN((cPages << PAGE_SHIFT) - (offFile & PAGE_OFFSET_MASK), len);
854 struct vbsf_reg_info *sf_r = (struct vbsf_reg_info *)file->private_data;
855 int vrc = VbglR0SfHostReqReadPgLst(pSuperInfo->map.root, pReq, sf_r->Handle.hHost, offFile, cbToRead, cPages);
856 if (RT_SUCCESS(vrc)) {
857 /*
858 * Get the number of bytes read, jettison the request
859 * and, in case of EOF, any unnecessary pages.
860 */
861 uint32_t cbActual = pReq->Parms.cb32Read.u.value32;
862 AssertStmt(cbActual <= cbToRead, cbActual = cbToRead);
863 SFLOG2(("vbsf_splice_read: read -> %#x bytes @ %#RX64\n", cbActual, offFile));
864
865 VbglR0PhysHeapFree(pReq);
866 pReq = NULL;
867
868 /*
869 * Now, feed it to the pipe thingy.
870 * This will take ownership of the all pages no matter what happens.
871 */
872 cbRet = vbsf_feed_pages_to_pipe(pipe, apPages, cPages, offFile & PAGE_OFFSET_MASK, cbActual, flags);
873 if (cbRet > 0)
874 *poffset = offFile + cbRet;
875 } else {
876 cbRet = -RTErrConvertToErrno(vrc);
877 SFLOGFLOW(("vbsf_splice_read: Read failed: %Rrc -> %zd\n", vrc, cbRet));
878 }
879 i = cPages;
880 }
881
882 while (i-- > 0)
883 if (apPages[i])
884 __free_pages(apPages[i], 0);
885 if (pReq)
886 VbglR0PhysHeapFree(pReq);
887 } else {
888 cbRet = -ENOMEM;
889 }
890 }
891 SFLOGFLOW(("vbsf_splice_read: returns %zd (%#zx), *poffset=%#RX64\n", cbRet, cbRet, *poffset));
892 return cbRet;
893}
894
895#endif /* 2.6.17 <= LINUX_VERSION_CODE < 2.6.31 */
896#if RTLNX_VER_RANGE(2,6,17, 3,16,0)
897
898/**
899 * For splicing from a pipe to a file.
900 *
901 * Since we can combine buffers and request allocations, this should be faster
902 * than the default implementation.
903 */
904static ssize_t vbsf_splice_write(struct pipe_inode_info *pPipe, struct file *file, loff_t *poffset, size_t len, unsigned int flags)
905{
906 struct inode *inode = VBSF_GET_F_DENTRY(file)->d_inode;
907 struct vbsf_super_info *pSuperInfo = VBSF_GET_SUPER_INFO(inode->i_sb);
908 ssize_t cbRet;
909
910 SFLOGFLOW(("vbsf_splice_write: pPipe=%p file=%p poffset=%p{%#RX64} len=%#zx flags=%#x\n", pPipe, file, poffset, *poffset, len, flags));
911 /** @todo later if (false) {
912 cbRet = generic_file_splice_write(pPipe, file, poffset, len, flags);
913 } else */ {
914 /*
915 * Prepare a write request.
916 */
917# ifdef PIPE_BUFFERS
918 uint32_t const cMaxPages = RT_MIN(PIPE_BUFFERS, RT_ALIGN_Z(len, PAGE_SIZE) >> PAGE_SHIFT);
919# else
920 uint32_t const cMaxPages = RT_MIN(RT_MAX(RT_MIN(pPipe->buffers, 256), PIPE_DEF_BUFFERS),
921 RT_ALIGN_Z(len, PAGE_SIZE) >> PAGE_SHIFT);
922# endif
923 VBOXSFWRITEPGLSTREQ *pReq = (VBOXSFWRITEPGLSTREQ *)VbglR0PhysHeapAlloc(RT_UOFFSETOF_DYN(VBOXSFREADPGLSTREQ,
924 PgLst.aPages[cMaxPages]));
925 if (pReq) {
926 /*
927 * Feed from the pipe.
928 */
929 struct vbsf_reg_info *sf_r = (struct vbsf_reg_info *)file->private_data;
930 struct address_space *mapping = inode->i_mapping;
931 loff_t offFile = *poffset;
932 bool fNeedWakeUp = false;
933 cbRet = 0;
934
935 LOCK_PIPE(pPipe);
936
937 for (;;) {
938 unsigned cBufs = pPipe->nrbufs;
939 /*SFLOG2(("vbsf_splice_write: nrbufs=%#x curbuf=%#x\n", cBufs, pPipe->curbuf));*/
940 if (cBufs) {
941 /*
942 * There is data available. Write it to the file.
943 */
944 int vrc;
945 struct pipe_buffer *pPipeBuf = &pPipe->bufs[pPipe->curbuf];
946 uint32_t cPagesToWrite = 1;
947 uint32_t cbToWrite = pPipeBuf->len;
948
949 Assert(pPipeBuf->offset < PAGE_SIZE);
950 Assert(pPipeBuf->offset + pPipeBuf->len <= PAGE_SIZE);
951
952 pReq->PgLst.offFirstPage = pPipeBuf->offset & PAGE_OFFSET;
953 pReq->PgLst.aPages[0] = page_to_phys(pPipeBuf->page);
954
955 /* Add any adjacent page buffers: */
956 while ( cPagesToWrite < cBufs
957 && cPagesToWrite < cMaxPages
958 && ((pReq->PgLst.offFirstPage + cbToWrite) & PAGE_OFFSET_MASK) == 0) {
959# ifdef PIPE_BUFFERS
960 struct pipe_buffer *pPipeBuf2 = &pPipe->bufs[(pPipe->curbuf + cPagesToWrite) % PIPE_BUFFERS];
961# else
962 struct pipe_buffer *pPipeBuf2 = &pPipe->bufs[(pPipe->curbuf + cPagesToWrite) % pPipe->buffers];
963# endif
964 Assert(pPipeBuf2->len <= PAGE_SIZE);
965 Assert(pPipeBuf2->offset < PAGE_SIZE);
966 if (pPipeBuf2->offset != 0)
967 break;
968 pReq->PgLst.aPages[cPagesToWrite] = page_to_phys(pPipeBuf2->page);
969 cbToWrite += pPipeBuf2->len;
970 cPagesToWrite += 1;
971 }
972
973 /* Check that we don't have signals pending before we issue the write, as
974 we'll only end up having to cancel the HGCM request 99% of the time: */
975 if (!signal_pending(current)) {
976 struct vbsf_inode_info *sf_i = VBSF_GET_INODE_INFO(inode);
977 vrc = VbglR0SfHostReqWritePgLst(pSuperInfo->map.root, pReq, sf_r->Handle.hHost, offFile,
978 cbToWrite, cPagesToWrite);
979 sf_i->ModificationTimeAtOurLastWrite = sf_i->ModificationTime;
980 } else
981 vrc = VERR_INTERRUPTED;
982 if (RT_SUCCESS(vrc)) {
983 /*
984 * Get the number of bytes actually written, update file position
985 * and return value, and advance the pipe buffer.
986 */
987 uint32_t cbActual = pReq->Parms.cb32Write.u.value32;
988 AssertStmt(cbActual <= cbToWrite, cbActual = cbToWrite);
989 SFLOG2(("vbsf_splice_write: write -> %#x bytes @ %#RX64\n", cbActual, offFile));
990
991 cbRet += cbActual;
992
993 while (cbActual > 0) {
994 uint32_t cbAdvance = RT_MIN(pPipeBuf->len, cbActual);
995
996 vbsf_reg_write_sync_page_cache(mapping, offFile, cbAdvance, NULL,
997 &pPipeBuf->page, pPipeBuf->offset, 1);
998
999 offFile += cbAdvance;
1000 cbActual -= cbAdvance;
1001 pPipeBuf->offset += cbAdvance;
1002 pPipeBuf->len -= cbAdvance;
1003
1004 if (!pPipeBuf->len) {
1005 struct pipe_buf_operations const *pOps = pPipeBuf->ops;
1006 pPipeBuf->ops = NULL;
1007 pOps->release(pPipe, pPipeBuf);
1008
1009# ifdef PIPE_BUFFERS
1010 pPipe->curbuf = (pPipe->curbuf + 1) % PIPE_BUFFERS;
1011# else
1012 pPipe->curbuf = (pPipe->curbuf + 1) % pPipe->buffers;
1013# endif
1014 pPipe->nrbufs -= 1;
1015 pPipeBuf = &pPipe->bufs[pPipe->curbuf];
1016
1017# if RTLNX_VER_MAX(2,6,30)
1018 fNeedWakeUp |= pPipe->inode != NULL;
1019# else
1020 fNeedWakeUp = true;
1021# endif
1022 } else {
1023 Assert(cbActual == 0);
1024 break;
1025 }
1026 }
1027
1028 *poffset = offFile;
1029 } else {
1030 if (cbRet == 0)
1031 cbRet = vrc == VERR_INTERRUPTED ? -ERESTARTSYS : -RTErrConvertToErrno(vrc);
1032 SFLOGFLOW(("vbsf_splice_write: Write failed: %Rrc -> %zd (cbRet=%#zx)\n",
1033 vrc, -RTErrConvertToErrno(vrc), cbRet));
1034 break;
1035 }
1036 } else {
1037 /*
1038 * Wait for data to become available, if there is chance that'll happen.
1039 */
1040 /* Quit if there are no writers (think EOF): */
1041 if (pPipe->writers == 0) {
1042 SFLOGFLOW(("vbsf_splice_write: No buffers. No writers. The show is done!\n"));
1043 break;
1044 }
1045
1046 /* Quit if if we've written some and no writers waiting on the lock: */
1047 if (cbRet > 0 && pPipe->waiting_writers == 0) {
1048 SFLOGFLOW(("vbsf_splice_write: No waiting writers, returning what we've got.\n"));
1049 break;
1050 }
1051
1052 /* Quit with EAGAIN if non-blocking: */
1053 if (flags & SPLICE_F_NONBLOCK) {
1054 if (cbRet == 0)
1055 cbRet = -EAGAIN;
1056 break;
1057 }
1058
1059 /* Quit if we've got pending signals: */
1060 if (signal_pending(current)) {
1061 if (cbRet == 0)
1062 cbRet = -ERESTARTSYS;
1063 SFLOGFLOW(("vbsf_splice_write: pending signal! (%zd)\n", cbRet));
1064 break;
1065 }
1066
1067 /* Wake up writers before we start waiting: */
1068 if (fNeedWakeUp) {
1069 vbsf_wake_up_pipe(pPipe, false /*fReaders*/);
1070 fNeedWakeUp = false;
1071 }
1072 vbsf_wait_pipe(pPipe);
1073 }
1074 } /* feed loop */
1075
1076 if (fNeedWakeUp)
1077 vbsf_wake_up_pipe(pPipe, false /*fReaders*/);
1078
1079 UNLOCK_PIPE(pPipe);
1080
1081 VbglR0PhysHeapFree(pReq);
1082 } else {
1083 cbRet = -ENOMEM;
1084 }
1085 }
1086 SFLOGFLOW(("vbsf_splice_write: returns %zd (%#zx), *poffset=%#RX64\n", cbRet, cbRet, *poffset));
1087 return cbRet;
1088}
1089
1090#endif /* 2.6.17 <= LINUX_VERSION_CODE < 3.16.0 */
1091
1092#if RTLNX_VER_RANGE(2,5,30, 2,6,23)
1093/**
1094 * Our own senfile implementation that does not go via the page cache like
1095 * generic_file_sendfile() does.
1096 */
1097static ssize_t vbsf_reg_sendfile(struct file *pFile, loff_t *poffFile, size_t cbToSend, read_actor_t pfnActor,
1098# if RTLNX_VER_MIN(2,6,8)
1099 void *pvUser
1100# else
1101 void __user *pvUser
1102# endif
1103 )
1104{
1105 struct inode *inode = VBSF_GET_F_DENTRY(pFile)->d_inode;
1106 struct vbsf_super_info *pSuperInfo = VBSF_GET_SUPER_INFO(inode->i_sb);
1107 ssize_t cbRet;
1108 SFLOGFLOW(("vbsf_reg_sendfile: pFile=%p poffFile=%p{%#RX64} cbToSend=%#zx pfnActor=%p pvUser=%p\n",
1109 pFile, poffFile, poffFile ? *poffFile : 0, cbToSend, pfnActor, pvUser));
1110 Assert(pSuperInfo);
1111
1112 /*
1113 * Return immediately if asked to send nothing.
1114 */
1115 if (cbToSend == 0)
1116 return 0;
1117
1118 /*
1119 * Like for vbsf_reg_read() and vbsf_reg_read_iter(), we allow going via
1120 * the page cache in some cases or configs.
1121 */
1122 if (vbsf_should_use_cached_read(pFile, inode->i_mapping, pSuperInfo)) {
1123 cbRet = generic_file_sendfile(pFile, poffFile, cbToSend, pfnActor, pvUser);
1124 SFLOGFLOW(("vbsf_reg_sendfile: returns %#zx *poffFile=%#RX64 [generic_file_sendfile]\n", cbRet, poffFile ? *poffFile : UINT64_MAX));
1125 } else {
1126 /*
1127 * Allocate a request and a bunch of pages for reading from the file.
1128 */
1129 struct page *apPages[16];
1130 loff_t offFile = poffFile ? *poffFile : 0;
1131 size_t const cPages = cbToSend + ((size_t)offFile & PAGE_OFFSET_MASK) >= RT_ELEMENTS(apPages) * PAGE_SIZE
1132 ? RT_ELEMENTS(apPages)
1133 : RT_ALIGN_Z(cbToSend + ((size_t)offFile & PAGE_OFFSET_MASK), PAGE_SIZE) >> PAGE_SHIFT;
1134 size_t iPage;
1135 VBOXSFREADPGLSTREQ *pReq = (VBOXSFREADPGLSTREQ *)VbglR0PhysHeapAlloc(RT_UOFFSETOF_DYN(VBOXSFREADPGLSTREQ,
1136 PgLst.aPages[cPages]));
1137 if (pReq) {
1138 Assert(cPages > 0);
1139 cbRet = 0;
1140 for (iPage = 0; iPage < cPages; iPage++) {
1141 struct page *pPage;
1142 apPages[iPage] = pPage = alloc_page(GFP_USER);
1143 if (pPage) {
1144 Assert(page_count(pPage) == 1);
1145 pReq->PgLst.aPages[iPage] = page_to_phys(pPage);
1146 } else {
1147 while (iPage-- > 0)
1148 vbsf_put_page(apPages[iPage]);
1149 cbRet = -ENOMEM;
1150 break;
1151 }
1152 }
1153 if (cbRet == 0) {
1154 /*
1155 * Do the job.
1156 */
1157 struct vbsf_reg_info *sf_r = (struct vbsf_reg_info *)pFile->private_data;
1158 read_descriptor_t RdDesc;
1159 RdDesc.count = cbToSend;
1160# if RTLNX_VER_MIN(2,6,8)
1161 RdDesc.arg.data = pvUser;
1162# else
1163 RdDesc.buf = pvUser;
1164# endif
1165 RdDesc.written = 0;
1166 RdDesc.error = 0;
1167
1168 Assert(sf_r);
1169 Assert((sf_r->Handle.fFlags & VBSF_HANDLE_F_MAGIC_MASK) == VBSF_HANDLE_F_MAGIC);
1170
1171 while (cbToSend > 0) {
1172 /*
1173 * Read another chunk. For paranoid reasons, we keep data where the page cache
1174 * would keep it, i.e. page offset bits corresponds to the file offset bits.
1175 */
1176 uint32_t const offPg0 = (uint32_t)offFile & (uint32_t)PAGE_OFFSET_MASK;
1177 uint32_t const cbToRead = RT_MIN((cPages << PAGE_SHIFT) - offPg0, cbToSend);
1178 uint32_t const cPagesToRead = RT_ALIGN_Z(cbToRead + offPg0, PAGE_SIZE) >> PAGE_SHIFT;
1179 int vrc;
1180 pReq->PgLst.offFirstPage = (uint16_t)offPg0;
1181 if (!signal_pending(current))
1182 vrc = VbglR0SfHostReqReadPgLst(pSuperInfo->map.root, pReq, sf_r->Handle.hHost, offFile,
1183 cbToRead, cPagesToRead);
1184 else
1185 vrc = VERR_INTERRUPTED;
1186 if (RT_SUCCESS(vrc)) {
1187 /*
1188 * Pass what we read to the actor.
1189 */
1190 uint32_t off = offPg0;
1191 uint32_t cbActual = pReq->Parms.cb32Read.u.value32;
1192 bool const fIsEof = cbActual < cbToRead;
1193 AssertStmt(cbActual <= cbToRead, cbActual = cbToRead);
1194 SFLOG3(("vbsf_reg_sendfile: Read %#x bytes (offPg0=%#x), wanted %#x ...\n", cbActual, offPg0, cbToRead));
1195
1196 iPage = 0;
1197 while (cbActual > 0) {
1198 uint32_t const cbPage = RT_MIN(cbActual, PAGE_SIZE - off);
1199 int const cbRetActor = pfnActor(&RdDesc, apPages[iPage], off, cbPage);
1200 Assert(cbRetActor >= 0); /* Returns zero on failure, with RdDesc.error holding the status code. */
1201
1202 AssertMsg(iPage < cPages && iPage < cPagesToRead, ("iPage=%#x cPages=%#x cPagesToRead=%#x\n", iPage, cPages, cPagesToRead));
1203
1204 offFile += cbRetActor;
1205 if ((uint32_t)cbRetActor == cbPage && RdDesc.count > 0) {
1206 cbActual -= cbPage;
1207 cbToSend -= cbPage;
1208 iPage++;
1209 } else {
1210 SFLOG3(("vbsf_reg_sendfile: cbRetActor=%#x (%d) cbPage=%#x RdDesc{count=%#lx error=%d} iPage=%#x/%#x/%#x cbToSend=%#zx\n",
1211 cbRetActor, cbRetActor, cbPage, RdDesc.count, RdDesc.error, iPage, cPagesToRead, cPages, cbToSend));
1212 vrc = VERR_CALLBACK_RETURN;
1213 break;
1214 }
1215 off = 0;
1216 }
1217
1218 /*
1219 * Are we done yet?
