xref: /linux/fs/netfs/write_issue.c (revision 2408a807bfc3f738850ef5ad5e3fd59d66168996)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Network filesystem high-level (buffered) writeback.
3  *
4  * Copyright (C) 2024 Red Hat, Inc. All Rights Reserved.
5  * Written by David Howells (dhowells@redhat.com)
6  *
7  *
8  * To support network filesystems with local caching, we manage a situation
9  * that can be envisioned like the following:
10  *
11  *               +---+---+-----+-----+---+----------+
12  *    Folios:    |   |   |     |     |   |          |
13  *               +---+---+-----+-----+---+----------+
14  *
15  *                 +------+------+     +----+----+
16  *    Upload:      |      |      |.....|    |    |
17  *  (Stream 0)     +------+------+     +----+----+
18  *
19  *               +------+------+------+------+------+
20  *    Cache:     |      |      |      |      |      |
21  *  (Stream 1)   +------+------+------+------+------+
22  *
23  * Where we have a sequence of folios of varying sizes that we need to overlay
24  * with multiple parallel streams of I/O requests, where the I/O requests in a
25  * stream may also be of various sizes (in cifs, for example, the sizes are
26  * negotiated with the server; in something like ceph, they may represent the
27  * sizes of storage objects).
28  *
29  * The sequence in each stream may contain gaps and noncontiguous subrequests
30  * may be glued together into single vectored write RPCs.
31  */
32 
33 #include <linux/export.h>
34 #include <linux/fs.h>
35 #include <linux/mm.h>
36 #include <linux/pagemap.h>
37 #include "internal.h"
38 
39 /*
40  * Kill all dirty folios in the event of an unrecoverable error, starting with
41  * a locked folio we've already obtained from writeback_iter().
42  */
netfs_kill_dirty_pages(struct address_space * mapping,struct writeback_control * wbc,struct folio * folio)43 static void netfs_kill_dirty_pages(struct address_space *mapping,
44 				   struct writeback_control *wbc,
45 				   struct folio *folio)
46 {
47 	int error = 0;
48 
49 	do {
50 		enum netfs_folio_trace why = netfs_folio_trace_kill;
51 		struct netfs_group *group = NULL;
52 		struct netfs_folio *finfo = NULL;
53 		void *priv;
54 
55 		priv = folio_detach_private(folio);
56 		if (priv) {
57 			finfo = __netfs_folio_info(priv);
58 			if (finfo) {
59 				/* Kill folio from streaming write. */
60 				group = finfo->netfs_group;
61 				why = netfs_folio_trace_kill_s;
62 			} else {
63 				group = priv;
64 				if (group == NETFS_FOLIO_COPY_TO_CACHE) {
65 					/* Kill copy-to-cache folio */
66 					why = netfs_folio_trace_kill_cc;
67 					group = NULL;
68 				} else {
69 					/* Kill folio with group */
70 					why = netfs_folio_trace_kill_g;
71 				}
72 			}
73 		}
74 
75 		trace_netfs_folio(folio, why);
76 
77 		folio_start_writeback(folio);
78 		folio_unlock(folio);
79 		folio_end_writeback(folio);
80 
81 		netfs_put_group(group);
82 		kfree(finfo);
83 
84 	} while ((folio = writeback_iter(mapping, wbc, folio, &error)));
85 }
86 
87 /*
88  * Create a write request and set it up appropriately for the origin type.
89  */
netfs_create_write_req(struct address_space * mapping,struct file * file,loff_t start,enum netfs_io_origin origin)90 struct netfs_io_request *netfs_create_write_req(struct address_space *mapping,
91 						struct file *file,
92 						loff_t start,
93 						enum netfs_io_origin origin)
94 {
95 	struct netfs_io_request *wreq;
96 	struct netfs_inode *ictx;
97 	bool is_cacheable = (origin == NETFS_WRITEBACK ||
98 			     origin == NETFS_WRITEBACK_SINGLE ||
99 			     origin == NETFS_WRITETHROUGH ||
100 			     origin == NETFS_PGPRIV2_COPY_TO_CACHE);
101 
102 	wreq = netfs_alloc_request(mapping, file, start, 0, origin);
103 	if (IS_ERR(wreq))
104 		return wreq;
105 
106 	_enter("R=%x", wreq->debug_id);
107 
108 	ictx = netfs_inode(wreq->inode);
109 	if (is_cacheable && netfs_is_cache_enabled(ictx))
110 		fscache_begin_write_operation(&wreq->cache_resources, netfs_i_cookie(ictx));
111 	if (rolling_buffer_init(&wreq->buffer, wreq->debug_id, ITER_SOURCE) < 0)
112 		goto nomem;
113 
114 	wreq->cleaned_to = wreq->start;
115 
116 	wreq->io_streams[0].stream_nr		= 0;
117 	wreq->io_streams[0].source		= NETFS_UPLOAD_TO_SERVER;
118 	wreq->io_streams[0].prepare_write	= ictx->ops->prepare_write;
119 	wreq->io_streams[0].issue_write		= ictx->ops->issue_write;
120 	wreq->io_streams[0].collected_to	= start;
121 	wreq->io_streams[0].transferred		= LONG_MAX;
122 
123 	wreq->io_streams[1].stream_nr		= 1;
124 	wreq->io_streams[1].source		= NETFS_WRITE_TO_CACHE;
125 	wreq->io_streams[1].collected_to	= start;
126 	wreq->io_streams[1].transferred		= LONG_MAX;
127 	if (fscache_resources_valid(&wreq->cache_resources)) {
128 		wreq->io_streams[1].avail	= true;
129 		wreq->io_streams[1].active	= true;
130 		wreq->io_streams[1].prepare_write = wreq->cache_resources.ops->prepare_write_subreq;
131 		wreq->io_streams[1].issue_write = wreq->cache_resources.ops->issue_write;
132 	}
133 
134 	return wreq;
135 nomem:
136 	wreq->error = -ENOMEM;
137 	netfs_put_request(wreq, false, netfs_rreq_trace_put_failed);
138 	return ERR_PTR(-ENOMEM);
139 }
140 
141 /**
142  * netfs_prepare_write_failed - Note write preparation failed
143  * @subreq: The subrequest to mark
144  *
145  * Mark a subrequest to note that preparation for write failed.
