xref: /linux/fs/cachefiles/io.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* kiocb-using read/write
3  *
4  * Copyright (C) 2021 Red Hat, Inc. All Rights Reserved.
5  * Written by David Howells (dhowells@redhat.com)
6  */
7 
8 #include <linux/mount.h>
9 #include <linux/slab.h>
10 #include <linux/file.h>
11 #include <linux/uio.h>
12 #include <linux/bio.h>
13 #include <linux/falloc.h>
14 #include <linux/sched/mm.h>
15 #include <trace/events/fscache.h>
16 #include "internal.h"
17 
18 struct cachefiles_kiocb {
19 	struct kiocb		iocb;
20 	refcount_t		ki_refcnt;
21 	loff_t			start;
22 	union {
23 		size_t		skipped;
24 		size_t		len;
25 	};
26 	struct cachefiles_object *object;
27 	netfs_io_terminated_t	term_func;
28 	void			*term_func_priv;
29 	bool			was_async;
30 	unsigned int		inval_counter;	/* Copy of cookie->inval_counter */
31 	u64			b_writing;
32 };
33 
34 static inline void cachefiles_put_kiocb(struct cachefiles_kiocb *ki)
35 {
36 	if (refcount_dec_and_test(&ki->ki_refcnt)) {
37 		cachefiles_put_object(ki->object, cachefiles_obj_put_ioreq);
38 		fput(ki->iocb.ki_filp);
39 		kfree(ki);
40 	}
41 }
42 
43 /*
44  * Handle completion of a read from the cache.
45  */
46 static void cachefiles_read_complete(struct kiocb *iocb, long ret)
47 {
48 	struct cachefiles_kiocb *ki = container_of(iocb, struct cachefiles_kiocb, iocb);
49 	struct inode *inode = file_inode(ki->iocb.ki_filp);
50 
51 	_enter("%ld", ret);
52 
53 	if (ret < 0)
54 		trace_cachefiles_io_error(ki->object, inode, ret,
55 					  cachefiles_trace_read_error);
56 
57 	if (ki->term_func) {
58 		if (ret >= 0) {
59 			if (ki->object->cookie->inval_counter == ki->inval_counter)
60 				ki->skipped += ret;
61 			else
62 				ret = -ESTALE;
63 		}
64 
65 		ki->term_func(ki->term_func_priv, ret, ki->was_async);
66 	}
67 
68 	cachefiles_put_kiocb(ki);
69 }
70 
71 /*
72  * Initiate a read from the cache.
73  */
74 static int cachefiles_read(struct netfs_cache_resources *cres,
75 			   loff_t start_pos,
76 			   struct iov_iter *iter,
77 			   enum netfs_read_from_hole read_hole,
78 			   netfs_io_terminated_t term_func,
79 			   void *term_func_priv)
80 {
81 	struct cachefiles_object *object;
82 	struct cachefiles_kiocb *ki;
83 	struct file *file;
84 	unsigned int old_nofs;
85 	ssize_t ret = -ENOBUFS;
86 	size_t len = iov_iter_count(iter), skipped = 0;
87 
88 	if (!fscache_wait_for_operation(cres, FSCACHE_WANT_READ))
89 		goto presubmission_error;
90 
91 	fscache_count_read();
92 	object = cachefiles_cres_object(cres);
93 	file = cachefiles_cres_file(cres);
94 
95 	_enter("%pD,%li,%llx,%zx/%llx",
96 	       file, file_inode(file)->i_ino, start_pos, len,
97 	       i_size_read(file_inode(file)));
98 
99 	/* If the caller asked us to seek for data before doing the read, then
100 	 * we should do that now.  If we find a gap, we fill it with zeros.
101 	 */
102 	if (read_hole != NETFS_READ_HOLE_IGNORE) {
103 		loff_t off = start_pos, off2;
104 
105 		off2 = cachefiles_inject_read_error();
106 		if (off2 == 0)
107 			off2 = vfs_llseek(file, off, SEEK_DATA);
108 		if (off2 < 0 && off2 >= (loff_t)-MAX_ERRNO && off2 != -ENXIO) {
109 			skipped = 0;
110 			ret = off2;
111 			goto presubmission_error;
112 		}
113 
114 		if (off2 == -ENXIO || off2 >= start_pos + len) {
115 			/* The region is beyond the EOF or there's no more data
116 			 * in the region, so clear the rest of the buffer and
117 			 * return success.