1220 */
1221 if (RT_FAILURE_NP(vrc) || cbToSend == 0 || RdDesc.error != 0 || fIsEof) {
1222 break;
1223 }
1224
1225 /*
1226 * Replace pages held by the actor.
1227 */
1228 vrc = VINF_SUCCESS;
1229 for (iPage = 0; iPage < cPages; iPage++) {
1230 struct page *pPage = apPages[iPage];
1231 if (page_count(pPage) != 1) {
1232 struct page *pNewPage = alloc_page(GFP_USER);
1233 if (pNewPage) {
1234 SFLOGFLOW(("vbsf_reg_sendfile: Replacing page #%x: %p -> %p\n", iPage, pPage, pNewPage));
1235 vbsf_put_page(pPage);
1236 apPages[iPage] = pNewPage;
1237 } else {
1238 SFLOGFLOW(("vbsf_reg_sendfile: Failed to allocate a replacement page.\n"));
1239 vrc = VERR_NO_MEMORY;
1240 break;
1241 }
1242 }
1243 }
1244 if (RT_FAILURE(vrc))
1245 break; /* RdDesc.written should be non-zero, so don't bother with setting error. */
1246 } else {
1247 RdDesc.error = vrc == VERR_INTERRUPTED ? -ERESTARTSYS : -RTErrConvertToErrno(vrc);
1248 SFLOGFLOW(("vbsf_reg_sendfile: Read failed: %Rrc -> %zd (RdDesc.error=%#d)\n",
1249 vrc, -RTErrConvertToErrno(vrc), RdDesc.error));
1250 break;
1251 }
1252 }
1253
1254 /*
1255 * Free memory.
1256 */
1257 for (iPage = 0; iPage < cPages; iPage++)
1258 vbsf_put_page(apPages[iPage]);
1259
1260 /*
1261 * Set the return values.
1262 */
1263 if (RdDesc.written) {
1264 cbRet = RdDesc.written;
1265 if (poffFile)
1266 *poffFile = offFile;
1267 } else {
1268 cbRet = RdDesc.error;
1269 }
1270 }
1271 VbglR0PhysHeapFree(pReq);
1272 } else {
1273 cbRet = -ENOMEM;
1274 }
1275 SFLOGFLOW(("vbsf_reg_sendfile: returns %#zx offFile=%#RX64\n", cbRet, offFile));
1276 }
1277 return cbRet;
1278}
1279#endif /* 2.5.30 <= LINUX_VERSION_CODE < 2.6.23 */
1280
1281
1282/*********************************************************************************************************************************
1283* File operations on regular files *
1284*********************************************************************************************************************************/
1285
1286/** Wrapper around put_page / page_cache_release. */
1287DECLINLINE(void) vbsf_put_page(struct page *pPage)
1288{
1289#if RTLNX_VER_MIN(4,6,0)
1290 put_page(pPage);
1291#else
1292 page_cache_release(pPage);
1293#endif
1294}
1295
1296
1297/** Wrapper around get_page / page_cache_get. */
1298DECLINLINE(void) vbsf_get_page(struct page *pPage)
1299{
1300#if RTLNX_VER_MIN(4,6,0)
1301 get_page(pPage);
1302#else
1303 page_cache_get(pPage);
1304#endif
1305}
1306
1307
1308/** Companion to vbsf_lock_user_pages(). */
1309static void vbsf_unlock_user_pages(struct page **papPages, size_t cPages, bool fSetDirty, bool fLockPgHack)
1310{
1311 /* We don't mark kernel pages dirty: */
1312 if (fLockPgHack)
1313 fSetDirty = false;
1314
1315 while (cPages-- > 0)
1316 {
1317 struct page *pPage = papPages[cPages];
1318 Assert((ssize_t)cPages >= 0);
1319 if (fSetDirty && !PageReserved(pPage))
1320 set_page_dirty(pPage);
1321 vbsf_put_page(pPage);
1322 }
1323}
1324
1325
1326/**
1327 * Worker for vbsf_lock_user_pages_failed_check_kernel() and
1328 * vbsf_iter_lock_pages().
1329 */
1330static int vbsf_lock_kernel_pages(uint8_t *pbStart, bool fWrite, size_t cPages, struct page **papPages)
1331{
1332 uintptr_t const uPtrFrom = (uintptr_t)pbStart;
1333 uintptr_t const uPtrLast = (uPtrFrom & ~(uintptr_t)PAGE_OFFSET_MASK) + (cPages << PAGE_SHIFT) - 1;
1334 uint8_t *pbPage = (uint8_t *)uPtrLast;
1335 size_t iPage = cPages;
1336
1337 /*
1338 * Touch the pages first (paranoia^2).
1339 */
1340 if (fWrite) {
1341 uint8_t volatile *pbProbe = (uint8_t volatile *)uPtrFrom;
1342 while (iPage-- > 0) {
1343 *pbProbe = *pbProbe;
1344 pbProbe += PAGE_SIZE;
1345 }
1346 } else {
1347 uint8_t const *pbProbe = (uint8_t const *)uPtrFrom;
1348 while (iPage-- > 0) {
1349 ASMProbeReadByte(pbProbe);
1350 pbProbe += PAGE_SIZE;
1351 }
1352 }
1353
1354 /*
1355 * Get the pages.
1356 * Note! Fixes here probably applies to rtR0MemObjNativeLockKernel as well.
1357 */
1358 iPage = cPages;
1359 if ( uPtrFrom >= (unsigned long)__va(0)
1360 && uPtrLast < (unsigned long)high_memory) {
1361 /* The physical page mapping area: */
1362 while (iPage-- > 0) {
1363 struct page *pPage = papPages[iPage] = virt_to_page(pbPage);
1364 vbsf_get_page(pPage);
1365 pbPage -= PAGE_SIZE;
1366 }
1367 } else {
1368 /* This is vmalloc or some such thing, so go thru page tables: */
1369 while (iPage-- > 0) {
1370 struct page *pPage = rtR0MemObjLinuxVirtToPage(pbPage);
1371 if (pPage) {
1372 papPages[iPage] = pPage;
1373 vbsf_get_page(pPage);
1374 pbPage -= PAGE_SIZE;
1375 } else {
1376 while (++iPage < cPages) {
1377 pPage = papPages[iPage];
1378 vbsf_put_page(pPage);
1379 }
1380 return -EFAULT;
1381 }
1382 }
1383 }
1384 return 0;
1385}
1386
1387
1388/**
1389 * Catches kernel_read() and kernel_write() calls and works around them.
1390 *
1391 * The file_operations::read and file_operations::write callbacks supposedly
1392 * hands us the user buffers to read into and write out of. To allow the kernel
1393 * to read and write without allocating buffers in userland, they kernel_read()
1394 * and kernel_write() increases the user space address limit before calling us
1395 * so that copyin/copyout won't reject it. Our problem is that get_user_pages()
1396 * works on the userspace address space structures and will not be fooled by an
1397 * increased addr_limit.
1398 *
1399 * This code tries to detect this situation and fake get_user_lock() for the
1400 * kernel buffer.
1401 */
1402static int vbsf_lock_user_pages_failed_check_kernel(uintptr_t uPtrFrom, size_t cPages, bool fWrite, int rcFailed,
1403 struct page **papPages, bool *pfLockPgHack)
1404{
1405 /*
1406 * Check that this is valid user memory that is actually in the kernel range.
1407 */
1408#if RTLNX_VER_MIN(5,10,0)
1409 if ( access_ok((void *)uPtrFrom, cPages << PAGE_SHIFT)
1410 && uPtrFrom >= TASK_SIZE_MAX)
1411#elif RTLNX_VER_MIN(5,0,0) || RTLNX_RHEL_MIN(8,1)
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 RTLNX_VER_MIN(4,9,0) \
1434 || (defined(CONFIG_SUSE_KERNEL) && RTLNX_VER_RANGE(4,4,73, 4,4,74) /** @todo Figure out when & what exactly. */) \
1435 || (defined(CONFIG_SUSE_KERNEL) && RTLNX_VER_RANGE(4,4,75, 4,4,90) /** @todo Figure out when & what exactly. */) \
1436 || (defined(CONFIG_SUSE_KERNEL) && RTLNX_VER_RANGE(4,4,92, 4,5,0) /** @todo Figure out when & what exactly. */)
1437 ssize_t cPagesLocked = get_user_pages_unlocked(uPtrFrom, cPages, papPages,
1438 fWrite ? FOLL_WRITE | FOLL_FORCE : FOLL_FORCE);
1439# elif RTLNX_VER_MIN(4,6,0)
1440 ssize_t cPagesLocked = get_user_pages_unlocked(uPtrFrom, cPages, fWrite, 1 /*force*/, papPages);
1441# elif RTLNX_VER_RANGE(4,4,168, 4,5,0)
1442 ssize_t cPagesLocked = get_user_pages_unlocked(current, current->mm, uPtrFrom, cPages, papPages,
1443 fWrite ? FOLL_WRITE | FOLL_FORCE : FOLL_FORCE);
1444# elif RTLNX_VER_MIN(4,0,0)
1445 ssize_t cPagesLocked = get_user_pages_unlocked(current, current->mm, uPtrFrom, cPages, fWrite, 1 /*force*/, papPages);
1446# else
1447 struct task_struct *pTask = current;
1448 ssize_t cPagesLocked;
1449 down_read(&pTask->mm->mmap_sem);
1450 cPagesLocked = get_user_pages(pTask, pTask->mm, uPtrFrom, cPages, fWrite, 1 /*force*/, papPages, NULL);
1451 up_read(&pTask->mm->mmap_sem);
1452# endif
1453 *pfLockPgHack = false;
1454 if (cPagesLocked == cPages)
1455 return 0;
1456
1457 /*
1458 * It failed.
1459 */
1460 if (cPagesLocked < 0)
1461 return vbsf_lock_user_pages_failed_check_kernel(uPtrFrom, cPages, fWrite, (int)cPagesLocked, papPages, pfLockPgHack);
1462
1463 vbsf_unlock_user_pages(papPages, cPagesLocked, false /*fSetDirty*/, false /*fLockPgHack*/);
1464
1465 /* We could use uPtrFrom + cPagesLocked to get the correct status here... */
1466 return -EFAULT;
1467}
1468
1469
1470#if RTLNX_VER_MAX(5,10,0)
1471/**
1472 * Read function used when accessing files that are memory mapped.
1473 *
1474 * We read from the page cache here to present the a cohertent picture of the
1475 * the file content.
1476 */
1477static ssize_t vbsf_reg_read_mapped(struct file *file, char /*__user*/ *buf, size_t size, loff_t *off)
1478{
1479# if RTLNX_VER_MIN(3,16,0)
1480 struct iovec iov = { .iov_base = buf, .iov_len = size };
1481 struct iov_iter iter;
1482 struct kiocb kiocb;
1483 ssize_t cbRet;
1484
1485 init_sync_kiocb(&kiocb, file);
1486 kiocb.ki_pos = *off;
1487 iov_iter_init(&iter, READ, &iov, 1, size);
1488
1489 cbRet = generic_file_read_iter(&kiocb, &iter);
1490
1491 *off = kiocb.ki_pos;
1492 return cbRet;
1493
1494# elif RTLNX_VER_MIN(2,6,19)
1495 struct iovec iov = { .iov_base = buf, .iov_len = size };
1496 struct kiocb kiocb;
1497 ssize_t cbRet;
1498
1499 init_sync_kiocb(&kiocb, file);
1500 kiocb.ki_pos = *off;
1501
1502 cbRet = generic_file_aio_read(&kiocb, &iov, 1, *off);
1503 if (cbRet == -EIOCBQUEUED)
1504 cbRet = wait_on_sync_kiocb(&kiocb);
1505
1506 *off = kiocb.ki_pos;
1507 return cbRet;
1508
1509# else /* 2.6.18 or earlier: */
1510 return generic_file_read(file, buf, size, off);
1511# endif
1512}
1513
1514
1515/**
1516 * Fallback case of vbsf_reg_read() that locks the user buffers and let the host
1517 * write directly to them.
1518 */
1519static ssize_t vbsf_reg_read_locking(struct file *file, char /*__user*/ *buf, size_t size, loff_t *off,
1520 struct vbsf_super_info *pSuperInfo, struct vbsf_reg_info *sf_r)
1521{
1522 /*
1523 * Lock pages and execute the read, taking care not to pass the host
1524 * more than it can handle in one go or more than we care to allocate
1525 * page arrays for. The latter limit is set at just short of 32KB due
1526 * to how the physical heap works.
1527 */
1528 struct page *apPagesStack[16];
1529 struct page **papPages = &apPagesStack[0];
1530 struct page **papPagesFree = NULL;
1531 VBOXSFREADPGLSTREQ *pReq;
1532 loff_t offFile = *off;
1533 ssize_t cbRet = -ENOMEM;
1534 size_t cPages = (((uintptr_t)buf & PAGE_OFFSET_MASK) + size + PAGE_OFFSET_MASK) >> PAGE_SHIFT;
1535 size_t cMaxPages = RT_MIN(RT_MAX(pSuperInfo->cMaxIoPages, 1), cPages);
1536 bool fLockPgHack;
1537
1538 pReq = (VBOXSFREADPGLSTREQ *)VbglR0PhysHeapAlloc(RT_UOFFSETOF_DYN(VBOXSFREADPGLSTREQ, PgLst.aPages[cMaxPages]));
1539 while (!pReq && cMaxPages > 4) {
1540 cMaxPages /= 2;
1541 pReq = (VBOXSFREADPGLSTREQ *)VbglR0PhysHeapAlloc(RT_UOFFSETOF_DYN(VBOXSFREADPGLSTREQ, PgLst.aPages[cMaxPages]));
1542 }
1543 if (pReq && cMaxPages > RT_ELEMENTS(apPagesStack))
1544 papPagesFree = papPages = kmalloc(cMaxPages * sizeof(sizeof(papPages[0])), GFP_KERNEL);
1545 if (pReq && papPages) {
1546 cbRet = 0;
1547 for (;;) {
1548 /*
1549 * Figure out how much to process now and lock the user pages.
1550 */
1551 int rc;
1552 size_t cbChunk = (uintptr_t)buf & PAGE_OFFSET_MASK;
1553 pReq->PgLst.offFirstPage = (uint16_t)cbChunk;
1554 cPages = RT_ALIGN_Z(cbChunk + size, PAGE_SIZE) >> PAGE_SHIFT;
1555 if (cPages <= cMaxPages)
1556 cbChunk = size;
1557 else {
1558 cPages = cMaxPages;
1559 cbChunk = (cMaxPages << PAGE_SHIFT) - cbChunk;
1560 }
1561
1562 rc = vbsf_lock_user_pages((uintptr_t)buf, cPages, true /*fWrite*/, papPages, &fLockPgHack);
1563 if (rc == 0) {
1564 size_t iPage = cPages;
1565 while (iPage-- > 0)
1566 pReq->PgLst.aPages[iPage] = page_to_phys(papPages[iPage]);
1567 } else {
1568 cbRet = rc;
1569 break;
1570 }
1571
1572 /*
1573 * Issue the request and unlock the pages.
1574 */
1575 rc = VbglR0SfHostReqReadPgLst(pSuperInfo->map.root, pReq, sf_r->Handle.hHost, offFile, cbChunk, cPages);
1576
1577 Assert(cPages <= cMaxPages);
1578 vbsf_unlock_user_pages(papPages, cPages, true /*fSetDirty*/, fLockPgHack);
1579
1580 if (RT_SUCCESS(rc)) {
1581 /*
1582 * Success, advance position and buffer.
1583 */
1584 uint32_t cbActual = pReq->Parms.cb32Read.u.value32;
1585 AssertStmt(cbActual <= cbChunk, cbActual = cbChunk);
1586 cbRet += cbActual;
1587 offFile += cbActual;
1588 buf = (uint8_t *)buf + cbActual;
1589 size -= cbActual;
1590
1591 /*
1592 * Are we done already? If so commit the new file offset.
1593 */
1594 if (!size || cbActual < cbChunk) {
1595 *off = offFile;
1596 break;
1597 }
1598 } else if (rc == VERR_NO_MEMORY && cMaxPages > 4) {
1599 /*
1600 * The host probably doesn't have enough heap to handle the
1601 * request, reduce the page count and retry.
1602 */
1603 cMaxPages /= 4;
1604 Assert(cMaxPages > 0);
1605 } else {
1606 /*
1607 * If we've successfully read stuff, return it rather than
1608 * the error. (Not sure if this is such a great idea...)
1609 */
1610 if (cbRet > 0) {
1611 SFLOGFLOW(("vbsf_reg_read: read at %#RX64 -> %Rrc; got cbRet=%#zx already\n", offFile, rc, cbRet));
1612 *off = offFile;
1613 } else {
1614 SFLOGFLOW(("vbsf_reg_read: read at %#RX64 -> %Rrc\n", offFile, rc));
1615 cbRet = -EPROTO;
1616 }
1617 break;
1618 }
1619 }
1620 }
1621 if (papPagesFree)
1622 kfree(papPages);
1623 if (pReq)
1624 VbglR0PhysHeapFree(pReq);
1625 SFLOGFLOW(("vbsf_reg_read: returns %zd (%#zx), *off=%RX64 [lock]\n", cbRet, cbRet, *off));
1626 return cbRet;
1627}
1628
1629/**
1630 * Read from a regular file.
