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);
58 ops->init_rreq(rreq, file);
59 netfs_stat(&netfs_n_rh_rreq);
65 static void netfs_get_read_request(struct netfs_read_request *rreq)
67 refcount_inc(&rreq->usage);
70 static void netfs_rreq_clear_subreqs(struct netfs_read_request *rreq,
73 struct netfs_read_subrequest *subreq;
75 while (!list_empty(&rreq->subrequests)) {
76 subreq = list_first_entry(&rreq->subrequests,
77 struct netfs_read_subrequest, rreq_link);
78 list_del(&subreq->rreq_link);
79 netfs_put_subrequest(subreq, was_async);
83 static void netfs_free_read_request(struct work_struct *work)
85 struct netfs_read_request *rreq =
86 container_of(work, struct netfs_read_request, work);
87 netfs_rreq_clear_subreqs(rreq, false);
89 rreq->netfs_ops->cleanup(rreq->mapping, rreq->netfs_priv);
90 trace_netfs_rreq(rreq, netfs_rreq_trace_free);
91 if (rreq->cache_resources.ops)
92 rreq->cache_resources.ops->end_operation(&rreq->cache_resources);
94 netfs_stat_d(&netfs_n_rh_rreq);
97 static void netfs_put_read_request(struct netfs_read_request *rreq, bool was_async)
99 if (refcount_dec_and_test(&rreq->usage)) {
101 rreq->work.func = netfs_free_read_request;
102 if (!queue_work(system_unbound_wq, &rreq->work))
105 netfs_free_read_request(&rreq->work);
111 * Allocate and partially initialise an I/O request structure.
113 static struct netfs_read_subrequest *netfs_alloc_subrequest(
114 struct netfs_read_request *rreq)
116 struct netfs_read_subrequest *subreq;
118 subreq = kzalloc(sizeof(struct netfs_read_subrequest), GFP_KERNEL);
120 INIT_LIST_HEAD(&subreq->rreq_link);
121 refcount_set(&subreq->usage, 2);
123 netfs_get_read_request(rreq);
124 netfs_stat(&netfs_n_rh_sreq);
130 static void netfs_get_read_subrequest(struct netfs_read_subrequest *subreq)
132 refcount_inc(&subreq->usage);
135 static void __netfs_put_subrequest(struct netfs_read_subrequest *subreq,
138 struct netfs_read_request *rreq = subreq->rreq;
140 trace_netfs_sreq(subreq, netfs_sreq_trace_free);
142 netfs_stat_d(&netfs_n_rh_sreq);
143 netfs_put_read_request(rreq, was_async);
147 * Clear the unread part of an I/O request.
149 static void netfs_clear_unread(struct netfs_read_subrequest *subreq)
151 struct iov_iter iter;
153 iov_iter_xarray(&iter, READ, &subreq->rreq->mapping->i_pages,
154 subreq->start + subreq->transferred,
155 subreq->len - subreq->transferred);
156 iov_iter_zero(iov_iter_count(&iter), &iter);
159 static void netfs_cache_read_terminated(void *priv, ssize_t transferred_or_error,
162 struct netfs_read_subrequest *subreq = priv;
164 netfs_subreq_terminated(subreq, transferred_or_error, was_async);
168 * Issue a read against the cache.
169 * - Eats the caller's ref on subreq.
171 static void netfs_read_from_cache(struct netfs_read_request *rreq,
172 struct netfs_read_subrequest *subreq,
175 struct netfs_cache_resources *cres = &rreq->cache_resources;
176 struct iov_iter iter;
178 netfs_stat(&netfs_n_rh_read);
179 iov_iter_xarray(&iter, READ, &rreq->mapping->i_pages,
180 subreq->start + subreq->transferred,
181 subreq->len - subreq->transferred);
183 cres->ops->read(cres, subreq->start, &iter, seek_data,
184 netfs_cache_read_terminated, subreq);
188 * Fill a subrequest region with zeroes.
190 static void netfs_fill_with_zeroes(struct netfs_read_request *rreq,
191 struct netfs_read_subrequest *subreq)
193 netfs_stat(&netfs_n_rh_zero);
194 __set_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags);
195 netfs_subreq_terminated(subreq, 0, false);
199 * Ask the netfs to issue a read request to the server for us.
201 * The netfs is expected to read from subreq->pos + subreq->transferred to
202 * subreq->pos + subreq->len - 1. It may not backtrack and write data into the
203 * buffer prior to the transferred point as it might clobber dirty data
204 * obtained from the cache.
206 * Alternatively, the netfs is allowed to indicate one of two things:
208 * - NETFS_SREQ_SHORT_READ: A short read - it will get called again to try and
211 * - NETFS_SREQ_CLEAR_TAIL: A short read - the rest of the buffer will be
214 static void netfs_read_from_server(struct netfs_read_request *rreq,
215 struct netfs_read_subrequest *subreq)
217 netfs_stat(&netfs_n_rh_download);
218 rreq->netfs_ops->issue_op(subreq);
222 * Release those waiting.
