1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Network filesystem high-level read support.
4 * Copyright (C) 2021 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
8 #include <linux/module.h>
9 #include <linux/export.h>
12 #include <linux/pagemap.h>
13 #include <linux/slab.h>
14 #include <linux/uio.h>
15 #include <linux/sched/mm.h>
16 #include <linux/task_io_accounting_ops.h>
17 #include <linux/netfs.h>
19 #define CREATE_TRACE_POINTS
20 #include <trace/events/netfs.h>
22 MODULE_DESCRIPTION("Network fs support");
23 MODULE_AUTHOR("Red Hat, Inc.");
24 MODULE_LICENSE("GPL");
27 module_param_named(debug, netfs_debug, uint, S_IWUSR | S_IRUGO);
28 MODULE_PARM_DESC(netfs_debug, "Netfs support debugging mask");
30 static void netfs_rreq_work(struct work_struct *);
31 static void __netfs_put_subrequest(struct netfs_read_subrequest *, bool);
33 static void netfs_put_subrequest(struct netfs_read_subrequest *subreq,
36 if (refcount_dec_and_test(&subreq->usage))
37 __netfs_put_subrequest(subreq, was_async);
40 static struct netfs_read_request *netfs_alloc_read_request(
41 const struct netfs_read_request_ops *ops, void *netfs_priv,
44 static atomic_t debug_ids;
45 struct netfs_read_request *rreq;
47 rreq = kzalloc(sizeof(struct netfs_read_request), GFP_KERNEL);
49 rreq->netfs_ops = ops;
50 rreq->netfs_priv = netfs_priv;
51 rreq->inode = file_inode(file);
52 rreq->i_size = i_size_read(rreq->inode);
53 rreq->debug_id = atomic_inc_return(&debug_ids);
54 INIT_LIST_HEAD(&rreq->subrequests);
55 INIT_WORK(&rreq->work, netfs_rreq_work);
56 refcount_set(&rreq->usage, 1);
57 __set_bit(NETFS_RREQ_IN_PROGRESS, &rreq->flags);
59 ops->init_rreq(rreq, file);
60 netfs_stat(&netfs_n_rh_rreq);
66 static void netfs_get_read_request(struct netfs_read_request *rreq)
68 refcount_inc(&rreq->usage);
71 static void netfs_rreq_clear_subreqs(struct netfs_read_request *rreq,
74 struct netfs_read_subrequest *subreq;
76 while (!list_empty(&rreq->subrequests)) {
77 subreq = list_first_entry(&rreq->subrequests,
78 struct netfs_read_subrequest, rreq_link);
79 list_del(&subreq->rreq_link);
80 netfs_put_subrequest(subreq, was_async);
84 static void netfs_free_read_request(struct work_struct *work)
86 struct netfs_read_request *rreq =
87 container_of(work, struct netfs_read_request, work);
88 netfs_rreq_clear_subreqs(rreq, false);
90 rreq->netfs_ops->cleanup(rreq->mapping, rreq->netfs_priv);
91 trace_netfs_rreq(rreq, netfs_rreq_trace_free);
92 if (rreq->cache_resources.ops)
93 rreq->cache_resources.ops->end_operation(&rreq->cache_resources);
95 netfs_stat_d(&netfs_n_rh_rreq);
98 static void netfs_put_read_request(struct netfs_read_request *rreq, bool was_async)
100 if (refcount_dec_and_test(&rreq->usage)) {
102 rreq->work.func = netfs_free_read_request;
103 if (!queue_work(system_unbound_wq, &rreq->work))
106 netfs_free_read_request(&rreq->work);
112 * Allocate and partially initialise an I/O request structure.
114 static struct netfs_read_subrequest *netfs_alloc_subrequest(
115 struct netfs_read_request *rreq)
117 struct netfs_read_subrequest *subreq;
119 subreq = kzalloc(sizeof(struct netfs_read_subrequest), GFP_KERNEL);
121 INIT_LIST_HEAD(&subreq->rreq_link);
122 refcount_set(&subreq->usage, 2);
124 netfs_get_read_request(rreq);
125 netfs_stat(&netfs_n_rh_sreq);
131 static void netfs_get_read_subrequest(struct netfs_read_subrequest *subreq)
133 refcount_inc(&subreq->usage);
136 static void __netfs_put_subrequest(struct netfs_read_subrequest *subreq,
139 struct netfs_read_request *rreq = subreq->rreq;
141 trace_netfs_sreq(subreq, netfs_sreq_trace_free);
143 netfs_stat_d(&netfs_n_rh_sreq);
144 netfs_put_read_request(rreq, was_async);
148 * Clear the unread part of an I/O request.
150 static void netfs_clear_unread(struct netfs_read_subrequest *subreq)
152 struct iov_iter iter;
154 iov_iter_xarray(&iter, READ, &subreq->rreq->mapping->i_pages,
155 subreq->start + subreq->transferred,
156 subreq->len - subreq->transferred);
157 iov_iter_zero(iov_iter_count(&iter), &iter);
160 static void netfs_cache_read_terminated(void *priv, ssize_t transferred_or_error,
163 struct netfs_read_subrequest *subreq = priv;
165 netfs_subreq_terminated(subreq, transferred_or_error, was_async);
169 * Issue a read against the cache.