1631 *
1632 * @param file the file
1633 * @param buf the buffer
1634 * @param size length of the buffer
1635 * @param off offset within the file (in/out).
1636 * @returns the number of read bytes on success, Linux error code otherwise
1637 */
1638static ssize_t vbsf_reg_read(struct file *file, char /*__user*/ *buf, size_t size, loff_t *off)
1639{
1640 struct inode *inode = VBSF_GET_F_DENTRY(file)->d_inode;
1641 struct vbsf_super_info *pSuperInfo = VBSF_GET_SUPER_INFO(inode->i_sb);
1642 struct vbsf_reg_info *sf_r = file->private_data;
1643 struct address_space *mapping = inode->i_mapping;
1644
1645 SFLOGFLOW(("vbsf_reg_read: inode=%p file=%p buf=%p size=%#zx off=%#llx\n", inode, file, buf, size, *off));
1646
1647 if (!S_ISREG(inode->i_mode)) {
1648 LogFunc(("read from non regular file %d\n", inode->i_mode));
1649 return -EINVAL;
1650 }
1651
1652 /** @todo XXX Check read permission according to inode->i_mode! */
1653
1654 if (!size)
1655 return 0;
1656
1657 /*
1658 * If there is a mapping and O_DIRECT isn't in effect, we must at a
1659 * heed dirty pages in the mapping and read from them. For simplicity
1660 * though, we just do page cache reading when there are writable
1661 * mappings around with any kind of pages loaded.
1662 */
1663 if (vbsf_should_use_cached_read(file, mapping, pSuperInfo))
1664 return vbsf_reg_read_mapped(file, buf, size, off);
1665
1666 /*
1667 * For small requests, try use an embedded buffer provided we get a heap block
1668 * that does not cross page boundraries (see host code).
1669 */
1670 if (size <= PAGE_SIZE / 4 * 3 - RT_UOFFSETOF(VBOXSFREADEMBEDDEDREQ, abData[0]) /* see allocator */) {
1671 uint32_t const cbReq = RT_UOFFSETOF(VBOXSFREADEMBEDDEDREQ, abData[0]) + size;
1672 VBOXSFREADEMBEDDEDREQ *pReq = (VBOXSFREADEMBEDDEDREQ *)VbglR0PhysHeapAlloc(cbReq);
1673 if (pReq) {
1674 if ((PAGE_SIZE - ((uintptr_t)pReq & PAGE_OFFSET_MASK)) >= cbReq) {
1675 ssize_t cbRet;
1676 int vrc = VbglR0SfHostReqReadEmbedded(pSuperInfo->map.root, pReq, sf_r->Handle.hHost, *off, (uint32_t)size);
1677 if (RT_SUCCESS(vrc)) {
1678 cbRet = pReq->Parms.cb32Read.u.value32;
1679 AssertStmt(cbRet <= (ssize_t)size, cbRet = size);
1680 if (copy_to_user(buf, pReq->abData, cbRet) == 0)
1681 *off += cbRet;
1682 else
1683 cbRet = -EFAULT;
1684 } else
1685 cbRet = -EPROTO;
1686 VbglR0PhysHeapFree(pReq);
1687 SFLOGFLOW(("vbsf_reg_read: returns %zd (%#zx), *off=%RX64 [embed]\n", cbRet, cbRet, *off));
1688 return cbRet;
1689 }
1690 VbglR0PhysHeapFree(pReq);
1691 }
1692 }
1693
1694# if 0 /* Turns out this is slightly slower than locking the pages even for 4KB reads (4.19/amd64). */
1695 /*
1696 * For medium sized requests try use a bounce buffer.
1697 */
1698 if (size <= _64K /** @todo make this configurable? */) {
1699 void *pvBounce = kmalloc(size, GFP_KERNEL);
1700 if (pvBounce) {
1701 VBOXSFREADPGLSTREQ *pReq = (VBOXSFREADPGLSTREQ *)VbglR0PhysHeapAlloc(sizeof(*pReq));
1702 if (pReq) {
1703 ssize_t cbRet;
1704 int vrc = VbglR0SfHostReqReadContig(pSuperInfo->map.root, pReq, sf_r->Handle.hHost, *off,
1705 (uint32_t)size, pvBounce, virt_to_phys(pvBounce));
1706 if (RT_SUCCESS(vrc)) {
1707 cbRet = pReq->Parms.cb32Read.u.value32;
1708 AssertStmt(cbRet <= (ssize_t)size, cbRet = size);
1709 if (copy_to_user(buf, pvBounce, cbRet) == 0)
1710 *off += cbRet;
1711 else
1712 cbRet = -EFAULT;
1713 } else
1714 cbRet = -EPROTO;
1715 VbglR0PhysHeapFree(pReq);
1716 kfree(pvBounce);
1717 SFLOGFLOW(("vbsf_reg_read: returns %zd (%#zx), *off=%RX64 [bounce]\n", cbRet, cbRet, *off));
1718 return cbRet;
1719 }
1720 kfree(pvBounce);
1721 }
1722 }
1723# endif
1724
1725 return vbsf_reg_read_locking(file, buf, size, off, pSuperInfo, sf_r);
1726}
1727#endif /* < 5.10.0 */
1728
1729
1730/**
1731 * Helper the synchronizes the page cache content with something we just wrote
1732 * to the host.
1733 */
1734static void vbsf_reg_write_sync_page_cache(struct address_space *mapping, loff_t offFile, uint32_t cbRange,
1735 uint8_t const *pbSrcBuf, struct page **papSrcPages,
1736 uint32_t offSrcPage, size_t cSrcPages)
1737{
1738 Assert(offSrcPage < PAGE_SIZE);
1739 if (mapping && mapping->nrpages > 0) {
1740 /*
1741 * Work the pages in the write range.
1742 */
1743 while (cbRange > 0) {
1744 /*
1745 * Lookup the page at offFile. We're fine if there aren't
1746 * any there. We're skip if it's dirty or is being written
1747 * back, at least for now.
1748 */
1749 size_t const offDstPage = offFile & PAGE_OFFSET_MASK;
1750 size_t const cbToCopy = RT_MIN(PAGE_SIZE - offDstPage, cbRange);
1751 pgoff_t const idxPage = offFile >> PAGE_SHIFT;
1752 struct page *pDstPage = find_lock_page(mapping, idxPage);
1753 if (pDstPage) {
1754 if ( pDstPage->mapping == mapping /* ignore if re-purposed (paranoia) */
1755 && pDstPage->index == idxPage
1756 && !PageDirty(pDstPage) /* ignore if dirty */
1757 && !PageWriteback(pDstPage) /* ignore if being written back */ ) {
1758 /*
1759 * Map the page and do the copying.
1760 */
1761 uint8_t *pbDst = (uint8_t *)kmap(pDstPage);
1762 if (pbSrcBuf)
1763 memcpy(&pbDst[offDstPage], pbSrcBuf, cbToCopy);
1764 else {
1765 uint32_t const cbSrc0 = PAGE_SIZE - offSrcPage;
1766 uint8_t const *pbSrc = (uint8_t const *)kmap(papSrcPages[0]);
1767 AssertMsg(cSrcPages >= 1, ("offFile=%#llx cbRange=%#zx cbToCopy=%#zx\n", offFile, cbRange, cbToCopy));
1768 memcpy(&pbDst[offDstPage], &pbSrc[offSrcPage], RT_MIN(cbToCopy, cbSrc0));
1769 kunmap(papSrcPages[0]);
1770 if (cbToCopy > cbSrc0) {
1771 AssertMsg(cSrcPages >= 2, ("offFile=%#llx cbRange=%#zx cbToCopy=%#zx\n", offFile, cbRange, cbToCopy));
1772 pbSrc = (uint8_t const *)kmap(papSrcPages[1]);
1773 memcpy(&pbDst[offDstPage + cbSrc0], pbSrc, cbToCopy - cbSrc0);
1774 kunmap(papSrcPages[1]);
1775 }
1776 }
1777 kunmap(pDstPage);
1778 flush_dcache_page(pDstPage);
1779 if (cbToCopy == PAGE_SIZE)
1780 SetPageUptodate(pDstPage);
1781# if RTLNX_VER_MIN(2,4,10)
1782 mark_page_accessed(pDstPage);
1783# endif
1784 } else
1785 SFLOGFLOW(("vbsf_reg_write_sync_page_cache: Skipping page %p: mapping=%p (vs %p) writeback=%d offset=%#lx (vs%#lx)\n",
1786 pDstPage, pDstPage->mapping, mapping, PageWriteback(pDstPage), pDstPage->index, idxPage));
1787 unlock_page(pDstPage);
1788 vbsf_put_page(pDstPage);
1789 }
1790
1791 /*
1792 * Advance.
1793 */
1794 if (pbSrcBuf)
1795 pbSrcBuf += cbToCopy;
1796 else
1797 {
1798 offSrcPage += cbToCopy;
1799 Assert(offSrcPage < PAGE_SIZE * 2);
1800 if (offSrcPage >= PAGE_SIZE) {
1801 offSrcPage &= PAGE_OFFSET_MASK;
1802 papSrcPages++;
1803# ifdef VBOX_STRICT
1804 Assert(cSrcPages > 0);
1805 cSrcPages--;
1806# endif
1807 }
1808 }
1809 offFile += cbToCopy;
1810 cbRange -= cbToCopy;
1811 }
1812 }
1813 RT_NOREF(cSrcPages);
1814}
1815
1816
1817#if RTLNX_VER_MAX(5,10,0)
1818/**
1819 * Fallback case of vbsf_reg_write() that locks the user buffers and let the host
1820 * write directly to them.
1821 */
1822static ssize_t vbsf_reg_write_locking(struct file *file, const char /*__user*/ *buf, size_t size, loff_t *off, loff_t offFile,
1823 struct inode *inode, struct vbsf_inode_info *sf_i,
1824 struct vbsf_super_info *pSuperInfo, struct vbsf_reg_info *sf_r)
1825{
1826 /*
1827 * Lock pages and execute the write, taking care not to pass the host
1828 * more than it can handle in one go or more than we care to allocate
1829 * page arrays for. The latter limit is set at just short of 32KB due
1830 * to how the physical heap works.
1831 */
1832 struct page *apPagesStack[16];
1833 struct page **papPages = &apPagesStack[0];
1834 struct page **papPagesFree = NULL;
1835 VBOXSFWRITEPGLSTREQ *pReq;
1836 ssize_t cbRet = -ENOMEM;
1837 size_t cPages = (((uintptr_t)buf & PAGE_OFFSET_MASK) + size + PAGE_OFFSET_MASK) >> PAGE_SHIFT;
1838 size_t cMaxPages = RT_MIN(RT_MAX(pSuperInfo->cMaxIoPages, 1), cPages);
1839 bool fLockPgHack;
1840
1841 pReq = (VBOXSFWRITEPGLSTREQ *)VbglR0PhysHeapAlloc(RT_UOFFSETOF_DYN(VBOXSFWRITEPGLSTREQ, PgLst.aPages[cMaxPages]));
1842 while (!pReq && cMaxPages > 4) {
1843 cMaxPages /= 2;
1844 pReq = (VBOXSFWRITEPGLSTREQ *)VbglR0PhysHeapAlloc(RT_UOFFSETOF_DYN(VBOXSFWRITEPGLSTREQ, PgLst.aPages[cMaxPages]));
1845 }
1846 if (pReq && cMaxPages > RT_ELEMENTS(apPagesStack))
1847 papPagesFree = papPages = kmalloc(cMaxPages * sizeof(sizeof(papPages[0])), GFP_KERNEL);
1848 if (pReq && papPages) {
1849 cbRet = 0;
1850 for (;;) {
1851 /*
1852 * Figure out how much to process now and lock the user pages.
1853 */
1854 int rc;
1855 size_t cbChunk = (uintptr_t)buf & PAGE_OFFSET_MASK;
1856 pReq->PgLst.offFirstPage = (uint16_t)cbChunk;
1857 cPages = RT_ALIGN_Z(cbChunk + size, PAGE_SIZE) >> PAGE_SHIFT;
1858 if (cPages <= cMaxPages)
1859 cbChunk = size;
1860 else {
1861 cPages = cMaxPages;
1862 cbChunk = (cMaxPages << PAGE_SHIFT) - cbChunk;
1863 }
1864
1865 rc = vbsf_lock_user_pages((uintptr_t)buf, cPages, false /*fWrite*/, papPages, &fLockPgHack);
1866 if (rc == 0) {
1867 size_t iPage = cPages;
1868 while (iPage-- > 0)
1869 pReq->PgLst.aPages[iPage] = page_to_phys(papPages[iPage]);
1870 } else {
1871 cbRet = rc;
1872 break;
1873 }
1874
1875 /*
1876 * Issue the request and unlock the pages.
1877 */
1878 rc = VbglR0SfHostReqWritePgLst(pSuperInfo->map.root, pReq, sf_r->Handle.hHost, offFile, cbChunk, cPages);
1879 sf_i->ModificationTimeAtOurLastWrite = sf_i->ModificationTime;
1880 if (RT_SUCCESS(rc)) {
1881 /*
1882 * Success, advance position and buffer.
1883 */
1884 uint32_t cbActual = pReq->Parms.cb32Write.u.value32;
1885 AssertStmt(cbActual <= cbChunk, cbActual = cbChunk);
1886
1887 vbsf_reg_write_sync_page_cache(inode->i_mapping, offFile, cbActual, NULL /*pbKrnlBuf*/,
1888 papPages, (uintptr_t)buf & PAGE_OFFSET_MASK, cPages);
1889 Assert(cPages <= cMaxPages);
1890 vbsf_unlock_user_pages(papPages, cPages, false /*fSetDirty*/, fLockPgHack);
1891
1892 cbRet += cbActual;
1893 buf = (uint8_t *)buf + cbActual;
1894 size -= cbActual;
1895
1896 offFile += cbActual;
1897 if ((file->f_flags & O_APPEND) && (g_fSfFeatures & SHFL_FEATURE_WRITE_UPDATES_OFFSET))
1898 offFile = pReq->Parms.off64Write.u.value64;
1899 if (offFile > i_size_read(inode))
1900 i_size_write(inode, offFile);
1901
1902 sf_i->force_restat = 1; /* mtime (and size) may have changed */
1903
1904 /*
1905 * Are we done already? If so commit the new file offset.
1906 */
1907 if (!size || cbActual < cbChunk) {
1908 *off = offFile;
1909 break;
1910 }
1911 } else {
1912 vbsf_unlock_user_pages(papPages, cPages, false /*fSetDirty*/, fLockPgHack);
1913 if (rc == VERR_NO_MEMORY && cMaxPages > 4) {
1914 /*
1915 * The host probably doesn't have enough heap to handle the
1916 * request, reduce the page count and retry.
1917 */
1918 cMaxPages /= 4;
1919 Assert(cMaxPages > 0);
1920 } else {
1921 /*
1922 * If we've successfully written stuff, return it rather than
1923 * the error. (Not sure if this is such a great idea...)
1924 */
1925 if (cbRet > 0) {
1926 SFLOGFLOW(("vbsf_reg_write: write at %#RX64 -> %Rrc; got cbRet=%#zx already\n", offFile, rc, cbRet));
1927 *off = offFile;
1928 } else {
1929 SFLOGFLOW(("vbsf_reg_write: write at %#RX64 -> %Rrc\n", offFile, rc));
1930 cbRet = -EPROTO;
1931 }
1932 break;
1933 }
1934 }
1935 }
1936 }
1937 if (papPagesFree)
1938 kfree(papPages);
1939 if (pReq)
1940 VbglR0PhysHeapFree(pReq);
1941 SFLOGFLOW(("vbsf_reg_write: returns %zd (%#zx), *off=%RX64 [lock]\n", cbRet, cbRet, *off));
1942 return cbRet;
1943}
1944
1945/**
1946 * Write to a regular file.
1947 *
1948 * @param file the file
1949 * @param buf the buffer
1950 * @param size length of the buffer
1951 * @param off offset within the file
1952 * @returns the number of written bytes on success, Linux error code otherwise
1953 */
1954static ssize_t vbsf_reg_write(struct file *file, const char *buf, size_t size, loff_t * off)
1955{
1956 struct inode *inode = VBSF_GET_F_DENTRY(file)->d_inode;
1957 struct vbsf_inode_info *sf_i = VBSF_GET_INODE_INFO(inode);
1958 struct vbsf_super_info *pSuperInfo = VBSF_GET_SUPER_INFO(inode->i_sb);
1959 struct vbsf_reg_info *sf_r = file->private_data;
1960 struct address_space *mapping = inode->i_mapping;
1961 loff_t pos;
1962
1963 SFLOGFLOW(("vbsf_reg_write: inode=%p file=%p buf=%p size=%#zx off=%#llx\n", inode, file, buf, size, *off));
1964 Assert(sf_i);
1965 Assert(pSuperInfo);
1966 Assert(sf_r);
1967 AssertReturn(S_ISREG(inode->i_mode), -EINVAL);
1968
1969 pos = *off;
1970 if (file->f_flags & O_APPEND)
1971 pos = i_size_read(inode);
1972
1973 /** @todo XXX Check write permission according to inode->i_mode! */
1974
1975 if (!size) {
1976 if (file->f_flags & O_APPEND) /** @todo check if this is the consensus behavior... */
1977 *off = pos;
1978 return 0;
1979 }
1980
1981 /** @todo Implement the read-write caching mode. */
1982
1983 /*
1984 * If there are active writable mappings, coordinate with any
1985 * pending writes via those.
