xref: /linux/fs/netfs/read_collect.c (revision b5329d5a35582abbef57562f9fb6cb26a643f252)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Network filesystem read subrequest result collection, assessment and
3  * 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 <linux/task_io_accounting_ops.h>
15 #include "internal.h"
16 
17 /* Notes made in the collector */
18 #define HIT_PENDING	0x01	/* A front op was still pending */
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 COPY_TO_CACHE	0x40	/* Need to copy subrequest to cache */
23 #define ABANDON_SREQ	0x80	/* Need to abandon untransferred part of subrequest */
24 
25 /*
26  * Clear the unread part of an I/O request.
27  */
netfs_clear_unread(struct netfs_io_subrequest * subreq)28 static void netfs_clear_unread(struct netfs_io_subrequest *subreq)
29 {
30 	netfs_reset_iter(subreq);
31 	WARN_ON_ONCE(subreq->len - subreq->transferred != iov_iter_count(&subreq->io_iter));
32 	iov_iter_zero(iov_iter_count(&subreq->io_iter), &subreq->io_iter);
33 	if (subreq->start + subreq->transferred >= subreq->rreq->i_size)
34 		__set_bit(NETFS_SREQ_HIT_EOF, &subreq->flags);
35 }
36 
37 /*
38  * Flush, mark and unlock a folio that's now completely read.  If we want to
39  * cache the folio, we set the group to NETFS_FOLIO_COPY_TO_CACHE, mark it
40  * dirty and let writeback handle it.
41  */
netfs_unlock_read_folio(struct netfs_io_request * rreq,struct folio_queue * folioq,int slot)42 static void netfs_unlock_read_folio(struct netfs_io_request *rreq,
43 				    struct folio_queue *folioq,
44 				    int slot)
45 {
46 	struct netfs_folio *finfo;
47 	struct folio *folio = folioq_folio(folioq, slot);
48 
49 	if (unlikely(folio_pos(folio) < rreq->abandon_to)) {
50 		trace_netfs_folio(folio, netfs_folio_trace_abandon);
51 		goto just_unlock;
52 	}
53 
54 	flush_dcache_folio(folio);
55 	folio_mark_uptodate(folio);
56 
57 	if (!test_bit(NETFS_RREQ_USE_PGPRIV2, &rreq->flags)) {
58 		finfo = netfs_folio_info(folio);
59 		if (finfo) {
60 			trace_netfs_folio(folio, netfs_folio_trace_filled_gaps);
61 			if (finfo->netfs_group)
62 				folio_change_private(folio, finfo->netfs_group);
63 			else
64 				folio_detach_private(folio);
65 			kfree(finfo);
66 		}
67 
68 		if (test_bit(NETFS_RREQ_FOLIO_COPY_TO_CACHE, &rreq->flags)) {
69 			if (!WARN_ON_ONCE(folio_get_private(folio) != NULL)) {
70 				trace_netfs_folio(folio, netfs_folio_trace_copy_to_cache);
71 				folio_attach_private(folio, NETFS_FOLIO_COPY_TO_CACHE);
72 				folio_mark_dirty(folio);
73 			}
74 		} else {
75 			trace_netfs_folio(folio, netfs_folio_trace_read_done);
76 		}
77 
78 		folioq_clear(folioq, slot);
79 	} else {
80 		// TODO: Use of PG_private_2 is deprecated.
81 		if (test_bit(NETFS_RREQ_FOLIO_COPY_TO_CACHE, &rreq->flags))
82 			netfs_pgpriv2_copy_to_cache(rreq, folio);
83 	}
84 
85 just_unlock:
86 	if (!test_bit(NETFS_RREQ_DONT_UNLOCK_FOLIOS, &rreq->flags)) {
87 		if (folio->index == rreq->no_unlock_folio &&
88 		    test_bit(NETFS_RREQ_NO_UNLOCK_FOLIO, &rreq->flags)) {
89 			_debug("no unlock");
90 		} else {
91 			trace_netfs_folio(folio, netfs_folio_trace_read_unlock);
92 			folio_unlock(folio);
93 		}
94 	}
95 
96 	folioq_clear(folioq, slot);
97 }
98 
99 /*
100  * Unlock any folios we've finished with.