170 * - Eats the caller's ref on subreq.
172 static void netfs_read_from_cache(struct netfs_read_request *rreq,
173 struct netfs_read_subrequest *subreq,
174 enum netfs_read_from_hole read_hole)
176 struct netfs_cache_resources *cres = &rreq->cache_resources;
177 struct iov_iter iter;
179 netfs_stat(&netfs_n_rh_read);
180 iov_iter_xarray(&iter, READ, &rreq->mapping->i_pages,
181 subreq->start + subreq->transferred,
182 subreq->len - subreq->transferred);
184 cres->ops->read(cres, subreq->start, &iter, read_hole,
185 netfs_cache_read_terminated, subreq);
189 * Fill a subrequest region with zeroes.
191 static void netfs_fill_with_zeroes(struct netfs_read_request *rreq,
192 struct netfs_read_subrequest *subreq)
194 netfs_stat(&netfs_n_rh_zero);
195 __set_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags);
196 netfs_subreq_terminated(subreq, 0, false);
200 * Ask the netfs to issue a read request to the server for us.
202 * The netfs is expected to read from subreq->pos + subreq->transferred to
203 * subreq->pos + subreq->len - 1. It may not backtrack and write data into the
204 * buffer prior to the transferred point as it might clobber dirty data
205 * obtained from the cache.
207 * Alternatively, the netfs is allowed to indicate one of two things:
209 * - NETFS_SREQ_SHORT_READ: A short read - it will get called again to try and
212 * - NETFS_SREQ_CLEAR_TAIL: A short read - the rest of the buffer will be
215 static void netfs_read_from_server(struct netfs_read_request *rreq,
216 struct netfs_read_subrequest *subreq)
218 netfs_stat(&netfs_n_rh_download);
219 rreq->netfs_ops->issue_op(subreq);
223 * Release those waiting.
225 static void netfs_rreq_completed(struct netfs_read_request *rreq, bool was_async)
227 trace_netfs_rreq(rreq, netfs_rreq_trace_done);
228 netfs_rreq_clear_subreqs(rreq, was_async);
229 netfs_put_read_request(rreq, was_async);
233 * Deal with the completion of writing the data to the cache. We have to clear
234 * the PG_fscache bits on the folios involved and release the caller's ref.
236 * May be called in softirq mode and we inherit a ref from the caller.
238 static void netfs_rreq_unmark_after_write(struct netfs_read_request *rreq,
241 struct netfs_read_subrequest *subreq;
243 pgoff_t unlocked = 0;
244 bool have_unlocked = false;
248 list_for_each_entry(subreq, &rreq->subrequests, rreq_link) {
249 XA_STATE(xas, &rreq->mapping->i_pages, subreq->start / PAGE_SIZE);
251 xas_for_each(&xas, folio, (subreq->start + subreq->len - 1) / PAGE_SIZE) {
252 /* We might have multiple writes from the same huge
253 * folio, but we mustn't unlock a folio more than once.
255 if (have_unlocked && folio_index(folio) <= unlocked)
257 unlocked = folio_index(folio);
258 folio_end_fscache(folio);
259 have_unlocked = true;
264 netfs_rreq_completed(rreq, was_async);
267 static void netfs_rreq_copy_terminated(void *priv, ssize_t transferred_or_error,
270 struct netfs_read_subrequest *subreq = priv;
271 struct netfs_read_request *rreq = subreq->rreq;
273 if (IS_ERR_VALUE(transferred_or_error)) {
274 netfs_stat(&netfs_n_rh_write_failed);
275 trace_netfs_failure(rreq, subreq, transferred_or_error,
276 netfs_fail_copy_to_cache);
278 netfs_stat(&netfs_n_rh_write_done);
281 trace_netfs_sreq(subreq, netfs_sreq_trace_write_term);
283 /* If we decrement nr_wr_ops to 0, the ref belongs to us. */
284 if (atomic_dec_and_test(&rreq->nr_wr_ops))
285 netfs_rreq_unmark_after_write(rreq, was_async);
287 netfs_put_subrequest(subreq, was_async);
291 * Perform any outstanding writes to the cache. We inherit a ref from the
294 static void netfs_rreq_do_write_to_cache(struct netfs_read_request *rreq)
296 struct netfs_cache_resources *cres = &rreq->cache_resources;
297 struct netfs_read_subrequest *subreq, *next, *p;
298 struct iov_iter iter;
301 trace_netfs_rreq(rreq, netfs_rreq_trace_write);
303 /* We don't want terminating writes trying to wake us up whilst we're
304 * still going through the list.
