1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/*
3 * Berkeley style UIO structures - Alan Cox 1994.
4 */
5#ifndef __LINUX_UIO_H
6#define __LINUX_UIO_H
7
8#include <linux/kernel.h>
9#include <linux/mm_types.h>
10#include <linux/ucopysize.h>
11#include <uapi/linux/uio.h>
12
13struct page;
14struct folio_queue;
15
16typedef unsigned int __bitwise iov_iter_extraction_t;
17
18struct kvec {
19 void *iov_base; /* and that should *never* hold a userland pointer */
20 size_t iov_len;
21};
22
23enum iter_type {
24 /* iter types */
25 ITER_UBUF,
26 ITER_IOVEC,
27 ITER_BVEC,
28 ITER_KVEC,
29 ITER_FOLIOQ,
30 ITER_XARRAY,
31 ITER_DISCARD,
32};
33
34#define ITER_SOURCE 1 // == WRITE
35#define ITER_DEST 0 // == READ
36
37struct iov_iter_state {
38 size_t iov_offset;
39 size_t count;
40 unsigned long nr_segs;
41};
42
43struct iov_iter {
44 u8 iter_type;
45 bool nofault;
46 bool data_source;
47 size_t iov_offset;
48 /*
49 * Hack alert: overlay ubuf_iovec with iovec + count, so
50 * that the members resolve correctly regardless of the type
51 * of iterator used. This means that you can use:
52 *
53 * &iter->__ubuf_iovec or iter->__iov
54 *
55 * interchangably for the user_backed cases, hence simplifying
56 * some of the cases that need to deal with both.
57 */
58 union {
59 /*
60 * This really should be a const, but we cannot do that without
61 * also modifying any of the zero-filling iter init functions.
62 * Leave it non-const for now, but it should be treated as such.
63 */
64 struct iovec __ubuf_iovec;
65 struct {
66 union {
67 /* use iter_iov() to get the current vec */
68 const struct iovec *__iov;
69 const struct kvec *kvec;
70 const struct bio_vec *bvec;
71 const struct folio_queue *folioq;
72 struct xarray *xarray;
73 void __user *ubuf;
74 };
75 size_t count;
76 };
77 };
78 union {
79 unsigned long nr_segs;
80 u8 folioq_slot;
81 loff_t xarray_start;
82 };
83};
84
85typedef __u16 uio_meta_flags_t;
86
87struct uio_meta {
88 uio_meta_flags_t flags;
89 u16 app_tag;
90 u64 seed;
91 struct iov_iter iter;
92};
93
94static inline const struct iovec *iter_iov(const struct iov_iter *iter)
95{
96 if (iter->iter_type == ITER_UBUF)
97 return (const struct iovec *) &iter->__ubuf_iovec;
98 return iter->__iov;
99}
100
101#define iter_iov_addr(iter) (iter_iov(iter)->iov_base + (iter)->iov_offset)
102
103static inline size_t iter_iov_len(const struct iov_iter *i)
104{
105 if (i->iter_type == ITER_UBUF)
106 return i->count;
107 return iter_iov(iter: i)->iov_len - i->iov_offset;
108}
109
110static inline enum iter_type iov_iter_type(const struct iov_iter *i)
111{
112 return i->iter_type;
113}
114
115static inline void iov_iter_save_state(struct iov_iter *iter,
116 struct iov_iter_state *state)
117{
118 state->iov_offset = iter->iov_offset;
119 state->count = iter->count;
120 state->nr_segs = iter->nr_segs;
121}
122
123static inline bool iter_is_ubuf(const struct iov_iter *i)
124{
125 return iov_iter_type(i) == ITER_UBUF;
126}
127
128static inline bool iter_is_iovec(const struct iov_iter *i)
129{
130 return iov_iter_type(i) == ITER_IOVEC;
131}
132
133static inline bool iov_iter_is_kvec(const struct iov_iter *i)
134{
135 return iov_iter_type(i) == ITER_KVEC;
136}
137
138static inline bool iov_iter_is_bvec(const struct iov_iter *i)
139{
140 return iov_iter_type(i) == ITER_BVEC;
141}
142
143static inline bool iov_iter_is_discard(const struct iov_iter *i)
144{
145 return iov_iter_type(i) == ITER_DISCARD;
146}
147
148static inline bool iov_iter_is_folioq(const struct iov_iter *i)
149{
150 return iov_iter_type(i) == ITER_FOLIOQ;
151}
152
153static inline bool iov_iter_is_xarray(const struct iov_iter *i)
154{
155 return iov_iter_type(i) == ITER_XARRAY;
156}
157
158static inline unsigned char iov_iter_rw(const struct iov_iter *i)
159{
160 return i->data_source ? WRITE : READ;
161}
162
163static inline bool user_backed_iter(const struct iov_iter *i)
164{
165 return iter_is_ubuf(i) || iter_is_iovec(i);
166}
167
168/*
169 * Total number of bytes covered by an iovec.