118 			 */
119 			ret = -ENODATA;
120 			if (read_hole == NETFS_READ_HOLE_FAIL)
121 				goto presubmission_error;
122 
123 			iov_iter_zero(len, iter);
124 			skipped = len;
125 			ret = 0;
126 			goto presubmission_error;
127 		}
128 
129 		skipped = off2 - off;
130 		iov_iter_zero(skipped, iter);
131 	}
132 
133 	ret = -ENOMEM;
134 	ki = kzalloc(sizeof(struct cachefiles_kiocb), GFP_KERNEL);
135 	if (!ki)
136 		goto presubmission_error;
137 
138 	refcount_set(&ki->ki_refcnt, 2);
139 	ki->iocb.ki_filp	= file;
140 	ki->iocb.ki_pos		= start_pos + skipped;
141 	ki->iocb.ki_flags	= IOCB_DIRECT;
142 	ki->iocb.ki_ioprio	= get_current_ioprio();
143 	ki->skipped		= skipped;
144 	ki->object		= object;
145 	ki->inval_counter	= cres->inval_counter;
146 	ki->term_func		= term_func;
147 	ki->term_func_priv	= term_func_priv;
148 	ki->was_async		= true;
149 
150 	if (ki->term_func)
151 		ki->iocb.ki_complete = cachefiles_read_complete;
152 
153 	get_file(ki->iocb.ki_filp);
154 	cachefiles_grab_object(object, cachefiles_obj_get_ioreq);
155 
156 	trace_cachefiles_read(object, file_inode(file), ki->iocb.ki_pos, len - skipped);
157 	old_nofs = memalloc_nofs_save();
158 	ret = cachefiles_inject_read_error();
159 	if (ret == 0)
160 		ret = vfs_iocb_iter_read(file, &ki->iocb, iter);
161 	memalloc_nofs_restore(old_nofs);
162 	switch (ret) {
163 	case -EIOCBQUEUED:
164 		goto in_progress;
165 
166 	case -ERESTARTSYS:
167 	case -ERESTARTNOINTR:
168 	case -ERESTARTNOHAND:
169 	case -ERESTART_RESTARTBLOCK:
170 		/* There's no easy way to restart the syscall since other AIO's
171 		 * may be already running. Just fail this IO with EINTR.
172 		 */
173 		ret = -EINTR;
174 		fallthrough;
175 	default:
176 		ki->was_async = false;
177 		cachefiles_read_complete(&ki->iocb, ret);
178 		if (ret > 0)
179 			ret = 0;
180 		break;
181 	}
182 
183 in_progress:
184 	cachefiles_put_kiocb(ki);
185 	_leave(" = %zd", ret);
186 	return ret;
187 
188 presubmission_error:
189 	if (term_func)
190 		term_func(term_func_priv, ret < 0 ? ret : skipped, false);
191 	return ret;
192 }
193 
194 /*
195  * Query the occupancy of the cache in a region, returning where the next chunk
196  * of data starts and how long it is.
197  */
198 static int cachefiles_query_occupancy(struct netfs_cache_resources *cres,
199 				      loff_t start, size_t len, size_t granularity,
200 				      loff_t *_data_start, size_t *_data_len)
201 {
202 	struct cachefiles_object *object;
203 	struct file *file;
204 	loff_t off, off2;
205 
206 	*_data_start = -1;
207 	*_data_len = 0;
208 
209 	if (!fscache_wait_for_operation(cres, FSCACHE_WANT_READ))
210 		return -ENOBUFS;
211 
212 	object = cachefiles_cres_object(cres);
213 	file = cachefiles_cres_file(cres);
214 	granularity = max_t(size_t, object->volume->cache->bsize, granularity);
215 
216 	_enter("%pD,%li,%llx,%zx/%llx",
217 	       file, file_inode(file)->i_ino, start, len,
218 	       i_size_read(file_inode(file)));
219 
220 	off = cachefiles_inject_read_error();
221 	if (off == 0)
222 		off = vfs_llseek(file, start, SEEK_DATA);
223 	if (off == -ENXIO)
224 		return -ENODATA; /* Beyond EOF */
225 	if (off < 0 && off >= (loff_t)-MAX_ERRNO)
226 		return -ENOBUFS; /* Error. */
227 	if (round_up(off, granularity) >= start + len)
228 		return -ENODATA; /* No data in range */
229 
230 	off2 = cachefiles_inject_read_error();
231 	if (off2 == 0)
232 		off2 = vfs_llseek(file, off, SEEK_HOLE);
233 	if (off2 == -ENXIO)
234 		return -ENODATA; /* Beyond EOF */
235 	if (off2 < 0 && off2 >= (loff_t)-MAX_ERRNO)
236 		return -ENOBUFS; /* Error. */
237 
238 	/* Round away partial blocks */
239 	off = round_up(off, granularity);
240 	off2 = round_down(off2, granularity);
241 	if (off2 <= off)
242 		return -ENODATA;
243 
244 	*_data_start = off;
245 	if (off2 > start + len)
246 		*_data_len = len;
247 	else
248 		*_data_len = off2 - off;
249 	return 0;
250 }
251 
252 /*
253  * Handle completion of a write to the cache.