101  */
netfs_read_unlock_folios(struct netfs_io_request * rreq,unsigned int * notes)102 static void netfs_read_unlock_folios(struct netfs_io_request *rreq,
103 				     unsigned int *notes)
104 {
105 	struct folio_queue *folioq = rreq->buffer.tail;
106 	unsigned long long collected_to = rreq->collected_to;
107 	unsigned int slot = rreq->buffer.first_tail_slot;
108 
109 	if (rreq->cleaned_to >= rreq->collected_to)
110 		return;
111 
112 	// TODO: Begin decryption
113 
114 	if (slot >= folioq_nr_slots(folioq)) {
115 		folioq = rolling_buffer_delete_spent(&rreq->buffer);
116 		if (!folioq) {
117 			rreq->front_folio_order = 0;
118 			return;
119 		}
120 		slot = 0;
121 	}
122 
123 	for (;;) {
124 		struct folio *folio;
125 		unsigned long long fpos, fend;
126 		unsigned int order;
127 		size_t fsize;
128 
129 		if (*notes & COPY_TO_CACHE)
130 			set_bit(NETFS_RREQ_FOLIO_COPY_TO_CACHE, &rreq->flags);
131 
132 		folio = folioq_folio(folioq, slot);
133 		if (WARN_ONCE(!folio_test_locked(folio),
134 			      "R=%08x: folio %lx is not locked\n",
135 			      rreq->debug_id, folio->index))
136 			trace_netfs_folio(folio, netfs_folio_trace_not_locked);
137 
138 		order = folioq_folio_order(folioq, slot);
139 		rreq->front_folio_order = order;
140 		fsize = PAGE_SIZE << order;
141 		fpos = folio_pos(folio);
142 		fend = umin(fpos + fsize, rreq->i_size);
143 
144 		trace_netfs_collect_folio(rreq, folio, fend, collected_to);
145 
146 		/* Unlock any folio we've transferred all of. */
147 		if (collected_to < fend)
148 			break;
149 
150 		netfs_unlock_read_folio(rreq, folioq, slot);
151 		WRITE_ONCE(rreq->cleaned_to, fpos + fsize);
152 		*notes |= MADE_PROGRESS;
153 
154 		clear_bit(NETFS_RREQ_FOLIO_COPY_TO_CACHE, &rreq->flags);
155 
156 		/* Clean up the head folioq.  If we clear an entire folioq, then
157 		 * we can get rid of it provided it's not also the tail folioq
158 		 * being filled by the issuer.
159 		 */
160 		folioq_clear(folioq, slot);
161 		slot++;
162 		if (slot >= folioq_nr_slots(folioq)) {
163 			folioq = rolling_buffer_delete_spent(&rreq->buffer);
164 			if (!folioq)
165 				goto done;
166 			slot = 0;
167 			trace_netfs_folioq(folioq, netfs_trace_folioq_read_progress);
168 		}
169 
170 		if (fpos + fsize >= collected_to)
171 			break;
172 	}
173 
174 	rreq->buffer.tail = folioq;
175 done:
176 	rreq->buffer.first_tail_slot = slot;
177 }
178 
179 /*
180  * Collect and assess the results of various read subrequests.  We may need to
181  * retry some of the results.
182  *
183  * Note that we have a sequence of subrequests, which may be drawing on
184  * different sources and may or may not be the same size or starting position
185  * and may not even correspond in boundary alignment.
