xref: /linux/fs/netfs/write_collect.c (revision c060f8168bdf22aa986970955af99702d142dfbe)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Network filesystem write subrequest result collection, assessment
3  * and retrying.
4  *
5  * Copyright (C) 2024 Red Hat, Inc. All Rights Reserved.
6  * Written by David Howells (dhowells@redhat.com)
7  */
8 
9 #include <linux/export.h>
10 #include <linux/fs.h>
11 #include <linux/mm.h>
12 #include <linux/pagemap.h>
13 #include <linux/slab.h>
14 #include "internal.h"
15 
16 /* Notes made in the collector */
17 #define HIT_PENDING		0x01	/* A front op was still pending */
18 #define NEED_REASSESS		0x02	/* Need to loop round and reassess */
19 #define MADE_PROGRESS		0x04	/* Made progress cleaning up a stream or the folio set */
20 #define BUFFERED		0x08	/* The pagecache needs cleaning up */
21 #define NEED_RETRY		0x10	/* A front op requests retrying */
22 #define SAW_FAILURE		0x20	/* One stream or hit a permanent failure */
23 
24 /*
25  * Successful completion of write of a folio to the server and/or cache.  Note
26  * that we are not allowed to lock the folio here on pain of deadlocking with
27  * truncate.
28  */
29 int netfs_folio_written_back(struct folio *folio)
30 {
31 	enum netfs_folio_trace why = netfs_folio_trace_clear;
32 	struct netfs_inode *ictx = netfs_inode(folio->mapping->host);
33 	struct netfs_folio *finfo;
34 	struct netfs_group *group = NULL;
35 	int gcount = 0;
36 
37 	if ((finfo = netfs_folio_info(folio))) {
38 		/* Streaming writes cannot be redirtied whilst under writeback,
39 		 * so discard the streaming record.
40 		 */
41 		unsigned long long fend;
42 
43 		fend = folio_pos(folio) + finfo->dirty_offset + finfo->dirty_len;
44 		if (fend > ictx->zero_point)
45 			ictx->zero_point = fend;
46 
47 		folio_detach_private(folio);
48 		group = finfo->netfs_group;
49 		gcount++;
50 		kfree(finfo);
51 		why = netfs_folio_trace_clear_s;
52 		goto end_wb;
53 	}
54 
55 	if ((group = netfs_folio_group(folio))) {
56 		if (group == NETFS_FOLIO_COPY_TO_CACHE) {
57 			why = netfs_folio_trace_clear_cc;
58 			folio_detach_private(folio);
59 			goto end_wb;
60 		}
61 
62 		/* Need to detach the group pointer if the page didn't get
63 		 * redirtied.  If it has been redirtied, then it must be within
64 		 * the same group.
65 		 */
66 		why = netfs_folio_trace_redirtied;
67 		if (!folio_test_dirty(folio)) {
68 			folio_detach_private(folio);
69 			gcount++;
70 			why = netfs_folio_trace_clear_g;
71 		}
72 	}
73 
74 end_wb:
75 	trace_netfs_folio(folio, why);
76 	folio_end_writeback(folio);
77 	return gcount;
78 }
79 
80 /*
81  * Unlock any folios we've finished with.