254  */
255 static void cachefiles_write_complete(struct kiocb *iocb, long ret)
256 {
257 	struct cachefiles_kiocb *ki = container_of(iocb, struct cachefiles_kiocb, iocb);
258 	struct cachefiles_object *object = ki->object;
259 	struct inode *inode = file_inode(ki->iocb.ki_filp);
260 
261 	_enter("%ld", ret);
262 
263 	if (ki->was_async)
264 		kiocb_end_write(iocb);
265 
266 	if (ret < 0)
267 		trace_cachefiles_io_error(object, inode, ret,
268 					  cachefiles_trace_write_error);
269 
270 	atomic_long_sub(ki->b_writing, &object->volume->cache->b_writing);
271 	set_bit(FSCACHE_COOKIE_HAVE_DATA, &object->cookie->flags);
272 	if (ki->term_func)
273 		ki->term_func(ki->term_func_priv, ret, ki->was_async);
274 	cachefiles_put_kiocb(ki);
275 }
276 
277 /*
278  * Initiate a write to the cache.
279  */
280 int __cachefiles_write(struct cachefiles_object *object,
281 		       struct file *file,
282 		       loff_t start_pos,
283 		       struct iov_iter *iter,
284 		       netfs_io_terminated_t term_func,
285 		       void *term_func_priv)
286 {
287 	struct cachefiles_cache *cache;
288 	struct cachefiles_kiocb *ki;
289 	unsigned int old_nofs;
290 	ssize_t ret;
291 	size_t len = iov_iter_count(iter);
292 
293 	fscache_count_write();
294 	cache = object->volume->cache;
295 
296 	_enter("%pD,%li,%llx,%zx/%llx",
297 	       file, file_inode(file)->i_ino, start_pos, len,
298 	       i_size_read(file_inode(file)));
299 
300 	ki = kzalloc(sizeof(struct cachefiles_kiocb), GFP_KERNEL);
301 	if (!ki) {
302 		if (term_func)
303 			term_func(term_func_priv, -ENOMEM, false);
304 		return -ENOMEM;
305 	}
306 
307 	refcount_set(&ki->ki_refcnt, 2);
308 	ki->iocb.ki_filp	= file;
309 	ki->iocb.ki_pos		= start_pos;
310 	ki->iocb.ki_flags	= IOCB_DIRECT | IOCB_WRITE;
311 	ki->iocb.ki_ioprio	= get_current_ioprio();
312 	ki->object		= object;
313 	ki->start		= start_pos;
314 	ki->len			= len;
315 	ki->term_func		= term_func;
316 	ki->term_func_priv	= term_func_priv;
317 	ki->was_async		= true;
318 	ki->b_writing		= (len + (1 << cache->bshift) - 1) >> cache->bshift;
319 
320 	if (ki->term_func)
321 		ki->iocb.ki_complete = cachefiles_write_complete;
322 	atomic_long_add(ki->b_writing, &cache->b_writing);
323 
324 	get_file(ki->iocb.ki_filp);
325 	cachefiles_grab_object(object, cachefiles_obj_get_ioreq);
326 
327 	trace_cachefiles_write(object, file_inode(file), ki->iocb.ki_pos, len);
328 	old_nofs = memalloc_nofs_save();
329 	ret = cachefiles_inject_write_error();
330 	if (ret == 0)
331 		ret = vfs_iocb_iter_write(file, &ki->iocb, iter);
332 	memalloc_nofs_restore(old_nofs);
333 	switch (ret) {
334 	case -EIOCBQUEUED:
335 		goto in_progress;
336 
337 	case -ERESTARTSYS:
338 	case -ERESTARTNOINTR:
339 	case -ERESTARTNOHAND:
340 	case -ERESTART_RESTARTBLOCK:
341 		/* There's no easy way to restart the syscall since other AIO's
342 		 * may be already running. Just fail this IO with EINTR.