224 static void netfs_rreq_completed(struct netfs_read_request *rreq, bool was_async)
226 trace_netfs_rreq(rreq, netfs_rreq_trace_done);
227 netfs_rreq_clear_subreqs(rreq, was_async);
228 netfs_put_read_request(rreq, was_async);
232 * Deal with the completion of writing the data to the cache. We have to clear
233 * the PG_fscache bits on the folios involved and release the caller's ref.
235 * May be called in softirq mode and we inherit a ref from the caller.
237 static void netfs_rreq_unmark_after_write(struct netfs_read_request *rreq,
240 struct netfs_read_subrequest *subreq;
242 pgoff_t unlocked = 0;
243 bool have_unlocked = false;
247 list_for_each_entry(subreq, &rreq->subrequests, rreq_link) {
248 XA_STATE(xas, &rreq->mapping->i_pages, subreq->start / PAGE_SIZE);
250 xas_for_each(&xas, folio, (subreq->start + subreq->len - 1) / PAGE_SIZE) {
251 /* We might have multiple writes from the same huge
252 * folio, but we mustn't unlock a folio more than once.
254 if (have_unlocked && folio_index(folio) <= unlocked)
256 unlocked = folio_index(folio);
257 folio_end_fscache(folio);
258 have_unlocked = true;
263 netfs_rreq_completed(rreq, was_async);
266 static void netfs_rreq_copy_terminated(void *priv, ssize_t transferred_or_error,
269 struct netfs_read_subrequest *subreq = priv;
270 struct netfs_read_request *rreq = subreq->rreq;
272 if (IS_ERR_VALUE(transferred_or_error)) {
273 netfs_stat(&netfs_n_rh_write_failed);
274 trace_netfs_failure(rreq, subreq, transferred_or_error,
275 netfs_fail_copy_to_cache);
277 netfs_stat(&netfs_n_rh_write_done);
280 trace_netfs_sreq(subreq, netfs_sreq_trace_write_term);
282 /* If we decrement nr_wr_ops to 0, the ref belongs to us. */
283 if (atomic_dec_and_test(&rreq->nr_wr_ops))
284 netfs_rreq_unmark_after_write(rreq, was_async);
286 netfs_put_subrequest(subreq, was_async);
290 * Perform any outstanding writes to the cache. We inherit a ref from the
293 static void netfs_rreq_do_write_to_cache(struct netfs_read_request *rreq)
295 struct netfs_cache_resources *cres = &rreq->cache_resources;
296 struct netfs_read_subrequest *subreq, *next, *p;
297 struct iov_iter iter;
300 trace_netfs_rreq(rreq, netfs_rreq_trace_write);
302 /* We don't want terminating writes trying to wake us up whilst we're
303 * still going through the list.
305 atomic_inc(&rreq->nr_wr_ops);
307 list_for_each_entry_safe(subreq, p, &rreq->subrequests, rreq_link) {
308 if (!test_bit(NETFS_SREQ_WRITE_TO_CACHE, &subreq->flags)) {
309 list_del_init(&subreq->rreq_link);
310 netfs_put_subrequest(subreq, false);
314 list_for_each_entry(subreq, &rreq->subrequests, rreq_link) {
315 /* Amalgamate adjacent writes */
316 while (!list_is_last(&subreq->rreq_link, &rreq->subrequests)) {
317 next = list_next_entry(subreq, rreq_link);
318 if (next->start != subreq->start + subreq->len)
320 subreq->len += next->len;
321 list_del_init(&next->rreq_link);
322 netfs_put_subrequest(next, false);
325 ret = cres->ops->prepare_write(cres, &subreq->start, &subreq->len,
328 trace_netfs_failure(rreq, subreq, ret, netfs_fail_prepare_write);
329 trace_netfs_sreq(subreq, netfs_sreq_trace_write_skip);
333 iov_iter_xarray(&iter, WRITE, &rreq->mapping->i_pages,
334 subreq->start, subreq->len);
336 atomic_inc(&rreq->nr_wr_ops);
337 netfs_stat(&netfs_n_rh_write);
338 netfs_get_read_subrequest(subreq);
339 trace_netfs_sreq(subreq, netfs_sreq_trace_write);
340 cres->ops->write(cres, subreq->start, &iter,
341 netfs_rreq_copy_terminated, subreq);
344 /* If we decrement nr_wr_ops to 0, the usage ref belongs to us. */
345 if (atomic_dec_and_test(&rreq->nr_wr_ops))
346 netfs_rreq_unmark_after_write(rreq, false);
349 static void netfs_rreq_write_to_cache_work(struct work_struct *work)
351 struct netfs_read_request *rreq =
352 container_of(work, struct netfs_read_request, work);
354 netfs_rreq_do_write_to_cache(rreq);
357 static void netfs_rreq_write_to_cache(struct netfs_read_request *rreq,
361 rreq->work.func = netfs_rreq_write_to_cache_work;
362 if (!queue_work(system_unbound_wq, &rreq->work))
365 netfs_rreq_do_write_to_cache(rreq);
370 * Unlock the folios in a read operation. We need to set PG_fscache on any
371 * folios we're going to write back before we unlock them.
