Commit | Line | Data |
---|---|---|
c38f4e96 DH |
1 | // SPDX-License-Identifier: GPL-2.0-only |
2 | /* Network filesystem high-level write support. | |
3 | * | |
4 | * Copyright (C) 2023 Red Hat, Inc. All Rights Reserved. | |
5 | * Written by David Howells (dhowells@redhat.com) | |
6 | */ | |
7 | ||
8 | #include <linux/export.h> | |
9 | #include <linux/fs.h> | |
10 | #include <linux/mm.h> | |
11 | #include <linux/pagemap.h> | |
12 | #include <linux/slab.h> | |
13 | #include <linux/pagevec.h> | |
14 | #include "internal.h" | |
15 | ||
16 | /* | |
17 | * Determined write method. Adjust netfs_folio_traces if this is changed. | |
18 | */ | |
19 | enum netfs_how_to_modify { | |
20 | NETFS_FOLIO_IS_UPTODATE, /* Folio is uptodate already */ | |
21 | NETFS_JUST_PREFETCH, /* We have to read the folio anyway */ | |
22 | NETFS_WHOLE_FOLIO_MODIFY, /* We're going to overwrite the whole folio */ | |
23 | NETFS_MODIFY_AND_CLEAR, /* We can assume there is no data to be downloaded. */ | |
24 | NETFS_STREAMING_WRITE, /* Store incomplete data in non-uptodate page. */ | |
25 | NETFS_STREAMING_WRITE_CONT, /* Continue streaming write. */ | |
26 | NETFS_FLUSH_CONTENT, /* Flush incompatible content. */ | |
27 | }; | |
28 | ||
41d8e767 DH |
29 | static void netfs_cleanup_buffered_write(struct netfs_io_request *wreq); |
30 | ||
c38f4e96 DH |
31 | static void netfs_set_group(struct folio *folio, struct netfs_group *netfs_group) |
32 | { | |
33 | if (netfs_group && !folio_get_private(folio)) | |
34 | folio_attach_private(folio, netfs_get_group(netfs_group)); | |
35 | } | |
36 | ||
62c3b748 DH |
37 | #if IS_ENABLED(CONFIG_FSCACHE) |
38 | static void netfs_folio_start_fscache(bool caching, struct folio *folio) | |
39 | { | |
40 | if (caching) | |
41 | folio_start_fscache(folio); | |
42 | } | |
43 | #else | |
44 | static void netfs_folio_start_fscache(bool caching, struct folio *folio) | |
45 | { | |
46 | } | |
47 | #endif | |
48 | ||
c38f4e96 DH |
49 | /* |
50 | * Decide how we should modify a folio. We might be attempting to do | |
51 | * write-streaming, in which case we don't want to a local RMW cycle if we can | |
52 | * avoid it. If we're doing local caching or content crypto, we award that | |
53 | * priority over avoiding RMW. If the file is open readably, then we also | |
54 | * assume that we may want to read what we wrote. | |
55 | */ | |
56 | static enum netfs_how_to_modify netfs_how_to_modify(struct netfs_inode *ctx, | |
57 | struct file *file, | |
58 | struct folio *folio, | |
59 | void *netfs_group, | |
60 | size_t flen, | |
61 | size_t offset, | |
62 | size_t len, | |
63 | bool maybe_trouble) | |
64 | { | |
65 | struct netfs_folio *finfo = netfs_folio_info(folio); | |
66 | loff_t pos = folio_file_pos(folio); | |
67 | ||
68 | _enter(""); | |
69 | ||
70 | if (netfs_folio_group(folio) != netfs_group) | |
71 | return NETFS_FLUSH_CONTENT; | |
72 | ||
73 | if (folio_test_uptodate(folio)) | |
74 | return NETFS_FOLIO_IS_UPTODATE; | |
75 | ||
100ccd18 | 76 | if (pos >= ctx->zero_point) |
c38f4e96 DH |
77 | return NETFS_MODIFY_AND_CLEAR; |
78 | ||
79 | if (!maybe_trouble && offset == 0 && len >= flen) | |
80 | return NETFS_WHOLE_FOLIO_MODIFY; | |
81 | ||
82 | if (file->f_mode & FMODE_READ) | |
92a714d7 DH |
83 | goto no_write_streaming; |
84 | if (test_bit(NETFS_ICTX_NO_WRITE_STREAMING, &ctx->flags)) | |
85 | goto no_write_streaming; | |
86 | ||
87 | if (netfs_is_cache_enabled(ctx)) { | |
88 | /* We don't want to get a streaming write on a file that loses | |
89 | * caching service temporarily because the backing store got | |
90 | * culled. | |
91 | */ | |
92 | if (!test_bit(NETFS_ICTX_NO_WRITE_STREAMING, &ctx->flags)) | |
93 | set_bit(NETFS_ICTX_NO_WRITE_STREAMING, &ctx->flags); | |
94 | goto no_write_streaming; | |
95 | } | |
c38f4e96 DH |
96 | |
97 | if (!finfo) | |
98 | return NETFS_STREAMING_WRITE; | |
99 | ||
100 | /* We can continue a streaming write only if it continues on from the | |
101 | * previous. If it overlaps, we must flush lest we suffer a partial | |
102 | * copy and disjoint dirty regions. | |
103 | */ | |
104 | if (offset == finfo->dirty_offset + finfo->dirty_len) | |
105 | return NETFS_STREAMING_WRITE_CONT; | |
106 | return NETFS_FLUSH_CONTENT; | |
92a714d7 DH |
107 | |
108 | no_write_streaming: | |
109 | if (finfo) { | |
110 | netfs_stat(&netfs_n_wh_wstream_conflict); | |
111 | return NETFS_FLUSH_CONTENT; | |
112 | } | |
113 | return NETFS_JUST_PREFETCH; | |
c38f4e96 DH |
114 | } |
115 | ||
116 | /* | |
e2e2e839 DH |
117 | * Grab a folio for writing and lock it. Attempt to allocate as large a folio |
118 | * as possible to hold as much of the remaining length as possible in one go. | |
c38f4e96 DH |
119 | */ |
120 | static struct folio *netfs_grab_folio_for_write(struct address_space *mapping, | |
121 | loff_t pos, size_t part) | |
122 | { | |
123 | pgoff_t index = pos / PAGE_SIZE; | |
e2e2e839 | 124 | fgf_t fgp_flags = FGP_WRITEBEGIN; |
c38f4e96 | 125 | |
e2e2e839 DH |
126 | if (mapping_large_folio_support(mapping)) |
127 | fgp_flags |= fgf_set_order(pos % PAGE_SIZE + part); | |
128 | ||
129 | return __filemap_get_folio(mapping, index, fgp_flags, | |
c38f4e96 DH |
130 | mapping_gfp_mask(mapping)); |
131 | } | |
132 | ||
5f24162f DH |
133 | /* |
134 | * Update i_size and estimate the update to i_blocks to reflect the additional | |
135 | * data written into the pagecache until we can find out from the server what | |
136 | * the values actually are. | |
137 | */ | |
138 | static void netfs_update_i_size(struct netfs_inode *ctx, struct inode *inode, | |
139 | loff_t i_size, loff_t pos, size_t copied) | |
140 | { | |