82  */
83 static void netfs_writeback_unlock_folios(struct netfs_io_request *wreq,
84 					  unsigned int *notes)
85 {
86 	struct folio_queue *folioq = wreq->buffer;
87 	unsigned long long collected_to = wreq->collected_to;
88 	unsigned int slot = wreq->buffer_head_slot;
89 
90 	if (wreq->origin == NETFS_PGPRIV2_COPY_TO_CACHE) {
91 		if (netfs_pgpriv2_unlock_copied_folios(wreq))
92 			*notes |= MADE_PROGRESS;
93 		return;
94 	}
95 
96 	if (slot >= folioq_nr_slots(folioq)) {
97 		folioq = netfs_delete_buffer_head(wreq);
98 		slot = 0;
99 	}
100 
101 	for (;;) {
102 		struct folio *folio;
103 		struct netfs_folio *finfo;
104 		unsigned long long fpos, fend;
105 		size_t fsize, flen;
106 
107 		folio = folioq_folio(folioq, slot);
108 		if (WARN_ONCE(!folio_test_writeback(folio),
109 			      "R=%08x: folio %lx is not under writeback\n",
110 			      wreq->debug_id, folio->index))
111 			trace_netfs_folio(folio, netfs_folio_trace_not_under_wback);
112 
113 		fpos = folio_pos(folio);
114 		fsize = folio_size(folio);
115 		finfo = netfs_folio_info(folio);
116 		flen = finfo ? finfo->dirty_offset + finfo->dirty_len : fsize;
117 
118 		fend = min_t(unsigned long long, fpos + flen, wreq->i_size);
119 
120 		trace_netfs_collect_folio(wreq, folio, fend, collected_to);
121 
122 		/* Unlock any folio we've transferred all of. */
123 		if (collected_to < fend)
124 			break;
125 
126 		wreq->nr_group_rel += netfs_folio_written_back(folio);
127 		wreq->cleaned_to = fpos + fsize;
128 		*notes |= MADE_PROGRESS;
129 
130 		/* Clean up the head folioq.  If we clear an entire folioq, then
131 		 * we can get rid of it provided it's not also the tail folioq
132 		 * being filled by the issuer.
133 		 */
134 		folioq_clear(folioq, slot);
135 		slot++;
136 		if (slot >= folioq_nr_slots(folioq)) {
137 			if (READ_ONCE(wreq->buffer_tail) == folioq)
138 				break;
139 			folioq = netfs_delete_buffer_head(wreq);
140 			slot = 0;
141 		}
142 
143 		if (fpos + fsize >= collected_to)
144 			break;
145 	}
146 
147 	wreq->buffer = folioq;
148 	wreq->buffer_head_slot = slot;
149 }
150 
151 /*
152  * Perform retries on the streams that need it.
153  */
154 static void netfs_retry_write_stream(struct netfs_io_request *wreq,
155 				     struct netfs_io_stream *stream)
156 {
157 	struct list_head *next;
158 
159 	_enter("R=%x[%x:]", wreq->debug_id, stream->stream_nr);
160 
161 	if (list_empty(&stream->subrequests))
162 		return;
163 
164 	if (stream->source == NETFS_UPLOAD_TO_SERVER &&
165 	    wreq->netfs_ops->retry_request)
166 		wreq->netfs_ops->retry_request(wreq, stream);
167 
168 	if (unlikely(stream->failed))
169 		return;
170 
171 	/* If there's no renegotiation to do, just resend each failed subreq. */
172 	if (!stream->prepare_write) {
173 		struct netfs_io_subrequest *subreq;
174 
175 		list_for_each_entry(subreq, &stream->subrequests, rreq_link) {
176 			if (test_bit(NETFS_SREQ_FAILED, &subreq->flags))
177 				break;
178 			if (__test_and_clear_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags)) {
179 				struct iov_iter source = subreq->io_iter;
180 
181 				iov_iter_revert(&source, subreq->len - source.count);
182 				__set_bit(NETFS_SREQ_RETRYING, &subreq->flags);
183 				netfs_get_subrequest(subreq, netfs_sreq_trace_get_resubmit);
184 				netfs_reissue_write(stream, subreq, &source);
185 			}
186 		}
187 		return;
188 	}
189 
190 	next = stream->subrequests.next;
191 
192 	do {
193 		struct netfs_io_subrequest *subreq = NULL, *from, *to, *tmp;
194 		struct iov_iter source;
195 		unsigned long long start, len;
196 		size_t part;
197 		bool boundary = false;
198 
199 		/* Go through the stream and find the next span of contiguous
200 		 * data that we then rejig (cifs, for example, needs the wsize
201 		 * renegotiating) and reissue.
202 		 */
203 		from = list_entry(next, struct netfs_io_subrequest, rreq_link);
204 		to = from;
205 		start = from->start + from->transferred;
206 		len   = from->len   - from->transferred;
207 
208 		if (test_bit(NETFS_SREQ_FAILED, &from->flags) ||
209 		    !test_bit(NETFS_SREQ_NEED_RETRY, &from->flags))
210 			return;
211 
212 		list_for_each_continue(next, &stream->subrequests) {
213 			subreq = list_entry(next, struct netfs_io_subrequest, rreq_link);
214 			if (subreq->start + subreq->transferred != start + len ||
215 			    test_bit(NETFS_SREQ_BOUNDARY, &subreq->flags) ||
216 			    !test_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags))
217 				break;
218 			to = subreq;
219 			len += to->len;
220 		}
221 
222 		/* Determine the set of buffers we're going to use.  Each
223 		 * subreq gets a subset of a single overall contiguous buffer.