373 static void netfs_rreq_unlock(struct netfs_read_request *rreq)
375 struct netfs_read_subrequest *subreq;
377 unsigned int iopos, account = 0;
378 pgoff_t start_page = rreq->start / PAGE_SIZE;
379 pgoff_t last_page = ((rreq->start + rreq->len) / PAGE_SIZE) - 1;
380 bool subreq_failed = false;
382 XA_STATE(xas, &rreq->mapping->i_pages, start_page);
384 if (test_bit(NETFS_RREQ_FAILED, &rreq->flags)) {
385 __clear_bit(NETFS_RREQ_WRITE_TO_CACHE, &rreq->flags);
386 list_for_each_entry(subreq, &rreq->subrequests, rreq_link) {
387 __clear_bit(NETFS_SREQ_WRITE_TO_CACHE, &subreq->flags);
391 /* Walk through the pagecache and the I/O request lists simultaneously.
392 * We may have a mixture of cached and uncached sections and we only
393 * really want to write out the uncached sections. This is slightly
394 * complicated by the possibility that we might have huge pages with a
397 subreq = list_first_entry(&rreq->subrequests,
398 struct netfs_read_subrequest, rreq_link);
400 subreq_failed = (subreq->error < 0);
402 trace_netfs_rreq(rreq, netfs_rreq_trace_unlock);
405 xas_for_each(&xas, folio, last_page) {
406 unsigned int pgpos = (folio_index(folio) - start_page) * PAGE_SIZE;
407 unsigned int pgend = pgpos + folio_size(folio);
408 bool pg_failed = false;
415 if (test_bit(NETFS_SREQ_WRITE_TO_CACHE, &subreq->flags))
416 folio_start_fscache(folio);
417 pg_failed |= subreq_failed;
418 if (pgend < iopos + subreq->len)
421 account += subreq->transferred;
422 iopos += subreq->len;
423 if (!list_is_last(&subreq->rreq_link, &rreq->subrequests)) {
424 subreq = list_next_entry(subreq, rreq_link);
425 subreq_failed = (subreq->error < 0);
428 subreq_failed = false;
435 flush_dcache_folio(folio);
436 folio_mark_uptodate(folio);
439 if (!test_bit(NETFS_RREQ_DONT_UNLOCK_FOLIOS, &rreq->flags)) {
440 if (folio_index(folio) == rreq->no_unlock_folio &&
441 test_bit(NETFS_RREQ_NO_UNLOCK_FOLIO, &rreq->flags))
449 task_io_account_read(account);
450 if (rreq->netfs_ops->done)
451 rreq->netfs_ops->done(rreq);
455 * Handle a short read.
457 static void netfs_rreq_short_read(struct netfs_read_request *rreq,
458 struct netfs_read_subrequest *subreq)
460 __clear_bit(NETFS_SREQ_SHORT_READ, &subreq->flags);
461 __set_bit(NETFS_SREQ_SEEK_DATA_READ, &subreq->flags);
463 netfs_stat(&netfs_n_rh_short_read);
464 trace_netfs_sreq(subreq, netfs_sreq_trace_resubmit_short);
466 netfs_get_read_subrequest(subreq);
467 atomic_inc(&rreq->nr_rd_ops);
468 if (subreq->source == NETFS_READ_FROM_CACHE)
469 netfs_read_from_cache(rreq, subreq, true);
471 netfs_read_from_server(rreq, subreq);
475 * Resubmit any short or failed operations. Returns true if we got the rreq
478 static bool netfs_rreq_perform_resubmissions(struct netfs_read_request *rreq)
480 struct netfs_read_subrequest *subreq;
482 WARN_ON(in_interrupt());
484 trace_netfs_rreq(rreq, netfs_rreq_trace_resubmit);
486 /* We don't want terminating submissions trying to wake us up whilst
487 * we're still going through the list.
