1 // SPDX-License-Identifier: GPL-2.0-only <<
2 /* 1 /*
3 * linux/mm/filemap.c 2 * linux/mm/filemap.c
4 * 3 *
5 * Copyright (C) 1994-1999 Linus Torvalds 4 * Copyright (C) 1994-1999 Linus Torvalds
6 */ 5 */
7 6
8 /* 7 /*
9 * This file handles the generic file mmap sem 8 * This file handles the generic file mmap semantics used by
10 * most "normal" filesystems (but you don't /h 9 * most "normal" filesystems (but you don't /have/ to use this:
11 * the NFS filesystem used to do this differen 10 * the NFS filesystem used to do this differently, for example)
12 */ 11 */
13 #include <linux/export.h> 12 #include <linux/export.h>
14 #include <linux/compiler.h> 13 #include <linux/compiler.h>
15 #include <linux/dax.h> 14 #include <linux/dax.h>
16 #include <linux/fs.h> 15 #include <linux/fs.h>
17 #include <linux/sched/signal.h> 16 #include <linux/sched/signal.h>
18 #include <linux/uaccess.h> 17 #include <linux/uaccess.h>
19 #include <linux/capability.h> 18 #include <linux/capability.h>
20 #include <linux/kernel_stat.h> 19 #include <linux/kernel_stat.h>
21 #include <linux/gfp.h> 20 #include <linux/gfp.h>
22 #include <linux/mm.h> 21 #include <linux/mm.h>
23 #include <linux/swap.h> 22 #include <linux/swap.h>
24 #include <linux/swapops.h> <<
25 #include <linux/syscalls.h> <<
26 #include <linux/mman.h> 23 #include <linux/mman.h>
27 #include <linux/pagemap.h> 24 #include <linux/pagemap.h>
28 #include <linux/file.h> 25 #include <linux/file.h>
29 #include <linux/uio.h> 26 #include <linux/uio.h>
30 #include <linux/error-injection.h> <<
31 #include <linux/hash.h> 27 #include <linux/hash.h>
32 #include <linux/writeback.h> 28 #include <linux/writeback.h>
33 #include <linux/backing-dev.h> 29 #include <linux/backing-dev.h>
34 #include <linux/pagevec.h> 30 #include <linux/pagevec.h>
>> 31 #include <linux/blkdev.h>
35 #include <linux/security.h> 32 #include <linux/security.h>
36 #include <linux/cpuset.h> 33 #include <linux/cpuset.h>
37 #include <linux/hugetlb.h> 34 #include <linux/hugetlb.h>
38 #include <linux/memcontrol.h> 35 #include <linux/memcontrol.h>
>> 36 #include <linux/cleancache.h>
39 #include <linux/shmem_fs.h> 37 #include <linux/shmem_fs.h>
40 #include <linux/rmap.h> 38 #include <linux/rmap.h>
41 #include <linux/delayacct.h> 39 #include <linux/delayacct.h>
42 #include <linux/psi.h> 40 #include <linux/psi.h>
43 #include <linux/ramfs.h> <<
44 #include <linux/page_idle.h> <<
45 #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> <<
51 #include <asm/tlbflush.h> <<
52 #include "internal.h" 41 #include "internal.h"
53 42
54 #define CREATE_TRACE_POINTS 43 #define CREATE_TRACE_POINTS
55 #include <trace/events/filemap.h> 44 #include <trace/events/filemap.h>
56 45
57 /* 46 /*
58 * FIXME: remove all knowledge of the buffer l 47 * FIXME: remove all knowledge of the buffer layer from the core VM
59 */ 48 */
60 #include <linux/buffer_head.h> /* for try_to_f 49 #include <linux/buffer_head.h> /* for try_to_free_buffers */
61 50
62 #include <asm/mman.h> 51 #include <asm/mman.h>
63 52
64 #include "swap.h" <<
65 <<
66 /* 53 /*
67 * Shared mappings implemented 30.11.1994. It' 54 * Shared mappings implemented 30.11.1994. It's not fully working yet,
68 * though. 55 * though.
69 * 56 *
70 * Shared mappings now work. 15.8.1995 Bruno. 57 * Shared mappings now work. 15.8.1995 Bruno.
71 * 58 *
72 * finished 'unifying' the page and buffer cac 59 * finished 'unifying' the page and buffer cache and SMP-threaded the
73 * page-cache, 21.05.1999, Ingo Molnar <mingo@ 60 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
74 * 61 *
75 * SMP-threaded pagemap-LRU 1999, Andrea Arcan 62 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
76 */ 63 */
77 64
78 /* 65 /*
79 * Lock ordering: 66 * Lock ordering:
80 * 67 *
81 * ->i_mmap_rwsem (truncate_page 68 * ->i_mmap_rwsem (truncate_pagecache)
82 * ->private_lock (__free_pte->b !! 69 * ->private_lock (__free_pte->__set_page_dirty_buffers)
83 * ->swap_lock (exclusive_swa 70 * ->swap_lock (exclusive_swap_page, others)
84 * ->i_pages lock 71 * ->i_pages lock
85 * 72 *
86 * ->i_rwsem !! 73 * ->i_mutex
87 * ->invalidate_lock (acquired by f !! 74 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
88 * ->i_mmap_rwsem (truncate->unm <<
89 * 75 *
90 * ->mmap_lock !! 76 * ->mmap_sem
91 * ->i_mmap_rwsem 77 * ->i_mmap_rwsem
92 * ->page_table_lock or pte_lock (vario 78 * ->page_table_lock or pte_lock (various, mainly in memory.c)
93 * ->i_pages lock (arch-dependen 79 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
94 * 80 *
95 * ->mmap_lock !! 81 * ->mmap_sem
96 * ->invalidate_lock (filemap_fault !! 82 * ->lock_page (access_process_vm)
97 * ->lock_page (filemap_fault <<
98 * 83 *
99 * ->i_rwsem (generic_perfo !! 84 * ->i_mutex (generic_perform_write)
100 * ->mmap_lock (fault_in_read !! 85 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
101 * 86 *
102 * bdi->wb.list_lock 87 * bdi->wb.list_lock
103 * sb_lock (fs/fs-writeba 88 * sb_lock (fs/fs-writeback.c)
104 * ->i_pages lock (__sync_single 89 * ->i_pages lock (__sync_single_inode)
105 * 90 *
106 * ->i_mmap_rwsem 91 * ->i_mmap_rwsem
107 * ->anon_vma.lock (vma_merge) !! 92 * ->anon_vma.lock (vma_adjust)
108 * 93 *
109 * ->anon_vma.lock 94 * ->anon_vma.lock
110 * ->page_table_lock or pte_lock (anon_ 95 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
111 * 96 *
112 * ->page_table_lock or pte_lock 97 * ->page_table_lock or pte_lock
113 * ->swap_lock (try_to_unmap_ 98 * ->swap_lock (try_to_unmap_one)
114 * ->private_lock (try_to_unmap_ 99 * ->private_lock (try_to_unmap_one)
115 * ->i_pages lock (try_to_unmap_ 100 * ->i_pages lock (try_to_unmap_one)
116 * ->lruvec->lru_lock (follow_page_m !! 101 * ->pgdat->lru_lock (follow_page->mark_page_accessed)
117 * ->lruvec->lru_lock (check_pte_ran !! 102 * ->pgdat->lru_lock (check_pte_range->isolate_lru_page)
118 * ->private_lock (folio_remove_ !! 103 * ->private_lock (page_remove_rmap->set_page_dirty)
119 * ->i_pages lock (folio_remove_ !! 104 * ->i_pages lock (page_remove_rmap->set_page_dirty)
120 * bdi.wb->list_lock (folio_remove_ !! 105 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
121 * ->inode->i_lock (folio_remove_ !! 106 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
122 * ->memcg->move_lock (folio_remove_ !! 107 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
123 * bdi.wb->list_lock (zap_pte_range 108 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
124 * ->inode->i_lock (zap_pte_range 109 * ->inode->i_lock (zap_pte_range->set_page_dirty)
125 * ->private_lock (zap_pte_range !! 110 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
>> 111 *
>> 112 * ->i_mmap_rwsem
>> 113 * ->tasklist_lock (memory_failure, collect_procs_ao)
126 */ 114 */
127 115
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 116 static void page_cache_delete(struct address_space *mapping,
138 struct foli !! 117 struct page *page, void *shadow)
139 { 118 {
140 XA_STATE(xas, &mapping->i_pages, folio !! 119 XA_STATE(xas, &mapping->i_pages, page->index);
141 long nr = 1; !! 120 unsigned int nr = 1;
142 121
143 mapping_set_update(&xas, mapping); 122 mapping_set_update(&xas, mapping);
144 123
145 xas_set_order(&xas, folio->index, foli !! 124 /* hugetlb pages are represented by a single entry in the xarray */
146 nr = folio_nr_pages(folio); !! 125 if (!PageHuge(page)) {
>> 126 xas_set_order(&xas, page->index, compound_order(page));
>> 127 nr = 1U << compound_order(page);
>> 128 }
147 129
148 VM_BUG_ON_FOLIO(!folio_test_locked(fol !! 130 VM_BUG_ON_PAGE(!PageLocked(page), page);
>> 131 VM_BUG_ON_PAGE(PageTail(page), page);
>> 132 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
149 133
150 xas_store(&xas, shadow); 134 xas_store(&xas, shadow);
151 xas_init_marks(&xas); 135 xas_init_marks(&xas);
152 136
153 folio->mapping = NULL; !! 137 page->mapping = NULL;
154 /* Leave page->index set: truncation l 138 /* Leave page->index set: truncation lookup relies upon it */
>> 139
>> 140 if (shadow) {
>> 141 mapping->nrexceptional += nr;
>> 142 /*
>> 143 * Make sure the nrexceptional update is committed before
>> 144 * the nrpages update so that final truncate racing
>> 145 * with reclaim does not see both counters 0 at the
>> 146 * same time and miss a shadow entry.
>> 147 */
>> 148 smp_wmb();
>> 149 }
155 mapping->nrpages -= nr; 150 mapping->nrpages -= nr;
156 } 151 }
157 152
158 static void filemap_unaccount_folio(struct add !! 153 static void unaccount_page_cache_page(struct address_space *mapping,
159 struct folio *folio) !! 154 struct page *page)
160 { 155 {
161 long nr; !! 156 int nr;
>> 157
>> 158 /*
>> 159 * if we're uptodate, flush out into the cleancache, otherwise
>> 160 * invalidate any existing cleancache entries. We can't leave
>> 161 * stale data around in the cleancache once our page is gone
>> 162 */
>> 163 if (PageUptodate(page) && PageMappedToDisk(page))
>> 164 cleancache_put_page(page);
>> 165 else
>> 166 cleancache_invalidate_page(mapping, page);
>> 167
>> 168 VM_BUG_ON_PAGE(PageTail(page), page);
>> 169 VM_BUG_ON_PAGE(page_mapped(page), page);
>> 170 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
>> 171 int mapcount;
162 172
163 VM_BUG_ON_FOLIO(folio_mapped(folio), f <<
164 if (!IS_ENABLED(CONFIG_DEBUG_VM) && un <<
165 pr_alert("BUG: Bad page cache 173 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
166 current->comm, folio_ !! 174 current->comm, page_to_pfn(page));
167 dump_page(&folio->page, "still !! 175 dump_page(page, "still mapped when deleted");
168 dump_stack(); 176 dump_stack();
169 add_taint(TAINT_BAD_PAGE, LOCK 177 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
170 178
171 if (mapping_exiting(mapping) & !! 179 mapcount = page_mapcount(page);
172 int mapcount = folio_m !! 180 if (mapping_exiting(mapping) &&
173 !! 181 page_count(page) >= mapcount + 2) {
174 if (folio_ref_count(fo !! 182 /*
175 /* !! 183 * All vmas have already been torn down, so it's
176 * All vmas ha !! 184 * a good bet that actually the page is unmapped,
177 * a good bet !! 185 * and we'd prefer not to leak it: if we're wrong,
178 * and we'd ra !! 186 * some other bad page check should catch it later.
179 * another bad !! 187 */
180 */ !! 188 page_mapcount_reset(page);
181 atomic_set(&fo !! 189 page_ref_sub(page, mapcount);
182 folio_ref_sub( <<
183 } <<
184 } 190 }
185 } 191 }
186 192
187 /* hugetlb folios do not participate i !! 193 /* hugetlb pages do not participate in page cache accounting. */
188 if (folio_test_hugetlb(folio)) !! 194 if (PageHuge(page))
189 return; 195 return;
190 196
191 nr = folio_nr_pages(folio); !! 197 nr = hpage_nr_pages(page);
192 198
193 __lruvec_stat_mod_folio(folio, NR_FILE !! 199 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
194 if (folio_test_swapbacked(folio)) { !! 200 if (PageSwapBacked(page)) {
195 __lruvec_stat_mod_folio(folio, !! 201 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
196 if (folio_test_pmd_mappable(fo !! 202 if (PageTransHuge(page))
197 __lruvec_stat_mod_foli !! 203 __dec_node_page_state(page, NR_SHMEM_THPS);
198 } else if (folio_test_pmd_mappable(fol !! 204 } else {
199 __lruvec_stat_mod_folio(folio, !! 205 VM_BUG_ON_PAGE(PageTransHuge(page), page);
200 filemap_nr_thps_dec(mapping); <<
201 } 206 }
202 207
203 /* 208 /*
204 * At this point folio must be either !! 209 * At this point page must be either written or cleaned by
205 * truncate. Dirty folio here signals !! 210 * truncate. Dirty page here signals a bug and loss of
206 * unwritten data - on ordinary filesy !! 211 * unwritten data.
207 * <<
208 * But it's harmless on in-memory file <<
209 * occur when a driver which did get_u <<
210 * before putting it, while the inode <<
211 * 212 *
212 * Below fixes dirty accounting after !! 213 * This fixes dirty accounting after removing the page entirely
213 * but leaves the dirty flag set: it h !! 214 * but leaves PageDirty set: it has no effect for truncated
214 * folio and anyway will be cleared be !! 215 * page and anyway will be cleared before returning page into
215 * buddy allocator. 216 * buddy allocator.
216 */ 217 */
217 if (WARN_ON_ONCE(folio_test_dirty(foli !! 218 if (WARN_ON_ONCE(PageDirty(page)))
218 mapping_can_writeback !! 219 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
219 folio_account_cleaned(folio, i <<
220 } 220 }
221 221
222 /* 222 /*
223 * Delete a page from the page cache and free 223 * Delete a page from the page cache and free it. Caller has to make
224 * sure the page is locked and that nobody els 224 * sure the page is locked and that nobody else uses it - or that usage
225 * is safe. The caller must hold the i_pages 225 * is safe. The caller must hold the i_pages lock.
226 */ 226 */
227 void __filemap_remove_folio(struct folio *foli !! 227 void __delete_from_page_cache(struct page *page, void *shadow)
228 { 228 {
229 struct address_space *mapping = folio- !! 229 struct address_space *mapping = page->mapping;
230 230
231 trace_mm_filemap_delete_from_page_cach !! 231 trace_mm_filemap_delete_from_page_cache(page);
232 filemap_unaccount_folio(mapping, folio !! 232
233 page_cache_delete(mapping, folio, shad !! 233 unaccount_page_cache_page(mapping, page);
>> 234 page_cache_delete(mapping, page, shadow);
234 } 235 }
235 236
236 void filemap_free_folio(struct address_space * !! 237 static void page_cache_free_page(struct address_space *mapping,
>> 238 struct page *page)
237 { 239 {
238 void (*free_folio)(struct folio *); !! 240 void (*freepage)(struct page *);
239 int refs = 1; <<
240 241
241 free_folio = mapping->a_ops->free_foli !! 242 freepage = mapping->a_ops->freepage;
242 if (free_folio) !! 243 if (freepage)
243 free_folio(folio); !! 244 freepage(page);
244 245
245 if (folio_test_large(folio)) !! 246 if (PageTransHuge(page) && !PageHuge(page)) {
246 refs = folio_nr_pages(folio); !! 247 page_ref_sub(page, HPAGE_PMD_NR);
247 folio_put_refs(folio, refs); !! 248 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
>> 249 } else {
>> 250 put_page(page);
>> 251 }
248 } 252 }
249 253
250 /** 254 /**
251 * filemap_remove_folio - Remove folio from pa !! 255 * delete_from_page_cache - delete page from page cache
252 * @folio: The folio. !! 256 * @page: the page which the kernel is trying to remove from page cache
253 * 257 *
254 * This must be called only on folios that are !! 258 * This must be called only on pages that have been verified to be in the page
255 * verified to be in the page cache. It will !! 259 * cache and locked. It will never put the page into the free list, the caller
256 * the free list because the caller has a refe !! 260 * has a reference on the page.
257 */ !! 261 */
258 void filemap_remove_folio(struct folio *folio) !! 262 void delete_from_page_cache(struct page *page)
259 { !! 263 {
260 struct address_space *mapping = folio- !! 264 struct address_space *mapping = page_mapping(page);
261 !! 265 unsigned long flags;
262 BUG_ON(!folio_test_locked(folio)); <<
263 spin_lock(&mapping->host->i_lock); <<
264 xa_lock_irq(&mapping->i_pages); <<
265 __filemap_remove_folio(folio, NULL); <<
266 xa_unlock_irq(&mapping->i_pages); <<
267 if (mapping_shrinkable(mapping)) <<
268 inode_add_lru(mapping->host); <<
269 spin_unlock(&mapping->host->i_lock); <<
270 <<
271 filemap_free_folio(mapping, folio); <<
272 } <<
273 266
274 /* !! 267 BUG_ON(!PageLocked(page));
275 * page_cache_delete_batch - delete several fo !! 268 xa_lock_irqsave(&mapping->i_pages, flags);
276 * @mapping: the mapping to which folios belon !! 269 __delete_from_page_cache(page, NULL);
277 * @fbatch: batch of folios to delete !! 270 xa_unlock_irqrestore(&mapping->i_pages, flags);
278 * !! 271
279 * The function walks over mapping->i_pages an !! 272 page_cache_free_page(mapping, page);
280 * @fbatch from the mapping. The function expe !! 273 }
281 * by page index and is optimised for it to be !! 274 EXPORT_SYMBOL(delete_from_page_cache);
282 * It tolerates holes in @fbatch (mapping entr !! 275
283 * modified). !! 276 /*
>> 277 * page_cache_delete_batch - delete several pages from page cache
>> 278 * @mapping: the mapping to which pages belong
>> 279 * @pvec: pagevec with pages to delete
>> 280 *
>> 281 * The function walks over mapping->i_pages and removes pages passed in @pvec
>> 282 * from the mapping. The function expects @pvec to be sorted by page index.
>> 283 * It tolerates holes in @pvec (mapping entries at those indices are not
>> 284 * modified). The function expects only THP head pages to be present in the
>> 285 * @pvec and takes care to delete all corresponding tail pages from the
>> 286 * mapping as well.
284 * 287 *
285 * The function expects the i_pages lock to be 288 * The function expects the i_pages lock to be held.
286 */ 289 */
287 static void page_cache_delete_batch(struct add 290 static void page_cache_delete_batch(struct address_space *mapping,
288 struct folio_batc !! 291 struct pagevec *pvec)
289 { 292 {
290 XA_STATE(xas, &mapping->i_pages, fbatc !! 293 XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
291 long total_pages = 0; !! 294 int total_pages = 0;
292 int i = 0; !! 295 int i = 0, tail_pages = 0;
293 struct folio *folio; !! 296 struct page *page;
294 297
295 mapping_set_update(&xas, mapping); 298 mapping_set_update(&xas, mapping);
296 xas_for_each(&xas, folio, ULONG_MAX) { !! 299 xas_for_each(&xas, page, ULONG_MAX) {
297 if (i >= folio_batch_count(fba !! 300 if (i >= pagevec_count(pvec) && !tail_pages)
298 break; 301 break;
299 !! 302 if (xa_is_value(page))
300 /* A swap/dax/shadow entry got <<
301 if (xa_is_value(folio)) <<
302 continue; <<
303 /* <<
304 * A page got inserted in our <<
305 * pages locked so they are pr <<
306 * If we see a page whose inde <<
307 * means our page has been rem <<
308 * possible because we're hold <<
309 */ <<
310 if (folio != fbatch->folios[i] <<
311 VM_BUG_ON_FOLIO(folio- <<
312 fbatch <<
313 continue; 303 continue;
>> 304 if (!tail_pages) {
>> 305 /*
>> 306 * Some page got inserted in our range? Skip it. We
>> 307 * have our pages locked so they are protected from
>> 308 * being removed.