146  */
netfs_prepare_write_failed(struct netfs_io_subrequest * subreq)147 void netfs_prepare_write_failed(struct netfs_io_subrequest *subreq)
148 {
149 	__set_bit(NETFS_SREQ_FAILED, &subreq->flags);
150 	trace_netfs_sreq(subreq, netfs_sreq_trace_prep_failed);
151 }
152 EXPORT_SYMBOL(netfs_prepare_write_failed);
153 
154 /*
155  * Prepare a write subrequest.  We need to allocate a new subrequest
156  * if we don't have one.
157  */
netfs_prepare_write(struct netfs_io_request * wreq,struct netfs_io_stream * stream,loff_t start)158 static void netfs_prepare_write(struct netfs_io_request *wreq,
159 				struct netfs_io_stream *stream,
160 				loff_t start)
161 {
162 	struct netfs_io_subrequest *subreq;
163 	struct iov_iter *wreq_iter = &wreq->buffer.iter;
164 
165 	/* Make sure we don't point the iterator at a used-up folio_queue
166 	 * struct being used as a placeholder to prevent the queue from
167 	 * collapsing.  In such a case, extend the queue.
168 	 */
169 	if (iov_iter_is_folioq(wreq_iter) &&
170 	    wreq_iter->folioq_slot >= folioq_nr_slots(wreq_iter->folioq))
171 		rolling_buffer_make_space(&wreq->buffer);
172 
173 	subreq = netfs_alloc_subrequest(wreq);
174 	subreq->source		= stream->source;
175 	subreq->start		= start;
176 	subreq->stream_nr	= stream->stream_nr;
177 	subreq->io_iter		= *wreq_iter;
178 
179 	_enter("R=%x[%x]", wreq->debug_id, subreq->debug_index);
180 
181 	trace_netfs_sreq(subreq, netfs_sreq_trace_prepare);
182 
183 	stream->sreq_max_len	= UINT_MAX;
184 	stream->sreq_max_segs	= INT_MAX;
185 	switch (stream->source) {
186 	case NETFS_UPLOAD_TO_SERVER:
187 		netfs_stat(&netfs_n_wh_upload);
188 		stream->sreq_max_len = wreq->wsize;
189 		break;
190 	case NETFS_WRITE_TO_CACHE:
191 		netfs_stat(&netfs_n_wh_write);
192 		break;
193 	default:
194 		WARN_ON_ONCE(1);
195 		break;
196 	}
197 
198 	if (stream->prepare_write)
199 		stream->prepare_write(subreq);
200 
201 	__set_bit(NETFS_SREQ_IN_PROGRESS, &subreq->flags);
202 
203 	/* We add to the end of the list whilst the collector may be walking
204 	 * the list.  The collector only goes nextwards and uses the lock to
205 	 * remove entries off of the front.
206 	 */
207 	spin_lock(&wreq->lock);
208 	list_add_tail(&subreq->rreq_link, &stream->subrequests);
209 	if (list_is_first(&subreq->rreq_link, &stream->subrequests)) {
210 		stream->front = subreq;
211 		if (!stream->active) {
212 			stream->collected_to = stream->front->start;
213 			/* Write list pointers before active flag */
214 			smp_store_release(&stream->active, true);
215 		}
216 	}
217 
218 	spin_unlock(&wreq->lock);
219 
220 	stream->construct = subreq;
221 }
222 
223 /*
224  * Set the I/O iterator for the filesystem/cache to use and dispatch the I/O
225  * operation.  The operation may be asynchronous and should call
226  * netfs_write_subrequest_terminated() when complete.
