1// SPDX-License-Identifier: GPL-2.0-or-later
  2/* Network filesystem high-level buffered read support.
  3 *
  4 * Copyright (C) 2021 Red Hat, Inc. All Rights Reserved.
  5 * Written by David Howells (dhowells@redhat.com)
  6 */
  7
  8#include <linux/export.h>
  9#include <linux/task_io_accounting_ops.h>
 10#include "internal.h"
 11
 12/*
 13 * Unlock the folios in a read operation.  We need to set PG_fscache on any
 14 * folios we're going to write back before we unlock them.
 15 */
 16void netfs_rreq_unlock_folios(struct netfs_io_request *rreq)
 17{
 18	struct netfs_io_subrequest *subreq;
 19	struct folio *folio;
 20	pgoff_t start_page = rreq->start / PAGE_SIZE;
 21	pgoff_t last_page = ((rreq->start + rreq->len) / PAGE_SIZE) - 1;
 22	size_t account = 0;
 23	bool subreq_failed = false;
 24
 25	XA_STATE(xas, &rreq->mapping->i_pages, start_page);
 26
 27	if (test_bit(NETFS_RREQ_FAILED, &rreq->flags)) {
 28		__clear_bit(NETFS_RREQ_COPY_TO_CACHE, &rreq->flags);
 29		list_for_each_entry(subreq, &rreq->subrequests, rreq_link) {
 30			__clear_bit(NETFS_SREQ_COPY_TO_CACHE, &subreq->flags);
 31		}
 32	}
 33
 34	/* Walk through the pagecache and the I/O request lists simultaneously.
 35	 * We may have a mixture of cached and uncached sections and we only
 36	 * really want to write out the uncached sections.  This is slightly
 37	 * complicated by the possibility that we might have huge pages with a
 38	 * mixture inside.
 39	 */
 40	subreq = list_first_entry(&rreq->subrequests,
 41				  struct netfs_io_subrequest, rreq_link);
 42	subreq_failed = (subreq->error < 0);
 43
 44	trace_netfs_rreq(rreq, netfs_rreq_trace_unlock);
 45
 46	rcu_read_lock();
 47	xas_for_each(&xas, folio, last_page) {
 48		loff_t pg_end;
 49		bool pg_failed = false;
 50
 51		if (xas_retry(&xas, folio))
 52			continue;
 53
 54		pg_end = folio_pos(folio) + folio_size(folio) - 1;
 55
 56		for (;;) {
 57			loff_t sreq_end;
 58
 59			if (!subreq) {
 60				pg_failed = true;
 61				break;
 62			}
 63			if (test_bit(NETFS_SREQ_COPY_TO_CACHE, &subreq->flags))
 64				folio_start_fscache(folio);
 65			pg_failed |= subreq_failed;
 66			sreq_end = subreq->start + subreq->len - 1;
 67			if (pg_end < sreq_end)
 68				break;
 69
 70			account += subreq->transferred;
 71			if (!list_is_last(&subreq->rreq_link, &rreq->subrequests)) {
 72				subreq = list_next_entry(subreq, rreq_link);
 73				subreq_failed = (subreq->error < 0);
 74			} else {
 75				subreq = NULL;
 76				subreq_failed = false;
 77			}
 78
 79			if (pg_end == sreq_end)
 80				break;
 81		}
 82
 83		if (!pg_failed) {
 84			flush_dcache_folio(folio);
 85			folio_mark_uptodate(folio);
 86		}
 87
 88		if (!test_bit(NETFS_RREQ_DONT_UNLOCK_FOLIOS, &rreq->flags)) {
 89			if (folio_index(folio) == rreq->no_unlock_folio &&
 90			    test_bit(NETFS_RREQ_NO_UNLOCK_FOLIO, &rreq->flags))
 91				_debug("no unlock");
 92			else
 93				folio_unlock(folio);
 94		}
 95	}
 96	rcu_read_unlock();
 97
 98	task_io_account_read(account);
 99	if (rreq->netfs_ops->done)
100		rreq->netfs_ops->done(rreq);
101}
102
103static void netfs_cache_expand_readahead(struct netfs_io_request *rreq,
104					 loff_t *_start, size_t *_len, loff_t i_size)
105{
106	struct netfs_cache_resources *cres = &rreq->cache_resources;
107
108	if (cres->ops && cres->ops->expand_readahead)
109		cres->ops->expand_readahead(cres, _start, _len, i_size);
110}
111
112static void netfs_rreq_expand(struct netfs_io_request *rreq,
113			      struct readahead_control *ractl)
114{
115	/* Give the cache a chance to change the request parameters.  The
116	 * resultant request must contain the original region.
