1 // SPDX-License-Identifier: GPL-2.0-only 1 // SPDX-License-Identifier: GPL-2.0-only
2 /* 2 /*
3 * linux/mm/filemap.c 3 * linux/mm/filemap.c
4 * 4 *
5 * Copyright (C) 1994-1999 Linus Torvalds 5 * Copyright (C) 1994-1999 Linus Torvalds
6 */ 6 */
7 7
8 /* 8 /*
9 * This file handles the generic file mmap sem 9 * This file handles the generic file mmap semantics used by
10 * most "normal" filesystems (but you don't /h 10 * most "normal" filesystems (but you don't /have/ to use this:
11 * the NFS filesystem used to do this differen 11 * the NFS filesystem used to do this differently, for example)
12 */ 12 */
13 #include <linux/export.h> 13 #include <linux/export.h>
14 #include <linux/compiler.h> 14 #include <linux/compiler.h>
15 #include <linux/dax.h> 15 #include <linux/dax.h>
16 #include <linux/fs.h> 16 #include <linux/fs.h>
17 #include <linux/sched/signal.h> 17 #include <linux/sched/signal.h>
18 #include <linux/uaccess.h> 18 #include <linux/uaccess.h>
19 #include <linux/capability.h> 19 #include <linux/capability.h>
20 #include <linux/kernel_stat.h> 20 #include <linux/kernel_stat.h>
21 #include <linux/gfp.h> 21 #include <linux/gfp.h>
22 #include <linux/mm.h> 22 #include <linux/mm.h>
23 #include <linux/swap.h> 23 #include <linux/swap.h>
24 #include <linux/swapops.h> 24 #include <linux/swapops.h>
25 #include <linux/syscalls.h> <<
26 #include <linux/mman.h> 25 #include <linux/mman.h>
27 #include <linux/pagemap.h> 26 #include <linux/pagemap.h>
28 #include <linux/file.h> 27 #include <linux/file.h>
29 #include <linux/uio.h> 28 #include <linux/uio.h>
30 #include <linux/error-injection.h> 29 #include <linux/error-injection.h>
31 #include <linux/hash.h> 30 #include <linux/hash.h>
32 #include <linux/writeback.h> 31 #include <linux/writeback.h>
33 #include <linux/backing-dev.h> 32 #include <linux/backing-dev.h>
34 #include <linux/pagevec.h> 33 #include <linux/pagevec.h>
35 #include <linux/security.h> 34 #include <linux/security.h>
36 #include <linux/cpuset.h> 35 #include <linux/cpuset.h>
37 #include <linux/hugetlb.h> 36 #include <linux/hugetlb.h>
38 #include <linux/memcontrol.h> 37 #include <linux/memcontrol.h>
39 #include <linux/shmem_fs.h> 38 #include <linux/shmem_fs.h>
40 #include <linux/rmap.h> 39 #include <linux/rmap.h>
41 #include <linux/delayacct.h> 40 #include <linux/delayacct.h>
42 #include <linux/psi.h> 41 #include <linux/psi.h>
43 #include <linux/ramfs.h> 42 #include <linux/ramfs.h>
44 #include <linux/page_idle.h> 43 #include <linux/page_idle.h>
45 #include <linux/migrate.h> 44 #include <linux/migrate.h>
46 #include <linux/pipe_fs_i.h> <<
47 #include <linux/splice.h> <<
48 #include <linux/rcupdate_wait.h> <<
49 #include <linux/sched/mm.h> <<
50 #include <asm/pgalloc.h> 45 #include <asm/pgalloc.h>
51 #include <asm/tlbflush.h> 46 #include <asm/tlbflush.h>
52 #include "internal.h" 47 #include "internal.h"
53 48
54 #define CREATE_TRACE_POINTS 49 #define CREATE_TRACE_POINTS
55 #include <trace/events/filemap.h> 50 #include <trace/events/filemap.h>
56 51
57 /* 52 /*
58 * FIXME: remove all knowledge of the buffer l 53 * FIXME: remove all knowledge of the buffer layer from the core VM
59 */ 54 */
60 #include <linux/buffer_head.h> /* for try_to_f 55 #include <linux/buffer_head.h> /* for try_to_free_buffers */
61 56
62 #include <asm/mman.h> 57 #include <asm/mman.h>
63 58
64 #include "swap.h" <<
65 <<
66 /* 59 /*
67 * Shared mappings implemented 30.11.1994. It' 60 * Shared mappings implemented 30.11.1994. It's not fully working yet,
68 * though. 61 * though.
69 * 62 *
70 * Shared mappings now work. 15.8.1995 Bruno. 63 * Shared mappings now work. 15.8.1995 Bruno.
71 * 64 *
72 * finished 'unifying' the page and buffer cac 65 * finished 'unifying' the page and buffer cache and SMP-threaded the
73 * page-cache, 21.05.1999, Ingo Molnar <mingo@ 66 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
74 * 67 *
75 * SMP-threaded pagemap-LRU 1999, Andrea Arcan 68 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
76 */ 69 */
77 70
78 /* 71 /*
79 * Lock ordering: 72 * Lock ordering:
80 * 73 *
81 * ->i_mmap_rwsem (truncate_page 74 * ->i_mmap_rwsem (truncate_pagecache)
82 * ->private_lock (__free_pte->b 75 * ->private_lock (__free_pte->block_dirty_folio)
83 * ->swap_lock (exclusive_swa 76 * ->swap_lock (exclusive_swap_page, others)
84 * ->i_pages lock 77 * ->i_pages lock
85 * 78 *
86 * ->i_rwsem 79 * ->i_rwsem
87 * ->invalidate_lock (acquired by f 80 * ->invalidate_lock (acquired by fs in truncate path)
88 * ->i_mmap_rwsem (truncate->unm 81 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
89 * 82 *
90 * ->mmap_lock 83 * ->mmap_lock
91 * ->i_mmap_rwsem 84 * ->i_mmap_rwsem
92 * ->page_table_lock or pte_lock (vario 85 * ->page_table_lock or pte_lock (various, mainly in memory.c)
93 * ->i_pages lock (arch-dependen 86 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
94 * 87 *
95 * ->mmap_lock 88 * ->mmap_lock
96 * ->invalidate_lock (filemap_fault 89 * ->invalidate_lock (filemap_fault)
97 * ->lock_page (filemap_fault 90 * ->lock_page (filemap_fault, access_process_vm)
98 * 91 *
99 * ->i_rwsem (generic_perfo 92 * ->i_rwsem (generic_perform_write)
100 * ->mmap_lock (fault_in_read 93 * ->mmap_lock (fault_in_readable->do_page_fault)
101 * 94 *
102 * bdi->wb.list_lock 95 * bdi->wb.list_lock
103 * sb_lock (fs/fs-writeba 96 * sb_lock (fs/fs-writeback.c)
104 * ->i_pages lock (__sync_single 97 * ->i_pages lock (__sync_single_inode)
105 * 98 *
106 * ->i_mmap_rwsem 99 * ->i_mmap_rwsem
107 * ->anon_vma.lock (vma_merge) !! 100 * ->anon_vma.lock (vma_adjust)
108 * 101 *
109 * ->anon_vma.lock 102 * ->anon_vma.lock
110 * ->page_table_lock or pte_lock (anon_ 103 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
111 * 104 *
112 * ->page_table_lock or pte_lock 105 * ->page_table_lock or pte_lock
113 * ->swap_lock (try_to_unmap_ 106 * ->swap_lock (try_to_unmap_one)
114 * ->private_lock (try_to_unmap_ 107 * ->private_lock (try_to_unmap_one)
115 * ->i_pages lock (try_to_unmap_ 108 * ->i_pages lock (try_to_unmap_one)
116 * ->lruvec->lru_lock (follow_page_m !! 109 * ->lruvec->lru_lock (follow_page->mark_page_accessed)
117 * ->lruvec->lru_lock (check_pte_ran !! 110 * ->lruvec->lru_lock (check_pte_range->isolate_lru_page)
118 * ->private_lock (folio_remove_ !! 111 * ->private_lock (page_remove_rmap->set_page_dirty)
119 * ->i_pages lock (folio_remove_ !! 112 * ->i_pages lock (page_remove_rmap->set_page_dirty)
120 * bdi.wb->list_lock (folio_remove_ !! 113 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
121 * ->inode->i_lock (folio_remove_ !! 114 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
122 * ->memcg->move_lock (folio_remove_ !! 115 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
123 * bdi.wb->list_lock (zap_pte_range 116 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
124 * ->inode->i_lock (zap_pte_range 117 * ->inode->i_lock (zap_pte_range->set_page_dirty)
125 * ->private_lock (zap_pte_range 118 * ->private_lock (zap_pte_range->block_dirty_folio)
>> 119 *
>> 120 * ->i_mmap_rwsem
>> 121 * ->tasklist_lock (memory_failure, collect_procs_ao)
126 */ 122 */
127 123
128 static void mapping_set_update(struct xa_state <<
129 struct address_space *mapping) <<
130 { <<
131 if (dax_mapping(mapping) || shmem_mapp <<
132 return; <<
133 xas_set_update(xas, workingset_update_ <<
134 xas_set_lru(xas, &shadow_nodes); <<
135 } <<
136 <<
137 static void page_cache_delete(struct address_s 124 static void page_cache_delete(struct address_space *mapping,
138 struct foli 125 struct folio *folio, void *shadow)
139 { 126 {
140 XA_STATE(xas, &mapping->i_pages, folio 127 XA_STATE(xas, &mapping->i_pages, folio->index);
141 long nr = 1; 128 long nr = 1;
142 129
143 mapping_set_update(&xas, mapping); 130 mapping_set_update(&xas, mapping);
144 131
145 xas_set_order(&xas, folio->index, foli !! 132 /* hugetlb pages are represented by a single entry in the xarray */
146 nr = folio_nr_pages(folio); !! 133 if (!folio_test_hugetlb(folio)) {
>> 134 xas_set_order(&xas, folio->index, folio_order(folio));
>> 135 nr = folio_nr_pages(folio);
>> 136 }
147 137
148 VM_BUG_ON_FOLIO(!folio_test_locked(fol 138 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
149 139
150 xas_store(&xas, shadow); 140 xas_store(&xas, shadow);
151 xas_init_marks(&xas); 141 xas_init_marks(&xas);
152 142
153 folio->mapping = NULL; 143 folio->mapping = NULL;
154 /* Leave page->index set: truncation l 144 /* Leave page->index set: truncation lookup relies upon it */
155 mapping->nrpages -= nr; 145 mapping->nrpages -= nr;
156 } 146 }
157 147
158 static void filemap_unaccount_folio(struct add 148 static void filemap_unaccount_folio(struct address_space *mapping,
159 struct folio *folio) 149 struct folio *folio)
160 { 150 {
161 long nr; 151 long nr;
162 152
163 VM_BUG_ON_FOLIO(folio_mapped(folio), f 153 VM_BUG_ON_FOLIO(folio_mapped(folio), folio);
164 if (!IS_ENABLED(CONFIG_DEBUG_VM) && un 154 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(folio_mapped(folio))) {
165 pr_alert("BUG: Bad page cache 155 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
166 current->comm, folio_ 156 current->comm, folio_pfn(folio));
167 dump_page(&folio->page, "still 157 dump_page(&folio->page, "still mapped when deleted");
168 dump_stack(); 158 dump_stack();
169 add_taint(TAINT_BAD_PAGE, LOCK 159 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
170 160
171 if (mapping_exiting(mapping) & 161 if (mapping_exiting(mapping) && !folio_test_large(folio)) {
172 int mapcount = folio_m !! 162 int mapcount = page_mapcount(&folio->page);
173 163
174 if (folio_ref_count(fo 164 if (folio_ref_count(folio) >= mapcount + 2) {
175 /* 165 /*
176 * All vmas ha 166 * All vmas have already been torn down, so it's
177 * a good bet 167 * a good bet that actually the page is unmapped
178 * and we'd ra 168 * and we'd rather not leak it: if we're wrong,
179 * another bad 169 * another bad page check should catch it later.
180 */ 170 */
181 atomic_set(&fo !! 171 page_mapcount_reset(&folio->page);
182 folio_ref_sub( 172 folio_ref_sub(folio, mapcount);
183 } 173 }
184 } 174 }
185 } 175 }
186 176
187 /* hugetlb folios do not participate i 177 /* hugetlb folios do not participate in page cache accounting. */
188 if (folio_test_hugetlb(folio)) 178 if (folio_test_hugetlb(folio))
189 return; 179 return;
190 180
191 nr = folio_nr_pages(folio); 181 nr = folio_nr_pages(folio);
192 182
193 __lruvec_stat_mod_folio(folio, NR_FILE 183 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
194 if (folio_test_swapbacked(folio)) { 184 if (folio_test_swapbacked(folio)) {
195 __lruvec_stat_mod_folio(folio, 185 __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
196 if (folio_test_pmd_mappable(fo 186 if (folio_test_pmd_mappable(folio))
197 __lruvec_stat_mod_foli 187 __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, -nr);
198 } else if (folio_test_pmd_mappable(fol 188 } else if (folio_test_pmd_mappable(folio)) {
199 __lruvec_stat_mod_folio(folio, 189 __lruvec_stat_mod_folio(folio, NR_FILE_THPS, -nr);
200 filemap_nr_thps_dec(mapping); 190 filemap_nr_thps_dec(mapping);
201 } 191 }
202 192
203 /* 193 /*
204 * At this point folio must be either 194 * At this point folio must be either written or cleaned by
205 * truncate. Dirty folio here signals 195 * truncate. Dirty folio here signals a bug and loss of
206 * unwritten data - on ordinary filesy 196 * unwritten data - on ordinary filesystems.
207 * 197 *
208 * But it's harmless on in-memory file 198 * But it's harmless on in-memory filesystems like tmpfs; and can
209 * occur when a driver which did get_u 199 * occur when a driver which did get_user_pages() sets page dirty
210 * before putting it, while the inode 200 * before putting it, while the inode is being finally evicted.
211 * 201 *
212 * Below fixes dirty accounting after 202 * Below fixes dirty accounting after removing the folio entirely
213 * but leaves the dirty flag set: it h 203 * but leaves the dirty flag set: it has no effect for truncated
214 * folio and anyway will be cleared be 204 * folio and anyway will be cleared before returning folio to
215 * buddy allocator. 205 * buddy allocator.
216 */ 206 */
217 if (WARN_ON_ONCE(folio_test_dirty(foli 207 if (WARN_ON_ONCE(folio_test_dirty(folio) &&
218 mapping_can_writeback 208 mapping_can_writeback(mapping)))
219 folio_account_cleaned(folio, i 209 folio_account_cleaned(folio, inode_to_wb(mapping->host));
220 } 210 }
221 211
222 /* 212 /*
223 * Delete a page from the page cache and free 213 * Delete a page from the page cache and free it. Caller has to make
224 * sure the page is locked and that nobody els 214 * sure the page is locked and that nobody else uses it - or that usage
225 * is safe. The caller must hold the i_pages 215 * is safe. The caller must hold the i_pages lock.
226 */ 216 */
227 void __filemap_remove_folio(struct folio *foli 217 void __filemap_remove_folio(struct folio *folio, void *shadow)
228 { 218 {
229 struct address_space *mapping = folio- 219 struct address_space *mapping = folio->mapping;
230 220
231 trace_mm_filemap_delete_from_page_cach 221 trace_mm_filemap_delete_from_page_cache(folio);
232 filemap_unaccount_folio(mapping, folio 222 filemap_unaccount_folio(mapping, folio);
233 page_cache_delete(mapping, folio, shad 223 page_cache_delete(mapping, folio, shadow);
234 } 224 }
235 225
236 void filemap_free_folio(struct address_space * 226 void filemap_free_folio(struct address_space *mapping, struct folio *folio)
237 { 227 {
238 void (*free_folio)(struct folio *); 228 void (*free_folio)(struct folio *);
239 int refs = 1; 229 int refs = 1;
240 230
241 free_folio = mapping->a_ops->free_foli 231 free_folio = mapping->a_ops->free_folio;
242 if (free_folio) 232 if (free_folio)
243 free_folio(folio); 233 free_folio(folio);
244 234
245 if (folio_test_large(folio)) !! 235 if (folio_test_large(folio) && !folio_test_hugetlb(folio))
246 refs = folio_nr_pages(folio); 236 refs = folio_nr_pages(folio);
247 folio_put_refs(folio, refs); 237 folio_put_refs(folio, refs);
248 } 238 }
249 239
250 /** 240 /**
251 * filemap_remove_folio - Remove folio from pa 241 * filemap_remove_folio - Remove folio from page cache.
252 * @folio: The folio. 242 * @folio: The folio.
253 * 243 *
254 * This must be called only on folios that are 244 * This must be called only on folios that are locked and have been
255 * verified to be in the page cache. It will 245 * verified to be in the page cache. It will never put the folio into
256 * the free list because the caller has a refe 246 * the free list because the caller has a reference on the page.
257 */ 247 */
258 void filemap_remove_folio(struct folio *folio) 248 void filemap_remove_folio(struct folio *folio)
259 { 249 {
260 struct address_space *mapping = folio- 250 struct address_space *mapping = folio->mapping;
261 251
262 BUG_ON(!folio_test_locked(folio)); 252 BUG_ON(!folio_test_locked(folio));
263 spin_lock(&mapping->host->i_lock); 253 spin_lock(&mapping->host->i_lock);
264 xa_lock_irq(&mapping->i_pages); 254 xa_lock_irq(&mapping->i_pages);
265 __filemap_remove_folio(folio, NULL); 255 __filemap_remove_folio(folio, NULL);
266 xa_unlock_irq(&mapping->i_pages); 256 xa_unlock_irq(&mapping->i_pages);
267 if (mapping_shrinkable(mapping)) 257 if (mapping_shrinkable(mapping))
268 inode_add_lru(mapping->host); 258 inode_add_lru(mapping->host);
269 spin_unlock(&mapping->host->i_lock); 259 spin_unlock(&mapping->host->i_lock);
270 260
271 filemap_free_folio(mapping, folio); 261 filemap_free_folio(mapping, folio);
272 } 262 }
273 263
274 /* 264 /*
275 * page_cache_delete_batch - delete several fo 265 * page_cache_delete_batch - delete several folios from page cache
276 * @mapping: the mapping to which folios belon 266 * @mapping: the mapping to which folios belong
277 * @fbatch: batch of folios to delete 267 * @fbatch: batch of folios to delete
278 * 268 *
279 * The function walks over mapping->i_pages an 269 * The function walks over mapping->i_pages and removes folios passed in
280 * @fbatch from the mapping. The function expe 270 * @fbatch from the mapping. The function expects @fbatch to be sorted
281 * by page index and is optimised for it to be 271 * by page index and is optimised for it to be dense.
282 * It tolerates holes in @fbatch (mapping entr 272 * It tolerates holes in @fbatch (mapping entries at those indices are not
283 * modified). 273 * modified).
284 * 274 *
285 * The function expects the i_pages lock to be 275 * The function expects the i_pages lock to be held.
286 */ 276 */
287 static void page_cache_delete_batch(struct add 277 static void page_cache_delete_batch(struct address_space *mapping,
288 struct folio_batc 278 struct folio_batch *fbatch)
289 { 279 {
290 XA_STATE(xas, &mapping->i_pages, fbatc 280 XA_STATE(xas, &mapping->i_pages, fbatch->folios[0]->index);
291 long total_pages = 0; 281 long total_pages = 0;
292 int i = 0; 282 int i = 0;
293 struct folio *folio; 283 struct folio *folio;
294 284
295 mapping_set_update(&xas, mapping); 285 mapping_set_update(&xas, mapping);
296 xas_for_each(&xas, folio, ULONG_MAX) { 286 xas_for_each(&xas, folio, ULONG_MAX) {
297 if (i >= folio_batch_count(fba 287 if (i >= folio_batch_count(fbatch))
298 break; 288 break;
299 289
300 /* A swap/dax/shadow entry got 290 /* A swap/dax/shadow entry got inserted? Skip it. */
301 if (xa_is_value(folio)) 291 if (xa_is_value(folio))
302 continue; 292 continue;
303 /* 293 /*
304 * A page got inserted in our 294 * A page got inserted in our range? Skip it. We have our
305 * pages locked so they are pr 295 * pages locked so they are protected from being removed.
306 * If we see a page whose inde 296 * If we see a page whose index is higher than ours, it
307 * means our page has been rem 297 * means our page has been removed, which shouldn't be
308 * possible because we're hold 298 * possible because we're holding the PageLock.
309 */ 299 */
310 if (folio != fbatch->folios[i] 300 if (folio != fbatch->folios[i]) {
311 VM_BUG_ON_FOLIO(folio- 301 VM_BUG_ON_FOLIO(folio->index >
312 fbatch 302 fbatch->folios[i]->index, folio);
313 continue; 303 continue;
314 } 304 }
315 305
316 WARN_ON_ONCE(!folio_test_locke 306 WARN_ON_ONCE(!folio_test_locked(folio));
317 307
318 folio->mapping = NULL; 308 folio->mapping = NULL;
319 /* Leave folio->index set: tru 309 /* Leave folio->index set: truncation lookup relies on it */
320 310
321 i++; 311 i++;
322 xas_store(&xas, NULL); 312 xas_store(&xas, NULL);
323 total_pages += folio_nr_pages( 313 total_pages += folio_nr_pages(folio);
324 } 314 }
325 mapping->nrpages -= total_pages; 315 mapping->nrpages -= total_pages;
326 } 316 }
327 317
328 void delete_from_page_cache_batch(struct addre 318 void delete_from_page_cache_batch(struct address_space *mapping,
329 struct folio 319 struct folio_batch *fbatch)
330 { 320 {
331 int i; 321 int i;
332 322
333 if (!folio_batch_count(fbatch)) 323 if (!folio_batch_count(fbatch))
334 return; 324 return;
335 325
336 spin_lock(&mapping->host->i_lock); 326 spin_lock(&mapping->host->i_lock);
337 xa_lock_irq(&mapping->i_pages); 327 xa_lock_irq(&mapping->i_pages);
338 for (i = 0; i < folio_batch_count(fbat 328 for (i = 0; i < folio_batch_count(fbatch); i++) {
339 struct folio *folio = fbatch-> 329 struct folio *folio = fbatch->folios[i];
340 330
341 trace_mm_filemap_delete_from_p 331 trace_mm_filemap_delete_from_page_cache(folio);
342 filemap_unaccount_folio(mappin 332 filemap_unaccount_folio(mapping, folio);
343 } 333 }
344 page_cache_delete_batch(mapping, fbatc 334 page_cache_delete_batch(mapping, fbatch);
345 xa_unlock_irq(&mapping->i_pages); 335 xa_unlock_irq(&mapping->i_pages);
346 if (mapping_shrinkable(mapping)) 336 if (mapping_shrinkable(mapping))
347 inode_add_lru(mapping->host); 337 inode_add_lru(mapping->host);
348 spin_unlock(&mapping->host->i_lock); 338 spin_unlock(&mapping->host->i_lock);
349 339
350 for (i = 0; i < folio_batch_count(fbat 340 for (i = 0; i < folio_batch_count(fbatch); i++)
351 filemap_free_folio(mapping, fb 341 filemap_free_folio(mapping, fbatch->folios[i]);
352 } 342 }
353 343
354 int filemap_check_errors(struct address_space 344 int filemap_check_errors(struct address_space *mapping)
355 { 345 {
356 int ret = 0; 346 int ret = 0;
357 /* Check for outstanding write errors 347 /* Check for outstanding write errors */
358 if (test_bit(AS_ENOSPC, &mapping->flag 348 if (test_bit(AS_ENOSPC, &mapping->flags) &&
359 test_and_clear_bit(AS_ENOSPC, &map 349 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
360 ret = -ENOSPC; 350 ret = -ENOSPC;
361 if (test_bit(AS_EIO, &mapping->flags) 351 if (test_bit(AS_EIO, &mapping->flags) &&
362 test_and_clear_bit(AS_EIO, &mappin 352 test_and_clear_bit(AS_EIO, &mapping->flags))
363 ret = -EIO; 353 ret = -EIO;
364 return ret; 354 return ret;
365 } 355 }
366 EXPORT_SYMBOL(filemap_check_errors); 356 EXPORT_SYMBOL(filemap_check_errors);
367 357
368 static int filemap_check_and_keep_errors(struc 358 static int filemap_check_and_keep_errors(struct address_space *mapping)
369 { 359 {
370 /* Check for outstanding write errors 360 /* Check for outstanding write errors */
371 if (test_bit(AS_EIO, &mapping->flags)) 361 if (test_bit(AS_EIO, &mapping->flags))
372 return -EIO; 362 return -EIO;
373 if (test_bit(AS_ENOSPC, &mapping->flag 363 if (test_bit(AS_ENOSPC, &mapping->flags))
374 return -ENOSPC; 364 return -ENOSPC;
375 return 0; 365 return 0;
376 } 366 }
377 367
378 /** 368 /**
379 * filemap_fdatawrite_wbc - start writeback on 369 * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range
380 * @mapping: address space structure to wri 370 * @mapping: address space structure to write
381 * @wbc: the writeback_control controll 371 * @wbc: the writeback_control controlling the writeout
382 * 372 *
383 * Call writepages on the mapping using the pr 373 * Call writepages on the mapping using the provided wbc to control the
384 * writeout. 374 * writeout.
385 * 375 *
386 * Return: %0 on success, negative error code 376 * Return: %0 on success, negative error code otherwise.
387 */ 377 */
388 int filemap_fdatawrite_wbc(struct address_spac 378 int filemap_fdatawrite_wbc(struct address_space *mapping,
389 struct writeback_co 379 struct writeback_control *wbc)
390 { 380 {
391 int ret; 381 int ret;
392 382
393 if (!mapping_can_writeback(mapping) || 383 if (!mapping_can_writeback(mapping) ||
394 !mapping_tagged(mapping, PAGECACHE 384 !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
395 return 0; 385 return 0;
396 386
397 wbc_attach_fdatawrite_inode(wbc, mappi 387 wbc_attach_fdatawrite_inode(wbc, mapping->host);
398 ret = do_writepages(mapping, wbc); 388 ret = do_writepages(mapping, wbc);
399 wbc_detach_inode(wbc); 389 wbc_detach_inode(wbc);
400 return ret; 390 return ret;
401 } 391 }
402 EXPORT_SYMBOL(filemap_fdatawrite_wbc); 392 EXPORT_SYMBOL(filemap_fdatawrite_wbc);
403 393
404 /** 394 /**
405 * __filemap_fdatawrite_range - start writebac 395 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
406 * @mapping: address space structure to wri 396 * @mapping: address space structure to write
407 * @start: offset in bytes where the rang 397 * @start: offset in bytes where the range starts
408 * @end: offset in bytes where the rang 398 * @end: offset in bytes where the range ends (inclusive)
409 * @sync_mode: enable synchronous operation 399 * @sync_mode: enable synchronous operation
410 * 400 *
411 * Start writeback against all of a mapping's 401 * Start writeback against all of a mapping's dirty pages that lie
412 * within the byte offsets <start, end> inclus 402 * within the byte offsets <start, end> inclusive.
413 * 403 *
414 * If sync_mode is WB_SYNC_ALL then this is a 404 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
415 * opposed to a regular memory cleansing write 405 * opposed to a regular memory cleansing writeback. The difference between
416 * these two operations is that if a dirty pag 406 * these two operations is that if a dirty page/buffer is encountered, it must
417 * be waited upon, and not just skipped over. 407 * be waited upon, and not just skipped over.
418 * 408 *
419 * Return: %0 on success, negative error code 409 * Return: %0 on success, negative error code otherwise.
420 */ 410 */
421 int __filemap_fdatawrite_range(struct address_ 411 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
422 loff_t end, in 412 loff_t end, int sync_mode)
423 { 413 {
424 struct writeback_control wbc = { 414 struct writeback_control wbc = {
425 .sync_mode = sync_mode, 415 .sync_mode = sync_mode,
426 .nr_to_write = LONG_MAX, 416 .nr_to_write = LONG_MAX,
427 .range_start = start, 417 .range_start = start,
428 .range_end = end, 418 .range_end = end,
429 }; 419 };
430 420
431 return filemap_fdatawrite_wbc(mapping, 421 return filemap_fdatawrite_wbc(mapping, &wbc);
432 } 422 }
433 423
434 static inline int __filemap_fdatawrite(struct 424 static inline int __filemap_fdatawrite(struct address_space *mapping,
435 int sync_mode) 425 int sync_mode)
436 { 426 {
437 return __filemap_fdatawrite_range(mapp 427 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
438 } 428 }
439 429
440 int filemap_fdatawrite(struct address_space *m 430 int filemap_fdatawrite(struct address_space *mapping)
441 { 431 {
442 return __filemap_fdatawrite(mapping, W 432 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
443 } 433 }
444 EXPORT_SYMBOL(filemap_fdatawrite); 434 EXPORT_SYMBOL(filemap_fdatawrite);
445 435
446 int filemap_fdatawrite_range(struct address_sp 436 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
447 loff_t end) 437 loff_t end)
448 { 438 {
449 return __filemap_fdatawrite_range(mapp 439 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
450 } 440 }
451 EXPORT_SYMBOL(filemap_fdatawrite_range); 441 EXPORT_SYMBOL(filemap_fdatawrite_range);
452 442
453 /** 443 /**
454 * filemap_flush - mostly a non-blocking flush 444 * filemap_flush - mostly a non-blocking flush
455 * @mapping: target address_space 445 * @mapping: target address_space
456 * 446 *
457 * This is a mostly non-blocking flush. Not s 447 * This is a mostly non-blocking flush. Not suitable for data-integrity
458 * purposes - I/O may not be started against a 448 * purposes - I/O may not be started against all dirty pages.
459 * 449 *
460 * Return: %0 on success, negative error code 450 * Return: %0 on success, negative error code otherwise.
461 */ 451 */
462 int filemap_flush(struct address_space *mappin 452 int filemap_flush(struct address_space *mapping)
463 { 453 {
464 return __filemap_fdatawrite(mapping, W 454 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
465 } 455 }
466 EXPORT_SYMBOL(filemap_flush); 456 EXPORT_SYMBOL(filemap_flush);
467 457
468 /** 458 /**
469 * filemap_range_has_page - check if a page ex 459 * filemap_range_has_page - check if a page exists in range.
470 * @mapping: address space within wh 460 * @mapping: address space within which to check
471 * @start_byte: offset in bytes where t 461 * @start_byte: offset in bytes where the range starts
472 * @end_byte: offset in bytes where t 462 * @end_byte: offset in bytes where the range ends (inclusive)
473 * 463 *
474 * Find at least one page in the range supplie 464 * Find at least one page in the range supplied, usually used to check if
475 * direct writing in this range will trigger a 465 * direct writing in this range will trigger a writeback.
476 * 466 *
477 * Return: %true if at least one page exists i 467 * Return: %true if at least one page exists in the specified range,
478 * %false otherwise. 468 * %false otherwise.
479 */ 469 */
480 bool filemap_range_has_page(struct address_spa 470 bool filemap_range_has_page(struct address_space *mapping,
481 loff_t start_byte, 471 loff_t start_byte, loff_t end_byte)
482 { 472 {
483 struct folio *folio; !! 473 struct page *page;
484 XA_STATE(xas, &mapping->i_pages, start 474 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
485 pgoff_t max = end_byte >> PAGE_SHIFT; 475 pgoff_t max = end_byte >> PAGE_SHIFT;
486 476
487 if (end_byte < start_byte) 477 if (end_byte < start_byte)
488 return false; 478 return false;
489 479
490 rcu_read_lock(); 480 rcu_read_lock();
491 for (;;) { 481 for (;;) {
492 folio = xas_find(&xas, max); !! 482 page = xas_find(&xas, max);
493 if (xas_retry(&xas, folio)) !! 483 if (xas_retry(&xas, page))
494 continue; 484 continue;
495 /* Shadow entries don't count 485 /* Shadow entries don't count */
496 if (xa_is_value(folio)) !! 486 if (xa_is_value(page))
497 continue; 487 continue;
498 /* 488 /*
499 * We don't need to try to pin 489 * We don't need to try to pin this page; we're about to
500 * release the RCU lock anyway 490 * release the RCU lock anyway. It is enough to know that
501 * there was a page here recen 491 * there was a page here recently.
502 */ 492 */
503 break; 493 break;
504 } 494 }
505 rcu_read_unlock(); 495 rcu_read_unlock();
506 496
507 return folio != NULL; !! 497 return page != NULL;
508 } 498 }
509 EXPORT_SYMBOL(filemap_range_has_page); 499 EXPORT_SYMBOL(filemap_range_has_page);
510 500
511 static void __filemap_fdatawait_range(struct a 501 static void __filemap_fdatawait_range(struct address_space *mapping,
512 loff_t st 502 loff_t start_byte, loff_t end_byte)
513 { 503 {
514 pgoff_t index = start_byte >> PAGE_SHI 504 pgoff_t index = start_byte >> PAGE_SHIFT;
515 pgoff_t end = end_byte >> PAGE_SHIFT; 505 pgoff_t end = end_byte >> PAGE_SHIFT;
516 struct folio_batch fbatch; !! 506 struct pagevec pvec;
517 unsigned nr_folios; !! 507 int nr_pages;
518 508
519 folio_batch_init(&fbatch); !! 509 if (end_byte < start_byte)
>> 510 return;
520 511
>> 512 pagevec_init(&pvec);
521 while (index <= end) { 513 while (index <= end) {
522 unsigned i; 514 unsigned i;
523 515
524 nr_folios = filemap_get_folios !! 516 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
525 PAGECACHE_TAG_ !! 517 end, PAGECACHE_TAG_WRITEBACK);
526 !! 518 if (!nr_pages)
527 if (!nr_folios) <<
528 break; 519 break;
529 520
530 for (i = 0; i < nr_folios; i++ !! 521 for (i = 0; i < nr_pages; i++) {
531 struct folio *folio = !! 522 struct page *page = pvec.pages[i];
532 523
533 folio_wait_writeback(f !! 524 wait_on_page_writeback(page);
>> 525 ClearPageError(page);
534 } 526 }
535 folio_batch_release(&fbatch); !! 527 pagevec_release(&pvec);
536 cond_resched(); 528 cond_resched();
537 } 529 }
538 } 530 }
539 531
540 /** 532 /**
541 * filemap_fdatawait_range - wait for writebac 533 * filemap_fdatawait_range - wait for writeback to complete
542 * @mapping: address space structur 534 * @mapping: address space structure to wait for
543 * @start_byte: offset in bytes where 535 * @start_byte: offset in bytes where the range starts
544 * @end_byte: offset in bytes where 536 * @end_byte: offset in bytes where the range ends (inclusive)
545 * 537 *
546 * Walk the list of under-writeback pages of t 538 * Walk the list of under-writeback pages of the given address space
547 * in the given range and wait for all of them 539 * in the given range and wait for all of them. Check error status of
548 * the address space and return it. 540 * the address space and return it.