1986 */
1987 if ( mapping
1988 && mapping->nrpages > 0
1989 && mapping_writably_mapped(mapping)) {
1990# if RTLNX_VER_MIN(2,6,32)
1991 int err = filemap_fdatawait_range(mapping, pos, pos + size - 1);
1992 if (err)
1993 return err;
1994# else
1995 /** @todo ... */
1996# endif
1997 }
1998
1999 /*
2000 * For small requests, try use an embedded buffer provided we get a heap block
2001 * that does not cross page boundraries (see host code).
2002 */
2003 if (size <= PAGE_SIZE / 4 * 3 - RT_UOFFSETOF(VBOXSFWRITEEMBEDDEDREQ, abData[0]) /* see allocator */) {
2004 uint32_t const cbReq = RT_UOFFSETOF(VBOXSFWRITEEMBEDDEDREQ, abData[0]) + size;
2005 VBOXSFWRITEEMBEDDEDREQ *pReq = (VBOXSFWRITEEMBEDDEDREQ *)VbglR0PhysHeapAlloc(cbReq);
2006 if ( pReq
2007 && (PAGE_SIZE - ((uintptr_t)pReq & PAGE_OFFSET_MASK)) >= cbReq) {
2008 ssize_t cbRet;
2009 if (copy_from_user(pReq->abData, buf, size) == 0) {
2010 int vrc = VbglR0SfHostReqWriteEmbedded(pSuperInfo->map.root, pReq, sf_r->Handle.hHost,
2011 pos, (uint32_t)size);
2012 sf_i->ModificationTimeAtOurLastWrite = sf_i->ModificationTime;
2013 if (RT_SUCCESS(vrc)) {
2014 cbRet = pReq->Parms.cb32Write.u.value32;
2015 AssertStmt(cbRet <= (ssize_t)size, cbRet = size);
2016 vbsf_reg_write_sync_page_cache(mapping, pos, (uint32_t)cbRet, pReq->abData,
2017 NULL /*papSrcPages*/, 0 /*offSrcPage0*/, 0 /*cSrcPages*/);
2018 pos += cbRet;
2019 if ((file->f_flags & O_APPEND) && (g_fSfFeatures & SHFL_FEATURE_WRITE_UPDATES_OFFSET))
2020 pos = pReq->Parms.off64Write.u.value64;
2021 *off = pos;
2022 if (pos > i_size_read(inode))
2023 i_size_write(inode, pos);
2024 } else
2025 cbRet = -EPROTO;
2026 sf_i->force_restat = 1; /* mtime (and size) may have changed */
2027 } else
2028 cbRet = -EFAULT;
2029
2030 VbglR0PhysHeapFree(pReq);
2031 SFLOGFLOW(("vbsf_reg_write: returns %zd (%#zx), *off=%RX64 [embed]\n", cbRet, cbRet, *off));
2032 return cbRet;
2033 }
2034 if (pReq)
2035 VbglR0PhysHeapFree(pReq);
2036 }
2037
2038# if 0 /* Turns out this is slightly slower than locking the pages even for 4KB reads (4.19/amd64). */
2039 /*
2040 * For medium sized requests try use a bounce buffer.
2041 */
2042 if (size <= _64K /** @todo make this configurable? */) {
2043 void *pvBounce = kmalloc(size, GFP_KERNEL);
2044 if (pvBounce) {
2045 if (copy_from_user(pvBounce, buf, size) == 0) {
2046 VBOXSFWRITEPGLSTREQ *pReq = (VBOXSFWRITEPGLSTREQ *)VbglR0PhysHeapAlloc(sizeof(*pReq));
2047 if (pReq) {
2048 ssize_t cbRet;
2049 int vrc = VbglR0SfHostReqWriteContig(pSuperInfo->map.root, pReq, sf_r->handle, pos,
2050 (uint32_t)size, pvBounce, virt_to_phys(pvBounce));
2051 sf_i->ModificationTimeAtOurLastWrite = sf_i->ModificationTime;
2052 if (RT_SUCCESS(vrc)) {
2053 cbRet = pReq->Parms.cb32Write.u.value32;
2054 AssertStmt(cbRet <= (ssize_t)size, cbRet = size);
2055 vbsf_reg_write_sync_page_cache(mapping, pos, (uint32_t)cbRet, (uint8_t const *)pvBounce,
2056 NULL /*papSrcPages*/, 0 /*offSrcPage0*/, 0 /*cSrcPages*/);
2057 pos += cbRet;
2058 *off = pos;
2059 if (pos > i_size_read(inode))
2060 i_size_write(inode, pos);
2061 } else
2062 cbRet = -EPROTO;
2063 sf_i->force_restat = 1; /* mtime (and size) may have changed */
2064 VbglR0PhysHeapFree(pReq);
2065 kfree(pvBounce);
2066 SFLOGFLOW(("vbsf_reg_write: returns %zd (%#zx), *off=%RX64 [bounce]\n", cbRet, cbRet, *off));
2067 return cbRet;
2068 }
2069 kfree(pvBounce);
2070 } else {
2071 kfree(pvBounce);
2072 SFLOGFLOW(("vbsf_reg_write: returns -EFAULT, *off=%RX64 [bounce]\n", *off));
2073 return -EFAULT;
2074 }
2075 }
2076 }
2077# endif
2078
2079 return vbsf_reg_write_locking(file, buf, size, off, pos, inode, sf_i, pSuperInfo, sf_r);
2080}
2081#endif /* < 5.10.0 */
2082
2083#if RTLNX_VER_MIN(2,6,19)
2084
2085/**
2086 * Companion to vbsf_iter_lock_pages().
2087 */
2088DECLINLINE(void) vbsf_iter_unlock_pages(struct iov_iter *iter, struct page **papPages, size_t cPages, bool fSetDirty)
2089{
2090 /* We don't mark kernel pages dirty (KVECs, BVECs, PIPEs): */
2091 if (!iter_is_iovec(iter))
2092 fSetDirty = false;
2093
2094 while (cPages-- > 0)
2095 {
2096 struct page *pPage = papPages[cPages];
2097 if (fSetDirty && !PageReserved(pPage))
2098 set_page_dirty(pPage);
2099 vbsf_put_page(pPage);
2100 }
2101}
2102
2103
2104/**
2105 * Locks up to @a cMaxPages from the I/O vector iterator, advancing the
2106 * iterator.
2107 *
2108 * @returns 0 on success, negative errno value on failure.
2109 * @param iter The iterator to lock pages from.
2110 * @param fWrite Whether to write (true) or read (false) lock the pages.
2111 * @param pStash Where we stash peek results.
2112 * @param cMaxPages The maximum number of pages to get.
2113 * @param papPages Where to return the locked pages.
2114 * @param pcPages Where to return the number of pages.
2115 * @param poffPage0 Where to return the offset into the first page.
2116 * @param pcbChunk Where to return the number of bytes covered.
2117 */
2118static int vbsf_iter_lock_pages(struct iov_iter *iter, bool fWrite, struct vbsf_iter_stash *pStash, size_t cMaxPages,
2119 struct page **papPages, size_t *pcPages, size_t *poffPage0, size_t *pcbChunk)
2120{
2121 size_t cbChunk = 0;
2122 size_t cPages = 0;
2123 size_t offPage0 = 0;
2124 int rc = 0;
2125
2126 Assert(iov_iter_count(iter) + pStash->cb > 0);
2127 if (!(iter->type & ITER_KVEC)) {
2128 /*
2129 * Do we have a stashed page?
2130 */
2131 if (pStash->pPage) {
2132 papPages[0] = pStash->pPage;
2133 offPage0 = pStash->off;
2134 cbChunk = pStash->cb;
2135 cPages = 1;
2136 pStash->pPage = NULL;
2137 pStash->off = 0;
2138 pStash->cb = 0;
2139 if ( offPage0 + cbChunk < PAGE_SIZE
2140 || iov_iter_count(iter) == 0) {
2141 *poffPage0 = offPage0;
2142 *pcbChunk = cbChunk;
2143 *pcPages = cPages;
2144 SFLOGFLOW(("vbsf_iter_lock_pages: returns %d - cPages=%#zx offPage0=%#zx cbChunk=%zx (stashed)\n",
2145 rc, cPages, offPage0, cbChunk));
2146 return 0;
2147 }
2148 cMaxPages -= 1;
2149 SFLOG3(("vbsf_iter_lock_pages: Picked up stashed page: %#zx LB %#zx\n", offPage0, cbChunk));
2150 } else {
2151# if RTLNX_VER_MAX(4,11,0)
2152 /*
2153 * Copy out our starting point to assist rewinding.
2154 */
2155 pStash->offFromEnd = iov_iter_count(iter);
2156 pStash->Copy = *iter;
2157# endif
2158 }
2159
2160 /*
2161 * Get pages segment by segment.
2162 */
2163 do {
2164 /*
2165 * Make a special case of the first time thru here, since that's
2166 * the most typical scenario.
2167 */
2168 ssize_t cbSegRet;
2169 if (cPages == 0) {
2170# if RTLNX_VER_MAX(3,19,0)
2171 while (!iov_iter_single_seg_count(iter)) /* Old code didn't skip empty segments which caused EFAULTs. */
2172 iov_iter_advance(iter, 0);
2173# endif
2174 cbSegRet = iov_iter_get_pages(iter, papPages, iov_iter_count(iter), cMaxPages, &offPage0);
2175 if (cbSegRet > 0) {
2176 iov_iter_advance(iter, cbSegRet);
2177 cbChunk = (size_t)cbSegRet;
2178 cPages = RT_ALIGN_Z(offPage0 + cbSegRet, PAGE_SIZE) >> PAGE_SHIFT;
2179 cMaxPages -= cPages;
2180 SFLOG3(("vbsf_iter_lock_pages: iov_iter_get_pages -> %#zx @ %#zx; %#zx pages [first]\n", cbSegRet, offPage0, cPages));
2181 if ( cMaxPages == 0
2182 || ((offPage0 + (size_t)cbSegRet) & PAGE_OFFSET_MASK))
2183 break;
2184 } else {
2185 AssertStmt(cbSegRet < 0, cbSegRet = -EFAULT);
2186 rc = (int)cbSegRet;
2187 break;
2188 }
2189 } else {
2190 /*
2191 * Probe first page of new segment to check that we've got a zero offset and
2192 * can continue on the current chunk. Stash the page if the offset isn't zero.
2193 */
2194 size_t offPgProbe;
2195 size_t cbSeg = iov_iter_single_seg_count(iter);
2196 while (!cbSeg) {
2197 iov_iter_advance(iter, 0);
2198 cbSeg = iov_iter_single_seg_count(iter);
2199 }
2200 cbSegRet = iov_iter_get_pages(iter, &papPages[cPages], iov_iter_count(iter), 1, &offPgProbe);
2201 if (cbSegRet > 0) {
2202 iov_iter_advance(iter, cbSegRet); /** @todo maybe not do this if we stash the page? */
2203 Assert(offPgProbe + cbSegRet <= PAGE_SIZE);
2204 if (offPgProbe == 0) {
2205 cbChunk += cbSegRet;
2206 cPages += 1;
2207 cMaxPages -= 1;
2208 SFLOG3(("vbsf_iter_lock_pages: iov_iter_get_pages(1) -> %#zx @ %#zx\n", cbSegRet, offPgProbe));
2209 if ( cMaxPages == 0
2210 || cbSegRet != PAGE_SIZE)
2211 break;
2212
2213 /*
2214 * Get the rest of the segment (if anything remaining).
2215 */
2216 cbSeg -= cbSegRet;
2217 if (cbSeg > 0) {
2218 cbSegRet = iov_iter_get_pages(iter, &papPages[cPages], iov_iter_count(iter), cMaxPages, &offPgProbe);
2219 if (cbSegRet > 0) {
2220 size_t const cPgRet = RT_ALIGN_Z((size_t)cbSegRet, PAGE_SIZE) >> PAGE_SHIFT;
2221 Assert(offPgProbe == 0);
2222 iov_iter_advance(iter, cbSegRet);
2223 SFLOG3(("vbsf_iter_lock_pages: iov_iter_get_pages() -> %#zx; %#zx pages\n", cbSegRet, cPgRet));
2224 cPages += cPgRet;
2225 cMaxPages -= cPgRet;
2226 cbChunk += cbSegRet;
2227 if ( cMaxPages == 0
2228 || ((size_t)cbSegRet & PAGE_OFFSET_MASK))
2229 break;
2230 } else {
2231 AssertStmt(cbSegRet < 0, cbSegRet = -EFAULT);
2232 rc = (int)cbSegRet;
2233 break;
2234 }
2235 }
2236 } else {
2237 /* The segment didn't start at a page boundrary, so stash it for
2238 the next round: */
2239 SFLOGFLOW(("vbsf_iter_lock_pages: iov_iter_get_pages(1) -> %#zx @ %#zx; stashed\n", cbSegRet, offPgProbe));
2240 Assert(papPages[cPages]);
2241 pStash->pPage = papPages[cPages];
2242 pStash->off = offPgProbe;
2243 pStash->cb = cbSegRet;
2244 break;
2245 }
2246 } else {
2247 AssertStmt(cbSegRet < 0, cbSegRet = -EFAULT);
2248 rc = (int)cbSegRet;
2249 break;
2250 }
2251 }
2252 Assert(cMaxPages > 0);
2253 } while (iov_iter_count(iter) > 0);
2254
2255 } else {
2256 /*
2257 * The silly iov_iter_get_pages_alloc() function doesn't handle KVECs,
2258 * so everyone needs to do that by themselves.
2259 *
2260 * Note! Fixes here may apply to rtR0MemObjNativeLockKernel()
2261 * and vbsf_lock_user_pages_failed_check_kernel() as well.
2262 */
2263# if RTLNX_VER_MAX(4,11,0)
2264 pStash->offFromEnd = iov_iter_count(iter);
2265 pStash->Copy = *iter;
2266# endif
2267 do {
2268 uint8_t *pbBuf;
2269 size_t offStart;
2270 size_t cPgSeg;
2271
2272 size_t cbSeg = iov_iter_single_seg_count(iter);
2273 while (!cbSeg) {
2274 iov_iter_advance(iter, 0);
2275 cbSeg = iov_iter_single_seg_count(iter);
2276 }
2277
2278# if RTLNX_VER_MIN(3,19,0)
2279 pbBuf = iter->kvec->iov_base + iter->iov_offset;
2280# else
2281 pbBuf = iter->iov->iov_base + iter->iov_offset;
2282# endif
2283 offStart = (uintptr_t)pbBuf & PAGE_OFFSET_MASK;
2284 if (!cPages)
2285 offPage0 = offStart;
2286 else if (offStart)
2287 break;
2288
2289 cPgSeg = RT_ALIGN_Z(cbSeg, PAGE_SIZE) >> PAGE_SHIFT;
2290 if (cPgSeg > cMaxPages) {
2291 cPgSeg = cMaxPages;
2292 cbSeg = (cPgSeg << PAGE_SHIFT) - offStart;
2293 }
2294
2295 rc = vbsf_lock_kernel_pages(pbBuf, fWrite, cPgSeg, &papPages[cPages]);
2296 if (rc == 0) {
2297 iov_iter_advance(iter, cbSeg);
2298 cbChunk += cbSeg;
2299 cPages += cPgSeg;
2300 cMaxPages -= cPgSeg;
2301 if ( cMaxPages == 0
2302 || ((offStart + cbSeg) & PAGE_OFFSET_MASK) != 0)
2303 break;
2304 } else
2305 break;
2306 } while (iov_iter_count(iter) > 0);
2307 }
2308
2309 /*
2310 * Clean up if we failed; set return values.
2311 */
2312 if (rc == 0) {
2313 /* likely */
2314 } else {
2315 if (cPages > 0)
2316 vbsf_iter_unlock_pages(iter, papPages, cPages, false /*fSetDirty*/);
2317 offPage0 = cbChunk = cPages = 0;
2318 }
2319 *poffPage0 = offPage0;
2320 *pcbChunk = cbChunk;
2321 *pcPages = cPages;
2322 SFLOGFLOW(("vbsf_iter_lock_pages: returns %d - cPages=%#zx offPage0=%#zx cbChunk=%zx\n", rc, cPages, offPage0, cbChunk));
2323 return rc;
2324}
2325
2326
2327/**
2328 * Rewinds the I/O vector.
2329 */
2330static bool vbsf_iter_rewind(struct iov_iter *iter, struct vbsf_iter_stash *pStash, size_t cbToRewind, size_t cbChunk)
2331{
2332 size_t cbExtra;
2333 if (!pStash->pPage) {
2334 cbExtra = 0;
2335 } else {
2336 cbExtra = pStash->cb;
2337 vbsf_put_page(pStash->pPage);
2338 pStash->pPage = NULL;
2339 pStash->cb = 0;
2340 pStash->off = 0;
2341 }
2342
2343# if RTLNX_VER_MIN(4,11,0) || RTLNX_VER_MAX(3,16,0)
2344 iov_iter_revert(iter, cbToRewind + cbExtra);
2345 return true;
2346# else
2347 /** @todo impl this */
2348 return false;
2349# endif
2350}
2351
2352
2353/**
2354 * Cleans up the page locking stash.
2355 */
2356DECLINLINE(void) vbsf_iter_cleanup_stash(struct iov_iter *iter, struct vbsf_iter_stash *pStash)
2357{
2358 if (pStash->pPage)
2359 vbsf_iter_rewind(iter, pStash, 0, 0);
2360}
2361
2362
2363/**
2364 * Calculates the longest span of pages we could transfer to the host in a
2365 * single request.
2366 *
2367 * @returns Page count, non-zero.
2368 * @param iter The I/O vector iterator to inspect.