227  */
netfs_do_issue_write(struct netfs_io_stream * stream,struct netfs_io_subrequest * subreq)228 static void netfs_do_issue_write(struct netfs_io_stream *stream,
229 				 struct netfs_io_subrequest *subreq)
230 {
231 	struct netfs_io_request *wreq = subreq->rreq;
232 
233 	_enter("R=%x[%x],%zx", wreq->debug_id, subreq->debug_index, subreq->len);
234 
235 	if (test_bit(NETFS_SREQ_FAILED, &subreq->flags))
236 		return netfs_write_subrequest_terminated(subreq, subreq->error, false);
237 
238 	trace_netfs_sreq(subreq, netfs_sreq_trace_submit);
239 	stream->issue_write(subreq);
240 }
241 
netfs_reissue_write(struct netfs_io_stream * stream,struct netfs_io_subrequest * subreq,struct iov_iter * source)242 void netfs_reissue_write(struct netfs_io_stream *stream,
243 			 struct netfs_io_subrequest *subreq,
244 			 struct iov_iter *source)
245 {
246 	size_t size = subreq->len - subreq->transferred;
247 
248 	// TODO: Use encrypted buffer
249 	subreq->io_iter = *source;
250 	iov_iter_advance(source, size);
251 	iov_iter_truncate(&subreq->io_iter, size);
252 
253 	subreq->retry_count++;
254 	__clear_bit(NETFS_SREQ_MADE_PROGRESS, &subreq->flags);
255 	__set_bit(NETFS_SREQ_IN_PROGRESS, &subreq->flags);
256 	netfs_stat(&netfs_n_wh_retry_write_subreq);
257 	netfs_do_issue_write(stream, subreq);
258 }
259 
netfs_issue_write(struct netfs_io_request * wreq,struct netfs_io_stream * stream)260 void netfs_issue_write(struct netfs_io_request *wreq,
261 		       struct netfs_io_stream *stream)
262 {
263 	struct netfs_io_subrequest *subreq = stream->construct;
264 
265 	if (!subreq)
266 		return;
267 	stream->construct = NULL;
268 	subreq->io_iter.count = subreq->len;
269 	netfs_do_issue_write(stream, subreq);
270 }
271 
272 /*
273  * Add data to the write subrequest, dispatching each as we fill it up or if it
274  * is discontiguous with the previous.  We only fill one part at a time so that
275  * we can avoid overrunning the credits obtained (cifs) and try to parallelise
276  * content-crypto preparation with network writes.
277  */
netfs_advance_write(struct netfs_io_request * wreq,struct netfs_io_stream * stream,loff_t start,size_t len,bool to_eof)278 size_t netfs_advance_write(struct netfs_io_request *wreq,
279 			   struct netfs_io_stream *stream,
280 			   loff_t start, size_t len, bool to_eof)
281 {
282 	struct netfs_io_subrequest *subreq = stream->construct;
283 	size_t part;
284 
285 	if (!stream->avail) {
286 		_leave("no write");
287 		return len;
288 	}
289 
290 	_enter("R=%x[%x]", wreq->debug_id, subreq ? subreq->debug_index : 0);
291 
292 	if (subreq && start != subreq->start + subreq->len) {
293 		netfs_issue_write(wreq, stream);
294 		subreq = NULL;
295 	}
296 
297 	if (!stream->construct)
298 		netfs_prepare_write(wreq, stream, start);
299 	subreq = stream->construct;
300 
301 	part = umin(stream->sreq_max_len - subreq->len, len);
302 	_debug("part %zx/%zx %zx/%zx", subreq->len, stream->sreq_max_len, part, len);
303 	subreq->len += part;
304 	subreq->nr_segs++;
305 	stream->submit_extendable_to -= part;
306 
307 	if (subreq->len >= stream->sreq_max_len ||
308 	    subreq->nr_segs >= stream->sreq_max_segs ||
309 	    to_eof) {
310 		netfs_issue_write(wreq, stream);
311 		subreq = NULL;
312 	}
313 
314 	return part;
315 }
316 
317 /*
318  * Write some of a pending folio data back to the server.
319  */
netfs_write_folio(struct netfs_io_request * wreq,struct writeback_control * wbc,struct folio * folio)320 static int netfs_write_folio(struct netfs_io_request *wreq,
321 			     struct writeback_control *wbc,
322 			     struct folio *folio)
323 {
324 	struct netfs_io_stream *upload = &wreq->io_streams[0];
325 	struct netfs_io_stream *cache  = &wreq->io_streams[1];
326 	struct netfs_io_stream *stream;
327 	struct netfs_group *fgroup; /* TODO: Use this with ceph */
328 	struct netfs_folio *finfo;
329 	size_t iter_off = 0;
330 	size_t fsize = folio_size(folio), flen = fsize, foff = 0;
331 	loff_t fpos = folio_pos(folio), i_size;
332 	bool to_eof = false, streamw = false;
333 	bool debug = false;
334 
335 	_enter("");
336 
337 	if (rolling_buffer_make_space(&wreq->buffer) < 0)
338 		return -ENOMEM;
339 
340 	/* netfs_perform_write() may shift i_size around the page or from out
341 	 * of the page to beyond it, but cannot move i_size into or through the
342 	 * page since we have it locked.