>> 309 */
>> 310 if (page != pvec->pages[i]) {
>> 311 VM_BUG_ON_PAGE(page->index >
>> 312 pvec->pages[i]->index, page);
>> 313 continue;
>> 314 }
>> 315 WARN_ON_ONCE(!PageLocked(page));
>> 316 if (PageTransHuge(page) && !PageHuge(page))
>> 317 tail_pages = HPAGE_PMD_NR - 1;
>> 318 page->mapping = NULL;
>> 319 /*
>> 320 * Leave page->index set: truncation lookup relies
>> 321 * upon it
>> 322 */
>> 323 i++;
>> 324 } else {
>> 325 VM_BUG_ON_PAGE(page->index + HPAGE_PMD_NR - tail_pages
>> 326 != pvec->pages[i]->index, page);
>> 327 tail_pages--;
314 } 328 }
315 <<
316 WARN_ON_ONCE(!folio_test_locke <<
317 <<
318 folio->mapping = NULL; <<
319 /* Leave folio->index set: tru <<
320 <<
321 i++; <<
322 xas_store(&xas, NULL); 329 xas_store(&xas, NULL);
323 total_pages += folio_nr_pages( !! 330 total_pages++;
324 } 331 }
325 mapping->nrpages -= total_pages; 332 mapping->nrpages -= total_pages;
326 } 333 }
327 334
328 void delete_from_page_cache_batch(struct addre 335 void delete_from_page_cache_batch(struct address_space *mapping,
329 struct folio !! 336 struct pagevec *pvec)
330 { 337 {
331 int i; 338 int i;
>> 339 unsigned long flags;
332 340
333 if (!folio_batch_count(fbatch)) !! 341 if (!pagevec_count(pvec))
334 return; 342 return;
335 343
336 spin_lock(&mapping->host->i_lock); !! 344 xa_lock_irqsave(&mapping->i_pages, flags);
337 xa_lock_irq(&mapping->i_pages); !! 345 for (i = 0; i < pagevec_count(pvec); i++) {
338 for (i = 0; i < folio_batch_count(fbat !! 346 trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
339 struct folio *folio = fbatch-> <<
340 <<
341 trace_mm_filemap_delete_from_p <<
342 filemap_unaccount_folio(mappin <<
343 } <<
344 page_cache_delete_batch(mapping, fbatc <<
345 xa_unlock_irq(&mapping->i_pages); <<
346 if (mapping_shrinkable(mapping)) <<
347 inode_add_lru(mapping->host); <<
348 spin_unlock(&mapping->host->i_lock); <<
349 347
350 for (i = 0; i < folio_batch_count(fbat !! 348 unaccount_page_cache_page(mapping, pvec->pages[i]);
351 filemap_free_folio(mapping, fb !! 349 }
>> 350 page_cache_delete_batch(mapping, pvec);
>> 351 xa_unlock_irqrestore(&mapping->i_pages, flags);
>> 352
>> 353 for (i = 0; i < pagevec_count(pvec); i++)
>> 354 page_cache_free_page(mapping, pvec->pages[i]);
352 } 355 }
353 356
354 int filemap_check_errors(struct address_space 357 int filemap_check_errors(struct address_space *mapping)
355 { 358 {
356 int ret = 0; 359 int ret = 0;
357 /* Check for outstanding write errors 360 /* Check for outstanding write errors */
358 if (test_bit(AS_ENOSPC, &mapping->flag 361 if (test_bit(AS_ENOSPC, &mapping->flags) &&
359 test_and_clear_bit(AS_ENOSPC, &map 362 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
360 ret = -ENOSPC; 363 ret = -ENOSPC;
361 if (test_bit(AS_EIO, &mapping->flags) 364 if (test_bit(AS_EIO, &mapping->flags) &&
362 test_and_clear_bit(AS_EIO, &mappin 365 test_and_clear_bit(AS_EIO, &mapping->flags))
363 ret = -EIO; 366 ret = -EIO;
364 return ret; 367 return ret;
365 } 368 }
366 EXPORT_SYMBOL(filemap_check_errors); 369 EXPORT_SYMBOL(filemap_check_errors);
367 370
368 static int filemap_check_and_keep_errors(struc 371 static int filemap_check_and_keep_errors(struct address_space *mapping)
369 { 372 {
370 /* Check for outstanding write errors 373 /* Check for outstanding write errors */
371 if (test_bit(AS_EIO, &mapping->flags)) 374 if (test_bit(AS_EIO, &mapping->flags))
372 return -EIO; 375 return -EIO;
373 if (test_bit(AS_ENOSPC, &mapping->flag 376 if (test_bit(AS_ENOSPC, &mapping->flags))
374 return -ENOSPC; 377 return -ENOSPC;
375 return 0; 378 return 0;
376 } 379 }
377 380
378 /** 381 /**
379 * filemap_fdatawrite_wbc - start writeback on <<
380 * @mapping: address space structure to wri <<
381 * @wbc: the writeback_control controll <<
382 * <<
383 * Call writepages on the mapping using the pr <<
384 * writeout. <<
385 * <<
386 * Return: %0 on success, negative error code <<
387 */ <<
388 int filemap_fdatawrite_wbc(struct address_spac <<
389 struct writeback_co <<
390 { <<
391 int ret; <<
392 <<
393 if (!mapping_can_writeback(mapping) || <<
394 !mapping_tagged(mapping, PAGECACHE <<
395 return 0; <<
396 <<
397 wbc_attach_fdatawrite_inode(wbc, mappi <<
398 ret = do_writepages(mapping, wbc); <<
399 wbc_detach_inode(wbc); <<
400 return ret; <<
401 } <<
402 EXPORT_SYMBOL(filemap_fdatawrite_wbc); <<
403 <<
404 /** <<
405 * __filemap_fdatawrite_range - start writebac 382 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
406 * @mapping: address space structure to wri 383 * @mapping: address space structure to write
407 * @start: offset in bytes where the rang 384 * @start: offset in bytes where the range starts
408 * @end: offset in bytes where the rang 385 * @end: offset in bytes where the range ends (inclusive)
409 * @sync_mode: enable synchronous operation 386 * @sync_mode: enable synchronous operation
410 * 387 *
411 * Start writeback against all of a mapping's 388 * Start writeback against all of a mapping's dirty pages that lie
412 * within the byte offsets <start, end> inclus 389 * within the byte offsets <start, end> inclusive.
413 * 390 *
414 * If sync_mode is WB_SYNC_ALL then this is a 391 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
415 * opposed to a regular memory cleansing write 392 * opposed to a regular memory cleansing writeback. The difference between
416 * these two operations is that if a dirty pag 393 * these two operations is that if a dirty page/buffer is encountered, it must
417 * be waited upon, and not just skipped over. 394 * be waited upon, and not just skipped over.
418 * 395 *
419 * Return: %0 on success, negative error code 396 * Return: %0 on success, negative error code otherwise.
420 */ 397 */
421 int __filemap_fdatawrite_range(struct address_ 398 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
422 loff_t end, in 399 loff_t end, int sync_mode)
423 { 400 {
>> 401 int ret;
424 struct writeback_control wbc = { 402 struct writeback_control wbc = {
425 .sync_mode = sync_mode, 403 .sync_mode = sync_mode,
426 .nr_to_write = LONG_MAX, 404 .nr_to_write = LONG_MAX,
427 .range_start = start, 405 .range_start = start,
428 .range_end = end, 406 .range_end = end,
429 }; 407 };
430 408
431 return filemap_fdatawrite_wbc(mapping, !! 409 if (!mapping_cap_writeback_dirty(mapping))
>> 410 return 0;
>> 411
>> 412 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
>> 413 ret = do_writepages(mapping, &wbc);
>> 414 wbc_detach_inode(&wbc);
>> 415 return ret;
432 } 416 }
433 417
434 static inline int __filemap_fdatawrite(struct 418 static inline int __filemap_fdatawrite(struct address_space *mapping,
435 int sync_mode) 419 int sync_mode)
436 { 420 {
437 return __filemap_fdatawrite_range(mapp 421 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
438 } 422 }
439 423
440 int filemap_fdatawrite(struct address_space *m 424 int filemap_fdatawrite(struct address_space *mapping)
441 { 425 {
442 return __filemap_fdatawrite(mapping, W 426 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
443 } 427 }
444 EXPORT_SYMBOL(filemap_fdatawrite); 428 EXPORT_SYMBOL(filemap_fdatawrite);
445 429
446 int filemap_fdatawrite_range(struct address_sp 430 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
447 loff_t end) 431 loff_t end)
448 { 432 {
449 return __filemap_fdatawrite_range(mapp 433 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
450 } 434 }
451 EXPORT_SYMBOL(filemap_fdatawrite_range); 435 EXPORT_SYMBOL(filemap_fdatawrite_range);
452 436
453 /** 437 /**
454 * filemap_flush - mostly a non-blocking flush 438 * filemap_flush - mostly a non-blocking flush
455 * @mapping: target address_space 439 * @mapping: target address_space
456 * 440 *
457 * This is a mostly non-blocking flush. Not s 441 * This is a mostly non-blocking flush. Not suitable for data-integrity
458 * purposes - I/O may not be started against a 442 * purposes - I/O may not be started against all dirty pages.
459 * 443 *
460 * Return: %0 on success, negative error code 444 * Return: %0 on success, negative error code otherwise.
461 */ 445 */
462 int filemap_flush(struct address_space *mappin 446 int filemap_flush(struct address_space *mapping)
463 { 447 {
464 return __filemap_fdatawrite(mapping, W 448 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
465 } 449 }
466 EXPORT_SYMBOL(filemap_flush); 450 EXPORT_SYMBOL(filemap_flush);
467 451
468 /** 452 /**
469 * filemap_range_has_page - check if a page ex 453 * filemap_range_has_page - check if a page exists in range.
470 * @mapping: address space within wh 454 * @mapping: address space within which to check
471 * @start_byte: offset in bytes where t 455 * @start_byte: offset in bytes where the range starts
472 * @end_byte: offset in bytes where t 456 * @end_byte: offset in bytes where the range ends (inclusive)
473 * 457 *
474 * Find at least one page in the range supplie 458 * Find at least one page in the range supplied, usually used to check if
475 * direct writing in this range will trigger a 459 * direct writing in this range will trigger a writeback.
476 * 460 *
477 * Return: %true if at least one page exists i 461 * Return: %true if at least one page exists in the specified range,
478 * %false otherwise. 462 * %false otherwise.
479 */ 463 */
480 bool filemap_range_has_page(struct address_spa 464 bool filemap_range_has_page(struct address_space *mapping,
481 loff_t start_byte, 465 loff_t start_byte, loff_t end_byte)
482 { 466 {
483 struct folio *folio; !! 467 struct page *page;
484 XA_STATE(xas, &mapping->i_pages, start 468 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
485 pgoff_t max = end_byte >> PAGE_SHIFT; 469 pgoff_t max = end_byte >> PAGE_SHIFT;
486 470
487 if (end_byte < start_byte) 471 if (end_byte < start_byte)
488 return false; 472 return false;
489 473
490 rcu_read_lock(); 474 rcu_read_lock();
491 for (;;) { 475 for (;;) {
492 folio = xas_find(&xas, max); !! 476 page = xas_find(&xas, max);
493 if (xas_retry(&xas, folio)) !! 477 if (xas_retry(&xas, page))
494 continue; 478 continue;
495 /* Shadow entries don't count 479 /* Shadow entries don't count */
496 if (xa_is_value(folio)) !! 480 if (xa_is_value(page))
497 continue; 481 continue;
498 /* 482 /*
499 * We don't need to try to pin 483 * We don't need to try to pin this page; we're about to
500 * release the RCU lock anyway 484 * release the RCU lock anyway. It is enough to know that
501 * there was a page here recen 485 * there was a page here recently.
502 */ 486 */
503 break; 487 break;
504 } 488 }
505 rcu_read_unlock(); 489 rcu_read_unlock();
506 490
507 return folio != NULL; !! 491 return page != NULL;
508 } 492 }
509 EXPORT_SYMBOL(filemap_range_has_page); 493 EXPORT_SYMBOL(filemap_range_has_page);
510 494
511 static void __filemap_fdatawait_range(struct a 495 static void __filemap_fdatawait_range(struct address_space *mapping,
512 loff_t st 496 loff_t start_byte, loff_t end_byte)
513 { 497 {
514 pgoff_t index = start_byte >> PAGE_SHI 498 pgoff_t index = start_byte >> PAGE_SHIFT;
515 pgoff_t end = end_byte >> PAGE_SHIFT; 499 pgoff_t end = end_byte >> PAGE_SHIFT;
516 struct folio_batch fbatch; !! 500 struct pagevec pvec;
517 unsigned nr_folios; !! 501 int nr_pages;
518 502
519 folio_batch_init(&fbatch); !! 503 if (end_byte < start_byte)
>> 504 return;
520 505
>> 506 pagevec_init(&pvec);
521 while (index <= end) { 507 while (index <= end) {
522 unsigned i; 508 unsigned i;
523 509
524 nr_folios = filemap_get_folios !! 510 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
525 PAGECACHE_TAG_ !! 511 end, PAGECACHE_TAG_WRITEBACK);
526 !! 512 if (!nr_pages)
527 if (!nr_folios) <<
528 break; 513 break;
529 514
530 for (i = 0; i < nr_folios; i++ !! 515 for (i = 0; i < nr_pages; i++) {
531 struct folio *folio = !! 516 struct page *page = pvec.pages[i];
532 517
533 folio_wait_writeback(f !! 518 wait_on_page_writeback(page);
>> 519 ClearPageError(page);
534 } 520 }
535 folio_batch_release(&fbatch); !! 521 pagevec_release(&pvec);
536 cond_resched(); 522 cond_resched();
537 } 523 }
538 } 524 }
539 525
540 /** 526 /**
541 * filemap_fdatawait_range - wait for writebac 527 * filemap_fdatawait_range - wait for writeback to complete
542 * @mapping: address space structur 528 * @mapping: address space structure to wait for
543 * @start_byte: offset in bytes where 529 * @start_byte: offset in bytes where the range starts
544 * @end_byte: offset in bytes where 530 * @end_byte: offset in bytes where the range ends (inclusive)
545 * 531 *
546 * Walk the list of under-writeback pages of t 532 * Walk the list of under-writeback pages of the given address space
547 * in the given range and wait for all of them 533 * in the given range and wait for all of them. Check error status of
548 * the address space and return it. 534 * the address space and return it.
549 * 535 *
550 * Since the error status of the address space 536 * Since the error status of the address space is cleared by this function,
551 * callers are responsible for checking the re 537 * callers are responsible for checking the return value and handling and/or
552 * reporting the error. 538 * reporting the error.
553 * 539 *
554 * Return: error status of the address space. 540 * Return: error status of the address space.
555 */ 541 */
556 int filemap_fdatawait_range(struct address_spa 542 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
557 loff_t end_byte) 543 loff_t end_byte)
558 { 544 {
559 __filemap_fdatawait_range(mapping, sta 545 __filemap_fdatawait_range(mapping, start_byte, end_byte);
560 return filemap_check_errors(mapping); 546 return filemap_check_errors(mapping);
561 } 547 }
562 EXPORT_SYMBOL(filemap_fdatawait_range); 548 EXPORT_SYMBOL(filemap_fdatawait_range);
563 549
564 /** 550 /**
565 * filemap_fdatawait_range_keep_errors - wait 551 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
566 * @mapping: address space structur 552 * @mapping: address space structure to wait for
567 * @start_byte: offset in bytes where 553 * @start_byte: offset in bytes where the range starts
568 * @end_byte: offset in bytes where 554 * @end_byte: offset in bytes where the range ends (inclusive)
569 * 555 *
570 * Walk the list of under-writeback pages of t 556 * Walk the list of under-writeback pages of the given address space in the
571 * given range and wait for all of them. Unli 557 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
572 * this function does not clear error status o 558 * this function does not clear error status of the address space.
573 * 559 *
574 * Use this function if callers don't handle e 560 * Use this function if callers don't handle errors themselves. Expected
575 * call sites are system-wide / filesystem-wid 561 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
576 * fsfreeze(8) 562 * fsfreeze(8)
577 */ 563 */
578 int filemap_fdatawait_range_keep_errors(struct 564 int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
579 loff_t start_byte, loff_t end_ 565 loff_t start_byte, loff_t end_byte)
580 { 566 {
581 __filemap_fdatawait_range(mapping, sta 567 __filemap_fdatawait_range(mapping, start_byte, end_byte);
582 return filemap_check_and_keep_errors(m 568 return filemap_check_and_keep_errors(mapping);
583 } 569 }
584 EXPORT_SYMBOL(filemap_fdatawait_range_keep_err 570 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
585 571
586 /** 572 /**
587 * file_fdatawait_range - wait for writeback t 573 * file_fdatawait_range - wait for writeback to complete
588 * @file: file pointing to addre 574 * @file: file pointing to address space structure to wait for
589 * @start_byte: offset in bytes where 575 * @start_byte: offset in bytes where the range starts
590 * @end_byte: offset in bytes where 576 * @end_byte: offset in bytes where the range ends (inclusive)
591 * 577 *
592 * Walk the list of under-writeback pages of t 578 * Walk the list of under-writeback pages of the address space that file
593 * refers to, in the given range and wait for 579 * refers to, in the given range and wait for all of them. Check error
594 * status of the address space vs. the file->f 580 * status of the address space vs. the file->f_wb_err cursor and return it.
595 * 581 *
596 * Since the error status of the file is advan 582 * Since the error status of the file is advanced by this function,
597 * callers are responsible for checking the re 583 * callers are responsible for checking the return value and handling and/or
598 * reporting the error. 584 * reporting the error.
599 * 585 *
600 * Return: error status of the address space v 586 * Return: error status of the address space vs. the file->f_wb_err cursor.
601 */ 587 */
602 int file_fdatawait_range(struct file *file, lo 588 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
603 { 589 {
604 struct address_space *mapping = file-> 590 struct address_space *mapping = file->f_mapping;
605 591
606 __filemap_fdatawait_range(mapping, sta 592 __filemap_fdatawait_range(mapping, start_byte, end_byte);
607 return file_check_and_advance_wb_err(f 593 return file_check_and_advance_wb_err(file);
608 } 594 }
609 EXPORT_SYMBOL(file_fdatawait_range); 595 EXPORT_SYMBOL(file_fdatawait_range);
610 596
611 /** 597 /**
612 * filemap_fdatawait_keep_errors - wait for wr 598 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
613 * @mapping: address space structure to wait f 599 * @mapping: address space structure to wait for
614 * 600 *
615 * Walk the list of under-writeback pages of t 601 * Walk the list of under-writeback pages of the given address space
616 * and wait for all of them. Unlike filemap_f 602 * and wait for all of them. Unlike filemap_fdatawait(), this function
617 * does not clear error status of the address 603 * does not clear error status of the address space.
618 * 604 *
619 * Use this function if callers don't handle e 605 * Use this function if callers don't handle errors themselves. Expected
620 * call sites are system-wide / filesystem-wid 606 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
621 * fsfreeze(8) 607 * fsfreeze(8)
622 * 608 *
623 * Return: error status of the address space. 609 * Return: error status of the address space.
624 */ 610 */
625 int filemap_fdatawait_keep_errors(struct addre 611 int filemap_fdatawait_keep_errors(struct address_space *mapping)
626 { 612 {
627 __filemap_fdatawait_range(mapping, 0, 613 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
628 return filemap_check_and_keep_errors(m 614 return filemap_check_and_keep_errors(mapping);
629 } 615 }
630 EXPORT_SYMBOL(filemap_fdatawait_keep_errors); 616 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
631 617
632 /* Returns true if writeback might be needed o <<
633 static bool mapping_needs_writeback(struct add 618 static bool mapping_needs_writeback(struct address_space *mapping)
634 { 619 {
635 return mapping->nrpages; !! 620 return (!dax_mapping(mapping) && mapping->nrpages) ||
>> 621 (dax_mapping(mapping) && mapping->nrexceptional);
636 } 622 }
637 623
638 bool filemap_range_has_writeback(struct addres !! 624 int filemap_write_and_wait(struct address_space *mapping)
639 loff_t start_ <<
640 { 625 {
641 XA_STATE(xas, &mapping->i_pages, start !! 626 int err = 0;
642 pgoff_t max = end_byte >> PAGE_SHIFT; <<
643 struct folio *folio; <<
644 <<
645 if (end_byte < start_byte) <<
646 return false; <<
647 627
648 rcu_read_lock(); !! 628 if (mapping_needs_writeback(mapping)) {
649 xas_for_each(&xas, folio, max) { !! 629 err = filemap_fdatawrite(mapping);
650 if (xas_retry(&xas, folio)) !! 630 /*
651 continue; !! 631 * Even if the above returned error, the pages may be
652 if (xa_is_value(folio)) !! 632 * written partially (e.g. -ENOSPC), so we wait for it.
653 continue; !! 633 * But the -EIO is special case, it may indicate the worst
654 if (folio_test_dirty(folio) || !! 634 * thing (e.g. bug) happened, so we avoid waiting for it.
655 folio_test_wri !! 635 */
656 break; !! 636 if (err != -EIO) {
>> 637 int err2 = filemap_fdatawait(mapping);
>> 638 if (!err)
>> 639 err = err2;
>> 640 } else {
>> 641 /* Clear any previously stored errors */
>> 642 filemap_check_errors(mapping);
>> 643 }
>> 644 } else {
>> 645 err = filemap_check_errors(mapping);
657 } 646 }
658 rcu_read_unlock(); !! 647 return err;
659 return folio != NULL; <<
660 } 648 }
661 EXPORT_SYMBOL_GPL(filemap_range_has_writeback) !! 649 EXPORT_SYMBOL(filemap_write_and_wait);
662 650
663 /** 651 /**
664 * filemap_write_and_wait_range - write out & 652 * filemap_write_and_wait_range - write out & wait on a file range
665 * @mapping: the address_space for the page 653 * @mapping: the address_space for the pages
666 * @lstart: offset in bytes where the rang 654 * @lstart: offset in bytes where the range starts
667 * @lend: offset in bytes where the rang 655 * @lend: offset in bytes where the range ends (inclusive)
668 * 656 *
669 * Write out and wait upon file offsets lstart 657 * Write out and wait upon file offsets lstart->lend, inclusive.
670 * 658 *
671 * Note that @lend is inclusive (describes the 659 * Note that @lend is inclusive (describes the last byte to be written) so
672 * that this function can be used to write to 660 * that this function can be used to write to the very end-of-file (end = -1).
673 * 661 *
674 * Return: error status of the address space. 662 * Return: error status of the address space.
675 */ 663 */
676 int filemap_write_and_wait_range(struct addres 664 int filemap_write_and_wait_range(struct address_space *mapping,
677 loff_t lstart 665 loff_t lstart, loff_t lend)
678 { 666 {
679 int err = 0, err2; !! 667 int err = 0;
680 <<
681 if (lend < lstart) <<
682 return 0; <<
683 668
684 if (mapping_needs_writeback(mapping)) 669 if (mapping_needs_writeback(mapping)) {
685 err = __filemap_fdatawrite_ran 670 err = __filemap_fdatawrite_range(mapping, lstart, lend,
686 671 WB_SYNC_ALL);
687 /* !! 672 /* See comment of filemap_write_and_wait() */
688 * Even if the above returned !! 673 if (err != -EIO) {
689 * written partially (e.g. -EN !! 674 int err2 = filemap_fdatawait_range(mapping,
690 * But the -EIO is special cas !! 675 lstart, lend);
691 * thing (e.g. bug) happened, !! 676 if (!err)
692 */ !! 677 err = err2;
693 if (err != -EIO) !! 678 } else {
694 __filemap_fdatawait_ra !! 679 /* Clear any previously stored errors */
>> 680 filemap_check_errors(mapping);
>> 681 }
>> 682 } else {
>> 683 err = filemap_check_errors(mapping);
695 } 684 }
696 err2 = filemap_check_errors(mapping); <<
697 if (!err) <<
698 err = err2; <<
699 return err; 685 return err;
700 } 686 }
701 EXPORT_SYMBOL(filemap_write_and_wait_range); 687 EXPORT_SYMBOL(filemap_write_and_wait_range);
702 688
703 void __filemap_set_wb_err(struct address_space 689 void __filemap_set_wb_err(struct address_space *mapping, int err)
704 { 690 {
705 errseq_t eseq = errseq_set(&mapping->w 691 errseq_t eseq = errseq_set(&mapping->wb_err, err);
706 692
707 trace_filemap_set_wb_err(mapping, eseq 693 trace_filemap_set_wb_err(mapping, eseq);
708 } 694 }
709 EXPORT_SYMBOL(__filemap_set_wb_err); 695 EXPORT_SYMBOL(__filemap_set_wb_err);
710 696
711 /** 697 /**
712 * file_check_and_advance_wb_err - report wb e 698 * file_check_and_advance_wb_err - report wb error (if any) that was previously
713 * and advance 699 * and advance wb_err to current one
714 * @file: struct file on which the error is be 700 * @file: struct file on which the error is being reported
715 * 701 *
716 * When userland calls fsync (or something lik 702 * When userland calls fsync (or something like nfsd does the equivalent), we
717 * want to report any writeback errors that oc 703 * want to report any writeback errors that occurred since the last fsync (or
718 * since the file was opened if there haven't 704 * since the file was opened if there haven't been any).
