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_buffered = (origin == NETFS_WRITEBACK ||
98 origin == NETFS_WRITETHROUGH ||
99 origin == NETFS_PGPRIV2_COPY_TO_CACHE);
100
101 wreq = netfs_alloc_request(mapping, file, start, 0, origin);
102 if (IS_ERR(wreq))
103 return wreq;
104
105 _enter("R=%x", wreq->debug_id);
106
107 ictx = netfs_inode(wreq->inode);
108 if (is_buffered && netfs_is_cache_enabled(ictx))
109 fscache_begin_write_operation(&wreq->cache_resources, netfs_i_cookie(ictx));
110
111 wreq->cleaned_to = wreq->start;
112
113 wreq->io_streams[0].stream_nr = 0;
114 wreq->io_streams[0].source = NETFS_UPLOAD_TO_SERVER;
115 wreq->io_streams[0].prepare_write = ictx->ops->prepare_write;
116 wreq->io_streams[0].issue_write = ictx->ops->issue_write;
117 wreq->io_streams[0].collected_to = start;
118 wreq->io_streams[0].transferred = LONG_MAX;
119
120 wreq->io_streams[1].stream_nr = 1;
121 wreq->io_streams[1].source = NETFS_WRITE_TO_CACHE;
122 wreq->io_streams[1].collected_to = start;
123 wreq->io_streams[1].transferred = LONG_MAX;
124 if (fscache_resources_valid(&wreq->cache_resources)) {
125 wreq->io_streams[1].avail = true;
126 wreq->io_streams[1].active = true;
127 wreq->io_streams[1].prepare_write = wreq->cache_resources.ops->prepare_write_subreq;
128 wreq->io_streams[1].issue_write = wreq->cache_resources.ops->issue_write;
129 }
130
131 return wreq;
132 }
133
134 /**
135 * netfs_prepare_write_failed - Note write preparation failed
136 * @subreq: The subrequest to mark
137 *
138 * Mark a subrequest to note that preparation for write failed.
139 */
netfs_prepare_write_failed(struct netfs_io_subrequest * subreq)140 void netfs_prepare_write_failed(struct netfs_io_subrequest *subreq)
141 {
142 __set_bit(NETFS_SREQ_FAILED, &subreq->flags);
143 trace_netfs_sreq(subreq, netfs_sreq_trace_prep_failed);
144 }
145 EXPORT_SYMBOL(netfs_prepare_write_failed);
146
147 /*
148 * Prepare a write subrequest. We need to allocate a new subrequest
149 * if we don't have one.
150 */
netfs_prepare_write(struct netfs_io_request * wreq,struct netfs_io_stream * stream,loff_t start)151 static void netfs_prepare_write(struct netfs_io_request *wreq,
152 struct netfs_io_stream *stream,
153 loff_t start)
154 {
155 struct netfs_io_subrequest *subreq;
156 struct iov_iter *wreq_iter = &wreq->io_iter;
157
158 /* Make sure we don't point the iterator at a used-up folio_queue
159 * struct being used as a placeholder to prevent the queue from
160 * collapsing. In such a case, extend the queue.
161 */
162 if (iov_iter_is_folioq(wreq_iter) &&
163 wreq_iter->folioq_slot >= folioq_nr_slots(wreq_iter->folioq)) {
164 netfs_buffer_make_space(wreq);
165 }
166
167 subreq = netfs_alloc_subrequest(wreq);
168 subreq->source = stream->source;
169 subreq->start = start;
170 subreq->stream_nr = stream->stream_nr;
171 subreq->io_iter = *wreq_iter;
172
173 _enter("R=%x[%x]", wreq->debug_id, subreq->debug_index);
174
175 trace_netfs_sreq(subreq, netfs_sreq_trace_prepare);
176
177 stream->sreq_max_len = UINT_MAX;
178 stream->sreq_max_segs = INT_MAX;
179 switch (stream->source) {
180 case NETFS_UPLOAD_TO_SERVER:
181 netfs_stat(&netfs_n_wh_upload);
182 stream->sreq_max_len = wreq->wsize;
183 break;
184 case NETFS_WRITE_TO_CACHE:
185 netfs_stat(&netfs_n_wh_write);
186 break;
187 default:
188 WARN_ON_ONCE(1);
189 break;
190 }
191
192 if (stream->prepare_write)
193 stream->prepare_write(subreq);
194
195 __set_bit(NETFS_SREQ_IN_PROGRESS, &subreq->flags);
196
197 /* We add to the end of the list whilst the collector may be walking
198 * the list. The collector only goes nextwards and uses the lock to
199 * remove entries off of the front.