186  */
netfs_collect_read_results(struct netfs_io_request * rreq)187 static void netfs_collect_read_results(struct netfs_io_request *rreq)
188 {
189 	struct netfs_io_subrequest *front, *remove;
190 	struct netfs_io_stream *stream = &rreq->io_streams[0];
191 	unsigned int notes;
192 
193 	_enter("%llx-%llx", rreq->start, rreq->start + rreq->len);
194 	trace_netfs_rreq(rreq, netfs_rreq_trace_collect);
195 	trace_netfs_collect(rreq);
196 
197 reassess:
198 	if (rreq->origin == NETFS_READAHEAD ||
199 	    rreq->origin == NETFS_READPAGE ||
200 	    rreq->origin == NETFS_READ_FOR_WRITE)
201 		notes = BUFFERED;
202 	else
203 		notes = 0;
204 
205 	/* Remove completed subrequests from the front of the stream and
206 	 * advance the completion point.  We stop when we hit something that's
207 	 * in progress.  The issuer thread may be adding stuff to the tail
208 	 * whilst we're doing this.
209 	 */
210 	front = READ_ONCE(stream->front);
211 	while (front) {
212 		size_t transferred;
213 
214 		trace_netfs_collect_sreq(rreq, front);
215 		_debug("sreq [%x] %llx %zx/%zx",
216 		       front->debug_index, front->start, front->transferred, front->len);
217 
218 		if (stream->collected_to < front->start) {
219 			trace_netfs_collect_gap(rreq, stream, front->start, 'F');
220 			stream->collected_to = front->start;
221 		}
222 
223 		if (test_bit(NETFS_SREQ_IN_PROGRESS, &front->flags))
224 			notes |= HIT_PENDING;
225 		smp_rmb(); /* Read counters after IN_PROGRESS flag. */
226 		transferred = READ_ONCE(front->transferred);
227 
228 		/* If we can now collect the next folio, do so.  We don't want
229 		 * to defer this as we have to decide whether we need to copy
230 		 * to the cache or not, and that may differ between adjacent
231 		 * subreqs.
232 		 */
233 		if (notes & BUFFERED) {
234 			size_t fsize = PAGE_SIZE << rreq->front_folio_order;
235 
236 			/* Clear the tail of a short read. */
237 			if (!(notes & HIT_PENDING) &&
238 			    front->error == 0 &&
239 			    transferred < front->len &&
240 			    (test_bit(NETFS_SREQ_HIT_EOF, &front->flags) ||
241 			     test_bit(NETFS_SREQ_CLEAR_TAIL, &front->flags))) {
242 				netfs_clear_unread(front);
243 				transferred = front->transferred = front->len;
244 				trace_netfs_sreq(front, netfs_sreq_trace_clear);
245 			}
246 
247 			stream->collected_to = front->start + transferred;
248 			rreq->collected_to = stream->collected_to;
249 
250 			if (test_bit(NETFS_SREQ_COPY_TO_CACHE, &front->flags))
251 				notes |= COPY_TO_CACHE;
252 
253 			if (test_bit(NETFS_SREQ_FAILED, &front->flags)) {
254 				rreq->abandon_to = front->start + front->len;
255 				front->transferred = front->len;
256 				transferred = front->len;
257 				trace_netfs_rreq(rreq, netfs_rreq_trace_set_abandon);
258 			}
259 			if (front->start + transferred >= rreq->cleaned_to + fsize ||
260 			    test_bit(NETFS_SREQ_HIT_EOF, &front->flags))
261 				netfs_read_unlock_folios(rreq, &notes);
262 		} else {
263 			stream->collected_to = front->start + transferred;
264 			rreq->collected_to = stream->collected_to;
265 		}
266 
267 		/* Stall if the front is still undergoing I/O. */
268 		if (notes & HIT_PENDING)
269 			break;
270 
271 		if (test_bit(NETFS_SREQ_FAILED, &front->flags)) {
272 			if (!stream->failed) {
273 				stream->error = front->error;
274 				rreq->error = front->error;
275 				set_bit(NETFS_RREQ_FAILED, &rreq->flags);
276 				stream->failed = true;
277 			}
278 			notes |= MADE_PROGRESS | ABANDON_SREQ;
279 		} else if (test_bit(NETFS_SREQ_NEED_RETRY, &front->flags)) {
280 			stream->need_retry = true;
281 			notes |= NEED_RETRY | MADE_PROGRESS;
282 			break;
283 		} else {
284 			if (!stream->failed)
285 				stream->transferred = stream->collected_to - rreq->start;
286 			notes |= MADE_PROGRESS;
287 		}
288 
289 		/* Remove if completely consumed. */
290 		stream->source = front->source;
291 		spin_lock(&rreq->lock);
292 
293 		remove = front;
294 		trace_netfs_sreq(front, netfs_sreq_trace_discard);
295 		list_del_init(&front->rreq_link);
296 		front = list_first_entry_or_null(&stream->subrequests,
297 						 struct netfs_io_subrequest, rreq_link);
298 		stream->front = front;
299 		spin_unlock(&rreq->lock);
300 		netfs_put_subrequest(remove, false,
301 				     notes & ABANDON_SREQ ?