141 | blkcnt_t add; | |
142 | size_t gap; | |
143 | ||
144 | if (ctx->ops->update_i_size) { | |
145 | ctx->ops->update_i_size(inode, pos); | |
146 | return; | |
147 | } | |
148 | ||
149 | i_size_write(inode, pos); | |
150 | #if IS_ENABLED(CONFIG_FSCACHE) | |
151 | fscache_update_cookie(ctx->cache, NULL, &pos); | |
152 | #endif | |
153 | ||
154 | gap = SECTOR_SIZE - (i_size & (SECTOR_SIZE - 1)); | |
155 | if (copied > gap) { | |
156 | add = DIV_ROUND_UP(copied - gap, SECTOR_SIZE); | |
157 | ||
158 | inode->i_blocks = min_t(blkcnt_t, | |
159 | DIV_ROUND_UP(pos, SECTOR_SIZE), | |
160 | inode->i_blocks + add); | |
161 | } | |
162 | } | |
163 | ||
c38f4e96 DH |
164 | /** |
165 | * netfs_perform_write - Copy data into the pagecache. | |
166 | * @iocb: The operation parameters | |
167 | * @iter: The source buffer | |
168 | * @netfs_group: Grouping for dirty pages (eg. ceph snaps). | |
169 | * | |
170 | * Copy data into pagecache pages attached to the inode specified by @iocb. | |
171 | * The caller must hold appropriate inode locks. | |
172 | * | |
173 | * Dirty pages are tagged with a netfs_folio struct if they're not up to date | |
174 | * to indicate the range modified. Dirty pages may also be tagged with a | |
175 | * netfs-specific grouping such that data from an old group gets flushed before | |
176 | * a new one is started. | |
177 | */ | |
178 | ssize_t netfs_perform_write(struct kiocb *iocb, struct iov_iter *iter, | |
179 | struct netfs_group *netfs_group) | |
180 | { | |
181 | struct file *file = iocb->ki_filp; | |
182 | struct inode *inode = file_inode(file); | |
183 | struct address_space *mapping = inode->i_mapping; | |
184 | struct netfs_inode *ctx = netfs_inode(inode); | |
41d8e767 DH |
185 | struct writeback_control wbc = { |
186 | .sync_mode = WB_SYNC_NONE, | |
187 | .for_sync = true, | |
188 | .nr_to_write = LONG_MAX, | |
189 | .range_start = iocb->ki_pos, | |
190 | .range_end = iocb->ki_pos + iter->count, | |
191 | }; | |
192 | struct netfs_io_request *wreq = NULL; | |
c38f4e96 DH |
193 | struct netfs_folio *finfo; |
194 | struct folio *folio; | |
195 | enum netfs_how_to_modify howto; | |
196 | enum netfs_folio_trace trace; | |
197 | unsigned int bdp_flags = (iocb->ki_flags & IOCB_SYNC) ? 0: BDP_ASYNC; | |
619606a7 | 198 | ssize_t written = 0, ret, ret2; |
c38f4e96 DH |
199 | loff_t i_size, pos = iocb->ki_pos, from, to; |
200 | size_t max_chunk = PAGE_SIZE << MAX_PAGECACHE_ORDER; | |
201 | bool maybe_trouble = false; | |
202 | ||
41d8e767 DH |
203 | if (unlikely(test_bit(NETFS_ICTX_WRITETHROUGH, &ctx->flags) || |
204 | iocb->ki_flags & (IOCB_DSYNC | IOCB_SYNC)) | |
205 | ) { | |
41d8e767 DH |
206 | wbc_attach_fdatawrite_inode(&wbc, mapping->host); |
207 | ||
c97f59e2 DH |
208 | ret = filemap_write_and_wait_range(mapping, pos, pos + iter->count); |
209 | if (ret < 0) { | |
210 | wbc_detach_inode(&wbc); | |
211 | goto out; | |
212 | } | |
213 | ||
41d8e767 DH |
214 | wreq = netfs_begin_writethrough(iocb, iter->count); |
215 | if (IS_ERR(wreq)) { | |
216 | wbc_detach_inode(&wbc); | |
217 | ret = PTR_ERR(wreq); | |
218 | wreq = NULL; | |
219 | goto out; | |
220 | } | |
221 | if (!is_sync_kiocb(iocb)) | |
222 | wreq->iocb = iocb; | |
223 | wreq->cleanup = netfs_cleanup_buffered_write; | |
224 | } | |
225 | ||
c38f4e96 DH |
226 | do { |
227 | size_t flen; | |
228 | size_t offset; /* Offset into pagecache folio */ | |
229 | size_t part; /* Bytes to write to folio */ | |
230 | size_t copied; /* Bytes copied from user */ | |
231 | ||
232 | ret = balance_dirty_pages_ratelimited_flags(mapping, bdp_flags); | |
233 | if (unlikely(ret < 0)) | |
234 | break; | |
235 | ||
236 | offset = pos & (max_chunk - 1); | |
237 | part = min(max_chunk - offset, iov_iter_count(iter)); | |
238 | ||
239 | /* Bring in the user pages that we will copy from _first_ lest | |
240 | * we hit a nasty deadlock on copying from the same page as | |
241 | * we're writing to, without it being marked uptodate. | |
242 | * | |
243 | * Not only is this an optimisation, but it is also required to | |
244 | * check that the address is actually valid, when atomic | |
245 | * usercopies are used below. | |
246 | * | |
247 | * We rely on the page being held onto long enough by the LRU | |
248 | * that we can grab it below if this causes it to be read. | |
249 | */ | |
250 | ret = -EFAULT; | |
251 | if (unlikely(fault_in_iov_iter_readable(iter, part) == part)) | |
252 | break; | |
253 | ||
c38f4e96 | 254 | folio = netfs_grab_folio_for_write(mapping, pos, part); |
843609df DC |
255 | if (IS_ERR(folio)) { |
256 | ret = PTR_ERR(folio); | |
c38f4e96 | 257 | break; |
843609df | 258 | } |
c38f4e96 DH |
259 | |
260 | flen = folio_size(folio); | |
261 | offset = pos & (flen - 1); | |
262 | part = min_t(size_t, flen - offset, part); | |
263 | ||
264 | if (signal_pending(current)) { | |
265 | ret = written ? -EINTR : -ERESTARTSYS; | |
266 | goto error_folio_unlock; | |
267 | } | |
268 | ||
269 | /* See if we need to prefetch the area we're going to modify. | |
270 | * We need to do this before we get a lock on the folio in case | |
271 | * there's more than one writer competing for the same cache | |
272 | * block. | |
273 | */ | |
274 | howto = netfs_how_to_modify(ctx, file, folio, netfs_group, | |
275 | flen, offset, part, maybe_trouble); | |
276 | _debug("howto %u", howto); | |
277 | switch (howto) { | |
278 | case NETFS_JUST_PREFETCH: | |
279 | ret = netfs_prefetch_for_write(file, folio, offset, part); | |
280 | if (ret < 0) { | |
281 | _debug("prefetch = %zd", ret); | |
282 | goto error_folio_unlock; | |
283 | } | |
284 | break; | |
285 | case NETFS_FOLIO_IS_UPTODATE: | |