200 */
201 spin_lock_bh(&wreq->lock);
202 list_add_tail(&subreq->rreq_link, &stream->subrequests);
203 if (list_is_first(&subreq->rreq_link, &stream->subrequests)) {
204 stream->front = subreq;
205 if (!stream->active) {
206 stream->collected_to = stream->front->start;
207 /* Write list pointers before active flag */
208 smp_store_release(&stream->active, true);
209 }
210 }
211
212 spin_unlock_bh(&wreq->lock);
213
214 stream->construct = subreq;
215 }
216
217 /*
218 * Set the I/O iterator for the filesystem/cache to use and dispatch the I/O
219 * operation. The operation may be asynchronous and should call
220 * netfs_write_subrequest_terminated() when complete.
221 */
netfs_do_issue_write(struct netfs_io_stream * stream,struct netfs_io_subrequest * subreq)222 static void netfs_do_issue_write(struct netfs_io_stream *stream,
223 struct netfs_io_subrequest *subreq)
224 {
225 struct netfs_io_request *wreq = subreq->rreq;
226
227 _enter("R=%x[%x],%zx", wreq->debug_id, subreq->debug_index, subreq->len);
228
229 if (test_bit(NETFS_SREQ_FAILED, &subreq->flags))
230 return netfs_write_subrequest_terminated(subreq, subreq->error, false);
231
232 trace_netfs_sreq(subreq, netfs_sreq_trace_submit);
233 stream->issue_write(subreq);
234 }
235
netfs_reissue_write(struct netfs_io_stream * stream,struct netfs_io_subrequest * subreq,struct iov_iter * source)236 void netfs_reissue_write(struct netfs_io_stream *stream,
237 struct netfs_io_subrequest *subreq,
238 struct iov_iter *source)
239 {
240 size_t size = subreq->len - subreq->transferred;
241
242 // TODO: Use encrypted buffer
243 subreq->io_iter = *source;
244 iov_iter_advance(source, size);
245 iov_iter_truncate(&subreq->io_iter, size);
246
247 __set_bit(NETFS_SREQ_IN_PROGRESS, &subreq->flags);
248 netfs_do_issue_write(stream, subreq);
249 }
250
netfs_issue_write(struct netfs_io_request * wreq,struct netfs_io_stream * stream)251 void netfs_issue_write(struct netfs_io_request *wreq,
252 struct netfs_io_stream *stream)
253 {
254 struct netfs_io_subrequest *subreq = stream->construct;
255
256 if (!subreq)
257 return;
258 stream->construct = NULL;
259 subreq->io_iter.count = subreq->len;
260 netfs_do_issue_write(stream, subreq);
261 }
262
263 /*
264 * Add data to the write subrequest, dispatching each as we fill it up or if it
265 * is discontiguous with the previous. We only fill one part at a time so that
266 * we can avoid overrunning the credits obtained (cifs) and try to parallelise
267 * content-crypto preparation with network writes.
268 */
netfs_advance_write(struct netfs_io_request * wreq,struct netfs_io_stream * stream,loff_t start,size_t len,bool to_eof)269 int netfs_advance_write(struct netfs_io_request *wreq,
270 struct netfs_io_stream *stream,
271 loff_t start, size_t len, bool to_eof)
272 {
273 struct netfs_io_subrequest *subreq = stream->construct;
274 size_t part;
275
276 if (!stream->avail) {
277 _leave("no write");
278 return len;
279 }
280
281 _enter("R=%x[%x]", wreq->debug_id, subreq ? subreq->debug_index : 0);
282
283 if (subreq && start != subreq->start + subreq->len) {
284 netfs_issue_write(wreq, stream);
285 subreq = NULL;
286 }
287
288 if (!stream->construct)
289 netfs_prepare_write(wreq, stream, start);
290 subreq = stream->construct;
291
292 part = umin(stream->sreq_max_len - subreq->len, len);
293 _debug("part %zx/%zx %zx/%zx", subreq->len, stream->sreq_max_len, part, len);
294 subreq->len += part;
295 subreq->nr_segs++;
296 stream->submit_extendable_to -= part;
297
298 if (subreq->len >= stream->sreq_max_len ||
299 subreq->nr_segs >= stream->sreq_max_segs ||
300 to_eof) {
301 netfs_issue_write(wreq, stream);
302 subreq = NULL;
303 }
304
305 return part;
306 }
307
308 /*
309 * Write some of a pending folio data back to the server.