719 * 705 *
720 * Grab the wb_err from the mapping. If it mat 706 * 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 707 * then just quickly return 0. The file is all caught up.
722 * 708 *
723 * If it doesn't match, then take the mapping 709 * 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 710 * 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 711 * 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 712 * portion. The error at this point must be reported via proper channels
727 * (a'la fsync, or NFS COMMIT operation, etc.) 713 * (a'la fsync, or NFS COMMIT operation, etc.).
728 * 714 *
729 * While we handle mapping->wb_err with atomic 715 * While we handle mapping->wb_err with atomic operations, the f_wb_err
730 * value is protected by the f_lock since we m 716 * value is protected by the f_lock since we must ensure that it reflects
731 * the latest value swapped in for this file d 717 * the latest value swapped in for this file descriptor.
732 * 718 *
733 * Return: %0 on success, negative error code 719 * Return: %0 on success, negative error code otherwise.
734 */ 720 */
735 int file_check_and_advance_wb_err(struct file 721 int file_check_and_advance_wb_err(struct file *file)
736 { 722 {
737 int err = 0; 723 int err = 0;
738 errseq_t old = READ_ONCE(file->f_wb_er 724 errseq_t old = READ_ONCE(file->f_wb_err);
739 struct address_space *mapping = file-> 725 struct address_space *mapping = file->f_mapping;
740 726
741 /* Locklessly handle the common case w 727 /* Locklessly handle the common case where nothing has changed */
742 if (errseq_check(&mapping->wb_err, old 728 if (errseq_check(&mapping->wb_err, old)) {
743 /* Something changed, must use 729 /* Something changed, must use slow path */
744 spin_lock(&file->f_lock); 730 spin_lock(&file->f_lock);
745 old = file->f_wb_err; 731 old = file->f_wb_err;
746 err = errseq_check_and_advance 732 err = errseq_check_and_advance(&mapping->wb_err,
747 733 &file->f_wb_err);
748 trace_file_check_and_advance_w 734 trace_file_check_and_advance_wb_err(file, old);
749 spin_unlock(&file->f_lock); 735 spin_unlock(&file->f_lock);
750 } 736 }
751 737
752 /* 738 /*
753 * We're mostly using this function as 739 * We're mostly using this function as a drop in replacement for
754 * filemap_check_errors. Clear AS_EIO/ 740 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
755 * that the legacy code would have had 741 * that the legacy code would have had on these flags.
756 */ 742 */
757 clear_bit(AS_EIO, &mapping->flags); 743 clear_bit(AS_EIO, &mapping->flags);
758 clear_bit(AS_ENOSPC, &mapping->flags); 744 clear_bit(AS_ENOSPC, &mapping->flags);
759 return err; 745 return err;
760 } 746 }
761 EXPORT_SYMBOL(file_check_and_advance_wb_err); 747 EXPORT_SYMBOL(file_check_and_advance_wb_err);
762 748
763 /** 749 /**
764 * file_write_and_wait_range - write out & wai 750 * file_write_and_wait_range - write out & wait on a file range
765 * @file: file pointing to address_space 751 * @file: file pointing to address_space with pages
766 * @lstart: offset in bytes where the rang 752 * @lstart: offset in bytes where the range starts
767 * @lend: offset in bytes where the rang 753 * @lend: offset in bytes where the range ends (inclusive)
768 * 754 *
769 * Write out and wait upon file offsets lstart 755 * Write out and wait upon file offsets lstart->lend, inclusive.
770 * 756 *
771 * Note that @lend is inclusive (describes the 757 * Note that @lend is inclusive (describes the last byte to be written) so
772 * that this function can be used to write to 758 * that this function can be used to write to the very end-of-file (end = -1).
773 * 759 *
774 * After writing out and waiting on the data, 760 * After writing out and waiting on the data, we check and advance the
775 * f_wb_err cursor to the latest value, and re 761 * f_wb_err cursor to the latest value, and return any errors detected there.
776 * 762 *
777 * Return: %0 on success, negative error code 763 * Return: %0 on success, negative error code otherwise.
778 */ 764 */
779 int file_write_and_wait_range(struct file *fil 765 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
780 { 766 {
781 int err = 0, err2; 767 int err = 0, err2;
782 struct address_space *mapping = file-> 768 struct address_space *mapping = file->f_mapping;
783 769
784 if (lend < lstart) <<
785 return 0; <<
786 <<
787 if (mapping_needs_writeback(mapping)) 770 if (mapping_needs_writeback(mapping)) {
788 err = __filemap_fdatawrite_ran 771 err = __filemap_fdatawrite_range(mapping, lstart, lend,
789 772 WB_SYNC_ALL);
790 /* See comment of filemap_writ 773 /* See comment of filemap_write_and_wait() */
791 if (err != -EIO) 774 if (err != -EIO)
792 __filemap_fdatawait_ra 775 __filemap_fdatawait_range(mapping, lstart, lend);
793 } 776 }
794 err2 = file_check_and_advance_wb_err(f 777 err2 = file_check_and_advance_wb_err(file);
795 if (!err) 778 if (!err)
796 err = err2; 779 err = err2;
797 return err; 780 return err;
798 } 781 }
799 EXPORT_SYMBOL(file_write_and_wait_range); 782 EXPORT_SYMBOL(file_write_and_wait_range);
800 783
801 /** 784 /**
802 * replace_page_cache_folio - replace a pageca !! 785 * replace_page_cache_page - replace a pagecache page with a new one
803 * @old: folio to be replaced !! 786 * @old: page to be replaced
804 * @new: folio to replace with !! 787 * @new: page to replace with
805 * !! 788 * @gfp_mask: allocation mode
806 * This function replaces a folio in the pagec !! 789 *
807 * success it acquires the pagecache reference !! 790 * This function replaces a page in the pagecache with a new one. On
808 * drops it for the old folio. Both the old a !! 791 * success it acquires the pagecache reference for the new page and
809 * locked. This function does not add the new !! 792 * drops it for the old page. Both the old and new pages must be
>> 793 * locked. This function does not add the new page to the LRU, the
810 * caller must do that. 794 * caller must do that.
811 * 795 *
812 * The remove + add is atomic. This function 796 * The remove + add is atomic. This function cannot fail.
>> 797 *
>> 798 * Return: %0
813 */ 799 */
814 void replace_page_cache_folio(struct folio *ol !! 800 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
815 { 801 {
816 struct address_space *mapping = old->m 802 struct address_space *mapping = old->mapping;
817 void (*free_folio)(struct folio *) = m !! 803 void (*freepage)(struct page *) = mapping->a_ops->freepage;
818 pgoff_t offset = old->index; 804 pgoff_t offset = old->index;
819 XA_STATE(xas, &mapping->i_pages, offse 805 XA_STATE(xas, &mapping->i_pages, offset);
>> 806 unsigned long flags;
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 xas_lock_irqsave(&xas, flags);
830 <<
831 xas_lock_irq(&xas); <<
832 xas_store(&xas, new); 817 xas_store(&xas, new);
833 818
834 old->mapping = NULL; 819 old->mapping = NULL;
835 /* hugetlb pages do not participate in 820 /* hugetlb pages do not participate in page cache accounting. */
836 if (!folio_test_hugetlb(old)) !! 821 if (!PageHuge(old))
837 __lruvec_stat_sub_folio(old, N !! 822 __dec_node_page_state(new, NR_FILE_PAGES);
838 if (!folio_test_hugetlb(new)) !! 823 if (!PageHuge(new))
839 __lruvec_stat_add_folio(new, N !! 824 __inc_node_page_state(new, NR_FILE_PAGES);
840 if (folio_test_swapbacked(old)) !! 825 if (PageSwapBacked(old))
841 __lruvec_stat_sub_folio(old, N !! 826 __dec_node_page_state(new, NR_SHMEM);
842 if (folio_test_swapbacked(new)) !! 827 if (PageSwapBacked(new))
843 __lruvec_stat_add_folio(new, N !! 828 __inc_node_page_state(new, NR_SHMEM);
844 xas_unlock_irq(&xas); !! 829 xas_unlock_irqrestore(&xas, flags);
845 if (free_folio) !! 830 mem_cgroup_migrate(old, new);
846 free_folio(old); !! 831 if (freepage)
847 folio_put(old); !! 832 freepage(old);
>> 833 put_page(old);
>> 834
>> 835 return 0;
848 } 836 }
849 EXPORT_SYMBOL_GPL(replace_page_cache_folio); !! 837 EXPORT_SYMBOL_GPL(replace_page_cache_page);
850 838
851 noinline int __filemap_add_folio(struct addres !! 839 static int __add_to_page_cache_locked(struct page *page,
852 struct folio *folio, pgoff_t i !! 840 struct address_space *mapping,
>> 841 pgoff_t offset, gfp_t gfp_mask,
>> 842 void **shadowp)
853 { 843 {
854 XA_STATE(xas, &mapping->i_pages, index !! 844 XA_STATE(xas, &mapping->i_pages, offset);
855 void *alloced_shadow = NULL; !! 845 int huge = PageHuge(page);
856 int alloced_order = 0; !! 846 struct mem_cgroup *memcg;
857 bool huge; !! 847 int error;
858 long nr; !! 848 void *old;
859 !! 849
860 VM_BUG_ON_FOLIO(!folio_test_locked(fol !! 850 VM_BUG_ON_PAGE(!PageLocked(page), page);
861 VM_BUG_ON_FOLIO(folio_test_swapbacked( !! 851 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
862 VM_BUG_ON_FOLIO(folio_order(folio) < m <<
863 folio); <<
864 mapping_set_update(&xas, mapping); 852 mapping_set_update(&xas, mapping);
865 853
866 VM_BUG_ON_FOLIO(index & (folio_nr_page !! 854 if (!huge) {
867 xas_set_order(&xas, index, folio_order !! 855 error = mem_cgroup_try_charge(page, current->mm,
868 huge = folio_test_hugetlb(folio); !! 856 gfp_mask, &memcg, false);
869 nr = folio_nr_pages(folio); !! 857 if (error)
870 !! 858 return error;
871 gfp &= GFP_RECLAIM_MASK; !! 859 }
872 folio_ref_add(folio, nr); <<
873 folio->mapping = mapping; <<
874 folio->index = xas.xa_index; <<
875 860
876 for (;;) { !! 861 get_page(page);
877 int order = -1, split_order = !! 862 page->mapping = mapping;
878 void *entry, *old = NULL; !! 863 page->index = offset;
879 864
>> 865 do {
880 xas_lock_irq(&xas); 866 xas_lock_irq(&xas);
881 xas_for_each_conflict(&xas, en !! 867 old = xas_load(&xas);
882 old = entry; !! 868 if (old && !xa_is_value(old))
883 if (!xa_is_value(entry !! 869 xas_set_err(&xas, -EEXIST);
884 xas_set_err(&x !! 870 xas_store(&xas, page);
885 goto unlock; !! 871 if (xas_error(&xas))
886 } !! 872 goto unlock;
887 /* <<
888 * If a larger entry e <<
889 * it will be the firs <<
890 */ <<
891 if (order == -1) <<
892 order = xas_ge <<
893 } <<
894 873
895 /* entry may have changed befo !! 874 if (xa_is_value(old)) {
896 if (alloced_order && (old != a !! 875 mapping->nrexceptional--;
897 xas_destroy(&xas); <<
898 alloced_order = 0; <<
899 } <<
900 <<
901 if (old) { <<
902 if (order > 0 && order <<
903 /* How to hand <<
904 BUG_ON(shmem_m <<
905 if (!alloced_o <<
906 split_ <<
907 goto u <<
908 } <<
909 xas_split(&xas <<
910 xas_reset(&xas <<
911 } <<
912 if (shadowp) 876 if (shadowp)
913 *shadowp = old 877 *shadowp = old;
914 } 878 }
915 !! 879 mapping->nrpages++;
916 xas_store(&xas, folio); <<
917 if (xas_error(&xas)) <<
918 goto unlock; <<
919 <<
920 mapping->nrpages += nr; <<
921 880
922 /* hugetlb pages do not partic 881 /* hugetlb pages do not participate in page cache accounting */
923 if (!huge) { !! 882 if (!huge)
924 __lruvec_stat_mod_foli !! 883 __inc_node_page_state(page, NR_FILE_PAGES);
925 if (folio_test_pmd_map <<
926 __lruvec_stat_ <<
927 <<
928 } <<
929 <<
930 unlock: 884 unlock:
931 xas_unlock_irq(&xas); 885 xas_unlock_irq(&xas);
932 !! 886 } while (xas_nomem(&xas, gfp_mask & GFP_RECLAIM_MASK));
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 } <<
947 887
948 if (xas_error(&xas)) 888 if (xas_error(&xas))
949 goto error; 889 goto error;
950 890
951 trace_mm_filemap_add_to_page_cache(fol !! 891 if (!huge)
>> 892 mem_cgroup_commit_charge(page, memcg, false, false);
>> 893 trace_mm_filemap_add_to_page_cache(page);
952 return 0; 894 return 0;
953 error: 895 error:
954 folio->mapping = NULL; !! 896 page->mapping = NULL;
955 /* Leave page->index set: truncation r 897 /* Leave page->index set: truncation relies upon it */
956 folio_put_refs(folio, nr); !! 898 if (!huge)
>> 899 mem_cgroup_cancel_charge(page, memcg, false);
>> 900 put_page(page);
957 return xas_error(&xas); 901 return xas_error(&xas);
958 } 902 }
959 ALLOW_ERROR_INJECTION(__filemap_add_folio, ERR <<
960 903
961 int filemap_add_folio(struct address_space *ma !! 904 /**
962 pgoff_t index, !! 905 * add_to_page_cache_locked - add a locked page to the pagecache
>> 906 * @page: page to add
>> 907 * @mapping: the page's address_space
>> 908 * @offset: page index
>> 909 * @gfp_mask: page allocation mode
>> 910 *
>> 911 * This function is used to add a page to the pagecache. It must be locked.
>> 912 * This function does not add the page to the LRU. The caller must do that.
>> 913 *
>> 914 * Return: %0 on success, negative error code otherwise.
>> 915 */
>> 916 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
>> 917 pgoff_t offset, gfp_t gfp_mask)
>> 918 {
>> 919 return __add_to_page_cache_locked(page, mapping, offset,
>> 920 gfp_mask, NULL);
>> 921 }
>> 922 EXPORT_SYMBOL(add_to_page_cache_locked);
>> 923
>> 924 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
>> 925 pgoff_t offset, gfp_t gfp_mask)
963 { 926 {
964 void *shadow = NULL; 927 void *shadow = NULL;
965 int ret; 928 int ret;
966 929
967 ret = mem_cgroup_charge(folio, NULL, g !! 930 __SetPageLocked(page);
968 if (ret) !! 931 ret = __add_to_page_cache_locked(page, mapping, offset,
969 return ret; !! 932 gfp_mask, &shadow);
970 !! 933 if (unlikely(ret))
971 __folio_set_locked(folio); !! 934 __ClearPageLocked(page);
972 ret = __filemap_add_folio(mapping, fol !! 935 else {
973 if (unlikely(ret)) { <<
974 mem_cgroup_uncharge(folio); <<
975 __folio_clear_locked(folio); <<
976 } else { <<
977 /* 936 /*
978 * The folio might have been e !! 937 * The page might have been evicted from cache only
979 * recently, in which case it 938 * recently, in which case it should be activated like
980 * any other repeatedly access !! 939 * any other repeatedly accessed page.
981 * The exception is folios get !! 940 * The exception is pages getting rewritten; evicting other
982 * data from the working set, 941 * data from the working set, only to cache data that will
983 * get overwritten with someth 942 * get overwritten with something else, is a waste of memory.
984 */ 943 */
985 WARN_ON_ONCE(folio_test_active !! 944 WARN_ON_ONCE(PageActive(page));
986 if (!(gfp & __GFP_WRITE) && sh !! 945 if (!(gfp_mask & __GFP_WRITE) && shadow)
987 workingset_refault(fol !! 946 workingset_refault(page, shadow);
988 folio_add_lru(folio); !! 947 lru_cache_add(page);
989 } 948 }
990 return ret; 949 return ret;
991 } 950 }
992 EXPORT_SYMBOL_GPL(filemap_add_folio); !! 951 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
993 952
994 #ifdef CONFIG_NUMA 953 #ifdef CONFIG_NUMA
995 struct folio *filemap_alloc_folio_noprof(gfp_t !! 954 struct page *__page_cache_alloc(gfp_t gfp)
996 { 955 {
997 int n; 956 int n;
998 struct folio *folio; !! 957 struct page *page;
999 958
1000 if (cpuset_do_page_mem_spread()) { 959 if (cpuset_do_page_mem_spread()) {
1001 unsigned int cpuset_mems_cook 960 unsigned int cpuset_mems_cookie;
1002 do { 961 do {
1003 cpuset_mems_cookie = 962 cpuset_mems_cookie = read_mems_allowed_begin();
1004 n = cpuset_mem_spread 963 n = cpuset_mem_spread_node();
1005 folio = __folio_alloc !! 964 page = __alloc_pages_node(n, gfp, 0);
1006 } while (!folio && read_mems_ !! 965 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
1007 966
1008 return folio; !! 967 return page;
1009 } 968 }
1010 return folio_alloc_noprof(gfp, order) !! 969 return alloc_pages(gfp, 0);
1011 } 970 }
1012 EXPORT_SYMBOL(filemap_alloc_folio_noprof); !! 971 EXPORT_SYMBOL(__page_cache_alloc);
1013 #endif 972 #endif
1014 973
1015 /* 974 /*
1016 * filemap_invalidate_lock_two - lock invalid <<
1017 * <<
1018 * Lock exclusively invalidate_lock of any pa <<
1019 * <<
1020 * @mapping1: the first mapping to lock <<
1021 * @mapping2: the second mapping to lock <<
1022 */ <<
1023 void filemap_invalidate_lock_two(struct addre <<
1024 struct addre <<
1025 { <<
1026 if (mapping1 > mapping2) <<
1027 swap(mapping1, mapping2); <<
1028 if (mapping1) <<
1029 down_write(&mapping1->invalid <<
1030 if (mapping2 && mapping1 != mapping2) <<
1031 down_write_nested(&mapping2-> <<
1032 } <<
1033 EXPORT_SYMBOL(filemap_invalidate_lock_two); <<
1034 <<
1035 /* <<
1036 * filemap_invalidate_unlock_two - unlock inv <<
1037 * <<
1038 * Unlock exclusive invalidate_lock of any pa <<
1039 * <<
1040 * @mapping1: the first mapping to unlock <<
1041 * @mapping2: the second mapping to unlock <<
1042 */ <<
1043 void filemap_invalidate_unlock_two(struct add <<
1044 struct add <<
1045 { <<
1046 if (mapping1) <<
1047 up_write(&mapping1->invalidat <<
1048 if (mapping2 && mapping1 != mapping2) <<
1049 up_write(&mapping2->invalidat <<
1050 } <<
1051 EXPORT_SYMBOL(filemap_invalidate_unlock_two); <<
1052 <<
1053 /* <<
1054 * In order to wait for pages to become avail 975 * In order to wait for pages to become available there must be
1055 * waitqueues associated with pages. By using 976 * waitqueues associated with pages. By using a hash table of
1056 * waitqueues where the bucket discipline is 977 * waitqueues where the bucket discipline is to maintain all
1057 * waiters on the same queue and wake all whe 978 * waiters on the same queue and wake all when any of the pages
1058 * become available, and for the woken contex 979 * become available, and for the woken contexts to check to be
1059 * sure the appropriate page became available 980 * sure the appropriate page became available, this saves space
1060 * at a cost of "thundering herd" phenomena d 981 * at a cost of "thundering herd" phenomena during rare hash
1061 * collisions. 982 * collisions.
1062 */ 983 */
1063 #define PAGE_WAIT_TABLE_BITS 8 984 #define PAGE_WAIT_TABLE_BITS 8
1064 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_ 985 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1065 static wait_queue_head_t folio_wait_table[PAG !! 986 static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1066 987
1067 static wait_queue_head_t *folio_waitqueue(str !! 988 static wait_queue_head_t *page_waitqueue(struct page *page)
1068 { 989 {
1069 return &folio_wait_table[hash_ptr(fol !! 990 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
1070 } 991 }
1071 992
1072 void __init pagecache_init(void) 993 void __init pagecache_init(void)
1073 { 994 {
1074 int i; 995 int i;
1075 996
1076 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; 997 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1077 init_waitqueue_head(&folio_wa !! 998 init_waitqueue_head(&page_wait_table[i]);
1078 999
1079 page_writeback_init(); 1000 page_writeback_init();
1080 } 1001 }
1081 1002
1082 /* !! 1003 /* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */
1083 * The page wait code treats the "wait->flags !! 1004 struct wait_page_key {
1084 * we have multiple different kinds of waits, !! 1005 struct page *page;
1085 * one. !! 1006 int bit_nr;
1086 * !! 1007 int page_match;
1087 * We have: !! 1008 };
1088 * !! 1009
1089 * (a) no special bits set: !! 1010 struct wait_page_queue {
1090 * !! 1011 struct page *page;
1091 * We're just waiting for the bit to be !! 1012 int bit_nr;
1092 * calls the wakeup function, we set WQ_ !! 1013 wait_queue_entry_t wait;
1093 * and remove it from the wait queue. !! 1014 };
1094 * !! 1015
1095 * Simple and straightforward. <<
1096 * <<
1097 * (b) WQ_FLAG_EXCLUSIVE: <<
1098 * <<
1099 * The waiter is waiting to get the lock <<
1100 * be woken up to avoid any thundering h <<
1101 * WQ_FLAG_WOKEN bit, wake it up, and re <<
1102 * <<
1103 * This is the traditional exclusive wai <<
1104 * <<
1105 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM: <<
1106 * <<
1107 * The waiter is waiting to get the bit, <<
1108 * lock to be transferred to it for fair <<
1109 * cannot be taken, we stop walking the <<
1110 * the waiter. <<
1111 * <<
1112 * This is the "fair lock handoff" case, <<
1113 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to <<
1114 * that it now has the lock. <<
1115 */ <<
1116 static int wake_page_function(wait_queue_entr 1016 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1117 { 1017 {
1118 unsigned int flags; <<
1119 struct wait_page_key *key = arg; 1018 struct wait_page_key *key = arg;
1120 struct wait_page_queue *wait_page 1019 struct wait_page_queue *wait_page
1121 = container_of(wait, struct w 1020 = container_of(wait, struct wait_page_queue, wait);
1122 1021
1123 if (!wake_page_match(wait_page, key)) !! 1022 if (wait_page->page != key->page)
1124 return 0; !! 1023 return 0;
>> 1024 key->page_match = 1;
1125 1025
1126 /* !! 1026 if (wait_page->bit_nr != key->bit_nr)
1127 * If it's a lock handoff wait, we ge !! 1027 return 0;
1128 * stop walking (and do not wake it u <<
1129 */ <<
1130 flags = wait->flags; <<
1131 if (flags & WQ_FLAG_EXCLUSIVE) { <<
1132 if (test_bit(key->bit_nr, &ke <<
1133 return -1; <<
1134 if (flags & WQ_FLAG_CUSTOM) { <<
1135 if (test_and_set_bit( <<
1136 return -1; <<
1137 flags |= WQ_FLAG_DONE <<
1138 } <<
1139 } <<
1140 1028
1141 /* 1029 /*
1142 * We are holding the wait-queue lock !! 1030 * Stop walking if it's locked.