117	 */
118	netfs_cache_expand_readahead(rreq, &rreq->start, &rreq->len, rreq->i_size);
119
120	/* Give the netfs a chance to change the request parameters.  The
121	 * resultant request must contain the original region.
122	 */
123	if (rreq->netfs_ops->expand_readahead)
124		rreq->netfs_ops->expand_readahead(rreq);
125
126	/* Expand the request if the cache wants it to start earlier.  Note
127	 * that the expansion may get further extended if the VM wishes to
128	 * insert THPs and the preferred start and/or end wind up in the middle
129	 * of THPs.
130	 *
131	 * If this is the case, however, the THP size should be an integer
132	 * multiple of the cache granule size, so we get a whole number of
133	 * granules to deal with.
134	 */
135	if (rreq->start  != readahead_pos(ractl) ||
136	    rreq->len != readahead_length(ractl)) {
137		readahead_expand(ractl, rreq->start, rreq->len);
138		rreq->start  = readahead_pos(ractl);
139		rreq->len = readahead_length(ractl);
140
141		trace_netfs_read(rreq, readahead_pos(ractl), readahead_length(ractl),
142				 netfs_read_trace_expanded);
143	}
144}
145
146/**
147 * netfs_readahead - Helper to manage a read request
148 * @ractl: The description of the readahead request
149 *
150 * Fulfil a readahead request by drawing data from the cache if possible, or
151 * the netfs if not.  Space beyond the EOF is zero-filled.  Multiple I/O
152 * requests from different sources will get munged together.  If necessary, the
153 * readahead window can be expanded in either direction to a more convenient
154 * alighment for RPC efficiency or to make storage in the cache feasible.
155 *
156 * The calling netfs must initialise a netfs context contiguous to the vfs
157 * inode before calling this.
158 *
159 * This is usable whether or not caching is enabled.
160 */
161void netfs_readahead(struct readahead_control *ractl)
162{
163	struct netfs_io_request *rreq;
164	struct netfs_inode *ctx = netfs_inode(ractl->mapping->host);
165	int ret;
166
167	_enter("%lx,%x", readahead_index(ractl), readahead_count(ractl));
168
169	if (readahead_count(ractl) == 0)
170		return;
171
172	rreq = netfs_alloc_request(ractl->mapping, ractl->file,
173				   readahead_pos(ractl),
174				   readahead_length(ractl),
175				   NETFS_READAHEAD);
176	if (IS_ERR(rreq))
177		return;
178
179	if (ctx->ops->begin_cache_operation) {
180		ret = ctx->ops->begin_cache_operation(rreq);
181		if (ret == -ENOMEM || ret == -EINTR || ret == -ERESTARTSYS)
182			goto cleanup_free;
183	}
184
185	netfs_stat(&netfs_n_rh_readahead);
186	trace_netfs_read(rreq, readahead_pos(ractl), readahead_length(ractl),
187			 netfs_read_trace_readahead);
188
189	netfs_rreq_expand(rreq, ractl);
190
191	/* Drop the refs on the folios here rather than in the cache or
192	 * filesystem.  The locks will be dropped in netfs_rreq_unlock().
193	 */
194	while (readahead_folio(ractl))
195		;
196
197	netfs_begin_read(rreq, false);
198	return;
199
200cleanup_free:
201	netfs_put_request(rreq, false, netfs_rreq_trace_put_failed);
202	return;
203}
204EXPORT_SYMBOL(netfs_readahead);
205
206/**
207 * netfs_read_folio - Helper to manage a read_folio request
208 * @file: The file to read from
209 * @folio: The folio to read
210 *
211 * Fulfil a read_folio request by drawing data from the cache if
212 * possible, or the netfs if not.  Space beyond the EOF is zero-filled.