489 atomic_inc(&rreq->nr_rd_ops);
491 __clear_bit(NETFS_RREQ_INCOMPLETE_IO, &rreq->flags);
492 list_for_each_entry(subreq, &rreq->subrequests, rreq_link) {
494 if (subreq->source != NETFS_READ_FROM_CACHE)
496 subreq->source = NETFS_DOWNLOAD_FROM_SERVER;
498 netfs_stat(&netfs_n_rh_download_instead);
499 trace_netfs_sreq(subreq, netfs_sreq_trace_download_instead);
500 netfs_get_read_subrequest(subreq);
501 atomic_inc(&rreq->nr_rd_ops);
502 netfs_read_from_server(rreq, subreq);
503 } else if (test_bit(NETFS_SREQ_SHORT_READ, &subreq->flags)) {
504 netfs_rreq_short_read(rreq, subreq);
508 /* If we decrement nr_rd_ops to 0, the usage ref belongs to us. */
509 if (atomic_dec_and_test(&rreq->nr_rd_ops))
512 wake_up_var(&rreq->nr_rd_ops);
517 * Check to see if the data read is still valid.
519 static void netfs_rreq_is_still_valid(struct netfs_read_request *rreq)
521 struct netfs_read_subrequest *subreq;
523 if (!rreq->netfs_ops->is_still_valid ||
524 rreq->netfs_ops->is_still_valid(rreq))
527 list_for_each_entry(subreq, &rreq->subrequests, rreq_link) {
528 if (subreq->source == NETFS_READ_FROM_CACHE) {
529 subreq->error = -ESTALE;
530 __set_bit(NETFS_RREQ_INCOMPLETE_IO, &rreq->flags);
536 * Assess the state of a read request and decide what to do next.
538 * Note that we could be in an ordinary kernel thread, on a workqueue or in
539 * softirq context at this point. We inherit a ref from the caller.
541 static void netfs_rreq_assess(struct netfs_read_request *rreq, bool was_async)
543 trace_netfs_rreq(rreq, netfs_rreq_trace_assess);
546 netfs_rreq_is_still_valid(rreq);
548 if (!test_bit(NETFS_RREQ_FAILED, &rreq->flags) &&
549 test_bit(NETFS_RREQ_INCOMPLETE_IO, &rreq->flags)) {
550 if (netfs_rreq_perform_resubmissions(rreq))
555 netfs_rreq_unlock(rreq);
557 clear_bit_unlock(NETFS_RREQ_IN_PROGRESS, &rreq->flags);
558 wake_up_bit(&rreq->flags, NETFS_RREQ_IN_PROGRESS);
560 if (test_bit(NETFS_RREQ_WRITE_TO_CACHE, &rreq->flags))
561 return netfs_rreq_write_to_cache(rreq, was_async);
563 netfs_rreq_completed(rreq, was_async);
566 static void netfs_rreq_work(struct work_struct *work)
568 struct netfs_read_request *rreq =
569 container_of(work, struct netfs_read_request, work);
570 netfs_rreq_assess(rreq, false);
574 * Handle the completion of all outstanding I/O operations on a read request.
575 * We inherit a ref from the caller.
577 static void netfs_rreq_terminated(struct netfs_read_request *rreq,
580 if (test_bit(NETFS_RREQ_INCOMPLETE_IO, &rreq->flags) &&
582 if (!queue_work(system_unbound_wq, &rreq->work))
585 netfs_rreq_assess(rreq, was_async);
590 * netfs_subreq_terminated - Note the termination of an I/O operation.
591 * @subreq: The I/O request that has terminated.
592 * @transferred_or_error: The amount of data transferred or an error code.
593 * @was_async: The termination was asynchronous
595 * This tells the read helper that a contributory I/O operation has terminated,
596 * one way or another, and that it should integrate the results.
598 * The caller indicates in @transferred_or_error the outcome of the operation,
599 * supplying a positive value to indicate the number of bytes transferred, 0 to
600 * indicate a failure to transfer anything that should be retried or a negative
601 * error code. The helper will look after reissuing I/O operations as
602 * appropriate and writing downloaded data to the cache.
604 * If @was_async is true, the caller might be running in softirq or interrupt
605 * context and we can't sleep.