286 | case NETFS_WHOLE_FOLIO_MODIFY: | |
287 | case NETFS_STREAMING_WRITE_CONT: | |
288 | break; | |
289 | case NETFS_MODIFY_AND_CLEAR: | |
290 | zero_user_segment(&folio->page, 0, offset); | |
291 | break; | |
292 | case NETFS_STREAMING_WRITE: | |
293 | ret = -EIO; | |
294 | if (WARN_ON(folio_get_private(folio))) | |
295 | goto error_folio_unlock; | |
296 | break; | |
297 | case NETFS_FLUSH_CONTENT: | |
298 | trace_netfs_folio(folio, netfs_flush_content); | |
299 | from = folio_pos(folio); | |
300 | to = from + folio_size(folio) - 1; | |
301 | folio_unlock(folio); | |
302 | folio_put(folio); | |
303 | ret = filemap_write_and_wait_range(mapping, from, to); | |
304 | if (ret < 0) | |
305 | goto error_folio_unlock; | |
306 | continue; | |
307 | } | |
308 | ||
309 | if (mapping_writably_mapped(mapping)) | |
310 | flush_dcache_folio(folio); | |
311 | ||
312 | copied = copy_folio_from_iter_atomic(folio, offset, part, iter); | |
313 | ||
314 | flush_dcache_folio(folio); | |
315 | ||
316 | /* Deal with a (partially) failed copy */ | |
317 | if (copied == 0) { | |
318 | ret = -EFAULT; | |
319 | goto error_folio_unlock; | |
320 | } | |
321 | ||
322 | trace = (enum netfs_folio_trace)howto; | |
323 | switch (howto) { | |
324 | case NETFS_FOLIO_IS_UPTODATE: | |
325 | case NETFS_JUST_PREFETCH: | |
326 | netfs_set_group(folio, netfs_group); | |
327 | break; | |
328 | case NETFS_MODIFY_AND_CLEAR: | |
329 | zero_user_segment(&folio->page, offset + copied, flen); | |
330 | netfs_set_group(folio, netfs_group); | |
331 | folio_mark_uptodate(folio); | |
332 | break; | |
333 | case NETFS_WHOLE_FOLIO_MODIFY: | |
334 | if (unlikely(copied < part)) { | |
335 | maybe_trouble = true; | |
336 | iov_iter_revert(iter, copied); | |
337 | copied = 0; | |
338 | goto retry; | |
339 | } | |
340 | netfs_set_group(folio, netfs_group); | |
341 | folio_mark_uptodate(folio); | |
342 | break; | |
343 | case NETFS_STREAMING_WRITE: | |
344 | if (offset == 0 && copied == flen) { | |
345 | netfs_set_group(folio, netfs_group); | |
346 | folio_mark_uptodate(folio); | |
347 | trace = netfs_streaming_filled_page; | |
348 | break; | |
349 | } | |
350 | finfo = kzalloc(sizeof(*finfo), GFP_KERNEL); | |
351 | if (!finfo) { | |
352 | iov_iter_revert(iter, copied); | |
353 | ret = -ENOMEM; | |
354 | goto error_folio_unlock; | |
355 | } | |
356 | finfo->netfs_group = netfs_get_group(netfs_group); | |
357 | finfo->dirty_offset = offset; | |
358 | finfo->dirty_len = copied; | |
359 | folio_attach_private(folio, (void *)((unsigned long)finfo | | |
360 | NETFS_FOLIO_INFO)); | |
361 | break; | |
362 | case NETFS_STREAMING_WRITE_CONT: | |
363 | finfo = netfs_folio_info(folio); | |
364 | finfo->dirty_len += copied; | |
365 | if (finfo->dirty_offset == 0 && finfo->dirty_len == flen) { | |
366 | if (finfo->netfs_group) | |
367 | folio_change_private(folio, finfo->netfs_group); | |
368 | else | |
369 | folio_detach_private(folio); | |
370 | folio_mark_uptodate(folio); | |
371 | kfree(finfo); | |
372 | trace = netfs_streaming_cont_filled_page; | |
373 | } | |
374 | break; | |
375 | default: | |
376 | WARN(true, "Unexpected modify type %u ix=%lx\n", | |
202bc57b | 377 | howto, folio->index); |
c38f4e96 DH |
378 | ret = -EIO; |
379 | goto error_folio_unlock; | |
380 | } | |
381 | ||
382 | trace_netfs_folio(folio, trace); | |
383 | ||
384 | /* Update the inode size if we moved the EOF marker */ | |
c38f4e96 | 385 | pos += copied; |
5f24162f DH |
386 | i_size = i_size_read(inode); |
387 | if (pos > i_size) | |
388 | netfs_update_i_size(ctx, inode, i_size, pos, copied); | |
c38f4e96 DH |
389 | written += copied; |
390 | ||
41d8e767 DH |
391 | if (likely(!wreq)) { |
392 | folio_mark_dirty(folio); | |
393 | } else { | |
394 | if (folio_test_dirty(folio)) | |
395 | /* Sigh. mmap. */ | |
396 | folio_clear_dirty_for_io(folio); | |
397 | /* We make multiple writes to the folio... */ | |
398 | if (!folio_test_writeback(folio)) { | |
399 | folio_wait_fscache(folio); | |
400 | folio_start_writeback(folio); | |
401 | folio_start_fscache(folio); | |
402 | if (wreq->iter.count == 0) | |
403 | trace_netfs_folio(folio, netfs_folio_trace_wthru); | |
404 | else | |
405 | trace_netfs_folio(folio, netfs_folio_trace_wthru_plus); | |
406 | } | |
407 | netfs_advance_writethrough(wreq, copied, | |
408 | offset + copied == flen); | |
409 | } | |
c38f4e96 DH |
410 | retry: |
411 | folio_unlock(folio); | |
412 | folio_put(folio); | |
413 | folio = NULL; | |
414 | ||
415 | cond_resched(); | |
416 | } while (iov_iter_count(iter)); | |
417 | ||
418 | out: | |
41d8e767 | 419 | if (unlikely(wreq)) { |
619606a7 | 420 | ret2 = netfs_end_writethrough(wreq, iocb); |
41d8e767 | 421 | wbc_detach_inode(&wbc); |
619606a7 DH |
422 | if (ret2 == -EIOCBQUEUED) |
423 | return ret2; | |
424 | if (ret == 0) | |
425 | ret = ret2; | |
c38f4e96 DH |
426 | } |
427 | ||
41d8e767 | 428 | iocb->ki_pos += written; |
c38f4e96 DH |
429 | _leave(" = %zd [%zd]", written, ret); |
430 | return written ? written : ret; | |
431 | ||
432 | error_folio_unlock: | |
433 | folio_unlock(folio); | |
434 | folio_put(folio); | |
435 | goto out; | |
436 | } | |
437 | EXPORT_SYMBOL(netfs_perform_write); | |
938e13a7 DH |
438 | |
439 | /** | |
440 | * netfs_buffered_write_iter_locked - write data to a file | |
441 | * @iocb: IO state structure (file, offset, etc.) | |
442 | * @from: iov_iter with data to write | |
443 | * @netfs_group: Grouping for dirty pages (eg. ceph snaps). | |
444 | * | |
445 | * This function does all the work needed for actually writing data to a | |
446 | * file. It does all basic checks, removes SUID from the file, updates | |