302 				     netfs_sreq_trace_put_abandon :
303 				     netfs_sreq_trace_put_done);
304 	}
305 
306 	trace_netfs_collect_stream(rreq, stream);
307 	trace_netfs_collect_state(rreq, rreq->collected_to, notes);
308 
309 	if (!(notes & BUFFERED))
310 		rreq->cleaned_to = rreq->collected_to;
311 
312 	if (notes & NEED_RETRY)
313 		goto need_retry;
314 	if ((notes & MADE_PROGRESS) && test_bit(NETFS_RREQ_PAUSE, &rreq->flags)) {
315 		trace_netfs_rreq(rreq, netfs_rreq_trace_unpause);
316 		clear_bit_unlock(NETFS_RREQ_PAUSE, &rreq->flags);
317 		smp_mb__after_atomic(); /* Set PAUSE before task state */
318 		wake_up(&rreq->waitq);
319 	}
320 
321 	if (notes & MADE_PROGRESS) {
322 		//cond_resched();
323 		goto reassess;
324 	}
325 
326 out:
327 	_leave(" = %x", notes);
328 	return;
329 
330 need_retry:
331 	/* Okay...  We're going to have to retry parts of the stream.  Note
332 	 * that any partially completed op will have had any wholly transferred
333 	 * folios removed from it.
334 	 */
335 	_debug("retry");
336 	netfs_retry_reads(rreq);
337 	goto out;
338 }
339 
340 /*
341  * Do page flushing and suchlike after DIO.
342  */
netfs_rreq_assess_dio(struct netfs_io_request * rreq)343 static void netfs_rreq_assess_dio(struct netfs_io_request *rreq)
344 {
345 	struct netfs_io_subrequest *subreq;
346 	struct netfs_io_stream *stream = &rreq->io_streams[0];
347 	unsigned int i;
348 
349 	/* Collect unbuffered reads and direct reads, adding up the transfer
350 	 * sizes until we find the first short or failed subrequest.
351 	 */
352 	list_for_each_entry(subreq, &stream->subrequests, rreq_link) {
353 		rreq->transferred += subreq->transferred;
354 
355 		if (subreq->transferred < subreq->len ||
356 		    test_bit(NETFS_SREQ_FAILED, &subreq->flags)) {
357 			rreq->error = subreq->error;
358 			break;
359 		}
360 	}
361 
362 	if (rreq->origin == NETFS_DIO_READ) {
363 		for (i = 0; i < rreq->direct_bv_count; i++) {
364 			flush_dcache_page(rreq->direct_bv[i].bv_page);
365 			// TODO: cifs marks pages in the destination buffer
366 			// dirty under some circumstances after a read.  Do we
367 			// need to do that too?