343 	 */
344 	i_size = i_size_read(wreq->inode);
345 
346 	if (fpos >= i_size) {
347 		/* mmap beyond eof. */
348 		_debug("beyond eof");
349 		folio_start_writeback(folio);
350 		folio_unlock(folio);
351 		wreq->nr_group_rel += netfs_folio_written_back(folio);
352 		netfs_put_group_many(wreq->group, wreq->nr_group_rel);
353 		wreq->nr_group_rel = 0;
354 		return 0;
355 	}
356 
357 	if (fpos + fsize > wreq->i_size)
358 		wreq->i_size = i_size;
359 
360 	fgroup = netfs_folio_group(folio);
361 	finfo = netfs_folio_info(folio);
362 	if (finfo) {
363 		foff = finfo->dirty_offset;
364 		flen = foff + finfo->dirty_len;
365 		streamw = true;
366 	}
367 
368 	if (wreq->origin == NETFS_WRITETHROUGH) {
369 		to_eof = false;
370 		if (flen > i_size - fpos)
371 			flen = i_size - fpos;
372 	} else if (flen > i_size - fpos) {
373 		flen = i_size - fpos;
374 		if (!streamw)
375 			folio_zero_segment(folio, flen, fsize);
376 		to_eof = true;
377 	} else if (flen == i_size - fpos) {
378 		to_eof = true;
379 	}
380 	flen -= foff;
381 
382 	_debug("folio %zx %zx %zx", foff, flen, fsize);
383 
384 	/* Deal with discontinuities in the stream of dirty pages.  These can
385 	 * arise from a number of sources:
386 	 *
387 	 * (1) Intervening non-dirty pages from random-access writes, multiple
388 	 *     flushers writing back different parts simultaneously and manual
389 	 *     syncing.
390 	 *
391 	 * (2) Partially-written pages from write-streaming.
392 	 *
393 	 * (3) Pages that belong to a different write-back group (eg.  Ceph
394 	 *     snapshots).
395 	 *
396 	 * (4) Actually-clean pages that were marked for write to the cache
397 	 *     when they were read.  Note that these appear as a special
398 	 *     write-back group.
399 	 */
400 	if (fgroup == NETFS_FOLIO_COPY_TO_CACHE) {
401 		netfs_issue_write(wreq, upload);
402 	} else if (fgroup != wreq->group) {
403 		/* We can't write this page to the server yet. */
404 		kdebug("wrong group");
405 		folio_redirty_for_writepage(wbc, folio);
406 		folio_unlock(folio);
407 		netfs_issue_write(wreq, upload);
408 		netfs_issue_write(wreq, cache);
409 		return 0;
410 	}
411 
412 	if (foff > 0)
413 		netfs_issue_write(wreq, upload);
414 	if (streamw)
415 		netfs_issue_write(wreq, cache);
416 
417 	/* Flip the page to the writeback state and unlock.  If we're called
418 	 * from write-through, then the page has already been put into the wb
419 	 * state.
420 	 */
421 	if (wreq->origin == NETFS_WRITEBACK)
422 		folio_start_writeback(folio);
423 	folio_unlock(folio);
424 
425 	if (fgroup == NETFS_FOLIO_COPY_TO_CACHE) {
426 		if (!cache->avail) {
427 			trace_netfs_folio(folio, netfs_folio_trace_cancel_copy);
428 			netfs_issue_write(wreq, upload);
429 			netfs_folio_written_back(folio);
430 			return 0;
431 		}
432 		trace_netfs_folio(folio, netfs_folio_trace_store_copy);
433 	} else if (!upload->avail && !cache->avail) {
434 		trace_netfs_folio(folio, netfs_folio_trace_cancel_store);
435 		netfs_folio_written_back(folio);
436 		return 0;
437 	} else if (!upload->construct) {
438 		trace_netfs_folio(folio, netfs_folio_trace_store);
439 	} else {
440 		trace_netfs_folio(folio, netfs_folio_trace_store_plus);
441 	}
442 
443 	/* Attach the folio to the rolling buffer. */
444 	rolling_buffer_append(&wreq->buffer, folio, 0);
445 
446 	/* Move the submission point forward to allow for write-streaming data
447 	 * not starting at the front of the page.  We don't do write-streaming
448 	 * with the cache as the cache requires DIO alignment.
449 	 *
450 	 * Also skip uploading for data that's been read and just needs copying
451 	 * to the cache.