310 */
netfs_write_folio(struct netfs_io_request * wreq,struct writeback_control * wbc,struct folio * folio)311 static int netfs_write_folio(struct netfs_io_request *wreq,
312 struct writeback_control *wbc,
313 struct folio *folio)
314 {
315 struct netfs_io_stream *upload = &wreq->io_streams[0];
316 struct netfs_io_stream *cache = &wreq->io_streams[1];
317 struct netfs_io_stream *stream;
318 struct netfs_group *fgroup; /* TODO: Use this with ceph */
319 struct netfs_folio *finfo;
320 size_t fsize = folio_size(folio), flen = fsize, foff = 0;
321 loff_t fpos = folio_pos(folio), i_size;
322 bool to_eof = false, streamw = false;
323 bool debug = false;
324
325 _enter("");
326
327 /* netfs_perform_write() may shift i_size around the page or from out
328 * of the page to beyond it, but cannot move i_size into or through the
329 * page since we have it locked.
330 */
331 i_size = i_size_read(wreq->inode);
332
333 if (fpos >= i_size) {
334 /* mmap beyond eof. */
335 _debug("beyond eof");
336 folio_start_writeback(folio);
337 folio_unlock(folio);
338 wreq->nr_group_rel += netfs_folio_written_back(folio);
339 netfs_put_group_many(wreq->group, wreq->nr_group_rel);
340 wreq->nr_group_rel = 0;
341 return 0;
342 }
343
344 if (fpos + fsize > wreq->i_size)
345 wreq->i_size = i_size;
346
347 fgroup = netfs_folio_group(folio);
348 finfo = netfs_folio_info(folio);
349 if (finfo) {
350 foff = finfo->dirty_offset;
351 flen = foff + finfo->dirty_len;
352 streamw = true;
353 }
354
355 if (wreq->origin == NETFS_WRITETHROUGH) {
356 to_eof = false;
357 if (flen > i_size - fpos)
358 flen = i_size - fpos;
359 } else if (flen > i_size - fpos) {
360 flen = i_size - fpos;
361 if (!streamw)
362 folio_zero_segment(folio, flen, fsize);
363 to_eof = true;
364 } else if (flen == i_size - fpos) {
365 to_eof = true;
366 }
367 flen -= foff;
368
369 _debug("folio %zx %zx %zx", foff, flen, fsize);
370
371 /* Deal with discontinuities in the stream of dirty pages. These can
372 * arise from a number of sources:
373 *
374 * (1) Intervening non-dirty pages from random-access writes, multiple
375 * flushers writing back different parts simultaneously and manual
376 * syncing.
377 *
378 * (2) Partially-written pages from write-streaming.
379 *
380 * (3) Pages that belong to a different write-back group (eg. Ceph
381 * snapshots).
382 *
383 * (4) Actually-clean pages that were marked for write to the cache
384 * when they were read. Note that these appear as a special
385 * write-back group.
386 */
387 if (fgroup == NETFS_FOLIO_COPY_TO_CACHE) {
388 netfs_issue_write(wreq, upload);
389 } else if (fgroup != wreq->group) {
390 /* We can't write this page to the server yet. */
391 kdebug("wrong group");
392 folio_redirty_for_writepage(wbc, folio);
393 folio_unlock(folio);
394 netfs_issue_write(wreq, upload);
395 netfs_issue_write(wreq, cache);
396 return 0;
397 }
398
399 if (foff > 0)
400 netfs_issue_write(wreq, upload);
401 if (streamw)
402 netfs_issue_write(wreq, cache);
403
404 /* Flip the page to the writeback state and unlock. If we're called
405 * from write-through, then the page has already been put into the wb
406 * state.