343 		 */
344 		ret = -EINTR;
345 		fallthrough;
346 	default:
347 		ki->was_async = false;
348 		cachefiles_write_complete(&ki->iocb, ret);
349 		if (ret > 0)
350 			ret = 0;
351 		break;
352 	}
353 
354 in_progress:
355 	cachefiles_put_kiocb(ki);
356 	_leave(" = %zd", ret);
357 	return ret;
358 }
359 
360 static int cachefiles_write(struct netfs_cache_resources *cres,
361 			    loff_t start_pos,
362 			    struct iov_iter *iter,
363 			    netfs_io_terminated_t term_func,
364 			    void *term_func_priv)
365 {
366 	if (!fscache_wait_for_operation(cres, FSCACHE_WANT_WRITE)) {
367 		if (term_func)
368 			term_func(term_func_priv, -ENOBUFS, false);
369 		return -ENOBUFS;
370 	}
371 
372 	return __cachefiles_write(cachefiles_cres_object(cres),
373 				  cachefiles_cres_file(cres),
374 				  start_pos, iter,
375 				  term_func, term_func_priv);
376 }
377 
378 static inline enum netfs_io_source
379 cachefiles_do_prepare_read(struct netfs_cache_resources *cres,
380 			   loff_t start, size_t *_len, loff_t i_size,
381 			   unsigned long *_flags, ino_t netfs_ino)
382 {
383 	enum cachefiles_prepare_read_trace why;
384 	struct cachefiles_object *object = NULL;
385 	struct cachefiles_cache *cache;
386 	struct fscache_cookie *cookie = fscache_cres_cookie(cres);
387 	const struct cred *saved_cred;
388 	struct file *file = cachefiles_cres_file(cres);
389 	enum netfs_io_source ret = NETFS_DOWNLOAD_FROM_SERVER;
390 	size_t len = *_len;
391 	loff_t off, to;
392 	ino_t ino = file ? file_inode(file)->i_ino : 0;
393 	int rc;
394 
395 	_enter("%zx @%llx/%llx", len, start, i_size);
396 
397 	if (start >= i_size) {
398 		ret = NETFS_FILL_WITH_ZEROES;
399 		why = cachefiles_trace_read_after_eof;
400 		goto out_no_object;
401 	}
402 
403 	if (test_bit(FSCACHE_COOKIE_NO_DATA_TO_READ, &cookie->flags)) {
404 		__set_bit(NETFS_SREQ_COPY_TO_CACHE, _flags);
405 		why = cachefiles_trace_read_no_data;
406 		if (!test_bit(NETFS_SREQ_ONDEMAND, _flags))
407 			goto out_no_object;
408 	}
409 
410 	/* The object and the file may be being created in the background. */
411 	if (!file) {
412 		why = cachefiles_trace_read_no_file;
413 		if (!fscache_wait_for_operation(cres, FSCACHE_WANT_READ))
414 			goto out_no_object;
415 		file = cachefiles_cres_file(cres);
416 		if (!file)
417 			goto out_no_object;
418 		ino = file_inode(file)->i_ino;
419 	}
420 
421 	object = cachefiles_cres_object(cres);
422 	cache = object->volume->cache;
423 	cachefiles_begin_secure(cache, &saved_cred);
424 retry:
425 	off = cachefiles_inject_read_error();
426 	if (off == 0)
427 		off = vfs_llseek(file, start, SEEK_DATA);
428 	if (off < 0 && off >= (loff_t)-MAX_ERRNO) {
429 		if (off == (loff_t)-ENXIO) {
430 			why = cachefiles_trace_read_seek_nxio;
431 			goto download_and_store;
432 		}
433 		trace_cachefiles_io_error(object, file_inode(file), off,
434 					  cachefiles_trace_seek_error);
435 		why = cachefiles_trace_read_seek_error;
436 		goto out;
437 	}
438 
439 	if (off >= start + len) {
440 		why = cachefiles_trace_read_found_hole;
441 		goto download_and_store;
442 	}
443 
444 	if (off > start) {
445 		off = round_up(off, cache->bsize);
446 		len = off - start;
447 		*_len = len;
448 		why = cachefiles_trace_read_found_part;
449 		goto download_and_store;
450 	}
451 
452 	to = cachefiles_inject_read_error();
453 	if (to == 0)
454 		to = vfs_llseek(file, start, SEEK_HOLE);
455 	if (to < 0 && to >= (loff_t)-MAX_ERRNO) {
456 		trace_cachefiles_io_error(object, file_inode(file), to,