452 	 */
453 	for (int s = 0; s < NR_IO_STREAMS; s++) {
454 		stream = &wreq->io_streams[s];
455 		stream->submit_off = foff;
456 		stream->submit_len = flen;
457 		if (!stream->avail ||
458 		    (stream->source == NETFS_WRITE_TO_CACHE && streamw) ||
459 		    (stream->source == NETFS_UPLOAD_TO_SERVER &&
460 		     fgroup == NETFS_FOLIO_COPY_TO_CACHE)) {
461 			stream->submit_off = UINT_MAX;
462 			stream->submit_len = 0;
463 		}
464 	}
465 
466 	/* Attach the folio to one or more subrequests.  For a big folio, we
467 	 * could end up with thousands of subrequests if the wsize is small -
468 	 * but we might need to wait during the creation of subrequests for
469 	 * network resources (eg. SMB credits).
470 	 */
471 	for (;;) {
472 		ssize_t part;
473 		size_t lowest_off = ULONG_MAX;
474 		int choose_s = -1;
475 
476 		/* Always add to the lowest-submitted stream first. */
477 		for (int s = 0; s < NR_IO_STREAMS; s++) {
478 			stream = &wreq->io_streams[s];
479 			if (stream->submit_len > 0 &&
480 			    stream->submit_off < lowest_off) {
481 				lowest_off = stream->submit_off;
482 				choose_s = s;
483 			}
484 		}
485 
486 		if (choose_s < 0)
487 			break;
488 		stream = &wreq->io_streams[choose_s];
489 
490 		/* Advance the iterator(s). */
491 		if (stream->submit_off > iter_off) {
492 			rolling_buffer_advance(&wreq->buffer, stream->submit_off - iter_off);
493 			iter_off = stream->submit_off;
494 		}
495 
496 		atomic64_set(&wreq->issued_to, fpos + stream->submit_off);
497 		stream->submit_extendable_to = fsize - stream->submit_off;
498 		part = netfs_advance_write(wreq, stream, fpos + stream->submit_off,
499 					   stream->submit_len, to_eof);
500 		stream->submit_off += part;
501 		if (part > stream->submit_len)
502 			stream->submit_len = 0;
503 		else
504 			stream->submit_len -= part;
505 		if (part > 0)
506 			debug = true;
507 	}
508 
509 	if (fsize > iter_off)
510 		rolling_buffer_advance(&wreq->buffer, fsize - iter_off);
511 	atomic64_set(&wreq->issued_to, fpos + fsize);
512 
513 	if (!debug)
514 		kdebug("R=%x: No submit", wreq->debug_id);
515 
516 	if (foff + flen < fsize)
517 		for (int s = 0; s < NR_IO_STREAMS; s++)
518 			netfs_issue_write(wreq, &wreq->io_streams[s]);
519 
520 	_leave(" = 0");
521 	return 0;
522 }
523 
524 /*
525  * End the issuing of writes, letting the collector know we're done.
526  */
netfs_end_issue_write(struct netfs_io_request * wreq)527 static void netfs_end_issue_write(struct netfs_io_request *wreq)
528 {
529 	bool needs_poke = true;
530 
531 	smp_wmb(); /* Write subreq lists before ALL_QUEUED. */
532 	set_bit(NETFS_RREQ_ALL_QUEUED, &wreq->flags);
533 
534 	for (int s = 0; s < NR_IO_STREAMS; s++) {
535 		struct netfs_io_stream *stream = &wreq->io_streams[s];
536 
537 		if (!stream->active)
538 			continue;
539 		if (!list_empty(&stream->subrequests))
540 			needs_poke = false;
541 		netfs_issue_write(wreq, stream);
542 	}
543 
544 	if (needs_poke)
545 		netfs_wake_write_collector(wreq, false);
546 }
547 
548 /*
549  * Write some of the pending data back to the server
550  */
netfs_writepages(struct address_space * mapping,struct writeback_control * wbc)551 int netfs_writepages(struct address_space *mapping,
552 		     struct writeback_control *wbc)
553 {
554 	struct netfs_inode *ictx = netfs_inode(mapping->host);
555 	struct netfs_io_request *wreq = NULL;
556 	struct folio *folio;
557 	int error = 0;
558 
559 	if (!mutex_trylock(&ictx->wb_lock)) {
560 		if (wbc->sync_mode == WB_SYNC_NONE) {
561 			netfs_stat(&netfs_n_wb_lock_skip);
562 			return 0;
563 		}
564 		netfs_stat(&netfs_n_wb_lock_wait);
565 		mutex_lock(&ictx->wb_lock);
566 	}
567 
568 	/* Need the first folio to be able to set up the op. */
569 	folio = writeback_iter(mapping, wbc, NULL, &error);
570 	if (!folio)
571 		goto out;
572 
573 	wreq = netfs_create_write_req(mapping, NULL, folio_pos(folio), NETFS_WRITEBACK);
574 	if (IS_ERR(wreq)) {
575 		error = PTR_ERR(wreq);
576 		goto couldnt_start;
577 	}
578 
579 	trace_netfs_write(wreq, netfs_write_trace_writeback);
580 	netfs_stat(&netfs_n_wh_writepages);
581 
582 	do {
583 		_debug("wbiter %lx %llx", folio->index, atomic64_read(&wreq->issued_to));