213 * Multiple I/O requests from different sources will get munged together.
214 *
215 * The calling netfs must initialise a netfs context contiguous to the vfs
216 * inode before calling this.
217 *
218 * This is usable whether or not caching is enabled.
219 */
220int netfs_read_folio(struct file *file, struct folio *folio)
221{
222	struct address_space *mapping = folio_file_mapping(folio);
223	struct netfs_io_request *rreq;
224	struct netfs_inode *ctx = netfs_inode(mapping->host);
225	int ret;
226
227	_enter("%lx", folio_index(folio));
228
229	rreq = netfs_alloc_request(mapping, file,
230				   folio_file_pos(folio), folio_size(folio),
231				   NETFS_READPAGE);
232	if (IS_ERR(rreq)) {
233		ret = PTR_ERR(rreq);
234		goto alloc_error;
235	}
236
237	if (ctx->ops->begin_cache_operation) {
238		ret = ctx->ops->begin_cache_operation(rreq);
239		if (ret == -ENOMEM || ret == -EINTR || ret == -ERESTARTSYS)
240			goto discard;
241	}
242
243	netfs_stat(&netfs_n_rh_readpage);
244	trace_netfs_read(rreq, rreq->start, rreq->len, netfs_read_trace_readpage);
245	return netfs_begin_read(rreq, true);
246
247discard:
248	netfs_put_request(rreq, false, netfs_rreq_trace_put_discard);
249alloc_error:
250	folio_unlock(folio);
251	return ret;
252}
253EXPORT_SYMBOL(netfs_read_folio);
254
255/*
256 * Prepare a folio for writing without reading first
257 * @folio: The folio being prepared
258 * @pos: starting position for the write
259 * @len: length of write
260 * @always_fill: T if the folio should always be completely filled/cleared
261 *
262 * In some cases, write_begin doesn't need to read at all:
263 * - full folio write
264 * - write that lies in a folio that is completely beyond EOF
265 * - write that covers the folio from start to EOF or beyond it
266 *
267 * If any of these criteria are met, then zero out the unwritten parts
268 * of the folio and return true. Otherwise, return false.
269 */
270static bool netfs_skip_folio_read(struct folio *folio, loff_t pos, size_t len,
271				 bool always_fill)
272{
273	struct inode *inode = folio_inode(folio);
274	loff_t i_size = i_size_read(inode);
275	size_t offset = offset_in_folio(folio, pos);
276	size_t plen = folio_size(folio);
277
278	if (unlikely(always_fill)) {
279		if (pos - offset + len <= i_size)
280			return false; /* Page entirely before EOF */
281		zero_user_segment(&folio->page, 0, plen);
282		folio_mark_uptodate(folio);
283		return true;
284	}
285
286	/* Full folio write */
287	if (offset == 0 && len >= plen)
288		return true;
289
290	/* Page entirely beyond the end of the file */
291	if (pos - offset >= i_size)
292		goto zero_out;
293
294	/* Write that covers from the start of the folio to EOF or beyond */
295	if (offset == 0 && (pos + len) >= i_size)
296		goto zero_out;
297
298	return false;
299zero_out:
300	zero_user_segments(&folio->page, 0, offset, offset + len, plen);
301	return true;
302}
303
304/**
305 * netfs_write_begin - Helper to prepare for writing
306 * @ctx: The netfs context
307 * @file: The file to read from
308 * @mapping: The mapping to read from
309 * @pos: File position at which the write will begin
310 * @len: The length of the write (may extend beyond the end of the folio chosen)
311 * @_folio: Where to put the resultant folio
312 * @_fsdata: Place for the netfs to store a cookie
313 *
314 * Pre-read data for a write-begin request by drawing data from the cache if
315 * possible, or the netfs if not.  Space beyond the EOF is zero-filled.
316 * Multiple I/O requests from different sources will get munged together.  If
317 * necessary, the readahead window can be expanded in either direction to a
318 * more convenient alighment for RPC efficiency or to make storage in the cache
319 * feasible.
320 *
321 * The calling netfs must provide a table of operations, only one of which,
322 * issue_op, is mandatory.