1143 * is waiting for this will be checki !! 1031 * Is this safe if put_and_wait_on_page_locked() is in use?
1144 * any locking. !! 1032 * Yes: the waker must hold a reference to this page, and if PG_locked
1145 * !! 1033 * has now already been set by another task, that task must also hold
1146 * So update the flags atomically, an !! 1034 * a reference to the *same usage* of this page; so there is no need
1147 * afterwards to avoid any races. Thi !! 1035 * to walk on to wake even the put_and_wait_on_page_locked() callers.
1148 * with the load-acquire in folio_wai <<
1149 */ 1036 */
1150 smp_store_release(&wait->flags, flags !! 1037 if (test_bit(key->bit_nr, &key->page->flags))
1151 wake_up_state(wait->private, mode); !! 1038 return -1;
1152 1039
1153 /* !! 1040 return autoremove_wake_function(wait, mode, sync, key);
1154 * Ok, we have successfully done what <<
1155 * and we can unconditionally remove <<
1156 * <<
1157 * Note that this pairs with the "fin <<
1158 * waiter, and has to be the absolute <<
1159 * After this list_del_init(&wait->en <<
1160 * might be de-allocated and the proc <<
1161 * exited. <<
1162 */ <<
1163 list_del_init_careful(&wait->entry); <<
1164 return (flags & WQ_FLAG_EXCLUSIVE) != <<
1165 } 1041 }
1166 1042
1167 static void folio_wake_bit(struct folio *foli !! 1043 static void wake_up_page_bit(struct page *page, int bit_nr)
1168 { 1044 {
1169 wait_queue_head_t *q = folio_waitqueu !! 1045 wait_queue_head_t *q = page_waitqueue(page);
1170 struct wait_page_key key; 1046 struct wait_page_key key;
1171 unsigned long flags; 1047 unsigned long flags;
>> 1048 wait_queue_entry_t bookmark;
1172 1049
1173 key.folio = folio; !! 1050 key.page = page;
1174 key.bit_nr = bit_nr; 1051 key.bit_nr = bit_nr;
1175 key.page_match = 0; 1052 key.page_match = 0;
1176 1053
>> 1054 bookmark.flags = 0;
>> 1055 bookmark.private = NULL;
>> 1056 bookmark.func = NULL;
>> 1057 INIT_LIST_HEAD(&bookmark.entry);
>> 1058
1177 spin_lock_irqsave(&q->lock, flags); 1059 spin_lock_irqsave(&q->lock, flags);
1178 __wake_up_locked_key(q, TASK_NORMAL, !! 1060 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
>> 1061
>> 1062 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
>> 1063 /*
>> 1064 * Take a breather from holding the lock,
>> 1065 * allow pages that finish wake up asynchronously
>> 1066 * to acquire the lock and remove themselves
>> 1067 * from wait queue
>> 1068 */
>> 1069 spin_unlock_irqrestore(&q->lock, flags);
>> 1070 cpu_relax();
>> 1071 spin_lock_irqsave(&q->lock, flags);
>> 1072 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
>> 1073 }
1179 1074
1180 /* 1075 /*
1181 * It's possible to miss clearing wai !! 1076 * It is possible for other pages to have collided on the waitqueue
1182 * waiters, but the hashed waitqueue !! 1077 * hash, so in that case check for a page match. That prevents a long-
1183 * That's okay, it's a rare case. The !! 1078 * term waiter
1184 * 1079 *
1185 * Note that, depending on the page p !! 1080 * It is still possible to miss a case here, when we woke page waiters
1186 * other), the flag may be cleared in !! 1081 * and removed them from the waitqueue, but there are still other
1187 * but that is not required for corre !! 1082 * page waiters.
1188 */ !! 1083 */
1189 if (!waitqueue_active(q) || !key.page !! 1084 if (!waitqueue_active(q) || !key.page_match) {
1190 folio_clear_waiters(folio); !! 1085 ClearPageWaiters(page);
1191 !! 1086 /*
>> 1087 * It's possible to miss clearing Waiters here, when we woke
>> 1088 * our page waiters, but the hashed waitqueue has waiters for
>> 1089 * other pages on it.
>> 1090 *
>> 1091 * That's okay, it's a rare case. The next waker will clear it.
>> 1092 */
>> 1093 }
1192 spin_unlock_irqrestore(&q->lock, flag 1094 spin_unlock_irqrestore(&q->lock, flags);
1193 } 1095 }
1194 1096
>> 1097 static void wake_up_page(struct page *page, int bit)
>> 1098 {
>> 1099 if (!PageWaiters(page))
>> 1100 return;
>> 1101 wake_up_page_bit(page, bit);
>> 1102 }
>> 1103
1195 /* 1104 /*
1196 * A choice of three behaviors for folio_wait !! 1105 * A choice of three behaviors for wait_on_page_bit_common():
1197 */ 1106 */
1198 enum behavior { 1107 enum behavior {
1199 EXCLUSIVE, /* Hold ref to page a 1108 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1200 * __folio_lock() wai !! 1109 * __lock_page() waiting on then setting PG_locked.
1201 */ 1110 */
1202 SHARED, /* Hold ref to page a 1111 SHARED, /* Hold ref to page and check the bit when woken, like
1203 * folio_wait_writeba !! 1112 * wait_on_page_writeback() waiting on PG_writeback.
1204 */ 1113 */
1205 DROP, /* Drop ref to page b 1114 DROP, /* Drop ref to page before wait, no check when woken,
1206 * like folio_put_wai !! 1115 * like put_and_wait_on_page_locked() on PG_locked.
1207 */ 1116 */
1208 }; 1117 };
1209 1118
1210 /* !! 1119 static inline int wait_on_page_bit_common(wait_queue_head_t *q,
1211 * Attempt to check (or get) the folio flag, !! 1120 struct page *page, int bit_nr, int state, enum behavior behavior)
1212 * if successful. <<
1213 */ <<
1214 static inline bool folio_trylock_flag(struct <<
1215 struc <<
1216 { 1121 {
1217 if (wait->flags & WQ_FLAG_EXCLUSIVE) <<
1218 if (test_and_set_bit(bit_nr, <<
1219 return false; <<
1220 } else if (test_bit(bit_nr, &folio->f <<
1221 return false; <<
1222 <<
1223 wait->flags |= WQ_FLAG_WOKEN | WQ_FLA <<
1224 return true; <<
1225 } <<
1226 <<
1227 /* How many times do we accept lock stealing <<
1228 int sysctl_page_lock_unfairness = 5; <<
1229 <<
1230 static inline int folio_wait_bit_common(struc <<
1231 int state, enum behavior beha <<
1232 { <<
1233 wait_queue_head_t *q = folio_waitqueu <<
1234 int unfairness = sysctl_page_lock_unf <<
1235 struct wait_page_queue wait_page; 1122 struct wait_page_queue wait_page;
1236 wait_queue_entry_t *wait = &wait_page 1123 wait_queue_entry_t *wait = &wait_page.wait;
>> 1124 bool bit_is_set;
1237 bool thrashing = false; 1125 bool thrashing = false;
>> 1126 bool delayacct = false;
1238 unsigned long pflags; 1127 unsigned long pflags;
1239 bool in_thrashing; !! 1128 int ret = 0;
1240 1129
1241 if (bit_nr == PG_locked && 1130 if (bit_nr == PG_locked &&
1242 !folio_test_uptodate(folio) && fo !! 1131 !PageUptodate(page) && PageWorkingset(page)) {
1243 delayacct_thrashing_start(&in !! 1132 if (!PageSwapBacked(page)) {
>> 1133 delayacct_thrashing_start();
>> 1134 delayacct = true;
>> 1135 }
1244 psi_memstall_enter(&pflags); 1136 psi_memstall_enter(&pflags);
1245 thrashing = true; 1137 thrashing = true;
1246 } 1138 }
1247 1139
1248 init_wait(wait); 1140 init_wait(wait);
>> 1141 wait->flags = behavior == EXCLUSIVE ? WQ_FLAG_EXCLUSIVE : 0;
1249 wait->func = wake_page_function; 1142 wait->func = wake_page_function;
1250 wait_page.folio = folio; !! 1143 wait_page.page = page;
1251 wait_page.bit_nr = bit_nr; 1144 wait_page.bit_nr = bit_nr;
1252 1145
1253 repeat: <<
1254 wait->flags = 0; <<
1255 if (behavior == EXCLUSIVE) { <<
1256 wait->flags = WQ_FLAG_EXCLUSI <<
1257 if (--unfairness < 0) <<
1258 wait->flags |= WQ_FLA <<
1259 } <<
1260 <<
1261 /* <<
1262 * Do one last check whether we can g <<
1263 * page bit synchronously. <<
1264 * <<
1265 * Do the folio_set_waiters() marking <<
1266 * to let any waker we _just_ missed <<
1267 * need to wake us up (otherwise they <<
1268 * even go to the slow case that look <<
1269 * page queue), and add ourselves to <<
1270 * queue if we need to sleep. <<
1271 * <<
1272 * This part needs to be done under t <<
1273 * lock to avoid races. <<
1274 */ <<
1275 spin_lock_irq(&q->lock); <<
1276 folio_set_waiters(folio); <<
1277 if (!folio_trylock_flag(folio, bit_nr <<
1278 __add_wait_queue_entry_tail(q <<
1279 spin_unlock_irq(&q->lock); <<
1280 <<
1281 /* <<
1282 * From now on, all the logic will be <<
1283 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE <<
1284 * see whether the page bit testing h <<
1285 * been done by the wake function. <<
1286 * <<
1287 * We can drop our reference to the f <<
1288 */ <<
1289 if (behavior == DROP) <<
1290 folio_put(folio); <<
1291 <<
1292 /* <<
1293 * Note that until the "finish_wait() <<
1294 * we see the WQ_FLAG_WOKEN flag, we <<
1295 * be very careful with the 'wait->fl <<
1296 * we may race with a waker that sets <<
1297 */ <<
1298 for (;;) { 1146 for (;;) {
1299 unsigned int flags; !! 1147 spin_lock_irq(&q->lock);
>> 1148
>> 1149 if (likely(list_empty(&wait->entry))) {
>> 1150 __add_wait_queue_entry_tail(q, wait);
>> 1151 SetPageWaiters(page);
>> 1152 }
1300 1153
1301 set_current_state(state); 1154 set_current_state(state);
1302 1155
1303 /* Loop until we've been woke !! 1156 spin_unlock_irq(&q->lock);
1304 flags = smp_load_acquire(&wai <<
1305 if (!(flags & WQ_FLAG_WOKEN)) <<
1306 if (signal_pending_st <<
1307 break; <<
1308 1157
>> 1158 bit_is_set = test_bit(bit_nr, &page->flags);
>> 1159 if (behavior == DROP)
>> 1160 put_page(page);
>> 1161
>> 1162 if (likely(bit_is_set))
1309 io_schedule(); 1163 io_schedule();
1310 continue; !! 1164
>> 1165 if (behavior == EXCLUSIVE) {
>> 1166 if (!test_and_set_bit_lock(bit_nr, &page->flags))
>> 1167 break;
>> 1168 } else if (behavior == SHARED) {
>> 1169 if (!test_bit(bit_nr, &page->flags))
>> 1170 break;
1311 } 1171 }
1312 1172
1313 /* If we were non-exclusive, !! 1173 if (signal_pending_state(state, current)) {
1314 if (behavior != EXCLUSIVE) !! 1174 ret = -EINTR;
1315 break; 1175 break;
>> 1176 }
1316 1177
1317 /* If the waker got the lock !! 1178 if (behavior == DROP) {
1318 if (flags & WQ_FLAG_DONE) !! 1179 /*
>> 1180 * We can no longer safely access page->flags:
>> 1181 * even if CONFIG_MEMORY_HOTREMOVE is not enabled,
>> 1182 * there is a risk of waiting forever on a page reused
>> 1183 * for something that keeps it locked indefinitely.
>> 1184 * But best check for -EINTR above before breaking.
>> 1185 */
1319 break; 1186 break;
1320 !! 1187 }
1321 /* <<
1322 * Otherwise, if we're gettin <<
1323 * try to get it ourselves. <<
1324 * <<
1325 * And if that fails, we'll h <<
1326 */ <<
1327 if (unlikely(test_and_set_bit <<
1328 goto repeat; <<
1329 <<
1330 wait->flags |= WQ_FLAG_DONE; <<
1331 break; <<
1332 } 1188 }
1333 1189
1334 /* <<
1335 * If a signal happened, this 'finish <<
1336 * waiter from the wait-queues, but t <<
1337 * set. That's ok. The next wakeup wi <<
1338 * to do it here would be difficult a <<
1339 */ <<
1340 finish_wait(q, wait); 1190 finish_wait(q, wait);
1341 1191
1342 if (thrashing) { 1192 if (thrashing) {
1343 delayacct_thrashing_end(&in_t !! 1193 if (delayacct)
>> 1194 delayacct_thrashing_end();
1344 psi_memstall_leave(&pflags); 1195 psi_memstall_leave(&pflags);
1345 } 1196 }
1346 1197
1347 /* 1198 /*
1348 * NOTE! The wait->flags weren't stab !! 1199 * A signal could leave PageWaiters set. Clearing it here if
1349 * 'finish_wait()', and we could have !! 1200 * !waitqueue_active would be possible (by open-coding finish_wait),
1350 * to a signal, and had a wakeup even !! 1201 * but still fail to catch it in the case of wait hash collision. We
1351 * test but before the 'finish_wait() !! 1202 * already can fail to clear wait hash collision cases, so don't
1352 * !! 1203 * bother with signals either.
1353 * So only after the finish_wait() ca <<
1354 * if we got woken up or not, so we c <<
1355 * return value based on that state w <<
1356 * <<
1357 * Also note that WQ_FLAG_WOKEN is su <<
1358 * waiter, but an exclusive one requi <<
1359 */ 1204 */
1360 if (behavior == EXCLUSIVE) <<
1361 return wait->flags & WQ_FLAG_ <<
1362 1205
1363 return wait->flags & WQ_FLAG_WOKEN ? !! 1206 return ret;
1364 } <<
1365 <<
1366 #ifdef CONFIG_MIGRATION <<
1367 /** <<
1368 * migration_entry_wait_on_locked - Wait for <<
1369 * @entry: migration swap entry. <<
1370 * @ptl: already locked ptl. This function wi <<
1371 * <<
1372 * Wait for a migration entry referencing the <<
1373 * equivalent to put_and_wait_on_page_locked( <<
1374 * this can be called without taking a refere <<
1375 * should be called while holding the ptl for <<
1376 * the page. <<
1377 * <<
1378 * Returns after unlocking the ptl. <<
1379 * <<
1380 * This follows the same logic as folio_wait_ <<
1381 * there. <<
1382 */ <<
1383 void migration_entry_wait_on_locked(swp_entry <<
1384 __releases(ptl) <<
1385 { <<
1386 struct wait_page_queue wait_page; <<
1387 wait_queue_entry_t *wait = &wait_page <<
1388 bool thrashing = false; <<
1389 unsigned long pflags; <<
1390 bool in_thrashing; <<
1391 wait_queue_head_t *q; <<
1392 struct folio *folio = pfn_swap_entry_ <<
1393 <<
1394 q = folio_waitqueue(folio); <<
1395 if (!folio_test_uptodate(folio) && fo <<
1396 delayacct_thrashing_start(&in <<
1397 psi_memstall_enter(&pflags); <<
1398 thrashing = true; <<
1399 } <<
1400 <<
1401 init_wait(wait); <<
1402 wait->func = wake_page_function; <<
1403 wait_page.folio = folio; <<
1404 wait_page.bit_nr = PG_locked; <<
1405 wait->flags = 0; <<
1406 <<
1407 spin_lock_irq(&q->lock); <<
1408 folio_set_waiters(folio); <<
1409 if (!folio_trylock_flag(folio, PG_loc <<
1410 __add_wait_queue_entry_tail(q <<
1411 spin_unlock_irq(&q->lock); <<
1412 <<
1413 /* <<
1414 * If a migration entry exists for th <<
1415 * a valid reference to the page, and <<
1416 * migration entry. So the page is va <<
1417 */ <<
1418 spin_unlock(ptl); <<
1419 <<
1420 for (;;) { <<
1421 unsigned int flags; <<
1422 <<
1423 set_current_state(TASK_UNINTE <<
1424 <<
1425 /* Loop until we've been woke <<
1426 flags = smp_load_acquire(&wai <<
1427 if (!(flags & WQ_FLAG_WOKEN)) <<
1428 if (signal_pending_st <<
1429 break; <<
1430 <<
1431 io_schedule(); <<
1432 continue; <<
1433 } <<
1434 break; <<
1435 } <<
1436 <<
1437 finish_wait(q, wait); <<
1438 <<
1439 if (thrashing) { <<
1440 delayacct_thrashing_end(&in_t <<
1441 psi_memstall_leave(&pflags); <<
1442 } <<
1443 } 1207 }
1444 #endif <<
1445 1208
1446 void folio_wait_bit(struct folio *folio, int !! 1209 void wait_on_page_bit(struct page *page, int bit_nr)
1447 { 1210 {
1448 folio_wait_bit_common(folio, bit_nr, !! 1211 wait_queue_head_t *q = page_waitqueue(page);
>> 1212 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1449 } 1213 }
1450 EXPORT_SYMBOL(folio_wait_bit); !! 1214 EXPORT_SYMBOL(wait_on_page_bit);
1451 1215
1452 int folio_wait_bit_killable(struct folio *fol !! 1216 int wait_on_page_bit_killable(struct page *page, int bit_nr)
1453 { 1217 {
1454 return folio_wait_bit_common(folio, b !! 1218 wait_queue_head_t *q = page_waitqueue(page);
>> 1219 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED);
1455 } 1220 }
1456 EXPORT_SYMBOL(folio_wait_bit_killable); !! 1221 EXPORT_SYMBOL(wait_on_page_bit_killable);
1457 1222
1458 /** 1223 /**
1459 * folio_put_wait_locked - Drop a reference a !! 1224 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1460 * @folio: The folio to wait for. !! 1225 * @page: The page to wait for.
1461 * @state: The sleep state (TASK_KILLABLE, TA <<
1462 * 1226 *
1463 * The caller should hold a reference on @fol !! 1227 * The caller should hold a reference on @page. They expect the page to
1464 * become unlocked relatively soon, but do no 1228 * become unlocked relatively soon, but do not wish to hold up migration
1465 * (for example) by holding the reference whi !! 1229 * (for example) by holding the reference while waiting for the page to
1466 * come unlocked. After this function return 1230 * come unlocked. After this function returns, the caller should not
1467 * dereference @folio. !! 1231 * dereference @page.
1468 * <<
1469 * Return: 0 if the folio was unlocked or -EI <<
1470 */ 1232 */
1471 static int folio_put_wait_locked(struct folio !! 1233 void put_and_wait_on_page_locked(struct page *page)
1472 { 1234 {
1473 return folio_wait_bit_common(folio, P !! 1235 wait_queue_head_t *q;
>> 1236
>> 1237 page = compound_head(page);
>> 1238 q = page_waitqueue(page);
>> 1239 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, DROP);
1474 } 1240 }
1475 1241
1476 /** 1242 /**
1477 * folio_add_wait_queue - Add an arbitrary wa !! 1243 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1478 * @folio: Folio defining the wait queue of i !! 1244 * @page: Page defining the wait queue of interest
1479 * @waiter: Waiter to add to the queue 1245 * @waiter: Waiter to add to the queue
1480 * 1246 *
1481 * Add an arbitrary @waiter to the wait queue !! 1247 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1482 */ 1248 */
1483 void folio_add_wait_queue(struct folio *folio !! 1249 void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1484 { 1250 {
1485 wait_queue_head_t *q = folio_waitqueu !! 1251 wait_queue_head_t *q = page_waitqueue(page);
1486 unsigned long flags; 1252 unsigned long flags;
1487 1253
1488 spin_lock_irqsave(&q->lock, flags); 1254 spin_lock_irqsave(&q->lock, flags);
1489 __add_wait_queue_entry_tail(q, waiter 1255 __add_wait_queue_entry_tail(q, waiter);
1490 folio_set_waiters(folio); !! 1256 SetPageWaiters(page);
1491 spin_unlock_irqrestore(&q->lock, flag 1257 spin_unlock_irqrestore(&q->lock, flags);
1492 } 1258 }
1493 EXPORT_SYMBOL_GPL(folio_add_wait_queue); !! 1259 EXPORT_SYMBOL_GPL(add_page_wait_queue);
1494 <<
1495 /** <<
1496 * folio_unlock - Unlock a locked folio. <<
1497 * @folio: The folio. <<
1498 * <<
1499 * Unlocks the folio and wakes up any thread <<
1500 * <<
1501 * Context: May be called from interrupt or p <<
1502 * called from NMI context. <<
1503 */ <<
1504 void folio_unlock(struct folio *folio) <<
1505 { <<
1506 /* Bit 7 allows x86 to check the byte <<
1507 BUILD_BUG_ON(PG_waiters != 7); <<
1508 BUILD_BUG_ON(PG_locked > 7); <<
1509 VM_BUG_ON_FOLIO(!folio_test_locked(fo <<
1510 if (folio_xor_flags_has_waiters(folio <<
1511 folio_wake_bit(folio, PG_lock <<
1512 } <<
1513 EXPORT_SYMBOL(folio_unlock); <<
1514 1260
1515 /** !! 1261 #ifndef clear_bit_unlock_is_negative_byte
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 1262
1545 /** !! 1263 /*
1546 * folio_end_private_2 - Clear PG_private_2 a !! 1264 * PG_waiters is the high bit in the same byte as PG_lock.
1547 * @folio: The folio. <<
1548 * 1265 *
1549 * Clear the PG_private_2 bit on a folio and !! 1266 * On x86 (and on many other architectures), we can clear PG_lock and
1550 * it. The folio reference held for PG_priva !! 1267 * test the sign bit at the same time. But if the architecture does
>> 1268 * not support that special operation, we just do this all by hand
>> 1269 * instead.