407 */
408 if (wreq->origin == NETFS_WRITEBACK)
409 folio_start_writeback(folio);
410 folio_unlock(folio);
411
412 if (fgroup == NETFS_FOLIO_COPY_TO_CACHE) {
413 if (!cache->avail) {
414 trace_netfs_folio(folio, netfs_folio_trace_cancel_copy);
415 netfs_issue_write(wreq, upload);
416 netfs_folio_written_back(folio);
417 return 0;
418 }
419 trace_netfs_folio(folio, netfs_folio_trace_store_copy);
420 } else if (!upload->avail && !cache->avail) {
421 trace_netfs_folio(folio, netfs_folio_trace_cancel_store);
422 netfs_folio_written_back(folio);
423 return 0;
424 } else if (!upload->construct) {
425 trace_netfs_folio(folio, netfs_folio_trace_store);
426 } else {
427 trace_netfs_folio(folio, netfs_folio_trace_store_plus);
428 }
429
430 /* Attach the folio to the rolling buffer. */
431 netfs_buffer_append_folio(wreq, folio, false);
432
433 /* Move the submission point forward to allow for write-streaming data
434 * not starting at the front of the page. We don't do write-streaming
435 * with the cache as the cache requires DIO alignment.
436 *
437 * Also skip uploading for data that's been read and just needs copying
438 * to the cache.
439 */
440 for (int s = 0; s < NR_IO_STREAMS; s++) {
441 stream = &wreq->io_streams[s];
442 stream->submit_off = foff;
443 stream->submit_len = flen;
444 if ((stream->source == NETFS_WRITE_TO_CACHE && streamw) ||
445 (stream->source == NETFS_UPLOAD_TO_SERVER &&
446 fgroup == NETFS_FOLIO_COPY_TO_CACHE)) {
447 stream->submit_off = UINT_MAX;
448 stream->submit_len = 0;
449 }
450 }
451
452 /* Attach the folio to one or more subrequests. For a big folio, we
453 * could end up with thousands of subrequests if the wsize is small -
454 * but we might need to wait during the creation of subrequests for
455 * network resources (eg. SMB credits).
456 */
457 for (;;) {
458 ssize_t part;
459 size_t lowest_off = ULONG_MAX;
460 int choose_s = -1;
461
462 /* Always add to the lowest-submitted stream first. */
463 for (int s = 0; s < NR_IO_STREAMS; s++) {
464 stream = &wreq->io_streams[s];
465 if (stream->submit_len > 0 &&
466 stream->submit_off < lowest_off) {
467 lowest_off = stream->submit_off;
468 choose_s = s;
469 }
470 }
471
472 if (choose_s < 0)
473 break;
474 stream = &wreq->io_streams[choose_s];
475 wreq->io_iter.iov_offset = stream->submit_off;
476
477 atomic64_set(&wreq->issued_to, fpos + stream->submit_off);
478 stream->submit_extendable_to = fsize - stream->submit_off;
479 part = netfs_advance_write(wreq, stream, fpos + stream->submit_off,
480 stream->submit_len, to_eof);
481 stream->submit_off += part;
482 if (part > stream->submit_len)
483 stream->submit_len = 0;
484 else
485 stream->submit_len -= part;
486 if (part > 0)
487 debug = true;
488 }
489
490 wreq->io_iter.iov_offset = 0;
491 iov_iter_advance(&wreq->io_iter, fsize);
492 atomic64_set(&wreq->issued_to, fpos + fsize);
493
494 if (!debug)
495 kdebug("R=%x: No submit", wreq->debug_id);
496
497 if (foff + flen < fsize)
498 for (int s = 0; s < NR_IO_STREAMS; s++)
499 netfs_issue_write(wreq, &wreq->io_streams[s]);
500
501 _leave(" = 0");
502 return 0;
503 }
504
505 /*
506 * Write some of the pending data back to the server