306 atomic_inc(&rreq->nr_wr_ops);
308 list_for_each_entry_safe(subreq, p, &rreq->subrequests, rreq_link) {
309 if (!test_bit(NETFS_SREQ_WRITE_TO_CACHE, &subreq->flags)) {
310 list_del_init(&subreq->rreq_link);
311 netfs_put_subrequest(subreq, false);
315 list_for_each_entry(subreq, &rreq->subrequests, rreq_link) {
316 /* Amalgamate adjacent writes */
317 while (!list_is_last(&subreq->rreq_link, &rreq->subrequests)) {
318 next = list_next_entry(subreq, rreq_link);
319 if (next->start != subreq->start + subreq->len)
321 subreq->len += next->len;
322 list_del_init(&next->rreq_link);
323 netfs_put_subrequest(next, false);
326 ret = cres->ops->prepare_write(cres, &subreq->start, &subreq->len,
329 trace_netfs_failure(rreq, subreq, ret, netfs_fail_prepare_write);
330 trace_netfs_sreq(subreq, netfs_sreq_trace_write_skip);
334 iov_iter_xarray(&iter, WRITE, &rreq->mapping->i_pages,
335 subreq->start, subreq->len);
337 atomic_inc(&rreq->nr_wr_ops);
338 netfs_stat(&netfs_n_rh_write);
339 netfs_get_read_subrequest(subreq);
340 trace_netfs_sreq(subreq, netfs_sreq_trace_write);
341 cres->ops->write(cres, subreq->start, &iter,
342 netfs_rreq_copy_terminated, subreq);
345 /* If we decrement nr_wr_ops to 0, the usage ref belongs to us. */
346 if (atomic_dec_and_test(&rreq->nr_wr_ops))
347 netfs_rreq_unmark_after_write(rreq, false);
350 static void netfs_rreq_write_to_cache_work(struct work_struct *work)
352 struct netfs_read_request *rreq =
353 container_of(work, struct netfs_read_request, work);
355 netfs_rreq_do_write_to_cache(rreq);
358 static void netfs_rreq_write_to_cache(struct netfs_read_request *rreq)
360 rreq->work.func = netfs_rreq_write_to_cache_work;
361 if (!queue_work(system_unbound_wq, &rreq->work))
366 * Unlock the folios in a read operation. We need to set PG_fscache on any
367 * folios we're going to write back before we unlock them.
369 static void netfs_rreq_unlock(struct netfs_read_request *rreq)
371 struct netfs_read_subrequest *subreq;
373 unsigned int iopos, account = 0;
374 pgoff_t start_page = rreq->start / PAGE_SIZE;
375 pgoff_t last_page = ((rreq->start + rreq->len) / PAGE_SIZE) - 1;
376 bool subreq_failed = false;
378 XA_STATE(xas, &rreq->mapping->i_pages, start_page);
380 if (test_bit(NETFS_RREQ_FAILED, &rreq->flags)) {
381 __clear_bit(NETFS_RREQ_WRITE_TO_CACHE, &rreq->flags);
382 list_for_each_entry(subreq, &rreq->subrequests, rreq_link) {
383 __clear_bit(NETFS_SREQ_WRITE_TO_CACHE, &subreq->flags);
387 /* Walk through the pagecache and the I/O request lists simultaneously.
388 * We may have a mixture of cached and uncached sections and we only
389 * really want to write out the uncached sections. This is slightly
390 * complicated by the possibility that we might have huge pages with a
393 subreq = list_first_entry(&rreq->subrequests,
394 struct netfs_read_subrequest, rreq_link);
396 subreq_failed = (subreq->error < 0);
398 trace_netfs_rreq(rreq, netfs_rreq_trace_unlock);
401 xas_for_each(&xas, folio, last_page) {
402 unsigned int pgpos = (folio_index(folio) - start_page) * PAGE_SIZE;
403 unsigned int pgend = pgpos + folio_size(folio);
404 bool pg_failed = false;
411 if (test_bit(NETFS_SREQ_WRITE_TO_CACHE, &subreq->flags))
412 folio_start_fscache(folio);
413 pg_failed |= subreq_failed;
414 if (pgend < iopos + subreq->len)
417 account += subreq->transferred;
418 iopos += subreq->len;
419 if (!list_is_last(&subreq->rreq_link, &rreq->subrequests)) {
420 subreq = list_next_entry(subreq, rreq_link);
421 subreq_failed = (subreq->error < 0);
424 subreq_failed = false;
431 flush_dcache_folio(folio);
432 folio_mark_uptodate(folio);
435 if (!test_bit(NETFS_RREQ_DONT_UNLOCK_FOLIOS, &rreq->flags)) {
436 if (folio_index(folio) == rreq->no_unlock_folio &&
437 test_bit(NETFS_RREQ_NO_UNLOCK_FOLIO, &rreq->flags))
445 task_io_account_read(account);
446 if (rreq->netfs_ops->done)
447 rreq->netfs_ops->done(rreq);
451 * Handle a short read.