549 * 541 *
550 * Since the error status of the address space 542 * Since the error status of the address space is cleared by this function,
551 * callers are responsible for checking the re 543 * callers are responsible for checking the return value and handling and/or
552 * reporting the error. 544 * reporting the error.
553 * 545 *
554 * Return: error status of the address space. 546 * Return: error status of the address space.
555 */ 547 */
556 int filemap_fdatawait_range(struct address_spa 548 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
557 loff_t end_byte) 549 loff_t end_byte)
558 { 550 {
559 __filemap_fdatawait_range(mapping, sta 551 __filemap_fdatawait_range(mapping, start_byte, end_byte);
560 return filemap_check_errors(mapping); 552 return filemap_check_errors(mapping);
561 } 553 }
562 EXPORT_SYMBOL(filemap_fdatawait_range); 554 EXPORT_SYMBOL(filemap_fdatawait_range);
563 555
564 /** 556 /**
565 * filemap_fdatawait_range_keep_errors - wait 557 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
566 * @mapping: address space structur 558 * @mapping: address space structure to wait for
567 * @start_byte: offset in bytes where 559 * @start_byte: offset in bytes where the range starts
568 * @end_byte: offset in bytes where 560 * @end_byte: offset in bytes where the range ends (inclusive)
569 * 561 *
570 * Walk the list of under-writeback pages of t 562 * Walk the list of under-writeback pages of the given address space in the
571 * given range and wait for all of them. Unli 563 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
572 * this function does not clear error status o 564 * this function does not clear error status of the address space.
573 * 565 *
574 * Use this function if callers don't handle e 566 * Use this function if callers don't handle errors themselves. Expected
575 * call sites are system-wide / filesystem-wid 567 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
576 * fsfreeze(8) 568 * fsfreeze(8)
577 */ 569 */
578 int filemap_fdatawait_range_keep_errors(struct 570 int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
579 loff_t start_byte, loff_t end_ 571 loff_t start_byte, loff_t end_byte)
580 { 572 {
581 __filemap_fdatawait_range(mapping, sta 573 __filemap_fdatawait_range(mapping, start_byte, end_byte);
582 return filemap_check_and_keep_errors(m 574 return filemap_check_and_keep_errors(mapping);
583 } 575 }
584 EXPORT_SYMBOL(filemap_fdatawait_range_keep_err 576 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
585 577
586 /** 578 /**
587 * file_fdatawait_range - wait for writeback t 579 * file_fdatawait_range - wait for writeback to complete
588 * @file: file pointing to addre 580 * @file: file pointing to address space structure to wait for
589 * @start_byte: offset in bytes where 581 * @start_byte: offset in bytes where the range starts
590 * @end_byte: offset in bytes where 582 * @end_byte: offset in bytes where the range ends (inclusive)
591 * 583 *
592 * Walk the list of under-writeback pages of t 584 * Walk the list of under-writeback pages of the address space that file
593 * refers to, in the given range and wait for 585 * refers to, in the given range and wait for all of them. Check error
594 * status of the address space vs. the file->f 586 * status of the address space vs. the file->f_wb_err cursor and return it.
595 * 587 *
596 * Since the error status of the file is advan 588 * Since the error status of the file is advanced by this function,
597 * callers are responsible for checking the re 589 * callers are responsible for checking the return value and handling and/or
598 * reporting the error. 590 * reporting the error.
599 * 591 *
600 * Return: error status of the address space v 592 * Return: error status of the address space vs. the file->f_wb_err cursor.
601 */ 593 */
602 int file_fdatawait_range(struct file *file, lo 594 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
603 { 595 {
604 struct address_space *mapping = file-> 596 struct address_space *mapping = file->f_mapping;
605 597
606 __filemap_fdatawait_range(mapping, sta 598 __filemap_fdatawait_range(mapping, start_byte, end_byte);
607 return file_check_and_advance_wb_err(f 599 return file_check_and_advance_wb_err(file);
608 } 600 }
609 EXPORT_SYMBOL(file_fdatawait_range); 601 EXPORT_SYMBOL(file_fdatawait_range);
610 602
611 /** 603 /**
612 * filemap_fdatawait_keep_errors - wait for wr 604 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
613 * @mapping: address space structure to wait f 605 * @mapping: address space structure to wait for
614 * 606 *
615 * Walk the list of under-writeback pages of t 607 * Walk the list of under-writeback pages of the given address space
616 * and wait for all of them. Unlike filemap_f 608 * and wait for all of them. Unlike filemap_fdatawait(), this function
617 * does not clear error status of the address 609 * does not clear error status of the address space.
618 * 610 *
619 * Use this function if callers don't handle e 611 * Use this function if callers don't handle errors themselves. Expected
620 * call sites are system-wide / filesystem-wid 612 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
621 * fsfreeze(8) 613 * fsfreeze(8)
622 * 614 *
623 * Return: error status of the address space. 615 * Return: error status of the address space.
624 */ 616 */
625 int filemap_fdatawait_keep_errors(struct addre 617 int filemap_fdatawait_keep_errors(struct address_space *mapping)
626 { 618 {
627 __filemap_fdatawait_range(mapping, 0, 619 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
628 return filemap_check_and_keep_errors(m 620 return filemap_check_and_keep_errors(mapping);
629 } 621 }
630 EXPORT_SYMBOL(filemap_fdatawait_keep_errors); 622 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
631 623
632 /* Returns true if writeback might be needed o 624 /* Returns true if writeback might be needed or already in progress. */
633 static bool mapping_needs_writeback(struct add 625 static bool mapping_needs_writeback(struct address_space *mapping)
634 { 626 {
635 return mapping->nrpages; 627 return mapping->nrpages;
636 } 628 }
637 629
638 bool filemap_range_has_writeback(struct addres 630 bool filemap_range_has_writeback(struct address_space *mapping,
639 loff_t start_ 631 loff_t start_byte, loff_t end_byte)
640 { 632 {
641 XA_STATE(xas, &mapping->i_pages, start 633 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
642 pgoff_t max = end_byte >> PAGE_SHIFT; 634 pgoff_t max = end_byte >> PAGE_SHIFT;
643 struct folio *folio; !! 635 struct page *page;
644 636
645 if (end_byte < start_byte) 637 if (end_byte < start_byte)
646 return false; 638 return false;
647 639
648 rcu_read_lock(); 640 rcu_read_lock();
649 xas_for_each(&xas, folio, max) { !! 641 xas_for_each(&xas, page, max) {
650 if (xas_retry(&xas, folio)) !! 642 if (xas_retry(&xas, page))
651 continue; 643 continue;
652 if (xa_is_value(folio)) !! 644 if (xa_is_value(page))
653 continue; 645 continue;
654 if (folio_test_dirty(folio) || !! 646 if (PageDirty(page) || PageLocked(page) || PageWriteback(page))
655 folio_test_wri <<
656 break; 647 break;
657 } 648 }
658 rcu_read_unlock(); 649 rcu_read_unlock();
659 return folio != NULL; !! 650 return page != NULL;
660 } 651 }
661 EXPORT_SYMBOL_GPL(filemap_range_has_writeback) 652 EXPORT_SYMBOL_GPL(filemap_range_has_writeback);
662 653
663 /** 654 /**
664 * filemap_write_and_wait_range - write out & 655 * filemap_write_and_wait_range - write out & wait on a file range
665 * @mapping: the address_space for the page 656 * @mapping: the address_space for the pages
666 * @lstart: offset in bytes where the rang 657 * @lstart: offset in bytes where the range starts
667 * @lend: offset in bytes where the rang 658 * @lend: offset in bytes where the range ends (inclusive)
668 * 659 *
669 * Write out and wait upon file offsets lstart 660 * Write out and wait upon file offsets lstart->lend, inclusive.
670 * 661 *
671 * Note that @lend is inclusive (describes the 662 * Note that @lend is inclusive (describes the last byte to be written) so
672 * that this function can be used to write to 663 * that this function can be used to write to the very end-of-file (end = -1).
673 * 664 *
674 * Return: error status of the address space. 665 * Return: error status of the address space.
675 */ 666 */
676 int filemap_write_and_wait_range(struct addres 667 int filemap_write_and_wait_range(struct address_space *mapping,
677 loff_t lstart 668 loff_t lstart, loff_t lend)
678 { 669 {
679 int err = 0, err2; 670 int err = 0, err2;
680 671
681 if (lend < lstart) <<
682 return 0; <<
683 <<
684 if (mapping_needs_writeback(mapping)) 672 if (mapping_needs_writeback(mapping)) {
685 err = __filemap_fdatawrite_ran 673 err = __filemap_fdatawrite_range(mapping, lstart, lend,
686 674 WB_SYNC_ALL);
687 /* 675 /*
688 * Even if the above returned 676 * Even if the above returned error, the pages may be
689 * written partially (e.g. -EN 677 * written partially (e.g. -ENOSPC), so we wait for it.
690 * But the -EIO is special cas 678 * But the -EIO is special case, it may indicate the worst
691 * thing (e.g. bug) happened, 679 * thing (e.g. bug) happened, so we avoid waiting for it.
692 */ 680 */
693 if (err != -EIO) 681 if (err != -EIO)
694 __filemap_fdatawait_ra 682 __filemap_fdatawait_range(mapping, lstart, lend);
695 } 683 }
696 err2 = filemap_check_errors(mapping); 684 err2 = filemap_check_errors(mapping);
697 if (!err) 685 if (!err)
698 err = err2; 686 err = err2;
699 return err; 687 return err;
700 } 688 }
701 EXPORT_SYMBOL(filemap_write_and_wait_range); 689 EXPORT_SYMBOL(filemap_write_and_wait_range);
702 690
703 void __filemap_set_wb_err(struct address_space 691 void __filemap_set_wb_err(struct address_space *mapping, int err)
704 { 692 {
705 errseq_t eseq = errseq_set(&mapping->w 693 errseq_t eseq = errseq_set(&mapping->wb_err, err);
706 694
707 trace_filemap_set_wb_err(mapping, eseq 695 trace_filemap_set_wb_err(mapping, eseq);
708 } 696 }
709 EXPORT_SYMBOL(__filemap_set_wb_err); 697 EXPORT_SYMBOL(__filemap_set_wb_err);
710 698
711 /** 699 /**
712 * file_check_and_advance_wb_err - report wb e 700 * file_check_and_advance_wb_err - report wb error (if any) that was previously
713 * and advance 701 * and advance wb_err to current one
714 * @file: struct file on which the error is be 702 * @file: struct file on which the error is being reported
715 * 703 *
716 * When userland calls fsync (or something lik 704 * When userland calls fsync (or something like nfsd does the equivalent), we
717 * want to report any writeback errors that oc 705 * want to report any writeback errors that occurred since the last fsync (or
718 * since the file was opened if there haven't 706 * since the file was opened if there haven't been any).
719 * 707 *
720 * Grab the wb_err from the mapping. If it mat 708 * Grab the wb_err from the mapping. If it matches what we have in the file,
721 * then just quickly return 0. The file is all 709 * then just quickly return 0. The file is all caught up.
722 * 710 *
723 * If it doesn't match, then take the mapping 711 * If it doesn't match, then take the mapping value, set the "seen" flag in
724 * it and try to swap it into place. If it wor 712 * it and try to swap it into place. If it works, or another task beat us
725 * to it with the new value, then update the f 713 * to it with the new value, then update the f_wb_err and return the error
726 * portion. The error at this point must be re 714 * portion. The error at this point must be reported via proper channels
727 * (a'la fsync, or NFS COMMIT operation, etc.) 715 * (a'la fsync, or NFS COMMIT operation, etc.).
728 * 716 *
729 * While we handle mapping->wb_err with atomic 717 * While we handle mapping->wb_err with atomic operations, the f_wb_err
730 * value is protected by the f_lock since we m 718 * value is protected by the f_lock since we must ensure that it reflects
731 * the latest value swapped in for this file d 719 * the latest value swapped in for this file descriptor.
732 * 720 *
733 * Return: %0 on success, negative error code 721 * Return: %0 on success, negative error code otherwise.
734 */ 722 */
735 int file_check_and_advance_wb_err(struct file 723 int file_check_and_advance_wb_err(struct file *file)
736 { 724 {
737 int err = 0; 725 int err = 0;
738 errseq_t old = READ_ONCE(file->f_wb_er 726 errseq_t old = READ_ONCE(file->f_wb_err);
739 struct address_space *mapping = file-> 727 struct address_space *mapping = file->f_mapping;
740 728
741 /* Locklessly handle the common case w 729 /* Locklessly handle the common case where nothing has changed */
742 if (errseq_check(&mapping->wb_err, old 730 if (errseq_check(&mapping->wb_err, old)) {
743 /* Something changed, must use 731 /* Something changed, must use slow path */
744 spin_lock(&file->f_lock); 732 spin_lock(&file->f_lock);
745 old = file->f_wb_err; 733 old = file->f_wb_err;
746 err = errseq_check_and_advance 734 err = errseq_check_and_advance(&mapping->wb_err,
747 735 &file->f_wb_err);
748 trace_file_check_and_advance_w 736 trace_file_check_and_advance_wb_err(file, old);
749 spin_unlock(&file->f_lock); 737 spin_unlock(&file->f_lock);
750 } 738 }
751 739
752 /* 740 /*
753 * We're mostly using this function as 741 * We're mostly using this function as a drop in replacement for
754 * filemap_check_errors. Clear AS_EIO/ 742 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
755 * that the legacy code would have had 743 * that the legacy code would have had on these flags.
756 */ 744 */
757 clear_bit(AS_EIO, &mapping->flags); 745 clear_bit(AS_EIO, &mapping->flags);
758 clear_bit(AS_ENOSPC, &mapping->flags); 746 clear_bit(AS_ENOSPC, &mapping->flags);
759 return err; 747 return err;
760 } 748 }
761 EXPORT_SYMBOL(file_check_and_advance_wb_err); 749 EXPORT_SYMBOL(file_check_and_advance_wb_err);
762 750
763 /** 751 /**
764 * file_write_and_wait_range - write out & wai 752 * file_write_and_wait_range - write out & wait on a file range
765 * @file: file pointing to address_space 753 * @file: file pointing to address_space with pages
766 * @lstart: offset in bytes where the rang 754 * @lstart: offset in bytes where the range starts
767 * @lend: offset in bytes where the rang 755 * @lend: offset in bytes where the range ends (inclusive)
768 * 756 *
769 * Write out and wait upon file offsets lstart 757 * Write out and wait upon file offsets lstart->lend, inclusive.
770 * 758 *
771 * Note that @lend is inclusive (describes the 759 * Note that @lend is inclusive (describes the last byte to be written) so
772 * that this function can be used to write to 760 * that this function can be used to write to the very end-of-file (end = -1).
773 * 761 *
774 * After writing out and waiting on the data, 762 * After writing out and waiting on the data, we check and advance the
775 * f_wb_err cursor to the latest value, and re 763 * f_wb_err cursor to the latest value, and return any errors detected there.
776 * 764 *
777 * Return: %0 on success, negative error code 765 * Return: %0 on success, negative error code otherwise.
778 */ 766 */
779 int file_write_and_wait_range(struct file *fil 767 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
780 { 768 {
781 int err = 0, err2; 769 int err = 0, err2;
782 struct address_space *mapping = file-> 770 struct address_space *mapping = file->f_mapping;
783 771
784 if (lend < lstart) <<
785 return 0; <<
786 <<
787 if (mapping_needs_writeback(mapping)) 772 if (mapping_needs_writeback(mapping)) {
788 err = __filemap_fdatawrite_ran 773 err = __filemap_fdatawrite_range(mapping, lstart, lend,
789 774 WB_SYNC_ALL);
790 /* See comment of filemap_writ 775 /* See comment of filemap_write_and_wait() */
791 if (err != -EIO) 776 if (err != -EIO)
792 __filemap_fdatawait_ra 777 __filemap_fdatawait_range(mapping, lstart, lend);
793 } 778 }
794 err2 = file_check_and_advance_wb_err(f 779 err2 = file_check_and_advance_wb_err(file);
795 if (!err) 780 if (!err)
796 err = err2; 781 err = err2;
797 return err; 782 return err;
798 } 783 }
799 EXPORT_SYMBOL(file_write_and_wait_range); 784 EXPORT_SYMBOL(file_write_and_wait_range);
800 785
801 /** 786 /**
802 * replace_page_cache_folio - replace a pageca !! 787 * replace_page_cache_page - replace a pagecache page with a new one
803 * @old: folio to be replaced !! 788 * @old: page to be replaced
804 * @new: folio to replace with !! 789 * @new: page to replace with
805 * !! 790 *
806 * This function replaces a folio in the pagec !! 791 * This function replaces a page in the pagecache with a new one. On
807 * success it acquires the pagecache reference !! 792 * success it acquires the pagecache reference for the new page and
808 * drops it for the old folio. Both the old a !! 793 * drops it for the old page. Both the old and new pages must be
809 * locked. This function does not add the new !! 794 * locked. This function does not add the new page to the LRU, the
810 * caller must do that. 795 * caller must do that.
811 * 796 *
812 * The remove + add is atomic. This function 797 * The remove + add is atomic. This function cannot fail.
813 */ 798 */
814 void replace_page_cache_folio(struct folio *ol !! 799 void replace_page_cache_page(struct page *old, struct page *new)
815 { 800 {
>> 801 struct folio *fold = page_folio(old);
>> 802 struct folio *fnew = page_folio(new);
816 struct address_space *mapping = old->m 803 struct address_space *mapping = old->mapping;
817 void (*free_folio)(struct folio *) = m 804 void (*free_folio)(struct folio *) = mapping->a_ops->free_folio;
818 pgoff_t offset = old->index; 805 pgoff_t offset = old->index;
819 XA_STATE(xas, &mapping->i_pages, offse 806 XA_STATE(xas, &mapping->i_pages, offset);
820 807
821 VM_BUG_ON_FOLIO(!folio_test_locked(old !! 808 VM_BUG_ON_PAGE(!PageLocked(old), old);
822 VM_BUG_ON_FOLIO(!folio_test_locked(new !! 809 VM_BUG_ON_PAGE(!PageLocked(new), new);
823 VM_BUG_ON_FOLIO(new->mapping, new); !! 810 VM_BUG_ON_PAGE(new->mapping, new);
824 811
825 folio_get(new); !! 812 get_page(new);
826 new->mapping = mapping; 813 new->mapping = mapping;
827 new->index = offset; 814 new->index = offset;
828 815
829 mem_cgroup_replace_folio(old, new); !! 816 mem_cgroup_migrate(fold, fnew);
830 817
831 xas_lock_irq(&xas); 818 xas_lock_irq(&xas);
832 xas_store(&xas, new); 819 xas_store(&xas, new);
833 820
834 old->mapping = NULL; 821 old->mapping = NULL;
835 /* hugetlb pages do not participate in 822 /* hugetlb pages do not participate in page cache accounting. */
836 if (!folio_test_hugetlb(old)) !! 823 if (!PageHuge(old))
837 __lruvec_stat_sub_folio(old, N !! 824 __dec_lruvec_page_state(old, NR_FILE_PAGES);
838 if (!folio_test_hugetlb(new)) !! 825 if (!PageHuge(new))
839 __lruvec_stat_add_folio(new, N !! 826 __inc_lruvec_page_state(new, NR_FILE_PAGES);
840 if (folio_test_swapbacked(old)) !! 827 if (PageSwapBacked(old))
841 __lruvec_stat_sub_folio(old, N !! 828 __dec_lruvec_page_state(old, NR_SHMEM);
842 if (folio_test_swapbacked(new)) !! 829 if (PageSwapBacked(new))
843 __lruvec_stat_add_folio(new, N !! 830 __inc_lruvec_page_state(new, NR_SHMEM);
844 xas_unlock_irq(&xas); 831 xas_unlock_irq(&xas);
845 if (free_folio) 832 if (free_folio)
846 free_folio(old); !! 833 free_folio(fold);
847 folio_put(old); !! 834 folio_put(fold);
848 } 835 }
849 EXPORT_SYMBOL_GPL(replace_page_cache_folio); !! 836 EXPORT_SYMBOL_GPL(replace_page_cache_page);
850 837
851 noinline int __filemap_add_folio(struct addres 838 noinline int __filemap_add_folio(struct address_space *mapping,
852 struct folio *folio, pgoff_t i 839 struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp)
853 { 840 {
854 XA_STATE(xas, &mapping->i_pages, index 841 XA_STATE(xas, &mapping->i_pages, index);
855 void *alloced_shadow = NULL; !! 842 int huge = folio_test_hugetlb(folio);
856 int alloced_order = 0; !! 843 bool charged = false;
857 bool huge; !! 844 long nr = 1;
858 long nr; <<
859 845
860 VM_BUG_ON_FOLIO(!folio_test_locked(fol 846 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
861 VM_BUG_ON_FOLIO(folio_test_swapbacked( 847 VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio);
862 VM_BUG_ON_FOLIO(folio_order(folio) < m <<
863 folio); <<
864 mapping_set_update(&xas, mapping); 848 mapping_set_update(&xas, mapping);
865 849
866 VM_BUG_ON_FOLIO(index & (folio_nr_page !! 850 if (!huge) {
867 xas_set_order(&xas, index, folio_order !! 851 int error = mem_cgroup_charge(folio, NULL, gfp);
868 huge = folio_test_hugetlb(folio); !! 852 VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio);
869 nr = folio_nr_pages(folio); !! 853 if (error)
>> 854 return error;
>> 855 charged = true;
>> 856 xas_set_order(&xas, index, folio_order(folio));
>> 857 nr = folio_nr_pages(folio);
>> 858 }
870 859
871 gfp &= GFP_RECLAIM_MASK; 860 gfp &= GFP_RECLAIM_MASK;
872 folio_ref_add(folio, nr); 861 folio_ref_add(folio, nr);
873 folio->mapping = mapping; 862 folio->mapping = mapping;
874 folio->index = xas.xa_index; 863 folio->index = xas.xa_index;
875 864
876 for (;;) { !! 865 do {
877 int order = -1, split_order = !! 866 unsigned int order = xa_get_order(xas.xa, xas.xa_index);
878 void *entry, *old = NULL; 867 void *entry, *old = NULL;
879 868
>> 869 if (order > folio_order(folio))
>> 870 xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
>> 871 order, gfp);
880 xas_lock_irq(&xas); 872 xas_lock_irq(&xas);
881 xas_for_each_conflict(&xas, en 873 xas_for_each_conflict(&xas, entry) {
882 old = entry; 874 old = entry;
883 if (!xa_is_value(entry 875 if (!xa_is_value(entry)) {
884 xas_set_err(&x 876 xas_set_err(&xas, -EEXIST);
885 goto unlock; 877 goto unlock;
886 } 878 }
887 /* <<
888 * If a larger entry e <<
889 * it will be the firs <<
890 */ <<
891 if (order == -1) <<
892 order = xas_ge <<
893 } <<
894 <<
895 /* entry may have changed befo <<
896 if (alloced_order && (old != a <<
897 xas_destroy(&xas); <<
898 alloced_order = 0; <<
899 } 879 }
900 880
901 if (old) { 881 if (old) {
902 if (order > 0 && order !! 882 if (shadowp)
>> 883 *shadowp = old;
>> 884 /* entry may have been split before we acquired lock */
>> 885 order = xa_get_order(xas.xa, xas.xa_index);
>> 886 if (order > folio_order(folio)) {
903 /* How to hand 887 /* How to handle large swap entries? */
904 BUG_ON(shmem_m 888 BUG_ON(shmem_mapping(mapping));
905 if (!alloced_o <<
906 split_ <<
907 goto u <<
908 } <<
909 xas_split(&xas 889 xas_split(&xas, old, order);
910 xas_reset(&xas 890 xas_reset(&xas);
911 } 891 }
912 if (shadowp) <<
913 *shadowp = old <<
914 } 892 }
915 893
916 xas_store(&xas, folio); 894 xas_store(&xas, folio);
917 if (xas_error(&xas)) 895 if (xas_error(&xas))
918 goto unlock; 896 goto unlock;
919 897
920 mapping->nrpages += nr; 898 mapping->nrpages += nr;
921 899
922 /* hugetlb pages do not partic 900 /* hugetlb pages do not participate in page cache accounting */
923 if (!huge) { 901 if (!huge) {
924 __lruvec_stat_mod_foli 902 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
925 if (folio_test_pmd_map 903 if (folio_test_pmd_mappable(folio))
926 __lruvec_stat_ 904 __lruvec_stat_mod_folio(folio,
927 905 NR_FILE_THPS, nr);
928 } 906 }
929 <<
930 unlock: 907 unlock:
931 xas_unlock_irq(&xas); 908 xas_unlock_irq(&xas);
932 !! 909 } while (xas_nomem(&xas, gfp));
933 /* split needed, alloc here an <<
934 if (split_order) { <<
935 xas_split_alloc(&xas, <<
936 if (xas_error(&xas)) <<
937 goto error; <<
938 alloced_shadow = old; <<
939 alloced_order = split_ <<
940 xas_reset(&xas); <<
941 continue; <<
942 } <<
943 <<
944 if (!xas_nomem(&xas, gfp)) <<
945 break; <<
946 } <<
947 910
948 if (xas_error(&xas)) 911 if (xas_error(&xas))
949 goto error; 912 goto error;
950 913
951 trace_mm_filemap_add_to_page_cache(fol 914 trace_mm_filemap_add_to_page_cache(folio);
952 return 0; 915 return 0;
953 error: 916 error:
>> 917 if (charged)
>> 918 mem_cgroup_uncharge(folio);
954 folio->mapping = NULL; 919 folio->mapping = NULL;
955 /* Leave page->index set: truncation r 920 /* Leave page->index set: truncation relies upon it */
956 folio_put_refs(folio, nr); 921 folio_put_refs(folio, nr);
957 return xas_error(&xas); 922 return xas_error(&xas);
958 } 923 }
959 ALLOW_ERROR_INJECTION(__filemap_add_folio, ERR 924 ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO);
960 925
961 int filemap_add_folio(struct address_space *ma 926 int filemap_add_folio(struct address_space *mapping, struct folio *folio,
962 pgoff_t index, 927 pgoff_t index, gfp_t gfp)
963 { 928 {
964 void *shadow = NULL; 929 void *shadow = NULL;
965 int ret; 930 int ret;
966 931
967 ret = mem_cgroup_charge(folio, NULL, g <<
968 if (ret) <<
969 return ret; <<
970 <<
971 __folio_set_locked(folio); 932 __folio_set_locked(folio);
972 ret = __filemap_add_folio(mapping, fol 933 ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow);
973 if (unlikely(ret)) { !! 934 if (unlikely(ret))
974 mem_cgroup_uncharge(folio); <<
975 __folio_clear_locked(folio); 935 __folio_clear_locked(folio);
976 } else { !! 936 else {
977 /* 937 /*
978 * The folio might have been e 938 * The folio might have been evicted from cache only
979 * recently, in which case it 939 * recently, in which case it should be activated like
980 * any other repeatedly access 940 * any other repeatedly accessed folio.
981 * The exception is folios get 941 * The exception is folios getting rewritten; evicting other
982 * data from the working set, 942 * data from the working set, only to cache data that will
983 * get overwritten with someth 943 * get overwritten with something else, is a waste of memory.
984 */ 944 */
985 WARN_ON_ONCE(folio_test_active 945 WARN_ON_ONCE(folio_test_active(folio));
986 if (!(gfp & __GFP_WRITE) && sh 946 if (!(gfp & __GFP_WRITE) && shadow)
987 workingset_refault(fol 947 workingset_refault(folio, shadow);
988 folio_add_lru(folio); 948 folio_add_lru(folio);
989 } 949 }
990 return ret; 950 return ret;
991 } 951 }
992 EXPORT_SYMBOL_GPL(filemap_add_folio); 952 EXPORT_SYMBOL_GPL(filemap_add_folio);
993 953
994 #ifdef CONFIG_NUMA 954 #ifdef CONFIG_NUMA
995 struct folio *filemap_alloc_folio_noprof(gfp_t !! 955 struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order)
996 { 956 {
997 int n; 957 int n;
998 struct folio *folio; 958 struct folio *folio;
999 959
1000 if (cpuset_do_page_mem_spread()) { 960 if (cpuset_do_page_mem_spread()) {
1001 unsigned int cpuset_mems_cook 961 unsigned int cpuset_mems_cookie;
1002 do { 962 do {
1003 cpuset_mems_cookie = 963 cpuset_mems_cookie = read_mems_allowed_begin();
1004 n = cpuset_mem_spread 964 n = cpuset_mem_spread_node();
1005 folio = __folio_alloc !! 965 folio = __folio_alloc_node(gfp, order, n);
1006 } while (!folio && read_mems_ 966 } while (!folio && read_mems_allowed_retry(cpuset_mems_cookie));
1007 967
1008 return folio; 968 return folio;
1009 } 969 }
1010 return folio_alloc_noprof(gfp, order) !! 970 return folio_alloc(gfp, order);
1011 } 971 }
1012 EXPORT_SYMBOL(filemap_alloc_folio_noprof); !! 972 EXPORT_SYMBOL(filemap_alloc_folio);
1013 #endif 973 #endif
1014 974
1015 /* 975 /*
1016 * filemap_invalidate_lock_two - lock invalid 976 * filemap_invalidate_lock_two - lock invalidate_lock for two mappings
1017 * 977 *
1018 * Lock exclusively invalidate_lock of any pa 978 * Lock exclusively invalidate_lock of any passed mapping that is not NULL.
1019 * 979 *
1020 * @mapping1: the first mapping to lock 980 * @mapping1: the first mapping to lock
1021 * @mapping2: the second mapping to lock 981 * @mapping2: the second mapping to lock
1022 */ 982 */
1023 void filemap_invalidate_lock_two(struct addre 983 void filemap_invalidate_lock_two(struct address_space *mapping1,
1024 struct addre 984 struct address_space *mapping2)
1025 { 985 {
1026 if (mapping1 > mapping2) 986 if (mapping1 > mapping2)
1027 swap(mapping1, mapping2); 987 swap(mapping1, mapping2);
1028 if (mapping1) 988 if (mapping1)
1029 down_write(&mapping1->invalid 989 down_write(&mapping1->invalidate_lock);
1030 if (mapping2 && mapping1 != mapping2) 990 if (mapping2 && mapping1 != mapping2)
1031 down_write_nested(&mapping2-> 991 down_write_nested(&mapping2->invalidate_lock, 1);
1032 } 992 }
1033 EXPORT_SYMBOL(filemap_invalidate_lock_two); 993 EXPORT_SYMBOL(filemap_invalidate_lock_two);
1034 994
1035 /* 995 /*
1036 * filemap_invalidate_unlock_two - unlock inv 996 * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings
1037 * 997 *
1038 * Unlock exclusive invalidate_lock of any pa 998 * Unlock exclusive invalidate_lock of any passed mapping that is not NULL.
1039 * 999 *
1040 * @mapping1: the first mapping to unlock 1000 * @mapping1: the first mapping to unlock
1041 * @mapping2: the second mapping to unlock 1001 * @mapping2: the second mapping to unlock
1042 */ 1002 */
1043 void filemap_invalidate_unlock_two(struct add 1003 void filemap_invalidate_unlock_two(struct address_space *mapping1,
1044 struct add 1004 struct address_space *mapping2)
1045 { 1005 {
1046 if (mapping1) 1006 if (mapping1)
1047 up_write(&mapping1->invalidat 1007 up_write(&mapping1->invalidate_lock);
1048 if (mapping2 && mapping1 != mapping2) 1008 if (mapping2 && mapping1 != mapping2)
1049 up_write(&mapping2->invalidat 1009 up_write(&mapping2->invalidate_lock);
1050 } 1010 }
1051 EXPORT_SYMBOL(filemap_invalidate_unlock_two); 1011 EXPORT_SYMBOL(filemap_invalidate_unlock_two);
1052 1012
1053 /* 1013 /*
1054 * In order to wait for pages to become avail 1014 * In order to wait for pages to become available there must be
1055 * waitqueues associated with pages. By using 1015 * waitqueues associated with pages. By using a hash table of
1056 * waitqueues where the bucket discipline is 1016 * waitqueues where the bucket discipline is to maintain all
1057 * waiters on the same queue and wake all whe 1017 * waiters on the same queue and wake all when any of the pages
1058 * become available, and for the woken contex 1018 * become available, and for the woken contexts to check to be
1059 * sure the appropriate page became available 1019 * sure the appropriate page became available, this saves space
1060 * at a cost of "thundering herd" phenomena d 1020 * at a cost of "thundering herd" phenomena during rare hash
1061 * collisions. 1021 * collisions.
1062 */ 1022 */
1063 #define PAGE_WAIT_TABLE_BITS 8 1023 #define PAGE_WAIT_TABLE_BITS 8
1064 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_ 1024 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1065 static wait_queue_head_t folio_wait_table[PAG 1025 static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1066 1026
1067 static wait_queue_head_t *folio_waitqueue(str 1027 static wait_queue_head_t *folio_waitqueue(struct folio *folio)
1068 { 1028 {
1069 return &folio_wait_table[hash_ptr(fol 1029 return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)];
1070 } 1030 }
1071 1031
1072 void __init pagecache_init(void) 1032 void __init pagecache_init(void)
1073 { 1033 {
1074 int i; 1034 int i;
1075 1035
1076 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; 1036 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1077 init_waitqueue_head(&folio_wa 1037 init_waitqueue_head(&folio_wait_table[i]);
1078 1038
1079 page_writeback_init(); 1039 page_writeback_init();
1080 } 1040 }
1081 1041
1082 /* 1042 /*
1083 * The page wait code treats the "wait->flags 1043 * The page wait code treats the "wait->flags" somewhat unusually, because
1084 * we have multiple different kinds of waits, 1044 * we have multiple different kinds of waits, not just the usual "exclusive"
1085 * one. 1045 * one.
1086 * 1046 *
1087 * We have: 1047 * We have:
1088 * 1048 *
1089 * (a) no special bits set: 1049 * (a) no special bits set:
1090 * 1050 *
1091 * We're just waiting for the bit to be 1051 * We're just waiting for the bit to be released, and when a waker
1092 * calls the wakeup function, we set WQ_ 1052 * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1093 * and remove it from the wait queue. 1053 * and remove it from the wait queue.
1094 * 1054 *
1095 * Simple and straightforward. 1055 * Simple and straightforward.
1096 * 1056 *
1097 * (b) WQ_FLAG_EXCLUSIVE: 1057 * (b) WQ_FLAG_EXCLUSIVE:
1098 * 1058 *
1099 * The waiter is waiting to get the lock 1059 * The waiter is waiting to get the lock, and only one waiter should
1100 * be woken up to avoid any thundering h 1060 * be woken up to avoid any thundering herd behavior. We'll set the
1101 * WQ_FLAG_WOKEN bit, wake it up, and re 1061 * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1102 * 1062 *
1103 * This is the traditional exclusive wai 1063 * This is the traditional exclusive wait.