607 void netfs_subreq_terminated(struct netfs_read_subrequest *subreq,
608 ssize_t transferred_or_error,
611 struct netfs_read_request *rreq = subreq->rreq;
614 _enter("[%u]{%llx,%lx},%zd",
615 subreq->debug_index, subreq->start, subreq->flags,
616 transferred_or_error);
618 switch (subreq->source) {
619 case NETFS_READ_FROM_CACHE:
620 netfs_stat(&netfs_n_rh_read_done);
622 case NETFS_DOWNLOAD_FROM_SERVER:
623 netfs_stat(&netfs_n_rh_download_done);
629 if (IS_ERR_VALUE(transferred_or_error)) {
630 subreq->error = transferred_or_error;
631 trace_netfs_failure(rreq, subreq, transferred_or_error,
636 if (WARN(transferred_or_error > subreq->len - subreq->transferred,
637 "Subreq overread: R%x[%x] %zd > %zu - %zu",
638 rreq->debug_id, subreq->debug_index,
639 transferred_or_error, subreq->len, subreq->transferred))
640 transferred_or_error = subreq->len - subreq->transferred;
643 subreq->transferred += transferred_or_error;
644 if (subreq->transferred < subreq->len)
648 __clear_bit(NETFS_SREQ_NO_PROGRESS, &subreq->flags);
649 if (test_bit(NETFS_SREQ_WRITE_TO_CACHE, &subreq->flags))
650 set_bit(NETFS_RREQ_WRITE_TO_CACHE, &rreq->flags);
653 trace_netfs_sreq(subreq, netfs_sreq_trace_terminated);
655 /* If we decrement nr_rd_ops to 0, the ref belongs to us. */
656 u = atomic_dec_return(&rreq->nr_rd_ops);
658 netfs_rreq_terminated(rreq, was_async);
660 wake_up_var(&rreq->nr_rd_ops);
662 netfs_put_subrequest(subreq, was_async);
666 if (test_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags)) {
667 netfs_clear_unread(subreq);
668 subreq->transferred = subreq->len;
672 if (transferred_or_error == 0) {
673 if (__test_and_set_bit(NETFS_SREQ_NO_PROGRESS, &subreq->flags)) {
674 subreq->error = -ENODATA;
678 __clear_bit(NETFS_SREQ_NO_PROGRESS, &subreq->flags);
681 __set_bit(NETFS_SREQ_SHORT_READ, &subreq->flags);
682 set_bit(NETFS_RREQ_INCOMPLETE_IO, &rreq->flags);
686 if (subreq->source == NETFS_READ_FROM_CACHE) {
687 netfs_stat(&netfs_n_rh_read_failed);
688 set_bit(NETFS_RREQ_INCOMPLETE_IO, &rreq->flags);
690 netfs_stat(&netfs_n_rh_download_failed);
691 set_bit(NETFS_RREQ_FAILED, &rreq->flags);
692 rreq->error = subreq->error;
696 EXPORT_SYMBOL(netfs_subreq_terminated);
698 static enum netfs_read_source netfs_cache_prepare_read(struct netfs_read_subrequest *subreq,
701 struct netfs_read_request *rreq = subreq->rreq;
702 struct netfs_cache_resources *cres = &rreq->cache_resources;
705 return cres->ops->prepare_read(subreq, i_size);
706 if (subreq->start >= rreq->i_size)
707 return NETFS_FILL_WITH_ZEROES;
708 return NETFS_DOWNLOAD_FROM_SERVER;
712 * Work out what sort of subrequest the next one will be.
714 static enum netfs_read_source
715 netfs_rreq_prepare_read(struct netfs_read_request *rreq,
716 struct netfs_read_subrequest *subreq)
718 enum netfs_read_source source;
720 _enter("%llx-%llx,%llx", subreq->start, subreq->start + subreq->len, rreq->i_size);
722 source = netfs_cache_prepare_read(subreq, rreq->i_size);
723 if (source == NETFS_INVALID_READ)
726 if (source == NETFS_DOWNLOAD_FROM_SERVER) {
727 /* Call out to the netfs to let it shrink the request to fit
728 * its own I/O sizes and boundaries. If it shinks it here, it
729 * will be called again to make simultaneous calls; if it wants
730 * to make serial calls, it can indicate a short read and then
731 * we will call it again.
733 if (subreq->len > rreq->i_size - subreq->start)
734 subreq->len = rreq->i_size - subreq->start;
736 if (rreq->netfs_ops->clamp_length &&
737 !rreq->netfs_ops->clamp_length(subreq)) {
738 source = NETFS_INVALID_READ;
743 if (WARN_ON(subreq->len == 0))
744 source = NETFS_INVALID_READ;
747 subreq->source = source;
748 trace_netfs_sreq(subreq, netfs_sreq_trace_prepare);
753 * Slice off a piece of a read request and submit an I/O request for it.
755 static bool netfs_rreq_submit_slice(struct netfs_read_request *rreq,
756 unsigned int *_debug_index)
758 struct netfs_read_subrequest *subreq;
759 enum netfs_read_source source;
761 subreq = netfs_alloc_subrequest(rreq);
765 subreq->debug_index = (*_debug_index)++;
766 subreq->start = rreq->start + rreq->submitted;
767 subreq->len = rreq->len - rreq->submitted;
769 _debug("slice %llx,%zx,%zx", subreq->start, subreq->len, rreq->submitted);
770 list_add_tail(&subreq->rreq_link, &rreq->subrequests);
772 /* Call out to the cache to find out what it can do with the remaining
773 * subset. It tells us in subreq->flags what it decided should be done
774 * and adjusts subreq->len down if the subset crosses a cache boundary.