447 | * modification times and calls proper subroutines depending on whether we | |
448 | * do direct IO or a standard buffered write. | |
449 | * | |
450 | * The caller must hold appropriate locks around this function and have called | |
451 | * generic_write_checks() already. The caller is also responsible for doing | |
452 | * any necessary syncing afterwards. | |
453 | * | |
454 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
455 | * A caller has to handle it. This is mainly due to the fact that we want to | |
456 | * avoid syncing under i_rwsem. | |
457 | * | |
458 | * Return: | |
459 | * * number of bytes written, even for truncated writes | |
460 | * * negative error code if no data has been written at all | |
461 | */ | |
462 | ssize_t netfs_buffered_write_iter_locked(struct kiocb *iocb, struct iov_iter *from, | |
463 | struct netfs_group *netfs_group) | |
464 | { | |
465 | struct file *file = iocb->ki_filp; | |
466 | ssize_t ret; | |
467 | ||
468 | trace_netfs_write_iter(iocb, from); | |
469 | ||
470 | ret = file_remove_privs(file); | |
471 | if (ret) | |
472 | return ret; | |
473 | ||
474 | ret = file_update_time(file); | |
475 | if (ret) | |
476 | return ret; | |
477 | ||
478 | return netfs_perform_write(iocb, from, netfs_group); | |
479 | } | |
480 | EXPORT_SYMBOL(netfs_buffered_write_iter_locked); | |
481 | ||
482 | /** | |
483 | * netfs_file_write_iter - write data to a file | |
484 | * @iocb: IO state structure | |
485 | * @from: iov_iter with data to write | |
486 | * | |
487 | * Perform a write to a file, writing into the pagecache if possible and doing | |
488 | * an unbuffered write instead if not. | |
489 | * | |
490 | * Return: | |
491 | * * Negative error code if no data has been written at all of | |
492 | * vfs_fsync_range() failed for a synchronous write | |
493 | * * Number of bytes written, even for truncated writes | |
494 | */ | |
495 | ssize_t netfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from) | |
496 | { | |
497 | struct file *file = iocb->ki_filp; | |
498 | struct inode *inode = file->f_mapping->host; | |
499 | struct netfs_inode *ictx = netfs_inode(inode); | |
500 | ssize_t ret; | |
501 | ||
502 | _enter("%llx,%zx,%llx", iocb->ki_pos, iov_iter_count(from), i_size_read(inode)); | |
503 | ||
ca9ca1a5 DH |
504 | if (!iov_iter_count(from)) |
505 | return 0; | |
506 | ||
938e13a7 DH |
507 | if ((iocb->ki_flags & IOCB_DIRECT) || |
508 | test_bit(NETFS_ICTX_UNBUFFERED, &ictx->flags)) | |
509 | return netfs_unbuffered_write_iter(iocb, from); | |
510 | ||
511 | ret = netfs_start_io_write(inode); | |
512 | if (ret < 0) | |
513 | return ret; | |
514 | ||
515 | ret = generic_write_checks(iocb, from); | |
516 | if (ret > 0) | |
517 | ret = netfs_buffered_write_iter_locked(iocb, from, NULL); | |
518 | netfs_end_io_write(inode); | |
519 | if (ret > 0) | |
520 | ret = generic_write_sync(iocb, ret); | |
521 | return ret; | |
522 | } | |
523 | EXPORT_SYMBOL(netfs_file_write_iter); | |
102a7e2c DH |
524 | |
525 | /* | |
526 | * Notification that a previously read-only page is about to become writable. | |
527 | * Note that the caller indicates a single page of a multipage folio. | |
528 | */ | |
529 | vm_fault_t netfs_page_mkwrite(struct vm_fault *vmf, struct netfs_group *netfs_group) | |
530 | { | |
531 | struct folio *folio = page_folio(vmf->page); | |
532 | struct file *file = vmf->vma->vm_file; | |
533 | struct inode *inode = file_inode(file); | |
534 | vm_fault_t ret = VM_FAULT_RETRY; | |
535 | int err; | |
536 | ||
537 | _enter("%lx", folio->index); | |
538 | ||
539 | sb_start_pagefault(inode->i_sb); | |
540 | ||
541 | if (folio_wait_writeback_killable(folio)) | |
542 | goto out; | |
543 | ||
544 | if (folio_lock_killable(folio) < 0) | |
545 | goto out; | |
546 | ||
547 | /* Can we see a streaming write here? */ | |
548 | if (WARN_ON(!folio_test_uptodate(folio))) { | |
549 | ret = VM_FAULT_SIGBUS | VM_FAULT_LOCKED; | |
550 | goto out; | |
551 | } | |
552 | ||
553 | if (netfs_folio_group(folio) != netfs_group) { | |
554 | folio_unlock(folio); | |
555 | err = filemap_fdatawait_range(inode->i_mapping, | |
556 | folio_pos(folio), | |
557 | folio_pos(folio) + folio_size(folio)); | |
558 | switch (err) { | |
559 | case 0: | |
560 | ret = VM_FAULT_RETRY; | |
561 | goto out; | |
562 | case -ENOMEM: | |
563 | ret = VM_FAULT_OOM; | |
564 | goto out; | |
565 | default: | |
566 | ret = VM_FAULT_SIGBUS; | |
567 | goto out; | |
568 | } | |
569 | } | |
570 | ||
571 | if (folio_test_dirty(folio)) | |
572 | trace_netfs_folio(folio, netfs_folio_trace_mkwrite_plus); | |
573 | else | |
574 | trace_netfs_folio(folio, netfs_folio_trace_mkwrite); | |
575 | netfs_set_group(folio, netfs_group); | |
576 | file_update_time(file); | |
577 | ret = VM_FAULT_LOCKED; | |
578 | out: | |
579 | sb_end_pagefault(inode->i_sb); | |
580 | return ret; | |
581 | } | |
582 | EXPORT_SYMBOL(netfs_page_mkwrite); | |
62c3b748 DH |
583 | |
584 | /* | |
585 | * Kill all the pages in the given range | |
586 | */ | |
587 | static void netfs_kill_pages(struct address_space *mapping, | |
588 | loff_t start, loff_t len) | |
589 | { | |
590 | struct folio *folio; | |
591 | pgoff_t index = start / PAGE_SIZE; | |
592 | pgoff_t last = (start + len - 1) / PAGE_SIZE, next; | |
593 | ||
594 | _enter("%llx-%llx", start, start + len - 1); | |
595 | ||
596 | do { | |
597 | _debug("kill %lx (to %lx)", index, last); | |
598 | ||
599 | folio = filemap_get_folio(mapping, index); | |
600 | if (IS_ERR(folio)) { | |
601 | next = index + 1; | |
602 | continue; | |
603 | } | |
604 | ||
605 | next = folio_next_index(folio); | |
606 | ||