224 		 */
225 		netfs_reset_iter(from);
226 		source = from->io_iter;
227 		source.count = len;
228 
229 		/* Work through the sublist. */
230 		subreq = from;
231 		list_for_each_entry_from(subreq, &stream->subrequests, rreq_link) {
232 			if (!len)
233 				break;
234 			/* Renegotiate max_len (wsize) */
235 			trace_netfs_sreq(subreq, netfs_sreq_trace_retry);
236 			__clear_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags);
237 			__set_bit(NETFS_SREQ_RETRYING, &subreq->flags);
238 			stream->prepare_write(subreq);
239 
240 			part = min(len, stream->sreq_max_len);
241 			subreq->len = part;
242 			subreq->start = start;
243 			subreq->transferred = 0;
244 			len -= part;
245 			start += part;
246 			if (len && subreq == to &&
247 			    __test_and_clear_bit(NETFS_SREQ_BOUNDARY, &to->flags))
248 				boundary = true;
249 
250 			netfs_get_subrequest(subreq, netfs_sreq_trace_get_resubmit);
251 			netfs_reissue_write(stream, subreq, &source);
252 			if (subreq == to)
253 				break;
254 		}
255 
256 		/* If we managed to use fewer subreqs, we can discard the
257 		 * excess; if we used the same number, then we're done.
258 		 */
259 		if (!len) {
260 			if (subreq == to)
261 				continue;
262 			list_for_each_entry_safe_from(subreq, tmp,
263 						      &stream->subrequests, rreq_link) {
264 				trace_netfs_sreq(subreq, netfs_sreq_trace_discard);
265 				list_del(&subreq->rreq_link);
266 				netfs_put_subrequest(subreq, false, netfs_sreq_trace_put_done);
267 				if (subreq == to)
268 					break;
269 			}
270 			continue;
271 		}
272 
273 		/* We ran out of subrequests, so we need to allocate some more
274 		 * and insert them after.
275 		 */
276 		do {
277 			subreq = netfs_alloc_subrequest(wreq);
278 			subreq->source		= to->source;
279 			subreq->start		= start;
280 			subreq->debug_index	= atomic_inc_return(&wreq->subreq_counter);
281 			subreq->stream_nr	= to->stream_nr;
282 			__set_bit(NETFS_SREQ_RETRYING, &subreq->flags);
283 
284 			trace_netfs_sreq_ref(wreq->debug_id, subreq->debug_index,
285 					     refcount_read(&subreq->ref),
286 					     netfs_sreq_trace_new);
287 			netfs_get_subrequest(subreq, netfs_sreq_trace_get_resubmit);
288 
289 			list_add(&subreq->rreq_link, &to->rreq_link);
290 			to = list_next_entry(to, rreq_link);
291 			trace_netfs_sreq(subreq, netfs_sreq_trace_retry);
292 
293 			stream->sreq_max_len	= len;
294 			stream->sreq_max_segs	= INT_MAX;
295 			switch (stream->source) {
296 			case NETFS_UPLOAD_TO_SERVER:
297 				netfs_stat(&netfs_n_wh_upload);
298 				stream->sreq_max_len = umin(len, wreq->wsize);
299 				break;
300 			case NETFS_WRITE_TO_CACHE:
301 				netfs_stat(&netfs_n_wh_write);
302 				break;
303 			default:
304 				WARN_ON_ONCE(1);
305 			}
306 
307 			stream->prepare_write(subreq);
308 
309 			part = umin(len, stream->sreq_max_len);
310 			subreq->len = subreq->transferred + part;
311 			len -= part;
312 			start += part;
313 			if (!len && boundary) {
314 				__set_bit(NETFS_SREQ_BOUNDARY, &to->flags);
315 				boundary = false;
316 			}
317 
318 			netfs_reissue_write(stream, subreq, &source);
319 			if (!len)
320 				break;
321 
322 		} while (len);
323 
324 	} while (!list_is_head(next, &stream->subrequests));
325 }
326 
327 /*
328  * Perform retries on the streams that need it.  If we're doing content
329  * encryption and the server copy changed due to a third-party write, we may
330  * need to do an RMW cycle and also rewrite the data to the cache.