776 * Then when we hand the subset, it can choose to take a subset of that
777 * (the starts must coincide), in which case, we go around the loop
778 * again and ask it to download the next piece.
780 source = netfs_rreq_prepare_read(rreq, subreq);
781 if (source == NETFS_INVALID_READ)
784 atomic_inc(&rreq->nr_rd_ops);
786 rreq->submitted += subreq->len;
788 trace_netfs_sreq(subreq, netfs_sreq_trace_submit);
790 case NETFS_FILL_WITH_ZEROES:
791 netfs_fill_with_zeroes(rreq, subreq);
793 case NETFS_DOWNLOAD_FROM_SERVER:
794 netfs_read_from_server(rreq, subreq);
796 case NETFS_READ_FROM_CACHE:
797 netfs_read_from_cache(rreq, subreq, false);
806 rreq->error = subreq->error;
807 netfs_put_subrequest(subreq, false);
811 static void netfs_cache_expand_readahead(struct netfs_read_request *rreq,
812 loff_t *_start, size_t *_len, loff_t i_size)
814 struct netfs_cache_resources *cres = &rreq->cache_resources;
816 if (cres->ops && cres->ops->expand_readahead)
817 cres->ops->expand_readahead(cres, _start, _len, i_size);
820 static void netfs_rreq_expand(struct netfs_read_request *rreq,
821 struct readahead_control *ractl)
823 /* Give the cache a chance to change the request parameters. The
824 * resultant request must contain the original region.
826 netfs_cache_expand_readahead(rreq, &rreq->start, &rreq->len, rreq->i_size);
828 /* Give the netfs a chance to change the request parameters. The
829 * resultant request must contain the original region.
831 if (rreq->netfs_ops->expand_readahead)
832 rreq->netfs_ops->expand_readahead(rreq);
834 /* Expand the request if the cache wants it to start earlier. Note
835 * that the expansion may get further extended if the VM wishes to
836 * insert THPs and the preferred start and/or end wind up in the middle
839 * If this is the case, however, the THP size should be an integer
840 * multiple of the cache granule size, so we get a whole number of
841 * granules to deal with.
843 if (rreq->start != readahead_pos(ractl) ||
844 rreq->len != readahead_length(ractl)) {
845 readahead_expand(ractl, rreq->start, rreq->len);
846 rreq->start = readahead_pos(ractl);
847 rreq->len = readahead_length(ractl);
849 trace_netfs_read(rreq, readahead_pos(ractl), readahead_length(ractl),
850 netfs_read_trace_expanded);
855 * netfs_readahead - Helper to manage a read request
856 * @ractl: The description of the readahead request
857 * @ops: The network filesystem's operations for the helper to use
858 * @netfs_priv: Private netfs data to be retained in the request
860 * Fulfil a readahead request by drawing data from the cache if possible, or
861 * the netfs if not. Space beyond the EOF is zero-filled. Multiple I/O
862 * requests from different sources will get munged together. If necessary, the
863 * readahead window can be expanded in either direction to a more convenient
864 * alighment for RPC efficiency or to make storage in the cache feasible.
866 * The calling netfs must provide a table of operations, only one of which,
867 * issue_op, is mandatory. It may also be passed a private token, which will
868 * be retained in rreq->netfs_priv and will be cleaned up by ops->cleanup().
870 * This is usable whether or not caching is enabled.
872 void netfs_readahead(struct readahead_control *ractl,
873 const struct netfs_read_request_ops *ops,
876 struct netfs_read_request *rreq;
877 unsigned int debug_index = 0;
880 _enter("%lx,%x", readahead_index(ractl), readahead_count(ractl));
882 if (readahead_count(ractl) == 0)
885 rreq = netfs_alloc_read_request(ops, netfs_priv, ractl->file);
888 rreq->mapping = ractl->mapping;
889 rreq->start = readahead_pos(ractl);
890 rreq->len = readahead_length(ractl);
892 if (ops->begin_cache_operation) {
893 ret = ops->begin_cache_operation(rreq);
894 if (ret == -ENOMEM || ret == -EINTR || ret == -ERESTARTSYS)
898 netfs_stat(&netfs_n_rh_readahead);
899 trace_netfs_read(rreq, readahead_pos(ractl), readahead_length(ractl),
900 netfs_read_trace_readahead);
902 netfs_rreq_expand(rreq, ractl);
904 atomic_set(&rreq->nr_rd_ops, 1);
906 if (!netfs_rreq_submit_slice(rreq, &debug_index))
909 } while (rreq->submitted < rreq->len);
911 /* Drop the refs on the folios here rather than in the cache or
912 * filesystem. The locks will be dropped in netfs_rreq_unlock().