457 					  cachefiles_trace_seek_error);
458 		why = cachefiles_trace_read_seek_error;
459 		goto out;
460 	}
461 
462 	if (to < start + len) {
463 		if (start + len >= i_size)
464 			to = round_up(to, cache->bsize);
465 		else
466 			to = round_down(to, cache->bsize);
467 		len = to - start;
468 		*_len = len;
469 	}
470 
471 	why = cachefiles_trace_read_have_data;
472 	ret = NETFS_READ_FROM_CACHE;
473 	goto out;
474 
475 download_and_store:
476 	__set_bit(NETFS_SREQ_COPY_TO_CACHE, _flags);
477 	if (test_bit(NETFS_SREQ_ONDEMAND, _flags)) {
478 		rc = cachefiles_ondemand_read(object, start, len);
479 		if (!rc) {
480 			__clear_bit(NETFS_SREQ_ONDEMAND, _flags);
481 			goto retry;
482 		}
483 		ret = NETFS_INVALID_READ;
484 	}
485 out:
486 	cachefiles_end_secure(cache, saved_cred);
487 out_no_object:
488 	trace_cachefiles_prep_read(object, start, len, *_flags, ret, why, ino, netfs_ino);
489 	return ret;
490 }
491 
492 /*
493  * Prepare a read operation, shortening it to a cached/uncached
494  * boundary as appropriate.
495  */
496 static enum netfs_io_source cachefiles_prepare_read(struct netfs_io_subrequest *subreq,
497 						    unsigned long long i_size)
498 {
499 	return cachefiles_do_prepare_read(&subreq->rreq->cache_resources,
500 					  subreq->start, &subreq->len, i_size,
501 					  &subreq->flags, subreq->rreq->inode->i_ino);
502 }
503 
504 /*
505  * Prepare an on-demand read operation, shortening it to a cached/uncached
506  * boundary as appropriate.
507  */
508 static enum netfs_io_source
509 cachefiles_prepare_ondemand_read(struct netfs_cache_resources *cres,
510 				 loff_t start, size_t *_len, loff_t i_size,
511 				 unsigned long *_flags, ino_t ino)
512 {
513 	return cachefiles_do_prepare_read(cres, start, _len, i_size, _flags, ino);
514 }
515 
516 /*
517  * Prepare for a write to occur.
518  */
519 int __cachefiles_prepare_write(struct cachefiles_object *object,
520 			       struct file *file,
521 			       loff_t *_start, size_t *_len, size_t upper_len,
522 			       bool no_space_allocated_yet)
523 {
524 	struct cachefiles_cache *cache = object->volume->cache;
525 	loff_t start = *_start, pos;
526 	size_t len = *_len;
527 	int ret;
528 
529 	/* Round to DIO size */
530 	start = round_down(*_start, PAGE_SIZE);
531 	if (start != *_start || *_len > upper_len) {
532 		/* Probably asked to cache a streaming write written into the
533 		 * pagecache when the cookie was temporarily out of service to
534 		 * culling.
535 		 */
536 		fscache_count_dio_misfit();
537 		return -ENOBUFS;
538 	}
539 
540 	*_len = round_up(len, PAGE_SIZE);
541 
542 	/* We need to work out whether there's sufficient disk space to perform
543 	 * the write - but we can skip that check if we have space already
544 	 * allocated.
545 	 */
546 	if (no_space_allocated_yet)
547 		goto check_space;
548 
549 	pos = cachefiles_inject_read_error();
550 	if (pos == 0)
551 		pos = vfs_llseek(file, start, SEEK_DATA);
552 	if (pos < 0 && pos >= (loff_t)-MAX_ERRNO) {
553 		if (pos == -ENXIO)
554 			goto check_space; /* Unallocated tail */
555 		trace_cachefiles_io_error(object, file_inode(file), pos,
556 					  cachefiles_trace_seek_error);
557 		return pos;
558 	}
559 	if ((u64)pos >= (u64)start + *_len)
560 		goto check_space; /* Unallocated region */
561 
562 	/* We have a block that's at least partially filled - if we're low on
563 	 * space, we need to see if it's fully allocated.  If it's not, we may
564 	 * want to cull it.