1551 * 1270 *
1552 * This is, for example, used when a netfs fo !! 1271 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1553 * disk cache, thereby allowing writes to the !! 1272 * being cleared, but a memory barrier should be unneccssary since it is
1554 * serialised. !! 1273 * in the same byte as PG_locked.
1555 */ 1274 */
1556 void folio_end_private_2(struct folio *folio) !! 1275 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1557 { 1276 {
1558 VM_BUG_ON_FOLIO(!folio_test_private_2 !! 1277 clear_bit_unlock(nr, mem);
1559 clear_bit_unlock(PG_private_2, folio_ !! 1278 /* smp_mb__after_atomic(); */
1560 folio_wake_bit(folio, PG_private_2); !! 1279 return test_bit(PG_waiters, mem);
1561 folio_put(folio); <<
1562 } 1280 }
1563 EXPORT_SYMBOL(folio_end_private_2); !! 1281
>> 1282 #endif
1564 1283
1565 /** 1284 /**
1566 * folio_wait_private_2 - Wait for PG_private !! 1285 * unlock_page - unlock a locked page
1567 * @folio: The folio to wait on. !! 1286 * @page: the page
1568 * 1287 *
1569 * Wait for PG_private_2 to be cleared on a f !! 1288 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
>> 1289 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
>> 1290 * mechanism between PageLocked pages and PageWriteback pages is shared.
>> 1291 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
>> 1292 *
>> 1293 * Note that this depends on PG_waiters being the sign bit in the byte
>> 1294 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
>> 1295 * clear the PG_locked bit and test PG_waiters at the same time fairly
>> 1296 * portably (architectures that do LL/SC can test any bit, while x86 can
>> 1297 * test the sign bit).
1570 */ 1298 */
1571 void folio_wait_private_2(struct folio *folio !! 1299 void unlock_page(struct page *page)
1572 { 1300 {
1573 while (folio_test_private_2(folio)) !! 1301 BUILD_BUG_ON(PG_waiters != 7);
1574 folio_wait_bit(folio, PG_priv !! 1302 page = compound_head(page);
>> 1303 VM_BUG_ON_PAGE(!PageLocked(page), page);
>> 1304 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
>> 1305 wake_up_page_bit(page, PG_locked);
1575 } 1306 }
1576 EXPORT_SYMBOL(folio_wait_private_2); !! 1307 EXPORT_SYMBOL(unlock_page);
1577 1308
1578 /** 1309 /**
1579 * folio_wait_private_2_killable - Wait for P !! 1310 * end_page_writeback - end writeback against a page
1580 * @folio: The folio to wait on. !! 1311 * @page: the page
1581 * <<
1582 * Wait for PG_private_2 to be cleared on a f <<
1583 * received by the calling task. <<
1584 * <<
1585 * Return: <<
1586 * - 0 if successful. <<
1587 * - -EINTR if a fatal signal was encountered <<
1588 */ 1312 */
1589 int folio_wait_private_2_killable(struct foli !! 1313 void end_page_writeback(struct page *page)
1590 { 1314 {
1591 int ret = 0; !! 1315 /*
1592 !! 1316 * TestClearPageReclaim could be used here but it is an atomic
1593 while (folio_test_private_2(folio)) { !! 1317 * operation and overkill in this particular case. Failing to
1594 ret = folio_wait_bit_killable !! 1318 * shuffle a page marked for immediate reclaim is too mild to
1595 if (ret < 0) !! 1319 * justify taking an atomic operation penalty at the end of
1596 break; !! 1320 * ever page writeback.
>> 1321 */
>> 1322 if (PageReclaim(page)) {
>> 1323 ClearPageReclaim(page);
>> 1324 rotate_reclaimable_page(page);
1597 } 1325 }
1598 1326
1599 return ret; !! 1327 if (!test_clear_page_writeback(page))
>> 1328 BUG();
>> 1329
>> 1330 smp_mb__after_atomic();
>> 1331 wake_up_page(page, PG_writeback);
1600 } 1332 }
1601 EXPORT_SYMBOL(folio_wait_private_2_killable); !! 1333 EXPORT_SYMBOL(end_page_writeback);
1602 1334
1603 /** !! 1335 /*
1604 * folio_end_writeback - End writeback agains !! 1336 * After completing I/O on a page, call this routine to update the page
1605 * @folio: The folio. !! 1337 * flags appropriately
1606 * <<
1607 * The folio must actually be under writeback <<
1608 * <<
1609 * Context: May be called from process or int <<
1610 */ 1338 */
1611 void folio_end_writeback(struct folio *folio) !! 1339 void page_endio(struct page *page, bool is_write, int err)
1612 { 1340 {
1613 VM_BUG_ON_FOLIO(!folio_test_writeback !! 1341 if (!is_write) {
>> 1342 if (!err) {
>> 1343 SetPageUptodate(page);
>> 1344 } else {
>> 1345 ClearPageUptodate(page);
>> 1346 SetPageError(page);
>> 1347 }
>> 1348 unlock_page(page);
>> 1349 } else {
>> 1350 if (err) {
>> 1351 struct address_space *mapping;
1614 1352
1615 /* !! 1353 SetPageError(page);
1616 * folio_test_clear_reclaim() could b !! 1354 mapping = page_mapping(page);
1617 * atomic operation and overkill in t !! 1355 if (mapping)
1618 * to shuffle a folio marked for imme !! 1356 mapping_set_error(mapping, err);
1619 * a gain to justify taking an atomic !! 1357 }
1620 * end of every folio writeback. !! 1358 end_page_writeback(page);
1621 */ <<
1622 if (folio_test_reclaim(folio)) { <<
1623 folio_clear_reclaim(folio); <<
1624 folio_rotate_reclaimable(foli <<
1625 } 1359 }
1626 <<
1627 /* <<
1628 * Writeback does not hold a folio re <<
1629 * on truncation to wait for the clea <<
1630 * But here we must make sure that th <<
1631 * reused before the folio_wake_bit() <<
1632 */ <<
1633 folio_get(folio); <<
1634 if (__folio_end_writeback(folio)) <<
1635 folio_wake_bit(folio, PG_writ <<
1636 acct_reclaim_writeback(folio); <<
1637 folio_put(folio); <<
1638 } 1360 }
1639 EXPORT_SYMBOL(folio_end_writeback); !! 1361 EXPORT_SYMBOL_GPL(page_endio);
1640 1362
1641 /** 1363 /**
1642 * __folio_lock - Get a lock on the folio, as !! 1364 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1643 * @folio: The folio to lock !! 1365 * @__page: the page to lock
1644 */ 1366 */
1645 void __folio_lock(struct folio *folio) !! 1367 void __lock_page(struct page *__page)
1646 { 1368 {
1647 folio_wait_bit_common(folio, PG_locke !! 1369 struct page *page = compound_head(__page);
>> 1370 wait_queue_head_t *q = page_waitqueue(page);
>> 1371 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE,
1648 EXCLUSIVE); 1372 EXCLUSIVE);
1649 } 1373 }
1650 EXPORT_SYMBOL(__folio_lock); !! 1374 EXPORT_SYMBOL(__lock_page);
1651 1375
1652 int __folio_lock_killable(struct folio *folio !! 1376 int __lock_page_killable(struct page *__page)
1653 { 1377 {
1654 return folio_wait_bit_common(folio, P !! 1378 struct page *page = compound_head(__page);
>> 1379 wait_queue_head_t *q = page_waitqueue(page);
>> 1380 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE,
1655 EXCLU 1381 EXCLUSIVE);
1656 } 1382 }
1657 EXPORT_SYMBOL_GPL(__folio_lock_killable); !! 1383 EXPORT_SYMBOL_GPL(__lock_page_killable);
1658 <<
1659 static int __folio_lock_async(struct folio *f <<
1660 { <<
1661 struct wait_queue_head *q = folio_wai <<
1662 int ret; <<
1663 <<
1664 wait->folio = folio; <<
1665 wait->bit_nr = PG_locked; <<
1666 <<
1667 spin_lock_irq(&q->lock); <<
1668 __add_wait_queue_entry_tail(q, &wait- <<
1669 folio_set_waiters(folio); <<
1670 ret = !folio_trylock(folio); <<
1671 /* <<
1672 * If we were successful now, we know <<
1673 * waitqueue as we're still under the <<
1674 * safe to remove and return success, <<
1675 * isn't going to trigger. <<
1676 */ <<
1677 if (!ret) <<
1678 __remove_wait_queue(q, &wait- <<
1679 else <<
1680 ret = -EIOCBQUEUED; <<
1681 spin_unlock_irq(&q->lock); <<
1682 return ret; <<
1683 } <<
1684 1384
1685 /* 1385 /*
1686 * Return values: 1386 * Return values:
1687 * 0 - folio is locked. !! 1387 * 1 - page is locked; mmap_sem is still held.
1688 * non-zero - folio is not locked. !! 1388 * 0 - page is not locked.
1689 * mmap_lock or per-VMA lock has been rel !! 1389 * mmap_sem has been released (up_read()), unless flags had both
1690 * vma_end_read()), unless flags had both !! 1390 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1691 * FAULT_FLAG_RETRY_NOWAIT set, in which !! 1391 * which case mmap_sem is still held.
1692 * !! 1392 *
1693 * If neither ALLOW_RETRY nor KILLABLE are se !! 1393 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1694 * with the folio locked and the mmap_lock/pe !! 1394 * with the page locked and the mmap_sem unperturbed.
1695 */ 1395 */
1696 vm_fault_t __folio_lock_or_retry(struct folio !! 1396 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
>> 1397 unsigned int flags)
1697 { 1398 {
1698 unsigned int flags = vmf->flags; !! 1399 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1699 <<
1700 if (fault_flag_allow_retry_first(flag <<
1701 /* 1400 /*
1702 * CAUTION! In this case, mma !! 1401 * CAUTION! In this case, mmap_sem is not released
1703 * released even though retur !! 1402 * even though return 0.
1704 */ 1403 */
1705 if (flags & FAULT_FLAG_RETRY_ 1404 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1706 return VM_FAULT_RETRY !! 1405 return 0;
1707 1406
1708 release_fault_lock(vmf); !! 1407 up_read(&mm->mmap_sem);
1709 if (flags & FAULT_FLAG_KILLAB 1408 if (flags & FAULT_FLAG_KILLABLE)
1710 folio_wait_locked_kil !! 1409 wait_on_page_locked_killable(page);
1711 else 1410 else
1712 folio_wait_locked(fol !! 1411 wait_on_page_locked(page);
1713 return VM_FAULT_RETRY; !! 1412 return 0;
1714 } <<
1715 if (flags & FAULT_FLAG_KILLABLE) { <<
1716 bool ret; <<
1717 <<
1718 ret = __folio_lock_killable(f <<
1719 if (ret) { <<
1720 release_fault_lock(vm <<
1721 return VM_FAULT_RETRY <<
1722 } <<
1723 } else { 1413 } else {
1724 __folio_lock(folio); !! 1414 if (flags & FAULT_FLAG_KILLABLE) {
1725 } !! 1415 int ret;
1726 1416
1727 return 0; !! 1417 ret = __lock_page_killable(page);
>> 1418 if (ret) {
>> 1419 up_read(&mm->mmap_sem);
>> 1420 return 0;
>> 1421 }
>> 1422 } else
>> 1423 __lock_page(page);
>> 1424 return 1;
>> 1425 }
1728 } 1426 }
1729 1427
1730 /** 1428 /**
1731 * page_cache_next_miss() - Find the next gap 1429 * page_cache_next_miss() - Find the next gap in the page cache.
1732 * @mapping: Mapping. 1430 * @mapping: Mapping.
1733 * @index: Index. 1431 * @index: Index.
1734 * @max_scan: Maximum range to search. 1432 * @max_scan: Maximum range to search.
1735 * 1433 *
1736 * Search the range [index, min(index + max_s 1434 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1737 * gap with the lowest index. 1435 * gap with the lowest index.
1738 * 1436 *
1739 * This function may be called under the rcu_ 1437 * This function may be called under the rcu_read_lock. However, this will
1740 * not atomically search a snapshot of the ca 1438 * not atomically search a snapshot of the cache at a single point in time.
1741 * For example, if a gap is created at index 1439 * For example, if a gap is created at index 5, then subsequently a gap is
1742 * created at index 10, page_cache_next_miss 1440 * created at index 10, page_cache_next_miss covering both indices may
1743 * return 10 if called under the rcu_read_loc 1441 * return 10 if called under the rcu_read_lock.
1744 * 1442 *
1745 * Return: The index of the gap if found, oth 1443 * Return: The index of the gap if found, otherwise an index outside the
1746 * range specified (in which case 'return - i 1444 * range specified (in which case 'return - index >= max_scan' will be true).
1747 * In the rare case of index wrap-around, 0 w 1445 * In the rare case of index wrap-around, 0 will be returned.
1748 */ 1446 */
1749 pgoff_t page_cache_next_miss(struct address_s 1447 pgoff_t page_cache_next_miss(struct address_space *mapping,
1750 pgoff_t index, u 1448 pgoff_t index, unsigned long max_scan)
1751 { 1449 {
1752 XA_STATE(xas, &mapping->i_pages, inde 1450 XA_STATE(xas, &mapping->i_pages, index);
1753 1451
1754 while (max_scan--) { 1452 while (max_scan--) {
1755 void *entry = xas_next(&xas); 1453 void *entry = xas_next(&xas);
1756 if (!entry || xa_is_value(ent 1454 if (!entry || xa_is_value(entry))
1757 return xas.xa_index; !! 1455 break;
1758 if (xas.xa_index == 0) 1456 if (xas.xa_index == 0)
1759 return 0; !! 1457 break;
1760 } 1458 }
1761 1459
1762 return index + max_scan; !! 1460 return xas.xa_index;
1763 } 1461 }
1764 EXPORT_SYMBOL(page_cache_next_miss); 1462 EXPORT_SYMBOL(page_cache_next_miss);
1765 1463
1766 /** 1464 /**
1767 * page_cache_prev_miss() - Find the previous !! 1465 * page_cache_prev_miss() - Find the next gap in the page cache.
1768 * @mapping: Mapping. 1466 * @mapping: Mapping.
1769 * @index: Index. 1467 * @index: Index.
1770 * @max_scan: Maximum range to search. 1468 * @max_scan: Maximum range to search.
1771 * 1469 *
1772 * Search the range [max(index - max_scan + 1 1470 * Search the range [max(index - max_scan + 1, 0), index] for the
1773 * gap with the highest index. 1471 * gap with the highest index.
1774 * 1472 *
1775 * This function may be called under the rcu_ 1473 * This function may be called under the rcu_read_lock. However, this will
1776 * not atomically search a snapshot of the ca 1474 * not atomically search a snapshot of the cache at a single point in time.
1777 * For example, if a gap is created at index 1475 * For example, if a gap is created at index 10, then subsequently a gap is
1778 * created at index 5, page_cache_prev_miss() 1476 * created at index 5, page_cache_prev_miss() covering both indices may
1779 * return 5 if called under the rcu_read_lock 1477 * return 5 if called under the rcu_read_lock.
1780 * 1478 *
1781 * Return: The index of the gap if found, oth 1479 * Return: The index of the gap if found, otherwise an index outside the
1782 * range specified (in which case 'index - re 1480 * range specified (in which case 'index - return >= max_scan' will be true).
1783 * In the rare case of wrap-around, ULONG_MAX 1481 * In the rare case of wrap-around, ULONG_MAX will be returned.
1784 */ 1482 */
1785 pgoff_t page_cache_prev_miss(struct address_s 1483 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1786 pgoff_t index, u 1484 pgoff_t index, unsigned long max_scan)
1787 { 1485 {
1788 XA_STATE(xas, &mapping->i_pages, inde 1486 XA_STATE(xas, &mapping->i_pages, index);
1789 1487
1790 while (max_scan--) { 1488 while (max_scan--) {
1791 void *entry = xas_prev(&xas); 1489 void *entry = xas_prev(&xas);
1792 if (!entry || xa_is_value(ent 1490 if (!entry || xa_is_value(entry))
1793 break; 1491 break;
1794 if (xas.xa_index == ULONG_MAX 1492 if (xas.xa_index == ULONG_MAX)
1795 break; 1493 break;
1796 } 1494 }
1797 1495
1798 return xas.xa_index; 1496 return xas.xa_index;
1799 } 1497 }
1800 EXPORT_SYMBOL(page_cache_prev_miss); 1498 EXPORT_SYMBOL(page_cache_prev_miss);
1801 1499
1802 /* !! 1500 /**
1803 * Lockless page cache protocol: !! 1501 * find_get_entry - find and get a page cache entry
1804 * On the lookup side: <<
1805 * 1. Load the folio from i_pages <<
1806 * 2. Increment the refcount if it's not zero <<
1807 * 3. If the folio is not found by xas_reload <<
1808 * <<
1809 * On the removal side: <<
1810 * A. Freeze the page (by zeroing the refcoun <<
1811 * B. Remove the page from i_pages <<
1812 * C. Return the page to the page allocator <<
1813 * <<
1814 * This means that any page may have its refe <<
1815 * increased by a speculative page cache (or <<
1816 * be allocated by another user before the RC <<
1817 * Because the refcount temporarily acquired <<
1818 * last refcount on the page, any page alloca <<
1819 * folio_put(). <<
1820 */ <<
1821 <<
1822 /* <<
1823 * filemap_get_entry - Get a page cache entry <<
1824 * @mapping: the address_space to search 1502 * @mapping: the address_space to search
1825 * @index: The page cache index. !! 1503 * @offset: the page cache index
1826 * 1504 *
1827 * Looks up the page cache entry at @mapping !! 1505 * Looks up the page cache slot at @mapping & @offset. If there is a
1828 * it is returned with an increased refcount. !! 1506 * page cache page, it is returned with an increased refcount.
1829 * of a previously evicted folio, or a swap e <<
1830 * it is returned without further action. <<
1831 * 1507 *
1832 * Return: The folio, swap or shadow entry, % !! 1508 * If the slot holds a shadow entry of a previously evicted page, or a
>> 1509 * swap entry from shmem/tmpfs, it is returned.
>> 1510 *
>> 1511 * Return: the found page or shadow entry, %NULL if nothing is found.
1833 */ 1512 */
1834 void *filemap_get_entry(struct address_space !! 1513 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1835 { 1514 {
1836 XA_STATE(xas, &mapping->i_pages, inde !! 1515 XA_STATE(xas, &mapping->i_pages, offset);
1837 struct folio *folio; !! 1516 struct page *head, *page;
1838 1517
1839 rcu_read_lock(); 1518 rcu_read_lock();
1840 repeat: 1519 repeat:
1841 xas_reset(&xas); 1520 xas_reset(&xas);
1842 folio = xas_load(&xas); !! 1521 page = xas_load(&xas);
1843 if (xas_retry(&xas, folio)) !! 1522 if (xas_retry(&xas, page))
1844 goto repeat; 1523 goto repeat;
1845 /* 1524 /*
1846 * A shadow entry of a recently evict 1525 * A shadow entry of a recently evicted page, or a swap entry from
1847 * shmem/tmpfs. Return it without at 1526 * shmem/tmpfs. Return it without attempting to raise page count.
1848 */ 1527 */
1849 if (!folio || xa_is_value(folio)) !! 1528 if (!page || xa_is_value(page))
1850 goto out; 1529 goto out;
1851 1530
1852 if (!folio_try_get(folio)) !! 1531 head = compound_head(page);
>> 1532 if (!page_cache_get_speculative(head))
>> 1533 goto repeat;
>> 1534
>> 1535 /* The page was split under us? */
>> 1536 if (compound_head(page) != head) {
>> 1537 put_page(head);
1853 goto repeat; 1538 goto repeat;
>> 1539 }
1854 1540
1855 if (unlikely(folio != xas_reload(&xas !! 1541 /*
1856 folio_put(folio); !! 1542 * Has the page moved?
>> 1543 * This is part of the lockless pagecache protocol. See
>> 1544 * include/linux/pagemap.h for details.
>> 1545 */
>> 1546 if (unlikely(page != xas_reload(&xas))) {
>> 1547 put_page(head);
1857 goto repeat; 1548 goto repeat;
1858 } 1549 }
1859 out: 1550 out:
1860 rcu_read_unlock(); 1551 rcu_read_unlock();
1861 1552
1862 return folio; !! 1553 return page;
1863 } 1554 }
>> 1555 EXPORT_SYMBOL(find_get_entry);
1864 1556
1865 /** 1557 /**
1866 * __filemap_get_folio - Find and get a refer !! 1558 * find_lock_entry - locate, pin and lock a page cache entry
1867 * @mapping: The address_space to search. !! 1559 * @mapping: the address_space to search
1868 * @index: The page index. !! 1560 * @offset: the page cache index
1869 * @fgp_flags: %FGP flags modify how the foli <<
1870 * @gfp: Memory allocation flags to use if %F <<
1871 * 1561 *
1872 * Looks up the page cache entry at @mapping !! 1562 * Looks up the page cache slot at @mapping & @offset. If there is a
>> 1563 * page cache page, it is returned locked and with an increased
>> 1564 * refcount.
1873 * 1565 *
1874 * If %FGP_LOCK or %FGP_CREAT are specified t !! 1566 * If the slot holds a shadow entry of a previously evicted page, or a
1875 * if the %GFP flags specified for %FGP_CREAT !! 1567 * swap entry from shmem/tmpfs, it is returned.
1876 * 1568 *
1877 * If this function returns a folio, it is re !! 1569 * find_lock_entry() may sleep.
1878 * 1570 *
1879 * Return: The found folio or an ERR_PTR() ot !! 1571 * Return: the found page or shadow entry, %NULL if nothing is found.
1880 */ 1572 */
1881 struct folio *__filemap_get_folio(struct addr !! 1573 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1882 fgf_t fgp_flags, gfp_t gfp) <<
1883 { 1574 {
1884 struct folio *folio; !! 1575 struct page *page;
1885 1576
1886 repeat: 1577 repeat:
1887 folio = filemap_get_entry(mapping, in !! 1578 page = find_get_entry(mapping, offset);
1888 if (xa_is_value(folio)) !! 1579 if (page && !xa_is_value(page)) {
1889 folio = NULL; !! 1580 lock_page(page);
1890 if (!folio) !! 1581 /* Has the page been truncated? */
>> 1582 if (unlikely(page_mapping(page) != mapping)) {
>> 1583 unlock_page(page);
>> 1584 put_page(page);
>> 1585 goto repeat;
>> 1586 }
>> 1587 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
>> 1588 }
>> 1589 return page;
>> 1590 }
>> 1591 EXPORT_SYMBOL(find_lock_entry);
>> 1592
>> 1593 /**
>> 1594 * pagecache_get_page - find and get a page reference
>> 1595 * @mapping: the address_space to search
>> 1596 * @offset: the page index
>> 1597 * @fgp_flags: PCG flags
>> 1598 * @gfp_mask: gfp mask to use for the page cache data page allocation
>> 1599 *
>> 1600 * Looks up the page cache slot at @mapping & @offset.