584 
585 		/* It appears we don't have to handle cyclic writeback wrapping. */
586 		WARN_ON_ONCE(wreq && folio_pos(folio) < atomic64_read(&wreq->issued_to));
587 
588 		if (netfs_folio_group(folio) != NETFS_FOLIO_COPY_TO_CACHE &&
589 		    unlikely(!test_bit(NETFS_RREQ_UPLOAD_TO_SERVER, &wreq->flags))) {
590 			set_bit(NETFS_RREQ_UPLOAD_TO_SERVER, &wreq->flags);
591 			wreq->netfs_ops->begin_writeback(wreq);
592 		}
593 
594 		error = netfs_write_folio(wreq, wbc, folio);
595 		if (error < 0)
596 			break;
597 	} while ((folio = writeback_iter(mapping, wbc, folio, &error)));
598 
599 	netfs_end_issue_write(wreq);
600 
601 	mutex_unlock(&ictx->wb_lock);
602 
603 	netfs_put_request(wreq, false, netfs_rreq_trace_put_return);
604 	_leave(" = %d", error);
605 	return error;
606 
607 couldnt_start:
608 	netfs_kill_dirty_pages(mapping, wbc, folio);
609 out:
610 	mutex_unlock(&ictx->wb_lock);
611 	_leave(" = %d", error);
612 	return error;
613 }
614 EXPORT_SYMBOL(netfs_writepages);
615 
616 /*
617  * Begin a write operation for writing through the pagecache.
618  */
netfs_begin_writethrough(struct kiocb * iocb,size_t len)619 struct netfs_io_request *netfs_begin_writethrough(struct kiocb *iocb, size_t len)
620 {
621 	struct netfs_io_request *wreq = NULL;
622 	struct netfs_inode *ictx = netfs_inode(file_inode(iocb->ki_filp));
623 
624 	mutex_lock(&ictx->wb_lock);
625 
626 	wreq = netfs_create_write_req(iocb->ki_filp->f_mapping, iocb->ki_filp,
627 				      iocb->ki_pos, NETFS_WRITETHROUGH);
628 	if (IS_ERR(wreq)) {
629 		mutex_unlock(&ictx->wb_lock);
630 		return wreq;
631 	}
632 
633 	wreq->io_streams[0].avail = true;
634 	trace_netfs_write(wreq, netfs_write_trace_writethrough);
635 	return wreq;
636 }
637 
638 /*
639  * Advance the state of the write operation used when writing through the
640  * pagecache.  Data has been copied into the pagecache that we need to append
641  * to the request.  If we've added more than wsize then we need to create a new
642  * subrequest.
643  */
netfs_advance_writethrough(struct netfs_io_request * wreq,struct writeback_control * wbc,struct folio * folio,size_t copied,bool to_page_end,struct folio ** writethrough_cache)644 int netfs_advance_writethrough(struct netfs_io_request *wreq, struct writeback_control *wbc,
645 			       struct folio *folio, size_t copied, bool to_page_end,
646 			       struct folio **writethrough_cache)
647 {
648 	_enter("R=%x ic=%zu ws=%u cp=%zu tp=%u",
649 	       wreq->debug_id, wreq->buffer.iter.count, wreq->wsize, copied, to_page_end);
650 
651 	if (!*writethrough_cache) {
652 		if (folio_test_dirty(folio))
653 			/* Sigh.  mmap. */
654 			folio_clear_dirty_for_io(folio);
655 
656 		/* We can make multiple writes to the folio... */
657 		folio_start_writeback(folio);
658 		if (wreq->len == 0)
659 			trace_netfs_folio(folio, netfs_folio_trace_wthru);
660 		else
661 			trace_netfs_folio(folio, netfs_folio_trace_wthru_plus);
662 		*writethrough_cache = folio;
663 	}
664 
665 	wreq->len += copied;
666 	if (!to_page_end)
667 		return 0;
668 
669 	*writethrough_cache = NULL;
670 	return netfs_write_folio(wreq, wbc, folio);
671 }
672 
673 /*
674  * End a write operation used when writing through the pagecache.
675  */
netfs_end_writethrough(struct netfs_io_request * wreq,struct writeback_control * wbc,struct folio * writethrough_cache)676 int netfs_end_writethrough(struct netfs_io_request *wreq, struct writeback_control *wbc,
677 			   struct folio *writethrough_cache)
678 {
679 	struct netfs_inode *ictx = netfs_inode(wreq->inode);
680 	int ret;
681 
682 	_enter("R=%x", wreq->debug_id);
683 
684 	if (writethrough_cache)
685 		netfs_write_folio(wreq, wbc, writethrough_cache);
686 
687 	netfs_end_issue_write(wreq);
688 
689 	mutex_unlock(&ictx->wb_lock);
690 
691 	if (wreq->iocb) {
692 		ret = -EIOCBQUEUED;
693 	} else {
694 		wait_on_bit(&wreq->flags, NETFS_RREQ_IN_PROGRESS, TASK_UNINTERRUPTIBLE);
695 		ret = wreq->error;
696 	}
697 	netfs_put_request(wreq, false, netfs_rreq_trace_put_return);
698 	return ret;
699 }
700 
701 /*
702  * Write data to the server without going through the pagecache and without
703  * writing it to the local cache.