507 */
netfs_writepages(struct address_space * mapping,struct writeback_control * wbc)508 int netfs_writepages(struct address_space *mapping,
509 struct writeback_control *wbc)
510 {
511 struct netfs_inode *ictx = netfs_inode(mapping->host);
512 struct netfs_io_request *wreq = NULL;
513 struct folio *folio;
514 int error = 0;
515
516 if (!mutex_trylock(&ictx->wb_lock)) {
517 if (wbc->sync_mode == WB_SYNC_NONE) {
518 netfs_stat(&netfs_n_wb_lock_skip);
519 return 0;
520 }
521 netfs_stat(&netfs_n_wb_lock_wait);
522 mutex_lock(&ictx->wb_lock);
523 }
524
525 /* Need the first folio to be able to set up the op. */
526 folio = writeback_iter(mapping, wbc, NULL, &error);
527 if (!folio)
528 goto out;
529
530 wreq = netfs_create_write_req(mapping, NULL, folio_pos(folio), NETFS_WRITEBACK);
531 if (IS_ERR(wreq)) {
532 error = PTR_ERR(wreq);
533 goto couldnt_start;
534 }
535
536 trace_netfs_write(wreq, netfs_write_trace_writeback);
537 netfs_stat(&netfs_n_wh_writepages);
538
539 do {
540 _debug("wbiter %lx %llx", folio->index, atomic64_read(&wreq->issued_to));
541
542 /* It appears we don't have to handle cyclic writeback wrapping. */
543 WARN_ON_ONCE(wreq && folio_pos(folio) < atomic64_read(&wreq->issued_to));
544
545 if (netfs_folio_group(folio) != NETFS_FOLIO_COPY_TO_CACHE &&
546 unlikely(!test_bit(NETFS_RREQ_UPLOAD_TO_SERVER, &wreq->flags))) {
547 set_bit(NETFS_RREQ_UPLOAD_TO_SERVER, &wreq->flags);
548 wreq->netfs_ops->begin_writeback(wreq);
549 }
550
551 error = netfs_write_folio(wreq, wbc, folio);
552 if (error < 0)
553 break;
554 } while ((folio = writeback_iter(mapping, wbc, folio, &error)));
555
556 for (int s = 0; s < NR_IO_STREAMS; s++)
557 netfs_issue_write(wreq, &wreq->io_streams[s]);
558 smp_wmb(); /* Write lists before ALL_QUEUED. */
559 set_bit(NETFS_RREQ_ALL_QUEUED, &wreq->flags);
560
561 mutex_unlock(&ictx->wb_lock);
562
563 netfs_put_request(wreq, false, netfs_rreq_trace_put_return);
564 _leave(" = %d", error);
565 return error;
566
567 couldnt_start:
568 netfs_kill_dirty_pages(mapping, wbc, folio);
569 out:
570 mutex_unlock(&ictx->wb_lock);
571 _leave(" = %d", error);
572 return error;
573 }
574 EXPORT_SYMBOL(netfs_writepages);
575
576 /*
577 * Begin a write operation for writing through the pagecache.
578 */
netfs_begin_writethrough(struct kiocb * iocb,size_t len)579 struct netfs_io_request *netfs_begin_writethrough(struct kiocb *iocb, size_t len)
580 {
581 struct netfs_io_request *wreq = NULL;
582 struct netfs_inode *ictx = netfs_inode(file_inode(iocb->ki_filp));
583
584 mutex_lock(&ictx->wb_lock);
585
586 wreq = netfs_create_write_req(iocb->ki_filp->f_mapping, iocb->ki_filp,
587 iocb->ki_pos, NETFS_WRITETHROUGH);
588 if (IS_ERR(wreq)) {
589 mutex_unlock(&ictx->wb_lock);
590 return wreq;
591 }
592
593 wreq->io_streams[0].avail = true;
594 trace_netfs_write(wreq, netfs_write_trace_writethrough);
595 return wreq;
596 }
597
598 /*
599 * Advance the state of the write operation used when writing through the
600 * pagecache. Data has been copied into the pagecache that we need to append
601 * to the request. If we've added more than wsize then we need to create a new
602 * subrequest.