>> 1601 *
>> 1602 * PCG flags modify how the page is returned.
>> 1603 *
>> 1604 * @fgp_flags can be:
>> 1605 *
>> 1606 * - FGP_ACCESSED: the page will be marked accessed
>> 1607 * - FGP_LOCK: Page is return locked
>> 1608 * - FGP_CREAT: If page is not present then a new page is allocated using
>> 1609 * @gfp_mask and added to the page cache and the VM's LRU
>> 1610 * list. The page is returned locked and with an increased
>> 1611 * refcount.
>> 1612 * - FGP_FOR_MMAP: Similar to FGP_CREAT, only we want to allow the caller to do
>> 1613 * its own locking dance if the page is already in cache, or unlock the page
>> 1614 * before returning if we had to add the page to pagecache.
>> 1615 *
>> 1616 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
>> 1617 * if the GFP flags specified for FGP_CREAT are atomic.
>> 1618 *
>> 1619 * If there is a page cache page, it is returned with an increased refcount.
>> 1620 *
>> 1621 * Return: the found page or %NULL otherwise.
>> 1622 */
>> 1623 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
>> 1624 int fgp_flags, gfp_t gfp_mask)
>> 1625 {
>> 1626 struct page *page;
>> 1627
>> 1628 repeat:
>> 1629 page = find_get_entry(mapping, offset);
>> 1630 if (xa_is_value(page))
>> 1631 page = NULL;
>> 1632 if (!page)
1891 goto no_page; 1633 goto no_page;
1892 1634
1893 if (fgp_flags & FGP_LOCK) { 1635 if (fgp_flags & FGP_LOCK) {
1894 if (fgp_flags & FGP_NOWAIT) { 1636 if (fgp_flags & FGP_NOWAIT) {
1895 if (!folio_trylock(fo !! 1637 if (!trylock_page(page)) {
1896 folio_put(fol !! 1638 put_page(page);
1897 return ERR_PT !! 1639 return NULL;
1898 } 1640 }
1899 } else { 1641 } else {
1900 folio_lock(folio); !! 1642 lock_page(page);
1901 } 1643 }
1902 1644
1903 /* Has the page been truncate 1645 /* Has the page been truncated? */
1904 if (unlikely(folio->mapping ! !! 1646 if (unlikely(page->mapping != mapping)) {
1905 folio_unlock(folio); !! 1647 unlock_page(page);
1906 folio_put(folio); !! 1648 put_page(page);
1907 goto repeat; 1649 goto repeat;
1908 } 1650 }
1909 VM_BUG_ON_FOLIO(!folio_contai !! 1651 VM_BUG_ON_PAGE(page->index != offset, page);
1910 } 1652 }
1911 1653
1912 if (fgp_flags & FGP_ACCESSED) 1654 if (fgp_flags & FGP_ACCESSED)
1913 folio_mark_accessed(folio); !! 1655 mark_page_accessed(page);
1914 else if (fgp_flags & FGP_WRITE) { <<
1915 /* Clear idle flag for buffer <<
1916 if (folio_test_idle(folio)) <<
1917 folio_clear_idle(foli <<
1918 } <<
1919 1656
1920 if (fgp_flags & FGP_STABLE) <<
1921 folio_wait_stable(folio); <<
1922 no_page: 1657 no_page:
1923 if (!folio && (fgp_flags & FGP_CREAT) !! 1658 if (!page && (fgp_flags & FGP_CREAT)) {
1924 unsigned int min_order = mapp <<
1925 unsigned int order = max(min_ <<
1926 int err; 1659 int err;
1927 index = mapping_align_index(m !! 1660 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1928 !! 1661 gfp_mask |= __GFP_WRITE;
1929 if ((fgp_flags & FGP_WRITE) & <<
1930 gfp |= __GFP_WRITE; <<
1931 if (fgp_flags & FGP_NOFS) 1662 if (fgp_flags & FGP_NOFS)
1932 gfp &= ~__GFP_FS; !! 1663 gfp_mask &= ~__GFP_FS;
1933 if (fgp_flags & FGP_NOWAIT) { <<
1934 gfp &= ~GFP_KERNEL; <<
1935 gfp |= GFP_NOWAIT | _ <<
1936 } <<
1937 if (WARN_ON_ONCE(!(fgp_flags <<
1938 fgp_flags |= FGP_LOCK <<
1939 1664
1940 if (order > mapping_max_folio !! 1665 page = __page_cache_alloc(gfp_mask);
1941 order = mapping_max_f !! 1666 if (!page)
1942 /* If we're not aligned, allo !! 1667 return NULL;
1943 if (index & ((1UL << order) - <<
1944 order = __ffs(index); <<
1945 <<
1946 do { <<
1947 gfp_t alloc_gfp = gfp <<
1948 1668
1949 err = -ENOMEM; !! 1669 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1950 if (order > min_order !! 1670 fgp_flags |= FGP_LOCK;
1951 alloc_gfp |= <<
1952 folio = filemap_alloc <<
1953 if (!folio) <<
1954 continue; <<
1955 1671
1956 /* Init accessed so a !! 1672 /* Init accessed so avoid atomic mark_page_accessed later */
1957 if (fgp_flags & FGP_A !! 1673 if (fgp_flags & FGP_ACCESSED)
1958 __folio_set_r !! 1674 __SetPageReferenced(page);
1959 1675
1960 err = filemap_add_fol !! 1676 err = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
1961 if (!err) !! 1677 if (unlikely(err)) {
1962 break; !! 1678 put_page(page);
1963 folio_put(folio); !! 1679 page = NULL;
1964 folio = NULL; !! 1680 if (err == -EEXIST)
1965 } while (order-- > min_order) !! 1681 goto repeat;
>> 1682 }
1966 1683
1967 if (err == -EEXIST) <<
1968 goto repeat; <<
1969 if (err) <<
1970 return ERR_PTR(err); <<
1971 /* 1684 /*
1972 * filemap_add_folio locks th !! 1685 * add_to_page_cache_lru locks the page, and for mmap we expect
1973 * we expect an unlocked page !! 1686 * an unlocked page.
1974 */ 1687 */
1975 if (folio && (fgp_flags & FGP !! 1688 if (page && (fgp_flags & FGP_FOR_MMAP))
1976 folio_unlock(folio); !! 1689 unlock_page(page);
1977 } 1690 }
1978 1691
1979 if (!folio) !! 1692 return page;
1980 return ERR_PTR(-ENOENT); <<
1981 return folio; <<
1982 } <<
1983 EXPORT_SYMBOL(__filemap_get_folio); <<
1984 <<
1985 static inline struct folio *find_get_entry(st <<
1986 xa_mark_t mark) <<
1987 { <<
1988 struct folio *folio; <<
1989 <<
1990 retry: <<
1991 if (mark == XA_PRESENT) <<
1992 folio = xas_find(xas, max); <<
1993 else <<
1994 folio = xas_find_marked(xas, <<
1995 <<
1996 if (xas_retry(xas, folio)) <<
1997 goto retry; <<
1998 /* <<
1999 * A shadow entry of a recently evict <<
2000 * entry from shmem/tmpfs or a DAX en <<
2001 * without attempting to raise page c <<
2002 */ <<
2003 if (!folio || xa_is_value(folio)) <<
2004 return folio; <<
2005 <<
2006 if (!folio_try_get(folio)) <<
2007 goto reset; <<
2008 <<
2009 if (unlikely(folio != xas_reload(xas) <<
2010 folio_put(folio); <<
2011 goto reset; <<
2012 } <<
2013 <<
2014 return folio; <<
2015 reset: <<
2016 xas_reset(xas); <<
2017 goto retry; <<
2018 } 1693 }
>> 1694 EXPORT_SYMBOL(pagecache_get_page);
2019 1695
2020 /** 1696 /**
2021 * find_get_entries - gang pagecache lookup 1697 * find_get_entries - gang pagecache lookup
2022 * @mapping: The address_space to search 1698 * @mapping: The address_space to search
2023 * @start: The starting page cache index 1699 * @start: The starting page cache index
2024 * @end: The final page index (inclusi !! 1700 * @nr_entries: The maximum number of entries
2025 * @fbatch: Where the resulting entries a !! 1701 * @entries: Where the resulting entries are placed
2026 * @indices: The cache indices correspondi 1702 * @indices: The cache indices corresponding to the entries in @entries
2027 * 1703 *
2028 * find_get_entries() will search for and ret !! 1704 * find_get_entries() will search for and return a group of up to
2029 * the mapping. The entries are placed in @f !! 1705 * @nr_entries entries in the mapping. The entries are placed at
2030 * takes a reference on any actual folios it !! 1706 * @entries. find_get_entries() takes a reference against any actual
2031 * !! 1707 * pages it returns.
2032 * The entries have ascending indexes. The i !! 1708 *
2033 * due to not-present entries or large folios !! 1709 * The search returns a group of mapping-contiguous page cache entries
>> 1710 * with ascending indexes. There may be holes in the indices due to
>> 1711 * not-present pages.
2034 * 1712 *
2035 * Any shadow entries of evicted folios, or s !! 1713 * Any shadow entries of evicted pages, or swap entries from
2036 * shmem/tmpfs, are included in the returned 1714 * shmem/tmpfs, are included in the returned array.
2037 * 1715 *
2038 * Return: The number of entries which were f !! 1716 * Return: the number of pages and shadow entries which were found.
2039 */ 1717 */
2040 unsigned find_get_entries(struct address_spac !! 1718 unsigned find_get_entries(struct address_space *mapping,
2041 pgoff_t end, struct folio_bat !! 1719 pgoff_t start, unsigned int nr_entries,
>> 1720 struct page **entries, pgoff_t *indices)
2042 { 1721 {
2043 XA_STATE(xas, &mapping->i_pages, *sta !! 1722 XA_STATE(xas, &mapping->i_pages, start);
2044 struct folio *folio; !! 1723 struct page *page;
2045 !! 1724 unsigned int ret = 0;
2046 rcu_read_lock(); <<
2047 while ((folio = find_get_entry(&xas, <<
2048 indices[fbatch->nr] = xas.xa_ <<
2049 if (!folio_batch_add(fbatch, <<
2050 break; <<
2051 } <<
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(); <<
2065 <<
2066 return folio_batch_count(fbatch); <<
2067 } <<
2068 1725
2069 /** !! 1726 if (!nr_entries)
2070 * find_lock_entries - Find a batch of pageca !! 1727 return 0;
2071 * @mapping: The address_space to search. <<
2072 * @start: The starting page cache index <<
2073 * @end: The final page index (inclusi <<
2074 * @fbatch: Where the resulting entries a <<
2075 * @indices: The cache indices of the entr <<
2076 * <<
2077 * find_lock_entries() will return a batch of <<
2078 * Swap, shadow and DAX entries are included. <<
2079 * locked and with an incremented refcount. <<
2080 * by somebody else or under writeback are sk <<
2081 * partially outside the range are not return <<
2082 * <<
2083 * The entries have ascending indexes. The i <<
2084 * due to not-present entries, large folios, <<
2085 * locked or folios under writeback. <<
2086 * <<
2087 * Return: The number of entries which were f <<
2088 */ <<
2089 unsigned find_lock_entries(struct address_spa <<
2090 pgoff_t end, struct folio_bat <<
2091 { <<
2092 XA_STATE(xas, &mapping->i_pages, *sta <<
2093 struct folio *folio; <<
2094 1728
2095 rcu_read_lock(); 1729 rcu_read_lock();
2096 while ((folio = find_get_entry(&xas, !! 1730 xas_for_each(&xas, page, ULONG_MAX) {
2097 unsigned long base; !! 1731 struct page *head;
2098 unsigned long nr; !! 1732 if (xas_retry(&xas, page))
2099 !! 1733 continue;
2100 if (!xa_is_value(folio)) { !! 1734 /*
2101 nr = folio_nr_pages(f !! 1735 * A shadow entry of a recently evicted page, a swap
2102 base = folio->index; !! 1736 * entry from shmem/tmpfs or a DAX entry. Return it
2103 /* Omit large folio w !! 1737 * without attempting to raise page count.
2104 if (base < *start) !! 1738 */
2105 goto put; !! 1739 if (xa_is_value(page))
2106 /* Omit large folio w !! 1740 goto export;
2107 if (base + nr - 1 > e !! 1741
2108 goto put; !! 1742 head = compound_head(page);
2109 if (!folio_trylock(fo !! 1743 if (!page_cache_get_speculative(head))
2110 goto put; !! 1744 goto retry;
2111 if (folio->mapping != <<
2112 folio_test_writeb <<
2113 goto unlock; <<
2114 VM_BUG_ON_FOLIO(!foli <<
2115 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 } <<
2126 1745
2127 /* Update start now so that l !! 1746 /* The page was split under us? */
2128 *start = base + nr; !! 1747 if (compound_head(page) != head)
2129 indices[fbatch->nr] = xas.xa_ !! 1748 goto put_page;
2130 if (!folio_batch_add(fbatch, !! 1749
>> 1750 /* Has the page moved? */
>> 1751 if (unlikely(page != xas_reload(&xas)))
>> 1752 goto put_page;
>> 1753
>> 1754 export:
>> 1755 indices[ret] = xas.xa_index;
>> 1756 entries[ret] = page;
>> 1757 if (++ret == nr_entries)
2131 break; 1758 break;
2132 continue; 1759 continue;
2133 unlock: !! 1760 put_page:
2134 folio_unlock(folio); !! 1761 put_page(head);
2135 put: !! 1762 retry:
2136 folio_put(folio); !! 1763 xas_reset(&xas);
2137 } 1764 }
2138 rcu_read_unlock(); 1765 rcu_read_unlock();
2139 !! 1766 return ret;
2140 return folio_batch_count(fbatch); <<
2141 } <<
2142 <<
2143 /** <<
2144 * filemap_get_folios - Get a batch of folios <<
2145 * @mapping: The address_space to search <<
2146 * @start: The starting page index <<
2147 * @end: The final page index (inclusi <<
2148 * @fbatch: The batch to fill. <<
2149 * <<
2150 * Search for and return a batch of folios in <<
2151 * index @start and up to index @end (inclusi <<
2152 * in @fbatch with an elevated reference coun <<
2153 * <<
2154 * Return: The number of folios which were fo <<
2155 * We also update @start to index the next fo <<
2156 */ <<
2157 unsigned filemap_get_folios(struct address_sp <<
2158 pgoff_t end, struct folio_bat <<
2159 { <<
2160 return filemap_get_folios_tag(mapping <<
2161 } 1767 }
2162 EXPORT_SYMBOL(filemap_get_folios); <<
2163 1768
2164 /** 1769 /**
2165 * filemap_get_folios_contig - Get a batch of !! 1770 * find_get_pages_range - gang pagecache lookup
2166 * @mapping: The address_space to search 1771 * @mapping: The address_space to search
2167 * @start: The starting page index 1772 * @start: The starting page index
2168 * @end: The final page index (inclusi 1773 * @end: The final page index (inclusive)
2169 * @fbatch: The batch to fill !! 1774 * @nr_pages: The maximum number of pages
>> 1775 * @pages: Where the resulting pages are placed
2170 * 1776 *
2171 * filemap_get_folios_contig() works exactly !! 1777 * find_get_pages_range() will search for and return a group of up to @nr_pages
2172 * except the returned folios are guaranteed !! 1778 * pages in the mapping starting at index @start and up to index @end
2173 * not return all contiguous folios if the ba !! 1779 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
2174 * !! 1780 * a reference against the returned pages.
2175 * Return: The number of folios found. !! 1781 *
2176 * Also update @start to be positioned for tr !! 1782 * The search returns a group of mapping-contiguous pages with ascending
2177 */ !! 1783 * indexes. There may be holes in the indices due to not-present pages.
2178 !! 1784 * We also update @start to index the next page for the traversal.
2179 unsigned filemap_get_folios_contig(struct add !! 1785 *
2180 pgoff_t *start, pgoff_t end, !! 1786 * Return: the number of pages which were found. If this number is
>> 1787 * smaller than @nr_pages, the end of specified range has been
>> 1788 * reached.
>> 1789 */
>> 1790 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
>> 1791 pgoff_t end, unsigned int nr_pages,
>> 1792 struct page **pages)
2181 { 1793 {
2182 XA_STATE(xas, &mapping->i_pages, *sta 1794 XA_STATE(xas, &mapping->i_pages, *start);
2183 unsigned long nr; !! 1795 struct page *page;
2184 struct folio *folio; !! 1796 unsigned ret = 0;
2185 1797
2186 rcu_read_lock(); !! 1798 if (unlikely(!nr_pages))
>> 1799 return 0;
2187 1800
2188 for (folio = xas_load(&xas); folio && !! 1801 rcu_read_lock();
2189 folio = xas_next(&xas !! 1802 xas_for_each(&xas, page, end) {
2190 if (xas_retry(&xas, folio)) !! 1803 struct page *head;
>> 1804 if (xas_retry(&xas, page))
>> 1805 continue;
>> 1806 /* Skip over shadow, swap and DAX entries */
>> 1807 if (xa_is_value(page))
2191 continue; 1808 continue;
2192 /* <<
2193 * If the entry has been swap <<
2194 * No current caller is looki <<
2195 */ <<
2196 if (xa_is_value(folio)) <<
2197 goto update_start; <<
2198 <<
2199 /* If we landed in the middle <<
2200 if (xa_is_sibling(folio)) <<
2201 goto update_start; <<
2202 1809
2203 if (!folio_try_get(folio)) !! 1810 head = compound_head(page);
>> 1811 if (!page_cache_get_speculative(head))
2204 goto retry; 1812 goto retry;
2205 1813
2206 if (unlikely(folio != xas_rel !! 1814 /* The page was split under us? */
2207 goto put_folio; !! 1815 if (compound_head(page) != head)
2208 !! 1816 goto put_page;
2209 if (!folio_batch_add(fbatch, !! 1817
2210 nr = folio_nr_pages(f !! 1818 /* Has the page moved? */
2211 *start = folio->index !! 1819 if (unlikely(page != xas_reload(&xas)))
>> 1820 goto put_page;
>> 1821
>> 1822 pages[ret] = page;
>> 1823 if (++ret == nr_pages) {
>> 1824 *start = xas.xa_index + 1;
2212 goto out; 1825 goto out;
2213 } 1826 }
2214 continue; 1827 continue;
2215 put_folio: !! 1828 put_page:
2216 folio_put(folio); !! 1829 put_page(head);
2217 <<
2218 retry: 1830 retry:
2219 xas_reset(&xas); 1831 xas_reset(&xas);
2220 } 1832 }
2221 1833
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(); <<
2231 return folio_batch_count(fbatch); <<
2232 } <<
2233 EXPORT_SYMBOL(filemap_get_folios_contig); <<
2234 <<
2235 /** <<
2236 * filemap_get_folios_tag - Get a batch of fo <<
2237 * @mapping: The address_space to search <<
2238 * @start: The starting page index <<
2239 * @end: The final page index (inclusi <<
2240 * @tag: The tag index <<
2241 * @fbatch: The batch to fill <<
2242 * <<
2243 * The first folio may start before @start; i <<
2244 * @start. The final folio may extend beyond <<
2245 * contain @end. The folios have ascending i <<
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 * <<
2251 * Return: The number of folios found. <<
2252 * Also update @start to index the next folio <<
2253 */ <<
2254 unsigned filemap_get_folios_tag(struct addres <<
2255 pgoff_t end, xa_mark_ <<
2256 { <<
2257 XA_STATE(xas, &mapping->i_pages, *sta <<
2258 struct folio *folio; <<
2259 <<
2260 rcu_read_lock(); <<
2261 while ((folio = find_get_entry(&xas, <<
2262 /* <<
2263 * Shadow entries should neve <<
2264 * is lockless so there is a <<
2265 * a page we saw tagged. Skip <<
2266 */ <<
2267 if (xa_is_value(folio)) <<
2268 continue; <<
2269 if (!folio_batch_add(fbatch, <<
2270 unsigned long nr = fo <<
2271 *start = folio->index <<
2272 goto out; <<
2273 } <<
2274 } <<
2275 /* 1834 /*
2276 * We come here when there is no page 1835 * We come here when there is no page beyond @end. We take care to not
2277 * overflow the index @start as it co 1836 * overflow the index @start as it confuses some of the callers. This
2278 * breaks the iteration when there is 1837 * breaks the iteration when there is a page at index -1 but that is
2279 * already broke anyway. !! 1838 * already broken anyway.
2280 */ 1839 */
2281 if (end == (pgoff_t)-1) 1840 if (end == (pgoff_t)-1)
2282 *start = (pgoff_t)-1; 1841 *start = (pgoff_t)-1;
2283 else 1842 else
2284 *start = end + 1; 1843 *start = end + 1;
2285 out: 1844 out:
2286 rcu_read_unlock(); 1845 rcu_read_unlock();
2287 1846
2288 return folio_batch_count(fbatch); !! 1847 return ret;
2289 } 1848 }
2290 EXPORT_SYMBOL(filemap_get_folios_tag); <<
2291 1849
2292 /* !! 1850 /**
2293 * CD/DVDs are error prone. When a medium err !! 1851 * find_get_pages_contig - gang contiguous pagecache lookup
2294 * a _large_ part of the i/o request. Imagine !! 1852 * @mapping: The address_space to search
2295 * !! 1853 * @index: The starting page index
2296 * ---R_________________________________ !! 1854 * @nr_pages: The maximum number of pages
2297 * ^ reading here !! 1855 * @pages: Where the resulting pages are placed
2298 * <<
2299 * read(R) => miss => readahead(R...B) => med <<
2300 * => failing the whole request => read(R) => <<
2301 * readahead(R+1...B+1) => bang => read(R+2) <<
2302 * readahead(R+3...B+2) => bang => read(R+3) <<
2303 * readahead(R+4...B+3) => bang => read(R+4) <<
2304 * 1856 *
2305 * It is going insane. Fix it by quickly scal !! 1857 * find_get_pages_contig() works exactly like find_get_pages(), except
2306 */ !! 1858 * that the returned number of pages are guaranteed to be contiguous.
2307 static void shrink_readahead_size_eio(struct <<
2308 { <<
2309 ra->ra_pages /= 4; <<
2310 } <<
2311 <<
2312 /* <<
2313 * filemap_get_read_batch - Get a batch of fo <<
2314 * 1859 *
2315 * Get a batch of folios which represent a co !! 1860 * Return: the number of pages which were found.