453 static void netfs_rreq_short_read(struct netfs_read_request *rreq,
454 struct netfs_read_subrequest *subreq)
456 __clear_bit(NETFS_SREQ_SHORT_READ, &subreq->flags);
457 __set_bit(NETFS_SREQ_SEEK_DATA_READ, &subreq->flags);
459 netfs_stat(&netfs_n_rh_short_read);
460 trace_netfs_sreq(subreq, netfs_sreq_trace_resubmit_short);
462 netfs_get_read_subrequest(subreq);
463 atomic_inc(&rreq->nr_rd_ops);
464 if (subreq->source == NETFS_READ_FROM_CACHE)
465 netfs_read_from_cache(rreq, subreq, NETFS_READ_HOLE_CLEAR);
467 netfs_read_from_server(rreq, subreq);
471 * Resubmit any short or failed operations. Returns true if we got the rreq
474 static bool netfs_rreq_perform_resubmissions(struct netfs_read_request *rreq)
476 struct netfs_read_subrequest *subreq;
478 WARN_ON(in_interrupt());
480 trace_netfs_rreq(rreq, netfs_rreq_trace_resubmit);
482 /* We don't want terminating submissions trying to wake us up whilst
483 * we're still going through the list.
485 atomic_inc(&rreq->nr_rd_ops);
487 __clear_bit(NETFS_RREQ_INCOMPLETE_IO, &rreq->flags);
488 list_for_each_entry(subreq, &rreq->subrequests, rreq_link) {
490 if (subreq->source != NETFS_READ_FROM_CACHE)
492 subreq->source = NETFS_DOWNLOAD_FROM_SERVER;
494 netfs_stat(&netfs_n_rh_download_instead);
495 trace_netfs_sreq(subreq, netfs_sreq_trace_download_instead);
496 netfs_get_read_subrequest(subreq);
497 atomic_inc(&rreq->nr_rd_ops);
498 netfs_read_from_server(rreq, subreq);
499 } else if (test_bit(NETFS_SREQ_SHORT_READ, &subreq->flags)) {
500 netfs_rreq_short_read(rreq, subreq);
504 /* If we decrement nr_rd_ops to 0, the usage ref belongs to us. */
505 if (atomic_dec_and_test(&rreq->nr_rd_ops))
508 wake_up_var(&rreq->nr_rd_ops);
513 * Check to see if the data read is still valid.
515 static void netfs_rreq_is_still_valid(struct netfs_read_request *rreq)
517 struct netfs_read_subrequest *subreq;
519 if (!rreq->netfs_ops->is_still_valid ||
520 rreq->netfs_ops->is_still_valid(rreq))
523 list_for_each_entry(subreq, &rreq->subrequests, rreq_link) {
524 if (subreq->source == NETFS_READ_FROM_CACHE) {
525 subreq->error = -ESTALE;
526 __set_bit(NETFS_RREQ_INCOMPLETE_IO, &rreq->flags);
532 * Assess the state of a read request and decide what to do next.
534 * Note that we could be in an ordinary kernel thread, on a workqueue or in
535 * softirq context at this point. We inherit a ref from the caller.
537 static void netfs_rreq_assess(struct netfs_read_request *rreq, bool was_async)
539 trace_netfs_rreq(rreq, netfs_rreq_trace_assess);
542 netfs_rreq_is_still_valid(rreq);
544 if (!test_bit(NETFS_RREQ_FAILED, &rreq->flags) &&
545 test_bit(NETFS_RREQ_INCOMPLETE_IO, &rreq->flags)) {
546 if (netfs_rreq_perform_resubmissions(rreq))
551 netfs_rreq_unlock(rreq);
553 clear_bit_unlock(NETFS_RREQ_IN_PROGRESS, &rreq->flags);
554 wake_up_bit(&rreq->flags, NETFS_RREQ_IN_PROGRESS);
556 if (test_bit(NETFS_RREQ_WRITE_TO_CACHE, &rreq->flags))
557 return netfs_rreq_write_to_cache(rreq);
559 netfs_rreq_completed(rreq, was_async);
562 static void netfs_rreq_work(struct work_struct *work)
564 struct netfs_read_request *rreq =
565 container_of(work, struct netfs_read_request, work);
566 netfs_rreq_assess(rreq, false);
570 * Handle the completion of all outstanding I/O operations on a read request.
571 * We inherit a ref from the caller.
573 static void netfs_rreq_terminated(struct netfs_read_request *rreq,
576 if (test_bit(NETFS_RREQ_INCOMPLETE_IO, &rreq->flags) &&
578 if (!queue_work(system_unbound_wq, &rreq->work))
581 netfs_rreq_assess(rreq, was_async);
586 * netfs_subreq_terminated - Note the termination of an I/O operation.
587 * @subreq: The I/O request that has terminated.
588 * @transferred_or_error: The amount of data transferred or an error code.
589 * @was_async: The termination was asynchronous
591 * This tells the read helper that a contributory I/O operation has terminated,
592 * one way or another, and that it should integrate the results.
594 * The caller indicates in @transferred_or_error the outcome of the operation,
595 * supplying a positive value to indicate the number of bytes transferred, 0 to
596 * indicate a failure to transfer anything that should be retried or a negative
597 * error code. The helper will look after reissuing I/O operations as
598 * appropriate and writing downloaded data to the cache.