1104 * 1064 *
1105 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM: 1065 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1106 * 1066 *
1107 * The waiter is waiting to get the bit, 1067 * The waiter is waiting to get the bit, and additionally wants the
1108 * lock to be transferred to it for fair 1068 * lock to be transferred to it for fair lock behavior. If the lock
1109 * cannot be taken, we stop walking the 1069 * cannot be taken, we stop walking the wait queue without waking
1110 * the waiter. 1070 * the waiter.
1111 * 1071 *
1112 * This is the "fair lock handoff" case, 1072 * This is the "fair lock handoff" case, and in addition to setting
1113 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to 1073 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1114 * that it now has the lock. 1074 * that it now has the lock.
1115 */ 1075 */
1116 static int wake_page_function(wait_queue_entr 1076 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1117 { 1077 {
1118 unsigned int flags; 1078 unsigned int flags;
1119 struct wait_page_key *key = arg; 1079 struct wait_page_key *key = arg;
1120 struct wait_page_queue *wait_page 1080 struct wait_page_queue *wait_page
1121 = container_of(wait, struct w 1081 = container_of(wait, struct wait_page_queue, wait);
1122 1082
1123 if (!wake_page_match(wait_page, key)) 1083 if (!wake_page_match(wait_page, key))
1124 return 0; 1084 return 0;
1125 1085
1126 /* 1086 /*
1127 * If it's a lock handoff wait, we ge 1087 * If it's a lock handoff wait, we get the bit for it, and
1128 * stop walking (and do not wake it u 1088 * stop walking (and do not wake it up) if we can't.
1129 */ 1089 */
1130 flags = wait->flags; 1090 flags = wait->flags;
1131 if (flags & WQ_FLAG_EXCLUSIVE) { 1091 if (flags & WQ_FLAG_EXCLUSIVE) {
1132 if (test_bit(key->bit_nr, &ke 1092 if (test_bit(key->bit_nr, &key->folio->flags))
1133 return -1; 1093 return -1;
1134 if (flags & WQ_FLAG_CUSTOM) { 1094 if (flags & WQ_FLAG_CUSTOM) {
1135 if (test_and_set_bit( 1095 if (test_and_set_bit(key->bit_nr, &key->folio->flags))
1136 return -1; 1096 return -1;
1137 flags |= WQ_FLAG_DONE 1097 flags |= WQ_FLAG_DONE;
1138 } 1098 }
1139 } 1099 }
1140 1100
1141 /* 1101 /*
1142 * We are holding the wait-queue lock 1102 * We are holding the wait-queue lock, but the waiter that
1143 * is waiting for this will be checki 1103 * is waiting for this will be checking the flags without
1144 * any locking. 1104 * any locking.
1145 * 1105 *
1146 * So update the flags atomically, an 1106 * So update the flags atomically, and wake up the waiter
1147 * afterwards to avoid any races. Thi 1107 * afterwards to avoid any races. This store-release pairs
1148 * with the load-acquire in folio_wai 1108 * with the load-acquire in folio_wait_bit_common().
1149 */ 1109 */
1150 smp_store_release(&wait->flags, flags 1110 smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1151 wake_up_state(wait->private, mode); 1111 wake_up_state(wait->private, mode);
1152 1112
1153 /* 1113 /*
1154 * Ok, we have successfully done what 1114 * Ok, we have successfully done what we're waiting for,
1155 * and we can unconditionally remove 1115 * and we can unconditionally remove the wait entry.
1156 * 1116 *
1157 * Note that this pairs with the "fin 1117 * Note that this pairs with the "finish_wait()" in the
1158 * waiter, and has to be the absolute 1118 * waiter, and has to be the absolute last thing we do.
1159 * After this list_del_init(&wait->en 1119 * After this list_del_init(&wait->entry) the wait entry
1160 * might be de-allocated and the proc 1120 * might be de-allocated and the process might even have
1161 * exited. 1121 * exited.
1162 */ 1122 */
1163 list_del_init_careful(&wait->entry); 1123 list_del_init_careful(&wait->entry);
1164 return (flags & WQ_FLAG_EXCLUSIVE) != 1124 return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1165 } 1125 }
1166 1126
1167 static void folio_wake_bit(struct folio *foli 1127 static void folio_wake_bit(struct folio *folio, int bit_nr)
1168 { 1128 {
1169 wait_queue_head_t *q = folio_waitqueu 1129 wait_queue_head_t *q = folio_waitqueue(folio);
1170 struct wait_page_key key; 1130 struct wait_page_key key;
1171 unsigned long flags; 1131 unsigned long flags;
>> 1132 wait_queue_entry_t bookmark;
1172 1133
1173 key.folio = folio; 1134 key.folio = folio;
1174 key.bit_nr = bit_nr; 1135 key.bit_nr = bit_nr;
1175 key.page_match = 0; 1136 key.page_match = 0;
1176 1137
>> 1138 bookmark.flags = 0;
>> 1139 bookmark.private = NULL;
>> 1140 bookmark.func = NULL;
>> 1141 INIT_LIST_HEAD(&bookmark.entry);
>> 1142
1177 spin_lock_irqsave(&q->lock, flags); 1143 spin_lock_irqsave(&q->lock, flags);
1178 __wake_up_locked_key(q, TASK_NORMAL, !! 1144 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
>> 1145
>> 1146 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
>> 1147 /*
>> 1148 * Take a breather from holding the lock,
>> 1149 * allow pages that finish wake up asynchronously
>> 1150 * to acquire the lock and remove themselves
>> 1151 * from wait queue
>> 1152 */
>> 1153 spin_unlock_irqrestore(&q->lock, flags);
>> 1154 cpu_relax();
>> 1155 spin_lock_irqsave(&q->lock, flags);
>> 1156 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
>> 1157 }
1179 1158
1180 /* 1159 /*
1181 * It's possible to miss clearing wai 1160 * It's possible to miss clearing waiters here, when we woke our page
1182 * waiters, but the hashed waitqueue 1161 * waiters, but the hashed waitqueue has waiters for other pages on it.
1183 * That's okay, it's a rare case. The 1162 * That's okay, it's a rare case. The next waker will clear it.
1184 * 1163 *
1185 * Note that, depending on the page p 1164 * Note that, depending on the page pool (buddy, hugetlb, ZONE_DEVICE,
1186 * other), the flag may be cleared in 1165 * other), the flag may be cleared in the course of freeing the page;
1187 * but that is not required for corre 1166 * but that is not required for correctness.
1188 */ 1167 */
1189 if (!waitqueue_active(q) || !key.page 1168 if (!waitqueue_active(q) || !key.page_match)
1190 folio_clear_waiters(folio); 1169 folio_clear_waiters(folio);
1191 1170
1192 spin_unlock_irqrestore(&q->lock, flag 1171 spin_unlock_irqrestore(&q->lock, flags);
1193 } 1172 }
1194 1173
>> 1174 static void folio_wake(struct folio *folio, int bit)
>> 1175 {
>> 1176 if (!folio_test_waiters(folio))
>> 1177 return;
>> 1178 folio_wake_bit(folio, bit);
>> 1179 }
>> 1180
1195 /* 1181 /*
1196 * A choice of three behaviors for folio_wait 1182 * A choice of three behaviors for folio_wait_bit_common():
1197 */ 1183 */
1198 enum behavior { 1184 enum behavior {
1199 EXCLUSIVE, /* Hold ref to page a 1185 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1200 * __folio_lock() wai 1186 * __folio_lock() waiting on then setting PG_locked.
1201 */ 1187 */
1202 SHARED, /* Hold ref to page a 1188 SHARED, /* Hold ref to page and check the bit when woken, like
1203 * folio_wait_writeba 1189 * folio_wait_writeback() waiting on PG_writeback.
1204 */ 1190 */
1205 DROP, /* Drop ref to page b 1191 DROP, /* Drop ref to page before wait, no check when woken,
1206 * like folio_put_wai 1192 * like folio_put_wait_locked() on PG_locked.
1207 */ 1193 */
1208 }; 1194 };
1209 1195
1210 /* 1196 /*
1211 * Attempt to check (or get) the folio flag, 1197 * Attempt to check (or get) the folio flag, and mark us done
1212 * if successful. 1198 * if successful.
1213 */ 1199 */
1214 static inline bool folio_trylock_flag(struct 1200 static inline bool folio_trylock_flag(struct folio *folio, int bit_nr,
1215 struc 1201 struct wait_queue_entry *wait)
1216 { 1202 {
1217 if (wait->flags & WQ_FLAG_EXCLUSIVE) 1203 if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1218 if (test_and_set_bit(bit_nr, 1204 if (test_and_set_bit(bit_nr, &folio->flags))
1219 return false; 1205 return false;
1220 } else if (test_bit(bit_nr, &folio->f 1206 } else if (test_bit(bit_nr, &folio->flags))
1221 return false; 1207 return false;
1222 1208
1223 wait->flags |= WQ_FLAG_WOKEN | WQ_FLA 1209 wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1224 return true; 1210 return true;
1225 } 1211 }
1226 1212
1227 /* How many times do we accept lock stealing 1213 /* How many times do we accept lock stealing from under a waiter? */
1228 int sysctl_page_lock_unfairness = 5; 1214 int sysctl_page_lock_unfairness = 5;
1229 1215
1230 static inline int folio_wait_bit_common(struc 1216 static inline int folio_wait_bit_common(struct folio *folio, int bit_nr,
1231 int state, enum behavior beha 1217 int state, enum behavior behavior)
1232 { 1218 {
1233 wait_queue_head_t *q = folio_waitqueu 1219 wait_queue_head_t *q = folio_waitqueue(folio);
1234 int unfairness = sysctl_page_lock_unf 1220 int unfairness = sysctl_page_lock_unfairness;
1235 struct wait_page_queue wait_page; 1221 struct wait_page_queue wait_page;
1236 wait_queue_entry_t *wait = &wait_page 1222 wait_queue_entry_t *wait = &wait_page.wait;
1237 bool thrashing = false; 1223 bool thrashing = false;
>> 1224 bool delayacct = false;
1238 unsigned long pflags; 1225 unsigned long pflags;
1239 bool in_thrashing; <<
1240 1226
1241 if (bit_nr == PG_locked && 1227 if (bit_nr == PG_locked &&
1242 !folio_test_uptodate(folio) && fo 1228 !folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1243 delayacct_thrashing_start(&in !! 1229 if (!folio_test_swapbacked(folio)) {
>> 1230 delayacct_thrashing_start();
>> 1231 delayacct = true;
>> 1232 }
1244 psi_memstall_enter(&pflags); 1233 psi_memstall_enter(&pflags);
1245 thrashing = true; 1234 thrashing = true;
1246 } 1235 }
1247 1236
1248 init_wait(wait); 1237 init_wait(wait);
1249 wait->func = wake_page_function; 1238 wait->func = wake_page_function;
1250 wait_page.folio = folio; 1239 wait_page.folio = folio;
1251 wait_page.bit_nr = bit_nr; 1240 wait_page.bit_nr = bit_nr;
1252 1241
1253 repeat: 1242 repeat:
1254 wait->flags = 0; 1243 wait->flags = 0;
1255 if (behavior == EXCLUSIVE) { 1244 if (behavior == EXCLUSIVE) {
1256 wait->flags = WQ_FLAG_EXCLUSI 1245 wait->flags = WQ_FLAG_EXCLUSIVE;
1257 if (--unfairness < 0) 1246 if (--unfairness < 0)
1258 wait->flags |= WQ_FLA 1247 wait->flags |= WQ_FLAG_CUSTOM;
1259 } 1248 }
1260 1249
1261 /* 1250 /*
1262 * Do one last check whether we can g 1251 * Do one last check whether we can get the
1263 * page bit synchronously. 1252 * page bit synchronously.
1264 * 1253 *
1265 * Do the folio_set_waiters() marking 1254 * Do the folio_set_waiters() marking before that
1266 * to let any waker we _just_ missed 1255 * to let any waker we _just_ missed know they
1267 * need to wake us up (otherwise they 1256 * need to wake us up (otherwise they'll never
1268 * even go to the slow case that look 1257 * even go to the slow case that looks at the
1269 * page queue), and add ourselves to 1258 * page queue), and add ourselves to the wait
1270 * queue if we need to sleep. 1259 * queue if we need to sleep.
1271 * 1260 *
1272 * This part needs to be done under t 1261 * This part needs to be done under the queue
1273 * lock to avoid races. 1262 * lock to avoid races.
1274 */ 1263 */
1275 spin_lock_irq(&q->lock); 1264 spin_lock_irq(&q->lock);
1276 folio_set_waiters(folio); 1265 folio_set_waiters(folio);
1277 if (!folio_trylock_flag(folio, bit_nr 1266 if (!folio_trylock_flag(folio, bit_nr, wait))
1278 __add_wait_queue_entry_tail(q 1267 __add_wait_queue_entry_tail(q, wait);
1279 spin_unlock_irq(&q->lock); 1268 spin_unlock_irq(&q->lock);
1280 1269
1281 /* 1270 /*
1282 * From now on, all the logic will be 1271 * From now on, all the logic will be based on
1283 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE 1272 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1284 * see whether the page bit testing h 1273 * see whether the page bit testing has already
1285 * been done by the wake function. 1274 * been done by the wake function.
1286 * 1275 *
1287 * We can drop our reference to the f 1276 * We can drop our reference to the folio.
1288 */ 1277 */
1289 if (behavior == DROP) 1278 if (behavior == DROP)
1290 folio_put(folio); 1279 folio_put(folio);
1291 1280
1292 /* 1281 /*
1293 * Note that until the "finish_wait() 1282 * Note that until the "finish_wait()", or until
1294 * we see the WQ_FLAG_WOKEN flag, we 1283 * we see the WQ_FLAG_WOKEN flag, we need to
1295 * be very careful with the 'wait->fl 1284 * be very careful with the 'wait->flags', because
1296 * we may race with a waker that sets 1285 * we may race with a waker that sets them.
1297 */ 1286 */
1298 for (;;) { 1287 for (;;) {
1299 unsigned int flags; 1288 unsigned int flags;
1300 1289
1301 set_current_state(state); 1290 set_current_state(state);
1302 1291
1303 /* Loop until we've been woke 1292 /* Loop until we've been woken or interrupted */
1304 flags = smp_load_acquire(&wai 1293 flags = smp_load_acquire(&wait->flags);
1305 if (!(flags & WQ_FLAG_WOKEN)) 1294 if (!(flags & WQ_FLAG_WOKEN)) {
1306 if (signal_pending_st 1295 if (signal_pending_state(state, current))
1307 break; 1296 break;
1308 1297
1309 io_schedule(); 1298 io_schedule();
1310 continue; 1299 continue;
1311 } 1300 }
1312 1301
1313 /* If we were non-exclusive, 1302 /* If we were non-exclusive, we're done */
1314 if (behavior != EXCLUSIVE) 1303 if (behavior != EXCLUSIVE)
1315 break; 1304 break;
1316 1305
1317 /* If the waker got the lock 1306 /* If the waker got the lock for us, we're done */
1318 if (flags & WQ_FLAG_DONE) 1307 if (flags & WQ_FLAG_DONE)
1319 break; 1308 break;
1320 1309
1321 /* 1310 /*
1322 * Otherwise, if we're gettin 1311 * Otherwise, if we're getting the lock, we need to
1323 * try to get it ourselves. 1312 * try to get it ourselves.
1324 * 1313 *
1325 * And if that fails, we'll h 1314 * And if that fails, we'll have to retry this all.
1326 */ 1315 */
1327 if (unlikely(test_and_set_bit 1316 if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0))))
1328 goto repeat; 1317 goto repeat;
1329 1318
1330 wait->flags |= WQ_FLAG_DONE; 1319 wait->flags |= WQ_FLAG_DONE;
1331 break; 1320 break;
1332 } 1321 }
1333 1322
1334 /* 1323 /*
1335 * If a signal happened, this 'finish 1324 * If a signal happened, this 'finish_wait()' may remove the last
1336 * waiter from the wait-queues, but t 1325 * waiter from the wait-queues, but the folio waiters bit will remain
1337 * set. That's ok. The next wakeup wi 1326 * set. That's ok. The next wakeup will take care of it, and trying
1338 * to do it here would be difficult a 1327 * to do it here would be difficult and prone to races.
1339 */ 1328 */
1340 finish_wait(q, wait); 1329 finish_wait(q, wait);
1341 1330
1342 if (thrashing) { 1331 if (thrashing) {
1343 delayacct_thrashing_end(&in_t !! 1332 if (delayacct)
>> 1333 delayacct_thrashing_end();
1344 psi_memstall_leave(&pflags); 1334 psi_memstall_leave(&pflags);
1345 } 1335 }
1346 1336
1347 /* 1337 /*
1348 * NOTE! The wait->flags weren't stab 1338 * NOTE! The wait->flags weren't stable until we've done the
1349 * 'finish_wait()', and we could have 1339 * 'finish_wait()', and we could have exited the loop above due
1350 * to a signal, and had a wakeup even 1340 * to a signal, and had a wakeup event happen after the signal
1351 * test but before the 'finish_wait() 1341 * test but before the 'finish_wait()'.
1352 * 1342 *
1353 * So only after the finish_wait() ca 1343 * So only after the finish_wait() can we reliably determine
1354 * if we got woken up or not, so we c 1344 * if we got woken up or not, so we can now figure out the final
1355 * return value based on that state w 1345 * return value based on that state without races.
1356 * 1346 *
1357 * Also note that WQ_FLAG_WOKEN is su 1347 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1358 * waiter, but an exclusive one requi 1348 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1359 */ 1349 */
1360 if (behavior == EXCLUSIVE) 1350 if (behavior == EXCLUSIVE)
1361 return wait->flags & WQ_FLAG_ 1351 return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1362 1352
1363 return wait->flags & WQ_FLAG_WOKEN ? 1353 return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1364 } 1354 }
1365 1355
1366 #ifdef CONFIG_MIGRATION 1356 #ifdef CONFIG_MIGRATION
1367 /** 1357 /**
1368 * migration_entry_wait_on_locked - Wait for 1358 * migration_entry_wait_on_locked - Wait for a migration entry to be removed
1369 * @entry: migration swap entry. 1359 * @entry: migration swap entry.
>> 1360 * @ptep: mapped pte pointer. Will return with the ptep unmapped. Only required
>> 1361 * for pte entries, pass NULL for pmd entries.
1370 * @ptl: already locked ptl. This function wi 1362 * @ptl: already locked ptl. This function will drop the lock.
1371 * 1363 *
1372 * Wait for a migration entry referencing the 1364 * Wait for a migration entry referencing the given page to be removed. This is
1373 * equivalent to put_and_wait_on_page_locked( 1365 * equivalent to put_and_wait_on_page_locked(page, TASK_UNINTERRUPTIBLE) except
1374 * this can be called without taking a refere 1366 * this can be called without taking a reference on the page. Instead this
1375 * should be called while holding the ptl for 1367 * should be called while holding the ptl for the migration entry referencing
1376 * the page. 1368 * the page.
1377 * 1369 *
1378 * Returns after unlocking the ptl. !! 1370 * Returns after unmapping and unlocking the pte/ptl with pte_unmap_unlock().
1379 * 1371 *
1380 * This follows the same logic as folio_wait_ 1372 * This follows the same logic as folio_wait_bit_common() so see the comments
1381 * there. 1373 * there.
1382 */ 1374 */
1383 void migration_entry_wait_on_locked(swp_entry !! 1375 void migration_entry_wait_on_locked(swp_entry_t entry, pte_t *ptep,
1384 __releases(ptl) !! 1376 spinlock_t *ptl)
1385 { 1377 {
1386 struct wait_page_queue wait_page; 1378 struct wait_page_queue wait_page;
1387 wait_queue_entry_t *wait = &wait_page 1379 wait_queue_entry_t *wait = &wait_page.wait;
1388 bool thrashing = false; 1380 bool thrashing = false;
>> 1381 bool delayacct = false;
1389 unsigned long pflags; 1382 unsigned long pflags;
1390 bool in_thrashing; <<
1391 wait_queue_head_t *q; 1383 wait_queue_head_t *q;
1392 struct folio *folio = pfn_swap_entry_ !! 1384 struct folio *folio = page_folio(pfn_swap_entry_to_page(entry));
1393 1385
1394 q = folio_waitqueue(folio); 1386 q = folio_waitqueue(folio);
1395 if (!folio_test_uptodate(folio) && fo 1387 if (!folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1396 delayacct_thrashing_start(&in !! 1388 if (!folio_test_swapbacked(folio)) {
>> 1389 delayacct_thrashing_start();
>> 1390 delayacct = true;
>> 1391 }
1397 psi_memstall_enter(&pflags); 1392 psi_memstall_enter(&pflags);
1398 thrashing = true; 1393 thrashing = true;
1399 } 1394 }
1400 1395
1401 init_wait(wait); 1396 init_wait(wait);
1402 wait->func = wake_page_function; 1397 wait->func = wake_page_function;
1403 wait_page.folio = folio; 1398 wait_page.folio = folio;
1404 wait_page.bit_nr = PG_locked; 1399 wait_page.bit_nr = PG_locked;
1405 wait->flags = 0; 1400 wait->flags = 0;
1406 1401
1407 spin_lock_irq(&q->lock); 1402 spin_lock_irq(&q->lock);
1408 folio_set_waiters(folio); 1403 folio_set_waiters(folio);
1409 if (!folio_trylock_flag(folio, PG_loc 1404 if (!folio_trylock_flag(folio, PG_locked, wait))
1410 __add_wait_queue_entry_tail(q 1405 __add_wait_queue_entry_tail(q, wait);
1411 spin_unlock_irq(&q->lock); 1406 spin_unlock_irq(&q->lock);
1412 1407
1413 /* 1408 /*
1414 * If a migration entry exists for th 1409 * If a migration entry exists for the page the migration path must hold
1415 * a valid reference to the page, and 1410 * a valid reference to the page, and it must take the ptl to remove the
1416 * migration entry. So the page is va 1411 * migration entry. So the page is valid until the ptl is dropped.
1417 */ 1412 */
1418 spin_unlock(ptl); !! 1413 if (ptep)
>> 1414 pte_unmap_unlock(ptep, ptl);
>> 1415 else
>> 1416 spin_unlock(ptl);
1419 1417
1420 for (;;) { 1418 for (;;) {
1421 unsigned int flags; 1419 unsigned int flags;
1422 1420
1423 set_current_state(TASK_UNINTE 1421 set_current_state(TASK_UNINTERRUPTIBLE);
1424 1422
1425 /* Loop until we've been woke 1423 /* Loop until we've been woken or interrupted */
1426 flags = smp_load_acquire(&wai 1424 flags = smp_load_acquire(&wait->flags);
1427 if (!(flags & WQ_FLAG_WOKEN)) 1425 if (!(flags & WQ_FLAG_WOKEN)) {
1428 if (signal_pending_st 1426 if (signal_pending_state(TASK_UNINTERRUPTIBLE, current))
1429 break; 1427 break;
1430 1428
1431 io_schedule(); 1429 io_schedule();
1432 continue; 1430 continue;
1433 } 1431 }
1434 break; 1432 break;
1435 } 1433 }
1436 1434
1437 finish_wait(q, wait); 1435 finish_wait(q, wait);
1438 1436
1439 if (thrashing) { 1437 if (thrashing) {
1440 delayacct_thrashing_end(&in_t !! 1438 if (delayacct)
>> 1439 delayacct_thrashing_end();
1441 psi_memstall_leave(&pflags); 1440 psi_memstall_leave(&pflags);
1442 } 1441 }
1443 } 1442 }
1444 #endif 1443 #endif
1445 1444
1446 void folio_wait_bit(struct folio *folio, int 1445 void folio_wait_bit(struct folio *folio, int bit_nr)
1447 { 1446 {
1448 folio_wait_bit_common(folio, bit_nr, 1447 folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1449 } 1448 }
1450 EXPORT_SYMBOL(folio_wait_bit); 1449 EXPORT_SYMBOL(folio_wait_bit);
1451 1450
1452 int folio_wait_bit_killable(struct folio *fol 1451 int folio_wait_bit_killable(struct folio *folio, int bit_nr)
1453 { 1452 {
1454 return folio_wait_bit_common(folio, b 1453 return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED);
1455 } 1454 }
1456 EXPORT_SYMBOL(folio_wait_bit_killable); 1455 EXPORT_SYMBOL(folio_wait_bit_killable);
1457 1456
1458 /** 1457 /**
1459 * folio_put_wait_locked - Drop a reference a 1458 * folio_put_wait_locked - Drop a reference and wait for it to be unlocked
1460 * @folio: The folio to wait for. 1459 * @folio: The folio to wait for.
1461 * @state: The sleep state (TASK_KILLABLE, TA 1460 * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1462 * 1461 *
1463 * The caller should hold a reference on @fol 1462 * The caller should hold a reference on @folio. They expect the page to
1464 * become unlocked relatively soon, but do no 1463 * become unlocked relatively soon, but do not wish to hold up migration
1465 * (for example) by holding the reference whi 1464 * (for example) by holding the reference while waiting for the folio to
1466 * come unlocked. After this function return 1465 * come unlocked. After this function returns, the caller should not
1467 * dereference @folio. 1466 * dereference @folio.
1468 * 1467 *
1469 * Return: 0 if the folio was unlocked or -EI 1468 * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal.
1470 */ 1469 */
1471 static int folio_put_wait_locked(struct folio !! 1470 int folio_put_wait_locked(struct folio *folio, int state)
1472 { 1471 {
1473 return folio_wait_bit_common(folio, P 1472 return folio_wait_bit_common(folio, PG_locked, state, DROP);
1474 } 1473 }
1475 1474
1476 /** 1475 /**
1477 * folio_add_wait_queue - Add an arbitrary wa 1476 * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue
1478 * @folio: Folio defining the wait queue of i 1477 * @folio: Folio defining the wait queue of interest
1479 * @waiter: Waiter to add to the queue 1478 * @waiter: Waiter to add to the queue
1480 * 1479 *
1481 * Add an arbitrary @waiter to the wait queue 1480 * Add an arbitrary @waiter to the wait queue for the nominated @folio.
1482 */ 1481 */
1483 void folio_add_wait_queue(struct folio *folio 1482 void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter)
1484 { 1483 {
1485 wait_queue_head_t *q = folio_waitqueu 1484 wait_queue_head_t *q = folio_waitqueue(folio);
1486 unsigned long flags; 1485 unsigned long flags;
1487 1486
1488 spin_lock_irqsave(&q->lock, flags); 1487 spin_lock_irqsave(&q->lock, flags);
1489 __add_wait_queue_entry_tail(q, waiter 1488 __add_wait_queue_entry_tail(q, waiter);
1490 folio_set_waiters(folio); 1489 folio_set_waiters(folio);
1491 spin_unlock_irqrestore(&q->lock, flag 1490 spin_unlock_irqrestore(&q->lock, flags);
1492 } 1491 }
1493 EXPORT_SYMBOL_GPL(folio_add_wait_queue); 1492 EXPORT_SYMBOL_GPL(folio_add_wait_queue);
1494 1493
>> 1494 #ifndef clear_bit_unlock_is_negative_byte
>> 1495
>> 1496 /*
>> 1497 * PG_waiters is the high bit in the same byte as PG_lock.
>> 1498 *
>> 1499 * On x86 (and on many other architectures), we can clear PG_lock and
>> 1500 * test the sign bit at the same time. But if the architecture does
>> 1501 * not support that special operation, we just do this all by hand
>> 1502 * instead.
>> 1503 *
>> 1504 * The read of PG_waiters has to be after (or concurrently with) PG_locked
>> 1505 * being cleared, but a memory barrier should be unnecessary since it is
>> 1506 * in the same byte as PG_locked.
>> 1507 */
>> 1508 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
>> 1509 {
>> 1510 clear_bit_unlock(nr, mem);
>> 1511 /* smp_mb__after_atomic(); */
>> 1512 return test_bit(PG_waiters, mem);
>> 1513 }
>> 1514
>> 1515 #endif
>> 1516
1495 /** 1517 /**
1496 * folio_unlock - Unlock a locked folio. 1518 * folio_unlock - Unlock a locked folio.
1497 * @folio: The folio. 1519 * @folio: The folio.
1498 * 1520 *
1499 * Unlocks the folio and wakes up any thread 1521 * Unlocks the folio and wakes up any thread sleeping on the page lock.
1500 * 1522 *
1501 * Context: May be called from interrupt or p 1523 * Context: May be called from interrupt or process context. May not be
1502 * called from NMI context. 1524 * called from NMI context.
1503 */ 1525 */
1504 void folio_unlock(struct folio *folio) 1526 void folio_unlock(struct folio *folio)
1505 { 1527 {
1506 /* Bit 7 allows x86 to check the byte 1528 /* Bit 7 allows x86 to check the byte's sign bit */
1507 BUILD_BUG_ON(PG_waiters != 7); 1529 BUILD_BUG_ON(PG_waiters != 7);
1508 BUILD_BUG_ON(PG_locked > 7); 1530 BUILD_BUG_ON(PG_locked > 7);
1509 VM_BUG_ON_FOLIO(!folio_test_locked(fo 1531 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1510 if (folio_xor_flags_has_waiters(folio !! 1532 if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0)))
1511 folio_wake_bit(folio, PG_lock 1533 folio_wake_bit(folio, PG_locked);
1512 } 1534 }
1513 EXPORT_SYMBOL(folio_unlock); 1535 EXPORT_SYMBOL(folio_unlock);
1514 1536
1515 /** 1537 /**
1516 * folio_end_read - End read on a folio. <<
1517 * @folio: The folio. <<
1518 * @success: True if all reads completed succ <<
1519 * <<
1520 * When all reads against a folio have comple <<
1521 * call this function to let the pagecache kn <<
1522 * are outstanding. This will unlock the fol <<
1523 * sleeping on the lock. The folio will also <<
1524 * reads succeeded. <<
1525 * <<
1526 * Context: May be called from interrupt or p <<
1527 * called from NMI context. <<
1528 */ <<
1529 void folio_end_read(struct folio *folio, bool <<
1530 { <<
1531 unsigned long mask = 1 << PG_locked; <<
1532 <<
1533 /* Must be in bottom byte for x86 to <<
1534 BUILD_BUG_ON(PG_uptodate > 7); <<
1535 VM_BUG_ON_FOLIO(!folio_test_locked(fo <<
1536 VM_BUG_ON_FOLIO(folio_test_uptodate(f <<
1537 <<
1538 if (likely(success)) <<
1539 mask |= 1 << PG_uptodate; <<
1540 if (folio_xor_flags_has_waiters(folio <<
1541 folio_wake_bit(folio, PG_lock <<
1542 } <<
1543 EXPORT_SYMBOL(folio_end_read); <<
1544 <<
1545 /** <<
1546 * folio_end_private_2 - Clear PG_private_2 a 1538 * folio_end_private_2 - Clear PG_private_2 and wake any waiters.
1547 * @folio: The folio. 1539 * @folio: The folio.
1548 * 1540 *
1549 * Clear the PG_private_2 bit on a folio and 1541 * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for
1550 * it. The folio reference held for PG_priva 1542 * it. The folio reference held for PG_private_2 being set is released.
1551 * 1543 *
1552 * This is, for example, used when a netfs fo 1544 * This is, for example, used when a netfs folio is being written to a local
1553 * disk cache, thereby allowing writes to the 1545 * disk cache, thereby allowing writes to the cache for the same folio to be
1554 * serialised. 1546 * serialised.
1555 */ 1547 */
1556 void folio_end_private_2(struct folio *folio) 1548 void folio_end_private_2(struct folio *folio)
1557 { 1549 {
1558 VM_BUG_ON_FOLIO(!folio_test_private_2 1550 VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio);
1559 clear_bit_unlock(PG_private_2, folio_ 1551 clear_bit_unlock(PG_private_2, folio_flags(folio, 0));
1560 folio_wake_bit(folio, PG_private_2); 1552 folio_wake_bit(folio, PG_private_2);
1561 folio_put(folio); 1553 folio_put(folio);
1562 } 1554 }
1563 EXPORT_SYMBOL(folio_end_private_2); 1555 EXPORT_SYMBOL(folio_end_private_2);
1564 1556
1565 /** 1557 /**
1566 * folio_wait_private_2 - Wait for PG_private 1558 * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio.
1567 * @folio: The folio to wait on. 1559 * @folio: The folio to wait on.
1568 * 1560 *
1569 * Wait for PG_private_2 to be cleared on a f !! 1561 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio.
1570 */ 1562 */
1571 void folio_wait_private_2(struct folio *folio 1563 void folio_wait_private_2(struct folio *folio)
1572 { 1564 {
1573 while (folio_test_private_2(folio)) 1565 while (folio_test_private_2(folio))
1574 folio_wait_bit(folio, PG_priv 1566 folio_wait_bit(folio, PG_private_2);
1575 } 1567 }
1576 EXPORT_SYMBOL(folio_wait_private_2); 1568 EXPORT_SYMBOL(folio_wait_private_2);
1577 1569
1578 /** 1570 /**
1579 * folio_wait_private_2_killable - Wait for P 1571 * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio.
1580 * @folio: The folio to wait on. 1572 * @folio: The folio to wait on.
1581 * 1573 *
1582 * Wait for PG_private_2 to be cleared on a f !! 1574 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a
1583 * received by the calling task. !! 1575 * fatal signal is received by the calling task.
1584 * 1576 *
1585 * Return: 1577 * Return:
1586 * - 0 if successful. 1578 * - 0 if successful.
1587 * - -EINTR if a fatal signal was encountered 1579 * - -EINTR if a fatal signal was encountered.
1588 */ 1580 */
1589 int folio_wait_private_2_killable(struct foli 1581 int folio_wait_private_2_killable(struct folio *folio)
1590 { 1582 {
1591 int ret = 0; 1583 int ret = 0;
1592 1584
1593 while (folio_test_private_2(folio)) { 1585 while (folio_test_private_2(folio)) {
1594 ret = folio_wait_bit_killable 1586 ret = folio_wait_bit_killable(folio, PG_private_2);
1595 if (ret < 0) 1587 if (ret < 0)
1596 break; 1588 break;
1597 } 1589 }
1598 1590
1599 return ret; 1591 return ret;
1600 } 1592 }
1601 EXPORT_SYMBOL(folio_wait_private_2_killable); 1593 EXPORT_SYMBOL(folio_wait_private_2_killable);
1602 1594
1603 /** 1595 /**
1604 * folio_end_writeback - End writeback agains 1596 * folio_end_writeback - End writeback against a folio.
1605 * @folio: The folio. 1597 * @folio: The folio.