368 			set_page_dirty(rreq->direct_bv[i].bv_page);
369 		}
370 	}
371 
372 	if (rreq->iocb) {
373 		rreq->iocb->ki_pos += rreq->transferred;
374 		if (rreq->iocb->ki_complete)
375 			rreq->iocb->ki_complete(
376 				rreq->iocb, rreq->error ? rreq->error : rreq->transferred);
377 	}
378 	if (rreq->netfs_ops->done)
379 		rreq->netfs_ops->done(rreq);
380 	if (rreq->origin == NETFS_DIO_READ)
381 		inode_dio_end(rreq->inode);
382 }
383 
384 /*
385  * Do processing after reading a monolithic single object.
386  */
netfs_rreq_assess_single(struct netfs_io_request * rreq)387 static void netfs_rreq_assess_single(struct netfs_io_request *rreq)
388 {
389 	struct netfs_io_stream *stream = &rreq->io_streams[0];
390 
391 	if (!rreq->error && stream->source == NETFS_DOWNLOAD_FROM_SERVER &&
392 	    fscache_resources_valid(&rreq->cache_resources)) {
393 		trace_netfs_rreq(rreq, netfs_rreq_trace_dirty);
394 		netfs_single_mark_inode_dirty(rreq->inode);
395 	}
396 
397 	if (rreq->iocb) {
398 		rreq->iocb->ki_pos += rreq->transferred;
399 		if (rreq->iocb->ki_complete)
400 			rreq->iocb->ki_complete(
401 				rreq->iocb, rreq->error ? rreq->error : rreq->transferred);
402 	}
403 	if (rreq->netfs_ops->done)
404 		rreq->netfs_ops->done(rreq);
405 }
406 
407 /*
408  * Perform the collection of subrequests and folios.
409  *
410  * Note that we're in normal kernel thread context at this point, possibly
411  * running on a workqueue.
412  */
netfs_read_collection(struct netfs_io_request * rreq)413 static void netfs_read_collection(struct netfs_io_request *rreq)
414 {
415 	struct netfs_io_stream *stream = &rreq->io_streams[0];
416 
417 	netfs_collect_read_results(rreq);
418 
419 	/* We're done when the app thread has finished posting subreqs and the
420 	 * queue is empty.
421 	 */
422 	if (!test_bit(NETFS_RREQ_ALL_QUEUED, &rreq->flags))
423 		return;
424 	smp_rmb(); /* Read ALL_QUEUED before subreq lists. */
425 
426 	if (!list_empty(&stream->subrequests))
427 		return;
428 
429 	/* Okay, declare that all I/O is complete. */
430 	rreq->transferred = stream->transferred;
431 	trace_netfs_rreq(rreq, netfs_rreq_trace_complete);
432 
433 	//netfs_rreq_is_still_valid(rreq);
434 
435 	switch (rreq->origin) {
436 	case NETFS_DIO_READ:
437 	case NETFS_READ_GAPS:
438 		netfs_rreq_assess_dio(rreq);
439 		break;
440 	case NETFS_READ_SINGLE:
441 		netfs_rreq_assess_single(rreq);
442 		break;
443 	default:
444 		break;
445 	}
446 	task_io_account_read(rreq->transferred);
447 
448 	trace_netfs_rreq(rreq, netfs_rreq_trace_wake_ip);
449 	clear_and_wake_up_bit(NETFS_RREQ_IN_PROGRESS, &rreq->flags);
450 
451 	trace_netfs_rreq(rreq, netfs_rreq_trace_done);
452 	netfs_clear_subrequests(rreq, false);
453 	netfs_unlock_abandoned_read_pages(rreq);
454 	if (unlikely(rreq->copy_to_cache))
455 		netfs_pgpriv2_end_copy_to_cache(rreq);
456 }
457 
netfs_read_collection_worker(struct work_struct * work)458 void netfs_read_collection_worker(struct work_struct *work)
459 {
460 	struct netfs_io_request *rreq = container_of(work, struct netfs_io_request, work);
461 
462 	netfs_see_request(rreq, netfs_rreq_trace_see_work);
463 	if (test_bit(NETFS_RREQ_IN_PROGRESS, &rreq->flags))
464 		netfs_read_collection(rreq);
465 	netfs_put_request(rreq, false, netfs_rreq_trace_put_work);
466 }
467 
468 /*
469  * Wake the collection work item.