600 * If @was_async is true, the caller might be running in softirq or interrupt
601 * context and we can't sleep.
603 void netfs_subreq_terminated(struct netfs_read_subrequest *subreq,
604 ssize_t transferred_or_error,
607 struct netfs_read_request *rreq = subreq->rreq;
610 _enter("[%u]{%llx,%lx},%zd",
611 subreq->debug_index, subreq->start, subreq->flags,
612 transferred_or_error);
614 switch (subreq->source) {
615 case NETFS_READ_FROM_CACHE:
616 netfs_stat(&netfs_n_rh_read_done);
618 case NETFS_DOWNLOAD_FROM_SERVER:
619 netfs_stat(&netfs_n_rh_download_done);
625 if (IS_ERR_VALUE(transferred_or_error)) {
626 subreq->error = transferred_or_error;
627 trace_netfs_failure(rreq, subreq, transferred_or_error,
632 if (WARN(transferred_or_error > subreq->len - subreq->transferred,
633 "Subreq overread: R%x[%x] %zd > %zu - %zu",
634 rreq->debug_id, subreq->debug_index,
635 transferred_or_error, subreq->len, subreq->transferred))
636 transferred_or_error = subreq->len - subreq->transferred;
639 subreq->transferred += transferred_or_error;
640 if (subreq->transferred < subreq->len)
644 __clear_bit(NETFS_SREQ_NO_PROGRESS, &subreq->flags);
645 if (test_bit(NETFS_SREQ_WRITE_TO_CACHE, &subreq->flags))
646 set_bit(NETFS_RREQ_WRITE_TO_CACHE, &rreq->flags);
649 trace_netfs_sreq(subreq, netfs_sreq_trace_terminated);
651 /* If we decrement nr_rd_ops to 0, the ref belongs to us. */
652 u = atomic_dec_return(&rreq->nr_rd_ops);
654 netfs_rreq_terminated(rreq, was_async);
656 wake_up_var(&rreq->nr_rd_ops);
658 netfs_put_subrequest(subreq, was_async);
662 if (test_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags)) {
663 netfs_clear_unread(subreq);
664 subreq->transferred = subreq->len;
668 if (transferred_or_error == 0) {
669 if (__test_and_set_bit(NETFS_SREQ_NO_PROGRESS, &subreq->flags)) {
670 subreq->error = -ENODATA;
674 __clear_bit(NETFS_SREQ_NO_PROGRESS, &subreq->flags);
677 __set_bit(NETFS_SREQ_SHORT_READ, &subreq->flags);
678 set_bit(NETFS_RREQ_INCOMPLETE_IO, &rreq->flags);
682 if (subreq->source == NETFS_READ_FROM_CACHE) {
683 netfs_stat(&netfs_n_rh_read_failed);
684 set_bit(NETFS_RREQ_INCOMPLETE_IO, &rreq->flags);
686 netfs_stat(&netfs_n_rh_download_failed);
687 set_bit(NETFS_RREQ_FAILED, &rreq->flags);
688 rreq->error = subreq->error;
692 EXPORT_SYMBOL(netfs_subreq_terminated);
694 static enum netfs_read_source netfs_cache_prepare_read(struct netfs_read_subrequest *subreq,
697 struct netfs_read_request *rreq = subreq->rreq;
698 struct netfs_cache_resources *cres = &rreq->cache_resources;
701 return cres->ops->prepare_read(subreq, i_size);
702 if (subreq->start >= rreq->i_size)
703 return NETFS_FILL_WITH_ZEROES;
704 return NETFS_DOWNLOAD_FROM_SERVER;
708 * Work out what sort of subrequest the next one will be.
710 static enum netfs_read_source
711 netfs_rreq_prepare_read(struct netfs_read_request *rreq,
712 struct netfs_read_subrequest *subreq)
714 enum netfs_read_source source;
716 _enter("%llx-%llx,%llx", subreq->start, subreq->start + subreq->len, rreq->i_size);
718 source = netfs_cache_prepare_read(subreq, rreq->i_size);
719 if (source == NETFS_INVALID_READ)
722 if (source == NETFS_DOWNLOAD_FROM_SERVER) {
723 /* Call out to the netfs to let it shrink the request to fit
724 * its own I/O sizes and boundaries. If it shinks it here, it
725 * will be called again to make simultaneous calls; if it wants
726 * to make serial calls, it can indicate a short read and then
727 * we will call it again.
729 if (subreq->len > rreq->i_size - subreq->start)
730 subreq->len = rreq->i_size - subreq->start;
732 if (rreq->netfs_ops->clamp_length &&
733 !rreq->netfs_ops->clamp_length(subreq)) {
734 source = NETFS_INVALID_READ;
739 if (WARN_ON(subreq->len == 0))
740 source = NETFS_INVALID_READ;
743 subreq->source = source;
744 trace_netfs_sreq(subreq, netfs_sreq_trace_prepare);
749 * Slice off a piece of a read request and submit an I/O request for it.