1606 * <<
1607 * The folio must actually be under writeback <<
1608 * <<
1609 * Context: May be called from process or int <<
1610 */ 1598 */
1611 void folio_end_writeback(struct folio *folio) 1599 void folio_end_writeback(struct folio *folio)
1612 { 1600 {
1613 VM_BUG_ON_FOLIO(!folio_test_writeback <<
1614 <<
1615 /* 1601 /*
1616 * folio_test_clear_reclaim() could b 1602 * folio_test_clear_reclaim() could be used here but it is an
1617 * atomic operation and overkill in t 1603 * atomic operation and overkill in this particular case. Failing
1618 * to shuffle a folio marked for imme 1604 * to shuffle a folio marked for immediate reclaim is too mild
1619 * a gain to justify taking an atomic 1605 * a gain to justify taking an atomic operation penalty at the
1620 * end of every folio writeback. 1606 * end of every folio writeback.
1621 */ 1607 */
1622 if (folio_test_reclaim(folio)) { 1608 if (folio_test_reclaim(folio)) {
1623 folio_clear_reclaim(folio); 1609 folio_clear_reclaim(folio);
1624 folio_rotate_reclaimable(foli 1610 folio_rotate_reclaimable(folio);
1625 } 1611 }
1626 1612
1627 /* 1613 /*
1628 * Writeback does not hold a folio re 1614 * Writeback does not hold a folio reference of its own, relying
1629 * on truncation to wait for the clea 1615 * on truncation to wait for the clearing of PG_writeback.
1630 * But here we must make sure that th 1616 * But here we must make sure that the folio is not freed and
1631 * reused before the folio_wake_bit() !! 1617 * reused before the folio_wake().
1632 */ 1618 */
1633 folio_get(folio); 1619 folio_get(folio);
1634 if (__folio_end_writeback(folio)) !! 1620 if (!__folio_end_writeback(folio))
1635 folio_wake_bit(folio, PG_writ !! 1621 BUG();
>> 1622
>> 1623 smp_mb__after_atomic();
>> 1624 folio_wake(folio, PG_writeback);
1636 acct_reclaim_writeback(folio); 1625 acct_reclaim_writeback(folio);
1637 folio_put(folio); 1626 folio_put(folio);
1638 } 1627 }
1639 EXPORT_SYMBOL(folio_end_writeback); 1628 EXPORT_SYMBOL(folio_end_writeback);
1640 1629
>> 1630 /*
>> 1631 * After completing I/O on a page, call this routine to update the page
>> 1632 * flags appropriately
>> 1633 */
>> 1634 void page_endio(struct page *page, bool is_write, int err)
>> 1635 {
>> 1636 if (!is_write) {
>> 1637 if (!err) {
>> 1638 SetPageUptodate(page);
>> 1639 } else {
>> 1640 ClearPageUptodate(page);
>> 1641 SetPageError(page);
>> 1642 }
>> 1643 unlock_page(page);
>> 1644 } else {
>> 1645 if (err) {
>> 1646 struct address_space *mapping;
>> 1647
>> 1648 SetPageError(page);
>> 1649 mapping = page_mapping(page);
>> 1650 if (mapping)
>> 1651 mapping_set_error(mapping, err);
>> 1652 }
>> 1653 end_page_writeback(page);
>> 1654 }
>> 1655 }
>> 1656 EXPORT_SYMBOL_GPL(page_endio);
>> 1657
1641 /** 1658 /**
1642 * __folio_lock - Get a lock on the folio, as 1659 * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it.
1643 * @folio: The folio to lock 1660 * @folio: The folio to lock
1644 */ 1661 */
1645 void __folio_lock(struct folio *folio) 1662 void __folio_lock(struct folio *folio)
1646 { 1663 {
1647 folio_wait_bit_common(folio, PG_locke 1664 folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE,
1648 EXCLUSIVE); 1665 EXCLUSIVE);
1649 } 1666 }
1650 EXPORT_SYMBOL(__folio_lock); 1667 EXPORT_SYMBOL(__folio_lock);
1651 1668
1652 int __folio_lock_killable(struct folio *folio 1669 int __folio_lock_killable(struct folio *folio)
1653 { 1670 {
1654 return folio_wait_bit_common(folio, P 1671 return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE,
1655 EXCLU 1672 EXCLUSIVE);
1656 } 1673 }
1657 EXPORT_SYMBOL_GPL(__folio_lock_killable); 1674 EXPORT_SYMBOL_GPL(__folio_lock_killable);
1658 1675
1659 static int __folio_lock_async(struct folio *f 1676 static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait)
1660 { 1677 {
1661 struct wait_queue_head *q = folio_wai 1678 struct wait_queue_head *q = folio_waitqueue(folio);
1662 int ret; !! 1679 int ret = 0;
1663 1680
1664 wait->folio = folio; 1681 wait->folio = folio;
1665 wait->bit_nr = PG_locked; 1682 wait->bit_nr = PG_locked;
1666 1683
1667 spin_lock_irq(&q->lock); 1684 spin_lock_irq(&q->lock);
1668 __add_wait_queue_entry_tail(q, &wait- 1685 __add_wait_queue_entry_tail(q, &wait->wait);
1669 folio_set_waiters(folio); 1686 folio_set_waiters(folio);
1670 ret = !folio_trylock(folio); 1687 ret = !folio_trylock(folio);
1671 /* 1688 /*
1672 * If we were successful now, we know 1689 * If we were successful now, we know we're still on the
1673 * waitqueue as we're still under the 1690 * waitqueue as we're still under the lock. This means it's
1674 * safe to remove and return success, 1691 * safe to remove and return success, we know the callback
1675 * isn't going to trigger. 1692 * isn't going to trigger.
1676 */ 1693 */
1677 if (!ret) 1694 if (!ret)
1678 __remove_wait_queue(q, &wait- 1695 __remove_wait_queue(q, &wait->wait);
1679 else 1696 else
1680 ret = -EIOCBQUEUED; 1697 ret = -EIOCBQUEUED;
1681 spin_unlock_irq(&q->lock); 1698 spin_unlock_irq(&q->lock);
1682 return ret; 1699 return ret;
1683 } 1700 }
1684 1701
1685 /* 1702 /*
1686 * Return values: 1703 * Return values:
1687 * 0 - folio is locked. !! 1704 * true - folio is locked; mmap_lock is still held.
1688 * non-zero - folio is not locked. !! 1705 * false - folio is not locked.
1689 * mmap_lock or per-VMA lock has been rel !! 1706 * mmap_lock has been released (mmap_read_unlock(), unless flags had both
1690 * vma_end_read()), unless flags had both !! 1707 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1691 * FAULT_FLAG_RETRY_NOWAIT set, in which !! 1708 * which case mmap_lock is still held.
1692 * 1709 *
1693 * If neither ALLOW_RETRY nor KILLABLE are se !! 1710 * If neither ALLOW_RETRY nor KILLABLE are set, will always return true
1694 * with the folio locked and the mmap_lock/pe !! 1711 * with the folio locked and the mmap_lock unperturbed.
1695 */ 1712 */
1696 vm_fault_t __folio_lock_or_retry(struct folio !! 1713 bool __folio_lock_or_retry(struct folio *folio, struct mm_struct *mm,
>> 1714 unsigned int flags)
1697 { 1715 {
1698 unsigned int flags = vmf->flags; <<
1699 <<
1700 if (fault_flag_allow_retry_first(flag 1716 if (fault_flag_allow_retry_first(flags)) {
1701 /* 1717 /*
1702 * CAUTION! In this case, mma !! 1718 * CAUTION! In this case, mmap_lock is not released
1703 * released even though retur !! 1719 * even though return 0.
1704 */ 1720 */
1705 if (flags & FAULT_FLAG_RETRY_ 1721 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1706 return VM_FAULT_RETRY !! 1722 return false;
1707 1723
1708 release_fault_lock(vmf); !! 1724 mmap_read_unlock(mm);
1709 if (flags & FAULT_FLAG_KILLAB 1725 if (flags & FAULT_FLAG_KILLABLE)
1710 folio_wait_locked_kil 1726 folio_wait_locked_killable(folio);
1711 else 1727 else
1712 folio_wait_locked(fol 1728 folio_wait_locked(folio);
1713 return VM_FAULT_RETRY; !! 1729 return false;
1714 } 1730 }
1715 if (flags & FAULT_FLAG_KILLABLE) { 1731 if (flags & FAULT_FLAG_KILLABLE) {
1716 bool ret; 1732 bool ret;
1717 1733
1718 ret = __folio_lock_killable(f 1734 ret = __folio_lock_killable(folio);
1719 if (ret) { 1735 if (ret) {
1720 release_fault_lock(vm !! 1736 mmap_read_unlock(mm);
1721 return VM_FAULT_RETRY !! 1737 return false;
1722 } 1738 }
1723 } else { 1739 } else {
1724 __folio_lock(folio); 1740 __folio_lock(folio);
1725 } 1741 }
1726 1742
1727 return 0; !! 1743 return true;
1728 } 1744 }
1729 1745
1730 /** 1746 /**
1731 * page_cache_next_miss() - Find the next gap 1747 * page_cache_next_miss() - Find the next gap in the page cache.
1732 * @mapping: Mapping. 1748 * @mapping: Mapping.
1733 * @index: Index. 1749 * @index: Index.
1734 * @max_scan: Maximum range to search. 1750 * @max_scan: Maximum range to search.
1735 * 1751 *
1736 * Search the range [index, min(index + max_s 1752 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1737 * gap with the lowest index. 1753 * gap with the lowest index.
1738 * 1754 *
1739 * This function may be called under the rcu_ 1755 * This function may be called under the rcu_read_lock. However, this will
1740 * not atomically search a snapshot of the ca 1756 * not atomically search a snapshot of the cache at a single point in time.
1741 * For example, if a gap is created at index 1757 * For example, if a gap is created at index 5, then subsequently a gap is
1742 * created at index 10, page_cache_next_miss 1758 * created at index 10, page_cache_next_miss covering both indices may
1743 * return 10 if called under the rcu_read_loc 1759 * return 10 if called under the rcu_read_lock.
1744 * 1760 *
1745 * Return: The index of the gap if found, oth 1761 * Return: The index of the gap if found, otherwise an index outside the
1746 * range specified (in which case 'return - i 1762 * range specified (in which case 'return - index >= max_scan' will be true).
1747 * In the rare case of index wrap-around, 0 w 1763 * In the rare case of index wrap-around, 0 will be returned.
1748 */ 1764 */
1749 pgoff_t page_cache_next_miss(struct address_s 1765 pgoff_t page_cache_next_miss(struct address_space *mapping,
1750 pgoff_t index, u 1766 pgoff_t index, unsigned long max_scan)
1751 { 1767 {
1752 XA_STATE(xas, &mapping->i_pages, inde 1768 XA_STATE(xas, &mapping->i_pages, index);
1753 1769
1754 while (max_scan--) { 1770 while (max_scan--) {
1755 void *entry = xas_next(&xas); 1771 void *entry = xas_next(&xas);
1756 if (!entry || xa_is_value(ent 1772 if (!entry || xa_is_value(entry))
1757 return xas.xa_index; !! 1773 break;
1758 if (xas.xa_index == 0) 1774 if (xas.xa_index == 0)
1759 return 0; !! 1775 break;
1760 } 1776 }
1761 1777
1762 return index + max_scan; !! 1778 return xas.xa_index;
1763 } 1779 }
1764 EXPORT_SYMBOL(page_cache_next_miss); 1780 EXPORT_SYMBOL(page_cache_next_miss);
1765 1781
1766 /** 1782 /**
1767 * page_cache_prev_miss() - Find the previous 1783 * page_cache_prev_miss() - Find the previous gap in the page cache.
1768 * @mapping: Mapping. 1784 * @mapping: Mapping.
1769 * @index: Index. 1785 * @index: Index.
1770 * @max_scan: Maximum range to search. 1786 * @max_scan: Maximum range to search.
1771 * 1787 *
1772 * Search the range [max(index - max_scan + 1 1788 * Search the range [max(index - max_scan + 1, 0), index] for the
1773 * gap with the highest index. 1789 * gap with the highest index.
1774 * 1790 *
1775 * This function may be called under the rcu_ 1791 * This function may be called under the rcu_read_lock. However, this will
1776 * not atomically search a snapshot of the ca 1792 * not atomically search a snapshot of the cache at a single point in time.
1777 * For example, if a gap is created at index 1793 * For example, if a gap is created at index 10, then subsequently a gap is
1778 * created at index 5, page_cache_prev_miss() 1794 * created at index 5, page_cache_prev_miss() covering both indices may
1779 * return 5 if called under the rcu_read_lock 1795 * return 5 if called under the rcu_read_lock.
1780 * 1796 *
1781 * Return: The index of the gap if found, oth 1797 * Return: The index of the gap if found, otherwise an index outside the
1782 * range specified (in which case 'index - re 1798 * range specified (in which case 'index - return >= max_scan' will be true).
1783 * In the rare case of wrap-around, ULONG_MAX 1799 * In the rare case of wrap-around, ULONG_MAX will be returned.
1784 */ 1800 */
1785 pgoff_t page_cache_prev_miss(struct address_s 1801 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1786 pgoff_t index, u 1802 pgoff_t index, unsigned long max_scan)
1787 { 1803 {
1788 XA_STATE(xas, &mapping->i_pages, inde 1804 XA_STATE(xas, &mapping->i_pages, index);
1789 1805
1790 while (max_scan--) { 1806 while (max_scan--) {
1791 void *entry = xas_prev(&xas); 1807 void *entry = xas_prev(&xas);
1792 if (!entry || xa_is_value(ent 1808 if (!entry || xa_is_value(entry))
1793 break; 1809 break;
1794 if (xas.xa_index == ULONG_MAX 1810 if (xas.xa_index == ULONG_MAX)
1795 break; 1811 break;
1796 } 1812 }
1797 1813
1798 return xas.xa_index; 1814 return xas.xa_index;
1799 } 1815 }
1800 EXPORT_SYMBOL(page_cache_prev_miss); 1816 EXPORT_SYMBOL(page_cache_prev_miss);
1801 1817
1802 /* 1818 /*
1803 * Lockless page cache protocol: 1819 * Lockless page cache protocol:
1804 * On the lookup side: 1820 * On the lookup side:
1805 * 1. Load the folio from i_pages 1821 * 1. Load the folio from i_pages
1806 * 2. Increment the refcount if it's not zero 1822 * 2. Increment the refcount if it's not zero
1807 * 3. If the folio is not found by xas_reload 1823 * 3. If the folio is not found by xas_reload(), put the refcount and retry
1808 * 1824 *
1809 * On the removal side: 1825 * On the removal side:
1810 * A. Freeze the page (by zeroing the refcoun 1826 * A. Freeze the page (by zeroing the refcount if nobody else has a reference)
1811 * B. Remove the page from i_pages 1827 * B. Remove the page from i_pages
1812 * C. Return the page to the page allocator 1828 * C. Return the page to the page allocator
1813 * 1829 *
1814 * This means that any page may have its refe 1830 * This means that any page may have its reference count temporarily
1815 * increased by a speculative page cache (or !! 1831 * increased by a speculative page cache (or fast GUP) lookup as it can
1816 * be allocated by another user before the RC 1832 * be allocated by another user before the RCU grace period expires.
1817 * Because the refcount temporarily acquired 1833 * Because the refcount temporarily acquired here may end up being the
1818 * last refcount on the page, any page alloca 1834 * last refcount on the page, any page allocation must be freeable by
1819 * folio_put(). 1835 * folio_put().
1820 */ 1836 */
1821 1837
1822 /* 1838 /*
1823 * filemap_get_entry - Get a page cache entry !! 1839 * mapping_get_entry - Get a page cache entry.
1824 * @mapping: the address_space to search 1840 * @mapping: the address_space to search
1825 * @index: The page cache index. 1841 * @index: The page cache index.
1826 * 1842 *
1827 * Looks up the page cache entry at @mapping 1843 * Looks up the page cache entry at @mapping & @index. If it is a folio,
1828 * it is returned with an increased refcount. 1844 * it is returned with an increased refcount. If it is a shadow entry
1829 * of a previously evicted folio, or a swap e 1845 * of a previously evicted folio, or a swap entry from shmem/tmpfs,
1830 * it is returned without further action. 1846 * it is returned without further action.
1831 * 1847 *
1832 * Return: The folio, swap or shadow entry, % 1848 * Return: The folio, swap or shadow entry, %NULL if nothing is found.
1833 */ 1849 */
1834 void *filemap_get_entry(struct address_space !! 1850 static void *mapping_get_entry(struct address_space *mapping, pgoff_t index)
1835 { 1851 {
1836 XA_STATE(xas, &mapping->i_pages, inde 1852 XA_STATE(xas, &mapping->i_pages, index);
1837 struct folio *folio; 1853 struct folio *folio;
1838 1854
1839 rcu_read_lock(); 1855 rcu_read_lock();
1840 repeat: 1856 repeat:
1841 xas_reset(&xas); 1857 xas_reset(&xas);
1842 folio = xas_load(&xas); 1858 folio = xas_load(&xas);
1843 if (xas_retry(&xas, folio)) 1859 if (xas_retry(&xas, folio))
1844 goto repeat; 1860 goto repeat;
1845 /* 1861 /*
1846 * A shadow entry of a recently evict 1862 * A shadow entry of a recently evicted page, or a swap entry from
1847 * shmem/tmpfs. Return it without at 1863 * shmem/tmpfs. Return it without attempting to raise page count.
1848 */ 1864 */
1849 if (!folio || xa_is_value(folio)) 1865 if (!folio || xa_is_value(folio))
1850 goto out; 1866 goto out;
1851 1867
1852 if (!folio_try_get(folio)) !! 1868 if (!folio_try_get_rcu(folio))
1853 goto repeat; 1869 goto repeat;
1854 1870
1855 if (unlikely(folio != xas_reload(&xas 1871 if (unlikely(folio != xas_reload(&xas))) {
1856 folio_put(folio); 1872 folio_put(folio);
1857 goto repeat; 1873 goto repeat;
1858 } 1874 }
1859 out: 1875 out:
1860 rcu_read_unlock(); 1876 rcu_read_unlock();
1861 1877
1862 return folio; 1878 return folio;
1863 } 1879 }
1864 1880
1865 /** 1881 /**
1866 * __filemap_get_folio - Find and get a refer 1882 * __filemap_get_folio - Find and get a reference to a folio.
1867 * @mapping: The address_space to search. 1883 * @mapping: The address_space to search.
1868 * @index: The page index. 1884 * @index: The page index.
1869 * @fgp_flags: %FGP flags modify how the foli 1885 * @fgp_flags: %FGP flags modify how the folio is returned.
1870 * @gfp: Memory allocation flags to use if %F 1886 * @gfp: Memory allocation flags to use if %FGP_CREAT is specified.
1871 * 1887 *
1872 * Looks up the page cache entry at @mapping 1888 * Looks up the page cache entry at @mapping & @index.
1873 * 1889 *
>> 1890 * @fgp_flags can be zero or more of these flags:
>> 1891 *
>> 1892 * * %FGP_ACCESSED - The folio will be marked accessed.
>> 1893 * * %FGP_LOCK - The folio is returned locked.
>> 1894 * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it
>> 1895 * instead of allocating a new folio to replace it.
>> 1896 * * %FGP_CREAT - If no page is present then a new page is allocated using
>> 1897 * @gfp and added to the page cache and the VM's LRU list.
>> 1898 * The page is returned locked and with an increased refcount.
>> 1899 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
>> 1900 * page is already in cache. If the page was allocated, unlock it before
>> 1901 * returning so the caller can do the same dance.
>> 1902 * * %FGP_WRITE - The page will be written to by the caller.
>> 1903 * * %FGP_NOFS - __GFP_FS will get cleared in gfp.
>> 1904 * * %FGP_NOWAIT - Don't get blocked by page lock.
>> 1905 * * %FGP_STABLE - Wait for the folio to be stable (finished writeback)
>> 1906 *
1874 * If %FGP_LOCK or %FGP_CREAT are specified t 1907 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1875 * if the %GFP flags specified for %FGP_CREAT 1908 * if the %GFP flags specified for %FGP_CREAT are atomic.
1876 * 1909 *
1877 * If this function returns a folio, it is re !! 1910 * If there is a page cache page, it is returned with an increased refcount.
1878 * 1911 *
1879 * Return: The found folio or an ERR_PTR() ot !! 1912 * Return: The found folio or %NULL otherwise.
1880 */ 1913 */
1881 struct folio *__filemap_get_folio(struct addr 1914 struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
1882 fgf_t fgp_flags, gfp_t gfp) !! 1915 int fgp_flags, gfp_t gfp)
1883 { 1916 {
1884 struct folio *folio; 1917 struct folio *folio;
1885 1918
1886 repeat: 1919 repeat:
1887 folio = filemap_get_entry(mapping, in !! 1920 folio = mapping_get_entry(mapping, index);
1888 if (xa_is_value(folio)) !! 1921 if (xa_is_value(folio)) {
>> 1922 if (fgp_flags & FGP_ENTRY)
>> 1923 return folio;
1889 folio = NULL; 1924 folio = NULL;
>> 1925 }
1890 if (!folio) 1926 if (!folio)
1891 goto no_page; 1927 goto no_page;
1892 1928
1893 if (fgp_flags & FGP_LOCK) { 1929 if (fgp_flags & FGP_LOCK) {
1894 if (fgp_flags & FGP_NOWAIT) { 1930 if (fgp_flags & FGP_NOWAIT) {
1895 if (!folio_trylock(fo 1931 if (!folio_trylock(folio)) {
1896 folio_put(fol 1932 folio_put(folio);
1897 return ERR_PT !! 1933 return NULL;
1898 } 1934 }
1899 } else { 1935 } else {
1900 folio_lock(folio); 1936 folio_lock(folio);
1901 } 1937 }
1902 1938
1903 /* Has the page been truncate 1939 /* Has the page been truncated? */
1904 if (unlikely(folio->mapping ! 1940 if (unlikely(folio->mapping != mapping)) {
1905 folio_unlock(folio); 1941 folio_unlock(folio);
1906 folio_put(folio); 1942 folio_put(folio);
1907 goto repeat; 1943 goto repeat;
1908 } 1944 }
1909 VM_BUG_ON_FOLIO(!folio_contai 1945 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
1910 } 1946 }
1911 1947
1912 if (fgp_flags & FGP_ACCESSED) 1948 if (fgp_flags & FGP_ACCESSED)
1913 folio_mark_accessed(folio); 1949 folio_mark_accessed(folio);
1914 else if (fgp_flags & FGP_WRITE) { 1950 else if (fgp_flags & FGP_WRITE) {
1915 /* Clear idle flag for buffer 1951 /* Clear idle flag for buffer write */
1916 if (folio_test_idle(folio)) 1952 if (folio_test_idle(folio))
1917 folio_clear_idle(foli 1953 folio_clear_idle(folio);
1918 } 1954 }
1919 1955
1920 if (fgp_flags & FGP_STABLE) 1956 if (fgp_flags & FGP_STABLE)
1921 folio_wait_stable(folio); 1957 folio_wait_stable(folio);
1922 no_page: 1958 no_page:
1923 if (!folio && (fgp_flags & FGP_CREAT) 1959 if (!folio && (fgp_flags & FGP_CREAT)) {
1924 unsigned int min_order = mapp <<
1925 unsigned int order = max(min_ <<
1926 int err; 1960 int err;
1927 index = mapping_align_index(m <<
1928 <<
1929 if ((fgp_flags & FGP_WRITE) & 1961 if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1930 gfp |= __GFP_WRITE; 1962 gfp |= __GFP_WRITE;
1931 if (fgp_flags & FGP_NOFS) 1963 if (fgp_flags & FGP_NOFS)
1932 gfp &= ~__GFP_FS; 1964 gfp &= ~__GFP_FS;
1933 if (fgp_flags & FGP_NOWAIT) { 1965 if (fgp_flags & FGP_NOWAIT) {
1934 gfp &= ~GFP_KERNEL; 1966 gfp &= ~GFP_KERNEL;
1935 gfp |= GFP_NOWAIT | _ 1967 gfp |= GFP_NOWAIT | __GFP_NOWARN;
1936 } 1968 }
1937 if (WARN_ON_ONCE(!(fgp_flags <<
1938 fgp_flags |= FGP_LOCK <<
1939 1969
1940 if (order > mapping_max_folio !! 1970 folio = filemap_alloc_folio(gfp, 0);
1941 order = mapping_max_f !! 1971 if (!folio)
1942 /* If we're not aligned, allo !! 1972 return NULL;
1943 if (index & ((1UL << order) - <<
1944 order = __ffs(index); <<
1945 1973
1946 do { !! 1974 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1947 gfp_t alloc_gfp = gfp !! 1975 fgp_flags |= FGP_LOCK;
1948 1976
1949 err = -ENOMEM; !! 1977 /* Init accessed so avoid atomic mark_page_accessed later */
1950 if (order > min_order !! 1978 if (fgp_flags & FGP_ACCESSED)
1951 alloc_gfp |= !! 1979 __folio_set_referenced(folio);
1952 folio = filemap_alloc <<
1953 if (!folio) <<
1954 continue; <<
1955 <<
1956 /* Init accessed so a <<
1957 if (fgp_flags & FGP_A <<
1958 __folio_set_r <<
1959 1980
1960 err = filemap_add_fol !! 1981 err = filemap_add_folio(mapping, folio, index, gfp);
1961 if (!err) !! 1982 if (unlikely(err)) {
1962 break; <<
1963 folio_put(folio); 1983 folio_put(folio);
1964 folio = NULL; 1984 folio = NULL;
1965 } while (order-- > min_order) !! 1985 if (err == -EEXIST)
>> 1986 goto repeat;
>> 1987 }
1966 1988
1967 if (err == -EEXIST) <<
1968 goto repeat; <<
1969 if (err) <<
1970 return ERR_PTR(err); <<
1971 /* 1989 /*
1972 * filemap_add_folio locks th 1990 * filemap_add_folio locks the page, and for mmap
1973 * we expect an unlocked page 1991 * we expect an unlocked page.
1974 */ 1992 */
1975 if (folio && (fgp_flags & FGP 1993 if (folio && (fgp_flags & FGP_FOR_MMAP))
1976 folio_unlock(folio); 1994 folio_unlock(folio);
1977 } 1995 }
1978 1996
1979 if (!folio) <<
1980 return ERR_PTR(-ENOENT); <<
1981 return folio; 1997 return folio;
1982 } 1998 }
1983 EXPORT_SYMBOL(__filemap_get_folio); 1999 EXPORT_SYMBOL(__filemap_get_folio);
1984 2000
1985 static inline struct folio *find_get_entry(st 2001 static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max,
1986 xa_mark_t mark) 2002 xa_mark_t mark)
1987 { 2003 {
1988 struct folio *folio; 2004 struct folio *folio;
1989 2005
1990 retry: 2006 retry:
1991 if (mark == XA_PRESENT) 2007 if (mark == XA_PRESENT)
1992 folio = xas_find(xas, max); 2008 folio = xas_find(xas, max);
1993 else 2009 else
1994 folio = xas_find_marked(xas, 2010 folio = xas_find_marked(xas, max, mark);
1995 2011
1996 if (xas_retry(xas, folio)) 2012 if (xas_retry(xas, folio))
1997 goto retry; 2013 goto retry;
1998 /* 2014 /*
1999 * A shadow entry of a recently evict 2015 * A shadow entry of a recently evicted page, a swap
2000 * entry from shmem/tmpfs or a DAX en 2016 * entry from shmem/tmpfs or a DAX entry. Return it
2001 * without attempting to raise page c 2017 * without attempting to raise page count.
2002 */ 2018 */
2003 if (!folio || xa_is_value(folio)) 2019 if (!folio || xa_is_value(folio))
2004 return folio; 2020 return folio;
2005 2021
2006 if (!folio_try_get(folio)) !! 2022 if (!folio_try_get_rcu(folio))
2007 goto reset; 2023 goto reset;
2008 2024
2009 if (unlikely(folio != xas_reload(xas) 2025 if (unlikely(folio != xas_reload(xas))) {
2010 folio_put(folio); 2026 folio_put(folio);
2011 goto reset; 2027 goto reset;
2012 } 2028 }
2013 2029
2014 return folio; 2030 return folio;
2015 reset: 2031 reset:
2016 xas_reset(xas); 2032 xas_reset(xas);
2017 goto retry; 2033 goto retry;
2018 } 2034 }
2019 2035
2020 /** 2036 /**
2021 * find_get_entries - gang pagecache lookup 2037 * find_get_entries - gang pagecache lookup
2022 * @mapping: The address_space to search 2038 * @mapping: The address_space to search
2023 * @start: The starting page cache index 2039 * @start: The starting page cache index
2024 * @end: The final page index (inclusi 2040 * @end: The final page index (inclusive).
2025 * @fbatch: Where the resulting entries a 2041 * @fbatch: Where the resulting entries are placed.
2026 * @indices: The cache indices correspondi 2042 * @indices: The cache indices corresponding to the entries in @entries
2027 * 2043 *
2028 * find_get_entries() will search for and ret 2044 * find_get_entries() will search for and return a batch of entries in
2029 * the mapping. The entries are placed in @f 2045 * the mapping. The entries are placed in @fbatch. find_get_entries()
2030 * takes a reference on any actual folios it 2046 * takes a reference on any actual folios it returns.
2031 * 2047 *
2032 * The entries have ascending indexes. The i 2048 * The entries have ascending indexes. The indices may not be consecutive
2033 * due to not-present entries or large folios 2049 * due to not-present entries or large folios.
2034 * 2050 *
2035 * Any shadow entries of evicted folios, or s 2051 * Any shadow entries of evicted folios, or swap entries from
2036 * shmem/tmpfs, are included in the returned 2052 * shmem/tmpfs, are included in the returned array.
2037 * 2053 *
2038 * Return: The number of entries which were f 2054 * Return: The number of entries which were found.
2039 */ 2055 */
2040 unsigned find_get_entries(struct address_spac !! 2056 unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
2041 pgoff_t end, struct folio_bat 2057 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2042 { 2058 {
2043 XA_STATE(xas, &mapping->i_pages, *sta !! 2059 XA_STATE(xas, &mapping->i_pages, start);
2044 struct folio *folio; 2060 struct folio *folio;
2045 2061
2046 rcu_read_lock(); 2062 rcu_read_lock();
2047 while ((folio = find_get_entry(&xas, 2063 while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
2048 indices[fbatch->nr] = xas.xa_ 2064 indices[fbatch->nr] = xas.xa_index;
2049 if (!folio_batch_add(fbatch, 2065 if (!folio_batch_add(fbatch, folio))
2050 break; 2066 break;
2051 } 2067 }
2052 <<
2053 if (folio_batch_count(fbatch)) { <<
2054 unsigned long nr; <<
2055 int idx = folio_batch_count(f <<
2056 <<
2057 folio = fbatch->folios[idx]; <<
2058 if (!xa_is_value(folio)) <<
2059 nr = folio_nr_pages(f <<
2060 else <<
2061 nr = 1 << xa_get_orde <<
2062 *start = round_down(indices[i <<
2063 } <<
2064 rcu_read_unlock(); 2068 rcu_read_unlock();
2065 2069
2066 return folio_batch_count(fbatch); 2070 return folio_batch_count(fbatch);
2067 } 2071 }
2068 2072
2069 /** 2073 /**
2070 * find_lock_entries - Find a batch of pageca 2074 * find_lock_entries - Find a batch of pagecache entries.
2071 * @mapping: The address_space to search. 2075 * @mapping: The address_space to search.
2072 * @start: The starting page cache index 2076 * @start: The starting page cache index.
2073 * @end: The final page index (inclusi 2077 * @end: The final page index (inclusive).
2074 * @fbatch: Where the resulting entries a 2078 * @fbatch: Where the resulting entries are placed.
2075 * @indices: The cache indices of the entr 2079 * @indices: The cache indices of the entries in @fbatch.
2076 * 2080 *
2077 * find_lock_entries() will return a batch of 2081 * find_lock_entries() will return a batch of entries from @mapping.
2078 * Swap, shadow and DAX entries are included. 2082 * Swap, shadow and DAX entries are included. Folios are returned
2079 * locked and with an incremented refcount. 2083 * locked and with an incremented refcount. Folios which are locked
2080 * by somebody else or under writeback are sk 2084 * by somebody else or under writeback are skipped. Folios which are
2081 * partially outside the range are not return 2085 * partially outside the range are not returned.
2082 * 2086 *
2083 * The entries have ascending indexes. The i 2087 * The entries have ascending indexes. The indices may not be consecutive
2084 * due to not-present entries, large folios, 2088 * due to not-present entries, large folios, folios which could not be
2085 * locked or folios under writeback. 2089 * locked or folios under writeback.
2086 * 2090 *
2087 * Return: The number of entries which were f 2091 * Return: The number of entries which were found.
2088 */ 2092 */
2089 unsigned find_lock_entries(struct address_spa !! 2093 unsigned find_lock_entries(struct address_space *mapping, pgoff_t start,
2090 pgoff_t end, struct folio_bat 2094 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2091 { 2095 {
2092 XA_STATE(xas, &mapping->i_pages, *sta !! 2096 XA_STATE(xas, &mapping->i_pages, start);
2093 struct folio *folio; 2097 struct folio *folio;
2094 2098
2095 rcu_read_lock(); 2099 rcu_read_lock();
2096 while ((folio = find_get_entry(&xas, 2100 while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
2097 unsigned long base; <<
2098 unsigned long nr; <<
2099 <<
2100 if (!xa_is_value(folio)) { 2101 if (!xa_is_value(folio)) {
2101 nr = folio_nr_pages(f !! 2102 if (folio->index < start)
2102 base = folio->index; <<
2103 /* Omit large folio w <<
2104 if (base < *start) <<
2105 goto put; 2103 goto put;
2106 /* Omit large folio w !! 2104 if (folio->index + folio_nr_pages(folio) - 1 > end)
2107 if (base + nr - 1 > e <<
2108 goto put; 2105 goto put;
2109 if (!folio_trylock(fo 2106 if (!folio_trylock(folio))
2110 goto put; 2107 goto put;
2111 if (folio->mapping != 2108 if (folio->mapping != mapping ||
2112 folio_test_writeb 2109 folio_test_writeback(folio))
2113 goto unlock; 2110 goto unlock;
2114 VM_BUG_ON_FOLIO(!foli 2111 VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index),
2115 folio 2112 folio);
2116 } else { <<
2117 nr = 1 << xas_get_ord <<
2118 base = xas.xa_index & <<
2119 /* Omit order>0 value <<
2120 if (base < *start) <<
2121 continue; <<
2122 /* Omit order>0 value <<
2123 if (base + nr - 1 > e <<
2124 break; <<
2125 } 2113 }
2126 <<
2127 /* Update start now so that l <<
2128 *start = base + nr; <<
2129 indices[fbatch->nr] = xas.xa_ 2114 indices[fbatch->nr] = xas.xa_index;
2130 if (!folio_batch_add(fbatch, 2115 if (!folio_batch_add(fbatch, folio))
2131 break; 2116 break;
2132 continue; 2117 continue;
2133 unlock: 2118 unlock:
2134 folio_unlock(folio); 2119 folio_unlock(folio);
2135 put: 2120 put:
2136 folio_put(folio); 2121 folio_put(folio);
2137 } 2122 }
2138 rcu_read_unlock(); 2123 rcu_read_unlock();
2139 2124
2140 return folio_batch_count(fbatch); 2125 return folio_batch_count(fbatch);
2141 } 2126 }
2142 2127
2143 /** 2128 /**
2144 * filemap_get_folios - Get a batch of folios 2129 * filemap_get_folios - Get a batch of folios
2145 * @mapping: The address_space to search 2130 * @mapping: The address_space to search
2146 * @start: The starting page index 2131 * @start: The starting page index
2147 * @end: The final page index (inclusi 2132 * @end: The final page index (inclusive)
2148 * @fbatch: The batch to fill. 2133 * @fbatch: The batch to fill.