704  */
netfs_unbuffered_write(struct netfs_io_request * wreq,bool may_wait,size_t len)705 int netfs_unbuffered_write(struct netfs_io_request *wreq, bool may_wait, size_t len)
706 {
707 	struct netfs_io_stream *upload = &wreq->io_streams[0];
708 	ssize_t part;
709 	loff_t start = wreq->start;
710 	int error = 0;
711 
712 	_enter("%zx", len);
713 
714 	if (wreq->origin == NETFS_DIO_WRITE)
715 		inode_dio_begin(wreq->inode);
716 
717 	while (len) {
718 		// TODO: Prepare content encryption
719 
720 		_debug("unbuffered %zx", len);
721 		part = netfs_advance_write(wreq, upload, start, len, false);
722 		start += part;
723 		len -= part;
724 		rolling_buffer_advance(&wreq->buffer, part);
725 		if (test_bit(NETFS_RREQ_PAUSE, &wreq->flags)) {
726 			trace_netfs_rreq(wreq, netfs_rreq_trace_wait_pause);
727 			wait_event(wreq->waitq, !test_bit(NETFS_RREQ_PAUSE, &wreq->flags));
728 		}
729 		if (test_bit(NETFS_RREQ_FAILED, &wreq->flags))
730 			break;
731 	}
732 
733 	netfs_end_issue_write(wreq);
734 	_leave(" = %d", error);
735 	return error;
736 }
737 
738 /*
739  * Write some of a pending folio data back to the server and/or the cache.
740  */
netfs_write_folio_single(struct netfs_io_request * wreq,struct folio * folio)741 static int netfs_write_folio_single(struct netfs_io_request *wreq,
742 				    struct folio *folio)
743 {
744 	struct netfs_io_stream *upload = &wreq->io_streams[0];
745 	struct netfs_io_stream *cache  = &wreq->io_streams[1];
746 	struct netfs_io_stream *stream;
747 	size_t iter_off = 0;
748 	size_t fsize = folio_size(folio), flen;
749 	loff_t fpos = folio_pos(folio);
750 	bool to_eof = false;
751 	bool no_debug = false;
752 
753 	_enter("");
754 
755 	flen = folio_size(folio);
756 	if (flen > wreq->i_size - fpos) {
757 		flen = wreq->i_size - fpos;
758 		folio_zero_segment(folio, flen, fsize);
759 		to_eof = true;
760 	} else if (flen == wreq->i_size - fpos) {
761 		to_eof = true;
762 	}
763 
764 	_debug("folio %zx/%zx", flen, fsize);
765 
766 	if (!upload->avail && !cache->avail) {
767 		trace_netfs_folio(folio, netfs_folio_trace_cancel_store);
768 		return 0;
769 	}
770 
771 	if (!upload->construct)
772 		trace_netfs_folio(folio, netfs_folio_trace_store);
773 	else
774 		trace_netfs_folio(folio, netfs_folio_trace_store_plus);
775 
776 	/* Attach the folio to the rolling buffer. */
777 	folio_get(folio);
778 	rolling_buffer_append(&wreq->buffer, folio, NETFS_ROLLBUF_PUT_MARK);
779 
780 	/* Move the submission point forward to allow for write-streaming data
781 	 * not starting at the front of the page.  We don't do write-streaming
782 	 * with the cache as the cache requires DIO alignment.
783 	 *
784 	 * Also skip uploading for data that's been read and just needs copying
785 	 * to the cache.
786 	 */
787 	for (int s = 0; s < NR_IO_STREAMS; s++) {
788 		stream = &wreq->io_streams[s];
789 		stream->submit_off = 0;
790 		stream->submit_len = flen;
791 		if (!stream->avail) {
792 			stream->submit_off = UINT_MAX;
793 			stream->submit_len = 0;
794 		}
795 	}
796 
797 	/* Attach the folio to one or more subrequests.  For a big folio, we
798 	 * could end up with thousands of subrequests if the wsize is small -
799 	 * but we might need to wait during the creation of subrequests for
800 	 * network resources (eg. SMB credits).