2369 */
2370static size_t vbsf_iter_max_span_of_pages(struct iov_iter *iter)
2371{
2372 size_t cPages;
2373# if RTLNX_VER_MIN(3,16,0)
2374 if (iter_is_iovec(iter) || (iter->type & ITER_KVEC)) {
2375#endif
2376 const struct iovec *pCurIov = iter->iov;
2377 size_t cLeft = iter->nr_segs;
2378 size_t cPagesSpan = 0;
2379
2380 /* iovect and kvec are identical, except for the __user tagging of iov_base. */
2381 AssertCompileMembersSameSizeAndOffset(struct iovec, iov_base, struct kvec, iov_base);
2382 AssertCompileMembersSameSizeAndOffset(struct iovec, iov_len, struct kvec, iov_len);
2383 AssertCompile(sizeof(struct iovec) == sizeof(struct kvec));
2384
2385 cPages = 1;
2386 AssertReturn(cLeft > 0, cPages);
2387
2388 /* Special case: segment offset. */
2389 if (iter->iov_offset > 0) {
2390 if (iter->iov_offset < pCurIov->iov_len) {
2391 size_t const cbSegLeft = pCurIov->iov_len - iter->iov_offset;
2392 size_t const offPage0 = ((uintptr_t)pCurIov->iov_base + iter->iov_offset) & PAGE_OFFSET_MASK;
2393 cPages = cPagesSpan = RT_ALIGN_Z(offPage0 + cbSegLeft, PAGE_SIZE) >> PAGE_SHIFT;
2394 if ((offPage0 + cbSegLeft) & PAGE_OFFSET_MASK)
2395 cPagesSpan = 0;
2396 }
2397 SFLOGFLOW(("vbsf_iter: seg[0]= %p LB %#zx\n", pCurIov->iov_base, pCurIov->iov_len));
2398 pCurIov++;
2399 cLeft--;
2400 }
2401
2402 /* Full segments. */
2403 while (cLeft-- > 0) {
2404 if (pCurIov->iov_len > 0) {
2405 size_t const offPage0 = (uintptr_t)pCurIov->iov_base & PAGE_OFFSET_MASK;
2406 if (offPage0 == 0) {
2407 if (!(pCurIov->iov_len & PAGE_OFFSET_MASK)) {
2408 cPagesSpan += pCurIov->iov_len >> PAGE_SHIFT;
2409 } else {
2410 cPagesSpan += RT_ALIGN_Z(pCurIov->iov_len, PAGE_SIZE) >> PAGE_SHIFT;
2411 if (cPagesSpan > cPages)
2412 cPages = cPagesSpan;
2413 cPagesSpan = 0;
2414 }
2415 } else {
2416 if (cPagesSpan > cPages)
2417 cPages = cPagesSpan;
2418 if (!((offPage0 + pCurIov->iov_len) & PAGE_OFFSET_MASK)) {
2419 cPagesSpan = pCurIov->iov_len >> PAGE_SHIFT;
2420 } else {
2421 cPagesSpan += RT_ALIGN_Z(offPage0 + pCurIov->iov_len, PAGE_SIZE) >> PAGE_SHIFT;
2422 if (cPagesSpan > cPages)
2423 cPages = cPagesSpan;
2424 cPagesSpan = 0;
2425 }
2426 }
2427 }
2428 SFLOGFLOW(("vbsf_iter: seg[%u]= %p LB %#zx\n", iter->nr_segs - cLeft, pCurIov->iov_base, pCurIov->iov_len));
2429 pCurIov++;
2430 }
2431 if (cPagesSpan > cPages)
2432 cPages = cPagesSpan;
2433# if RTLNX_VER_MIN(3,16,0)
2434 } else {
2435 /* Won't bother with accurate counts for the next two types, just make
2436 some rough estimates (does pipes have segments?): */
2437 size_t cSegs = iter->type & ITER_BVEC ? RT_MAX(1, iter->nr_segs) : 1;
2438 cPages = (iov_iter_count(iter) + (PAGE_SIZE * 2 - 2) * cSegs) >> PAGE_SHIFT;
2439 }
2440# endif
2441 SFLOGFLOW(("vbsf_iter_max_span_of_pages: returns %#zx\n", cPages));
2442 return cPages;
2443}
2444
2445
2446/**
2447 * Worker for vbsf_reg_read_iter() that deals with larger reads using page
2448 * locking.
2449 */
2450static ssize_t vbsf_reg_read_iter_locking(struct kiocb *kio, struct iov_iter *iter, size_t cbToRead,
2451 struct vbsf_super_info *pSuperInfo, struct vbsf_reg_info *sf_r)
2452{
2453 /*
2454 * Estimate how many pages we may possible submit in a single request so
2455 * that we can allocate matching request buffer and page array.
2456 */
2457 struct page *apPagesStack[16];
2458 struct page **papPages = &apPagesStack[0];
2459 struct page **papPagesFree = NULL;
2460 VBOXSFREADPGLSTREQ *pReq;
2461 ssize_t cbRet = 0;
2462 size_t cMaxPages = vbsf_iter_max_span_of_pages(iter);
2463 cMaxPages = RT_MIN(RT_MAX(pSuperInfo->cMaxIoPages, 2), cMaxPages);
2464
2465 pReq = (VBOXSFREADPGLSTREQ *)VbglR0PhysHeapAlloc(RT_UOFFSETOF_DYN(VBOXSFREADPGLSTREQ, PgLst.aPages[cMaxPages]));
2466 while (!pReq && cMaxPages > 4) {
2467 cMaxPages /= 2;
2468 pReq = (VBOXSFREADPGLSTREQ *)VbglR0PhysHeapAlloc(RT_UOFFSETOF_DYN(VBOXSFREADPGLSTREQ, PgLst.aPages[cMaxPages]));
2469 }
2470 if (pReq && cMaxPages > RT_ELEMENTS(apPagesStack))
2471 papPagesFree = papPages = kmalloc(cMaxPages * sizeof(sizeof(papPages[0])), GFP_KERNEL);
2472 if (pReq && papPages) {
2473
2474 /*
2475 * The read loop.
2476 */
2477 struct vbsf_iter_stash Stash = VBSF_ITER_STASH_INITIALIZER;
2478 do {
2479 /*
2480 * Grab as many pages as we can. This means that if adjacent
2481 * segments both starts and ends at a page boundrary, we can
2482 * do them both in the same transfer from the host.
2483 */
2484 size_t cPages = 0;
2485 size_t cbChunk = 0;
2486 size_t offPage0 = 0;
2487 int rc = vbsf_iter_lock_pages(iter, true /*fWrite*/, &Stash, cMaxPages, papPages, &cPages, &offPage0, &cbChunk);
2488 if (rc == 0) {
2489 size_t iPage = cPages;
2490 while (iPage-- > 0)
2491 pReq->PgLst.aPages[iPage] = page_to_phys(papPages[iPage]);
2492 pReq->PgLst.offFirstPage = (uint16_t)offPage0;
2493 AssertStmt(cbChunk <= cbToRead, cbChunk = cbToRead);
2494 } else {
2495 cbRet = rc;
2496 break;
2497 }
2498
2499 /*
2500 * Issue the request and unlock the pages.
2501 */
2502 rc = VbglR0SfHostReqReadPgLst(pSuperInfo->map.root, pReq, sf_r->Handle.hHost, kio->ki_pos, cbChunk, cPages);
2503 SFLOGFLOW(("vbsf_reg_read_iter_locking: VbglR0SfHostReqReadPgLst -> %d (cbActual=%#x cbChunk=%#zx of %#zx cPages=%#zx offPage0=%#x\n",
2504 rc, pReq->Parms.cb32Read.u.value32, cbChunk, cbToRead, cPages, offPage0));
2505
2506 vbsf_iter_unlock_pages(iter, papPages, cPages, true /*fSetDirty*/);
2507
2508 if (RT_SUCCESS(rc)) {
2509 /*
2510 * Success, advance position and buffer.
2511 */
2512 uint32_t cbActual = pReq->Parms.cb32Read.u.value32;
2513 AssertStmt(cbActual <= cbChunk, cbActual = cbChunk);
2514 cbRet += cbActual;
2515 kio->ki_pos += cbActual;
2516 cbToRead -= cbActual;
2517
2518 /*
2519 * Are we done already?
2520 */
2521 if (!cbToRead)
2522 break;
2523 if (cbActual < cbChunk) { /* We ASSUME end-of-file here. */
2524 if (vbsf_iter_rewind(iter, &Stash, cbChunk - cbActual, cbActual))
2525 iov_iter_truncate(iter, 0);
2526 break;
2527 }
2528 } else {
2529 /*
2530 * Try rewind the iter structure.
2531 */
2532 bool const fRewindOkay = vbsf_iter_rewind(iter, &Stash, cbChunk, cbChunk);
2533 if (rc == VERR_NO_MEMORY && cMaxPages > 4 && fRewindOkay) {
2534 /*
2535 * The host probably doesn't have enough heap to handle the
2536 * request, reduce the page count and retry.
2537 */
2538 cMaxPages /= 4;
2539 Assert(cMaxPages > 0);
2540 } else {
2541 /*
2542 * If we've successfully read stuff, return it rather than
2543 * the error. (Not sure if this is such a great idea...)
2544 */
2545 if (cbRet <= 0)
2546 cbRet = -EPROTO;
2547 break;
2548 }
2549 }
2550 } while (cbToRead > 0);
2551
2552 vbsf_iter_cleanup_stash(iter, &Stash);
2553 }
2554 else
2555 cbRet = -ENOMEM;
2556 if (papPagesFree)
2557 kfree(papPages);
2558 if (pReq)
2559 VbglR0PhysHeapFree(pReq);
2560 SFLOGFLOW(("vbsf_reg_read_iter_locking: returns %#zx (%zd)\n", cbRet, cbRet));
2561 return cbRet;
2562}
2563
2564
2565/**
2566 * Read into I/O vector iterator.
2567 *
2568 * @returns Number of bytes read on success, negative errno on error.
2569 * @param kio The kernel I/O control block (or something like that).
2570 * @param iter The I/O vector iterator describing the buffer.
2571 */
2572# if RTLNX_VER_MIN(3,16,0)
2573static ssize_t vbsf_reg_read_iter(struct kiocb *kio, struct iov_iter *iter)
2574# else
2575static ssize_t vbsf_reg_aio_read(struct kiocb *kio, const struct iovec *iov, unsigned long cSegs, loff_t offFile)
2576# endif
2577{
2578# if RTLNX_VER_MAX(3,16,0)
2579 struct vbsf_iov_iter fake_iter = VBSF_IOV_ITER_INITIALIZER(cSegs, iov, 0 /*write*/);
2580 struct vbsf_iov_iter *iter = &fake_iter;
2581# endif
2582 size_t cbToRead = iov_iter_count(iter);
2583 struct inode *inode = VBSF_GET_F_DENTRY(kio->ki_filp)->d_inode;
2584 struct address_space *mapping = inode->i_mapping;
2585
2586 struct vbsf_reg_info *sf_r = kio->ki_filp->private_data;
2587 struct vbsf_super_info *pSuperInfo = VBSF_GET_SUPER_INFO(inode->i_sb);
2588
2589 SFLOGFLOW(("vbsf_reg_read_iter: inode=%p file=%p size=%#zx off=%#llx type=%#x\n",
2590 inode, kio->ki_filp, cbToRead, kio->ki_pos, iter->type));
2591 AssertReturn(S_ISREG(inode->i_mode), -EINVAL);
2592
2593 /*
2594 * Do we have anything at all to do here?
2595 */
2596 if (!cbToRead)
2597 return 0;
2598
2599 /*
2600 * If there is a mapping and O_DIRECT isn't in effect, we must at a
2601 * heed dirty pages in the mapping and read from them. For simplicity
2602 * though, we just do page cache reading when there are writable
2603 * mappings around with any kind of pages loaded.
2604 */
2605 if (vbsf_should_use_cached_read(kio->ki_filp, mapping, pSuperInfo)) {
2606# if RTLNX_VER_MIN(3,16,0)
2607 return generic_file_read_iter(kio, iter);
2608# else
2609 return generic_file_aio_read(kio, iov, cSegs, offFile);
2610# endif
2611 }
2612
2613 /*
2614 * Now now we reject async I/O requests.
2615 */
2616 if (!is_sync_kiocb(kio)) {
2617 SFLOGFLOW(("vbsf_reg_read_iter: async I/O not yet supported\n")); /** @todo extend FsPerf with AIO tests. */
2618 return -EOPNOTSUPP;
2619 }
2620
2621 /*
2622 * For small requests, try use an embedded buffer provided we get a heap block
2623 * that does not cross page boundraries (see host code).
2624 */
2625 if (cbToRead <= PAGE_SIZE / 4 * 3 - RT_UOFFSETOF(VBOXSFREADEMBEDDEDREQ, abData[0]) /* see allocator */) {
2626 uint32_t const cbReq = RT_UOFFSETOF(VBOXSFREADEMBEDDEDREQ, abData[0]) + cbToRead;
2627 VBOXSFREADEMBEDDEDREQ *pReq = (VBOXSFREADEMBEDDEDREQ *)VbglR0PhysHeapAlloc(cbReq);
2628 if (pReq) {
2629 if ((PAGE_SIZE - ((uintptr_t)pReq & PAGE_OFFSET_MASK)) >= cbReq) {
2630 ssize_t cbRet;
2631 int vrc = VbglR0SfHostReqReadEmbedded(pSuperInfo->map.root, pReq, sf_r->Handle.hHost,
2632 kio->ki_pos, (uint32_t)cbToRead);
2633 if (RT_SUCCESS(vrc)) {
2634 cbRet = pReq->Parms.cb32Read.u.value32;
2635 AssertStmt(cbRet <= (ssize_t)cbToRead, cbRet = cbToRead);
2636 if (copy_to_iter(pReq->abData, cbRet, iter) == cbRet) {
2637 kio->ki_pos += cbRet;
2638 if (cbRet < cbToRead)
2639 iov_iter_truncate(iter, 0);
2640 } else
2641 cbRet = -EFAULT;
2642 } else
2643 cbRet = -EPROTO;
2644 VbglR0PhysHeapFree(pReq);
2645 SFLOGFLOW(("vbsf_reg_read_iter: returns %#zx (%zd)\n", cbRet, cbRet));
2646 return cbRet;
2647 }
2648 VbglR0PhysHeapFree(pReq);
2649 }
2650 }
2651
2652 /*
2653 * Otherwise do the page locking thing.
2654 */
2655 return vbsf_reg_read_iter_locking(kio, iter, cbToRead, pSuperInfo, sf_r);
2656}
2657
2658
2659/**
2660 * Worker for vbsf_reg_write_iter() that deals with larger writes using page
2661 * locking.
2662 */
2663static ssize_t vbsf_reg_write_iter_locking(struct kiocb *kio, struct iov_iter *iter, size_t cbToWrite, loff_t offFile,
2664 struct vbsf_super_info *pSuperInfo, struct vbsf_reg_info *sf_r, struct inode *inode,
2665 struct vbsf_inode_info *sf_i, struct address_space *mapping, bool fAppend)
2666{
2667 /*
2668 * Estimate how many pages we may possible submit in a single request so
2669 * that we can allocate matching request buffer and page array.
2670 */
2671 struct page *apPagesStack[16];
2672 struct page **papPages = &apPagesStack[0];
2673 struct page **papPagesFree = NULL;
2674 VBOXSFWRITEPGLSTREQ *pReq;
2675 ssize_t cbRet = 0;
2676 size_t cMaxPages = vbsf_iter_max_span_of_pages(iter);
2677 cMaxPages = RT_MIN(RT_MAX(pSuperInfo->cMaxIoPages, 2), cMaxPages);
2678
2679 pReq = (VBOXSFWRITEPGLSTREQ *)VbglR0PhysHeapAlloc(RT_UOFFSETOF_DYN(VBOXSFWRITEPGLSTREQ, PgLst.aPages[cMaxPages]));
2680 while (!pReq && cMaxPages > 4) {
2681 cMaxPages /= 2;
2682 pReq = (VBOXSFWRITEPGLSTREQ *)VbglR0PhysHeapAlloc(RT_UOFFSETOF_DYN(VBOXSFWRITEPGLSTREQ, PgLst.aPages[cMaxPages]));
2683 }
2684 if (pReq && cMaxPages > RT_ELEMENTS(apPagesStack))
2685 papPagesFree = papPages = kmalloc(cMaxPages * sizeof(sizeof(papPages[0])), GFP_KERNEL);
2686 if (pReq && papPages) {
2687
2688 /*
2689 * The write loop.
2690 */
2691 struct vbsf_iter_stash Stash = VBSF_ITER_STASH_INITIALIZER;
2692 do {
2693 /*
2694 * Grab as many pages as we can. This means that if adjacent
2695 * segments both starts and ends at a page boundrary, we can
2696 * do them both in the same transfer from the host.
2697 */
2698 size_t cPages = 0;
2699 size_t cbChunk = 0;
2700 size_t offPage0 = 0;
2701 int rc = vbsf_iter_lock_pages(iter, false /*fWrite*/, &Stash, cMaxPages, papPages, &cPages, &offPage0, &cbChunk);
2702 if (rc == 0) {
2703 size_t iPage = cPages;
2704 while (iPage-- > 0)
2705 pReq->PgLst.aPages[iPage] = page_to_phys(papPages[iPage]);
2706 pReq->PgLst.offFirstPage = (uint16_t)offPage0;
2707 AssertStmt(cbChunk <= cbToWrite, cbChunk = cbToWrite);
2708 } else {
2709 cbRet = rc;
2710 break;
2711 }
2712
2713 /*
2714 * Issue the request and unlock the pages.