331  */
332 static void netfs_retry_writes(struct netfs_io_request *wreq)
333 {
334 	struct netfs_io_subrequest *subreq;
335 	struct netfs_io_stream *stream;
336 	int s;
337 
338 	/* Wait for all outstanding I/O to quiesce before performing retries as
339 	 * we may need to renegotiate the I/O sizes.
340 	 */
341 	for (s = 0; s < NR_IO_STREAMS; s++) {
342 		stream = &wreq->io_streams[s];
343 		if (!stream->active)
344 			continue;
345 
346 		list_for_each_entry(subreq, &stream->subrequests, rreq_link) {
347 			wait_on_bit(&subreq->flags, NETFS_SREQ_IN_PROGRESS,
348 				    TASK_UNINTERRUPTIBLE);
349 		}
350 	}
351 
352 	// TODO: Enc: Fetch changed partial pages
353 	// TODO: Enc: Reencrypt content if needed.
354 	// TODO: Enc: Wind back transferred point.
355 	// TODO: Enc: Mark cache pages for retry.
356 
357 	for (s = 0; s < NR_IO_STREAMS; s++) {
358 		stream = &wreq->io_streams[s];
359 		if (stream->need_retry) {
360 			stream->need_retry = false;
361 			netfs_retry_write_stream(wreq, stream);
362 		}
363 	}
364 }
365 
366 /*
367  * Collect and assess the results of various write subrequests.  We may need to
368  * retry some of the results - or even do an RMW cycle for content crypto.
369  *
370  * Note that we have a number of parallel, overlapping lists of subrequests,
371  * one to the server and one to the local cache for example, which may not be
372  * the same size or starting position and may not even correspond in boundary
373  * alignment.
374  */
375 static void netfs_collect_write_results(struct netfs_io_request *wreq)
376 {
377 	struct netfs_io_subrequest *front, *remove;
378 	struct netfs_io_stream *stream;
379 	unsigned long long collected_to, issued_to;
380 	unsigned int notes;
381 	int s;
382 
383 	_enter("%llx-%llx", wreq->start, wreq->start + wreq->len);
384 	trace_netfs_collect(wreq);
385 	trace_netfs_rreq(wreq, netfs_rreq_trace_collect);
386 
387 reassess_streams:
388 	issued_to = atomic64_read(&wreq->issued_to);
389 	smp_rmb();
390 	collected_to = ULLONG_MAX;
391 	if (wreq->origin == NETFS_WRITEBACK ||
392 	    wreq->origin == NETFS_WRITETHROUGH ||
393 	    wreq->origin == NETFS_PGPRIV2_COPY_TO_CACHE)
394 		notes = BUFFERED;
395 	else
396 		notes = 0;
397 
398 	/* Remove completed subrequests from the front of the streams and
399 	 * advance the completion point on each stream.  We stop when we hit
400 	 * something that's in progress.  The issuer thread may be adding stuff
401 	 * to the tail whilst we're doing this.