751 static bool netfs_rreq_submit_slice(struct netfs_read_request *rreq,
752 unsigned int *_debug_index)
754 struct netfs_read_subrequest *subreq;
755 enum netfs_read_source source;
757 subreq = netfs_alloc_subrequest(rreq);
761 subreq->debug_index = (*_debug_index)++;
762 subreq->start = rreq->start + rreq->submitted;
763 subreq->len = rreq->len - rreq->submitted;
765 _debug("slice %llx,%zx,%zx", subreq->start, subreq->len, rreq->submitted);
766 list_add_tail(&subreq->rreq_link, &rreq->subrequests);
768 /* Call out to the cache to find out what it can do with the remaining
769 * subset. It tells us in subreq->flags what it decided should be done
770 * and adjusts subreq->len down if the subset crosses a cache boundary.
772 * Then when we hand the subset, it can choose to take a subset of that
773 * (the starts must coincide), in which case, we go around the loop
774 * again and ask it to download the next piece.
776 source = netfs_rreq_prepare_read(rreq, subreq);
777 if (source == NETFS_INVALID_READ)
780 atomic_inc(&rreq->nr_rd_ops);
782 rreq->submitted += subreq->len;
784 trace_netfs_sreq(subreq, netfs_sreq_trace_submit);
786 case NETFS_FILL_WITH_ZEROES:
787 netfs_fill_with_zeroes(rreq, subreq);
789 case NETFS_DOWNLOAD_FROM_SERVER:
790 netfs_read_from_server(rreq, subreq);
792 case NETFS_READ_FROM_CACHE:
793 netfs_read_from_cache(rreq, subreq, NETFS_READ_HOLE_IGNORE);
802 rreq->error = subreq->error;
803 netfs_put_subrequest(subreq, false);
807 static void netfs_cache_expand_readahead(struct netfs_read_request *rreq,
808 loff_t *_start, size_t *_len, loff_t i_size)
810 struct netfs_cache_resources *cres = &rreq->cache_resources;
812 if (cres->ops && cres->ops->expand_readahead)
813 cres->ops->expand_readahead(cres, _start, _len, i_size);
816 static void netfs_rreq_expand(struct netfs_read_request *rreq,
817 struct readahead_control *ractl)
819 /* Give the cache a chance to change the request parameters. The
820 * resultant request must contain the original region.
822 netfs_cache_expand_readahead(rreq, &rreq->start, &rreq->len, rreq->i_size);
824 /* Give the netfs a chance to change the request parameters. The
825 * resultant request must contain the original region.
827 if (rreq->netfs_ops->expand_readahead)
828 rreq->netfs_ops->expand_readahead(rreq);
830 /* Expand the request if the cache wants it to start earlier. Note
831 * that the expansion may get further extended if the VM wishes to
832 * insert THPs and the preferred start and/or end wind up in the middle
835 * If this is the case, however, the THP size should be an integer
836 * multiple of the cache granule size, so we get a whole number of
837 * granules to deal with.
839 if (rreq->start != readahead_pos(ractl) ||
840 rreq->len != readahead_length(ractl)) {
841 readahead_expand(ractl, rreq->start, rreq->len);
842 rreq->start = readahead_pos(ractl);
843 rreq->len = readahead_length(ractl);
845 trace_netfs_read(rreq, readahead_pos(ractl), readahead_length(ractl),
846 netfs_read_trace_expanded);
851 * netfs_readahead - Helper to manage a read request
852 * @ractl: The description of the readahead request
853 * @ops: The network filesystem's operations for the helper to use
854 * @netfs_priv: Private netfs data to be retained in the request
856 * Fulfil a readahead request by drawing data from the cache if possible, or
857 * the netfs if not. Space beyond the EOF is zero-filled. Multiple I/O
858 * requests from different sources will get munged together. If necessary, the
859 * readahead window can be expanded in either direction to a more convenient
860 * alighment for RPC efficiency or to make storage in the cache feasible.
862 * The calling netfs must provide a table of operations, only one of which,
863 * issue_op, is mandatory. It may also be passed a private token, which will
864 * be retained in rreq->netfs_priv and will be cleaned up by ops->cleanup().
866 * This is usable whether or not caching is enabled.
868 void netfs_readahead(struct readahead_control *ractl,
869 const struct netfs_read_request_ops *ops,
872 struct netfs_read_request *rreq;
873 unsigned int debug_index = 0;
876 _enter("%lx,%x", readahead_index(ractl), readahead_count(ractl));
878 if (readahead_count(ractl) == 0)
881 rreq = netfs_alloc_read_request(ops, netfs_priv, ractl->file);
884 rreq->mapping = ractl->mapping;
885 rreq->start = readahead_pos(ractl);
886 rreq->len = readahead_length(ractl);
888 if (ops->begin_cache_operation) {
889 ret = ops->begin_cache_operation(rreq);
890 if (ret == -ENOMEM || ret == -EINTR || ret == -ERESTARTSYS)
894 netfs_stat(&netfs_n_rh_readahead);
895 trace_netfs_read(rreq, readahead_pos(ractl), readahead_length(ractl),
896 netfs_read_trace_readahead);
898 netfs_rreq_expand(rreq, ractl);
900 atomic_set(&rreq->nr_rd_ops, 1);
902 if (!netfs_rreq_submit_slice(rreq, &debug_index))
905 } while (rreq->submitted < rreq->len);
907 /* Drop the refs on the folios here rather than in the cache or
908 * filesystem. The locks will be dropped in netfs_rreq_unlock().