2715 */
2716 rc = VbglR0SfHostReqWritePgLst(pSuperInfo->map.root, pReq, sf_r->Handle.hHost, offFile, cbChunk, cPages);
2717 sf_i->ModificationTimeAtOurLastWrite = sf_i->ModificationTime;
2718 SFLOGFLOW(("vbsf_reg_write_iter_locking: VbglR0SfHostReqWritePgLst -> %d (cbActual=%#x cbChunk=%#zx of %#zx cPages=%#zx offPage0=%#x\n",
2719 rc, pReq->Parms.cb32Write.u.value32, cbChunk, cbToWrite, cPages, offPage0));
2720 if (RT_SUCCESS(rc)) {
2721 /*
2722 * Success, advance position and buffer.
2723 */
2724 uint32_t cbActual = pReq->Parms.cb32Write.u.value32;
2725 AssertStmt(cbActual <= cbChunk, cbActual = cbChunk);
2726
2727 vbsf_reg_write_sync_page_cache(mapping, offFile, cbActual, NULL /*pbSrcBuf*/, papPages, offPage0, cPages);
2728 vbsf_iter_unlock_pages(iter, papPages, cPages, false /*fSetDirty*/);
2729
2730 cbRet += cbActual;
2731 cbToWrite -= cbActual;
2732
2733 offFile += cbActual;
2734 if (fAppend && (g_fSfFeatures & SHFL_FEATURE_WRITE_UPDATES_OFFSET))
2735 offFile = pReq->Parms.off64Write.u.value64;
2736 kio->ki_pos = offFile;
2737 if (offFile > i_size_read(inode))
2738 i_size_write(inode, offFile);
2739
2740 sf_i->force_restat = 1; /* mtime (and size) may have changed */
2741
2742 /*
2743 * Are we done already?
2744 */
2745 if (!cbToWrite)
2746 break;
2747 if (cbActual < cbChunk) { /* We ASSUME end-of-file here. */
2748 if (vbsf_iter_rewind(iter, &Stash, cbChunk - cbActual, cbActual))
2749 iov_iter_truncate(iter, 0);
2750 break;
2751 }
2752 } else {
2753 /*
2754 * Try rewind the iter structure.
2755 */
2756 bool fRewindOkay;
2757 vbsf_iter_unlock_pages(iter, papPages, cPages, false /*fSetDirty*/);
2758 fRewindOkay = vbsf_iter_rewind(iter, &Stash, cbChunk, cbChunk);
2759 if (rc == VERR_NO_MEMORY && cMaxPages > 4 && fRewindOkay) {
2760 /*
2761 * The host probably doesn't have enough heap to handle the
2762 * request, reduce the page count and retry.
2763 */
2764 cMaxPages /= 4;
2765 Assert(cMaxPages > 0);
2766 } else {
2767 /*
2768 * If we've successfully written stuff, return it rather than
2769 * the error. (Not sure if this is such a great idea...)
2770 */
2771 if (cbRet <= 0)
2772 cbRet = -EPROTO;
2773 break;
2774 }
2775 }
2776 } while (cbToWrite > 0);
2777
2778 vbsf_iter_cleanup_stash(iter, &Stash);
2779 }
2780 else
2781 cbRet = -ENOMEM;
2782 if (papPagesFree)
2783 kfree(papPages);
2784 if (pReq)
2785 VbglR0PhysHeapFree(pReq);
2786 SFLOGFLOW(("vbsf_reg_write_iter_locking: returns %#zx (%zd)\n", cbRet, cbRet));
2787 return cbRet;
2788}
2789
2790
2791/**
2792 * Write from I/O vector iterator.
2793 *
2794 * @returns Number of bytes written on success, negative errno on error.
2795 * @param kio The kernel I/O control block (or something like that).
2796 * @param iter The I/O vector iterator describing the buffer.
2797 */
2798# if RTLNX_VER_MIN(3,16,0)
2799static ssize_t vbsf_reg_write_iter(struct kiocb *kio, struct iov_iter *iter)
2800# else
2801static ssize_t vbsf_reg_aio_write(struct kiocb *kio, const struct iovec *iov, unsigned long cSegs, loff_t offFile)
2802# endif
2803{
2804# if RTLNX_VER_MAX(3,16,0)
2805 struct vbsf_iov_iter fake_iter = VBSF_IOV_ITER_INITIALIZER(cSegs, iov, 1 /*write*/);
2806 struct vbsf_iov_iter *iter = &fake_iter;
2807# endif
2808 size_t cbToWrite = iov_iter_count(iter);
2809 struct inode *inode = VBSF_GET_F_DENTRY(kio->ki_filp)->d_inode;
2810 struct vbsf_inode_info *sf_i = VBSF_GET_INODE_INFO(inode);
2811 struct address_space *mapping = inode->i_mapping;
2812
2813 struct vbsf_reg_info *sf_r = kio->ki_filp->private_data;
2814 struct vbsf_super_info *pSuperInfo = VBSF_GET_SUPER_INFO(inode->i_sb);
2815# if RTLNX_VER_MIN(3,16,0)
2816 loff_t offFile = kio->ki_pos;
2817# endif
2818# if RTLNX_VER_MIN(4,1,0)
2819 bool const fAppend = RT_BOOL(kio->ki_flags & IOCB_APPEND);
2820# else
2821 bool const fAppend = RT_BOOL(kio->ki_filp->f_flags & O_APPEND);
2822# endif
2823
2824
2825 SFLOGFLOW(("vbsf_reg_write_iter: inode=%p file=%p size=%#zx off=%#llx type=%#x\n",
2826 inode, kio->ki_filp, cbToWrite, offFile, iter->type));
2827 AssertReturn(S_ISREG(inode->i_mode), -EINVAL);
2828
2829 /*
2830 * Enforce APPEND flag (more later).
2831 */
2832 if (fAppend)
2833 kio->ki_pos = offFile = i_size_read(inode);
2834
2835 /*
2836 * Do we have anything at all to do here?
2837 */
2838 if (!cbToWrite)
2839 return 0;
2840
2841 /** @todo Implement the read-write caching mode. */
2842
2843 /*
2844 * Now now we reject async I/O requests.
2845 */
2846 if (!is_sync_kiocb(kio)) {
2847 SFLOGFLOW(("vbsf_reg_write_iter: async I/O not yet supported\n")); /** @todo extend FsPerf with AIO tests. */
2848 return -EOPNOTSUPP;
2849 }
2850
2851 /*
2852 * If there are active writable mappings, coordinate with any
2853 * pending writes via those.
2854 */
2855 if ( mapping
2856 && mapping->nrpages > 0
2857 && mapping_writably_mapped(mapping)) {
2858# if RTLNX_VER_MIN(2,6,32)
2859 int err = filemap_fdatawait_range(mapping, offFile, offFile + cbToWrite - 1);
2860 if (err)
2861 return err;
2862# else
2863 /** @todo ... */
2864# endif
2865 }
2866
2867 /*
2868 * For small requests, try use an embedded buffer provided we get a heap block
2869 * that does not cross page boundraries (see host code).
2870 */
2871 if (cbToWrite <= PAGE_SIZE / 4 * 3 - RT_UOFFSETOF(VBOXSFWRITEEMBEDDEDREQ, abData[0]) /* see allocator */) {
2872 uint32_t const cbReq = RT_UOFFSETOF(VBOXSFWRITEEMBEDDEDREQ, abData[0]) + cbToWrite;
2873 VBOXSFWRITEEMBEDDEDREQ *pReq = (VBOXSFWRITEEMBEDDEDREQ *)VbglR0PhysHeapAlloc(cbReq);
2874 if (pReq) {
2875 if ((PAGE_SIZE - ((uintptr_t)pReq & PAGE_OFFSET_MASK)) >= cbReq) {
2876 ssize_t cbRet;
2877 if (copy_from_iter(pReq->abData, cbToWrite, iter) == cbToWrite) {
2878 int vrc = VbglR0SfHostReqWriteEmbedded(pSuperInfo->map.root, pReq, sf_r->Handle.hHost,
2879 offFile, (uint32_t)cbToWrite);
2880 sf_i->ModificationTimeAtOurLastWrite = sf_i->ModificationTime;
2881 if (RT_SUCCESS(vrc)) {
2882 cbRet = pReq->Parms.cb32Write.u.value32;
2883 AssertStmt(cbRet <= (ssize_t)cbToWrite, cbRet = cbToWrite);
2884 vbsf_reg_write_sync_page_cache(mapping, offFile, (uint32_t)cbRet, pReq->abData,
2885 NULL /*papSrcPages*/, 0 /*offSrcPage0*/, 0 /*cSrcPages*/);
2886
2887 offFile += cbRet;
2888 if (fAppend && (g_fSfFeatures & SHFL_FEATURE_WRITE_UPDATES_OFFSET))
2889 offFile = pReq->Parms.off64Write.u.value64;
2890 kio->ki_pos = offFile;
2891 if (offFile > i_size_read(inode))
2892 i_size_write(inode, offFile);
2893
2894# if RTLNX_VER_MIN(4,11,0)
2895 if ((size_t)cbRet < cbToWrite)
2896 iov_iter_revert(iter, cbToWrite - cbRet);
2897# endif
2898 } else
2899 cbRet = -EPROTO;
2900 sf_i->force_restat = 1; /* mtime (and size) may have changed */
2901 } else
2902 cbRet = -EFAULT;
2903 VbglR0PhysHeapFree(pReq);
2904 SFLOGFLOW(("vbsf_reg_write_iter: returns %#zx (%zd)\n", cbRet, cbRet));
2905 return cbRet;
2906 }
2907 VbglR0PhysHeapFree(pReq);
2908 }
2909 }
2910
2911 /*
2912 * Otherwise do the page locking thing.
2913 */
2914 return vbsf_reg_write_iter_locking(kio, iter, cbToWrite, offFile, pSuperInfo, sf_r, inode, sf_i, mapping, fAppend);
2915}
2916
2917#endif /* >= 2.6.19 */
2918
2919/**
2920 * Used by vbsf_reg_open() and vbsf_inode_atomic_open() to
2921 *
2922 * @returns shared folders create flags.
2923 * @param fLnxOpen The linux O_XXX flags to convert.
2924 * @param pfHandle Pointer to vbsf_handle::fFlags.
2925 * @param pszCaller Caller, for logging purposes.
2926 */
2927uint32_t vbsf_linux_oflags_to_vbox(unsigned fLnxOpen, uint32_t *pfHandle, const char *pszCaller)
2928{
2929 uint32_t fVBoxFlags = SHFL_CF_ACCESS_DENYNONE;
2930
2931 /*
2932 * Disposition.
2933 */
2934 if (fLnxOpen & O_CREAT) {
2935 Log(("%s: O_CREAT set\n", pszCaller));
2936 fVBoxFlags |= SHFL_CF_ACT_CREATE_IF_NEW;
2937 if (fLnxOpen & O_EXCL) {
2938 Log(("%s: O_EXCL set\n", pszCaller));
2939 fVBoxFlags |= SHFL_CF_ACT_FAIL_IF_EXISTS;
2940 } else if (fLnxOpen & O_TRUNC) {
2941 Log(("%s: O_TRUNC set\n", pszCaller));
2942 fVBoxFlags |= SHFL_CF_ACT_OVERWRITE_IF_EXISTS;
2943 } else
2944 fVBoxFlags |= SHFL_CF_ACT_OPEN_IF_EXISTS;
2945 } else {
2946 fVBoxFlags |= SHFL_CF_ACT_FAIL_IF_NEW;
2947 if (fLnxOpen & O_TRUNC) {
2948 Log(("%s: O_TRUNC set\n", pszCaller));
2949 fVBoxFlags |= SHFL_CF_ACT_OVERWRITE_IF_EXISTS;
2950 }
2951 }
2952
2953 /*
2954 * Access.
2955 */
2956 switch (fLnxOpen & O_ACCMODE) {
2957 case O_RDONLY:
2958 fVBoxFlags |= SHFL_CF_ACCESS_READ;
2959 *pfHandle |= VBSF_HANDLE_F_READ;
2960 break;
2961
2962 case O_WRONLY:
2963 fVBoxFlags |= SHFL_CF_ACCESS_WRITE;
2964 *pfHandle |= VBSF_HANDLE_F_WRITE;
2965 break;
2966
2967 case O_RDWR:
2968 fVBoxFlags |= SHFL_CF_ACCESS_READWRITE;
2969 *pfHandle |= VBSF_HANDLE_F_READ | VBSF_HANDLE_F_WRITE;
2970 break;
2971
2972 default:
2973 BUG();
2974 }
2975
2976 if (fLnxOpen & O_APPEND) {
2977 Log(("%s: O_APPEND set\n", pszCaller));
2978 fVBoxFlags |= SHFL_CF_ACCESS_APPEND;
2979 *pfHandle |= VBSF_HANDLE_F_APPEND;
2980 }
2981
2982 /*
2983 * Only directories?
2984 */
2985 if (fLnxOpen & O_DIRECTORY) {
2986 Log(("%s: O_DIRECTORY set\n", pszCaller));
2987 fVBoxFlags |= SHFL_CF_DIRECTORY;
2988 }
2989
2990 return fVBoxFlags;
2991}
2992
2993
2994/**
2995 * Open a regular file.
2996 *
2997 * @param inode the inode
2998 * @param file the file
2999 * @returns 0 on success, Linux error code otherwise
3000 */
3001static int vbsf_reg_open(struct inode *inode, struct file *file)
3002{
3003 int rc, rc_linux = 0;
3004 struct vbsf_super_info *pSuperInfo = VBSF_GET_SUPER_INFO(inode->i_sb);
3005 struct vbsf_inode_info *sf_i = VBSF_GET_INODE_INFO(inode);
3006 struct dentry *dentry = VBSF_GET_F_DENTRY(file);
3007 struct vbsf_reg_info *sf_r;
3008 VBOXSFCREATEREQ *pReq;
3009
3010 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));
3011 Assert(pSuperInfo);
3012 Assert(sf_i);
3013
3014 sf_r = kmalloc(sizeof(*sf_r), GFP_KERNEL);
3015 if (!sf_r) {
3016 LogRelFunc(("could not allocate reg info\n"));
3017 return -ENOMEM;
3018 }
3019
3020 RTListInit(&sf_r->Handle.Entry);
3021 sf_r->Handle.cRefs = 1;
3022 sf_r->Handle.fFlags = VBSF_HANDLE_F_FILE | VBSF_HANDLE_F_MAGIC;
3023 sf_r->Handle.hHost = SHFL_HANDLE_NIL;
3024
3025 /* Already open? */
3026 if (sf_i->handle != SHFL_HANDLE_NIL) {
3027 /*
3028 * This inode was created with vbsf_create_worker(). Check the CreateFlags:
3029 * O_CREAT, O_TRUNC: inherent true (file was just created). Not sure
3030 * about the access flags (SHFL_CF_ACCESS_*).
3031 */
3032 sf_i->force_restat = 1;
3033 sf_r->Handle.hHost = sf_i->handle;
3034 sf_i->handle = SHFL_HANDLE_NIL;
3035 file->private_data = sf_r;
3036
3037 sf_r->Handle.fFlags |= VBSF_HANDLE_F_READ | VBSF_HANDLE_F_WRITE; /** @todo fix */
3038 vbsf_handle_append(sf_i, &sf_r->Handle);
3039 SFLOGFLOW(("vbsf_reg_open: returns 0 (#1) - sf_i=%p hHost=%#llx\n", sf_i, sf_r->Handle.hHost));
3040 return 0;
3041 }
3042
3043 pReq = (VBOXSFCREATEREQ *)VbglR0PhysHeapAlloc(sizeof(*pReq) + sf_i->path->u16Size);
3044 if (!pReq) {
3045 kfree(sf_r);
3046 LogRelFunc(("Failed to allocate a VBOXSFCREATEREQ buffer!\n"));
3047 return -ENOMEM;
3048 }
3049 memcpy(&pReq->StrPath, sf_i->path, SHFLSTRING_HEADER_SIZE + sf_i->path->u16Size);
3050 RT_ZERO(pReq->CreateParms);
3051 pReq->CreateParms.Handle = SHFL_HANDLE_NIL;
3052
3053 /* We check the value of pReq->CreateParms.Handle afterwards to
3054 * find out if the call succeeded or failed, as the API does not seem
3055 * to cleanly distinguish error and informational messages.
3056 *
3057 * Furthermore, we must set pReq->CreateParms.Handle to SHFL_HANDLE_NIL
3058 * to make the shared folders host service use our fMode parameter */
3059
3060 /* We ignore O_EXCL, as the Linux kernel seems to call create
3061 beforehand itself, so O_EXCL should always fail. */
3062 pReq->CreateParms.CreateFlags = vbsf_linux_oflags_to_vbox(file->f_flags & ~O_EXCL, &sf_r->Handle.fFlags, __FUNCTION__);
3063 pReq->CreateParms.Info.Attr.fMode = inode->i_mode;
3064 LogFunc(("vbsf_reg_open: calling VbglR0SfHostReqCreate, file %s, flags=%#x, %#x\n",
3065 sf_i->path->String.utf8, file->f_flags, pReq->CreateParms.CreateFlags));
3066 rc = VbglR0SfHostReqCreate(pSuperInfo->map.root, pReq);
3067 if (RT_FAILURE(rc)) {
3068 LogFunc(("VbglR0SfHostReqCreate failed flags=%d,%#x rc=%Rrc\n", file->f_flags, pReq->CreateParms.CreateFlags, rc));
3069 kfree(sf_r);
3070 VbglR0PhysHeapFree(pReq);
3071 return -RTErrConvertToErrno(rc);
3072 }
3073
3074 if (pReq->CreateParms.Handle != SHFL_HANDLE_NIL) {
3075 vbsf_dentry_chain_increase_ttl(dentry);
3076 vbsf_update_inode(inode, sf_i, &pReq->CreateParms.Info, pSuperInfo, false /*fInodeLocked*/, 0 /*fSetAttrs*/);
3077 rc_linux = 0;
3078 } else {
3079 switch (pReq->CreateParms.Result) {
3080 case SHFL_PATH_NOT_FOUND:
3081 vbsf_dentry_invalidate_ttl(dentry);
3082 rc_linux = -ENOENT;
3083 break;
3084 case SHFL_FILE_NOT_FOUND:
3085 vbsf_dentry_invalidate_ttl(dentry);
3086 /** @todo sf_dentry_increase_parent_ttl(file->f_dentry); if we can trust it. */
3087 rc_linux = -ENOENT;
3088 break;
3089 case SHFL_FILE_EXISTS:
3090 vbsf_dentry_chain_increase_ttl(dentry);
3091 vbsf_update_inode(inode, sf_i, &pReq->CreateParms.Info, pSuperInfo, false /*fInodeLocked*/, 0 /*fSetAttrs*/);
3092 rc_linux = -EEXIST;
3093 break;
3094 default:
3095 vbsf_dentry_chain_increase_parent_ttl(dentry);
3096 rc_linux = 0;
3097 break;
3098 }
3099 }
3100
3101 sf_r->Handle.hHost = pReq->CreateParms.Handle;
3102 file->private_data = sf_r;
3103 vbsf_handle_append(sf_i, &sf_r->Handle);
3104 VbglR0PhysHeapFree(pReq);
3105 SFLOGFLOW(("vbsf_reg_open: returns 0 (#2) - sf_i=%p hHost=%#llx\n", sf_i, sf_r->Handle.hHost));
3106 return rc_linux;
3107}
3108
3109
3110/**
3111 * Close a regular file.