470  */
netfs_wake_read_collector(struct netfs_io_request * rreq)471 void netfs_wake_read_collector(struct netfs_io_request *rreq)
472 {
473 	if (test_bit(NETFS_RREQ_OFFLOAD_COLLECTION, &rreq->flags) &&
474 	    !test_bit(NETFS_RREQ_RETRYING, &rreq->flags)) {
475 		if (!work_pending(&rreq->work)) {
476 			netfs_get_request(rreq, netfs_rreq_trace_get_work);
477 			if (!queue_work(system_unbound_wq, &rreq->work))
478 				netfs_put_request(rreq, true, netfs_rreq_trace_put_work_nq);
479 		}
480 	} else {
481 		trace_netfs_rreq(rreq, netfs_rreq_trace_wake_queue);
482 		wake_up(&rreq->waitq);
483 	}
484 }
485 
486 /**
487  * netfs_read_subreq_progress - Note progress of a read operation.
488  * @subreq: The read request that has terminated.
489  *
490  * This tells the read side of netfs lib that a contributory I/O operation has
491  * made some progress and that it may be possible to unlock some folios.
492  *
493  * Before calling, the filesystem should update subreq->transferred to track
494  * the amount of data copied into the output buffer.
495  */
netfs_read_subreq_progress(struct netfs_io_subrequest * subreq)496 void netfs_read_subreq_progress(struct netfs_io_subrequest *subreq)
497 {
498 	struct netfs_io_request *rreq = subreq->rreq;
499 	struct netfs_io_stream *stream = &rreq->io_streams[0];
500 	size_t fsize = PAGE_SIZE << rreq->front_folio_order;
501 
502 	trace_netfs_sreq(subreq, netfs_sreq_trace_progress);
503 
504 	/* If we are at the head of the queue, wake up the collector,
505 	 * getting a ref to it if we were the ones to do so.
506 	 */
507 	if (subreq->start + subreq->transferred > rreq->cleaned_to + fsize &&
508 	    (rreq->origin == NETFS_READAHEAD ||
509 	     rreq->origin == NETFS_READPAGE ||
510 	     rreq->origin == NETFS_READ_FOR_WRITE) &&
511 	    list_is_first(&subreq->rreq_link, &stream->subrequests)
512 	    ) {
513 		__set_bit(NETFS_SREQ_MADE_PROGRESS, &subreq->flags);
514 		netfs_wake_read_collector(rreq);
515 	}
516 }
517 EXPORT_SYMBOL(netfs_read_subreq_progress);
518 
519 /**
520  * netfs_read_subreq_terminated - Note the termination of an I/O operation.
521  * @subreq: The I/O request that has terminated.
522  *
523  * This tells the read helper that a contributory I/O operation has terminated,
524  * one way or another, and that it should integrate the results.
525  *
526  * The caller indicates the outcome of the operation through @subreq->error,
527  * supplying 0 to indicate a successful or retryable transfer (if
528  * NETFS_SREQ_NEED_RETRY is set) or a negative error code.  The helper will
529  * look after reissuing I/O operations as appropriate and writing downloaded
530  * data to the cache.
531  *
532  * Before calling, the filesystem should update subreq->transferred to track
533  * the amount of data copied into the output buffer.