2149 * 2134 *
2150 * Search for and return a batch of folios in 2135 * Search for and return a batch of folios in the mapping starting at
2151 * index @start and up to index @end (inclusi 2136 * index @start and up to index @end (inclusive). The folios are returned
2152 * in @fbatch with an elevated reference coun 2137 * in @fbatch with an elevated reference count.
2153 * 2138 *
>> 2139 * The first folio may start before @start; if it does, it will contain
>> 2140 * @start. The final folio may extend beyond @end; if it does, it will
>> 2141 * contain @end. The folios have ascending indices. There may be gaps
>> 2142 * between the folios if there are indices which have no folio in the
>> 2143 * page cache. If folios are added to or removed from the page cache
>> 2144 * while this is running, they may or may not be found by this call.
>> 2145 *
2154 * Return: The number of folios which were fo 2146 * Return: The number of folios which were found.
2155 * We also update @start to index the next fo 2147 * We also update @start to index the next folio for the traversal.
2156 */ 2148 */
2157 unsigned filemap_get_folios(struct address_sp 2149 unsigned filemap_get_folios(struct address_space *mapping, pgoff_t *start,
2158 pgoff_t end, struct folio_bat 2150 pgoff_t end, struct folio_batch *fbatch)
2159 { 2151 {
2160 return filemap_get_folios_tag(mapping !! 2152 XA_STATE(xas, &mapping->i_pages, *start);
>> 2153 struct folio *folio;
>> 2154
>> 2155 rcu_read_lock();
>> 2156 while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
>> 2157 /* Skip over shadow, swap and DAX entries */
>> 2158 if (xa_is_value(folio))
>> 2159 continue;
>> 2160 if (!folio_batch_add(fbatch, folio)) {
>> 2161 unsigned long nr = folio_nr_pages(folio);
>> 2162
>> 2163 if (folio_test_hugetlb(folio))
>> 2164 nr = 1;
>> 2165 *start = folio->index + nr;
>> 2166 goto out;
>> 2167 }
>> 2168 }
>> 2169
>> 2170 /*
>> 2171 * We come here when there is no page beyond @end. We take care to not
>> 2172 * overflow the index @start as it confuses some of the callers. This
>> 2173 * breaks the iteration when there is a page at index -1 but that is
>> 2174 * already broken anyway.
>> 2175 */
>> 2176 if (end == (pgoff_t)-1)
>> 2177 *start = (pgoff_t)-1;
>> 2178 else
>> 2179 *start = end + 1;
>> 2180 out:
>> 2181 rcu_read_unlock();
>> 2182
>> 2183 return folio_batch_count(fbatch);
2161 } 2184 }
2162 EXPORT_SYMBOL(filemap_get_folios); 2185 EXPORT_SYMBOL(filemap_get_folios);
2163 2186
>> 2187 static inline
>> 2188 bool folio_more_pages(struct folio *folio, pgoff_t index, pgoff_t max)
>> 2189 {
>> 2190 if (!folio_test_large(folio) || folio_test_hugetlb(folio))
>> 2191 return false;
>> 2192 if (index >= max)
>> 2193 return false;
>> 2194 return index < folio->index + folio_nr_pages(folio) - 1;
>> 2195 }
>> 2196
2164 /** 2197 /**
2165 * filemap_get_folios_contig - Get a batch of !! 2198 * find_get_pages_contig - gang contiguous pagecache lookup
2166 * @mapping: The address_space to search 2199 * @mapping: The address_space to search
2167 * @start: The starting page index !! 2200 * @index: The starting page index
2168 * @end: The final page index (inclusi !! 2201 * @nr_pages: The maximum number of pages
2169 * @fbatch: The batch to fill !! 2202 * @pages: Where the resulting pages are placed
>> 2203 *
>> 2204 * find_get_pages_contig() works exactly like find_get_pages_range(),
>> 2205 * except that the returned number of pages are guaranteed to be
>> 2206 * contiguous.
2170 * 2207 *
2171 * filemap_get_folios_contig() works exactly !! 2208 * Return: the number of pages which were found.
2172 * except the returned folios are guaranteed <<
2173 * not return all contiguous folios if the ba <<
2174 * <<
2175 * Return: The number of folios found. <<
2176 * Also update @start to be positioned for tr <<
2177 */ 2209 */
2178 !! 2210 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
2179 unsigned filemap_get_folios_contig(struct add !! 2211 unsigned int nr_pages, struct page **pages)
2180 pgoff_t *start, pgoff_t end, <<
2181 { 2212 {
2182 XA_STATE(xas, &mapping->i_pages, *sta !! 2213 XA_STATE(xas, &mapping->i_pages, index);
2183 unsigned long nr; <<
2184 struct folio *folio; 2214 struct folio *folio;
>> 2215 unsigned int ret = 0;
2185 2216
2186 rcu_read_lock(); !! 2217 if (unlikely(!nr_pages))
>> 2218 return 0;
2187 2219
2188 for (folio = xas_load(&xas); folio && !! 2220 rcu_read_lock();
2189 folio = xas_next(&xas !! 2221 for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2190 if (xas_retry(&xas, folio)) 2222 if (xas_retry(&xas, folio))
2191 continue; 2223 continue;
2192 /* 2224 /*
2193 * If the entry has been swap 2225 * If the entry has been swapped out, we can stop looking.
2194 * No current caller is looki 2226 * No current caller is looking for DAX entries.
2195 */ 2227 */
2196 if (xa_is_value(folio)) 2228 if (xa_is_value(folio))
2197 goto update_start; !! 2229 break;
2198 <<
2199 /* If we landed in the middle <<
2200 if (xa_is_sibling(folio)) <<
2201 goto update_start; <<
2202 2230
2203 if (!folio_try_get(folio)) !! 2231 if (!folio_try_get_rcu(folio))
2204 goto retry; 2232 goto retry;
2205 2233
2206 if (unlikely(folio != xas_rel 2234 if (unlikely(folio != xas_reload(&xas)))
2207 goto put_folio; !! 2235 goto put_page;
2208 2236
2209 if (!folio_batch_add(fbatch, !! 2237 again:
2210 nr = folio_nr_pages(f !! 2238 pages[ret] = folio_file_page(folio, xas.xa_index);
2211 *start = folio->index !! 2239 if (++ret == nr_pages)
2212 goto out; !! 2240 break;
>> 2241 if (folio_more_pages(folio, xas.xa_index, ULONG_MAX)) {
>> 2242 xas.xa_index++;
>> 2243 folio_ref_inc(folio);
>> 2244 goto again;
2213 } 2245 }
2214 continue; 2246 continue;
2215 put_folio: !! 2247 put_page:
2216 folio_put(folio); 2248 folio_put(folio);
2217 <<
2218 retry: 2249 retry:
2219 xas_reset(&xas); 2250 xas_reset(&xas);
2220 } 2251 }
2221 <<
2222 update_start: <<
2223 nr = folio_batch_count(fbatch); <<
2224 <<
2225 if (nr) { <<
2226 folio = fbatch->folios[nr - 1 <<
2227 *start = folio_next_index(fol <<
2228 } <<
2229 out: <<
2230 rcu_read_unlock(); 2252 rcu_read_unlock();
2231 return folio_batch_count(fbatch); !! 2253 return ret;
2232 } 2254 }
2233 EXPORT_SYMBOL(filemap_get_folios_contig); !! 2255 EXPORT_SYMBOL(find_get_pages_contig);
2234 2256
2235 /** 2257 /**
2236 * filemap_get_folios_tag - Get a batch of fo !! 2258 * find_get_pages_range_tag - Find and return head pages matching @tag.
2237 * @mapping: The address_space to search !! 2259 * @mapping: the address_space to search
2238 * @start: The starting page index !! 2260 * @index: the starting page index
2239 * @end: The final page index (inclusi !! 2261 * @end: The final page index (inclusive)
2240 * @tag: The tag index !! 2262 * @tag: the tag index
2241 * @fbatch: The batch to fill !! 2263 * @nr_pages: the maximum number of pages
>> 2264 * @pages: where the resulting pages are placed
2242 * 2265 *
2243 * The first folio may start before @start; i !! 2266 * Like find_get_pages_range(), except we only return head pages which are
2244 * @start. The final folio may extend beyond !! 2267 * tagged with @tag. @index is updated to the index immediately after the
2245 * contain @end. The folios have ascending i !! 2268 * last page we return, ready for the next iteration.
2246 * between the folios if there are indices wh <<
2247 * page cache. If folios are added to or rem <<
2248 * while this is running, they may or may not <<
2249 * Only returns folios that are tagged with @ <<
2250 * 2269 *
2251 * Return: The number of folios found. !! 2270 * Return: the number of pages which were found.
2252 * Also update @start to index the next folio <<
2253 */ 2271 */
2254 unsigned filemap_get_folios_tag(struct addres !! 2272 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
2255 pgoff_t end, xa_mark_ !! 2273 pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
>> 2274 struct page **pages)
2256 { 2275 {
2257 XA_STATE(xas, &mapping->i_pages, *sta !! 2276 XA_STATE(xas, &mapping->i_pages, *index);
2258 struct folio *folio; 2277 struct folio *folio;
>> 2278 unsigned ret = 0;
>> 2279
>> 2280 if (unlikely(!nr_pages))
>> 2281 return 0;
2259 2282
2260 rcu_read_lock(); 2283 rcu_read_lock();
2261 while ((folio = find_get_entry(&xas, !! 2284 while ((folio = find_get_entry(&xas, end, tag))) {
2262 /* 2285 /*
2263 * Shadow entries should neve 2286 * Shadow entries should never be tagged, but this iteration
2264 * is lockless so there is a 2287 * is lockless so there is a window for page reclaim to evict
2265 * a page we saw tagged. Skip !! 2288 * a page we saw tagged. Skip over it.
2266 */ 2289 */
2267 if (xa_is_value(folio)) 2290 if (xa_is_value(folio))
2268 continue; 2291 continue;
2269 if (!folio_batch_add(fbatch, !! 2292
2270 unsigned long nr = fo !! 2293 pages[ret] = &folio->page;
2271 *start = folio->index !! 2294 if (++ret == nr_pages) {
>> 2295 *index = folio->index + folio_nr_pages(folio);
2272 goto out; 2296 goto out;
2273 } 2297 }
2274 } 2298 }
>> 2299
2275 /* 2300 /*
2276 * We come here when there is no page !! 2301 * We come here when we got to @end. We take care to not overflow the
2277 * overflow the index @start as it co !! 2302 * index @index as it confuses some of the callers. This breaks the
2278 * breaks the iteration when there is !! 2303 * iteration when there is a page at index -1 but that is already
2279 * already broke anyway. !! 2304 * broken anyway.
2280 */ 2305 */
2281 if (end == (pgoff_t)-1) 2306 if (end == (pgoff_t)-1)
2282 *start = (pgoff_t)-1; !! 2307 *index = (pgoff_t)-1;
2283 else 2308 else
2284 *start = end + 1; !! 2309 *index = end + 1;
2285 out: 2310 out:
2286 rcu_read_unlock(); 2311 rcu_read_unlock();
2287 2312
2288 return folio_batch_count(fbatch); !! 2313 return ret;
2289 } 2314 }
2290 EXPORT_SYMBOL(filemap_get_folios_tag); !! 2315 EXPORT_SYMBOL(find_get_pages_range_tag);
2291 2316
2292 /* 2317 /*
2293 * CD/DVDs are error prone. When a medium err 2318 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2294 * a _large_ part of the i/o request. Imagine 2319 * a _large_ part of the i/o request. Imagine the worst scenario:
2295 * 2320 *
2296 * ---R_________________________________ 2321 * ---R__________________________________________B__________
2297 * ^ reading here 2322 * ^ reading here ^ bad block(assume 4k)
2298 * 2323 *
2299 * read(R) => miss => readahead(R...B) => med 2324 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2300 * => failing the whole request => read(R) => 2325 * => failing the whole request => read(R) => read(R+1) =>
2301 * readahead(R+1...B+1) => bang => read(R+2) 2326 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2302 * readahead(R+3...B+2) => bang => read(R+3) 2327 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2303 * readahead(R+4...B+3) => bang => read(R+4) 2328 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2304 * 2329 *
2305 * It is going insane. Fix it by quickly scal 2330 * It is going insane. Fix it by quickly scaling down the readahead size.
2306 */ 2331 */
2307 static void shrink_readahead_size_eio(struct 2332 static void shrink_readahead_size_eio(struct file_ra_state *ra)
2308 { 2333 {
2309 ra->ra_pages /= 4; 2334 ra->ra_pages /= 4;
2310 } 2335 }
2311 2336
2312 /* 2337 /*
2313 * filemap_get_read_batch - Get a batch of fo 2338 * filemap_get_read_batch - Get a batch of folios for read
2314 * 2339 *
2315 * Get a batch of folios which represent a co 2340 * Get a batch of folios which represent a contiguous range of bytes in
2316 * the file. No exceptional entries will be 2341 * the file. No exceptional entries will be returned. If @index is in
2317 * the middle of a folio, the entire folio wi 2342 * the middle of a folio, the entire folio will be returned. The last
2318 * folio in the batch may have the readahead 2343 * folio in the batch may have the readahead flag set or the uptodate flag
2319 * clear so that the caller can take the appr 2344 * clear so that the caller can take the appropriate action.
2320 */ 2345 */
2321 static void filemap_get_read_batch(struct add 2346 static void filemap_get_read_batch(struct address_space *mapping,
2322 pgoff_t index, pgoff_t max, s 2347 pgoff_t index, pgoff_t max, struct folio_batch *fbatch)
2323 { 2348 {
2324 XA_STATE(xas, &mapping->i_pages, inde 2349 XA_STATE(xas, &mapping->i_pages, index);
2325 struct folio *folio; 2350 struct folio *folio;
2326 2351
2327 rcu_read_lock(); 2352 rcu_read_lock();
2328 for (folio = xas_load(&xas); folio; f 2353 for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2329 if (xas_retry(&xas, folio)) 2354 if (xas_retry(&xas, folio))
2330 continue; 2355 continue;
2331 if (xas.xa_index > max || xa_ 2356 if (xas.xa_index > max || xa_is_value(folio))
2332 break; 2357 break;
2333 if (xa_is_sibling(folio)) 2358 if (xa_is_sibling(folio))
2334 break; 2359 break;
2335 if (!folio_try_get(folio)) !! 2360 if (!folio_try_get_rcu(folio))
2336 goto retry; 2361 goto retry;
2337 2362
2338 if (unlikely(folio != xas_rel 2363 if (unlikely(folio != xas_reload(&xas)))
2339 goto put_folio; 2364 goto put_folio;
2340 2365
2341 if (!folio_batch_add(fbatch, 2366 if (!folio_batch_add(fbatch, folio))
2342 break; 2367 break;
2343 if (!folio_test_uptodate(foli 2368 if (!folio_test_uptodate(folio))
2344 break; 2369 break;
2345 if (folio_test_readahead(foli 2370 if (folio_test_readahead(folio))
2346 break; 2371 break;
2347 xas_advance(&xas, folio_next_ !! 2372 xas_advance(&xas, folio->index + folio_nr_pages(folio) - 1);
2348 continue; 2373 continue;
2349 put_folio: 2374 put_folio:
2350 folio_put(folio); 2375 folio_put(folio);
2351 retry: 2376 retry:
2352 xas_reset(&xas); 2377 xas_reset(&xas);
2353 } 2378 }
2354 rcu_read_unlock(); 2379 rcu_read_unlock();
2355 } 2380 }
2356 2381
2357 static int filemap_read_folio(struct file *fi 2382 static int filemap_read_folio(struct file *file, filler_t filler,
2358 struct folio *folio) 2383 struct folio *folio)
2359 { 2384 {
2360 bool workingset = folio_test_workings <<
2361 unsigned long pflags; <<
2362 int error; 2385 int error;
2363 2386
>> 2387 /*
>> 2388 * A previous I/O error may have been due to temporary failures,
>> 2389 * eg. multipath errors. PG_error will be set again if read_folio
>> 2390 * fails.
>> 2391 */
>> 2392 folio_clear_error(folio);
2364 /* Start the actual read. The read wi 2393 /* Start the actual read. The read will unlock the page. */
2365 if (unlikely(workingset)) <<
2366 psi_memstall_enter(&pflags); <<
2367 error = filler(file, folio); 2394 error = filler(file, folio);
2368 if (unlikely(workingset)) <<
2369 psi_memstall_leave(&pflags); <<
2370 if (error) 2395 if (error)
2371 return error; 2396 return error;
2372 2397
2373 error = folio_wait_locked_killable(fo 2398 error = folio_wait_locked_killable(folio);
2374 if (error) 2399 if (error)
2375 return error; 2400 return error;
2376 if (folio_test_uptodate(folio)) 2401 if (folio_test_uptodate(folio))
2377 return 0; 2402 return 0;
2378 if (file) 2403 if (file)
2379 shrink_readahead_size_eio(&fi 2404 shrink_readahead_size_eio(&file->f_ra);
2380 return -EIO; 2405 return -EIO;
2381 } 2406 }
2382 2407
2383 static bool filemap_range_uptodate(struct add 2408 static bool filemap_range_uptodate(struct address_space *mapping,
2384 loff_t pos, size_t count, str !! 2409 loff_t pos, struct iov_iter *iter, struct folio *folio)
2385 bool need_uptodate) <<
2386 { 2410 {
>> 2411 int count;
>> 2412
2387 if (folio_test_uptodate(folio)) 2413 if (folio_test_uptodate(folio))
2388 return true; 2414 return true;
2389 /* pipes can't handle partially uptod 2415 /* pipes can't handle partially uptodate pages */
2390 if (need_uptodate) !! 2416 if (iov_iter_is_pipe(iter))
2391 return false; 2417 return false;
2392 if (!mapping->a_ops->is_partially_upt 2418 if (!mapping->a_ops->is_partially_uptodate)
2393 return false; 2419 return false;
2394 if (mapping->host->i_blkbits >= folio 2420 if (mapping->host->i_blkbits >= folio_shift(folio))
2395 return false; 2421 return false;
2396 2422
>> 2423 count = iter->count;
2397 if (folio_pos(folio) > pos) { 2424 if (folio_pos(folio) > pos) {
2398 count -= folio_pos(folio) - p 2425 count -= folio_pos(folio) - pos;
2399 pos = 0; 2426 pos = 0;
2400 } else { 2427 } else {
2401 pos -= folio_pos(folio); 2428 pos -= folio_pos(folio);
2402 } 2429 }
2403 2430
2404 return mapping->a_ops->is_partially_u 2431 return mapping->a_ops->is_partially_uptodate(folio, pos, count);
2405 } 2432 }
2406 2433
2407 static int filemap_update_page(struct kiocb * 2434 static int filemap_update_page(struct kiocb *iocb,
2408 struct address_space *mapping !! 2435 struct address_space *mapping, struct iov_iter *iter,
2409 struct folio *folio, bool nee !! 2436 struct folio *folio)
2410 { 2437 {
2411 int error; 2438 int error;
2412 2439
2413 if (iocb->ki_flags & IOCB_NOWAIT) { 2440 if (iocb->ki_flags & IOCB_NOWAIT) {
2414 if (!filemap_invalidate_trylo 2441 if (!filemap_invalidate_trylock_shared(mapping))
2415 return -EAGAIN; 2442 return -EAGAIN;
2416 } else { 2443 } else {
2417 filemap_invalidate_lock_share 2444 filemap_invalidate_lock_shared(mapping);
2418 } 2445 }
2419 2446
2420 if (!folio_trylock(folio)) { 2447 if (!folio_trylock(folio)) {
2421 error = -EAGAIN; 2448 error = -EAGAIN;
2422 if (iocb->ki_flags & (IOCB_NO 2449 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2423 goto unlock_mapping; 2450 goto unlock_mapping;
2424 if (!(iocb->ki_flags & IOCB_W 2451 if (!(iocb->ki_flags & IOCB_WAITQ)) {
2425 filemap_invalidate_un 2452 filemap_invalidate_unlock_shared(mapping);
2426 /* 2453 /*
2427 * This is where we u 2454 * This is where we usually end up waiting for a
2428 * previously submitt 2455 * previously submitted readahead to finish.
2429 */ 2456 */
2430 folio_put_wait_locked 2457 folio_put_wait_locked(folio, TASK_KILLABLE);
2431 return AOP_TRUNCATED_ 2458 return AOP_TRUNCATED_PAGE;
2432 } 2459 }
2433 error = __folio_lock_async(fo 2460 error = __folio_lock_async(folio, iocb->ki_waitq);
2434 if (error) 2461 if (error)
2435 goto unlock_mapping; 2462 goto unlock_mapping;
2436 } 2463 }
2437 2464
2438 error = AOP_TRUNCATED_PAGE; 2465 error = AOP_TRUNCATED_PAGE;
2439 if (!folio->mapping) 2466 if (!folio->mapping)
2440 goto unlock; 2467 goto unlock;
2441 2468
2442 error = 0; 2469 error = 0;
2443 if (filemap_range_uptodate(mapping, i !! 2470 if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, folio))
2444 need_uptod <<
2445 goto unlock; 2471 goto unlock;
2446 2472
2447 error = -EAGAIN; 2473 error = -EAGAIN;
2448 if (iocb->ki_flags & (IOCB_NOIO | IOC 2474 if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2449 goto unlock; 2475 goto unlock;
2450 2476
2451 error = filemap_read_folio(iocb->ki_f 2477 error = filemap_read_folio(iocb->ki_filp, mapping->a_ops->read_folio,
2452 folio); 2478 folio);
2453 goto unlock_mapping; 2479 goto unlock_mapping;
2454 unlock: 2480 unlock:
2455 folio_unlock(folio); 2481 folio_unlock(folio);
2456 unlock_mapping: 2482 unlock_mapping:
2457 filemap_invalidate_unlock_shared(mapp 2483 filemap_invalidate_unlock_shared(mapping);
2458 if (error == AOP_TRUNCATED_PAGE) 2484 if (error == AOP_TRUNCATED_PAGE)
2459 folio_put(folio); 2485 folio_put(folio);
2460 return error; 2486 return error;
2461 } 2487 }
2462 2488
2463 static int filemap_create_folio(struct file * 2489 static int filemap_create_folio(struct file *file,
2464 struct address_space *mapping !! 2490 struct address_space *mapping, pgoff_t index,
2465 struct folio_batch *fbatch) 2491 struct folio_batch *fbatch)
2466 { 2492 {
2467 struct folio *folio; 2493 struct folio *folio;
2468 int error; 2494 int error;
2469 unsigned int min_order = mapping_min_ <<
2470 pgoff_t index; <<
2471 2495
2472 folio = filemap_alloc_folio(mapping_g !! 2496 folio = filemap_alloc_folio(mapping_gfp_mask(mapping), 0);
2473 if (!folio) 2497 if (!folio)
2474 return -ENOMEM; 2498 return -ENOMEM;
2475 2499
2476 /* 2500 /*
2477 * Protect against truncate / hole pu 2501 * Protect against truncate / hole punch. Grabbing invalidate_lock
2478 * here assures we cannot instantiate 2502 * here assures we cannot instantiate and bring uptodate new
2479 * pagecache folios after evicting pa 2503 * pagecache folios after evicting page cache during truncate
2480 * and before actually freeing blocks 2504 * and before actually freeing blocks. Note that we could
2481 * release invalidate_lock after inse 2505 * release invalidate_lock after inserting the folio into
2482 * the page cache as the locked folio 2506 * the page cache as the locked folio would then be enough to
2483 * synchronize with hole punching. Bu 2507 * synchronize with hole punching. But there are code paths
2484 * such as filemap_update_page() fill 2508 * such as filemap_update_page() filling in partially uptodate
2485 * pages or ->readahead() that need t 2509 * pages or ->readahead() that need to hold invalidate_lock
2486 * while mapping blocks for IO so let 2510 * while mapping blocks for IO so let's hold the lock here as
2487 * well to keep locking rules simple. 2511 * well to keep locking rules simple.
2488 */ 2512 */
2489 filemap_invalidate_lock_shared(mappin 2513 filemap_invalidate_lock_shared(mapping);
2490 index = (pos >> (PAGE_SHIFT + min_ord <<
2491 error = filemap_add_folio(mapping, fo 2514 error = filemap_add_folio(mapping, folio, index,
2492 mapping_gfp_constrain 2515 mapping_gfp_constraint(mapping, GFP_KERNEL));
2493 if (error == -EEXIST) 2516 if (error == -EEXIST)
2494 error = AOP_TRUNCATED_PAGE; 2517 error = AOP_TRUNCATED_PAGE;
2495 if (error) 2518 if (error)
2496 goto error; 2519 goto error;
2497 2520
2498 error = filemap_read_folio(file, mapp 2521 error = filemap_read_folio(file, mapping->a_ops->read_folio, folio);
2499 if (error) 2522 if (error)
2500 goto error; 2523 goto error;
2501 2524
2502 filemap_invalidate_unlock_shared(mapp 2525 filemap_invalidate_unlock_shared(mapping);
2503 folio_batch_add(fbatch, folio); 2526 folio_batch_add(fbatch, folio);
2504 return 0; 2527 return 0;
2505 error: 2528 error:
2506 filemap_invalidate_unlock_shared(mapp 2529 filemap_invalidate_unlock_shared(mapping);
2507 folio_put(folio); 2530 folio_put(folio);
2508 return error; 2531 return error;
2509 } 2532 }
2510 2533
2511 static int filemap_readahead(struct kiocb *io 2534 static int filemap_readahead(struct kiocb *iocb, struct file *file,
2512 struct address_space *mapping 2535 struct address_space *mapping, struct folio *folio,
2513 pgoff_t last_index) 2536 pgoff_t last_index)
2514 { 2537 {
2515 DEFINE_READAHEAD(ractl, file, &file-> 2538 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, folio->index);
2516 2539
2517 if (iocb->ki_flags & IOCB_NOIO) 2540 if (iocb->ki_flags & IOCB_NOIO)
2518 return -EAGAIN; 2541 return -EAGAIN;
2519 page_cache_async_ra(&ractl, folio, la 2542 page_cache_async_ra(&ractl, folio, last_index - folio->index);
2520 return 0; 2543 return 0;
2521 } 2544 }
2522 2545
2523 static int filemap_get_pages(struct kiocb *io !! 2546 static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter,
2524 struct folio_batch *fbatch, b !! 2547 struct folio_batch *fbatch)
2525 { 2548 {
2526 struct file *filp = iocb->ki_filp; 2549 struct file *filp = iocb->ki_filp;
2527 struct address_space *mapping = filp- 2550 struct address_space *mapping = filp->f_mapping;
2528 struct file_ra_state *ra = &filp->f_r 2551 struct file_ra_state *ra = &filp->f_ra;
2529 pgoff_t index = iocb->ki_pos >> PAGE_ 2552 pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2530 pgoff_t last_index; 2553 pgoff_t last_index;
2531 struct folio *folio; 2554 struct folio *folio;
2532 unsigned int flags; <<
2533 int err = 0; 2555 int err = 0;
2534 2556
2535 /* "last_index" is the index of the p !! 2557 last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE);
2536 last_index = DIV_ROUND_UP(iocb->ki_po <<
2537 retry: 2558 retry:
2538 if (fatal_signal_pending(current)) 2559 if (fatal_signal_pending(current))
2539 return -EINTR; 2560 return -EINTR;
2540 2561
2541 filemap_get_read_batch(mapping, index !! 2562 filemap_get_read_batch(mapping, index, last_index, fbatch);
2542 if (!folio_batch_count(fbatch)) { 2563 if (!folio_batch_count(fbatch)) {
2543 if (iocb->ki_flags & IOCB_NOI 2564 if (iocb->ki_flags & IOCB_NOIO)
2544 return -EAGAIN; 2565 return -EAGAIN;
2545 if (iocb->ki_flags & IOCB_NOW <<
2546 flags = memalloc_noio <<
2547 page_cache_sync_readahead(map 2566 page_cache_sync_readahead(mapping, ra, filp, index,
2548 last_index - 2567 last_index - index);
2549 if (iocb->ki_flags & IOCB_NOW !! 2568 filemap_get_read_batch(mapping, index, last_index, fbatch);
2550 memalloc_noio_restore <<
2551 filemap_get_read_batch(mappin <<
2552 } 2569 }
2553 if (!folio_batch_count(fbatch)) { 2570 if (!folio_batch_count(fbatch)) {
2554 if (iocb->ki_flags & (IOCB_NO 2571 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2555 return -EAGAIN; 2572 return -EAGAIN;
2556 err = filemap_create_folio(fi !! 2573 err = filemap_create_folio(filp, mapping,
>> 2574 iocb->ki_pos >> PAGE_SHIFT, fbatch);
2557 if (err == AOP_TRUNCATED_PAGE 2575 if (err == AOP_TRUNCATED_PAGE)
2558 goto retry; 2576 goto retry;
2559 return err; 2577 return err;
2560 } 2578 }
2561 2579
2562 folio = fbatch->folios[folio_batch_co 2580 folio = fbatch->folios[folio_batch_count(fbatch) - 1];
2563 if (folio_test_readahead(folio)) { 2581 if (folio_test_readahead(folio)) {
2564 err = filemap_readahead(iocb, 2582 err = filemap_readahead(iocb, filp, mapping, folio, last_index);
2565 if (err) 2583 if (err)
2566 goto err; 2584 goto err;
2567 } 2585 }
2568 if (!folio_test_uptodate(folio)) { 2586 if (!folio_test_uptodate(folio)) {
2569 if ((iocb->ki_flags & IOCB_WA 2587 if ((iocb->ki_flags & IOCB_WAITQ) &&
2570 folio_batch_count(fbatch) 2588 folio_batch_count(fbatch) > 1)
2571 iocb->ki_flags |= IOC 2589 iocb->ki_flags |= IOCB_NOWAIT;
2572 err = filemap_update_page(ioc !! 2590 err = filemap_update_page(iocb, mapping, iter, folio);
2573 nee <<
2574 if (err) 2591 if (err)
2575 goto err; 2592 goto err;
2576 } 2593 }
2577 2594
2578 trace_mm_filemap_get_pages(mapping, i <<
2579 return 0; 2595 return 0;
2580 err: 2596 err:
2581 if (err < 0) 2597 if (err < 0)
2582 folio_put(folio); 2598 folio_put(folio);
2583 if (likely(--fbatch->nr)) 2599 if (likely(--fbatch->nr))
2584 return 0; 2600 return 0;
2585 if (err == AOP_TRUNCATED_PAGE) 2601 if (err == AOP_TRUNCATED_PAGE)
2586 goto retry; 2602 goto retry;
2587 return err; 2603 return err;
2588 } 2604 }
2589 2605
2590 static inline bool pos_same_folio(loff_t pos1 2606 static inline bool pos_same_folio(loff_t pos1, loff_t pos2, struct folio *folio)
2591 { 2607 {
2592 unsigned int shift = folio_shift(foli 2608 unsigned int shift = folio_shift(folio);
2593 2609
2594 return (pos1 >> shift == pos2 >> shif 2610 return (pos1 >> shift == pos2 >> shift);
2595 } 2611 }
2596 2612
2597 /** 2613 /**
2598 * filemap_read - Read data from the page cac 2614 * filemap_read - Read data from the page cache.
2599 * @iocb: The iocb to read. 2615 * @iocb: The iocb to read.
2600 * @iter: Destination for the data. 2616 * @iter: Destination for the data.
2601 * @already_read: Number of bytes already rea 2617 * @already_read: Number of bytes already read by the caller.
2602 * 2618 *
2603 * Copies data from the page cache. If the d 2619 * Copies data from the page cache. If the data is not currently present,
2604 * uses the readahead and read_folio address_ 2620 * uses the readahead and read_folio address_space operations to fetch it.
2605 * 2621 *
2606 * Return: Total number of bytes copied, incl 2622 * Return: Total number of bytes copied, including those already read by
2607 * the caller. If an error happens before an 2623 * the caller. If an error happens before any bytes are copied, returns
2608 * a negative error number. 2624 * a negative error number.
2609 */ 2625 */
2610 ssize_t filemap_read(struct kiocb *iocb, stru 2626 ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2611 ssize_t already_read) 2627 ssize_t already_read)
2612 { 2628 {
2613 struct file *filp = iocb->ki_filp; 2629 struct file *filp = iocb->ki_filp;
2614 struct file_ra_state *ra = &filp->f_r 2630 struct file_ra_state *ra = &filp->f_ra;
2615 struct address_space *mapping = filp- 2631 struct address_space *mapping = filp->f_mapping;
2616 struct inode *inode = mapping->host; 2632 struct inode *inode = mapping->host;
2617 struct folio_batch fbatch; 2633 struct folio_batch fbatch;
2618 int i, error = 0; 2634 int i, error = 0;
2619 bool writably_mapped; 2635 bool writably_mapped;
2620 loff_t isize, end_offset; 2636 loff_t isize, end_offset;
2621 loff_t last_pos = ra->prev_pos; <<
2622 2637
2623 if (unlikely(iocb->ki_pos >= inode->i 2638 if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2624 return 0; 2639 return 0;
2625 if (unlikely(!iov_iter_count(iter))) 2640 if (unlikely(!iov_iter_count(iter)))
2626 return 0; 2641 return 0;
2627 2642
2628 iov_iter_truncate(iter, inode->i_sb-> !! 2643 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2629 folio_batch_init(&fbatch); 2644 folio_batch_init(&fbatch);
2630 2645
2631 do { 2646 do {
2632 cond_resched(); 2647 cond_resched();
2633 2648
2634 /* 2649 /*
2635 * If we've already successfu 2650 * If we've already successfully copied some data, then we
2636 * can no longer safely retur 2651 * can no longer safely return -EIOCBQUEUED. Hence mark
2637 * an async read NOWAIT at th 2652 * an async read NOWAIT at that point.