607 | trace_netfs_folio(folio, netfs_folio_trace_kill); | |
608 | folio_clear_uptodate(folio); | |
609 | if (folio_test_fscache(folio)) | |
610 | folio_end_fscache(folio); | |
611 | folio_end_writeback(folio); | |
612 | folio_lock(folio); | |
16df6e07 | 613 | generic_error_remove_folio(mapping, folio); |
62c3b748 DH |
614 | folio_unlock(folio); |
615 | folio_put(folio); | |
616 | ||
617 | } while (index = next, index <= last); | |
618 | ||
619 | _leave(""); | |
620 | } | |
621 | ||
622 | /* | |
623 | * Redirty all the pages in a given range. | |
624 | */ | |
625 | static void netfs_redirty_pages(struct address_space *mapping, | |
626 | loff_t start, loff_t len) | |
627 | { | |
628 | struct folio *folio; | |
629 | pgoff_t index = start / PAGE_SIZE; | |
630 | pgoff_t last = (start + len - 1) / PAGE_SIZE, next; | |
631 | ||
632 | _enter("%llx-%llx", start, start + len - 1); | |
633 | ||
634 | do { | |
635 | _debug("redirty %llx @%llx", len, start); | |
636 | ||
637 | folio = filemap_get_folio(mapping, index); | |
638 | if (IS_ERR(folio)) { | |
639 | next = index + 1; | |
640 | continue; | |
641 | } | |
642 | ||
643 | next = folio_next_index(folio); | |
644 | trace_netfs_folio(folio, netfs_folio_trace_redirty); | |
645 | filemap_dirty_folio(mapping, folio); | |
646 | if (folio_test_fscache(folio)) | |
647 | folio_end_fscache(folio); | |
648 | folio_end_writeback(folio); | |
649 | folio_put(folio); | |
650 | } while (index = next, index <= last); | |
651 | ||
652 | balance_dirty_pages_ratelimited(mapping); | |
653 | ||
654 | _leave(""); | |
655 | } | |
656 | ||
657 | /* | |
658 | * Completion of write to server | |
659 | */ | |
660 | static void netfs_pages_written_back(struct netfs_io_request *wreq) | |
661 | { | |
662 | struct address_space *mapping = wreq->mapping; | |
663 | struct netfs_folio *finfo; | |
664 | struct netfs_group *group = NULL; | |
665 | struct folio *folio; | |
666 | pgoff_t last; | |
667 | int gcount = 0; | |
668 | ||
669 | XA_STATE(xas, &mapping->i_pages, wreq->start / PAGE_SIZE); | |
670 | ||
671 | _enter("%llx-%llx", wreq->start, wreq->start + wreq->len); | |
672 | ||
673 | rcu_read_lock(); | |
674 | ||
675 | last = (wreq->start + wreq->len - 1) / PAGE_SIZE; | |
676 | xas_for_each(&xas, folio, last) { | |
677 | WARN(!folio_test_writeback(folio), | |
678 | "bad %zx @%llx page %lx %lx\n", | |
202bc57b | 679 | wreq->len, wreq->start, folio->index, last); |
62c3b748 DH |
680 | |
681 | if ((finfo = netfs_folio_info(folio))) { | |
682 | /* Streaming writes cannot be redirtied whilst under | |
683 | * writeback, so discard the streaming record. | |
684 | */ | |
685 | folio_detach_private(folio); | |
686 | group = finfo->netfs_group; | |
687 | gcount++; | |
688 | trace_netfs_folio(folio, netfs_folio_trace_clear_s); | |
689 | kfree(finfo); | |
690 | } else if ((group = netfs_folio_group(folio))) { | |
691 | /* Need to detach the group pointer if the page didn't | |
692 | * get redirtied. If it has been redirtied, then it | |
693 | * must be within the same group. | |
694 | */ | |
695 | if (folio_test_dirty(folio)) { | |
696 | trace_netfs_folio(folio, netfs_folio_trace_redirtied); | |
697 | goto end_wb; | |
698 | } | |
699 | if (folio_trylock(folio)) { | |
700 | if (!folio_test_dirty(folio)) { | |
701 | folio_detach_private(folio); | |
702 | gcount++; | |
703 | trace_netfs_folio(folio, netfs_folio_trace_clear_g); | |
704 | } else { | |
705 | trace_netfs_folio(folio, netfs_folio_trace_redirtied); | |
706 | } | |
707 | folio_unlock(folio); | |
708 | goto end_wb; | |
709 | } | |
710 | ||
711 | xas_pause(&xas); | |
712 | rcu_read_unlock(); | |
713 | folio_lock(folio); | |
714 | if (!folio_test_dirty(folio)) { | |
715 | folio_detach_private(folio); | |
716 | gcount++; | |
717 | trace_netfs_folio(folio, netfs_folio_trace_clear_g); | |
718 | } else { | |
719 | trace_netfs_folio(folio, netfs_folio_trace_redirtied); | |
720 | } | |
721 | folio_unlock(folio); | |
722 | rcu_read_lock(); | |
723 | } else { | |
724 | trace_netfs_folio(folio, netfs_folio_trace_clear); | |
725 | } | |
726 | end_wb: | |
727 | if (folio_test_fscache(folio)) | |
728 | folio_end_fscache(folio); | |
807c6d09 | 729 | xas_advance(&xas, folio_next_index(folio) - 1); |
62c3b748 DH |
730 | folio_end_writeback(folio); |
731 | } | |
732 | ||
733 | rcu_read_unlock(); | |
734 | netfs_put_group_many(group, gcount); | |
735 | _leave(""); | |
736 | } | |
737 | ||
738 | /* | |
739 | * Deal with the disposition of the folios that are under writeback to close | |
740 | * out the operation. | |
741 | */ | |
742 | static void netfs_cleanup_buffered_write(struct netfs_io_request *wreq) | |
743 | { | |
744 | struct address_space *mapping = wreq->mapping; | |
745 | ||
746 | _enter(""); | |
747 | ||
748 | switch (wreq->error) { | |
749 | case 0: | |
750 | netfs_pages_written_back(wreq); | |
751 | break; | |
752 | ||
753 | default: | |
754 | pr_notice("R=%08x Unexpected error %d\n", wreq->debug_id, wreq->error); | |
755 | fallthrough; | |
756 | case -EACCES: | |
757 | case -EPERM: | |
758 | case -ENOKEY: | |
759 | case -EKEYEXPIRED: | |
760 | case -EKEYREJECTED: | |
761 | case -EKEYREVOKED: | |
762 | case -ENETRESET: | |
763 | case -EDQUOT: | |
764 | case -ENOSPC: | |
765 | netfs_redirty_pages(mapping, wreq->start, wreq->len); | |
766 | break; | |
767 | ||
768 | case -EROFS: | |
769 | case -EIO: | |
770 | case -EREMOTEIO: | |
771 | case -EFBIG: | |
772 | case -ENOENT: | |
773 | case -ENOMEDIUM: | |
774 | case -ENXIO: | |
775 | netfs_kill_pages(mapping, wreq->start, wreq->len); | |
776 | break; | |
777 | } | |
778 | ||
779 | if (wreq->error) | |
780 | mapping_set_error(mapping, wreq->error); | |