170 *
171 * NOTE that it is not safe to use this function until all the iovec's
172 * segment lengths have been validated. Because the individual lengths can
173 * overflow a size_t when added together.
174 */
175static inline size_t iov_length(const struct iovec *iov, unsigned long nr_segs)
176{
177 unsigned long seg;
178 size_t ret = 0;
179
180 for (seg = 0; seg < nr_segs; seg++)
181 ret += iov[seg].iov_len;
182 return ret;
183}
184
185void iov_iter_advance(struct iov_iter *i, size_t bytes);
186void iov_iter_revert(struct iov_iter *i, size_t bytes);
187size_t fault_in_iov_iter_readable(const struct iov_iter *i, size_t bytes);
188size_t fault_in_iov_iter_writeable(const struct iov_iter *i, size_t bytes);
189size_t iov_iter_single_seg_count(const struct iov_iter *i);
190size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
191 struct iov_iter *i);
192size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
193 struct iov_iter *i);
194size_t copy_folio_from_iter_atomic(struct folio *folio, size_t offset,
195 size_t bytes, struct iov_iter *i);
196
197size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i);
198size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i);
199size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i);
200
201static inline size_t copy_folio_to_iter(struct folio *folio, size_t offset,
202 size_t bytes, struct iov_iter *i)
203{
204 return copy_page_to_iter(page: &folio->page, offset, bytes, i);
205}
206
207static inline size_t copy_folio_from_iter(struct folio *folio, size_t offset,
208 size_t bytes, struct iov_iter *i)
209{
210 return copy_page_from_iter(page: &folio->page, offset, bytes, i);
211}
212
213size_t copy_page_to_iter_nofault(struct page *page, unsigned offset,
214 size_t bytes, struct iov_iter *i);
215
216static __always_inline __must_check
217size_t copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
218{
219 if (check_copy_size(addr, bytes, is_source: true))
220 return _copy_to_iter(addr, bytes, i);
221 return 0;
222}
223
224static __always_inline __must_check
225size_t copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
226{
227 if (check_copy_size(addr, bytes, is_source: false))
228 return _copy_from_iter(addr, bytes, i);
229 return 0;
230}
231
232static __always_inline __must_check
233bool copy_to_iter_full(const void *addr, size_t bytes, struct iov_iter *i)
234{
235 size_t copied = copy_to_iter(addr, bytes, i);
236 if (likely(copied == bytes))
237 return true;
238 iov_iter_revert(i, bytes: copied);
239 return false;
240}
241
242static __always_inline __must_check
243bool copy_from_iter_full(void *addr, size_t bytes, struct iov_iter *i)
244{
245 size_t copied = copy_from_iter(addr, bytes, i);
246 if (likely(copied == bytes))
247 return true;
248 iov_iter_revert(i, bytes: copied);
249 return false;
250}
251
252static __always_inline __must_check
253size_t copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
254{
255 if (check_copy_size(addr, bytes, is_source: false))
256 return _copy_from_iter_nocache(addr, bytes, i);
257 return 0;
258}
259
260static __always_inline __must_check
261bool copy_from_iter_full_nocache(void *addr, size_t bytes, struct iov_iter *i)
262{
263 size_t copied = copy_from_iter_nocache(addr, bytes, i);
264 if (likely(copied == bytes))
265 return true;
266 iov_iter_revert(i, bytes: copied);
267 return false;
268}
269
270#ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
271/*
272 * Note, users like pmem that depend on the stricter semantics of
273 * _copy_from_iter_flushcache() than _copy_from_iter_nocache() must check for
274 * IS_ENABLED(CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE) before assuming that the
275 * destination is flushed from the cache on return.