914 while (readahead_folio(ractl))
917 /* If we decrement nr_rd_ops to 0, the ref belongs to us. */
918 if (atomic_dec_and_test(&rreq->nr_rd_ops))
919 netfs_rreq_assess(rreq, false);
923 netfs_put_read_request(rreq, false);
927 ops->cleanup(ractl->mapping, netfs_priv);
930 EXPORT_SYMBOL(netfs_readahead);
933 * netfs_readpage - Helper to manage a readpage request
934 * @file: The file to read from
935 * @folio: The folio to read
936 * @ops: The network filesystem's operations for the helper to use
937 * @netfs_priv: Private netfs data to be retained in the request
939 * Fulfil a readpage request by drawing data from the cache if possible, or the
940 * netfs if not. Space beyond the EOF is zero-filled. Multiple I/O requests
941 * from different sources will get munged together.
943 * The calling netfs must provide a table of operations, only one of which,
944 * issue_op, is mandatory. It may also be passed a private token, which will
945 * be retained in rreq->netfs_priv and will be cleaned up by ops->cleanup().
947 * This is usable whether or not caching is enabled.
949 int netfs_readpage(struct file *file,
951 const struct netfs_read_request_ops *ops,
954 struct netfs_read_request *rreq;
955 unsigned int debug_index = 0;
958 _enter("%lx", folio_index(folio));
960 rreq = netfs_alloc_read_request(ops, netfs_priv, file);
963 ops->cleanup(netfs_priv, folio_file_mapping(folio));
967 rreq->mapping = folio_file_mapping(folio);
968 rreq->start = folio_file_pos(folio);
969 rreq->len = folio_size(folio);
971 if (ops->begin_cache_operation) {
972 ret = ops->begin_cache_operation(rreq);
973 if (ret == -ENOMEM || ret == -EINTR || ret == -ERESTARTSYS) {
979 netfs_stat(&netfs_n_rh_readpage);
980 trace_netfs_read(rreq, rreq->start, rreq->len, netfs_read_trace_readpage);
982 netfs_get_read_request(rreq);
984 atomic_set(&rreq->nr_rd_ops, 1);
986 if (!netfs_rreq_submit_slice(rreq, &debug_index))
989 } while (rreq->submitted < rreq->len);
991 /* Keep nr_rd_ops incremented so that the ref always belongs to us, and
992 * the service code isn't punted off to a random thread pool to
996 wait_var_event(&rreq->nr_rd_ops, atomic_read(&rreq->nr_rd_ops) == 1);
997 netfs_rreq_assess(rreq, false);
998 } while (test_bit(NETFS_RREQ_IN_PROGRESS, &rreq->flags));
1001 if (ret == 0 && rreq->submitted < rreq->len) {
1002 trace_netfs_failure(rreq, NULL, ret, netfs_fail_short_readpage);
1006 netfs_put_read_request(rreq, false);
1009 EXPORT_SYMBOL(netfs_readpage);
1012 * netfs_skip_folio_read - prep a folio for writing without reading first
1013 * @folio: The folio being prepared
1014 * @pos: starting position for the write
1015 * @len: length of write
1017 * In some cases, write_begin doesn't need to read at all:
1018 * - full folio write
1019 * - write that lies in a folio that is completely beyond EOF
1020 * - write that covers the folio from start to EOF or beyond it
1022 * If any of these criteria are met, then zero out the unwritten parts
1023 * of the folio and return true. Otherwise, return false.
1025 static bool netfs_skip_folio_read(struct folio *folio, loff_t pos, size_t len)
1027 struct inode *inode = folio_inode(folio);
1028 loff_t i_size = i_size_read(inode);
1029 size_t offset = offset_in_folio(folio, pos);
1031 /* Full folio write */
1032 if (offset == 0 && len >= folio_size(folio))
1035 /* pos beyond last folio in the file */
1036 if (pos - offset >= i_size)
1039 /* Write that covers from the start of the folio to EOF or beyond */
1040 if (offset == 0 && (pos + len) >= i_size)
1045 zero_user_segments(&folio->page, 0, offset, offset + len, folio_size(folio));
1050 * netfs_write_begin - Helper to prepare for writing
1051 * @file: The file to read from
1052 * @mapping: The mapping to read from
1053 * @pos: File position at which the write will begin
1054 * @len: The length of the write (may extend beyond the end of the folio chosen)
1055 * @aop_flags: AOP_* flags
1056 * @_folio: Where to put the resultant folio
1057 * @_fsdata: Place for the netfs to store a cookie
1058 * @ops: The network filesystem's operations for the helper to use
1059 * @netfs_priv: Private netfs data to be retained in the request
1061 * Pre-read data for a write-begin request by drawing data from the cache if
1062 * possible, or the netfs if not. Space beyond the EOF is zero-filled.