781 | if (wreq->netfs_ops->done) | |
782 | wreq->netfs_ops->done(wreq); | |
783 | } | |
784 | ||
785 | /* | |
786 | * Extend the region to be written back to include subsequent contiguously | |
787 | * dirty pages if possible, but don't sleep while doing so. | |
788 | * | |
789 | * If this page holds new content, then we can include filler zeros in the | |
790 | * writeback. | |
791 | */ | |
792 | static void netfs_extend_writeback(struct address_space *mapping, | |
793 | struct netfs_group *group, | |
794 | struct xa_state *xas, | |
795 | long *_count, | |
796 | loff_t start, | |
797 | loff_t max_len, | |
798 | bool caching, | |
799 | size_t *_len, | |
800 | size_t *_top) | |
801 | { | |
802 | struct netfs_folio *finfo; | |
803 | struct folio_batch fbatch; | |
804 | struct folio *folio; | |
805 | unsigned int i; | |
806 | pgoff_t index = (start + *_len) / PAGE_SIZE; | |
807 | size_t len; | |
808 | void *priv; | |
809 | bool stop = true; | |
810 | ||
811 | folio_batch_init(&fbatch); | |
812 | ||
813 | do { | |
814 | /* Firstly, we gather up a batch of contiguous dirty pages | |
815 | * under the RCU read lock - but we can't clear the dirty flags | |
816 | * there if any of those pages are mapped. | |
817 | */ | |
818 | rcu_read_lock(); | |
819 | ||
820 | xas_for_each(xas, folio, ULONG_MAX) { | |
821 | stop = true; | |
822 | if (xas_retry(xas, folio)) | |
823 | continue; | |
824 | if (xa_is_value(folio)) | |
825 | break; | |
202bc57b | 826 | if (folio->index != index) { |
62c3b748 DH |
827 | xas_reset(xas); |
828 | break; | |
829 | } | |
830 | ||
831 | if (!folio_try_get_rcu(folio)) { | |
832 | xas_reset(xas); | |
833 | continue; | |
834 | } | |
835 | ||
836 | /* Has the folio moved or been split? */ | |
837 | if (unlikely(folio != xas_reload(xas))) { | |
838 | folio_put(folio); | |
839 | xas_reset(xas); | |
840 | break; | |
841 | } | |
842 | ||
843 | if (!folio_trylock(folio)) { | |
844 | folio_put(folio); | |
845 | xas_reset(xas); | |
846 | break; | |
847 | } | |
848 | if (!folio_test_dirty(folio) || | |
849 | folio_test_writeback(folio) || | |
850 | folio_test_fscache(folio)) { | |
851 | folio_unlock(folio); | |
852 | folio_put(folio); | |
853 | xas_reset(xas); | |
854 | break; | |
855 | } | |
856 | ||
857 | stop = false; | |
858 | len = folio_size(folio); | |
859 | priv = folio_get_private(folio); | |
860 | if ((const struct netfs_group *)priv != group) { | |
861 | stop = true; | |
862 | finfo = netfs_folio_info(folio); | |
863 | if (finfo->netfs_group != group || | |
864 | finfo->dirty_offset > 0) { | |
865 | folio_unlock(folio); | |
866 | folio_put(folio); | |
867 | xas_reset(xas); | |
868 | break; | |
869 | } | |
870 | len = finfo->dirty_len; | |
871 | } | |
872 | ||
873 | *_top += folio_size(folio); | |
874 | index += folio_nr_pages(folio); | |
875 | *_count -= folio_nr_pages(folio); | |
876 | *_len += len; | |
877 | if (*_len >= max_len || *_count <= 0) | |
878 | stop = true; | |
879 | ||
880 | if (!folio_batch_add(&fbatch, folio)) | |
881 | break; | |
882 | if (stop) | |
883 | break; | |
884 | } | |
885 | ||
886 | xas_pause(xas); | |
887 | rcu_read_unlock(); | |
888 | ||
889 | /* Now, if we obtained any folios, we can shift them to being | |
890 | * writable and mark them for caching. | |
891 | */ | |
892 | if (!folio_batch_count(&fbatch)) | |
893 | break; | |
894 | ||
895 | for (i = 0; i < folio_batch_count(&fbatch); i++) { | |
896 | folio = fbatch.folios[i]; | |
897 | trace_netfs_folio(folio, netfs_folio_trace_store_plus); | |
898 | ||
899 | if (!folio_clear_dirty_for_io(folio)) | |
900 | BUG(); | |
901 | folio_start_writeback(folio); | |
902 | netfs_folio_start_fscache(caching, folio); | |
903 | folio_unlock(folio); | |
904 | } | |
905 | ||
906 | folio_batch_release(&fbatch); | |
907 | cond_resched(); | |
908 | } while (!stop); | |
909 | } | |
910 | ||
911 | /* | |
912 | * Synchronously write back the locked page and any subsequent non-locked dirty | |
913 | * pages. | |
914 | */ | |
915 | static ssize_t netfs_write_back_from_locked_folio(struct address_space *mapping, | |
916 | struct writeback_control *wbc, | |
917 | struct netfs_group *group, | |
918 | struct xa_state *xas, | |
919 | struct folio *folio, | |
920 | unsigned long long start, | |
921 | unsigned long long end) | |
922 | { | |
923 | struct netfs_io_request *wreq; | |
924 | struct netfs_folio *finfo; | |
925 | struct netfs_inode *ctx = netfs_inode(mapping->host); | |
926 | unsigned long long i_size = i_size_read(&ctx->inode); | |
927 | size_t len, max_len; | |
928 | bool caching = netfs_is_cache_enabled(ctx); | |
929 | long count = wbc->nr_to_write; | |
930 | int ret; | |
931 | ||
202bc57b | 932 | _enter(",%lx,%llx-%llx,%u", folio->index, start, end, caching); |
62c3b748 DH |
933 | |
934 | wreq = netfs_alloc_request(mapping, NULL, start, folio_size(folio), | |
935 | NETFS_WRITEBACK); | |
936 | if (IS_ERR(wreq)) { | |
937 | folio_unlock(folio); | |
938 | return PTR_ERR(wreq); | |
939 | } | |
940 | ||
941 | if (!folio_clear_dirty_for_io(folio)) | |
942 | BUG(); | |
943 | folio_start_writeback(folio); | |
944 | netfs_folio_start_fscache(caching, folio); | |
945 | ||
946 | count -= folio_nr_pages(folio); | |
947 | ||
948 | /* Find all consecutive lockable dirty pages that have contiguous | |
949 | * written regions, stopping when we find a page that is not | |
950 | * immediately lockable, is not dirty or is missing, or we reach the | |
951 | * end of the range. | |
952 | */ | |
953 | trace_netfs_folio(folio, netfs_folio_trace_store); | |
954 | ||
955 | len = wreq->len; | |
956 | finfo = netfs_folio_info(folio); | |