276 */
277size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i);
278#else
279#define _copy_from_iter_flushcache _copy_from_iter_nocache
280#endif
281
282#ifdef CONFIG_ARCH_HAS_COPY_MC
283size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i);
284#else
285#define _copy_mc_to_iter _copy_to_iter
286#endif
287
288size_t iov_iter_zero(size_t bytes, struct iov_iter *);
289unsigned long iov_iter_alignment(const struct iov_iter *i);
290unsigned long iov_iter_gap_alignment(const struct iov_iter *i);
291void iov_iter_init(struct iov_iter *i, unsigned int direction, const struct iovec *iov,
292 unsigned long nr_segs, size_t count);
293void iov_iter_kvec(struct iov_iter *i, unsigned int direction, const struct kvec *kvec,
294 unsigned long nr_segs, size_t count);
295void iov_iter_bvec(struct iov_iter *i, unsigned int direction, const struct bio_vec *bvec,
296 unsigned long nr_segs, size_t count);
297void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count);
298void iov_iter_folio_queue(struct iov_iter *i, unsigned int direction,
299 const struct folio_queue *folioq,
300 unsigned int first_slot, unsigned int offset, size_t count);
301void iov_iter_xarray(struct iov_iter *i, unsigned int direction, struct xarray *xarray,
302 loff_t start, size_t count);
303ssize_t iov_iter_get_pages2(struct iov_iter *i, struct page **pages,
304 size_t maxsize, unsigned maxpages, size_t *start);
305ssize_t iov_iter_get_pages_alloc2(struct iov_iter *i, struct page ***pages,
306 size_t maxsize, size_t *start);
307int iov_iter_npages(const struct iov_iter *i, int maxpages);
308void iov_iter_restore(struct iov_iter *i, struct iov_iter_state *state);
309
310const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags);
311
312static inline size_t iov_iter_count(const struct iov_iter *i)
313{
314 return i->count;
315}
316
317/*
318 * Cap the iov_iter by given limit; note that the second argument is
319 * *not* the new size - it's upper limit for such. Passing it a value
320 * greater than the amount of data in iov_iter is fine - it'll just do
321 * nothing in that case.
322 */
323static inline void iov_iter_truncate(struct iov_iter *i, u64 count)
324{
325 /*
326 * count doesn't have to fit in size_t - comparison extends both
327 * operands to u64 here and any value that would be truncated by
328 * conversion in assignement is by definition greater than all
329 * values of size_t, including old i->count.
330 */
331 if (i->count > count)
332 i->count = count;
333}
334
335/*
336 * reexpand a previously truncated iterator; count must be no more than how much
337 * we had shrunk it.
338 */
339static inline void iov_iter_reexpand(struct iov_iter *i, size_t count)
340{
341 i->count = count;
342}
343
344static inline int
345iov_iter_npages_cap(struct iov_iter *i, int maxpages, size_t max_bytes)
346{
347 size_t shorted = 0;
348 int npages;
349
350 if (iov_iter_count(i) > max_bytes) {
351 shorted = iov_iter_count(i) - max_bytes;
352 iov_iter_truncate(i, count: max_bytes);
353 }
354 npages = iov_iter_npages(i, maxpages);
355 if (shorted)
356 iov_iter_reexpand(i, count: iov_iter_count(i) + shorted);
357
358 return npages;
359}
360
361struct iovec *iovec_from_user(const struct iovec __user *uvector,
362 unsigned long nr_segs, unsigned long fast_segs,
363 struct iovec *fast_iov, bool compat);
364ssize_t import_iovec(int type, const struct iovec __user *uvec,
365 unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
366 struct iov_iter *i);
367ssize_t __import_iovec(int type, const struct iovec __user *uvec,
368 unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
369 struct iov_iter *i, bool compat);
370int import_ubuf(int type, void __user *buf, size_t len, struct iov_iter *i);
371
372static inline void iov_iter_ubuf(struct iov_iter *i, unsigned int direction,
373 void __user *buf, size_t count)
374{
375 WARN_ON(direction & ~(READ | WRITE));
376 *i = (struct iov_iter) {
377 .iter_type = ITER_UBUF,
378 .data_source = direction,
379 .ubuf = buf,
380 .count = count,
381 .nr_segs = 1
382 };
383}
384/* Flags for iov_iter_get/extract_pages*() */
385/* Allow P2PDMA on the extracted pages */
386#define ITER_ALLOW_P2PDMA ((__force iov_iter_extraction_t)0x01)
387
388ssize_t iov_iter_extract_pages(struct iov_iter *i, struct page ***pages,
389 size_t maxsize, unsigned int maxpages,
390 iov_iter_extraction_t extraction_flags,
391 size_t *offset0);
392
393/**
394 * iov_iter_extract_will_pin - Indicate how pages from the iterator will be retained
395 * @iter: The iterator
396 *
397 * Examine the iterator and indicate by returning true or false as to how, if
398 * at all, pages extracted from the iterator will be retained by the extraction
399 * function.
400 *
401 * %true indicates that the pages will have a pin placed in them that the
402 * caller must unpin. This is must be done for DMA/async DIO to force fork()
403 * to forcibly copy a page for the child (the parent must retain the original
404 * page).
405 *
406 * %false indicates that no measures are taken and that it's up to the caller
407 * to retain the pages.
408 */
409static inline bool iov_iter_extract_will_pin(const struct iov_iter *iter)
410{
411 return user_backed_iter(i: iter);
412}
413
414struct sg_table;
415ssize_t extract_iter_to_sg(struct iov_iter *iter, size_t len,
416 struct sg_table *sgtable, unsigned int sg_max,
417 iov_iter_extraction_t extraction_flags);
418
419#endif
420