3112 *
3113 * @param inode the inode
3114 * @param file the file
3115 * @returns 0 on success, Linux error code otherwise
3116 */
3117static int vbsf_reg_release(struct inode *inode, struct file *file)
3118{
3119 struct vbsf_inode_info *sf_i = VBSF_GET_INODE_INFO(inode);
3120 struct vbsf_reg_info *sf_r = file->private_data;
3121
3122 SFLOGFLOW(("vbsf_reg_release: inode=%p file=%p\n", inode, file));
3123 if (sf_r) {
3124 struct vbsf_super_info *pSuperInfo = VBSF_GET_SUPER_INFO(inode->i_sb);
3125 struct address_space *mapping = inode->i_mapping;
3126 Assert(pSuperInfo);
3127
3128 /* If we're closing the last handle for this inode, make sure the flush
3129 the mapping or we'll end up in vbsf_writepage without a handle. */
3130 if ( mapping
3131 && mapping->nrpages > 0
3132 /** @todo && last writable handle */ ) {
3133#if RTLNX_VER_MIN(2,4,25)
3134 if (filemap_fdatawrite(mapping) != -EIO)
3135#else
3136 if ( filemap_fdatasync(mapping) == 0
3137 && fsync_inode_data_buffers(inode) == 0)
3138#endif
3139 filemap_fdatawait(inode->i_mapping);
3140 }
3141
3142 /* Release sf_r, closing the handle if we're the last user. */
3143 file->private_data = NULL;
3144 vbsf_handle_release(&sf_r->Handle, pSuperInfo, "vbsf_reg_release");
3145
3146 sf_i->handle = SHFL_HANDLE_NIL;
3147 }
3148 return 0;
3149}
3150
3151
3152/**
3153 * Wrapper around generic/default seek function that ensures that we've got
3154 * the up-to-date file size when doing anything relative to EOF.
3155 *
3156 * The issue is that the host may extend the file while we weren't looking and
3157 * if the caller wishes to append data, it may end up overwriting existing data
3158 * if we operate with a stale size. So, we always retrieve the file size on EOF
3159 * relative seeks.
3160 */
3161static loff_t vbsf_reg_llseek(struct file *file, loff_t off, int whence)
3162{
3163 SFLOGFLOW(("vbsf_reg_llseek: file=%p off=%lld whence=%d\n", file, off, whence));
3164
3165 switch (whence) {
3166#ifdef SEEK_HOLE
3167 case SEEK_HOLE:
3168 case SEEK_DATA:
3169#endif
3170 case SEEK_END: {
3171 struct vbsf_reg_info *sf_r = file->private_data;
3172 int rc = vbsf_inode_revalidate_with_handle(VBSF_GET_F_DENTRY(file), sf_r->Handle.hHost,
3173 true /*fForce*/, false /*fInodeLocked*/);
3174 if (rc == 0)
3175 break;
3176 return rc;
3177 }
3178 }
3179
3180#if RTLNX_VER_MIN(2,4,8)
3181 return generic_file_llseek(file, off, whence);
3182#else
3183 return default_llseek(file, off, whence);
3184#endif
3185}
3186
3187
3188/**
3189 * Flush region of file - chiefly mmap/msync.
3190 *
3191 * We cannot use the noop_fsync / simple_sync_file here as that means
3192 * msync(,,MS_SYNC) will return before the data hits the host, thereby
3193 * causing coherency issues with O_DIRECT access to the same file as
3194 * well as any host interaction with the file.
3195 */
3196#if RTLNX_VER_MIN(3,1,0) \
3197 || (defined(CONFIG_SUSE_KERNEL) && RTLNX_VER_MIN(3,0,101) /** @todo figure when exactly */)
3198static int vbsf_reg_fsync(struct file *file, loff_t start, loff_t end, int datasync)
3199{
3200# if RTLNX_VER_MIN(3,16,0)
3201 return __generic_file_fsync(file, start, end, datasync);
3202# else
3203 return generic_file_fsync(file, start, end, datasync);
3204# endif
3205}
3206#elif RTLNX_VER_MIN(2,6,35)
3207static int vbsf_reg_fsync(struct file *file, int datasync)
3208{
3209 return generic_file_fsync(file, datasync);
3210}
3211#else /* < 2.6.35 */
3212static int vbsf_reg_fsync(struct file *file, struct dentry *dentry, int datasync)
3213{
3214# if RTLNX_VER_MIN(2,6,31)
3215 return simple_fsync(file, dentry, datasync);
3216# else
3217 int rc;
3218 struct inode *inode = dentry->d_inode;
3219 AssertReturn(inode, -EINVAL);
3220
3221 /** @todo What about file_fsync()? (<= 2.5.11) */
3222
3223# if RTLNX_VER_MIN(2,5,12)
3224 rc = sync_mapping_buffers(inode->i_mapping);
3225 if ( rc == 0
3226 && (inode->i_state & I_DIRTY)
3227 && ((inode->i_state & I_DIRTY_DATASYNC) || !datasync)
3228 ) {
3229 struct writeback_control wbc = {
3230 .sync_mode = WB_SYNC_ALL,
3231 .nr_to_write = 0
3232 };
3233 rc = sync_inode(inode, &wbc);
3234 }
3235# else /* < 2.5.12 */
3236 /** @todo
3237 * Somethings is buggy here or in the 2.4.21-27.EL kernel I'm testing on.
3238 *
3239 * In theory we shouldn't need to do anything here, since msync will call
3240 * writepage() on each dirty page and we write them out synchronously. So, the
3241 * problem is elsewhere... Doesn't happen all the time either. Sigh.
3242 */
3243 rc = fsync_inode_buffers(inode);
3244# if RTLNX_VER_MIN(2,4,10)
3245 if (rc == 0 && datasync)
3246 rc = fsync_inode_data_buffers(inode);
3247# endif
3248
3249# endif /* < 2.5.12 */
3250 return rc;
3251# endif
3252}
3253#endif /* < 2.6.35 */
3254
3255
3256#if RTLNX_VER_MIN(4,5,0)
3257/**
3258 * Copy a datablock from one file to another on the host side.
3259 */
3260static ssize_t vbsf_reg_copy_file_range(struct file *pFileSrc, loff_t offSrc, struct file *pFileDst, loff_t offDst,
3261 size_t cbRange, unsigned int fFlags)
3262{
3263 ssize_t cbRet;
3264 if (g_uSfLastFunction >= SHFL_FN_COPY_FILE_PART) {
3265 struct inode *pInodeSrc = pFileSrc->f_inode;
3266 struct vbsf_inode_info *pInodeInfoSrc = VBSF_GET_INODE_INFO(pInodeSrc);
3267 struct vbsf_super_info *pSuperInfoSrc = VBSF_GET_SUPER_INFO(pInodeSrc->i_sb);
3268 struct vbsf_reg_info *pFileInfoSrc = (struct vbsf_reg_info *)pFileSrc->private_data;
3269 struct inode *pInodeDst = pInodeSrc;
3270 struct vbsf_inode_info *pInodeInfoDst = VBSF_GET_INODE_INFO(pInodeDst);
3271 struct vbsf_super_info *pSuperInfoDst = VBSF_GET_SUPER_INFO(pInodeDst->i_sb);
3272 struct vbsf_reg_info *pFileInfoDst = (struct vbsf_reg_info *)pFileDst->private_data;
3273 VBOXSFCOPYFILEPARTREQ *pReq;
3274
3275 /*
3276 * Some extra validation.
3277 */
3278 AssertPtrReturn(pInodeInfoSrc, -EOPNOTSUPP);
3279 Assert(pInodeInfoSrc->u32Magic == SF_INODE_INFO_MAGIC);
3280 AssertPtrReturn(pInodeInfoDst, -EOPNOTSUPP);
3281 Assert(pInodeInfoDst->u32Magic == SF_INODE_INFO_MAGIC);
3282
3283# if RTLNX_VER_MAX(4,11,0)
3284 if (!S_ISREG(pInodeSrc->i_mode) || !S_ISREG(pInodeDst->i_mode))
3285 return S_ISDIR(pInodeSrc->i_mode) || S_ISDIR(pInodeDst->i_mode) ? -EISDIR : -EINVAL;
3286# endif
3287
3288 /*
3289 * Allocate the request and issue it.
3290 */
3291 pReq = (VBOXSFCOPYFILEPARTREQ *)VbglR0PhysHeapAlloc(sizeof(*pReq));
3292 if (pReq) {
3293 int vrc = VbglR0SfHostReqCopyFilePart(pSuperInfoSrc->map.root, pFileInfoSrc->Handle.hHost, offSrc,
3294 pSuperInfoDst->map.root, pFileInfoDst->Handle.hHost, offDst,
3295 cbRange, 0 /*fFlags*/, pReq);
3296 if (RT_SUCCESS(vrc))
3297 cbRet = pReq->Parms.cb64ToCopy.u.value64;
3298 else if (vrc == VERR_NOT_IMPLEMENTED)
3299 cbRet = -EOPNOTSUPP;
3300 else
3301 cbRet = -RTErrConvertToErrno(vrc);
3302
3303 VbglR0PhysHeapFree(pReq);
3304 } else
3305 cbRet = -ENOMEM;
3306 } else {
3307 cbRet = -EOPNOTSUPP;
3308 }
3309 SFLOGFLOW(("vbsf_reg_copy_file_range: returns %zd\n", cbRet));
3310 return cbRet;
3311}
3312#endif /* > 4.5 */
3313
3314
3315#ifdef SFLOG_ENABLED
3316/*
3317 * This is just for logging page faults and such.
3318 */
3319
3320/** Pointer to the ops generic_file_mmap returns the first time it's called. */
3321static struct vm_operations_struct const *g_pGenericFileVmOps = NULL;
3322/** Merge of g_LoggingVmOpsTemplate and g_pGenericFileVmOps. */
3323static struct vm_operations_struct g_LoggingVmOps;
3324
3325
3326/* Generic page fault callback: */
3327# if RTLNX_VER_MIN(4,11,0)
3328static vm_fault_t vbsf_vmlog_fault(struct vm_fault *vmf)
3329{
3330 vm_fault_t rc;
3331 SFLOGFLOW(("vbsf_vmlog_fault: vmf=%p flags=%#x addr=%p\n", vmf, vmf->flags, vmf->address));
3332 rc = g_pGenericFileVmOps->fault(vmf);
3333 SFLOGFLOW(("vbsf_vmlog_fault: returns %d\n", rc));
3334 return rc;
3335}
3336# elif RTLNX_VER_MIN(2,6,23)
3337static int vbsf_vmlog_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
3338{
3339 int rc;
3340# if RTLNX_VER_MIN(4,10,0)
3341 SFLOGFLOW(("vbsf_vmlog_fault: vma=%p vmf=%p flags=%#x addr=%p\n", vma, vmf, vmf->flags, vmf->address));
3342# else
3343 SFLOGFLOW(("vbsf_vmlog_fault: vma=%p vmf=%p flags=%#x addr=%p\n", vma, vmf, vmf->flags, vmf->virtual_address));
3344# endif
3345 rc = g_pGenericFileVmOps->fault(vma, vmf);
3346 SFLOGFLOW(("vbsf_vmlog_fault: returns %d\n", rc));
3347 return rc;
3348}
3349# endif
3350
3351
3352/* Special/generic page fault handler: */
3353# if RTLNX_VER_MIN(2,6,26)
3354# elif RTLNX_VER_MIN(2,6,1)
3355static struct page *vbsf_vmlog_nopage(struct vm_area_struct *vma, unsigned long address, int *type)
3356{
3357 struct page *page;
3358 SFLOGFLOW(("vbsf_vmlog_nopage: vma=%p address=%p type=%p:{%#x}\n", vma, address, type, type ? *type : 0));
3359 page = g_pGenericFileVmOps->nopage(vma, address, type);
3360 SFLOGFLOW(("vbsf_vmlog_nopage: returns %p\n", page));
3361 return page;
3362}
3363# else
3364static struct page *vbsf_vmlog_nopage(struct vm_area_struct *vma, unsigned long address, int write_access_or_unused)
3365{
3366 struct page *page;
3367 SFLOGFLOW(("vbsf_vmlog_nopage: vma=%p address=%p wau=%d\n", vma, address, write_access_or_unused));
3368 page = g_pGenericFileVmOps->nopage(vma, address, write_access_or_unused);
3369 SFLOGFLOW(("vbsf_vmlog_nopage: returns %p\n", page));
3370 return page;
3371}
3372# endif /* < 2.6.26 */
3373
3374
3375/* Special page fault callback for making something writable: */
3376# if RTLNX_VER_MIN(4,11,0)
3377static vm_fault_t vbsf_vmlog_page_mkwrite(struct vm_fault *vmf)
3378{
3379 vm_fault_t rc;
3380 SFLOGFLOW(("vbsf_vmlog_page_mkwrite: vmf=%p flags=%#x addr=%p\n", vmf, vmf->flags, vmf->address));
3381 rc = g_pGenericFileVmOps->page_mkwrite(vmf);
3382 SFLOGFLOW(("vbsf_vmlog_page_mkwrite: returns %d\n", rc));
3383 return rc;
3384}
3385# elif RTLNX_VER_MIN(2,6,30)
3386static int vbsf_vmlog_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
3387{
3388 int rc;
3389# if RTLNX_VER_MIN(4,10,0)
3390 SFLOGFLOW(("vbsf_vmlog_page_mkwrite: vma=%p vmf=%p flags=%#x addr=%p\n", vma, vmf, vmf->flags, vmf->address));
3391# else
3392 SFLOGFLOW(("vbsf_vmlog_page_mkwrite: vma=%p vmf=%p flags=%#x addr=%p\n", vma, vmf, vmf->flags, vmf->virtual_address));
3393# endif
3394 rc = g_pGenericFileVmOps->page_mkwrite(vma, vmf);
3395 SFLOGFLOW(("vbsf_vmlog_page_mkwrite: returns %d\n", rc));
3396 return rc;
3397}
3398# elif RTLNX_VER_MIN(2,6,18)
3399static int vbsf_vmlog_page_mkwrite(struct vm_area_struct *vma, struct page *page)
3400{
3401 int rc;
3402 SFLOGFLOW(("vbsf_vmlog_page_mkwrite: vma=%p page=%p\n", vma, page));
3403 rc = g_pGenericFileVmOps->page_mkwrite(vma, page);
3404 SFLOGFLOW(("vbsf_vmlog_page_mkwrite: returns %d\n", rc));
3405 return rc;
3406}
3407# endif
3408
3409
3410/* Special page fault callback for mapping pages: */
3411# if RTLNX_VER_MIN(5,12,0)
3412static vm_fault_t vbsf_vmlog_map_pages(struct vm_fault *vmf, pgoff_t start, pgoff_t end)
3413{
3414 vm_fault_t rc;
3415 SFLOGFLOW(("vbsf_vmlog_map_pages: vmf=%p (flags=%#x addr=%p) start=%p end=%p\n", vmf, vmf->flags, vmf->address, start, end));
3416 rc = g_pGenericFileVmOps->map_pages(vmf, start, end);
3417 SFLOGFLOW(("vbsf_vmlog_map_pages: returns\n"));
3418 return rc;
3419}
3420# elif RTLNX_VER_MIN(4,10,0)
3421static void vbsf_vmlog_map_pages(struct vm_fault *vmf, pgoff_t start, pgoff_t end)
3422{
3423 SFLOGFLOW(("vbsf_vmlog_map_pages: vmf=%p (flags=%#x addr=%p) start=%p end=%p\n", vmf, vmf->flags, vmf->address, start, end));
3424 g_pGenericFileVmOps->map_pages(vmf, start, end);
3425 SFLOGFLOW(("vbsf_vmlog_map_pages: returns\n"));
3426}
3427# elif RTLNX_VER_MIN(4,8,0)
3428static void vbsf_vmlog_map_pages(struct fault_env *fenv, pgoff_t start, pgoff_t end)
3429{
3430 SFLOGFLOW(("vbsf_vmlog_map_pages: fenv=%p (flags=%#x addr=%p) start=%p end=%p\n", fenv, fenv->flags, fenv->address, start, end));
3431 g_pGenericFileVmOps->map_pages(fenv, start, end);
3432 SFLOGFLOW(("vbsf_vmlog_map_pages: returns\n"));
3433}
3434# elif RTLNX_VER_MIN(3,15,0)
3435static void vbsf_vmlog_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
3436{
3437 SFLOGFLOW(("vbsf_vmlog_map_pages: vma=%p vmf=%p (flags=%#x addr=%p)\n", vma, vmf, vmf->flags, vmf->virtual_address));
3438 g_pGenericFileVmOps->map_pages(vma, vmf);
3439 SFLOGFLOW(("vbsf_vmlog_map_pages: returns\n"));
3440}
3441# endif
3442
3443
3444/** Overload template. */
3445static struct vm_operations_struct const g_LoggingVmOpsTemplate = {
3446# if RTLNX_VER_MIN(2,6,23)
3447 .fault = vbsf_vmlog_fault,
3448# endif
3449# if RTLNX_VER_MAX(2,6,26)
3450 .nopage = vbsf_vmlog_nopage,
3451# endif
3452# if RTLNX_VER_MIN(2,6,18)
3453 .page_mkwrite = vbsf_vmlog_page_mkwrite,
3454# endif
3455# if RTLNX_VER_MIN(3,15,0)
3456 .map_pages = vbsf_vmlog_map_pages,
3457# endif
3458};
3459
3460/** file_operations::mmap wrapper for logging purposes. */
3461extern int vbsf_reg_mmap(struct file *file, struct vm_area_struct *vma)
3462{
3463 int rc;
3464 SFLOGFLOW(("vbsf_reg_mmap: file=%p vma=%p\n", file, vma));
3465 rc = generic_file_mmap(file, vma);
3466 if (rc == 0) {
3467 /* Merge the ops and template the first time thru (there's a race here). */
3468 if (g_pGenericFileVmOps == NULL) {
3469 uintptr_t const *puSrc1 = (uintptr_t *)vma->vm_ops;
3470 uintptr_t const *puSrc2 = (uintptr_t *)&g_LoggingVmOpsTemplate;
3471 uintptr_t volatile *puDst = (uintptr_t *)&g_LoggingVmOps;
3472 size_t cbLeft = sizeof(g_LoggingVmOps) / sizeof(*puDst);
3473 while (cbLeft-- > 0) {
3474 *puDst = *puSrc2 && *puSrc1 ? *puSrc2 : *puSrc1;
3475 puSrc1++;
3476 puSrc2++;
3477 puDst++;
3478 }
3479 g_pGenericFileVmOps = vma->vm_ops;
3480 vma->vm_ops = &g_LoggingVmOps;
3481 } else if (g_pGenericFileVmOps == vma->vm_ops)
3482 vma->vm_ops = &g_LoggingVmOps;
3483 else
3484 SFLOGFLOW(("vbsf_reg_mmap: Warning: vm_ops=%p, expected %p!\n", vma->vm_ops, g_pGenericFileVmOps));
3485 }
3486 SFLOGFLOW(("vbsf_reg_mmap: returns %d\n", rc));
3487 return rc;
3488}
3489
3490#endif /* SFLOG_ENABLED */
3491
3492
3493/**
3494 * File operations for regular files.