957 | if (finfo) { | |
958 | start += finfo->dirty_offset; | |
959 | if (finfo->dirty_offset + finfo->dirty_len != len) { | |
960 | len = finfo->dirty_len; | |
961 | goto cant_expand; | |
962 | } | |
963 | len = finfo->dirty_len; | |
964 | } | |
965 | ||
966 | if (start < i_size) { | |
967 | /* Trim the write to the EOF; the extra data is ignored. Also | |
968 | * put an upper limit on the size of a single storedata op. | |
969 | */ | |
970 | max_len = 65536 * 4096; | |
971 | max_len = min_t(unsigned long long, max_len, end - start + 1); | |
972 | max_len = min_t(unsigned long long, max_len, i_size - start); | |
973 | ||
974 | if (len < max_len) | |
975 | netfs_extend_writeback(mapping, group, xas, &count, start, | |
976 | max_len, caching, &len, &wreq->upper_len); | |
977 | } | |
978 | ||
979 | cant_expand: | |
980 | len = min_t(unsigned long long, len, i_size - start); | |
981 | ||
982 | /* We now have a contiguous set of dirty pages, each with writeback | |
983 | * set; the first page is still locked at this point, but all the rest | |
984 | * have been unlocked. | |
985 | */ | |
986 | folio_unlock(folio); | |
987 | wreq->start = start; | |
988 | wreq->len = len; | |
989 | ||
990 | if (start < i_size) { | |
991 | _debug("write back %zx @%llx [%llx]", len, start, i_size); | |
992 | ||
993 | /* Speculatively write to the cache. We have to fix this up | |
994 | * later if the store fails. | |
995 | */ | |
996 | wreq->cleanup = netfs_cleanup_buffered_write; | |
997 | ||
998 | iov_iter_xarray(&wreq->iter, ITER_SOURCE, &mapping->i_pages, start, | |
999 | wreq->upper_len); | |
1000 | __set_bit(NETFS_RREQ_UPLOAD_TO_SERVER, &wreq->flags); | |
1001 | ret = netfs_begin_write(wreq, true, netfs_write_trace_writeback); | |
1002 | if (ret == 0 || ret == -EIOCBQUEUED) | |
1003 | wbc->nr_to_write -= len / PAGE_SIZE; | |
1004 | } else { | |
1005 | _debug("write discard %zx @%llx [%llx]", len, start, i_size); | |
1006 | ||
1007 | /* The dirty region was entirely beyond the EOF. */ | |
1008 | fscache_clear_page_bits(mapping, start, len, caching); | |
1009 | netfs_pages_written_back(wreq); | |
1010 | ret = 0; | |
1011 | } | |
1012 | ||
1013 | netfs_put_request(wreq, false, netfs_rreq_trace_put_return); | |
1014 | _leave(" = 1"); | |
1015 | return 1; | |
1016 | } | |
1017 | ||
1018 | /* | |
1019 | * Write a region of pages back to the server | |
1020 | */ | |
1021 | static ssize_t netfs_writepages_begin(struct address_space *mapping, | |
1022 | struct writeback_control *wbc, | |
1023 | struct netfs_group *group, | |
1024 | struct xa_state *xas, | |
1025 | unsigned long long *_start, | |
1026 | unsigned long long end) | |
1027 | { | |
1028 | const struct netfs_folio *finfo; | |
1029 | struct folio *folio; | |
1030 | unsigned long long start = *_start; | |
1031 | ssize_t ret; | |
1032 | void *priv; | |
1033 | int skips = 0; | |
1034 | ||
1035 | _enter("%llx,%llx,", start, end); | |
1036 | ||
1037 | search_again: | |
1038 | /* Find the first dirty page in the group. */ | |
1039 | rcu_read_lock(); | |
1040 | ||
1041 | for (;;) { | |
1042 | folio = xas_find_marked(xas, end / PAGE_SIZE, PAGECACHE_TAG_DIRTY); | |
1043 | if (xas_retry(xas, folio) || xa_is_value(folio)) | |
1044 | continue; | |
1045 | if (!folio) | |
1046 | break; | |
1047 | ||
1048 | if (!folio_try_get_rcu(folio)) { | |
1049 | xas_reset(xas); | |
1050 | continue; | |
1051 | } | |
1052 | ||
1053 | if (unlikely(folio != xas_reload(xas))) { | |
1054 | folio_put(folio); | |
1055 | xas_reset(xas); | |
1056 | continue; | |
1057 | } | |
1058 | ||
1059 | /* Skip any dirty folio that's not in the group of interest. */ | |
1060 | priv = folio_get_private(folio); | |
1061 | if ((const struct netfs_group *)priv != group) { | |
1062 | finfo = netfs_folio_info(folio); | |
1063 | if (finfo->netfs_group != group) { | |
1064 | folio_put(folio); | |
1065 | continue; | |
1066 | } | |
1067 | } | |
1068 | ||
1069 | xas_pause(xas); | |
1070 | break; | |
1071 | } | |
1072 | rcu_read_unlock(); | |
1073 | if (!folio) | |
1074 | return 0; | |
1075 | ||
1076 | start = folio_pos(folio); /* May regress with THPs */ | |
1077 | ||
202bc57b | 1078 | _debug("wback %lx", folio->index); |
62c3b748 DH |
1079 | |
1080 | /* At this point we hold neither the i_pages lock nor the page lock: | |
1081 | * the page may be truncated or invalidated (changing page->mapping to | |
1082 | * NULL), or even swizzled back from swapper_space to tmpfs file | |
1083 | * mapping | |
1084 | */ | |
1085 | lock_again: | |
1086 | if (wbc->sync_mode != WB_SYNC_NONE) { | |
1087 | ret = folio_lock_killable(folio); | |
1088 | if (ret < 0) | |
1089 | return ret; | |
1090 | } else { | |
1091 | if (!folio_trylock(folio)) | |
1092 | goto search_again; | |
1093 | } | |
1094 | ||
1095 | if (folio->mapping != mapping || | |
1096 | !folio_test_dirty(folio)) { | |
1097 | start += folio_size(folio); | |
1098 | folio_unlock(folio); | |
1099 | goto search_again; | |
1100 | } | |
1101 | ||
1102 | if (folio_test_writeback(folio) || | |
1103 | folio_test_fscache(folio)) { | |
1104 | folio_unlock(folio); | |
1105 | if (wbc->sync_mode != WB_SYNC_NONE) { | |
1106 | folio_wait_writeback(folio); | |
e2bdb527 | 1107 | #ifdef CONFIG_FSCACHE |
62c3b748 DH |
1108 | folio_wait_fscache(folio); |
1109 | #endif | |
1110 | goto lock_again; | |
1111 | } | |
1112 | ||
1113 | start += folio_size(folio); | |
1114 | if (wbc->sync_mode == WB_SYNC_NONE) { | |
1115 | if (skips >= 5 || need_resched()) { | |
1116 | ret = 0; | |
1117 | goto out; | |
1118 | } | |
1119 | skips++; | |
1120 | } | |
1121 | goto search_again; | |
1122 | } | |
1123 | ||