1063 * Multiple I/O requests from different sources will get munged together. If
1064 * necessary, the readahead window can be expanded in either direction to a
1065 * more convenient alighment for RPC efficiency or to make storage in the cache
1068 * The calling netfs must provide a table of operations, only one of which,
1069 * issue_op, is mandatory.
1071 * The check_write_begin() operation can be provided to check for and flush
1072 * conflicting writes once the folio is grabbed and locked. It is passed a
1073 * pointer to the fsdata cookie that gets returned to the VM to be passed to
1074 * write_end. It is permitted to sleep. It should return 0 if the request
1075 * should go ahead; unlock the folio and return -EAGAIN to cause the folio to
1076 * be regot; or return an error.
1078 * This is usable whether or not caching is enabled.
1080 int netfs_write_begin(struct file *file, struct address_space *mapping,
1081 loff_t pos, unsigned int len, unsigned int aop_flags,
1082 struct folio **_folio, void **_fsdata,
1083 const struct netfs_read_request_ops *ops,
1086 struct netfs_read_request *rreq;
1087 struct folio *folio;
1088 struct inode *inode = file_inode(file);
1089 unsigned int debug_index = 0, fgp_flags;
1090 pgoff_t index = pos >> PAGE_SHIFT;
1093 DEFINE_READAHEAD(ractl, file, NULL, mapping, index);
1096 fgp_flags = FGP_LOCK | FGP_WRITE | FGP_CREAT | FGP_STABLE;
1097 if (aop_flags & AOP_FLAG_NOFS)
1098 fgp_flags |= FGP_NOFS;
1099 folio = __filemap_get_folio(mapping, index, fgp_flags,
1100 mapping_gfp_mask(mapping));
1104 if (ops->check_write_begin) {
1105 /* Allow the netfs (eg. ceph) to flush conflicts. */
1106 ret = ops->check_write_begin(file, pos, len, folio, _fsdata);
1108 trace_netfs_failure(NULL, NULL, ret, netfs_fail_check_write_begin);
1115 if (folio_test_uptodate(folio))
1118 /* If the page is beyond the EOF, we want to clear it - unless it's
1119 * within the cache granule containing the EOF, in which case we need
1120 * to preload the granule.
1122 if (!ops->is_cache_enabled(inode) &&
1123 netfs_skip_folio_read(folio, pos, len)) {
1124 netfs_stat(&netfs_n_rh_write_zskip);
1125 goto have_folio_no_wait;
1129 rreq = netfs_alloc_read_request(ops, netfs_priv, file);
1132 rreq->mapping = folio_file_mapping(folio);
1133 rreq->start = folio_file_pos(folio);
1134 rreq->len = folio_size(folio);
1135 rreq->no_unlock_folio = folio_index(folio);
1136 __set_bit(NETFS_RREQ_NO_UNLOCK_FOLIO, &rreq->flags);
1139 if (ops->begin_cache_operation) {
1140 ret = ops->begin_cache_operation(rreq);
1141 if (ret == -ENOMEM || ret == -EINTR || ret == -ERESTARTSYS)
1145 netfs_stat(&netfs_n_rh_write_begin);
1146 trace_netfs_read(rreq, pos, len, netfs_read_trace_write_begin);
1148 /* Expand the request to meet caching requirements and download
1151 ractl._nr_pages = folio_nr_pages(folio);
1152 netfs_rreq_expand(rreq, &ractl);
1153 netfs_get_read_request(rreq);
1155 /* We hold the folio locks, so we can drop the references */
1157 while (readahead_folio(&ractl))
1160 atomic_set(&rreq->nr_rd_ops, 1);
1162 if (!netfs_rreq_submit_slice(rreq, &debug_index))
1165 } while (rreq->submitted < rreq->len);
1167 /* Keep nr_rd_ops incremented so that the ref always belongs to us, and
1168 * the service code isn't punted off to a random thread pool to
1172 wait_var_event(&rreq->nr_rd_ops, atomic_read(&rreq->nr_rd_ops) == 1);
1173 netfs_rreq_assess(rreq, false);
1174 if (!test_bit(NETFS_RREQ_IN_PROGRESS, &rreq->flags))
1180 if (ret == 0 && rreq->submitted < rreq->len) {
1181 trace_netfs_failure(rreq, NULL, ret, netfs_fail_short_write_begin);
1184 netfs_put_read_request(rreq, false);
1189 ret = folio_wait_fscache_killable(folio);
1194 ops->cleanup(netfs_priv, mapping);
1200 netfs_put_read_request(rreq, false);
1202 folio_unlock(folio);
1205 ops->cleanup(netfs_priv, mapping);
1206 _leave(" = %d", ret);
1209 EXPORT_SYMBOL(netfs_write_begin);