565 	 */
566 	if (cachefiles_has_space(cache, 0, *_len / PAGE_SIZE,
567 				 cachefiles_has_space_check) == 0)
568 		return 0; /* Enough space to simply overwrite the whole block */
569 
570 	pos = cachefiles_inject_read_error();
571 	if (pos == 0)
572 		pos = vfs_llseek(file, start, SEEK_HOLE);
573 	if (pos < 0 && pos >= (loff_t)-MAX_ERRNO) {
574 		trace_cachefiles_io_error(object, file_inode(file), pos,
575 					  cachefiles_trace_seek_error);
576 		return pos;
577 	}
578 	if ((u64)pos >= (u64)start + *_len)
579 		return 0; /* Fully allocated */
580 
581 	/* Partially allocated, but insufficient space: cull. */
582 	fscache_count_no_write_space();
583 	ret = cachefiles_inject_remove_error();
584 	if (ret == 0)
585 		ret = vfs_fallocate(file, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
586 				    start, *_len);
587 	if (ret < 0) {
588 		trace_cachefiles_io_error(object, file_inode(file), ret,
589 					  cachefiles_trace_fallocate_error);
590 		cachefiles_io_error_obj(object,
591 					"CacheFiles: fallocate failed (%d)\n", ret);
592 		ret = -EIO;
593 	}
594 
595 	return ret;
596 
597 check_space:
598 	return cachefiles_has_space(cache, 0, *_len / PAGE_SIZE,
599 				    cachefiles_has_space_for_write);
600 }
601 
602 static int cachefiles_prepare_write(struct netfs_cache_resources *cres,
603 				    loff_t *_start, size_t *_len, size_t upper_len,
604 				    loff_t i_size, bool no_space_allocated_yet)
605 {
606 	struct cachefiles_object *object = cachefiles_cres_object(cres);
607 	struct cachefiles_cache *cache = object->volume->cache;
608 	const struct cred *saved_cred;
609 	int ret;
610 
611 	if (!cachefiles_cres_file(cres)) {
612 		if (!fscache_wait_for_operation(cres, FSCACHE_WANT_WRITE))
613 			return -ENOBUFS;
614 		if (!cachefiles_cres_file(cres))
615 			return -ENOBUFS;
616 	}
617 
618 	cachefiles_begin_secure(cache, &saved_cred);
619 	ret = __cachefiles_prepare_write(object, cachefiles_cres_file(cres),
620 					 _start, _len, upper_len,
621 					 no_space_allocated_yet);
622 	cachefiles_end_secure(cache, saved_cred);
623 	return ret;
624 }
625 
626 static void cachefiles_prepare_write_subreq(struct netfs_io_subrequest *subreq)
627 {
628 	struct netfs_io_request *wreq = subreq->rreq;
629 	struct netfs_cache_resources *cres = &wreq->cache_resources;
630 	struct netfs_io_stream *stream = &wreq->io_streams[subreq->stream_nr];
631 
632 	_enter("W=%x[%x] %llx", wreq->debug_id, subreq->debug_index, subreq->start);
633 
634 	stream->sreq_max_len = MAX_RW_COUNT;
635 	stream->sreq_max_segs = BIO_MAX_VECS;
636 
637 	if (!cachefiles_cres_file(cres)) {
638 		if (!fscache_wait_for_operation(cres, FSCACHE_WANT_WRITE))
639 			return netfs_prepare_write_failed(subreq);
640 		if (!cachefiles_cres_file(cres))
641 			return netfs_prepare_write_failed(subreq);
642 	}
643 }
644 
645 static void cachefiles_issue_write(struct netfs_io_subrequest *subreq)
646 {
647 	struct netfs_io_request *wreq = subreq->rreq;
648 	struct netfs_cache_resources *cres = &wreq->cache_resources;
649 	struct cachefiles_object *object = cachefiles_cres_object(cres);
650 	struct cachefiles_cache *cache = object->volume->cache;
651 	struct netfs_io_stream *stream = &wreq->io_streams[subreq->stream_nr];
652 	const struct cred *saved_cred;
653 	size_t off, pre, post, len = subreq->len;