2316 * the file. No exceptional entries will be <<
2317 * the middle of a folio, the entire folio wi <<
2318 * folio in the batch may have the readahead <<
2319 * clear so that the caller can take the appr <<
2320 */ 1861 */
2321 static void filemap_get_read_batch(struct add !! 1862 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
2322 pgoff_t index, pgoff_t max, s !! 1863 unsigned int nr_pages, struct page **pages)
2323 { 1864 {
2324 XA_STATE(xas, &mapping->i_pages, inde 1865 XA_STATE(xas, &mapping->i_pages, index);
2325 struct folio *folio; !! 1866 struct page *page;
>> 1867 unsigned int ret = 0;
>> 1868
>> 1869 if (unlikely(!nr_pages))
>> 1870 return 0;
2326 1871
2327 rcu_read_lock(); 1872 rcu_read_lock();
2328 for (folio = xas_load(&xas); folio; f !! 1873 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
2329 if (xas_retry(&xas, folio)) !! 1874 struct page *head;
>> 1875 if (xas_retry(&xas, page))
2330 continue; 1876 continue;
2331 if (xas.xa_index > max || xa_ !! 1877 /*
2332 break; !! 1878 * If the entry has been swapped out, we can stop looking.
2333 if (xa_is_sibling(folio)) !! 1879 * No current caller is looking for DAX entries.
>> 1880 */
>> 1881 if (xa_is_value(page))
2334 break; 1882 break;
2335 if (!folio_try_get(folio)) !! 1883
>> 1884 head = compound_head(page);
>> 1885 if (!page_cache_get_speculative(head))
2336 goto retry; 1886 goto retry;
2337 1887
2338 if (unlikely(folio != xas_rel !! 1888 /* The page was split under us? */
2339 goto put_folio; !! 1889 if (compound_head(page) != head)
>> 1890 goto put_page;
>> 1891
>> 1892 /* Has the page moved? */
>> 1893 if (unlikely(page != xas_reload(&xas)))
>> 1894 goto put_page;
2340 1895
2341 if (!folio_batch_add(fbatch, !! 1896 pages[ret] = page;
2342 break; !! 1897 if (++ret == nr_pages)
2343 if (!folio_test_uptodate(foli <<
2344 break; <<
2345 if (folio_test_readahead(foli <<
2346 break; 1898 break;
2347 xas_advance(&xas, folio_next_ <<
2348 continue; 1899 continue;
2349 put_folio: !! 1900 put_page:
2350 folio_put(folio); !! 1901 put_page(head);
2351 retry: 1902 retry:
2352 xas_reset(&xas); 1903 xas_reset(&xas);
2353 } 1904 }
2354 rcu_read_unlock(); 1905 rcu_read_unlock();
>> 1906 return ret;
2355 } 1907 }
>> 1908 EXPORT_SYMBOL(find_get_pages_contig);
2356 1909
2357 static int filemap_read_folio(struct file *fi !! 1910 /**
2358 struct folio *folio) !! 1911 * find_get_pages_range_tag - find and return pages in given range matching @tag
>> 1912 * @mapping: the address_space to search
>> 1913 * @index: the starting page index
>> 1914 * @end: The final page index (inclusive)
>> 1915 * @tag: the tag index
>> 1916 * @nr_pages: the maximum number of pages
>> 1917 * @pages: where the resulting pages are placed
>> 1918 *
>> 1919 * Like find_get_pages, except we only return pages which are tagged with
>> 1920 * @tag. We update @index to index the next page for the traversal.
>> 1921 *
>> 1922 * Return: the number of pages which were found.
>> 1923 */
>> 1924 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
>> 1925 pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
>> 1926 struct page **pages)
2359 { 1927 {
2360 bool workingset = folio_test_workings !! 1928 XA_STATE(xas, &mapping->i_pages, *index);
2361 unsigned long pflags; !! 1929 struct page *page;
2362 int error; !! 1930 unsigned ret = 0;
2363 <<
2364 /* Start the actual read. The read wi <<
2365 if (unlikely(workingset)) <<
2366 psi_memstall_enter(&pflags); <<
2367 error = filler(file, folio); <<
2368 if (unlikely(workingset)) <<
2369 psi_memstall_leave(&pflags); <<
2370 if (error) <<
2371 return error; <<
2372 1931
2373 error = folio_wait_locked_killable(fo !! 1932 if (unlikely(!nr_pages))
2374 if (error) <<
2375 return error; <<
2376 if (folio_test_uptodate(folio)) <<
2377 return 0; 1933 return 0;
2378 if (file) <<
2379 shrink_readahead_size_eio(&fi <<
2380 return -EIO; <<
2381 } <<
2382 1934
2383 static bool filemap_range_uptodate(struct add !! 1935 rcu_read_lock();
2384 loff_t pos, size_t count, str !! 1936 xas_for_each_marked(&xas, page, end, tag) {
2385 bool need_uptodate) !! 1937 struct page *head;
2386 { !! 1938 if (xas_retry(&xas, page))
2387 if (folio_test_uptodate(folio)) !! 1939 continue;
2388 return true; !! 1940 /*
2389 /* pipes can't handle partially uptod !! 1941 * Shadow entries should never be tagged, but this iteration
2390 if (need_uptodate) !! 1942 * is lockless so there is a window for page reclaim to evict
2391 return false; !! 1943 * a page we saw tagged. Skip over it.
2392 if (!mapping->a_ops->is_partially_upt !! 1944 */
2393 return false; !! 1945 if (xa_is_value(page))
2394 if (mapping->host->i_blkbits >= folio !! 1946 continue;
2395 return false; <<
2396 <<
2397 if (folio_pos(folio) > pos) { <<
2398 count -= folio_pos(folio) - p <<
2399 pos = 0; <<
2400 } else { <<
2401 pos -= folio_pos(folio); <<
2402 } <<
2403 <<
2404 return mapping->a_ops->is_partially_u <<
2405 } <<
2406 <<
2407 static int filemap_update_page(struct kiocb * <<
2408 struct address_space *mapping <<
2409 struct folio *folio, bool nee <<
2410 { <<
2411 int error; <<
2412 1947
2413 if (iocb->ki_flags & IOCB_NOWAIT) { !! 1948 head = compound_head(page);
2414 if (!filemap_invalidate_trylo !! 1949 if (!page_cache_get_speculative(head))
2415 return -EAGAIN; !! 1950 goto retry;
2416 } else { <<
2417 filemap_invalidate_lock_share <<
2418 } <<
2419 1951
2420 if (!folio_trylock(folio)) { !! 1952 /* The page was split under us? */
2421 error = -EAGAIN; !! 1953 if (compound_head(page) != head)
2422 if (iocb->ki_flags & (IOCB_NO !! 1954 goto put_page;
2423 goto unlock_mapping; !! 1955
2424 if (!(iocb->ki_flags & IOCB_W !! 1956 /* Has the page moved? */
2425 filemap_invalidate_un !! 1957 if (unlikely(page != xas_reload(&xas)))
2426 /* !! 1958 goto put_page;
2427 * This is where we u !! 1959
2428 * previously submitt !! 1960 pages[ret] = page;
2429 */ !! 1961 if (++ret == nr_pages) {
2430 folio_put_wait_locked !! 1962 *index = xas.xa_index + 1;
2431 return AOP_TRUNCATED_ !! 1963 goto out;
2432 } 1964 }
2433 error = __folio_lock_async(fo !! 1965 continue;
2434 if (error) !! 1966 put_page:
2435 goto unlock_mapping; !! 1967 put_page(head);
>> 1968 retry:
>> 1969 xas_reset(&xas);
2436 } 1970 }
2437 1971
2438 error = AOP_TRUNCATED_PAGE; <<
2439 if (!folio->mapping) <<
2440 goto unlock; <<
2441 <<
2442 error = 0; <<
2443 if (filemap_range_uptodate(mapping, i <<
2444 need_uptod <<
2445 goto unlock; <<
2446 <<
2447 error = -EAGAIN; <<
2448 if (iocb->ki_flags & (IOCB_NOIO | IOC <<
2449 goto unlock; <<
2450 <<
2451 error = filemap_read_folio(iocb->ki_f <<
2452 folio); <<
2453 goto unlock_mapping; <<
2454 unlock: <<
2455 folio_unlock(folio); <<
2456 unlock_mapping: <<
2457 filemap_invalidate_unlock_shared(mapp <<
2458 if (error == AOP_TRUNCATED_PAGE) <<
2459 folio_put(folio); <<
2460 return error; <<
2461 } <<
2462 <<
2463 static int filemap_create_folio(struct file * <<
2464 struct address_space *mapping <<
2465 struct folio_batch *fbatch) <<
2466 { <<
2467 struct folio *folio; <<
2468 int error; <<
2469 unsigned int min_order = mapping_min_ <<
2470 pgoff_t index; <<
2471 <<
2472 folio = filemap_alloc_folio(mapping_g <<
2473 if (!folio) <<
2474 return -ENOMEM; <<
2475 <<
2476 /* 1972 /*
2477 * Protect against truncate / hole pu !! 1973 * We come here when we got to @end. We take care to not overflow the
2478 * here assures we cannot instantiate !! 1974 * index @index as it confuses some of the callers. This breaks the
2479 * pagecache folios after evicting pa !! 1975 * iteration when there is a page at index -1 but that is already
2480 * and before actually freeing blocks !! 1976 * broken anyway.
2481 * release invalidate_lock after inse <<
2482 * the page cache as the locked folio <<
2483 * synchronize with hole punching. Bu <<
2484 * such as filemap_update_page() fill <<
2485 * pages or ->readahead() that need t <<
2486 * while mapping blocks for IO so let <<
2487 * well to keep locking rules simple. <<
2488 */ 1977 */
2489 filemap_invalidate_lock_shared(mappin !! 1978 if (end == (pgoff_t)-1)
2490 index = (pos >> (PAGE_SHIFT + min_ord !! 1979 *index = (pgoff_t)-1;
2491 error = filemap_add_folio(mapping, fo !! 1980 else
2492 mapping_gfp_constrain !! 1981 *index = end + 1;
2493 if (error == -EEXIST) !! 1982 out:
2494 error = AOP_TRUNCATED_PAGE; !! 1983 rcu_read_unlock();
2495 if (error) <<
2496 goto error; <<
2497 <<
2498 error = filemap_read_folio(file, mapp <<
2499 if (error) <<
2500 goto error; <<
2501 1984
2502 filemap_invalidate_unlock_shared(mapp !! 1985 return ret;
2503 folio_batch_add(fbatch, folio); <<
2504 return 0; <<
2505 error: <<
2506 filemap_invalidate_unlock_shared(mapp <<
2507 folio_put(folio); <<
2508 return error; <<
2509 } 1986 }
>> 1987 EXPORT_SYMBOL(find_get_pages_range_tag);
2510 1988
2511 static int filemap_readahead(struct kiocb *io !! 1989 /**
2512 struct address_space *mapping !! 1990 * find_get_entries_tag - find and return entries that match @tag
2513 pgoff_t last_index) !! 1991 * @mapping: the address_space to search
>> 1992 * @start: the starting page cache index
>> 1993 * @tag: the tag index
>> 1994 * @nr_entries: the maximum number of entries
>> 1995 * @entries: where the resulting entries are placed
>> 1996 * @indices: the cache indices corresponding to the entries in @entries
>> 1997 *
>> 1998 * Like find_get_entries, except we only return entries which are tagged with
>> 1999 * @tag.
>> 2000 *
>> 2001 * Return: the number of entries which were found.
>> 2002 */
>> 2003 unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
>> 2004 xa_mark_t tag, unsigned int nr_entries,
>> 2005 struct page **entries, pgoff_t *indices)
2514 { 2006 {
2515 DEFINE_READAHEAD(ractl, file, &file-> !! 2007 XA_STATE(xas, &mapping->i_pages, start);
2516 !! 2008 struct page *page;
2517 if (iocb->ki_flags & IOCB_NOIO) !! 2009 unsigned int ret = 0;
2518 return -EAGAIN; <<
2519 page_cache_async_ra(&ractl, folio, la <<
2520 return 0; <<
2521 } <<
2522 2010
2523 static int filemap_get_pages(struct kiocb *io !! 2011 if (!nr_entries)
2524 struct folio_batch *fbatch, b !! 2012 return 0;
2525 { <<
2526 struct file *filp = iocb->ki_filp; <<
2527 struct address_space *mapping = filp- <<
2528 struct file_ra_state *ra = &filp->f_r <<
2529 pgoff_t index = iocb->ki_pos >> PAGE_ <<
2530 pgoff_t last_index; <<
2531 struct folio *folio; <<
2532 unsigned int flags; <<
2533 int err = 0; <<
2534 2013
2535 /* "last_index" is the index of the p !! 2014 rcu_read_lock();
2536 last_index = DIV_ROUND_UP(iocb->ki_po !! 2015 xas_for_each_marked(&xas, page, ULONG_MAX, tag) {
2537 retry: !! 2016 struct page *head;
2538 if (fatal_signal_pending(current)) !! 2017 if (xas_retry(&xas, page))
2539 return -EINTR; !! 2018 continue;
>> 2019 /*
>> 2020 * A shadow entry of a recently evicted page, a swap
>> 2021 * entry from shmem/tmpfs or a DAX entry. Return it
>> 2022 * without attempting to raise page count.
>> 2023 */
>> 2024 if (xa_is_value(page))
>> 2025 goto export;
2540 2026
2541 filemap_get_read_batch(mapping, index !! 2027 head = compound_head(page);
2542 if (!folio_batch_count(fbatch)) { !! 2028 if (!page_cache_get_speculative(head))
2543 if (iocb->ki_flags & IOCB_NOI <<
2544 return -EAGAIN; <<
2545 if (iocb->ki_flags & IOCB_NOW <<
2546 flags = memalloc_noio <<
2547 page_cache_sync_readahead(map <<
2548 last_index - <<
2549 if (iocb->ki_flags & IOCB_NOW <<
2550 memalloc_noio_restore <<
2551 filemap_get_read_batch(mappin <<
2552 } <<
2553 if (!folio_batch_count(fbatch)) { <<
2554 if (iocb->ki_flags & (IOCB_NO <<
2555 return -EAGAIN; <<
2556 err = filemap_create_folio(fi <<
2557 if (err == AOP_TRUNCATED_PAGE <<
2558 goto retry; 2029 goto retry;
2559 return err; <<
2560 } <<
2561 2030
2562 folio = fbatch->folios[folio_batch_co !! 2031 /* The page was split under us? */
2563 if (folio_test_readahead(folio)) { !! 2032 if (compound_head(page) != head)
2564 err = filemap_readahead(iocb, !! 2033 goto put_page;
2565 if (err) !! 2034
2566 goto err; !! 2035 /* Has the page moved? */
2567 } !! 2036 if (unlikely(page != xas_reload(&xas)))
2568 if (!folio_test_uptodate(folio)) { !! 2037 goto put_page;
2569 if ((iocb->ki_flags & IOCB_WA !! 2038
2570 folio_batch_count(fbatch) !! 2039 export:
2571 iocb->ki_flags |= IOC !! 2040 indices[ret] = xas.xa_index;
2572 err = filemap_update_page(ioc !! 2041 entries[ret] = page;
2573 nee !! 2042 if (++ret == nr_entries)
2574 if (err) !! 2043 break;
2575 goto err; !! 2044 continue;
>> 2045 put_page:
>> 2046 put_page(head);
>> 2047 retry:
>> 2048 xas_reset(&xas);
2576 } 2049 }
2577 !! 2050 rcu_read_unlock();
2578 trace_mm_filemap_get_pages(mapping, i !! 2051 return ret;
2579 return 0; <<
2580 err: <<
2581 if (err < 0) <<
2582 folio_put(folio); <<
2583 if (likely(--fbatch->nr)) <<
2584 return 0; <<
2585 if (err == AOP_TRUNCATED_PAGE) <<
2586 goto retry; <<
2587 return err; <<
2588 } 2052 }
>> 2053 EXPORT_SYMBOL(find_get_entries_tag);
2589 2054
2590 static inline bool pos_same_folio(loff_t pos1 !! 2055 /*
>> 2056 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
>> 2057 * a _large_ part of the i/o request. Imagine the worst scenario:
>> 2058 *
>> 2059 * ---R__________________________________________B__________
>> 2060 * ^ reading here ^ bad block(assume 4k)
>> 2061 *
>> 2062 * read(R) => miss => readahead(R...B) => media error => frustrating retries
>> 2063 * => failing the whole request => read(R) => read(R+1) =>
>> 2064 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
>> 2065 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
>> 2066 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
>> 2067 *
>> 2068 * It is going insane. Fix it by quickly scaling down the readahead size.
>> 2069 */
>> 2070 static void shrink_readahead_size_eio(struct file *filp,
>> 2071 struct file_ra_state *ra)
2591 { 2072 {
2592 unsigned int shift = folio_shift(foli !! 2073 ra->ra_pages /= 4;
2593 <<
2594 return (pos1 >> shift == pos2 >> shif <<
2595 } 2074 }
2596 2075
2597 /** 2076 /**
2598 * filemap_read - Read data from the page cac !! 2077 * generic_file_buffered_read - generic file read routine
2599 * @iocb: The iocb to read. !! 2078 * @iocb: the iocb to read
2600 * @iter: Destination for the data. !! 2079 * @iter: data destination
2601 * @already_read: Number of bytes already rea !! 2080 * @written: already copied
2602 * !! 2081 *
2603 * Copies data from the page cache. If the d !! 2082 * This is a generic file read routine, and uses the
2604 * uses the readahead and read_folio address_ !! 2083 * mapping->a_ops->readpage() function for the actual low-level stuff.
2605 * !! 2084 *
2606 * Return: Total number of bytes copied, incl !! 2085 * This is really ugly. But the goto's actually try to clarify some
2607 * the caller. If an error happens before an !! 2086 * of the logic when it comes to error handling etc.
2608 * a negative error number. !! 2087 *
>> 2088 * Return:
>> 2089 * * total number of bytes copied, including those the were already @written
>> 2090 * * negative error code if nothing was copied
2609 */ 2091 */
2610 ssize_t filemap_read(struct kiocb *iocb, stru !! 2092 static ssize_t generic_file_buffered_read(struct kiocb *iocb,
2611 ssize_t already_read) !! 2093 struct iov_iter *iter, ssize_t written)
2612 { 2094 {
2613 struct file *filp = iocb->ki_filp; 2095 struct file *filp = iocb->ki_filp;
2614 struct file_ra_state *ra = &filp->f_r <<
2615 struct address_space *mapping = filp- 2096 struct address_space *mapping = filp->f_mapping;
2616 struct inode *inode = mapping->host; 2097 struct inode *inode = mapping->host;
2617 struct folio_batch fbatch; !! 2098 struct file_ra_state *ra = &filp->f_ra;
2618 int i, error = 0; !! 2099 loff_t *ppos = &iocb->ki_pos;
2619 bool writably_mapped; !! 2100 pgoff_t index;
2620 loff_t isize, end_offset; !! 2101 pgoff_t last_index;
2621 loff_t last_pos = ra->prev_pos; !! 2102 pgoff_t prev_index;
>> 2103 unsigned long offset; /* offset into pagecache page */
>> 2104 unsigned int prev_offset;
>> 2105 int error = 0;
2622 2106
2623 if (unlikely(iocb->ki_pos >= inode->i !! 2107 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
2624 return 0; <<
2625 if (unlikely(!iov_iter_count(iter))) <<
2626 return 0; 2108 return 0;
>> 2109 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2627 2110
2628 iov_iter_truncate(iter, inode->i_sb-> !! 2111 index = *ppos >> PAGE_SHIFT;
2629 folio_batch_init(&fbatch); !! 2112 prev_index = ra->prev_pos >> PAGE_SHIFT;
>> 2113 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
>> 2114 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
>> 2115 offset = *ppos & ~PAGE_MASK;
2630 2116
2631 do { !! 2117 for (;;) {
2632 cond_resched(); !! 2118 struct page *page;
2633 !! 2119 pgoff_t end_index;
2634 /* !! 2120 loff_t isize;
2635 * If we've already successfu !! 2121 unsigned long nr, ret;
2636 * can no longer safely retur <<
2637 * an async read NOWAIT at th <<
2638 */ <<
2639 if ((iocb->ki_flags & IOCB_WA <<
2640 iocb->ki_flags |= IOC <<
2641 2122
2642 if (unlikely(iocb->ki_pos >= !! 2123 cond_resched();
2643 break; !! 2124 find_page:
>> 2125 if (fatal_signal_pending(current)) {
>> 2126 error = -EINTR;
>> 2127 goto out;
>> 2128 }
2644 2129
2645 error = filemap_get_pages(ioc !! 2130 page = find_get_page(mapping, index);
2646 if (error < 0) !! 2131 if (!page) {
2647 break; !! 2132 if (iocb->ki_flags & IOCB_NOWAIT)
>> 2133 goto would_block;
>> 2134 page_cache_sync_readahead(mapping,
>> 2135 ra, filp,
>> 2136 index, last_index - index);
>> 2137 page = find_get_page(mapping, index);
>> 2138 if (unlikely(page == NULL))
>> 2139 goto no_cached_page;
>> 2140 }
>> 2141 if (PageReadahead(page)) {
>> 2142 page_cache_async_readahead(mapping,
>> 2143 ra, filp, page,
>> 2144 index, last_index - index);
>> 2145 }
>> 2146 if (!PageUptodate(page)) {
>> 2147 if (iocb->ki_flags & IOCB_NOWAIT) {
>> 2148 put_page(page);
>> 2149 goto would_block;
>> 2150 }
2648 2151
>> 2152 /*
>> 2153 * See comment in do_read_cache_page on why
>> 2154 * wait_on_page_locked is used to avoid unnecessarily
>> 2155 * serialisations and why it's safe.
>> 2156 */
>> 2157 error = wait_on_page_locked_killable(page);
>> 2158 if (unlikely(error))
>> 2159 goto readpage_error;
>> 2160 if (PageUptodate(page))
>> 2161 goto page_ok;
>> 2162
>> 2163 if (inode->i_blkbits == PAGE_SHIFT ||
>> 2164 !mapping->a_ops->is_partially_uptodate)
>> 2165 goto page_not_up_to_date;
>> 2166 /* pipes can't handle partially uptodate pages */
>> 2167 if (unlikely(iov_iter_is_pipe(iter)))
>> 2168 goto page_not_up_to_date;
>> 2169 if (!trylock_page(page))
>> 2170 goto page_not_up_to_date;
>> 2171 /* Did it get truncated before we got the lock? */
>> 2172 if (!page->mapping)
>> 2173 goto page_not_up_to_date_locked;
>> 2174 if (!mapping->a_ops->is_partially_uptodate(page,
>> 2175 offset, iter->count))
>> 2176 goto page_not_up_to_date_locked;
>> 2177 unlock_page(page);
>> 2178 }
>> 2179 page_ok:
2649 /* 2180 /*
2650 * i_size must be checked aft !! 2181 * i_size must be checked after we know the page is Uptodate.