603 */
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)604 int netfs_advance_writethrough(struct netfs_io_request *wreq, struct writeback_control *wbc,
605 struct folio *folio, size_t copied, bool to_page_end,
606 struct folio **writethrough_cache)
607 {
608 _enter("R=%x ic=%zu ws=%u cp=%zu tp=%u",
609 wreq->debug_id, wreq->iter.count, wreq->wsize, copied, to_page_end);
610
611 if (!*writethrough_cache) {
612 if (folio_test_dirty(folio))
613 /* Sigh. mmap. */
614 folio_clear_dirty_for_io(folio);
615
616 /* We can make multiple writes to the folio... */
617 folio_start_writeback(folio);
618 if (wreq->len == 0)
619 trace_netfs_folio(folio, netfs_folio_trace_wthru);
620 else
621 trace_netfs_folio(folio, netfs_folio_trace_wthru_plus);
622 *writethrough_cache = folio;
623 }
624
625 wreq->len += copied;
626 if (!to_page_end)
627 return 0;
628
629 *writethrough_cache = NULL;
630 return netfs_write_folio(wreq, wbc, folio);
631 }
632
633 /*
634 * End a write operation used when writing through the pagecache.
635 */
netfs_end_writethrough(struct netfs_io_request * wreq,struct writeback_control * wbc,struct folio * writethrough_cache)636 int netfs_end_writethrough(struct netfs_io_request *wreq, struct writeback_control *wbc,
637 struct folio *writethrough_cache)
638 {
639 struct netfs_inode *ictx = netfs_inode(wreq->inode);
640 int ret;
641
642 _enter("R=%x", wreq->debug_id);
643
644 if (writethrough_cache)
645 netfs_write_folio(wreq, wbc, writethrough_cache);
646
647 netfs_issue_write(wreq, &wreq->io_streams[0]);
648 netfs_issue_write(wreq, &wreq->io_streams[1]);
649 smp_wmb(); /* Write lists before ALL_QUEUED. */
650 set_bit(NETFS_RREQ_ALL_QUEUED, &wreq->flags);
651
652 mutex_unlock(&ictx->wb_lock);
653
654 if (wreq->iocb) {
655 ret = -EIOCBQUEUED;
656 } else {
657 wait_on_bit(&wreq->flags, NETFS_RREQ_IN_PROGRESS, TASK_UNINTERRUPTIBLE);
658 ret = wreq->error;
659 }
660 netfs_put_request(wreq, false, netfs_rreq_trace_put_return);
661 return ret;
662 }
663
664 /*
665 * Write data to the server without going through the pagecache and without
666 * writing it to the local cache.
667 */
netfs_unbuffered_write(struct netfs_io_request * wreq,bool may_wait,size_t len)668 int netfs_unbuffered_write(struct netfs_io_request *wreq, bool may_wait, size_t len)
669 {
670 struct netfs_io_stream *upload = &wreq->io_streams[0];
671 ssize_t part;
672 loff_t start = wreq->start;
673 int error = 0;
674
675 _enter("%zx", len);
676
677 if (wreq->origin == NETFS_DIO_WRITE)
678 inode_dio_begin(wreq->inode);
679
680 while (len) {
681 // TODO: Prepare content encryption
682
683 _debug("unbuffered %zx", len);
684 part = netfs_advance_write(wreq, upload, start, len, false);
685 start += part;
686 len -= part;
687 iov_iter_advance(&wreq->io_iter, part);
688 if (test_bit(NETFS_RREQ_PAUSE, &wreq->flags)) {
689 trace_netfs_rreq(wreq, netfs_rreq_trace_wait_pause);
690 wait_on_bit(&wreq->flags, NETFS_RREQ_PAUSE, TASK_UNINTERRUPTIBLE);
691 }
692 if (test_bit(NETFS_RREQ_FAILED, &wreq->flags))
693 break;
694 }
695
696 netfs_issue_write(wreq, upload);
697
698 smp_wmb(); /* Write lists before ALL_QUEUED. */
699 set_bit(NETFS_RREQ_ALL_QUEUED, &wreq->flags);
700 if (list_empty(&upload->subrequests))
701 netfs_wake_write_collector(wreq, false);
702
703 _leave(" = %d", error);
704 return error;
705 }
706