654 	loff_t start = subreq->start;
655 	int ret;
656 
657 	_enter("W=%x[%x] %llx-%llx",
658 	       wreq->debug_id, subreq->debug_index, start, start + len - 1);
659 
660 	/* We need to start on the cache granularity boundary */
661 	off = start & (CACHEFILES_DIO_BLOCK_SIZE - 1);
662 	if (off) {
663 		pre = CACHEFILES_DIO_BLOCK_SIZE - off;
664 		if (pre >= len) {
665 			fscache_count_dio_misfit();
666 			netfs_write_subrequest_terminated(subreq, len, false);
667 			return;
668 		}
669 		subreq->transferred += pre;
670 		start += pre;
671 		len -= pre;
672 		iov_iter_advance(&subreq->io_iter, pre);
673 	}
674 
675 	/* We also need to end on the cache granularity boundary */
676 	if (start + len == wreq->i_size) {
677 		size_t part = len % CACHEFILES_DIO_BLOCK_SIZE;
678 		size_t need = CACHEFILES_DIO_BLOCK_SIZE - part;
679 
680 		if (part && stream->submit_extendable_to >= need) {
681 			len += need;
682 			subreq->len += need;
683 			subreq->io_iter.count += need;
684 		}
685 	}
686 
687 	post = len & (CACHEFILES_DIO_BLOCK_SIZE - 1);
688 	if (post) {
689 		len -= post;
690 		if (len == 0) {
691 			fscache_count_dio_misfit();
692 			netfs_write_subrequest_terminated(subreq, post, false);
693 			return;
694 		}
695 		iov_iter_truncate(&subreq->io_iter, len);
696 	}
697 
698 	cachefiles_begin_secure(cache, &saved_cred);
699 	ret = __cachefiles_prepare_write(object, cachefiles_cres_file(cres),
700 					 &start, &len, len, true);
701 	cachefiles_end_secure(cache, saved_cred);
702 	if (ret < 0) {
703 		netfs_write_subrequest_terminated(subreq, ret, false);
704 		return;
705 	}
706 
707 	cachefiles_write(&subreq->rreq->cache_resources,
708 			 subreq->start, &subreq->io_iter,
709 			 netfs_write_subrequest_terminated, subreq);
710 }
711 
712 /*
713  * Clean up an operation.
714  */
715 static void cachefiles_end_operation(struct netfs_cache_resources *cres)
716 {
717 	struct file *file = cachefiles_cres_file(cres);
718 
719 	if (file)
720 		fput(file);
721 	fscache_end_cookie_access(fscache_cres_cookie(cres), fscache_access_io_end);
722 }
723 
724 static const struct netfs_cache_ops cachefiles_netfs_cache_ops = {
725 	.end_operation		= cachefiles_end_operation,
726 	.read			= cachefiles_read,
727 	.write			= cachefiles_write,
728 	.issue_write		= cachefiles_issue_write,
729 	.prepare_read		= cachefiles_prepare_read,
730 	.prepare_write		= cachefiles_prepare_write,
731 	.prepare_write_subreq	= cachefiles_prepare_write_subreq,
732 	.prepare_ondemand_read	= cachefiles_prepare_ondemand_read,
733 	.query_occupancy	= cachefiles_query_occupancy,
734 };
735 
736 /*
737  * Open the cache file when beginning a cache operation.
738  */
739 bool cachefiles_begin_operation(struct netfs_cache_resources *cres,
740 				enum fscache_want_state want_state)
741 {
742 	struct cachefiles_object *object = cachefiles_cres_object(cres);
743 
744 	if (!cachefiles_cres_file(cres)) {
745 		cres->ops = &cachefiles_netfs_cache_ops;
746 		if (object->file) {
747 			spin_lock(&object->lock);
748 			if (!cres->cache_priv2 && object->file)
749 				cres->cache_priv2 = get_file(object->file);
750 			spin_unlock(&object->lock);
751 		}
752 	}
753 
754 	if (!cachefiles_cres_file(cres) && want_state != FSCACHE_WANT_PARAMS) {
755 		pr_err("failed to get cres->file\n");
756 		return false;
757 	}
758 
759 	return true;
760 }
761