2638 */ 2653 */
2639 if ((iocb->ki_flags & IOCB_WA 2654 if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
2640 iocb->ki_flags |= IOC 2655 iocb->ki_flags |= IOCB_NOWAIT;
2641 2656
2642 if (unlikely(iocb->ki_pos >= 2657 if (unlikely(iocb->ki_pos >= i_size_read(inode)))
2643 break; 2658 break;
2644 2659
2645 error = filemap_get_pages(ioc !! 2660 error = filemap_get_pages(iocb, iter, &fbatch);
2646 if (error < 0) 2661 if (error < 0)
2647 break; 2662 break;
2648 2663
2649 /* 2664 /*
2650 * i_size must be checked aft 2665 * i_size must be checked after we know the pages are Uptodate.
2651 * 2666 *
2652 * Checking i_size after the 2667 * Checking i_size after the check allows us to calculate
2653 * the correct value for "nr" 2668 * the correct value for "nr", which means the zero-filled
2654 * part of the page is not co 2669 * part of the page is not copied back to userspace (unless
2655 * another truncate extends t 2670 * another truncate extends the file - this is desired though).
2656 */ 2671 */
2657 isize = i_size_read(inode); 2672 isize = i_size_read(inode);
2658 if (unlikely(iocb->ki_pos >= 2673 if (unlikely(iocb->ki_pos >= isize))
2659 goto put_folios; 2674 goto put_folios;
2660 end_offset = min_t(loff_t, is 2675 end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2661 2676
2662 /* 2677 /*
2663 * Once we start copying data 2678 * Once we start copying data, we don't want to be touching any
2664 * cachelines that might be c 2679 * cachelines that might be contended:
2665 */ 2680 */
2666 writably_mapped = mapping_wri 2681 writably_mapped = mapping_writably_mapped(mapping);
2667 2682
2668 /* 2683 /*
2669 * When a read accesses the s 2684 * When a read accesses the same folio several times, only
2670 * mark it as accessed the fi 2685 * mark it as accessed the first time.
2671 */ 2686 */
2672 if (!pos_same_folio(iocb->ki_ !! 2687 if (!pos_same_folio(iocb->ki_pos, ra->prev_pos - 1,
2673 fbatch.fo !! 2688 fbatch.folios[0]))
2674 folio_mark_accessed(f 2689 folio_mark_accessed(fbatch.folios[0]);
2675 2690
2676 for (i = 0; i < folio_batch_c 2691 for (i = 0; i < folio_batch_count(&fbatch); i++) {
2677 struct folio *folio = 2692 struct folio *folio = fbatch.folios[i];
2678 size_t fsize = folio_ 2693 size_t fsize = folio_size(folio);
2679 size_t offset = iocb- 2694 size_t offset = iocb->ki_pos & (fsize - 1);
2680 size_t bytes = min_t( 2695 size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2681 2696 fsize - offset);
2682 size_t copied; 2697 size_t copied;
2683 2698
2684 if (end_offset < foli 2699 if (end_offset < folio_pos(folio))
2685 break; 2700 break;
2686 if (i > 0) 2701 if (i > 0)
2687 folio_mark_ac 2702 folio_mark_accessed(folio);
2688 /* 2703 /*
2689 * If users can be wr 2704 * If users can be writing to this folio using arbitrary
2690 * virtual addresses, 2705 * virtual addresses, take care of potential aliasing
2691 * before reading the 2706 * before reading the folio on the kernel side.
2692 */ 2707 */
2693 if (writably_mapped) 2708 if (writably_mapped)
2694 flush_dcache_ 2709 flush_dcache_folio(folio);
2695 2710
2696 copied = copy_folio_t 2711 copied = copy_folio_to_iter(folio, offset, bytes, iter);
2697 2712
2698 already_read += copie 2713 already_read += copied;
2699 iocb->ki_pos += copie 2714 iocb->ki_pos += copied;
2700 last_pos = iocb->ki_p !! 2715 ra->prev_pos = iocb->ki_pos;
2701 2716
2702 if (copied < bytes) { 2717 if (copied < bytes) {
2703 error = -EFAU 2718 error = -EFAULT;
2704 break; 2719 break;
2705 } 2720 }
2706 } 2721 }
2707 put_folios: 2722 put_folios:
2708 for (i = 0; i < folio_batch_c 2723 for (i = 0; i < folio_batch_count(&fbatch); i++)
2709 folio_put(fbatch.foli 2724 folio_put(fbatch.folios[i]);
2710 folio_batch_init(&fbatch); 2725 folio_batch_init(&fbatch);
2711 } while (iov_iter_count(iter) && iocb 2726 } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2712 2727
2713 file_accessed(filp); 2728 file_accessed(filp);
2714 ra->prev_pos = last_pos; !! 2729
2715 return already_read ? already_read : 2730 return already_read ? already_read : error;
2716 } 2731 }
2717 EXPORT_SYMBOL_GPL(filemap_read); 2732 EXPORT_SYMBOL_GPL(filemap_read);
2718 2733
2719 int kiocb_write_and_wait(struct kiocb *iocb, <<
2720 { <<
2721 struct address_space *mapping = iocb- <<
2722 loff_t pos = iocb->ki_pos; <<
2723 loff_t end = pos + count - 1; <<
2724 <<
2725 if (iocb->ki_flags & IOCB_NOWAIT) { <<
2726 if (filemap_range_needs_write <<
2727 return -EAGAIN; <<
2728 return 0; <<
2729 } <<
2730 <<
2731 return filemap_write_and_wait_range(m <<
2732 } <<
2733 EXPORT_SYMBOL_GPL(kiocb_write_and_wait); <<
2734 <<
2735 int filemap_invalidate_pages(struct address_s <<
2736 loff_t pos, loff <<
2737 { <<
2738 int ret; <<
2739 <<
2740 if (nowait) { <<
2741 /* we could block if there ar <<
2742 if (filemap_range_has_page(ma <<
2743 return -EAGAIN; <<
2744 } else { <<
2745 ret = filemap_write_and_wait_ <<
2746 if (ret) <<
2747 return ret; <<
2748 } <<
2749 <<
2750 /* <<
2751 * After a write we want buffered rea <<
2752 * the new data. We invalidate clean <<
2753 * about to write. We do this *befor <<
2754 * without clobbering -EIOCBQUEUED fr <<
2755 */ <<
2756 return invalidate_inode_pages2_range( <<
2757 <<
2758 } <<
2759 <<
2760 int kiocb_invalidate_pages(struct kiocb *iocb <<
2761 { <<
2762 struct address_space *mapping = iocb- <<
2763 <<
2764 return filemap_invalidate_pages(mappi <<
2765 iocb- <<
2766 iocb- <<
2767 } <<
2768 EXPORT_SYMBOL_GPL(kiocb_invalidate_pages); <<
2769 <<
2770 /** 2734 /**
2771 * generic_file_read_iter - generic filesyste 2735 * generic_file_read_iter - generic filesystem read routine
2772 * @iocb: kernel I/O control block 2736 * @iocb: kernel I/O control block
2773 * @iter: destination for the data read 2737 * @iter: destination for the data read
2774 * 2738 *
2775 * This is the "read_iter()" routine for all 2739 * This is the "read_iter()" routine for all filesystems
2776 * that can use the page cache directly. 2740 * that can use the page cache directly.
2777 * 2741 *
2778 * The IOCB_NOWAIT flag in iocb->ki_flags ind 2742 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2779 * be returned when no data can be read witho 2743 * be returned when no data can be read without waiting for I/O requests
2780 * to complete; it doesn't prevent readahead. 2744 * to complete; it doesn't prevent readahead.
2781 * 2745 *
2782 * The IOCB_NOIO flag in iocb->ki_flags indic 2746 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2783 * requests shall be made for the read or for 2747 * requests shall be made for the read or for readahead. When no data
2784 * can be read, -EAGAIN shall be returned. W 2748 * can be read, -EAGAIN shall be returned. When readahead would be
2785 * triggered, a partial, possibly empty read 2749 * triggered, a partial, possibly empty read shall be returned.
2786 * 2750 *
2787 * Return: 2751 * Return:
2788 * * number of bytes copied, even for partial 2752 * * number of bytes copied, even for partial reads
2789 * * negative error code (or 0 if IOCB_NOIO) 2753 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2790 */ 2754 */
2791 ssize_t 2755 ssize_t
2792 generic_file_read_iter(struct kiocb *iocb, st 2756 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2793 { 2757 {
2794 size_t count = iov_iter_count(iter); 2758 size_t count = iov_iter_count(iter);
2795 ssize_t retval = 0; 2759 ssize_t retval = 0;
2796 2760
2797 if (!count) 2761 if (!count)
2798 return 0; /* skip atime */ 2762 return 0; /* skip atime */
2799 2763
2800 if (iocb->ki_flags & IOCB_DIRECT) { 2764 if (iocb->ki_flags & IOCB_DIRECT) {
2801 struct file *file = iocb->ki_ 2765 struct file *file = iocb->ki_filp;
2802 struct address_space *mapping 2766 struct address_space *mapping = file->f_mapping;
2803 struct inode *inode = mapping 2767 struct inode *inode = mapping->host;
2804 2768
2805 retval = kiocb_write_and_wait !! 2769 if (iocb->ki_flags & IOCB_NOWAIT) {
2806 if (retval < 0) !! 2770 if (filemap_range_needs_writeback(mapping, iocb->ki_pos,
2807 return retval; !! 2771 iocb->ki_pos + count - 1))
>> 2772 return -EAGAIN;
>> 2773 } else {
>> 2774 retval = filemap_write_and_wait_range(mapping,
>> 2775 iocb->ki_pos,
>> 2776 iocb->ki_pos + count - 1);
>> 2777 if (retval < 0)
>> 2778 return retval;
>> 2779 }
>> 2780
2808 file_accessed(file); 2781 file_accessed(file);
2809 2782
2810 retval = mapping->a_ops->dire 2783 retval = mapping->a_ops->direct_IO(iocb, iter);
2811 if (retval >= 0) { 2784 if (retval >= 0) {
2812 iocb->ki_pos += retva 2785 iocb->ki_pos += retval;
2813 count -= retval; 2786 count -= retval;
2814 } 2787 }
2815 if (retval != -EIOCBQUEUED) 2788 if (retval != -EIOCBQUEUED)
2816 iov_iter_revert(iter, 2789 iov_iter_revert(iter, count - iov_iter_count(iter));
2817 2790
2818 /* 2791 /*
2819 * Btrfs can have a short DIO 2792 * Btrfs can have a short DIO read if we encounter
2820 * compressed extents, so if 2793 * compressed extents, so if there was an error, or if
2821 * we've already read everyth 2794 * we've already read everything we wanted to, or if
2822 * there was a short read bec 2795 * there was a short read because we hit EOF, go ahead
2823 * and return. Otherwise fal 2796 * and return. Otherwise fallthrough to buffered io for
2824 * the rest of the read. Buf 2797 * the rest of the read. Buffered reads will not work for
2825 * DAX files, so don't bother 2798 * DAX files, so don't bother trying.
2826 */ 2799 */
2827 if (retval < 0 || !count || I 2800 if (retval < 0 || !count || IS_DAX(inode))
2828 return retval; 2801 return retval;
2829 if (iocb->ki_pos >= i_size_re 2802 if (iocb->ki_pos >= i_size_read(inode))
2830 return retval; 2803 return retval;
2831 } 2804 }
2832 2805
2833 return filemap_read(iocb, iter, retva 2806 return filemap_read(iocb, iter, retval);
2834 } 2807 }
2835 EXPORT_SYMBOL(generic_file_read_iter); 2808 EXPORT_SYMBOL(generic_file_read_iter);
2836 2809
2837 /* <<
2838 * Splice subpages from a folio into a pipe. <<
2839 */ <<
2840 size_t splice_folio_into_pipe(struct pipe_ino <<
2841 struct folio *f <<
2842 { <<
2843 struct page *page; <<
2844 size_t spliced = 0, offset = offset_i <<
2845 <<
2846 page = folio_page(folio, offset / PAG <<
2847 size = min(size, folio_size(folio) - <<
2848 offset %= PAGE_SIZE; <<
2849 <<
2850 while (spliced < size && <<
2851 !pipe_full(pipe->head, pipe->t <<
2852 struct pipe_buffer *buf = pip <<
2853 size_t part = min_t(size_t, P <<
2854 <<
2855 *buf = (struct pipe_buffer) { <<
2856 .ops = &page_cache <<
2857 .page = page, <<
2858 .offset = offset, <<
2859 .len = part, <<
2860 }; <<
2861 folio_get(folio); <<
2862 pipe->head++; <<
2863 page++; <<
2864 spliced += part; <<
2865 offset = 0; <<
2866 } <<
2867 <<
2868 return spliced; <<
2869 } <<
2870 <<
2871 /** <<
2872 * filemap_splice_read - Splice data from a <<
2873 * @in: The file to read from <<
2874 * @ppos: Pointer to the file position to rea <<
2875 * @pipe: The pipe to splice into <<
2876 * @len: The amount to splice <<
2877 * @flags: The SPLICE_F_* flags <<
2878 * <<
2879 * This function gets folios from a file's pa <<
2880 * pipe. Readahead will be called as necessa <<
2881 * be used for blockdevs also. <<
2882 * <<
2883 * Return: On success, the number of bytes re <<
2884 * will be updated if appropriate; 0 will be <<
2885 * to be read; -EAGAIN will be returned if th <<
2886 * other negative error code will be returned <<
2887 * if the pipe has insufficient space, we rea <<
2888 * hole. <<
2889 */ <<
2890 ssize_t filemap_splice_read(struct file *in, <<
2891 struct pipe_inode <<
2892 size_t len, unsig <<
2893 { <<
2894 struct folio_batch fbatch; <<
2895 struct kiocb iocb; <<
2896 size_t total_spliced = 0, used, npage <<
2897 loff_t isize, end_offset; <<
2898 bool writably_mapped; <<
2899 int i, error = 0; <<
2900 <<
2901 if (unlikely(*ppos >= in->f_mapping-> <<
2902 return 0; <<
2903 <<
2904 init_sync_kiocb(&iocb, in); <<
2905 iocb.ki_pos = *ppos; <<
2906 <<
2907 /* Work out how much data we can actu <<
2908 used = pipe_occupancy(pipe->head, pip <<
2909 npages = max_t(ssize_t, pipe->max_usa <<
2910 len = min_t(size_t, len, npages * PAG <<
2911 <<
2912 folio_batch_init(&fbatch); <<
2913 <<
2914 do { <<
2915 cond_resched(); <<
2916 <<
2917 if (*ppos >= i_size_read(in-> <<
2918 break; <<
2919 <<
2920 iocb.ki_pos = *ppos; <<
2921 error = filemap_get_pages(&io <<
2922 if (error < 0) <<
2923 break; <<
2924 <<
2925 /* <<
2926 * i_size must be checked aft <<
2927 * <<
2928 * Checking i_size after the <<
2929 * the correct value for "nr" <<
2930 * part of the page is not co <<
2931 * another truncate extends t <<
2932 */ <<
2933 isize = i_size_read(in->f_map <<
2934 if (unlikely(*ppos >= isize)) <<
2935 break; <<
2936 end_offset = min_t(loff_t, is <<
2937 <<
2938 /* <<
2939 * Once we start copying data <<
2940 * cachelines that might be c <<
2941 */ <<
2942 writably_mapped = mapping_wri <<
2943 <<
2944 for (i = 0; i < folio_batch_c <<
2945 struct folio *folio = <<
2946 size_t n; <<
2947 <<
2948 if (folio_pos(folio) <<
2949 goto out; <<
2950 folio_mark_accessed(f <<
2951 <<
2952 /* <<
2953 * If users can be wr <<
2954 * virtual addresses, <<
2955 * before reading the <<
2956 */ <<
2957 if (writably_mapped) <<
2958 flush_dcache_ <<
2959 <<
2960 n = min_t(loff_t, len <<
2961 n = splice_folio_into <<
2962 if (!n) <<
2963 goto out; <<
2964 len -= n; <<
2965 total_spliced += n; <<
2966 *ppos += n; <<
2967 in->f_ra.prev_pos = * <<
2968 if (pipe_full(pipe->h <<
2969 goto out; <<
2970 } <<
2971 <<
2972 folio_batch_release(&fbatch); <<
2973 } while (len); <<
2974 <<
2975 out: <<
2976 folio_batch_release(&fbatch); <<
2977 file_accessed(in); <<
2978 <<
2979 return total_spliced ? total_spliced <<
2980 } <<
2981 EXPORT_SYMBOL(filemap_splice_read); <<
2982 <<
2983 static inline loff_t folio_seek_hole_data(str 2810 static inline loff_t folio_seek_hole_data(struct xa_state *xas,
2984 struct address_space *mapping 2811 struct address_space *mapping, struct folio *folio,
2985 loff_t start, loff_t end, boo 2812 loff_t start, loff_t end, bool seek_data)
2986 { 2813 {
2987 const struct address_space_operations 2814 const struct address_space_operations *ops = mapping->a_ops;
2988 size_t offset, bsz = i_blocksize(mapp 2815 size_t offset, bsz = i_blocksize(mapping->host);
2989 2816
2990 if (xa_is_value(folio) || folio_test_ 2817 if (xa_is_value(folio) || folio_test_uptodate(folio))
2991 return seek_data ? start : en 2818 return seek_data ? start : end;
2992 if (!ops->is_partially_uptodate) 2819 if (!ops->is_partially_uptodate)
2993 return seek_data ? end : star 2820 return seek_data ? end : start;
2994 2821
2995 xas_pause(xas); 2822 xas_pause(xas);
2996 rcu_read_unlock(); 2823 rcu_read_unlock();
2997 folio_lock(folio); 2824 folio_lock(folio);
2998 if (unlikely(folio->mapping != mappin 2825 if (unlikely(folio->mapping != mapping))
2999 goto unlock; 2826 goto unlock;
3000 2827
3001 offset = offset_in_folio(folio, start 2828 offset = offset_in_folio(folio, start) & ~(bsz - 1);
3002 2829
3003 do { 2830 do {
3004 if (ops->is_partially_uptodat 2831 if (ops->is_partially_uptodate(folio, offset, bsz) ==
3005 2832 seek_data)
3006 break; 2833 break;
3007 start = (start + bsz) & ~(bsz 2834 start = (start + bsz) & ~(bsz - 1);
3008 offset += bsz; 2835 offset += bsz;
3009 } while (offset < folio_size(folio)); 2836 } while (offset < folio_size(folio));
3010 unlock: 2837 unlock:
3011 folio_unlock(folio); 2838 folio_unlock(folio);
3012 rcu_read_lock(); 2839 rcu_read_lock();
3013 return start; 2840 return start;
3014 } 2841 }
3015 2842
3016 static inline size_t seek_folio_size(struct x 2843 static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio)
3017 { 2844 {
3018 if (xa_is_value(folio)) 2845 if (xa_is_value(folio))
3019 return PAGE_SIZE << xas_get_o !! 2846 return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
3020 return folio_size(folio); 2847 return folio_size(folio);
3021 } 2848 }
3022 2849
3023 /** 2850 /**
3024 * mapping_seek_hole_data - Seek for SEEK_DAT 2851 * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
3025 * @mapping: Address space to search. 2852 * @mapping: Address space to search.
3026 * @start: First byte to consider. 2853 * @start: First byte to consider.
3027 * @end: Limit of search (exclusive). 2854 * @end: Limit of search (exclusive).
3028 * @whence: Either SEEK_HOLE or SEEK_DATA. 2855 * @whence: Either SEEK_HOLE or SEEK_DATA.
3029 * 2856 *
3030 * If the page cache knows which blocks conta 2857 * If the page cache knows which blocks contain holes and which blocks
3031 * contain data, your filesystem can use this 2858 * contain data, your filesystem can use this function to implement
3032 * SEEK_HOLE and SEEK_DATA. This is useful f 2859 * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are
3033 * entirely memory-based such as tmpfs, and f 2860 * entirely memory-based such as tmpfs, and filesystems which support
3034 * unwritten extents. 2861 * unwritten extents.
3035 * 2862 *
3036 * Return: The requested offset on success, o 2863 * Return: The requested offset on success, or -ENXIO if @whence specifies
3037 * SEEK_DATA and there is no data after @star 2864 * SEEK_DATA and there is no data after @start. There is an implicit hole
3038 * after @end - 1, so SEEK_HOLE returns @end 2865 * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
3039 * and @end contain data. 2866 * and @end contain data.
3040 */ 2867 */
3041 loff_t mapping_seek_hole_data(struct address_ 2868 loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
3042 loff_t end, int whence) 2869 loff_t end, int whence)
3043 { 2870 {
3044 XA_STATE(xas, &mapping->i_pages, star 2871 XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
3045 pgoff_t max = (end - 1) >> PAGE_SHIFT 2872 pgoff_t max = (end - 1) >> PAGE_SHIFT;
3046 bool seek_data = (whence == SEEK_DATA 2873 bool seek_data = (whence == SEEK_DATA);
3047 struct folio *folio; 2874 struct folio *folio;
3048 2875
3049 if (end <= start) 2876 if (end <= start)
3050 return -ENXIO; 2877 return -ENXIO;
3051 2878
3052 rcu_read_lock(); 2879 rcu_read_lock();
3053 while ((folio = find_get_entry(&xas, 2880 while ((folio = find_get_entry(&xas, max, XA_PRESENT))) {
3054 loff_t pos = (u64)xas.xa_inde 2881 loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
3055 size_t seek_size; 2882 size_t seek_size;
3056 2883
3057 if (start < pos) { 2884 if (start < pos) {
3058 if (!seek_data) 2885 if (!seek_data)
3059 goto unlock; 2886 goto unlock;
3060 start = pos; 2887 start = pos;
3061 } 2888 }
3062 2889
3063 seek_size = seek_folio_size(& 2890 seek_size = seek_folio_size(&xas, folio);
3064 pos = round_up((u64)pos + 1, 2891 pos = round_up((u64)pos + 1, seek_size);
3065 start = folio_seek_hole_data( 2892 start = folio_seek_hole_data(&xas, mapping, folio, start, pos,
3066 seek_data); 2893 seek_data);
3067 if (start < pos) 2894 if (start < pos)
3068 goto unlock; 2895 goto unlock;
3069 if (start >= end) 2896 if (start >= end)
3070 break; 2897 break;
3071 if (seek_size > PAGE_SIZE) 2898 if (seek_size > PAGE_SIZE)
3072 xas_set(&xas, pos >> 2899 xas_set(&xas, pos >> PAGE_SHIFT);
3073 if (!xa_is_value(folio)) 2900 if (!xa_is_value(folio))
3074 folio_put(folio); 2901 folio_put(folio);
3075 } 2902 }
3076 if (seek_data) 2903 if (seek_data)
3077 start = -ENXIO; 2904 start = -ENXIO;
3078 unlock: 2905 unlock:
3079 rcu_read_unlock(); 2906 rcu_read_unlock();
3080 if (folio && !xa_is_value(folio)) 2907 if (folio && !xa_is_value(folio))
3081 folio_put(folio); 2908 folio_put(folio);
3082 if (start > end) 2909 if (start > end)
3083 return end; 2910 return end;
3084 return start; 2911 return start;
3085 } 2912 }
3086 2913
3087 #ifdef CONFIG_MMU 2914 #ifdef CONFIG_MMU
3088 #define MMAP_LOTSAMISS (100) 2915 #define MMAP_LOTSAMISS (100)
3089 /* 2916 /*
3090 * lock_folio_maybe_drop_mmap - lock the page 2917 * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
3091 * @vmf - the vm_fault for this fault. 2918 * @vmf - the vm_fault for this fault.
3092 * @folio - the folio to lock. 2919 * @folio - the folio to lock.
3093 * @fpin - the pointer to the file we may pin 2920 * @fpin - the pointer to the file we may pin (or is already pinned).
3094 * 2921 *
3095 * This works similar to lock_folio_or_retry 2922 * This works similar to lock_folio_or_retry in that it can drop the
3096 * mmap_lock. It differs in that it actually 2923 * mmap_lock. It differs in that it actually returns the folio locked
3097 * if it returns 1 and 0 if it couldn't lock 2924 * if it returns 1 and 0 if it couldn't lock the folio. If we did have
3098 * to drop the mmap_lock then fpin will point 2925 * to drop the mmap_lock then fpin will point to the pinned file and
3099 * needs to be fput()'ed at a later point. 2926 * needs to be fput()'ed at a later point.
3100 */ 2927 */
3101 static int lock_folio_maybe_drop_mmap(struct 2928 static int lock_folio_maybe_drop_mmap(struct vm_fault *vmf, struct folio *folio,
3102 struct f 2929 struct file **fpin)
3103 { 2930 {
3104 if (folio_trylock(folio)) 2931 if (folio_trylock(folio))
3105 return 1; 2932 return 1;
3106 2933
3107 /* 2934 /*
3108 * NOTE! This will make us return wit 2935 * NOTE! This will make us return with VM_FAULT_RETRY, but with
3109 * the fault lock still held. That's !! 2936 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
3110 * is supposed to work. We have way t 2937 * is supposed to work. We have way too many special cases..
3111 */ 2938 */
3112 if (vmf->flags & FAULT_FLAG_RETRY_NOW 2939 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
3113 return 0; 2940 return 0;
3114 2941
3115 *fpin = maybe_unlock_mmap_for_io(vmf, 2942 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
3116 if (vmf->flags & FAULT_FLAG_KILLABLE) 2943 if (vmf->flags & FAULT_FLAG_KILLABLE) {
3117 if (__folio_lock_killable(fol 2944 if (__folio_lock_killable(folio)) {
3118 /* 2945 /*
3119 * We didn't have the !! 2946 * We didn't have the right flags to drop the mmap_lock,
3120 * fault lock, but al !! 2947 * but all fault_handlers only check for fatal signals
3121 * for fatal signals !! 2948 * if we return VM_FAULT_RETRY, so we need to drop the
3122 * so we need to drop !! 2949 * mmap_lock here and return 0 if we don't have a fpin.
3123 * return 0 if we don <<
3124 */ 2950 */
3125 if (*fpin == NULL) 2951 if (*fpin == NULL)
3126 release_fault !! 2952 mmap_read_unlock(vmf->vma->vm_mm);
3127 return 0; 2953 return 0;
3128 } 2954 }
3129 } else 2955 } else
3130 __folio_lock(folio); 2956 __folio_lock(folio);
3131 2957
3132 return 1; 2958 return 1;
3133 } 2959 }
3134 2960
3135 /* 2961 /*
3136 * Synchronous readahead happens when we don' 2962 * Synchronous readahead happens when we don't even find a page in the page
3137 * cache at all. We don't want to perform IO 2963 * cache at all. We don't want to perform IO under the mmap sem, so if we have
3138 * to drop the mmap sem we return the file th 2964 * to drop the mmap sem we return the file that was pinned in order for us to do
3139 * that. If we didn't pin a file then we ret 2965 * that. If we didn't pin a file then we return NULL. The file that is
3140 * returned needs to be fput()'ed when we're 2966 * returned needs to be fput()'ed when we're done with it.
3141 */ 2967 */
3142 static struct file *do_sync_mmap_readahead(st 2968 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
3143 { 2969 {
3144 struct file *file = vmf->vma->vm_file 2970 struct file *file = vmf->vma->vm_file;
3145 struct file_ra_state *ra = &file->f_r 2971 struct file_ra_state *ra = &file->f_ra;
3146 struct address_space *mapping = file- 2972 struct address_space *mapping = file->f_mapping;
3147 DEFINE_READAHEAD(ractl, file, ra, map 2973 DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
3148 struct file *fpin = NULL; 2974 struct file *fpin = NULL;
3149 unsigned long vm_flags = vmf->vma->vm 2975 unsigned long vm_flags = vmf->vma->vm_flags;
3150 unsigned int mmap_miss; 2976 unsigned int mmap_miss;
3151 2977
3152 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 2978 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3153 /* Use the readahead code, even if re 2979 /* Use the readahead code, even if readahead is disabled */
3154 if ((vm_flags & VM_HUGEPAGE) && HPAGE !! 2980 if (vm_flags & VM_HUGEPAGE) {
3155 fpin = maybe_unlock_mmap_for_ 2981 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3156 ractl._index &= ~((unsigned l 2982 ractl._index &= ~((unsigned long)HPAGE_PMD_NR - 1);
3157 ra->size = HPAGE_PMD_NR; 2983 ra->size = HPAGE_PMD_NR;
3158 /* 2984 /*
3159 * Fetch two PMD folios, so w 2985 * Fetch two PMD folios, so we get the chance to actually
3160 * readahead, unless we've be 2986 * readahead, unless we've been told not to.
3161 */ 2987 */
3162 if (!(vm_flags & VM_RAND_READ 2988 if (!(vm_flags & VM_RAND_READ))
3163 ra->size *= 2; 2989 ra->size *= 2;
3164 ra->async_size = HPAGE_PMD_NR 2990 ra->async_size = HPAGE_PMD_NR;
3165 page_cache_ra_order(&ractl, r 2991 page_cache_ra_order(&ractl, ra, HPAGE_PMD_ORDER);
3166 return fpin; 2992 return fpin;
3167 } 2993 }
3168 #endif 2994 #endif
3169 2995
3170 /* If we don't want any read-ahead, d 2996 /* If we don't want any read-ahead, don't bother */
3171 if (vm_flags & VM_RAND_READ) 2997 if (vm_flags & VM_RAND_READ)
3172 return fpin; 2998 return fpin;
3173 if (!ra->ra_pages) 2999 if (!ra->ra_pages)
3174 return fpin; 3000 return fpin;
3175 3001
3176 if (vm_flags & VM_SEQ_READ) { 3002 if (vm_flags & VM_SEQ_READ) {
3177 fpin = maybe_unlock_mmap_for_ 3003 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3178 page_cache_sync_ra(&ractl, ra 3004 page_cache_sync_ra(&ractl, ra->ra_pages);
3179 return fpin; 3005 return fpin;
3180 } 3006 }
3181 3007
3182 /* Avoid banging the cache line if no 3008 /* Avoid banging the cache line if not needed */
3183 mmap_miss = READ_ONCE(ra->mmap_miss); 3009 mmap_miss = READ_ONCE(ra->mmap_miss);
3184 if (mmap_miss < MMAP_LOTSAMISS * 10) 3010 if (mmap_miss < MMAP_LOTSAMISS * 10)
3185 WRITE_ONCE(ra->mmap_miss, ++m 3011 WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
3186 3012
3187 /* 3013 /*
3188 * Do we miss much more than hit in t 3014 * Do we miss much more than hit in this file? If so,
3189 * stop bothering with read-ahead. It 3015 * stop bothering with read-ahead. It will only hurt.
3190 */ 3016 */
3191 if (mmap_miss > MMAP_LOTSAMISS) 3017 if (mmap_miss > MMAP_LOTSAMISS)
3192 return fpin; 3018 return fpin;
3193 3019
3194 /* 3020 /*
3195 * mmap read-around 3021 * mmap read-around
3196 */ 3022 */
3197 fpin = maybe_unlock_mmap_for_io(vmf, 3023 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3198 ra->start = max_t(long, 0, vmf->pgoff 3024 ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
3199 ra->size = ra->ra_pages; 3025 ra->size = ra->ra_pages;
3200 ra->async_size = ra->ra_pages / 4; 3026 ra->async_size = ra->ra_pages / 4;
3201 ractl._index = ra->start; 3027 ractl._index = ra->start;
3202 page_cache_ra_order(&ractl, ra, 0); 3028 page_cache_ra_order(&ractl, ra, 0);
3203 return fpin; 3029 return fpin;
3204 } 3030 }
3205 3031
3206 /* 3032 /*
3207 * Asynchronous readahead happens when we fin 3033 * Asynchronous readahead happens when we find the page and PG_readahead,
3208 * so we want to possibly extend the readahea 3034 * so we want to possibly extend the readahead further. We return the file that
3209 * was pinned if we have to drop the mmap_loc 3035 * was pinned if we have to drop the mmap_lock in order to do IO.
3210 */ 3036 */
3211 static struct file *do_async_mmap_readahead(s 3037 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
3212 s 3038 struct folio *folio)
3213 { 3039 {
3214 struct file *file = vmf->vma->vm_file 3040 struct file *file = vmf->vma->vm_file;
3215 struct file_ra_state *ra = &file->f_r 3041 struct file_ra_state *ra = &file->f_ra;
3216 DEFINE_READAHEAD(ractl, file, ra, fil 3042 DEFINE_READAHEAD(ractl, file, ra, file->f_mapping, vmf->pgoff);
3217 struct file *fpin = NULL; 3043 struct file *fpin = NULL;
3218 unsigned int mmap_miss; 3044 unsigned int mmap_miss;
3219 3045
3220 /* If we don't want any read-ahead, d 3046 /* If we don't want any read-ahead, don't bother */
3221 if (vmf->vma->vm_flags & VM_RAND_READ 3047 if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
3222 return fpin; 3048 return fpin;
3223 3049
3224 mmap_miss = READ_ONCE(ra->mmap_miss); 3050 mmap_miss = READ_ONCE(ra->mmap_miss);
3225 if (mmap_miss) 3051 if (mmap_miss)
3226 WRITE_ONCE(ra->mmap_miss, --m 3052 WRITE_ONCE(ra->mmap_miss, --mmap_miss);
3227 3053
3228 if (folio_test_readahead(folio)) { 3054 if (folio_test_readahead(folio)) {
3229 fpin = maybe_unlock_mmap_for_ 3055 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3230 page_cache_async_ra(&ractl, f 3056 page_cache_async_ra(&ractl, folio, ra->ra_pages);
3231 } 3057 }
3232 return fpin; 3058 return fpin;
3233 } 3059 }
3234 3060
3235 static vm_fault_t filemap_fault_recheck_pte_n <<
3236 { <<
3237 struct vm_area_struct *vma = vmf->vma <<
3238 vm_fault_t ret = 0; <<
3239 pte_t *ptep; <<
3240 <<
3241 /* <<
3242 * We might have COW'ed a pagecache f <<
3243 * anon folio mapped. The original pa <<
3244 * might have been evicted. During a <<
3245 * the PTE, such as done in do_numa_p <<
3246 * temporarily clear the PTE under PT <<
3247 * "none" when not holding the PT loc <<
3248 * <<
3249 * Not rechecking the PTE under PT lo <<
3250 * major fault in an mlock'ed region. <<
3251 * scenario while holding the PT lock <<
3252 * scenarios. Recheck the PTE without <<
3253 * the number of times we hold PT loc <<
3254 */ <<
3255 if (!(vma->vm_flags & VM_LOCKED)) <<
3256 return 0; <<
3257 <<
3258 if (!(vmf->flags & FAULT_FLAG_ORIG_PT <<
3259 return 0; <<
3260 <<
3261 ptep = pte_offset_map_nolock(vma->vm_ <<
3262 &vmf->pt <<
3263 if (unlikely(!ptep)) <<
3264 return VM_FAULT_NOPAGE; <<
3265 <<
3266 if (unlikely(!pte_none(ptep_get_lockl <<
3267 ret = VM_FAULT_NOPAGE; <<
3268 } else { <<
3269 spin_lock(vmf->ptl); <<
3270 if (unlikely(!pte_none(ptep_g <<
3271 ret = VM_FAULT_NOPAGE <<
3272 spin_unlock(vmf->ptl); <<
3273 } <<
3274 pte_unmap(ptep); <<
3275 return ret; <<
3276 } <<
3277 <<
3278 /** 3061 /**
3279 * filemap_fault - read in file data for page 3062 * filemap_fault - read in file data for page fault handling
3280 * @vmf: struct vm_fault containing de 3063 * @vmf: struct vm_fault containing details of the fault
3281 * 3064 *
3282 * filemap_fault() is invoked via the vma ope 3065 * filemap_fault() is invoked via the vma operations vector for a
3283 * mapped memory region to read in file data 3066 * mapped memory region to read in file data during a page fault.