534  */
netfs_read_subreq_terminated(struct netfs_io_subrequest * subreq)535 void netfs_read_subreq_terminated(struct netfs_io_subrequest *subreq)
536 {
537 	struct netfs_io_request *rreq = subreq->rreq;
538 	struct netfs_io_stream *stream = &rreq->io_streams[0];
539 
540 	switch (subreq->source) {
541 	case NETFS_READ_FROM_CACHE:
542 		netfs_stat(&netfs_n_rh_read_done);
543 		break;
544 	case NETFS_DOWNLOAD_FROM_SERVER:
545 		netfs_stat(&netfs_n_rh_download_done);
546 		break;
547 	default:
548 		break;
549 	}
550 
551 	/* Deal with retry requests, short reads and errors.  If we retry
552 	 * but don't make progress, we abandon the attempt.
553 	 */
554 	if (!subreq->error && subreq->transferred < subreq->len) {
555 		if (test_bit(NETFS_SREQ_HIT_EOF, &subreq->flags)) {
556 			trace_netfs_sreq(subreq, netfs_sreq_trace_hit_eof);
557 		} else if (test_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags)) {
558 			trace_netfs_sreq(subreq, netfs_sreq_trace_need_clear);
559 		} else if (test_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags)) {
560 			trace_netfs_sreq(subreq, netfs_sreq_trace_need_retry);
561 		} else if (test_bit(NETFS_SREQ_MADE_PROGRESS, &subreq->flags)) {
562 			__set_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags);
563 			trace_netfs_sreq(subreq, netfs_sreq_trace_partial_read);
564 		} else {
565 			__set_bit(NETFS_SREQ_FAILED, &subreq->flags);
566 			subreq->error = -ENODATA;
567 			trace_netfs_sreq(subreq, netfs_sreq_trace_short);
568 		}
569 	}
570 
571 	if (unlikely(subreq->error < 0)) {
572 		trace_netfs_failure(rreq, subreq, subreq->error, netfs_fail_read);
573 		if (subreq->source == NETFS_READ_FROM_CACHE) {
574 			netfs_stat(&netfs_n_rh_read_failed);
575 			__set_bit(NETFS_SREQ_NEED_RETRY, &subreq->flags);
576 		} else {
577 			netfs_stat(&netfs_n_rh_download_failed);
578 			__set_bit(NETFS_SREQ_FAILED, &subreq->flags);
579 		}
580 		trace_netfs_rreq(rreq, netfs_rreq_trace_set_pause);
581 		set_bit(NETFS_RREQ_PAUSE, &rreq->flags);
582 	}
583 
584 	trace_netfs_sreq(subreq, netfs_sreq_trace_terminated);
585 
586 	clear_bit_unlock(NETFS_SREQ_IN_PROGRESS, &subreq->flags);
587 	smp_mb__after_atomic(); /* Clear IN_PROGRESS before task state */
588 
589 	/* If we are at the head of the queue, wake up the collector. */
590 	if (list_is_first(&subreq->rreq_link, &stream->subrequests) ||
591 	    test_bit(NETFS_RREQ_RETRYING, &rreq->flags))
592 		netfs_wake_read_collector(rreq);
593 
594 	netfs_put_subrequest(subreq, true, netfs_sreq_trace_put_terminated);
595 }
596 EXPORT_SYMBOL(netfs_read_subreq_terminated);
597 
598 /*
599  * Handle termination of a read from the cache.
600  */
netfs_cache_read_terminated(void * priv,ssize_t transferred_or_error,bool was_async)601 void netfs_cache_read_terminated(void *priv, ssize_t transferred_or_error, bool was_async)
602 {
603 	struct netfs_io_subrequest *subreq = priv;
604 
605 	if (transferred_or_error > 0) {
606 		subreq->error = 0;
607 		if (transferred_or_error > 0) {
608 			subreq->transferred += transferred_or_error;
609 			__set_bit(NETFS_SREQ_MADE_PROGRESS, &subreq->flags);
610 		}
611 	} else {
612 		subreq->error = transferred_or_error;
613 	}
614 	netfs_read_subreq_terminated(subreq);
615 }
616 
617 /*
618  * Wait for the read operation to complete, successfully or otherwise.