3495 *
3496 * Note on splice_read/splice_write/sendfile:
3497 * - Splice was introduced in 2.6.17. The generic_file_splice_read/write
3498 * methods go thru the page cache, which is undesirable and is why we
3499 * need to cook our own versions of the code as long as we cannot track
3500 * host-side writes and correctly invalidate the guest page-cache.
3501 * - Sendfile reimplemented using splice in 2.6.23.
3502 * - The default_file_splice_read/write no-page-cache fallback functions,
3503 * were introduced in 2.6.31. The write one work in page units.
3504 * - Since linux 3.16 there is iter_file_splice_write that uses iter_write.
3505 * - Since linux 4.9 the generic_file_splice_read function started using
3506 * read_iter.
3507 */
3508struct file_operations vbsf_reg_fops = {
3509 .open = vbsf_reg_open,
3510#if RTLNX_VER_MAX(5,10,0)
3511 .read = vbsf_reg_read,
3512 .write = vbsf_reg_write,
3513#endif
3514#if RTLNX_VER_MIN(3,16,0)
3515 .read_iter = vbsf_reg_read_iter,
3516 .write_iter = vbsf_reg_write_iter,
3517#elif RTLNX_VER_MIN(2,6,19)
3518 .aio_read = vbsf_reg_aio_read,
3519 .aio_write = vbsf_reg_aio_write,
3520#endif
3521 .release = vbsf_reg_release,
3522#ifdef SFLOG_ENABLED
3523 .mmap = vbsf_reg_mmap,
3524#else
3525 .mmap = generic_file_mmap,
3526#endif
3527#if RTLNX_VER_RANGE(2,6,17, 2,6,31)
3528 .splice_read = vbsf_splice_read,
3529#endif
3530#if RTLNX_VER_MIN(3,16,0)
3531 .splice_write = iter_file_splice_write,
3532#elif RTLNX_VER_MIN(2,6,17)
3533 .splice_write = vbsf_splice_write,
3534#endif
3535#if RTLNX_VER_RANGE(2,5,30, 2,6,23)
3536 .sendfile = vbsf_reg_sendfile,
3537#endif
3538 .llseek = vbsf_reg_llseek,
3539 .fsync = vbsf_reg_fsync,
3540#if RTLNX_VER_MIN(4,5,0)
3541 .copy_file_range = vbsf_reg_copy_file_range,
3542#endif
3543};
3544
3545
3546/**
3547 * Inodes operations for regular files.
3548 */
3549struct inode_operations vbsf_reg_iops = {
3550#if RTLNX_VER_MIN(2,5,18)
3551 .getattr = vbsf_inode_getattr,
3552#else
3553 .revalidate = vbsf_inode_revalidate,
3554#endif
3555 .setattr = vbsf_inode_setattr,
3556};
3557
3558
3559
3560/*********************************************************************************************************************************
3561* Address Space Operations on Regular Files (for mmap, sendfile, direct I/O) *
3562*********************************************************************************************************************************/
3563
3564/**
3565 * Used to read the content of a page into the page cache.
3566 *
3567 * Needed for mmap and reads+writes when the file is mmapped in a
3568 * shared+writeable fashion.
3569 */
3570static int vbsf_readpage(struct file *file, struct page *page)
3571{
3572 struct inode *inode = VBSF_GET_F_DENTRY(file)->d_inode;
3573 int err;
3574
3575 SFLOGFLOW(("vbsf_readpage: inode=%p file=%p page=%p off=%#llx\n", inode, file, page, (uint64_t)page->index << PAGE_SHIFT));
3576 Assert(PageLocked(page));
3577
3578 if (PageUptodate(page)) {
3579 unlock_page(page);
3580 return 0;
3581 }
3582
3583 if (!is_bad_inode(inode)) {
3584 VBOXSFREADPGLSTREQ *pReq = (VBOXSFREADPGLSTREQ *)VbglR0PhysHeapAlloc(sizeof(*pReq));
3585 if (pReq) {
3586 struct vbsf_super_info *pSuperInfo = VBSF_GET_SUPER_INFO(inode->i_sb);
3587 struct vbsf_reg_info *sf_r = file->private_data;
3588 uint32_t cbRead;
3589 int vrc;
3590
3591 pReq->PgLst.offFirstPage = 0;
3592 pReq->PgLst.aPages[0] = page_to_phys(page);
3593 vrc = VbglR0SfHostReqReadPgLst(pSuperInfo->map.root,
3594 pReq,
3595 sf_r->Handle.hHost,
3596 (uint64_t)page->index << PAGE_SHIFT,
3597 PAGE_SIZE,
3598 1 /*cPages*/);
3599
3600 cbRead = pReq->Parms.cb32Read.u.value32;
3601 AssertStmt(cbRead <= PAGE_SIZE, cbRead = PAGE_SIZE);
3602 VbglR0PhysHeapFree(pReq);
3603
3604 if (RT_SUCCESS(vrc)) {
3605 if (cbRead == PAGE_SIZE) {
3606 /* likely */
3607 } else {
3608 uint8_t *pbMapped = (uint8_t *)kmap(page);
3609 RT_BZERO(&pbMapped[cbRead], PAGE_SIZE - cbRead);
3610 kunmap(page);
3611 /** @todo truncate the inode file size? */
3612 }
3613
3614 flush_dcache_page(page);
3615 SetPageUptodate(page);
3616 unlock_page(page);
3617 return 0;
3618 }
3619 err = -RTErrConvertToErrno(vrc);
3620 } else
3621 err = -ENOMEM;
3622 } else
3623 err = -EIO;
3624 SetPageError(page);
3625 unlock_page(page);
3626 return err;
3627}
3628
3629
3630/**
3631 * Used to write out the content of a dirty page cache page to the host file.
3632 *
3633 * Needed for mmap and writes when the file is mmapped in a shared+writeable
3634 * fashion.
3635 */
3636#if RTLNX_VER_MIN(2,5,52)
3637static int vbsf_writepage(struct page *page, struct writeback_control *wbc)
3638#else
3639static int vbsf_writepage(struct page *page)
3640#endif
3641{
3642 struct address_space *mapping = page->mapping;
3643 struct inode *inode = mapping->host;
3644 struct vbsf_inode_info *sf_i = VBSF_GET_INODE_INFO(inode);
3645 struct vbsf_handle *pHandle = vbsf_handle_find(sf_i, VBSF_HANDLE_F_WRITE, VBSF_HANDLE_F_APPEND);
3646 int err;
3647
3648 SFLOGFLOW(("vbsf_writepage: inode=%p page=%p off=%#llx pHandle=%p (%#llx)\n",
3649 inode, page, (uint64_t)page->index << PAGE_SHIFT, pHandle, pHandle ? pHandle->hHost : 0));
3650
3651 if (pHandle) {
3652 struct vbsf_super_info *pSuperInfo = VBSF_GET_SUPER_INFO(inode->i_sb);
3653 VBOXSFWRITEPGLSTREQ *pReq = (VBOXSFWRITEPGLSTREQ *)VbglR0PhysHeapAlloc(sizeof(*pReq));
3654 if (pReq) {
3655 uint64_t const cbFile = i_size_read(inode);
3656 uint64_t const offInFile = (uint64_t)page->index << PAGE_SHIFT;
3657 uint32_t const cbToWrite = page->index != (cbFile >> PAGE_SHIFT) ? PAGE_SIZE
3658 : (uint32_t)cbFile & (uint32_t)PAGE_OFFSET_MASK;
3659 int vrc;
3660
3661 pReq->PgLst.offFirstPage = 0;
3662 pReq->PgLst.aPages[0] = page_to_phys(page);
3663 vrc = VbglR0SfHostReqWritePgLst(pSuperInfo->map.root,
3664 pReq,
3665 pHandle->hHost,
3666 offInFile,
3667 cbToWrite,
3668 1 /*cPages*/);
3669 sf_i->ModificationTimeAtOurLastWrite = sf_i->ModificationTime;
3670 AssertMsgStmt(pReq->Parms.cb32Write.u.value32 == cbToWrite || RT_FAILURE(vrc), /* lazy bird */
3671 ("%#x vs %#x\n", pReq->Parms.cb32Write, cbToWrite),
3672 vrc = VERR_WRITE_ERROR);
3673 VbglR0PhysHeapFree(pReq);
3674
3675 if (RT_SUCCESS(vrc)) {
3676 /* Update the inode if we've extended the file. */
3677 /** @todo is this necessary given the cbToWrite calc above? */
3678 uint64_t const offEndOfWrite = offInFile + cbToWrite;
3679 if ( offEndOfWrite > cbFile
3680 && offEndOfWrite > i_size_read(inode))
3681 i_size_write(inode, offEndOfWrite);
3682
3683 /* Update and unlock the page. */
3684 if (PageError(page))
3685 ClearPageError(page);
3686 SetPageUptodate(page);
3687 unlock_page(page);
3688
3689 vbsf_handle_release(pHandle, pSuperInfo, "vbsf_writepage");
3690 return 0;
3691 }
3692
3693 /*
3694 * We failed.
3695 */
3696 err = -EIO;
3697 } else
3698 err = -ENOMEM;
3699 vbsf_handle_release(pHandle, pSuperInfo, "vbsf_writepage");
3700 } else {
3701 /** @todo we could re-open the file here and deal with this... */
3702 static uint64_t volatile s_cCalls = 0;
3703 if (s_cCalls++ < 16)
3704 printk("vbsf_writepage: no writable handle for %s..\n", sf_i->path->String.ach);
3705 err = -EIO;
3706 }
3707 SetPageError(page);
3708 unlock_page(page);
3709 return err;
3710}
3711
3712
3713#if RTLNX_VER_MIN(2,6,24)
3714/**
3715 * Called when writing thru the page cache (which we shouldn't be doing).
3716 */
3717int vbsf_write_begin(struct file *file, struct address_space *mapping, loff_t pos,
3718 unsigned len, unsigned flags, struct page **pagep, void **fsdata)
3719{
3720 /** @todo r=bird: We shouldn't ever get here, should we? Because we don't use
3721 * the page cache for any writes AFAIK. We could just as well use
3722 * simple_write_begin & simple_write_end here if we think we really
3723 * need to have non-NULL function pointers in the table... */
3724 static uint64_t volatile s_cCalls = 0;
3725 if (s_cCalls++ < 16) {
3726 printk("vboxsf: Unexpected call to vbsf_write_begin(pos=%#llx len=%#x flags=%#x)! Please report.\n",
3727 (unsigned long long)pos, len, flags);
3728 RTLogBackdoorPrintf("vboxsf: Unexpected call to vbsf_write_begin(pos=%#llx len=%#x flags=%#x)! Please report.\n",
3729 (unsigned long long)pos, len, flags);
3730# ifdef WARN_ON
3731 WARN_ON(1);
3732# endif
3733 }
3734 return simple_write_begin(file, mapping, pos, len, flags, pagep, fsdata);
3735}
3736#endif /* KERNEL_VERSION >= 2.6.24 */
3737
3738
3739#if RTLNX_VER_MIN(2,4,10)
3740
3741# ifdef VBOX_UEK
3742# undef iov_iter /* HACK ALERT! Don't put anything needing vbsf_iov_iter after this fun! */
3743# endif
3744
3745/**
3746 * This is needed to make open accept O_DIRECT as well as dealing with direct
3747 * I/O requests if we don't intercept them earlier.
3748 */
3749# if RTLNX_VER_MIN(4, 7, 0) \
3750 || (defined(CONFIG_SUSE_KERNEL) && RTLNX_VER_RANGE(4,4,73, 4,4,74) /** @todo Figure out when exactly. */) \
3751 || (defined(CONFIG_SUSE_KERNEL) && RTLNX_VER_RANGE(4,4,75, 4,4,90) /** @todo Figure out when exactly. */) \
3752 || (defined(CONFIG_SUSE_KERNEL) && RTLNX_VER_RANGE(4,4,92, 4,5,0) /** @todo Figure out when exactly. */)
3753static ssize_t vbsf_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3754# elif RTLNX_VER_MIN(4, 1, 0)
3755static ssize_t vbsf_direct_IO(struct kiocb *iocb, struct iov_iter *iter, loff_t offset)
3756# elif RTLNX_VER_MIN(3, 16, 0) || defined(VBOX_UEK)
3757static ssize_t vbsf_direct_IO(int rw, struct kiocb *iocb, struct iov_iter *iter, loff_t offset)
3758# elif RTLNX_VER_MIN(2, 6, 6)
3759static ssize_t vbsf_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, loff_t offset, unsigned long nr_segs)
3760# elif RTLNX_VER_MIN(2, 5, 55)
3761static int vbsf_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, loff_t offset, unsigned long nr_segs)
3762# elif RTLNX_VER_MIN(2, 5, 41)
3763static int vbsf_direct_IO(int rw, struct file *file, const struct iovec *iov, loff_t offset, unsigned long nr_segs)
3764# elif RTLNX_VER_MIN(2, 5, 35)
3765static int vbsf_direct_IO(int rw, struct inode *inode, const struct iovec *iov, loff_t offset, unsigned long nr_segs)
3766# elif RTLNX_VER_MIN(2, 5, 26)
3767static int vbsf_direct_IO(int rw, struct inode *inode, char *buf, loff_t offset, size_t count)
3768# elif LINUX_VERSION_CODE == KERNEL_VERSION(2, 4, 21) && defined(I_NEW) /* RHEL3 Frankenkernel. */
3769static int vbsf_direct_IO(int rw, struct file *file, struct kiobuf *buf, unsigned long whatever1, int whatever2)
3770# else
3771static int vbsf_direct_IO(int rw, struct inode *inode, struct kiobuf *buf, unsigned long whatever1, int whatever2)
3772# endif
3773{
3774 TRACE();
3775 return -EINVAL;
3776}
3777
3778#endif
3779
3780/**
3781 * Address space (for the page cache) operations for regular files.
3782 *
3783 * @todo the FsPerf touch/flush (mmap) test fails on 4.4.0 (ubuntu 16.04 lts).
3784 */
3785struct address_space_operations vbsf_reg_aops = {
3786 .readpage = vbsf_readpage,
3787 .writepage = vbsf_writepage,
3788 /** @todo Need .writepages if we want msync performance... */
3789#if RTLNX_VER_MIN(2,5,12)
3790 .set_page_dirty = __set_page_dirty_buffers,
3791#endif
3792#if RTLNX_VER_MIN(2,6,24)
3793 .write_begin = vbsf_write_begin,
3794 .write_end = simple_write_end,
3795#elif RTLNX_VER_MIN(2,5,45)
3796 .prepare_write = simple_prepare_write,
3797 .commit_write = simple_commit_write,
3798#endif
3799#if RTLNX_VER_MIN(2,4,10)
3800 .direct_IO = vbsf_direct_IO,
3801#endif
3802};
3803
Note: See TracBrowser for help on using the repository browser.

© 2023 Oracle
Contact – Privacy policy – Terms of Use