3284 * 3067 *
3285 * The goto's are kind of ugly, but this stre 3068 * The goto's are kind of ugly, but this streamlines the normal case of having
3286 * it in the page cache, and handles the spec 3069 * it in the page cache, and handles the special cases reasonably without
3287 * having a lot of duplicated code. 3070 * having a lot of duplicated code.
3288 * 3071 *
3289 * vma->vm_mm->mmap_lock must be held on entr 3072 * vma->vm_mm->mmap_lock must be held on entry.
3290 * 3073 *
3291 * If our return value has VM_FAULT_RETRY set 3074 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
3292 * may be dropped before doing I/O or by lock 3075 * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap().
3293 * 3076 *
3294 * If our return value does not have VM_FAULT 3077 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
3295 * has not been released. 3078 * has not been released.
3296 * 3079 *
3297 * We never return with VM_FAULT_RETRY and a 3080 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3298 * 3081 *
3299 * Return: bitwise-OR of %VM_FAULT_ codes. 3082 * Return: bitwise-OR of %VM_FAULT_ codes.
3300 */ 3083 */
3301 vm_fault_t filemap_fault(struct vm_fault *vmf 3084 vm_fault_t filemap_fault(struct vm_fault *vmf)
3302 { 3085 {
3303 int error; 3086 int error;
3304 struct file *file = vmf->vma->vm_file 3087 struct file *file = vmf->vma->vm_file;
3305 struct file *fpin = NULL; 3088 struct file *fpin = NULL;
3306 struct address_space *mapping = file- 3089 struct address_space *mapping = file->f_mapping;
3307 struct inode *inode = mapping->host; 3090 struct inode *inode = mapping->host;
3308 pgoff_t max_idx, index = vmf->pgoff; 3091 pgoff_t max_idx, index = vmf->pgoff;
3309 struct folio *folio; 3092 struct folio *folio;
3310 vm_fault_t ret = 0; 3093 vm_fault_t ret = 0;
3311 bool mapping_locked = false; 3094 bool mapping_locked = false;
3312 3095
3313 max_idx = DIV_ROUND_UP(i_size_read(in 3096 max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3314 if (unlikely(index >= max_idx)) 3097 if (unlikely(index >= max_idx))
3315 return VM_FAULT_SIGBUS; 3098 return VM_FAULT_SIGBUS;
3316 3099
3317 trace_mm_filemap_fault(mapping, index <<
3318 <<
3319 /* 3100 /*
3320 * Do we have something in the page c 3101 * Do we have something in the page cache already?
3321 */ 3102 */
3322 folio = filemap_get_folio(mapping, in 3103 folio = filemap_get_folio(mapping, index);
3323 if (likely(!IS_ERR(folio))) { !! 3104 if (likely(folio)) {
3324 /* 3105 /*
3325 * We found the page, so try 3106 * We found the page, so try async readahead before waiting for
3326 * the lock. 3107 * the lock.
3327 */ 3108 */
3328 if (!(vmf->flags & FAULT_FLAG 3109 if (!(vmf->flags & FAULT_FLAG_TRIED))
3329 fpin = do_async_mmap_ 3110 fpin = do_async_mmap_readahead(vmf, folio);
3330 if (unlikely(!folio_test_upto 3111 if (unlikely(!folio_test_uptodate(folio))) {
3331 filemap_invalidate_lo 3112 filemap_invalidate_lock_shared(mapping);
3332 mapping_locked = true 3113 mapping_locked = true;
3333 } 3114 }
3334 } else { 3115 } else {
3335 ret = filemap_fault_recheck_p <<
3336 if (unlikely(ret)) <<
3337 return ret; <<
3338 <<
3339 /* No page in the page cache 3116 /* No page in the page cache at all */
3340 count_vm_event(PGMAJFAULT); 3117 count_vm_event(PGMAJFAULT);
3341 count_memcg_event_mm(vmf->vma 3118 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3342 ret = VM_FAULT_MAJOR; 3119 ret = VM_FAULT_MAJOR;
3343 fpin = do_sync_mmap_readahead 3120 fpin = do_sync_mmap_readahead(vmf);
3344 retry_find: 3121 retry_find:
3345 /* 3122 /*
3346 * See comment in filemap_cre 3123 * See comment in filemap_create_folio() why we need
3347 * invalidate_lock 3124 * invalidate_lock
3348 */ 3125 */
3349 if (!mapping_locked) { 3126 if (!mapping_locked) {
3350 filemap_invalidate_lo 3127 filemap_invalidate_lock_shared(mapping);
3351 mapping_locked = true 3128 mapping_locked = true;
3352 } 3129 }
3353 folio = __filemap_get_folio(m 3130 folio = __filemap_get_folio(mapping, index,
3354 FGP 3131 FGP_CREAT|FGP_FOR_MMAP,
3355 vmf 3132 vmf->gfp_mask);
3356 if (IS_ERR(folio)) { !! 3133 if (!folio) {
3357 if (fpin) 3134 if (fpin)
3358 goto out_retr 3135 goto out_retry;
3359 filemap_invalidate_un 3136 filemap_invalidate_unlock_shared(mapping);
3360 return VM_FAULT_OOM; 3137 return VM_FAULT_OOM;
3361 } 3138 }
3362 } 3139 }
3363 3140
3364 if (!lock_folio_maybe_drop_mmap(vmf, 3141 if (!lock_folio_maybe_drop_mmap(vmf, folio, &fpin))
3365 goto out_retry; 3142 goto out_retry;
3366 3143
3367 /* Did it get truncated? */ 3144 /* Did it get truncated? */
3368 if (unlikely(folio->mapping != mappin 3145 if (unlikely(folio->mapping != mapping)) {
3369 folio_unlock(folio); 3146 folio_unlock(folio);
3370 folio_put(folio); 3147 folio_put(folio);
3371 goto retry_find; 3148 goto retry_find;
3372 } 3149 }
3373 VM_BUG_ON_FOLIO(!folio_contains(folio 3150 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
3374 3151
3375 /* 3152 /*
3376 * We have a locked folio in the page !! 3153 * We have a locked page in the page cache, now we need to check
3377 * that it's up-to-date. If not, it i !! 3154 * that it's up-to-date. If not, it is going to be due to an error.
3378 * or because readahead was otherwise <<
3379 */ 3155 */
3380 if (unlikely(!folio_test_uptodate(fol 3156 if (unlikely(!folio_test_uptodate(folio))) {
3381 /* 3157 /*
3382 * If the invalidate lock is !! 3158 * The page was in cache and uptodate and now it is not.
3383 * and uptodate and now it is !! 3159 * Strange but possible since we didn't hold the page lock all
3384 * didn't hold the page lock !! 3160 * the time. Let's drop everything get the invalidate lock and
3385 * everything, get the invali !! 3161 * try again.
3386 */ 3162 */
3387 if (!mapping_locked) { 3163 if (!mapping_locked) {
3388 folio_unlock(folio); 3164 folio_unlock(folio);
3389 folio_put(folio); 3165 folio_put(folio);
3390 goto retry_find; 3166 goto retry_find;
3391 } 3167 }
3392 <<
3393 /* <<
3394 * OK, the folio is really no <<
3395 * VMA has the VM_RAND_READ f <<
3396 * arose. Let's read it in di <<
3397 */ <<
3398 goto page_not_uptodate; 3168 goto page_not_uptodate;
3399 } 3169 }
3400 3170
3401 /* 3171 /*
3402 * We've made it this far and we had 3172 * We've made it this far and we had to drop our mmap_lock, now is the
3403 * time to return to the upper layer 3173 * time to return to the upper layer and have it re-find the vma and
3404 * redo the fault. 3174 * redo the fault.
3405 */ 3175 */
3406 if (fpin) { 3176 if (fpin) {
3407 folio_unlock(folio); 3177 folio_unlock(folio);
3408 goto out_retry; 3178 goto out_retry;
3409 } 3179 }
3410 if (mapping_locked) 3180 if (mapping_locked)
3411 filemap_invalidate_unlock_sha 3181 filemap_invalidate_unlock_shared(mapping);
3412 3182
3413 /* 3183 /*
3414 * Found the page and have a referenc 3184 * Found the page and have a reference on it.
3415 * We must recheck i_size under page 3185 * We must recheck i_size under page lock.
3416 */ 3186 */
3417 max_idx = DIV_ROUND_UP(i_size_read(in 3187 max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3418 if (unlikely(index >= max_idx)) { 3188 if (unlikely(index >= max_idx)) {
3419 folio_unlock(folio); 3189 folio_unlock(folio);
3420 folio_put(folio); 3190 folio_put(folio);
3421 return VM_FAULT_SIGBUS; 3191 return VM_FAULT_SIGBUS;
3422 } 3192 }
3423 3193
3424 vmf->page = folio_file_page(folio, in 3194 vmf->page = folio_file_page(folio, index);
3425 return ret | VM_FAULT_LOCKED; 3195 return ret | VM_FAULT_LOCKED;
3426 3196
3427 page_not_uptodate: 3197 page_not_uptodate:
3428 /* 3198 /*
3429 * Umm, take care of errors if the pa 3199 * Umm, take care of errors if the page isn't up-to-date.
3430 * Try to re-read it _once_. We do th 3200 * Try to re-read it _once_. We do this synchronously,
3431 * because there really aren't any pe 3201 * because there really aren't any performance issues here
3432 * and we need to check for errors. 3202 * and we need to check for errors.
3433 */ 3203 */
3434 fpin = maybe_unlock_mmap_for_io(vmf, 3204 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3435 error = filemap_read_folio(file, mapp 3205 error = filemap_read_folio(file, mapping->a_ops->read_folio, folio);
3436 if (fpin) 3206 if (fpin)
3437 goto out_retry; 3207 goto out_retry;
3438 folio_put(folio); 3208 folio_put(folio);
3439 3209
3440 if (!error || error == AOP_TRUNCATED_ 3210 if (!error || error == AOP_TRUNCATED_PAGE)
3441 goto retry_find; 3211 goto retry_find;
3442 filemap_invalidate_unlock_shared(mapp 3212 filemap_invalidate_unlock_shared(mapping);
3443 3213
3444 return VM_FAULT_SIGBUS; 3214 return VM_FAULT_SIGBUS;
3445 3215
3446 out_retry: 3216 out_retry:
3447 /* 3217 /*
3448 * We dropped the mmap_lock, we need 3218 * We dropped the mmap_lock, we need to return to the fault handler to
3449 * re-find the vma and come back and 3219 * re-find the vma and come back and find our hopefully still populated
3450 * page. 3220 * page.
3451 */ 3221 */
3452 if (!IS_ERR(folio)) !! 3222 if (folio)
3453 folio_put(folio); 3223 folio_put(folio);
3454 if (mapping_locked) 3224 if (mapping_locked)
3455 filemap_invalidate_unlock_sha 3225 filemap_invalidate_unlock_shared(mapping);
3456 if (fpin) 3226 if (fpin)
3457 fput(fpin); 3227 fput(fpin);
3458 return ret | VM_FAULT_RETRY; 3228 return ret | VM_FAULT_RETRY;
3459 } 3229 }
3460 EXPORT_SYMBOL(filemap_fault); 3230 EXPORT_SYMBOL(filemap_fault);
3461 3231
3462 static bool filemap_map_pmd(struct vm_fault * !! 3232 static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page)
3463 pgoff_t start) <<
3464 { 3233 {
3465 struct mm_struct *mm = vmf->vma->vm_m 3234 struct mm_struct *mm = vmf->vma->vm_mm;
3466 3235
3467 /* Huge page is mapped? No need to pr 3236 /* Huge page is mapped? No need to proceed. */
3468 if (pmd_trans_huge(*vmf->pmd)) { 3237 if (pmd_trans_huge(*vmf->pmd)) {
3469 folio_unlock(folio); !! 3238 unlock_page(page);
3470 folio_put(folio); !! 3239 put_page(page);
3471 return true; 3240 return true;
3472 } 3241 }
3473 3242
3474 if (pmd_none(*vmf->pmd) && folio_test !! 3243 if (pmd_none(*vmf->pmd) && PageTransHuge(page)) {
3475 struct page *page = folio_fil <<
3476 vm_fault_t ret = do_set_pmd(v 3244 vm_fault_t ret = do_set_pmd(vmf, page);
3477 if (!ret) { 3245 if (!ret) {
3478 /* The page is mapped 3246 /* The page is mapped successfully, reference consumed. */
3479 folio_unlock(folio); !! 3247 unlock_page(page);
3480 return true; 3248 return true;
3481 } 3249 }
3482 } 3250 }
3483 3251
3484 if (pmd_none(*vmf->pmd) && vmf->preal !! 3252 if (pmd_none(*vmf->pmd))
3485 pmd_install(mm, vmf->pmd, &vm 3253 pmd_install(mm, vmf->pmd, &vmf->prealloc_pte);
3486 3254
>> 3255 /* See comment in handle_pte_fault() */
>> 3256 if (pmd_devmap_trans_unstable(vmf->pmd)) {
>> 3257 unlock_page(page);
>> 3258 put_page(page);
>> 3259 return true;
>> 3260 }
>> 3261
3487 return false; 3262 return false;
3488 } 3263 }
3489 3264
3490 static struct folio *next_uptodate_folio(stru !! 3265 static struct folio *next_uptodate_page(struct folio *folio,
3491 struct address_space *mapping !! 3266 struct address_space *mapping,
>> 3267 struct xa_state *xas, pgoff_t end_pgoff)
3492 { 3268 {
3493 struct folio *folio = xas_next_entry( <<
3494 unsigned long max_idx; 3269 unsigned long max_idx;
3495 3270
3496 do { 3271 do {
3497 if (!folio) 3272 if (!folio)
3498 return NULL; 3273 return NULL;
3499 if (xas_retry(xas, folio)) 3274 if (xas_retry(xas, folio))
3500 continue; 3275 continue;
3501 if (xa_is_value(folio)) 3276 if (xa_is_value(folio))
3502 continue; 3277 continue;
3503 if (folio_test_locked(folio)) 3278 if (folio_test_locked(folio))
3504 continue; 3279 continue;
3505 if (!folio_try_get(folio)) !! 3280 if (!folio_try_get_rcu(folio))
3506 continue; 3281 continue;
3507 /* Has the page moved or been 3282 /* Has the page moved or been split? */
3508 if (unlikely(folio != xas_rel 3283 if (unlikely(folio != xas_reload(xas)))
3509 goto skip; 3284 goto skip;
3510 if (!folio_test_uptodate(foli 3285 if (!folio_test_uptodate(folio) || folio_test_readahead(folio))
3511 goto skip; 3286 goto skip;
3512 if (!folio_trylock(folio)) 3287 if (!folio_trylock(folio))
3513 goto skip; 3288 goto skip;
3514 if (folio->mapping != mapping 3289 if (folio->mapping != mapping)
3515 goto unlock; 3290 goto unlock;
3516 if (!folio_test_uptodate(foli 3291 if (!folio_test_uptodate(folio))
3517 goto unlock; 3292 goto unlock;
3518 max_idx = DIV_ROUND_UP(i_size 3293 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3519 if (xas->xa_index >= max_idx) 3294 if (xas->xa_index >= max_idx)
3520 goto unlock; 3295 goto unlock;
3521 return folio; 3296 return folio;
3522 unlock: 3297 unlock:
3523 folio_unlock(folio); 3298 folio_unlock(folio);
3524 skip: 3299 skip:
3525 folio_put(folio); 3300 folio_put(folio);
3526 } while ((folio = xas_next_entry(xas, 3301 } while ((folio = xas_next_entry(xas, end_pgoff)) != NULL);
3527 3302
3528 return NULL; 3303 return NULL;
3529 } 3304 }
3530 3305
3531 /* !! 3306 static inline struct folio *first_map_page(struct address_space *mapping,
3532 * Map page range [start_page, start_page + n !! 3307 struct xa_state *xas,
3533 * start_page is gotten from start by folio_p !! 3308 pgoff_t end_pgoff)
3534 */ <<
3535 static vm_fault_t filemap_map_folio_range(str <<
3536 struct folio *folio, <<
3537 unsigned long addr, u <<
3538 unsigned long *rss, u <<
3539 { 3309 {
3540 vm_fault_t ret = 0; !! 3310 return next_uptodate_page(xas_find(xas, end_pgoff),
3541 struct page *page = folio_page(folio, !! 3311 mapping, xas, end_pgoff);
3542 unsigned int count = 0; <<
3543 pte_t *old_ptep = vmf->pte; <<
3544 <<
3545 do { <<
3546 if (PageHWPoison(page + count <<
3547 goto skip; <<
3548 <<
3549 /* <<
3550 * If there are too many foli <<
3551 * in a file, they will proba <<
3552 * In such situation, read-ah <<
3553 * Don't decrease mmap_miss i <<
3554 * we can stop read-ahead. <<
3555 */ <<
3556 if (!folio_test_workingset(fo <<
3557 (*mmap_miss)++; <<
3558 <<
3559 /* <<
3560 * NOTE: If there're PTE mark <<
3561 * handled in the specific fa <<
3562 * fault-around logic. <<
3563 */ <<
3564 if (!pte_none(ptep_get(&vmf-> <<
3565 goto skip; <<
3566 <<
3567 count++; <<
3568 continue; <<
3569 skip: <<
3570 if (count) { <<
3571 set_pte_range(vmf, fo <<
3572 *rss += count; <<
3573 folio_ref_add(folio, <<
3574 if (in_range(vmf->add <<
3575 ret = VM_FAUL <<
3576 } <<
3577 <<
3578 count++; <<
3579 page += count; <<
3580 vmf->pte += count; <<
3581 addr += count * PAGE_SIZE; <<
3582 count = 0; <<
3583 } while (--nr_pages > 0); <<
3584 <<
3585 if (count) { <<
3586 set_pte_range(vmf, folio, pag <<
3587 *rss += count; <<
3588 folio_ref_add(folio, count); <<
3589 if (in_range(vmf->address, ad <<
3590 ret = VM_FAULT_NOPAGE <<
3591 } <<
3592 <<
3593 vmf->pte = old_ptep; <<
3594 <<
3595 return ret; <<
3596 } 3312 }
3597 3313
3598 static vm_fault_t filemap_map_order0_folio(st !! 3314 static inline struct folio *next_map_page(struct address_space *mapping,
3599 struct folio *folio, unsigned !! 3315 struct xa_state *xas,
3600 unsigned long *rss, unsigned !! 3316 pgoff_t end_pgoff)
3601 { 3317 {
3602 vm_fault_t ret = 0; !! 3318 return next_uptodate_page(xas_next_entry(xas, end_pgoff),
3603 struct page *page = &folio->page; !! 3319 mapping, xas, end_pgoff);
3604 <<
3605 if (PageHWPoison(page)) <<
3606 return ret; <<
3607 <<
3608 /* See comment of filemap_map_folio_r <<
3609 if (!folio_test_workingset(folio)) <<
3610 (*mmap_miss)++; <<
3611 <<
3612 /* <<
3613 * NOTE: If there're PTE markers, we' <<
3614 * handled in the specific fault path <<
3615 * the fault-around logic. <<
3616 */ <<
3617 if (!pte_none(ptep_get(vmf->pte))) <<
3618 return ret; <<
3619 <<
3620 if (vmf->address == addr) <<
3621 ret = VM_FAULT_NOPAGE; <<
3622 <<
3623 set_pte_range(vmf, folio, page, 1, ad <<
3624 (*rss)++; <<
3625 folio_ref_inc(folio); <<
3626 <<
3627 return ret; <<
3628 } 3320 }
3629 3321
3630 vm_fault_t filemap_map_pages(struct vm_fault 3322 vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3631 pgoff_t start_pg 3323 pgoff_t start_pgoff, pgoff_t end_pgoff)
3632 { 3324 {
3633 struct vm_area_struct *vma = vmf->vma 3325 struct vm_area_struct *vma = vmf->vma;
3634 struct file *file = vma->vm_file; 3326 struct file *file = vma->vm_file;
3635 struct address_space *mapping = file- 3327 struct address_space *mapping = file->f_mapping;
3636 pgoff_t file_end, last_pgoff = start_ !! 3328 pgoff_t last_pgoff = start_pgoff;
3637 unsigned long addr; 3329 unsigned long addr;
3638 XA_STATE(xas, &mapping->i_pages, star 3330 XA_STATE(xas, &mapping->i_pages, start_pgoff);
3639 struct folio *folio; 3331 struct folio *folio;
>> 3332 struct page *page;
>> 3333 unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
3640 vm_fault_t ret = 0; 3334 vm_fault_t ret = 0;
3641 unsigned long rss = 0; <<
3642 unsigned int nr_pages = 0, mmap_miss <<
3643 3335
3644 rcu_read_lock(); 3336 rcu_read_lock();
3645 folio = next_uptodate_folio(&xas, map !! 3337 folio = first_map_page(mapping, &xas, end_pgoff);
3646 if (!folio) 3338 if (!folio)
3647 goto out; 3339 goto out;
3648 3340
3649 if (filemap_map_pmd(vmf, folio, start !! 3341 if (filemap_map_pmd(vmf, &folio->page)) {
3650 ret = VM_FAULT_NOPAGE; 3342 ret = VM_FAULT_NOPAGE;
3651 goto out; 3343 goto out;
3652 } 3344 }
3653 3345
3654 addr = vma->vm_start + ((start_pgoff 3346 addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3655 vmf->pte = pte_offset_map_lock(vma->v 3347 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3656 if (!vmf->pte) { <<
3657 folio_unlock(folio); <<
3658 folio_put(folio); <<
3659 goto out; <<
3660 } <<
3661 <<
3662 file_end = DIV_ROUND_UP(i_size_read(m <<
3663 if (end_pgoff > file_end) <<
3664 end_pgoff = file_end; <<
3665 <<
3666 folio_type = mm_counter_file(folio); <<
3667 do { 3348 do {
3668 unsigned long end; !! 3349 again:
>> 3350 page = folio_file_page(folio, xas.xa_index);
>> 3351 if (PageHWPoison(page))
>> 3352 goto unlock;
>> 3353
>> 3354 if (mmap_miss > 0)
>> 3355 mmap_miss--;
3669 3356
3670 addr += (xas.xa_index - last_ 3357 addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3671 vmf->pte += xas.xa_index - la 3358 vmf->pte += xas.xa_index - last_pgoff;
3672 last_pgoff = xas.xa_index; 3359 last_pgoff = xas.xa_index;
3673 end = folio_next_index(folio) <<
3674 nr_pages = min(end, end_pgoff <<
3675 3360
3676 if (!folio_test_large(folio)) !! 3361 /*
3677 ret |= filemap_map_or !! 3362 * NOTE: If there're PTE markers, we'll leave them to be
3678 folio !! 3363 * handled in the specific fault path, and it'll prohibit the
3679 else !! 3364 * fault-around logic.
3680 ret |= filemap_map_fo !! 3365 */
3681 xas.x !! 3366 if (!pte_none(*vmf->pte))
3682 nr_pa !! 3367 goto unlock;
>> 3368
>> 3369 /* We're about to handle the fault */
>> 3370 if (vmf->address == addr)
>> 3371 ret = VM_FAULT_NOPAGE;
3683 3372
>> 3373 do_set_pte(vmf, page, addr);
>> 3374 /* no need to invalidate: a not-present page won't be cached */
>> 3375 update_mmu_cache(vma, addr, vmf->pte);
>> 3376 if (folio_more_pages(folio, xas.xa_index, end_pgoff)) {
>> 3377 xas.xa_index++;
>> 3378 folio_ref_inc(folio);
>> 3379 goto again;
>> 3380 }
>> 3381 folio_unlock(folio);
>> 3382 continue;
>> 3383 unlock:
>> 3384 if (folio_more_pages(folio, xas.xa_index, end_pgoff)) {
>> 3385 xas.xa_index++;
>> 3386 goto again;
>> 3387 }
3684 folio_unlock(folio); 3388 folio_unlock(folio);
3685 folio_put(folio); 3389 folio_put(folio);
3686 } while ((folio = next_uptodate_folio !! 3390 } while ((folio = next_map_page(mapping, &xas, end_pgoff)) != NULL);
3687 add_mm_counter(vma->vm_mm, folio_type <<
3688 pte_unmap_unlock(vmf->pte, vmf->ptl); 3391 pte_unmap_unlock(vmf->pte, vmf->ptl);
3689 trace_mm_filemap_map_pages(mapping, s <<
3690 out: 3392 out:
3691 rcu_read_unlock(); 3393 rcu_read_unlock();
3692 !! 3394 WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
3693 mmap_miss_saved = READ_ONCE(file->f_r <<
3694 if (mmap_miss >= mmap_miss_saved) <<
3695 WRITE_ONCE(file->f_ra.mmap_mi <<
3696 else <<
3697 WRITE_ONCE(file->f_ra.mmap_mi <<
3698 <<
3699 return ret; 3395 return ret;
3700 } 3396 }
3701 EXPORT_SYMBOL(filemap_map_pages); 3397 EXPORT_SYMBOL(filemap_map_pages);
3702 3398
3703 vm_fault_t filemap_page_mkwrite(struct vm_fau 3399 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3704 { 3400 {
3705 struct address_space *mapping = vmf-> 3401 struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3706 struct folio *folio = page_folio(vmf- 3402 struct folio *folio = page_folio(vmf->page);
3707 vm_fault_t ret = VM_FAULT_LOCKED; 3403 vm_fault_t ret = VM_FAULT_LOCKED;
3708 3404
3709 sb_start_pagefault(mapping->host->i_s 3405 sb_start_pagefault(mapping->host->i_sb);
3710 file_update_time(vmf->vma->vm_file); 3406 file_update_time(vmf->vma->vm_file);
3711 folio_lock(folio); 3407 folio_lock(folio);
3712 if (folio->mapping != mapping) { 3408 if (folio->mapping != mapping) {
3713 folio_unlock(folio); 3409 folio_unlock(folio);
3714 ret = VM_FAULT_NOPAGE; 3410 ret = VM_FAULT_NOPAGE;
3715 goto out; 3411 goto out;
3716 } 3412 }
3717 /* 3413 /*
3718 * We mark the folio dirty already he 3414 * We mark the folio dirty already here so that when freeze is in
3719 * progress, we are guaranteed that w 3415 * progress, we are guaranteed that writeback during freezing will
3720 * see the dirty folio and writeprote 3416 * see the dirty folio and writeprotect it again.
3721 */ 3417 */
3722 folio_mark_dirty(folio); 3418 folio_mark_dirty(folio);
3723 folio_wait_stable(folio); 3419 folio_wait_stable(folio);
3724 out: 3420 out:
3725 sb_end_pagefault(mapping->host->i_sb) 3421 sb_end_pagefault(mapping->host->i_sb);
3726 return ret; 3422 return ret;
3727 } 3423 }
3728 3424
3729 const struct vm_operations_struct generic_fil 3425 const struct vm_operations_struct generic_file_vm_ops = {
3730 .fault = filemap_fault, 3426 .fault = filemap_fault,
3731 .map_pages = filemap_map_pages, 3427 .map_pages = filemap_map_pages,
3732 .page_mkwrite = filemap_page_mkwrit 3428 .page_mkwrite = filemap_page_mkwrite,
3733 }; 3429 };
3734 3430
3735 /* This is used for a general mmap of a disk 3431 /* This is used for a general mmap of a disk file */
3736 3432
3737 int generic_file_mmap(struct file *file, stru 3433 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3738 { 3434 {
3739 struct address_space *mapping = file- 3435 struct address_space *mapping = file->f_mapping;
3740 3436
3741 if (!mapping->a_ops->read_folio) 3437 if (!mapping->a_ops->read_folio)
3742 return -ENOEXEC; 3438 return -ENOEXEC;
3743 file_accessed(file); 3439 file_accessed(file);
3744 vma->vm_ops = &generic_file_vm_ops; 3440 vma->vm_ops = &generic_file_vm_ops;
3745 return 0; 3441 return 0;
3746 } 3442 }
3747 3443
3748 /* 3444 /*
3749 * This is for filesystems which do not imple 3445 * This is for filesystems which do not implement ->writepage.
3750 */ 3446 */
3751 int generic_file_readonly_mmap(struct file *f 3447 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3752 { 3448 {
3753 if (vma_is_shared_maywrite(vma)) !! 3449 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3754 return -EINVAL; 3450 return -EINVAL;
3755 return generic_file_mmap(file, vma); 3451 return generic_file_mmap(file, vma);
3756 } 3452 }
3757 #else 3453 #else
3758 vm_fault_t filemap_page_mkwrite(struct vm_fau 3454 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3759 { 3455 {
3760 return VM_FAULT_SIGBUS; 3456 return VM_FAULT_SIGBUS;
3761 } 3457 }
3762 int generic_file_mmap(struct file *file, stru 3458 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3763 { 3459 {
3764 return -ENOSYS; 3460 return -ENOSYS;
3765 } 3461 }
3766 int generic_file_readonly_mmap(struct file *f 3462 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3767 { 3463 {
3768 return -ENOSYS; 3464 return -ENOSYS;
3769 } 3465 }
3770 #endif /* CONFIG_MMU */ 3466 #endif /* CONFIG_MMU */
3771 3467
3772 EXPORT_SYMBOL(filemap_page_mkwrite); 3468 EXPORT_SYMBOL(filemap_page_mkwrite);
3773 EXPORT_SYMBOL(generic_file_mmap); 3469 EXPORT_SYMBOL(generic_file_mmap);
3774 EXPORT_SYMBOL(generic_file_readonly_mmap); 3470 EXPORT_SYMBOL(generic_file_readonly_mmap);
3775 3471
3776 static struct folio *do_read_cache_folio(stru 3472 static struct folio *do_read_cache_folio(struct address_space *mapping,
3777 pgoff_t index, filler_t fille 3473 pgoff_t index, filler_t filler, struct file *file, gfp_t gfp)
3778 { 3474 {
3779 struct folio *folio; 3475 struct folio *folio;
3780 int err; 3476 int err;
3781 3477
3782 if (!filler) 3478 if (!filler)
3783 filler = mapping->a_ops->read 3479 filler = mapping->a_ops->read_folio;
3784 repeat: 3480 repeat:
3785 folio = filemap_get_folio(mapping, in 3481 folio = filemap_get_folio(mapping, index);
3786 if (IS_ERR(folio)) { !! 3482 if (!folio) {
3787 folio = filemap_alloc_folio(g !! 3483 folio = filemap_alloc_folio(gfp, 0);
3788 m <<
3789 if (!folio) 3484 if (!folio)
3790 return ERR_PTR(-ENOME 3485 return ERR_PTR(-ENOMEM);
3791 index = mapping_align_index(m <<
3792 err = filemap_add_folio(mappi 3486 err = filemap_add_folio(mapping, folio, index, gfp);
3793 if (unlikely(err)) { 3487 if (unlikely(err)) {
3794 folio_put(folio); 3488 folio_put(folio);
3795 if (err == -EEXIST) 3489 if (err == -EEXIST)
3796 goto repeat; 3490 goto repeat;
3797 /* Presumably ENOMEM 3491 /* Presumably ENOMEM for xarray node */
3798 return ERR_PTR(err); 3492 return ERR_PTR(err);
3799 } 3493 }
3800 3494
3801 goto filler; 3495 goto filler;
3802 } 3496 }
3803 if (folio_test_uptodate(folio)) 3497 if (folio_test_uptodate(folio))
3804 goto out; 3498 goto out;
3805 3499
3806 if (!folio_trylock(folio)) { 3500 if (!folio_trylock(folio)) {
3807 folio_put_wait_locked(folio, 3501 folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE);
3808 goto repeat; 3502 goto repeat;
3809 } 3503 }
3810 3504
3811 /* Folio was truncated from mapping * 3505 /* Folio was truncated from mapping */
3812 if (!folio->mapping) { 3506 if (!folio->mapping) {
3813 folio_unlock(folio); 3507 folio_unlock(folio);
3814 folio_put(folio); 3508 folio_put(folio);
3815 goto repeat; 3509 goto repeat;
3816 } 3510 }
3817 3511
3818 /* Someone else locked and filled the 3512 /* Someone else locked and filled the page in a very small window */
3819 if (folio_test_uptodate(folio)) { 3513 if (folio_test_uptodate(folio)) {
3820 folio_unlock(folio); 3514 folio_unlock(folio);
3821 goto out; 3515 goto out;
3822 } 3516 }
3823 3517
3824 filler: 3518 filler:
3825 err = filemap_read_folio(file, filler 3519 err = filemap_read_folio(file, filler, folio);
3826 if (err) { 3520 if (err) {
3827 folio_put(folio); 3521 folio_put(folio);
3828 if (err == AOP_TRUNCATED_PAGE 3522 if (err == AOP_TRUNCATED_PAGE)
3829 goto repeat; 3523 goto repeat;
3830 return ERR_PTR(err); 3524 return ERR_PTR(err);
3831 } 3525 }
3832 3526
3833 out: 3527 out:
3834 folio_mark_accessed(folio); 3528 folio_mark_accessed(folio);
3835 return folio; 3529 return folio;
3836 } 3530 }
3837 3531
3838 /** 3532 /**
3839 * read_cache_folio - Read into page cache, f 3533 * read_cache_folio - Read into page cache, fill it if needed.
3840 * @mapping: The address_space to read from. 3534 * @mapping: The address_space to read from.
3841 * @index: The index to read. 3535 * @index: The index to read.
3842 * @filler: Function to perform the read, or 3536 * @filler: Function to perform the read, or NULL to use aops->read_folio().
3843 * @file: Passed to filler function, may be N 3537 * @file: Passed to filler function, may be NULL if not required.
3844 * 3538 *
3845 * Read one page into the page cache. If it 3539 * Read one page into the page cache. If it succeeds, the folio returned
3846 * will contain @index, but it may not be the 3540 * will contain @index, but it may not be the first page of the folio.
3847 * 3541 *
3848 * If the filler function returns an error, i 3542 * If the filler function returns an error, it will be returned to the
3849 * caller. 3543 * caller.