910 while (readahead_folio(ractl))
913 /* If we decrement nr_rd_ops to 0, the ref belongs to us. */
914 if (atomic_dec_and_test(&rreq->nr_rd_ops))
915 netfs_rreq_assess(rreq, false);
919 netfs_put_read_request(rreq, false);
923 ops->cleanup(ractl->mapping, netfs_priv);
926 EXPORT_SYMBOL(netfs_readahead);
929 * netfs_readpage - Helper to manage a readpage request
930 * @file: The file to read from
931 * @folio: The folio to read
932 * @ops: The network filesystem's operations for the helper to use
933 * @netfs_priv: Private netfs data to be retained in the request
935 * Fulfil a readpage request by drawing data from the cache if possible, or the
936 * netfs if not. Space beyond the EOF is zero-filled. Multiple I/O requests
937 * from different sources will get munged together.
939 * The calling netfs must provide a table of operations, only one of which,
940 * issue_op, is mandatory. It may also be passed a private token, which will
941 * be retained in rreq->netfs_priv and will be cleaned up by ops->cleanup().
943 * This is usable whether or not caching is enabled.
945 int netfs_readpage(struct file *file,
947 const struct netfs_read_request_ops *ops,
950 struct netfs_read_request *rreq;
951 unsigned int debug_index = 0;
954 _enter("%lx", folio_index(folio));
956 rreq = netfs_alloc_read_request(ops, netfs_priv, file);
959 ops->cleanup(folio_file_mapping(folio), netfs_priv);
963 rreq->mapping = folio_file_mapping(folio);
964 rreq->start = folio_file_pos(folio);
965 rreq->len = folio_size(folio);
967 if (ops->begin_cache_operation) {
968 ret = ops->begin_cache_operation(rreq);
969 if (ret == -ENOMEM || ret == -EINTR || ret == -ERESTARTSYS) {
975 netfs_stat(&netfs_n_rh_readpage);
976 trace_netfs_read(rreq, rreq->start, rreq->len, netfs_read_trace_readpage);
978 netfs_get_read_request(rreq);
980 atomic_set(&rreq->nr_rd_ops, 1);
982 if (!netfs_rreq_submit_slice(rreq, &debug_index))
985 } while (rreq->submitted < rreq->len);
987 /* Keep nr_rd_ops incremented so that the ref always belongs to us, and
988 * the service code isn't punted off to a random thread pool to
992 wait_var_event(&rreq->nr_rd_ops, atomic_read(&rreq->nr_rd_ops) == 1);
993 netfs_rreq_assess(rreq, false);
994 } while (test_bit(NETFS_RREQ_IN_PROGRESS, &rreq->flags));
997 if (ret == 0 && rreq->submitted < rreq->len) {
998 trace_netfs_failure(rreq, NULL, ret, netfs_fail_short_readpage);
1002 netfs_put_read_request(rreq, false);
1005 EXPORT_SYMBOL(netfs_readpage);
1008 * Prepare a folio for writing without reading first
1009 * @folio: The folio being prepared
1010 * @pos: starting position for the write
1011 * @len: length of write
1013 * In some cases, write_begin doesn't need to read at all:
1014 * - full folio write
1015 * - write that lies in a folio that is completely beyond EOF
1016 * - write that covers the folio from start to EOF or beyond it
1018 * If any of these criteria are met, then zero out the unwritten parts
1019 * of the folio and return true. Otherwise, return false.
1021 static bool netfs_skip_folio_read(struct folio *folio, loff_t pos, size_t len)
1023 struct inode *inode = folio_inode(folio);
1024 loff_t i_size = i_size_read(inode);
1025 size_t offset = offset_in_folio(folio, pos);
1027 /* Full folio write */
1028 if (offset == 0 && len >= folio_size(folio))
1031 /* pos beyond last folio in the file */
1032 if (pos - offset >= i_size)
1035 /* Write that covers from the start of the folio to EOF or beyond */
1036 if (offset == 0 && (pos + len) >= i_size)
1041 zero_user_segments(&folio->page, 0, offset, offset + len, folio_size(folio));
1046 * netfs_write_begin - Helper to prepare for writing
1047 * @file: The file to read from
1048 * @mapping: The mapping to read from
1049 * @pos: File position at which the write will begin
1050 * @len: The length of the write (may extend beyond the end of the folio chosen)
1051 * @aop_flags: AOP_* flags
1052 * @_folio: Where to put the resultant folio
1053 * @_fsdata: Place for the netfs to store a cookie
1054 * @ops: The network filesystem's operations for the helper to use
1055 * @netfs_priv: Private netfs data to be retained in the request
1057 * Pre-read data for a write-begin request by drawing data from the cache if
1058 * possible, or the netfs if not. Space beyond the EOF is zero-filled.