801 	 */
802 	for (;;) {
803 		ssize_t part;
804 		size_t lowest_off = ULONG_MAX;
805 		int choose_s = -1;
806 
807 		/* Always add to the lowest-submitted stream first. */
808 		for (int s = 0; s < NR_IO_STREAMS; s++) {
809 			stream = &wreq->io_streams[s];
810 			if (stream->submit_len > 0 &&
811 			    stream->submit_off < lowest_off) {
812 				lowest_off = stream->submit_off;
813 				choose_s = s;
814 			}
815 		}
816 
817 		if (choose_s < 0)
818 			break;
819 		stream = &wreq->io_streams[choose_s];
820 
821 		/* Advance the iterator(s). */
822 		if (stream->submit_off > iter_off) {
823 			rolling_buffer_advance(&wreq->buffer, stream->submit_off - iter_off);
824 			iter_off = stream->submit_off;
825 		}
826 
827 		atomic64_set(&wreq->issued_to, fpos + stream->submit_off);
828 		stream->submit_extendable_to = fsize - stream->submit_off;
829 		part = netfs_advance_write(wreq, stream, fpos + stream->submit_off,
830 					   stream->submit_len, to_eof);
831 		stream->submit_off += part;
832 		if (part > stream->submit_len)
833 			stream->submit_len = 0;
834 		else
835 			stream->submit_len -= part;
836 		if (part > 0)
837 			no_debug = true;
838 	}
839 
840 	wreq->buffer.iter.iov_offset = 0;
841 	if (fsize > iter_off)
842 		rolling_buffer_advance(&wreq->buffer, fsize - iter_off);
843 	atomic64_set(&wreq->issued_to, fpos + fsize);
844 
845 	if (!no_debug)
846 		kdebug("R=%x: No submit", wreq->debug_id);
847 	_leave(" = 0");
848 	return 0;
849 }
850 
851 /**
852  * netfs_writeback_single - Write back a monolithic payload
853  * @mapping: The mapping to write from
854  * @wbc: Hints from the VM
855  * @iter: Data to write, must be ITER_FOLIOQ.
856  *
857  * Write a monolithic, non-pagecache object back to the server and/or
858  * the cache.
859  */
netfs_writeback_single(struct address_space * mapping,struct writeback_control * wbc,struct iov_iter * iter)860 int netfs_writeback_single(struct address_space *mapping,
861 			   struct writeback_control *wbc,
862 			   struct iov_iter *iter)
863 {
864 	struct netfs_io_request *wreq;
865 	struct netfs_inode *ictx = netfs_inode(mapping->host);
866 	struct folio_queue *fq;
867 	size_t size = iov_iter_count(iter);
868 	int ret;
869 
870 	if (WARN_ON_ONCE(!iov_iter_is_folioq(iter)))
871 		return -EIO;
872 
873 	if (!mutex_trylock(&ictx->wb_lock)) {
874 		if (wbc->sync_mode == WB_SYNC_NONE) {
875 			netfs_stat(&netfs_n_wb_lock_skip);
876 			return 0;
877 		}
878 		netfs_stat(&netfs_n_wb_lock_wait);
879 		mutex_lock(&ictx->wb_lock);
880 	}
881 
882 	wreq = netfs_create_write_req(mapping, NULL, 0, NETFS_WRITEBACK_SINGLE);
883 	if (IS_ERR(wreq)) {
884 		ret = PTR_ERR(wreq);
885 		goto couldnt_start;
886 	}
887 
888 	trace_netfs_write(wreq, netfs_write_trace_writeback);
889 	netfs_stat(&netfs_n_wh_writepages);
890 
891 	if (__test_and_set_bit(NETFS_RREQ_UPLOAD_TO_SERVER, &wreq->flags))
892 		wreq->netfs_ops->begin_writeback(wreq);
893 
894 	for (fq = (struct folio_queue *)iter->folioq; fq; fq = fq->next) {
895 		for (int slot = 0; slot < folioq_count(fq); slot++) {
896 			struct folio *folio = folioq_folio(fq, slot);
897 			size_t part = umin(folioq_folio_size(fq, slot), size);
898 
899 			_debug("wbiter %lx %llx", folio->index, atomic64_read(&wreq->issued_to));
900 
901 			ret = netfs_write_folio_single(wreq, folio);
902 			if (ret < 0)
903 				goto stop;
904 			size -= part;
905 			if (size <= 0)
906 				goto stop;
907 		}
908 	}
909 
910 stop:
911 	for (int s = 0; s < NR_IO_STREAMS; s++)
912 		netfs_issue_write(wreq, &wreq->io_streams[s]);
913 	smp_wmb(); /* Write lists before ALL_QUEUED. */
914 	set_bit(NETFS_RREQ_ALL_QUEUED, &wreq->flags);
915 
916 	mutex_unlock(&ictx->wb_lock);
917 
918 	netfs_put_request(wreq, false, netfs_rreq_trace_put_return);
919 	_leave(" = %d", ret);
920 	return ret;
921 
922 couldnt_start:
923 	mutex_unlock(&ictx->wb_lock);
924 	_leave(" = %d", ret);
925 	return ret;
926 }
927 EXPORT_SYMBOL(netfs_writeback_single);
928