323 *
324 * The check_write_begin() operation can be provided to check for and flush
325 * conflicting writes once the folio is grabbed and locked.  It is passed a
326 * pointer to the fsdata cookie that gets returned to the VM to be passed to
327 * write_end.  It is permitted to sleep.  It should return 0 if the request
328 * should go ahead or it may return an error.  It may also unlock and put the
329 * folio, provided it sets ``*foliop`` to NULL, in which case a return of 0
330 * will cause the folio to be re-got and the process to be retried.
331 *
332 * The calling netfs must initialise a netfs context contiguous to the vfs
333 * inode before calling this.
334 *
335 * This is usable whether or not caching is enabled.
336 */
337int netfs_write_begin(struct netfs_inode *ctx,
338		      struct file *file, struct address_space *mapping,
339		      loff_t pos, unsigned int len, struct folio **_folio,
340		      void **_fsdata)
341{
342	struct netfs_io_request *rreq;
343	struct folio *folio;
344	unsigned int fgp_flags = FGP_LOCK | FGP_WRITE | FGP_CREAT | FGP_STABLE;
345	pgoff_t index = pos >> PAGE_SHIFT;
346	int ret;
347
348	DEFINE_READAHEAD(ractl, file, NULL, mapping, index);
349
350retry:
351	folio = __filemap_get_folio(mapping, index, fgp_flags,
352				    mapping_gfp_mask(mapping));
353	if (!folio)
354		return -ENOMEM;
355
356	if (ctx->ops->check_write_begin) {
357		/* Allow the netfs (eg. ceph) to flush conflicts. */
358		ret = ctx->ops->check_write_begin(file, pos, len, &folio, _fsdata);
359		if (ret < 0) {
360			trace_netfs_failure(NULL, NULL, ret, netfs_fail_check_write_begin);
361			goto error;
362		}
363		if (!folio)
364			goto retry;
365	}
366
367	if (folio_test_uptodate(folio))
368		goto have_folio;
369
370	/* If the page is beyond the EOF, we want to clear it - unless it's
371	 * within the cache granule containing the EOF, in which case we need
372	 * to preload the granule.
373	 */
374	if (!netfs_is_cache_enabled(ctx) &&
375	    netfs_skip_folio_read(folio, pos, len, false)) {
376		netfs_stat(&netfs_n_rh_write_zskip);
377		goto have_folio_no_wait;
378	}
379
380	rreq = netfs_alloc_request(mapping, file,
381				   folio_file_pos(folio), folio_size(folio),
382				   NETFS_READ_FOR_WRITE);
383	if (IS_ERR(rreq)) {
384		ret = PTR_ERR(rreq);
385		goto error;
386	}
387	rreq->no_unlock_folio	= folio_index(folio);
388	__set_bit(NETFS_RREQ_NO_UNLOCK_FOLIO, &rreq->flags);
389
390	if (ctx->ops->begin_cache_operation) {
391		ret = ctx->ops->begin_cache_operation(rreq);
392		if (ret == -ENOMEM || ret == -EINTR || ret == -ERESTARTSYS)
393			goto error_put;
394	}
395
396	netfs_stat(&netfs_n_rh_write_begin);
397	trace_netfs_read(rreq, pos, len, netfs_read_trace_write_begin);
398
399	/* Expand the request to meet caching requirements and download
400	 * preferences.
401	 */
402	ractl._nr_pages = folio_nr_pages(folio);
403	netfs_rreq_expand(rreq, &ractl);
404
405	/* We hold the folio locks, so we can drop the references */
406	folio_get(folio);
407	while (readahead_folio(&ractl))
408		;
409
410	ret = netfs_begin_read(rreq, true);
411	if (ret < 0)
412		goto error;
413
414have_folio:
415	ret = folio_wait_fscache_killable(folio);
416	if (ret < 0)
417		goto error;
418have_folio_no_wait:
419	*_folio = folio;
420	_leave(" = 0");
421	return 0;
422
423error_put:
424	netfs_put_request(rreq, false, netfs_rreq_trace_put_failed);
425error:
426	if (folio) {
427		folio_unlock(folio);
428		folio_put(folio);
429	}
430	_leave(" = %d", ret);
431	return ret;
432}
433EXPORT_SYMBOL(netfs_write_begin);