1059 * Multiple I/O requests from different sources will get munged together. If
1060 * necessary, the readahead window can be expanded in either direction to a
1061 * more convenient alighment for RPC efficiency or to make storage in the cache
1064 * The calling netfs must provide a table of operations, only one of which,
1065 * issue_op, is mandatory.
1067 * The check_write_begin() operation can be provided to check for and flush
1068 * conflicting writes once the folio is grabbed and locked. It is passed a
1069 * pointer to the fsdata cookie that gets returned to the VM to be passed to
1070 * write_end. It is permitted to sleep. It should return 0 if the request
1071 * should go ahead; unlock the folio and return -EAGAIN to cause the folio to
1072 * be regot; or return an error.
1074 * This is usable whether or not caching is enabled.
1076 int netfs_write_begin(struct file *file, struct address_space *mapping,
1077 loff_t pos, unsigned int len, unsigned int aop_flags,
1078 struct folio **_folio, void **_fsdata,
1079 const struct netfs_read_request_ops *ops,
1082 struct netfs_read_request *rreq;
1083 struct folio *folio;
1084 struct inode *inode = file_inode(file);
1085 unsigned int debug_index = 0, fgp_flags;
1086 pgoff_t index = pos >> PAGE_SHIFT;
1089 DEFINE_READAHEAD(ractl, file, NULL, mapping, index);
1092 fgp_flags = FGP_LOCK | FGP_WRITE | FGP_CREAT | FGP_STABLE;
1093 if (aop_flags & AOP_FLAG_NOFS)
1094 fgp_flags |= FGP_NOFS;
1095 folio = __filemap_get_folio(mapping, index, fgp_flags,
1096 mapping_gfp_mask(mapping));
1100 if (ops->check_write_begin) {
1101 /* Allow the netfs (eg. ceph) to flush conflicts. */
1102 ret = ops->check_write_begin(file, pos, len, folio, _fsdata);
1104 trace_netfs_failure(NULL, NULL, ret, netfs_fail_check_write_begin);
1111 if (folio_test_uptodate(folio))
1114 /* If the page is beyond the EOF, we want to clear it - unless it's
1115 * within the cache granule containing the EOF, in which case we need
1116 * to preload the granule.
1118 if (!ops->is_cache_enabled(inode) &&
1119 netfs_skip_folio_read(folio, pos, len)) {
1120 netfs_stat(&netfs_n_rh_write_zskip);
1121 goto have_folio_no_wait;
1125 rreq = netfs_alloc_read_request(ops, netfs_priv, file);
1128 rreq->mapping = folio_file_mapping(folio);
1129 rreq->start = folio_file_pos(folio);
1130 rreq->len = folio_size(folio);
1131 rreq->no_unlock_folio = folio_index(folio);
1132 __set_bit(NETFS_RREQ_NO_UNLOCK_FOLIO, &rreq->flags);
1135 if (ops->begin_cache_operation) {
1136 ret = ops->begin_cache_operation(rreq);
1137 if (ret == -ENOMEM || ret == -EINTR || ret == -ERESTARTSYS)
1141 netfs_stat(&netfs_n_rh_write_begin);
1142 trace_netfs_read(rreq, pos, len, netfs_read_trace_write_begin);
1144 /* Expand the request to meet caching requirements and download
1147 ractl._nr_pages = folio_nr_pages(folio);
1148 netfs_rreq_expand(rreq, &ractl);
1149 netfs_get_read_request(rreq);
1151 /* We hold the folio locks, so we can drop the references */
1153 while (readahead_folio(&ractl))
1156 atomic_set(&rreq->nr_rd_ops, 1);
1158 if (!netfs_rreq_submit_slice(rreq, &debug_index))
1161 } while (rreq->submitted < rreq->len);
1163 /* Keep nr_rd_ops incremented so that the ref always belongs to us, and
1164 * the service code isn't punted off to a random thread pool to
1168 wait_var_event(&rreq->nr_rd_ops, atomic_read(&rreq->nr_rd_ops) == 1);
1169 netfs_rreq_assess(rreq, false);
1170 if (!test_bit(NETFS_RREQ_IN_PROGRESS, &rreq->flags))
1176 if (ret == 0 && rreq->submitted < rreq->len) {
1177 trace_netfs_failure(rreq, NULL, ret, netfs_fail_short_write_begin);
1180 netfs_put_read_request(rreq, false);
1185 ret = folio_wait_fscache_killable(folio);
1190 ops->cleanup(mapping, netfs_priv);
1196 netfs_put_read_request(rreq, false);
1198 folio_unlock(folio);
1201 ops->cleanup(mapping, netfs_priv);
1202 _leave(" = %d", ret);
1205 EXPORT_SYMBOL(netfs_write_begin);