402 	 */
403 	for (s = 0; s < NR_IO_STREAMS; s++) {
404 		stream = &wreq->io_streams[s];
405 		/* Read active flag before list pointers */
406 		if (!smp_load_acquire(&stream->active))
407 			continue;
408 
409 		front = stream->front;
410 		while (front) {
411 			trace_netfs_collect_sreq(wreq, front);
412 			//_debug("sreq [%x] %llx %zx/%zx",
413 			//       front->debug_index, front->start, front->transferred, front->len);
414 
415 			if (stream->collected_to < front->start) {
416 				trace_netfs_collect_gap(wreq, stream, issued_to, 'F');
417 				stream->collected_to = front->start;
418 			}
419 
420 			/* Stall if the front is still undergoing I/O. */
421 			if (test_bit(NETFS_SREQ_IN_PROGRESS, &front->flags)) {
422 				notes |= HIT_PENDING;
423 				break;
424 			}
425 			smp_rmb(); /* Read counters after I-P flag. */
426 
427 			if (stream->failed) {
428 				stream->collected_to = front->start + front->len;
429 				notes |= MADE_PROGRESS | SAW_FAILURE;
430 				goto cancel;
431 			}
432 			if (front->start + front->transferred > stream->collected_to) {
433 				stream->collected_to = front->start + front->transferred;
434 				stream->transferred = stream->collected_to - wreq->start;
435 				notes |= MADE_PROGRESS;
436 			}
437 			if (test_bit(NETFS_SREQ_FAILED, &front->flags)) {
438 				stream->failed = true;
439 				stream->error = front->error;
440 				if (stream->source == NETFS_UPLOAD_TO_SERVER)
441 					mapping_set_error(wreq->mapping, front->error);
442 				notes |= NEED_REASSESS | SAW_FAILURE;
443 				break;
444 			}
445 			if (front->transferred < front->len) {
446 				stream->need_retry = true;
447 				notes |= NEED_RETRY | MADE_PROGRESS;
448 				break;
449 			}
450 
451 		cancel:
452 			/* Remove if completely consumed. */
453 			spin_lock_bh(&wreq->lock);
454 
455 			remove = front;
456 			list_del_init(&front->rreq_link);
457 			front = list_first_entry_or_null(&stream->subrequests,
458 							 struct netfs_io_subrequest, rreq_link);
459 			stream->front = front;
460 			spin_unlock_bh(&wreq->lock);
461 			netfs_put_subrequest(remove, false,
462 					     notes & SAW_FAILURE ?
463 					     netfs_sreq_trace_put_cancel :
464 					     netfs_sreq_trace_put_done);
465 		}
466 
467 		/* If we have an empty stream, we need to jump it forward
468 		 * otherwise the collection point will never advance.
469 		 */
470 		if (!front && issued_to > stream->collected_to) {
471 			trace_netfs_collect_gap(wreq, stream, issued_to, 'E');
472 			stream->collected_to = issued_to;
473 		}
474 
475 		if (stream->collected_to < collected_to)
476 			collected_to = stream->collected_to;
477 	}
478 
479 	if (collected_to != ULLONG_MAX && collected_to > wreq->collected_to)
480 		wreq->collected_to = collected_to;
481 
482 	for (s = 0; s < NR_IO_STREAMS; s++) {
483 		stream = &wreq->io_streams[s];
484 		if (stream->active)
485 			trace_netfs_collect_stream(wreq, stream);
486 	}
487 
488 	trace_netfs_collect_state(wreq, wreq->collected_to, notes);
489 
490 	/* Unlock any folios that we have now finished with. */
491 	if (notes & BUFFERED) {
492 		if (wreq->cleaned_to < wreq->collected_to)
493 			netfs_writeback_unlock_folios(wreq, &notes);
494 	} else {
495 		wreq->cleaned_to = wreq->collected_to;
496 	}
497 
498 	// TODO: Discard encryption buffers
499 
500 	if (notes & NEED_RETRY)
501 		goto need_retry;
502 	if ((notes & MADE_PROGRESS) && test_bit(NETFS_RREQ_PAUSE, &wreq->flags)) {
503 		trace_netfs_rreq(wreq, netfs_rreq_trace_unpause);
504 		clear_bit_unlock(NETFS_RREQ_PAUSE, &wreq->flags);
505 		wake_up_bit(&wreq->flags, NETFS_RREQ_PAUSE);
506 	}
507 
508 	if (notes & NEED_REASSESS) {
509 		//cond_resched();
510 		goto reassess_streams;
511 	}
512 	if (notes & MADE_PROGRESS) {
513 		//cond_resched();
514 		goto reassess_streams;
515 	}
516 
517 out:
518 	netfs_put_group_many(wreq->group, wreq->nr_group_rel);
519 	wreq->nr_group_rel = 0;
520 	_leave(" = %x", notes);
521 	return;
522 
523 need_retry:
524 	/* Okay...  We're going to have to retry one or both streams.  Note
525 	 * that any partially completed op will have had any wholly transferred
526 	 * folios removed from it.
527 	 */
528 	_debug("retry");
529 	netfs_retry_writes(wreq);
530 	goto out;
531 }
532 
533 /*
534  * Perform the collection of subrequests, folios and encryption buffers.