3850 * 3544 *
3851 * Context: May sleep. Expects mapping->inva 3545 * Context: May sleep. Expects mapping->invalidate_lock to be held.
3852 * Return: An uptodate folio on success, ERR_ 3546 * Return: An uptodate folio on success, ERR_PTR() on failure.
3853 */ 3547 */
3854 struct folio *read_cache_folio(struct address 3548 struct folio *read_cache_folio(struct address_space *mapping, pgoff_t index,
3855 filler_t filler, struct file 3549 filler_t filler, struct file *file)
3856 { 3550 {
3857 return do_read_cache_folio(mapping, i 3551 return do_read_cache_folio(mapping, index, filler, file,
3858 mapping_gfp_mask(mapp 3552 mapping_gfp_mask(mapping));
3859 } 3553 }
3860 EXPORT_SYMBOL(read_cache_folio); 3554 EXPORT_SYMBOL(read_cache_folio);
3861 3555
3862 /** <<
3863 * mapping_read_folio_gfp - Read into page ca <<
3864 * @mapping: The address_space for the fol <<
3865 * @index: The index that the allocated <<
3866 * @gfp: The page allocator flags to u <<
3867 * <<
3868 * This is the same as "read_cache_folio(mapp <<
3869 * any new memory allocations done using the <<
3870 * <<
3871 * The most likely error from this function i <<
3872 * possible and so is EINTR. If ->read_folio <<
3873 * that will be returned to the caller. <<
3874 * <<
3875 * The function expects mapping->invalidate_l <<
3876 * <<
3877 * Return: Uptodate folio on success, ERR_PTR <<
3878 */ <<
3879 struct folio *mapping_read_folio_gfp(struct a <<
3880 pgoff_t index, gfp_t gfp) <<
3881 { <<
3882 return do_read_cache_folio(mapping, i <<
3883 } <<
3884 EXPORT_SYMBOL(mapping_read_folio_gfp); <<
3885 <<
3886 static struct page *do_read_cache_page(struct 3556 static struct page *do_read_cache_page(struct address_space *mapping,
3887 pgoff_t index, filler_t *fill 3557 pgoff_t index, filler_t *filler, struct file *file, gfp_t gfp)
3888 { 3558 {
3889 struct folio *folio; 3559 struct folio *folio;
3890 3560
3891 folio = do_read_cache_folio(mapping, 3561 folio = do_read_cache_folio(mapping, index, filler, file, gfp);
3892 if (IS_ERR(folio)) 3562 if (IS_ERR(folio))
3893 return &folio->page; 3563 return &folio->page;
3894 return folio_file_page(folio, index); 3564 return folio_file_page(folio, index);
3895 } 3565 }
3896 3566
3897 struct page *read_cache_page(struct address_s 3567 struct page *read_cache_page(struct address_space *mapping,
3898 pgoff_t index, filler 3568 pgoff_t index, filler_t *filler, struct file *file)
3899 { 3569 {
3900 return do_read_cache_page(mapping, in 3570 return do_read_cache_page(mapping, index, filler, file,
3901 mapping_gfp_mask(mapp 3571 mapping_gfp_mask(mapping));
3902 } 3572 }
3903 EXPORT_SYMBOL(read_cache_page); 3573 EXPORT_SYMBOL(read_cache_page);
3904 3574
3905 /** 3575 /**
3906 * read_cache_page_gfp - read into page cache 3576 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3907 * @mapping: the page's address_space 3577 * @mapping: the page's address_space
3908 * @index: the page index 3578 * @index: the page index
3909 * @gfp: the page allocator flags to u 3579 * @gfp: the page allocator flags to use if allocating
3910 * 3580 *
3911 * This is the same as "read_mapping_page(map 3581 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3912 * any new page allocations done using the sp 3582 * any new page allocations done using the specified allocation flags.
3913 * 3583 *
3914 * If the page does not get brought uptodate, 3584 * If the page does not get brought uptodate, return -EIO.
3915 * 3585 *
3916 * The function expects mapping->invalidate_l 3586 * The function expects mapping->invalidate_lock to be already held.
3917 * 3587 *
3918 * Return: up to date page on success, ERR_PT 3588 * Return: up to date page on success, ERR_PTR() on failure.
3919 */ 3589 */
3920 struct page *read_cache_page_gfp(struct addre 3590 struct page *read_cache_page_gfp(struct address_space *mapping,
3921 pgoff_t index 3591 pgoff_t index,
3922 gfp_t gfp) 3592 gfp_t gfp)
3923 { 3593 {
3924 return do_read_cache_page(mapping, in 3594 return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3925 } 3595 }
3926 EXPORT_SYMBOL(read_cache_page_gfp); 3596 EXPORT_SYMBOL(read_cache_page_gfp);
3927 3597
3928 /* 3598 /*
3929 * Warn about a page cache invalidation failu 3599 * Warn about a page cache invalidation failure during a direct I/O write.
3930 */ 3600 */
3931 static void dio_warn_stale_pagecache(struct f !! 3601 void dio_warn_stale_pagecache(struct file *filp)
3932 { 3602 {
3933 static DEFINE_RATELIMIT_STATE(_rs, 86 3603 static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3934 char pathname[128]; 3604 char pathname[128];
3935 char *path; 3605 char *path;
3936 3606
3937 errseq_set(&filp->f_mapping->wb_err, 3607 errseq_set(&filp->f_mapping->wb_err, -EIO);
3938 if (__ratelimit(&_rs)) { 3608 if (__ratelimit(&_rs)) {
3939 path = file_path(filp, pathna 3609 path = file_path(filp, pathname, sizeof(pathname));
3940 if (IS_ERR(path)) 3610 if (IS_ERR(path))
3941 path = "(unknown)"; 3611 path = "(unknown)";
3942 pr_crit("Page cache invalidat 3612 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3943 pr_crit("File: %s PID: %d Com 3613 pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3944 current->comm); 3614 current->comm);
3945 } 3615 }
3946 } 3616 }
3947 3617
3948 void kiocb_invalidate_post_direct_write(struc <<
3949 { <<
3950 struct address_space *mapping = iocb- <<
3951 <<
3952 if (mapping->nrpages && <<
3953 invalidate_inode_pages2_range(map <<
3954 iocb->ki_pos >> PAGE_ <<
3955 (iocb->ki_pos + count <<
3956 dio_warn_stale_pagecache(iocb <<
3957 } <<
3958 <<
3959 ssize_t 3618 ssize_t
3960 generic_file_direct_write(struct kiocb *iocb, 3619 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3961 { 3620 {
3962 struct address_space *mapping = iocb- !! 3621 struct file *file = iocb->ki_filp;
3963 size_t write_len = iov_iter_count(fro !! 3622 struct address_space *mapping = file->f_mapping;
3964 ssize_t written; !! 3623 struct inode *inode = mapping->host;
>> 3624 loff_t pos = iocb->ki_pos;
>> 3625 ssize_t written;
>> 3626 size_t write_len;
>> 3627 pgoff_t end;
>> 3628
>> 3629 write_len = iov_iter_count(from);
>> 3630 end = (pos + write_len - 1) >> PAGE_SHIFT;
>> 3631
>> 3632 if (iocb->ki_flags & IOCB_NOWAIT) {
>> 3633 /* If there are pages to writeback, return */
>> 3634 if (filemap_range_has_page(file->f_mapping, pos,
>> 3635 pos + write_len - 1))
>> 3636 return -EAGAIN;
>> 3637 } else {
>> 3638 written = filemap_write_and_wait_range(mapping, pos,
>> 3639 pos + write_len - 1);
>> 3640 if (written)
>> 3641 goto out;
>> 3642 }
3965 3643
3966 /* 3644 /*
>> 3645 * After a write we want buffered reads to be sure to go to disk to get
>> 3646 * the new data. We invalidate clean cached page from the region we're
>> 3647 * about to write. We do this *before* the write so that we can return
>> 3648 * without clobbering -EIOCBQUEUED from ->direct_IO().
>> 3649 */
>> 3650 written = invalidate_inode_pages2_range(mapping,
>> 3651 pos >> PAGE_SHIFT, end);
>> 3652 /*
3967 * If a page can not be invalidated, 3653 * If a page can not be invalidated, return 0 to fall back
3968 * to buffered write. 3654 * to buffered write.
3969 */ 3655 */
3970 written = kiocb_invalidate_pages(iocb <<
3971 if (written) { 3656 if (written) {
3972 if (written == -EBUSY) 3657 if (written == -EBUSY)
3973 return 0; 3658 return 0;
3974 return written; !! 3659 goto out;
3975 } 3660 }
3976 3661
3977 written = mapping->a_ops->direct_IO(i 3662 written = mapping->a_ops->direct_IO(iocb, from);
3978 3663
3979 /* 3664 /*
3980 * Finally, try again to invalidate c 3665 * Finally, try again to invalidate clean pages which might have been
3981 * cached by non-direct readahead, or 3666 * cached by non-direct readahead, or faulted in by get_user_pages()
3982 * if the source of the write was an 3667 * if the source of the write was an mmap'ed region of the file
3983 * we're writing. Either one is a pr 3668 * we're writing. Either one is a pretty crazy thing to do,
3984 * so we don't support it 100%. If t 3669 * so we don't support it 100%. If this invalidation
3985 * fails, tough, the write still work 3670 * fails, tough, the write still worked...
3986 * 3671 *
3987 * Most of the time we do not need th 3672 * Most of the time we do not need this since dio_complete() will do
3988 * the invalidation for us. However t 3673 * the invalidation for us. However there are some file systems that
3989 * do not end up with dio_complete() 3674 * do not end up with dio_complete() being called, so let's not break
3990 * them by removing it completely. 3675 * them by removing it completely.
3991 * 3676 *
3992 * Noticeable example is a blkdev_dir 3677 * Noticeable example is a blkdev_direct_IO().
3993 * 3678 *
3994 * Skip invalidation for async writes 3679 * Skip invalidation for async writes or if mapping has no pages.
3995 */ 3680 */
3996 if (written > 0) { !! 3681 if (written > 0 && mapping->nrpages &&
3997 struct inode *inode = mapping !! 3682 invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3998 loff_t pos = iocb->ki_pos; !! 3683 dio_warn_stale_pagecache(file);
3999 3684
4000 kiocb_invalidate_post_direct_ !! 3685 if (written > 0) {
4001 pos += written; 3686 pos += written;
4002 write_len -= written; 3687 write_len -= written;
4003 if (pos > i_size_read(inode) 3688 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
4004 i_size_write(inode, p 3689 i_size_write(inode, pos);
4005 mark_inode_dirty(inod 3690 mark_inode_dirty(inode);
4006 } 3691 }
4007 iocb->ki_pos = pos; 3692 iocb->ki_pos = pos;
4008 } 3693 }
4009 if (written != -EIOCBQUEUED) 3694 if (written != -EIOCBQUEUED)
4010 iov_iter_revert(from, write_l 3695 iov_iter_revert(from, write_len - iov_iter_count(from));
>> 3696 out:
4011 return written; 3697 return written;
4012 } 3698 }
4013 EXPORT_SYMBOL(generic_file_direct_write); 3699 EXPORT_SYMBOL(generic_file_direct_write);
4014 3700
4015 ssize_t generic_perform_write(struct kiocb *i 3701 ssize_t generic_perform_write(struct kiocb *iocb, struct iov_iter *i)
4016 { 3702 {
4017 struct file *file = iocb->ki_filp; 3703 struct file *file = iocb->ki_filp;
4018 loff_t pos = iocb->ki_pos; 3704 loff_t pos = iocb->ki_pos;
4019 struct address_space *mapping = file- 3705 struct address_space *mapping = file->f_mapping;
4020 const struct address_space_operations 3706 const struct address_space_operations *a_ops = mapping->a_ops;
4021 size_t chunk = mapping_max_folio_size <<
4022 long status = 0; 3707 long status = 0;
4023 ssize_t written = 0; 3708 ssize_t written = 0;
4024 3709
4025 do { 3710 do {
4026 struct folio *folio; !! 3711 struct page *page;
4027 size_t offset; /* Of !! 3712 unsigned long offset; /* Offset into pagecache page */
4028 size_t bytes; /* By !! 3713 unsigned long bytes; /* Bytes to write to page */
4029 size_t copied; /* By 3714 size_t copied; /* Bytes copied from user */
4030 void *fsdata = NULL; 3715 void *fsdata = NULL;
4031 3716
4032 bytes = iov_iter_count(i); !! 3717 offset = (pos & (PAGE_SIZE - 1));
4033 retry: !! 3718 bytes = min_t(unsigned long, PAGE_SIZE - offset,
4034 offset = pos & (chunk - 1); !! 3719 iov_iter_count(i));
4035 bytes = min(chunk - offset, b <<
4036 balance_dirty_pages_ratelimit <<
4037 3720
>> 3721 again:
4038 /* 3722 /*
4039 * Bring in the user page tha 3723 * Bring in the user page that we will copy from _first_.
4040 * Otherwise there's a nasty 3724 * Otherwise there's a nasty deadlock on copying from the
4041 * same page as we're writing 3725 * same page as we're writing to, without it being marked
4042 * up-to-date. 3726 * up-to-date.
4043 */ 3727 */
4044 if (unlikely(fault_in_iov_ite 3728 if (unlikely(fault_in_iov_iter_readable(i, bytes) == bytes)) {
4045 status = -EFAULT; 3729 status = -EFAULT;
4046 break; 3730 break;
4047 } 3731 }
4048 3732
4049 if (fatal_signal_pending(curr 3733 if (fatal_signal_pending(current)) {
4050 status = -EINTR; 3734 status = -EINTR;
4051 break; 3735 break;
4052 } 3736 }
4053 3737
4054 status = a_ops->write_begin(f 3738 status = a_ops->write_begin(file, mapping, pos, bytes,
4055 !! 3739 &page, &fsdata);
4056 if (unlikely(status < 0)) 3740 if (unlikely(status < 0))
4057 break; 3741 break;
4058 3742
4059 offset = offset_in_folio(foli <<
4060 if (bytes > folio_size(folio) <<
4061 bytes = folio_size(fo <<
4062 <<
4063 if (mapping_writably_mapped(m 3743 if (mapping_writably_mapped(mapping))
4064 flush_dcache_folio(fo !! 3744 flush_dcache_page(page);
4065 3745
4066 copied = copy_folio_from_iter !! 3746 copied = copy_page_from_iter_atomic(page, offset, bytes, i);
4067 flush_dcache_folio(folio); !! 3747 flush_dcache_page(page);
4068 3748
4069 status = a_ops->write_end(fil 3749 status = a_ops->write_end(file, mapping, pos, bytes, copied,
4070 !! 3750 page, fsdata);
4071 if (unlikely(status != copied 3751 if (unlikely(status != copied)) {
4072 iov_iter_revert(i, co 3752 iov_iter_revert(i, copied - max(status, 0L));
4073 if (unlikely(status < 3753 if (unlikely(status < 0))
4074 break; 3754 break;
4075 } 3755 }
4076 cond_resched(); 3756 cond_resched();
4077 3757
4078 if (unlikely(status == 0)) { 3758 if (unlikely(status == 0)) {
4079 /* 3759 /*
4080 * A short copy made 3760 * A short copy made ->write_end() reject the
4081 * thing entirely. M 3761 * thing entirely. Might be memory poisoning
4082 * halfway through, m 3762 * halfway through, might be a race with munmap,
4083 * might be severe me 3763 * might be severe memory pressure.
4084 */ 3764 */
4085 if (chunk > PAGE_SIZE !! 3765 if (copied)
4086 chunk /= 2; <<
4087 if (copied) { <<
4088 bytes = copie 3766 bytes = copied;
4089 goto retry; !! 3767 goto again;
4090 } <<
4091 } else { <<
4092 pos += status; <<
4093 written += status; <<
4094 } 3768 }
>> 3769 pos += status;
>> 3770 written += status;
>> 3771
>> 3772 balance_dirty_pages_ratelimited(mapping);
4095 } while (iov_iter_count(i)); 3773 } while (iov_iter_count(i));
4096 3774
4097 if (!written) !! 3775 return written ? written : status;
4098 return status; <<
4099 iocb->ki_pos += written; <<
4100 return written; <<
4101 } 3776 }
4102 EXPORT_SYMBOL(generic_perform_write); 3777 EXPORT_SYMBOL(generic_perform_write);
4103 3778
4104 /** 3779 /**
4105 * __generic_file_write_iter - write data to 3780 * __generic_file_write_iter - write data to a file
4106 * @iocb: IO state structure (file, off 3781 * @iocb: IO state structure (file, offset, etc.)
4107 * @from: iov_iter with data to write 3782 * @from: iov_iter with data to write
4108 * 3783 *
4109 * This function does all the work needed for 3784 * This function does all the work needed for actually writing data to a
4110 * file. It does all basic checks, removes SU 3785 * file. It does all basic checks, removes SUID from the file, updates
4111 * modification times and calls proper subrou 3786 * modification times and calls proper subroutines depending on whether we
4112 * do direct IO or a standard buffered write. 3787 * do direct IO or a standard buffered write.
4113 * 3788 *
4114 * It expects i_rwsem to be grabbed unless we 3789 * It expects i_rwsem to be grabbed unless we work on a block device or similar
4115 * object which does not need locking at all. 3790 * object which does not need locking at all.
4116 * 3791 *
4117 * This function does *not* take care of sync 3792 * This function does *not* take care of syncing data in case of O_SYNC write.
4118 * A caller has to handle it. This is mainly 3793 * A caller has to handle it. This is mainly due to the fact that we want to
4119 * avoid syncing under i_rwsem. 3794 * avoid syncing under i_rwsem.
4120 * 3795 *
4121 * Return: 3796 * Return:
4122 * * number of bytes written, even for trunca 3797 * * number of bytes written, even for truncated writes
4123 * * negative error code if no data has been 3798 * * negative error code if no data has been written at all
4124 */ 3799 */
4125 ssize_t __generic_file_write_iter(struct kioc 3800 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4126 { 3801 {
4127 struct file *file = iocb->ki_filp; 3802 struct file *file = iocb->ki_filp;
4128 struct address_space *mapping = file- 3803 struct address_space *mapping = file->f_mapping;
4129 struct inode *inode = mapping->host; !! 3804 struct inode *inode = mapping->host;
4130 ssize_t ret; !! 3805 ssize_t written = 0;
>> 3806 ssize_t err;
>> 3807 ssize_t status;
>> 3808
>> 3809 /* We can write back this queue in page reclaim */
>> 3810 current->backing_dev_info = inode_to_bdi(inode);
>> 3811 err = file_remove_privs(file);
>> 3812 if (err)
>> 3813 goto out;
4131 3814
4132 ret = file_remove_privs(file); !! 3815 err = file_update_time(file);
4133 if (ret) !! 3816 if (err)
4134 return ret; !! 3817 goto out;
4135 <<
4136 ret = file_update_time(file); <<
4137 if (ret) <<
4138 return ret; <<
4139 3818
4140 if (iocb->ki_flags & IOCB_DIRECT) { 3819 if (iocb->ki_flags & IOCB_DIRECT) {
4141 ret = generic_file_direct_wri !! 3820 loff_t pos, endbyte;
>> 3821
>> 3822 written = generic_file_direct_write(iocb, from);
4142 /* 3823 /*
4143 * If the write stopped short 3824 * If the write stopped short of completing, fall back to
4144 * buffered writes. Some fil 3825 * buffered writes. Some filesystems do this for writes to
4145 * holes, for example. For D 3826 * holes, for example. For DAX files, a buffered write will
4146 * not succeed (even if it di 3827 * not succeed (even if it did, DAX does not handle dirty
4147 * page-cache pages correctly 3828 * page-cache pages correctly).
4148 */ 3829 */
4149 if (ret < 0 || !iov_iter_coun !! 3830 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
4150 return ret; !! 3831 goto out;
4151 return direct_write_fallback( <<
4152 generic_perfo <<
4153 } <<
4154 3832
4155 return generic_perform_write(iocb, fr !! 3833 pos = iocb->ki_pos;
>> 3834 status = generic_perform_write(iocb, from);
>> 3835 /*
>> 3836 * If generic_perform_write() returned a synchronous error
>> 3837 * then we want to return the number of bytes which were
>> 3838 * direct-written, or the error code if that was zero. Note
>> 3839 * that this differs from normal direct-io semantics, which
>> 3840 * will return -EFOO even if some bytes were written.
>> 3841 */
>> 3842 if (unlikely(status < 0)) {
>> 3843 err = status;
>> 3844 goto out;
>> 3845 }
>> 3846 /*
>> 3847 * We need to ensure that the page cache pages are written to
>> 3848 * disk and invalidated to preserve the expected O_DIRECT
>> 3849 * semantics.
>> 3850 */
>> 3851 endbyte = pos + status - 1;
>> 3852 err = filemap_write_and_wait_range(mapping, pos, endbyte);
>> 3853 if (err == 0) {
>> 3854 iocb->ki_pos = endbyte + 1;
>> 3855 written += status;
>> 3856 invalidate_mapping_pages(mapping,
>> 3857 pos >> PAGE_SHIFT,
>> 3858 endbyte >> PAGE_SHIFT);
>> 3859 } else {
>> 3860 /*
>> 3861 * We don't know how much we wrote, so just return
>> 3862 * the number of bytes which were direct-written
>> 3863 */
>> 3864 }
>> 3865 } else {
>> 3866 written = generic_perform_write(iocb, from);
>> 3867 if (likely(written > 0))
>> 3868 iocb->ki_pos += written;
>> 3869 }
>> 3870 out:
>> 3871 current->backing_dev_info = NULL;
>> 3872 return written ? written : err;
4156 } 3873 }
4157 EXPORT_SYMBOL(__generic_file_write_iter); 3874 EXPORT_SYMBOL(__generic_file_write_iter);
4158 3875
4159 /** 3876 /**
4160 * generic_file_write_iter - write data to a 3877 * generic_file_write_iter - write data to a file
4161 * @iocb: IO state structure 3878 * @iocb: IO state structure
4162 * @from: iov_iter with data to write 3879 * @from: iov_iter with data to write
4163 * 3880 *
4164 * This is a wrapper around __generic_file_wr 3881 * This is a wrapper around __generic_file_write_iter() to be used by most
4165 * filesystems. It takes care of syncing the 3882 * filesystems. It takes care of syncing the file in case of O_SYNC file
4166 * and acquires i_rwsem as needed. 3883 * and acquires i_rwsem as needed.
4167 * Return: 3884 * Return:
4168 * * negative error code if no data has been 3885 * * negative error code if no data has been written at all of
4169 * vfs_fsync_range() failed for a synchrono 3886 * vfs_fsync_range() failed for a synchronous write
4170 * * number of bytes written, even for trunca 3887 * * number of bytes written, even for truncated writes
4171 */ 3888 */
4172 ssize_t generic_file_write_iter(struct kiocb 3889 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4173 { 3890 {
4174 struct file *file = iocb->ki_filp; 3891 struct file *file = iocb->ki_filp;
4175 struct inode *inode = file->f_mapping 3892 struct inode *inode = file->f_mapping->host;
4176 ssize_t ret; 3893 ssize_t ret;
4177 3894
4178 inode_lock(inode); 3895 inode_lock(inode);
4179 ret = generic_write_checks(iocb, from 3896 ret = generic_write_checks(iocb, from);
4180 if (ret > 0) 3897 if (ret > 0)
4181 ret = __generic_file_write_it 3898 ret = __generic_file_write_iter(iocb, from);
4182 inode_unlock(inode); 3899 inode_unlock(inode);
4183 3900
4184 if (ret > 0) 3901 if (ret > 0)
4185 ret = generic_write_sync(iocb 3902 ret = generic_write_sync(iocb, ret);
4186 return ret; 3903 return ret;
4187 } 3904 }
4188 EXPORT_SYMBOL(generic_file_write_iter); 3905 EXPORT_SYMBOL(generic_file_write_iter);
4189 3906
4190 /** 3907 /**
4191 * filemap_release_folio() - Release fs-speci 3908 * filemap_release_folio() - Release fs-specific metadata on a folio.
4192 * @folio: The folio which the kernel is tryi 3909 * @folio: The folio which the kernel is trying to free.
4193 * @gfp: Memory allocation flags (and I/O mod 3910 * @gfp: Memory allocation flags (and I/O mode).
4194 * 3911 *
4195 * The address_space is trying to release any 3912 * The address_space is trying to release any data attached to a folio
4196 * (presumably at folio->private). 3913 * (presumably at folio->private).
4197 * 3914 *
4198 * This will also be called if the private_2 3915 * This will also be called if the private_2 flag is set on a page,
4199 * indicating that the folio has other metada 3916 * indicating that the folio has other metadata associated with it.
4200 * 3917 *
4201 * The @gfp argument specifies whether I/O ma 3918 * The @gfp argument specifies whether I/O may be performed to release
4202 * this page (__GFP_IO), and whether the call 3919 * this page (__GFP_IO), and whether the call may block
4203 * (__GFP_RECLAIM & __GFP_FS). 3920 * (__GFP_RECLAIM & __GFP_FS).
4204 * 3921 *
4205 * Return: %true if the release was successfu 3922 * Return: %true if the release was successful, otherwise %false.
4206 */ 3923 */
4207 bool filemap_release_folio(struct folio *foli 3924 bool filemap_release_folio(struct folio *folio, gfp_t gfp)
4208 { 3925 {
4209 struct address_space * const mapping 3926 struct address_space * const mapping = folio->mapping;
4210 3927
4211 BUG_ON(!folio_test_locked(folio)); 3928 BUG_ON(!folio_test_locked(folio));
4212 if (!folio_needs_release(folio)) <<
4213 return true; <<
4214 if (folio_test_writeback(folio)) 3929 if (folio_test_writeback(folio))
4215 return false; 3930 return false;
4216 3931
4217 if (mapping && mapping->a_ops->releas 3932 if (mapping && mapping->a_ops->release_folio)
4218 return mapping->a_ops->releas 3933 return mapping->a_ops->release_folio(folio, gfp);
4219 return try_to_free_buffers(folio); 3934 return try_to_free_buffers(folio);
4220 } 3935 }
4221 EXPORT_SYMBOL(filemap_release_folio); 3936 EXPORT_SYMBOL(filemap_release_folio);
4222 <<
4223 /** <<
4224 * filemap_invalidate_inode - Invalidate/forc <<
4225 * @inode: The inode to flush <<
4226 * @flush: Set to write back rather than simp <<
4227 * @start: First byte to in range. <<
4228 * @end: Last byte in range (inclusive), or L <<
4229 * onwards. <<
4230 * <<
4231 * Invalidate all the folios on an inode that <<
4232 * range, possibly writing them back first. <<
4233 * undertaken, the invalidate lock is held to <<
4234 * installed. <<
4235 */ <<
4236 int filemap_invalidate_inode(struct inode *in <<
4237 loff_t start, lo <<
4238 { <<
4239 struct address_space *mapping = inode <<
4240 pgoff_t first = start >> PAGE_SHIFT; <<
4241 pgoff_t last = end >> PAGE_SHIFT; <<
4242 pgoff_t nr = end == LLONG_MAX ? ULONG <<
4243 <<
4244 if (!mapping || !mapping->nrpages || <<
4245 goto out; <<
4246 <<
4247 /* Prevent new folios from being adde <<
4248 filemap_invalidate_lock(mapping); <<
4249 <<
4250 if (!mapping->nrpages) <<
4251 goto unlock; <<
4252 <<
4253 unmap_mapping_pages(mapping, first, n <<
4254 <<
4255 /* Write back the data if we're asked <<
4256 if (flush) { <<
4257 struct writeback_control wbc <<
4258 .sync_mode = WB_ <<
4259 .nr_to_write = LON <<
4260 .range_start = sta <<
4261 .range_end = end <<
4262 }; <<
4263 <<
4264 filemap_fdatawrite_wbc(mappin <<
4265 } <<
4266 <<
4267 /* Wait for writeback to complete on <<
4268 invalidate_inode_pages2_range(mapping <<
4269 <<
4270 unlock: <<
4271 filemap_invalidate_unlock(mapping); <<
4272 out: <<
4273 return filemap_check_errors(mapping); <<
4274 } <<
4275 EXPORT_SYMBOL_GPL(filemap_invalidate_inode); <<
4276 <<
4277 #ifdef CONFIG_CACHESTAT_SYSCALL <<
4278 /** <<
4279 * filemap_cachestat() - compute the page cac <<
4280 * @mapping: The mapping to compute the st <<
4281 * @first_index: The starting page cac <<
4282 * @last_index: The final page index (inclusi <<
4283 * @cs: the cachestat struct to write the res <<
4284 * <<
4285 * This will query the page cache statistics <<
4286 * page range of [first_index, last_index] (i <<
4287 * queried include: number of dirty pages, nu <<
4288 * writeback, and the number of (recently) ev <<
4289 */ <<
4290 static void filemap_cachestat(struct address_ <<
4291 pgoff_t first_index, pgoff_t <<
4292 { <<
4293 XA_STATE(xas, &mapping->i_pages, firs <<
4294 struct folio *folio; <<
4295 <<
4296 /* Flush stats (and potentially sleep <<
4297 mem_cgroup_flush_stats_ratelimited(NU <<
4298 <<
4299 rcu_read_lock(); <<
4300 xas_for_each(&xas, folio, last_index) <<
4301 int order; <<
4302 unsigned long nr_pages; <<
4303 pgoff_t folio_first_index, fo <<
4304 <<
4305 /* <<
4306 * Don't deref the folio. It <<
4307 * get freed (and reused) und <<
4308 * <<
4309 * We *could* pin it, but tha <<
4310 * what should be a fast and <<
4311 * <<
4312 * Instead, derive all inform <<
4313 * the rcu-protected xarray. <<
4314 */ <<
4315 <<
4316 if (xas_retry(&xas, folio)) <<
4317 continue; <<
4318 <<
4319 order = xas_get_order(&xas); <<
4320 nr_pages = 1 << order; <<
4321 folio_first_index = round_dow <<
4322 folio_last_index = folio_firs <<
4323 <<
4324 /* Folios might straddle the <<
4325 if (folio_first_index < first <<
4326 nr_pages -= first_ind <<
4327 <<
4328 if (folio_last_index > last_i <<
4329 nr_pages -= folio_las <<
4330 <<
4331 if (xa_is_value(folio)) { <<
4332 /* page is evicted */ <<
4333 void *shadow = (void <<
4334 bool workingset; /* n <<
4335 <<
4336 cs->nr_evicted += nr_ <<
4337 <<
4338 #ifdef CONFIG_SWAP /* implies CONFIG_MMU */ <<
4339 if (shmem_mapping(map <<
4340 /* shmem file <<
4341 swp_entry_t s <<
4342 <<
4343 /* swapin err <<
4344 if (non_swap_ <<
4345 goto <<
4346 <<
4347 /* <<
4348 * Getting a <<
4349 * inode mean <<
4350 * shmem_unus <<
4351 * ensures sw <<
4352 * freeing th <<
4353 * we can rac <<
4354 * invalidati <<
4355 * a shadow i <<
4356 */ <<
4357 shadow = get_ <<
4358 if (!shadow) <<
4359 goto <<
4360 } <<
4361 #endif <<
4362 if (workingset_test_r <<
4363 cs->nr_recent <<
4364 <<
4365 goto resched; <<
4366 } <<
4367 <<
4368 /* page is in cache */ <<
4369 cs->nr_cache += nr_pages; <<
4370 <<
4371 if (xas_get_mark(&xas, PAGECA <<
4372 cs->nr_dirty += nr_pa <<
4373 <<
4374 if (xas_get_mark(&xas, PAGECA <<
4375 cs->nr_writeback += n <<
4376 <<
4377 resched: <<
4378 if (need_resched()) { <<
4379 xas_pause(&xas); <<
4380 cond_resched_rcu(); <<
4381 } <<
4382 } <<
4383 rcu_read_unlock(); <<
4384 } <<
4385 <<
4386 /* <<
4387 * The cachestat(2) system call. <<
4388 * <<
4389 * cachestat() returns the page cache statist <<
4390 * bytes range specified by `off` and `len`: <<
4391 * number of dirty pages, number of pages mar <<
4392 * number of evicted pages, and number of rec <<
4393 * <<
4394 * An evicted page is a page that is previous <<
4395 * but has been evicted since. A page is rece <<
4396 * eviction was recent enough that its reentr <<
4397 * indicate that it is actively being used by <<
4398 * there is memory pressure on the system. <<
4399 * <<
4400 * `off` and `len` must be non-negative integ <<
4401 * the queried range is [`off`, `off` + `len` <<
4402 * we will query in the range from `off` to t <<
4403 * <<
4404 * The `flags` argument is unused for now, bu <<
4405 * extensibility. User should pass 0 (i.e no <<
4406 * <<
4407 * Currently, hugetlbfs is not supported. <<
4408 * <<
4409 * Because the status of a page can change af <<
4410 * but before it returns to the application, <<
4411 * contain stale information. <<
4412 * <<
4413 * return values: <<
4414 * zero - success <<
4415 * -EFAULT - cstat or cstat_range points <<
4416 * -EINVAL - invalid flags <<
4417 * -EBADF - invalid file descriptor <<
4418 * -EOPNOTSUPP - file descriptor is of a hug <<
4419 */ <<
4420 SYSCALL_DEFINE4(cachestat, unsigned int, fd, <<
4421 struct cachestat_range __user <<
4422 struct cachestat __user *, cs <<
4423 { <<
4424 struct fd f = fdget(fd); <<
4425 struct address_space *mapping; <<
4426 struct cachestat_range csr; <<
4427 struct cachestat cs; <<
4428 pgoff_t first_index, last_index; <<
4429 <<
4430 if (!fd_file(f)) <<
4431 return -EBADF; <<
4432 <<
4433 if (copy_from_user(&csr, cstat_range, <<
4434 sizeof(struct cachest <<
4435 fdput(f); <<
4436 return -EFAULT; <<
4437 } <<
4438 <<
4439 /* hugetlbfs is not supported */ <<
4440 if (is_file_hugepages(fd_file(f))) { <<
4441 fdput(f); <<
4442 return -EOPNOTSUPP; <<
4443 } <<
4444 <<
4445 if (flags != 0) { <<
4446 fdput(f); <<
4447 return -EINVAL; <<
4448 } <<
4449 <<
4450 first_index = csr.off >> PAGE_SHIFT; <<
4451 last_index = <<
4452 csr.len == 0 ? ULONG_MAX : (c <<
4453 memset(&cs, 0, sizeof(struct cachesta <<
4454 mapping = fd_file(f)->f_mapping; <<
4455 filemap_cachestat(mapping, first_inde <<
4456 fdput(f); <<
4457 <<
4458 if (copy_to_user(cstat, &cs, sizeof(s <<
4459 return -EFAULT; <<
4460 <<
4461 return 0; <<
4462 } <<
4463 #endif /* CONFIG_CACHESTAT_SYSCALL */ <<
4464 3937
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.