619  */
netfs_wait_for_read(struct netfs_io_request * rreq)620 ssize_t netfs_wait_for_read(struct netfs_io_request *rreq)
621 {
622 	struct netfs_io_subrequest *subreq;
623 	struct netfs_io_stream *stream = &rreq->io_streams[0];
624 	DEFINE_WAIT(myself);
625 	ssize_t ret;
626 
627 	for (;;) {
628 		trace_netfs_rreq(rreq, netfs_rreq_trace_wait_queue);
629 		prepare_to_wait(&rreq->waitq, &myself, TASK_UNINTERRUPTIBLE);
630 
631 		subreq = list_first_entry_or_null(&stream->subrequests,
632 						  struct netfs_io_subrequest, rreq_link);
633 		if (subreq &&
634 		    (!test_bit(NETFS_SREQ_IN_PROGRESS, &subreq->flags) ||
635 		     test_bit(NETFS_SREQ_MADE_PROGRESS, &subreq->flags))) {
636 			__set_current_state(TASK_RUNNING);
637 			netfs_read_collection(rreq);
638 			continue;
639 		}
640 
641 		if (!test_bit(NETFS_RREQ_IN_PROGRESS, &rreq->flags))
642 			break;
643 
644 		schedule();
645 		trace_netfs_rreq(rreq, netfs_rreq_trace_woke_queue);
646 	}
647 
648 	finish_wait(&rreq->waitq, &myself);
649 
650 	ret = rreq->error;
651 	if (ret == 0) {
652 		ret = rreq->transferred;
653 		switch (rreq->origin) {
654 		case NETFS_DIO_READ:
655 		case NETFS_READ_SINGLE:
656 			ret = rreq->transferred;
657 			break;
658 		default:
659 			if (rreq->submitted < rreq->len) {
660 				trace_netfs_failure(rreq, NULL, ret, netfs_fail_short_read);
661 				ret = -EIO;
662 			}
663 			break;
664 		}
665 	}
666 
667 	return ret;
668 }
669 
670 /*
671  * Wait for a paused read operation to unpause or complete in some manner.
672  */
netfs_wait_for_pause(struct netfs_io_request * rreq)673 void netfs_wait_for_pause(struct netfs_io_request *rreq)
674 {
675 	struct netfs_io_subrequest *subreq;
676 	struct netfs_io_stream *stream = &rreq->io_streams[0];
677 	DEFINE_WAIT(myself);
678 
679 	trace_netfs_rreq(rreq, netfs_rreq_trace_wait_pause);
680 
681 	for (;;) {
682 		trace_netfs_rreq(rreq, netfs_rreq_trace_wait_queue);
683 		prepare_to_wait(&rreq->waitq, &myself, TASK_UNINTERRUPTIBLE);
684 
685 		if (!test_bit(NETFS_RREQ_OFFLOAD_COLLECTION, &rreq->flags)) {
686 			subreq = list_first_entry_or_null(&stream->subrequests,
687 							  struct netfs_io_subrequest, rreq_link);
688 			if (subreq &&
689 			    (!test_bit(NETFS_SREQ_IN_PROGRESS, &subreq->flags) ||
690 			     test_bit(NETFS_SREQ_MADE_PROGRESS, &subreq->flags))) {
691 				__set_current_state(TASK_RUNNING);
692 				netfs_read_collection(rreq);
693 				continue;
694 			}
695 		}
696 
697 		if (!test_bit(NETFS_RREQ_IN_PROGRESS, &rreq->flags) ||
698 		    !test_bit(NETFS_RREQ_PAUSE, &rreq->flags))
699 			break;
700 
701 		schedule();
702 		trace_netfs_rreq(rreq, netfs_rreq_trace_woke_queue);
703 	}
704 
705 	finish_wait(&rreq->waitq, &myself);
706 }
707