535  */
536 void netfs_write_collection_worker(struct work_struct *work)
537 {
538 	struct netfs_io_request *wreq = container_of(work, struct netfs_io_request, work);
539 	struct netfs_inode *ictx = netfs_inode(wreq->inode);
540 	size_t transferred;
541 	int s;
542 
543 	_enter("R=%x", wreq->debug_id);
544 
545 	netfs_see_request(wreq, netfs_rreq_trace_see_work);
546 	if (!test_bit(NETFS_RREQ_IN_PROGRESS, &wreq->flags)) {
547 		netfs_put_request(wreq, false, netfs_rreq_trace_put_work);
548 		return;
549 	}
550 
551 	netfs_collect_write_results(wreq);
552 
553 	/* We're done when the app thread has finished posting subreqs and all
554 	 * the queues in all the streams are empty.
555 	 */
556 	if (!test_bit(NETFS_RREQ_ALL_QUEUED, &wreq->flags)) {
557 		netfs_put_request(wreq, false, netfs_rreq_trace_put_work);
558 		return;
559 	}
560 	smp_rmb(); /* Read ALL_QUEUED before lists. */
561 
562 	transferred = LONG_MAX;
563 	for (s = 0; s < NR_IO_STREAMS; s++) {
564 		struct netfs_io_stream *stream = &wreq->io_streams[s];
565 		if (!stream->active)
566 			continue;
567 		if (!list_empty(&stream->subrequests)) {
568 			netfs_put_request(wreq, false, netfs_rreq_trace_put_work);
569 			return;
570 		}
571 		if (stream->transferred < transferred)
572 			transferred = stream->transferred;
573 	}
574 
575 	/* Okay, declare that all I/O is complete. */
576 	wreq->transferred = transferred;
577 	trace_netfs_rreq(wreq, netfs_rreq_trace_write_done);
578 
579 	if (wreq->io_streams[1].active &&
580 	    wreq->io_streams[1].failed) {
581 		/* Cache write failure doesn't prevent writeback completion
582 		 * unless we're in disconnected mode.
583 		 */
584 		ictx->ops->invalidate_cache(wreq);
585 	}
586 
587 	if (wreq->cleanup)
588 		wreq->cleanup(wreq);
589 
590 	if (wreq->origin == NETFS_DIO_WRITE &&
591 	    wreq->mapping->nrpages) {
592 		/* mmap may have got underfoot and we may now have folios
593 		 * locally covering the region we just wrote.  Attempt to
594 		 * discard the folios, but leave in place any modified locally.
595 		 * ->write_iter() is prevented from interfering by the DIO
596 		 * counter.
597 		 */
598 		pgoff_t first = wreq->start >> PAGE_SHIFT;
599 		pgoff_t last = (wreq->start + wreq->transferred - 1) >> PAGE_SHIFT;
600 		invalidate_inode_pages2_range(wreq->mapping, first, last);
601 	}
602 
603 	if (wreq->origin == NETFS_DIO_WRITE)
604 		inode_dio_end(wreq->inode);
605 
606 	_debug("finished");
607 	trace_netfs_rreq(wreq, netfs_rreq_trace_wake_ip);
608 	clear_bit_unlock(NETFS_RREQ_IN_PROGRESS, &wreq->flags);
609 	wake_up_bit(&wreq->flags, NETFS_RREQ_IN_PROGRESS);
610 
611 	if (wreq->iocb) {
612 		size_t written = min(wreq->transferred, wreq->len);
613 		wreq->iocb->ki_pos += written;
614 		if (wreq->iocb->ki_complete)
615 			wreq->iocb->ki_complete(
616 				wreq->iocb, wreq->error ? wreq->error : written);
617 		wreq->iocb = VFS_PTR_POISON;
618 	}
619 
620 	netfs_clear_subrequests(wreq, false);
621 	netfs_put_request(wreq, false, netfs_rreq_trace_put_work_complete);
622 }
623 
624 /*
625  * Wake the collection work item.
626  */
627 void netfs_wake_write_collector(struct netfs_io_request *wreq, bool was_async)
628 {
629 	if (!work_pending(&wreq->work)) {
630 		netfs_get_request(wreq, netfs_rreq_trace_get_work);
631 		if (!queue_work(system_unbound_wq, &wreq->work))
632 			netfs_put_request(wreq, was_async, netfs_rreq_trace_put_work_nq);
633 	}
634 }
635 
636 /**
637  * netfs_write_subrequest_terminated - Note the termination of a write operation.