1124 | ret = netfs_write_back_from_locked_folio(mapping, wbc, group, xas, | |
1125 | folio, start, end); | |
1126 | out: | |
1127 | if (ret > 0) | |
1128 | *_start = start + ret; | |
1129 | _leave(" = %zd [%llx]", ret, *_start); | |
1130 | return ret; | |
1131 | } | |
1132 | ||
1133 | /* | |
1134 | * Write a region of pages back to the server | |
1135 | */ | |
1136 | static int netfs_writepages_region(struct address_space *mapping, | |
1137 | struct writeback_control *wbc, | |
1138 | struct netfs_group *group, | |
1139 | unsigned long long *_start, | |
1140 | unsigned long long end) | |
1141 | { | |
1142 | ssize_t ret; | |
1143 | ||
1144 | XA_STATE(xas, &mapping->i_pages, *_start / PAGE_SIZE); | |
1145 | ||
1146 | do { | |
1147 | ret = netfs_writepages_begin(mapping, wbc, group, &xas, | |
1148 | _start, end); | |
1149 | if (ret > 0 && wbc->nr_to_write > 0) | |
1150 | cond_resched(); | |
1151 | } while (ret > 0 && wbc->nr_to_write > 0); | |
1152 | ||
1153 | return ret > 0 ? 0 : ret; | |
1154 | } | |
1155 | ||
1156 | /* | |
1157 | * write some of the pending data back to the server | |
1158 | */ | |
1159 | int netfs_writepages(struct address_space *mapping, | |
1160 | struct writeback_control *wbc) | |
1161 | { | |
1162 | struct netfs_group *group = NULL; | |
1163 | loff_t start, end; | |
1164 | int ret; | |
1165 | ||
1166 | _enter(""); | |
1167 | ||
1168 | /* We have to be careful as we can end up racing with setattr() | |
1169 | * truncating the pagecache since the caller doesn't take a lock here | |
1170 | * to prevent it. | |
1171 | */ | |
1172 | ||
1173 | if (wbc->range_cyclic && mapping->writeback_index) { | |
1174 | start = mapping->writeback_index * PAGE_SIZE; | |
1175 | ret = netfs_writepages_region(mapping, wbc, group, | |
1176 | &start, LLONG_MAX); | |
1177 | if (ret < 0) | |
1178 | goto out; | |
1179 | ||
1180 | if (wbc->nr_to_write <= 0) { | |
1181 | mapping->writeback_index = start / PAGE_SIZE; | |
1182 | goto out; | |
1183 | } | |
1184 | ||
1185 | start = 0; | |
1186 | end = mapping->writeback_index * PAGE_SIZE; | |
1187 | mapping->writeback_index = 0; | |
1188 | ret = netfs_writepages_region(mapping, wbc, group, &start, end); | |
1189 | if (ret == 0) | |
1190 | mapping->writeback_index = start / PAGE_SIZE; | |
1191 | } else if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) { | |
1192 | start = 0; | |
1193 | ret = netfs_writepages_region(mapping, wbc, group, | |
1194 | &start, LLONG_MAX); | |
1195 | if (wbc->nr_to_write > 0 && ret == 0) | |
1196 | mapping->writeback_index = start / PAGE_SIZE; | |
1197 | } else { | |
1198 | start = wbc->range_start; | |
1199 | ret = netfs_writepages_region(mapping, wbc, group, | |
1200 | &start, wbc->range_end); | |
1201 | } | |
1202 | ||
1203 | out: | |
1204 | _leave(" = %d", ret); | |
1205 | return ret; | |
1206 | } | |
1207 | EXPORT_SYMBOL(netfs_writepages); | |
4a79616c DH |
1208 | |
1209 | /* | |
1210 | * Deal with the disposition of a laundered folio. | |
1211 | */ | |
1212 | static void netfs_cleanup_launder_folio(struct netfs_io_request *wreq) | |
1213 | { | |
1214 | if (wreq->error) { | |
1215 | pr_notice("R=%08x Laundering error %d\n", wreq->debug_id, wreq->error); | |
1216 | mapping_set_error(wreq->mapping, wreq->error); | |
1217 | } | |
1218 | } | |
1219 | ||
1220 | /** | |
1221 | * netfs_launder_folio - Clean up a dirty folio that's being invalidated | |
1222 | * @folio: The folio to clean | |
1223 | * | |
1224 | * This is called to write back a folio that's being invalidated when an inode | |
1225 | * is getting torn down. Ideally, writepages would be used instead. | |
1226 | */ | |
1227 | int netfs_launder_folio(struct folio *folio) | |
1228 | { | |
1229 | struct netfs_io_request *wreq; | |
1230 | struct address_space *mapping = folio->mapping; | |
1231 | struct netfs_folio *finfo = netfs_folio_info(folio); | |
1232 | struct netfs_group *group = netfs_folio_group(folio); | |
1233 | struct bio_vec bvec; | |
1234 | unsigned long long i_size = i_size_read(mapping->host); | |
1235 | unsigned long long start = folio_pos(folio); | |
1236 | size_t offset = 0, len; | |
1237 | int ret = 0; | |
1238 | ||
1239 | if (finfo) { | |
1240 | offset = finfo->dirty_offset; | |
1241 | start += offset; | |
1242 | len = finfo->dirty_len; | |
1243 | } else { | |
1244 | len = folio_size(folio); | |
1245 | } | |
1246 | len = min_t(unsigned long long, len, i_size - start); | |
1247 | ||
1248 | wreq = netfs_alloc_request(mapping, NULL, start, len, NETFS_LAUNDER_WRITE); | |
1249 | if (IS_ERR(wreq)) { | |
1250 | ret = PTR_ERR(wreq); | |
1251 | goto out; | |
1252 | } | |
1253 | ||
1254 | if (!folio_clear_dirty_for_io(folio)) | |
1255 | goto out_put; | |
1256 | ||
1257 | trace_netfs_folio(folio, netfs_folio_trace_launder); | |
1258 | ||
1259 | _debug("launder %llx-%llx", start, start + len - 1); | |
1260 | ||
1261 | /* Speculatively write to the cache. We have to fix this up later if | |
1262 | * the store fails. | |
1263 | */ | |
1264 | wreq->cleanup = netfs_cleanup_launder_folio; | |
1265 | ||
1266 | bvec_set_folio(&bvec, folio, len, offset); | |
1267 | iov_iter_bvec(&wreq->iter, ITER_SOURCE, &bvec, 1, len); | |
1268 | __set_bit(NETFS_RREQ_UPLOAD_TO_SERVER, &wreq->flags); | |
1269 | ret = netfs_begin_write(wreq, true, netfs_write_trace_launder); | |
1270 | ||
1271 | out_put: | |
1272 | folio_detach_private(folio); | |
1273 | netfs_put_group(group); | |
1274 | kfree(finfo); | |
1275 | netfs_put_request(wreq, false, netfs_rreq_trace_put_return); | |
1276 | out: | |
1277 | folio_wait_fscache(folio); | |
1278 | _leave(" = %d", ret); | |
1279 | return ret; | |
1280 | } | |
1281 | EXPORT_SYMBOL(netfs_launder_folio); |