638  * @_op: The I/O request that has terminated.
639  * @transferred_or_error: The amount of data transferred or an error code.
640  * @was_async: The termination was asynchronous
641  *
642  * This tells the library that a contributory write I/O operation has
643  * terminated, one way or another, and that it should collect the results.
644  *
645  * The caller indicates in @transferred_or_error the outcome of the operation,
646  * supplying a positive value to indicate the number of bytes transferred or a
647  * negative error code.  The library will look after reissuing I/O operations
648  * as appropriate and writing downloaded data to the cache.
649  *
650  * If @was_async is true, the caller might be running in softirq or interrupt
651  * context and we can't sleep.
652  *
653  * When this is called, ownership of the subrequest is transferred back to the
654  * library, along with a ref.
655  *
656  * Note that %_op is a void* so that the function can be passed to
657  * kiocb::term_func without the need for a casting wrapper.
658  */
659 void netfs_write_subrequest_terminated(void *_op, ssize_t transferred_or_error,
660 				       bool was_async)
661 {
662 	struct netfs_io_subrequest *subreq = _op;
663 	struct netfs_io_request *wreq = subreq->rreq;
664 	struct netfs_io_stream *stream = &wreq->io_streams[subreq->stream_nr];
665 
666 	_enter("%x[%x] %zd", wreq->debug_id, subreq->debug_index, transferred_or_error);
667 
668 	switch (subreq->source) {
669 	case NETFS_UPLOAD_TO_SERVER:
670 		netfs_stat(&netfs_n_wh_upload_done);
671 		break;
672 	case NETFS_WRITE_TO_CACHE:
673 		netfs_stat(&netfs_n_wh_write_done);
674 		break;
675 	case NETFS_INVALID_WRITE:
676 		break;
677 	default:
678 		BUG();
679 	}
680 
681 	if (IS_ERR_VALUE(transferred_or_error)) {
682 		subreq->error = transferred_or_error;
683 		if (subreq->error == -EAGAIN)
684 			set_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags);
685 		else
686 			set_bit(NETFS_SREQ_FAILED, &subreq->flags);
687 		trace_netfs_failure(wreq, subreq, transferred_or_error, netfs_fail_write);
688 
689 		switch (subreq->source) {
690 		case NETFS_WRITE_TO_CACHE:
691 			netfs_stat(&netfs_n_wh_write_failed);
692 			break;
693 		case NETFS_UPLOAD_TO_SERVER:
694 			netfs_stat(&netfs_n_wh_upload_failed);
695 			break;
696 		default:
697 			break;
698 		}
699 		trace_netfs_rreq(wreq, netfs_rreq_trace_set_pause);
700 		set_bit(NETFS_RREQ_PAUSE, &wreq->flags);
701 	} else {
702 		if (WARN(transferred_or_error > subreq->len - subreq->transferred,
703 			 "Subreq excess write: R=%x[%x] %zd > %zu - %zu",
704 			 wreq->debug_id, subreq->debug_index,
705 			 transferred_or_error, subreq->len, subreq->transferred))
706 			transferred_or_error = subreq->len - subreq->transferred;
707 
708 		subreq->error = 0;
709 		subreq->transferred += transferred_or_error;
710 
711 		if (subreq->transferred < subreq->len)
712 			set_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags);
713 	}
714 
715 	trace_netfs_sreq(subreq, netfs_sreq_trace_terminated);
716 
717 	clear_bit_unlock(NETFS_SREQ_IN_PROGRESS, &subreq->flags);
718 	wake_up_bit(&subreq->flags, NETFS_SREQ_IN_PROGRESS);
719 
720 	/* If we are at the head of the queue, wake up the collector,
721 	 * transferring a ref to it if we were the ones to do so.
722 	 */
723 	if (list_is_first(&subreq->rreq_link, &stream->subrequests))
724 		netfs_wake_write_collector(wreq, was_async);
725 
726 	netfs_put_subrequest(subreq, was_async, netfs_sreq_trace_put_terminated);
727 }
728 EXPORT_SYMBOL(netfs_write_subrequest_terminated);
729