2651 * 2182 *
2652 * Checking i_size after the 2183 * Checking i_size after the check allows us to calculate
2653 * the correct value for "nr" 2184 * the correct value for "nr", which means the zero-filled
2654 * part of the page is not co 2185 * part of the page is not copied back to userspace (unless
2655 * another truncate extends t 2186 * another truncate extends the file - this is desired though).
2656 */ 2187 */
>> 2188
2657 isize = i_size_read(inode); 2189 isize = i_size_read(inode);
2658 if (unlikely(iocb->ki_pos >= !! 2190 end_index = (isize - 1) >> PAGE_SHIFT;
2659 goto put_folios; !! 2191 if (unlikely(!isize || index > end_index)) {
2660 end_offset = min_t(loff_t, is !! 2192 put_page(page);
>> 2193 goto out;
>> 2194 }
>> 2195
>> 2196 /* nr is the maximum number of bytes to copy from this page */
>> 2197 nr = PAGE_SIZE;
>> 2198 if (index == end_index) {
>> 2199 nr = ((isize - 1) & ~PAGE_MASK) + 1;
>> 2200 if (nr <= offset) {
>> 2201 put_page(page);
>> 2202 goto out;
>> 2203 }
>> 2204 }
>> 2205 nr = nr - offset;
>> 2206
>> 2207 /* If users can be writing to this page using arbitrary
>> 2208 * virtual addresses, take care about potential aliasing
>> 2209 * before reading the page on the kernel side.
>> 2210 */
>> 2211 if (mapping_writably_mapped(mapping))
>> 2212 flush_dcache_page(page);
2661 2213
2662 /* 2214 /*
2663 * Once we start copying data !! 2215 * When a sequential read accesses a page several times,
2664 * cachelines that might be c !! 2216 * only mark it as accessed the first time.
2665 */ 2217 */
2666 writably_mapped = mapping_wri !! 2218 if (prev_index != index || offset != prev_offset)
>> 2219 mark_page_accessed(page);
>> 2220 prev_index = index;
2667 2221
2668 /* 2222 /*
2669 * When a read accesses the s !! 2223 * Ok, we have the page, and it's up-to-date, so
2670 * mark it as accessed the fi !! 2224 * now we can copy it to user space...
2671 */ 2225 */
2672 if (!pos_same_folio(iocb->ki_ <<
2673 fbatch.fo <<
2674 folio_mark_accessed(f <<
2675 <<
2676 for (i = 0; i < folio_batch_c <<
2677 struct folio *folio = <<
2678 size_t fsize = folio_ <<
2679 size_t offset = iocb- <<
2680 size_t bytes = min_t( <<
2681 <<
2682 size_t copied; <<
2683 2226
2684 if (end_offset < foli !! 2227 ret = copy_page_to_iter(page, offset, nr, iter);
2685 break; !! 2228 offset += ret;
2686 if (i > 0) !! 2229 index += offset >> PAGE_SHIFT;
2687 folio_mark_ac !! 2230 offset &= ~PAGE_MASK;
2688 /* !! 2231 prev_offset = offset;
2689 * If users can be wr <<
2690 * virtual addresses, <<
2691 * before reading the <<
2692 */ <<
2693 if (writably_mapped) <<
2694 flush_dcache_ <<
2695 2232
2696 copied = copy_folio_t !! 2233 put_page(page);
>> 2234 written += ret;
>> 2235 if (!iov_iter_count(iter))
>> 2236 goto out;
>> 2237 if (ret < nr) {
>> 2238 error = -EFAULT;
>> 2239 goto out;
>> 2240 }
>> 2241 continue;
2697 2242
2698 already_read += copie !! 2243 page_not_up_to_date:
2699 iocb->ki_pos += copie !! 2244 /* Get exclusive access to the page ... */
2700 last_pos = iocb->ki_p !! 2245 error = lock_page_killable(page);
>> 2246 if (unlikely(error))
>> 2247 goto readpage_error;
>> 2248
>> 2249 page_not_up_to_date_locked:
>> 2250 /* Did it get truncated before we got the lock? */
>> 2251 if (!page->mapping) {
>> 2252 unlock_page(page);
>> 2253 put_page(page);
>> 2254 continue;
>> 2255 }
2701 2256
2702 if (copied < bytes) { !! 2257 /* Did somebody else fill it already? */
2703 error = -EFAU !! 2258 if (PageUptodate(page)) {
2704 break; !! 2259 unlock_page(page);
>> 2260 goto page_ok;
>> 2261 }
>> 2262
>> 2263 readpage:
>> 2264 /*
>> 2265 * A previous I/O error may have been due to temporary
>> 2266 * failures, eg. multipath errors.
>> 2267 * PG_error will be set again if readpage fails.
>> 2268 */
>> 2269 ClearPageError(page);
>> 2270 /* Start the actual read. The read will unlock the page. */
>> 2271 error = mapping->a_ops->readpage(filp, page);
>> 2272
>> 2273 if (unlikely(error)) {
>> 2274 if (error == AOP_TRUNCATED_PAGE) {
>> 2275 put_page(page);
>> 2276 error = 0;
>> 2277 goto find_page;
2705 } 2278 }
>> 2279 goto readpage_error;
2706 } 2280 }
2707 put_folios: <<
2708 for (i = 0; i < folio_batch_c <<
2709 folio_put(fbatch.foli <<
2710 folio_batch_init(&fbatch); <<
2711 } while (iov_iter_count(iter) && iocb <<
2712 <<
2713 file_accessed(filp); <<
2714 ra->prev_pos = last_pos; <<
2715 return already_read ? already_read : <<
2716 } <<
2717 EXPORT_SYMBOL_GPL(filemap_read); <<
2718 2281
2719 int kiocb_write_and_wait(struct kiocb *iocb, !! 2282 if (!PageUptodate(page)) {
2720 { !! 2283 error = lock_page_killable(page);
2721 struct address_space *mapping = iocb- !! 2284 if (unlikely(error))
2722 loff_t pos = iocb->ki_pos; !! 2285 goto readpage_error;
2723 loff_t end = pos + count - 1; !! 2286 if (!PageUptodate(page)) {
2724 !! 2287 if (page->mapping == NULL) {
2725 if (iocb->ki_flags & IOCB_NOWAIT) { !! 2288 /*
2726 if (filemap_range_needs_write !! 2289 * invalidate_mapping_pages got it
2727 return -EAGAIN; !! 2290 */
2728 return 0; !! 2291 unlock_page(page);
2729 } !! 2292 put_page(page);
>> 2293 goto find_page;
>> 2294 }
>> 2295 unlock_page(page);
>> 2296 shrink_readahead_size_eio(filp, ra);
>> 2297 error = -EIO;
>> 2298 goto readpage_error;
>> 2299 }
>> 2300 unlock_page(page);
>> 2301 }
2730 2302
2731 return filemap_write_and_wait_range(m !! 2303 goto page_ok;
2732 } <<
2733 EXPORT_SYMBOL_GPL(kiocb_write_and_wait); <<
2734 2304
2735 int filemap_invalidate_pages(struct address_s !! 2305 readpage_error:
2736 loff_t pos, loff !! 2306 /* UHHUH! A synchronous read error occurred. Report it */
2737 { !! 2307 put_page(page);
2738 int ret; !! 2308 goto out;
2739 2309
2740 if (nowait) { !! 2310 no_cached_page:
2741 /* we could block if there ar !! 2311 /*
2742 if (filemap_range_has_page(ma !! 2312 * Ok, it wasn't cached, so we need to create a new
2743 return -EAGAIN; !! 2313 * page..
2744 } else { !! 2314 */
2745 ret = filemap_write_and_wait_ !! 2315 page = page_cache_alloc(mapping);
2746 if (ret) !! 2316 if (!page) {
2747 return ret; !! 2317 error = -ENOMEM;
>> 2318 goto out;
>> 2319 }
>> 2320 error = add_to_page_cache_lru(page, mapping, index,
>> 2321 mapping_gfp_constraint(mapping, GFP_KERNEL));
>> 2322 if (error) {
>> 2323 put_page(page);
>> 2324 if (error == -EEXIST) {
>> 2325 error = 0;
>> 2326 goto find_page;
>> 2327 }
>> 2328 goto out;
>> 2329 }
>> 2330 goto readpage;
2748 } 2331 }
2749 2332
2750 /* !! 2333 would_block:
2751 * After a write we want buffered rea !! 2334 error = -EAGAIN;
2752 * the new data. We invalidate clean !! 2335 out:
2753 * about to write. We do this *befor !! 2336 ra->prev_pos = prev_index;
2754 * without clobbering -EIOCBQUEUED fr !! 2337 ra->prev_pos <<= PAGE_SHIFT;
2755 */ !! 2338 ra->prev_pos |= prev_offset;
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 2339
2764 return filemap_invalidate_pages(mappi !! 2340 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2765 iocb- !! 2341 file_accessed(filp);
2766 iocb- !! 2342 return written ? written : error;
2767 } 2343 }
2768 EXPORT_SYMBOL_GPL(kiocb_invalidate_pages); <<
2769 2344
2770 /** 2345 /**
2771 * generic_file_read_iter - generic filesyste 2346 * generic_file_read_iter - generic filesystem read routine
2772 * @iocb: kernel I/O control block 2347 * @iocb: kernel I/O control block
2773 * @iter: destination for the data read 2348 * @iter: destination for the data read
2774 * 2349 *
2775 * This is the "read_iter()" routine for all 2350 * This is the "read_iter()" routine for all filesystems
2776 * that can use the page cache directly. 2351 * that can use the page cache directly.
2777 * <<
2778 * The IOCB_NOWAIT flag in iocb->ki_flags ind <<
2779 * be returned when no data can be read witho <<
2780 * to complete; it doesn't prevent readahead. <<
2781 * <<
2782 * The IOCB_NOIO flag in iocb->ki_flags indic <<
2783 * requests shall be made for the read or for <<
2784 * can be read, -EAGAIN shall be returned. W <<
2785 * triggered, a partial, possibly empty read <<
2786 * <<
2787 * Return: 2352 * Return:
2788 * * number of bytes copied, even for partial 2353 * * number of bytes copied, even for partial reads
2789 * * negative error code (or 0 if IOCB_NOIO) !! 2354 * * negative error code if nothing was read
2790 */ 2355 */
2791 ssize_t 2356 ssize_t
2792 generic_file_read_iter(struct kiocb *iocb, st 2357 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2793 { 2358 {
2794 size_t count = iov_iter_count(iter); 2359 size_t count = iov_iter_count(iter);
2795 ssize_t retval = 0; 2360 ssize_t retval = 0;
2796 2361
2797 if (!count) 2362 if (!count)
2798 return 0; /* skip atime */ !! 2363 goto out; /* skip atime */
2799 2364
2800 if (iocb->ki_flags & IOCB_DIRECT) { 2365 if (iocb->ki_flags & IOCB_DIRECT) {
2801 struct file *file = iocb->ki_ 2366 struct file *file = iocb->ki_filp;
2802 struct address_space *mapping 2367 struct address_space *mapping = file->f_mapping;
2803 struct inode *inode = mapping 2368 struct inode *inode = mapping->host;
>> 2369 loff_t size;
>> 2370
>> 2371 size = i_size_read(inode);
>> 2372 if (iocb->ki_flags & IOCB_NOWAIT) {
>> 2373 if (filemap_range_has_page(mapping, iocb->ki_pos,
>> 2374 iocb->ki_pos + count - 1))
>> 2375 return -EAGAIN;
>> 2376 } else {
>> 2377 retval = filemap_write_and_wait_range(mapping,
>> 2378 iocb->ki_pos,
>> 2379 iocb->ki_pos + count - 1);
>> 2380 if (retval < 0)
>> 2381 goto out;
>> 2382 }
2804 2383
2805 retval = kiocb_write_and_wait <<
2806 if (retval < 0) <<
2807 return retval; <<
2808 file_accessed(file); 2384 file_accessed(file);
2809 2385
2810 retval = mapping->a_ops->dire 2386 retval = mapping->a_ops->direct_IO(iocb, iter);
2811 if (retval >= 0) { 2387 if (retval >= 0) {
2812 iocb->ki_pos += retva 2388 iocb->ki_pos += retval;
2813 count -= retval; 2389 count -= retval;
2814 } 2390 }
2815 if (retval != -EIOCBQUEUED) !! 2391 iov_iter_revert(iter, count - iov_iter_count(iter));
2816 iov_iter_revert(iter, <<
2817 2392
2818 /* 2393 /*
2819 * Btrfs can have a short DIO 2394 * Btrfs can have a short DIO read if we encounter
2820 * compressed extents, so if 2395 * compressed extents, so if there was an error, or if
2821 * we've already read everyth 2396 * we've already read everything we wanted to, or if
2822 * there was a short read bec 2397 * there was a short read because we hit EOF, go ahead
2823 * and return. Otherwise fal 2398 * and return. Otherwise fallthrough to buffered io for
2824 * the rest of the read. Buf 2399 * the rest of the read. Buffered reads will not work for
2825 * DAX files, so don't bother 2400 * DAX files, so don't bother trying.
2826 */ 2401 */
2827 if (retval < 0 || !count || I !! 2402 if (retval < 0 || !count || iocb->ki_pos >= size ||
2828 return retval; !! 2403 IS_DAX(inode))
2829 if (iocb->ki_pos >= i_size_re !! 2404 goto out;
2830 return retval; <<
2831 } <<
2832 <<
2833 return filemap_read(iocb, iter, retva <<
2834 } <<
2835 EXPORT_SYMBOL(generic_file_read_iter); <<
2836 <<
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 } 2405 }
2867 2406
2868 return spliced; !! 2407 retval = generic_file_buffered_read(iocb, iter, retval);
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: 2408 out:
2976 folio_batch_release(&fbatch); !! 2409 return retval;
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 <<
2984 struct address_space *mapping <<
2985 loff_t start, loff_t end, boo <<
2986 { <<
2987 const struct address_space_operations <<
2988 size_t offset, bsz = i_blocksize(mapp <<
2989 <<
2990 if (xa_is_value(folio) || folio_test_ <<
2991 return seek_data ? start : en <<
2992 if (!ops->is_partially_uptodate) <<
2993 return seek_data ? end : star <<
2994 <<
2995 xas_pause(xas); <<
2996 rcu_read_unlock(); <<
2997 folio_lock(folio); <<
2998 if (unlikely(folio->mapping != mappin <<
2999 goto unlock; <<
3000 <<
3001 offset = offset_in_folio(folio, start <<
3002 <<
3003 do { <<
3004 if (ops->is_partially_uptodat <<
3005 <<
3006 break; <<
3007 start = (start + bsz) & ~(bsz <<
3008 offset += bsz; <<
3009 } while (offset < folio_size(folio)); <<
3010 unlock: <<
3011 folio_unlock(folio); <<
3012 rcu_read_lock(); <<
3013 return start; <<
3014 } 2410 }
>> 2411 EXPORT_SYMBOL(generic_file_read_iter);
3015 2412
3016 static inline size_t seek_folio_size(struct x !! 2413 #ifdef CONFIG_MMU
>> 2414 #define MMAP_LOTSAMISS (100)
>> 2415 static struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
>> 2416 struct file *fpin)
3017 { 2417 {
3018 if (xa_is_value(folio)) !! 2418 int flags = vmf->flags;
3019 return PAGE_SIZE << xas_get_o <<
3020 return folio_size(folio); <<
3021 } <<
3022 <<
3023 /** <<
3024 * mapping_seek_hole_data - Seek for SEEK_DAT <<
3025 * @mapping: Address space to search. <<
3026 * @start: First byte to consider. <<
3027 * @end: Limit of search (exclusive). <<
3028 * @whence: Either SEEK_HOLE or SEEK_DATA. <<
3029 * <<
3030 * If the page cache knows which blocks conta <<
3031 * contain data, your filesystem can use this <<
3032 * SEEK_HOLE and SEEK_DATA. This is useful f <<
3033 * entirely memory-based such as tmpfs, and f <<
3034 * unwritten extents. <<
3035 * <<
3036 * Return: The requested offset on success, o <<
3037 * SEEK_DATA and there is no data after @star <<
3038 * after @end - 1, so SEEK_HOLE returns @end <<
3039 * and @end contain data. <<
3040 */ <<
3041 loff_t mapping_seek_hole_data(struct address_ <<
3042 loff_t end, int whence) <<
3043 { <<
3044 XA_STATE(xas, &mapping->i_pages, star <<
3045 pgoff_t max = (end - 1) >> PAGE_SHIFT <<
3046 bool seek_data = (whence == SEEK_DATA <<
3047 struct folio *folio; <<
3048 <<
3049 if (end <= start) <<
3050 return -ENXIO; <<
3051 2419
3052 rcu_read_lock(); !! 2420 if (fpin)
3053 while ((folio = find_get_entry(&xas, !! 2421 return fpin;
3054 loff_t pos = (u64)xas.xa_inde <<
3055 size_t seek_size; <<
3056 <<
3057 if (start < pos) { <<
3058 if (!seek_data) <<
3059 goto unlock; <<
3060 start = pos; <<
3061 } <<
3062 2422
3063 seek_size = seek_folio_size(& !! 2423 /*
3064 pos = round_up((u64)pos + 1, !! 2424 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
3065 start = folio_seek_hole_data( !! 2425 * anything, so we only pin the file and drop the mmap_sem if only
3066 seek_data); !! 2426 * FAULT_FLAG_ALLOW_RETRY is set.
3067 if (start < pos) !! 2427 */
3068 goto unlock; !! 2428 if ((flags & (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT)) ==
3069 if (start >= end) !! 2429 FAULT_FLAG_ALLOW_RETRY) {
3070 break; !! 2430 fpin = get_file(vmf->vma->vm_file);
3071 if (seek_size > PAGE_SIZE) !! 2431 up_read(&vmf->vma->vm_mm->mmap_sem);
3072 xas_set(&xas, pos >> <<
3073 if (!xa_is_value(folio)) <<
3074 folio_put(folio); <<
3075 } 2432 }
3076 if (seek_data) !! 2433 return fpin;
3077 start = -ENXIO; <<
3078 unlock: <<
3079 rcu_read_unlock(); <<
3080 if (folio && !xa_is_value(folio)) <<
3081 folio_put(folio); <<
3082 if (start > end) <<
3083 return end; <<
3084 return start; <<
3085 } 2434 }
3086 2435
3087 #ifdef CONFIG_MMU <<
3088 #define MMAP_LOTSAMISS (100) <<
3089 /* 2436 /*
3090 * lock_folio_maybe_drop_mmap - lock the page !! 2437 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_sem
3091 * @vmf - the vm_fault for this fault. 2438 * @vmf - the vm_fault for this fault.
3092 * @folio - the folio to lock. !! 2439 * @page - the page to lock.
3093 * @fpin - the pointer to the file we may pin 2440 * @fpin - the pointer to the file we may pin (or is already pinned).
3094 * 2441 *
3095 * This works similar to lock_folio_or_retry !! 2442 * This works similar to lock_page_or_retry in that it can drop the mmap_sem.
3096 * mmap_lock. It differs in that it actually !! 2443 * It differs in that it actually returns the page locked if it returns 1 and 0
3097 * if it returns 1 and 0 if it couldn't lock !! 2444 * if it couldn't lock the page. If we did have to drop the mmap_sem then fpin
3098 * to drop the mmap_lock then fpin will point !! 2445 * will point to the pinned file and needs to be fput()'ed at a later point.
3099 * needs to be fput()'ed at a later point. <<
3100 */ 2446 */
3101 static int lock_folio_maybe_drop_mmap(struct !! 2447 static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page,
3102 struct f 2448 struct file **fpin)
3103 { 2449 {
3104 if (folio_trylock(folio)) !! 2450 if (trylock_page(page))
3105 return 1; 2451 return 1;
3106 2452
3107 /* 2453 /*
3108 * NOTE! This will make us return wit 2454 * NOTE! This will make us return with VM_FAULT_RETRY, but with
3109 * the fault lock still held. That's !! 2455 * the mmap_sem still held. That's how FAULT_FLAG_RETRY_NOWAIT
3110 * is supposed to work. We have way t 2456 * is supposed to work. We have way too many special cases..
3111 */ 2457 */
3112 if (vmf->flags & FAULT_FLAG_RETRY_NOW 2458 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
3113 return 0; 2459 return 0;
3114 2460
3115 *fpin = maybe_unlock_mmap_for_io(vmf, 2461 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
3116 if (vmf->flags & FAULT_FLAG_KILLABLE) 2462 if (vmf->flags & FAULT_FLAG_KILLABLE) {
3117 if (__folio_lock_killable(fol !! 2463 if (__lock_page_killable(page)) {
3118 /* 2464 /*
3119 * We didn't have the !! 2465 * We didn't have the right flags to drop the mmap_sem,
3120 * fault lock, but al !! 2466 * but all fault_handlers only check for fatal signals
3121 * for fatal signals !! 2467 * if we return VM_FAULT_RETRY, so we need to drop the
3122 * so we need to drop !! 2468 * mmap_sem here and return 0 if we don't have a fpin.
3123 * return 0 if we don <<
3124 */ 2469 */
3125 if (*fpin == NULL) 2470 if (*fpin == NULL)
3126 release_fault !! 2471 up_read(&vmf->vma->vm_mm->mmap_sem);
3127 return 0; 2472 return 0;
3128 } 2473 }
3129 } else 2474 } else
3130 __folio_lock(folio); !! 2475 __lock_page(page);
3131 <<
3132 return 1; 2476 return 1;
3133 } 2477 }
3134 2478
>> 2479
3135 /* 2480 /*
3136 * Synchronous readahead happens when we don' 2481 * 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 2482 * 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 2483 * 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 2484 * 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 2485 * returned needs to be fput()'ed when we're done with it.
3141 */ 2486 */
3142 static struct file *do_sync_mmap_readahead(st 2487 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
3143 { 2488 {
3144 struct file *file = vmf->vma->vm_file 2489 struct file *file = vmf->vma->vm_file;
3145 struct file_ra_state *ra = &file->f_r 2490 struct file_ra_state *ra = &file->f_ra;
3146 struct address_space *mapping = file- 2491 struct address_space *mapping = file->f_mapping;
3147 DEFINE_READAHEAD(ractl, file, ra, map <<
3148 struct file *fpin = NULL; 2492 struct file *fpin = NULL;
3149 unsigned long vm_flags = vmf->vma->vm !! 2493 pgoff_t offset = vmf->pgoff;
3150 unsigned int mmap_miss; <<
3151 <<
3152 #ifdef CONFIG_TRANSPARENT_HUGEPAGE <<
3153 /* Use the readahead code, even if re <<
3154 if ((vm_flags & VM_HUGEPAGE) && HPAGE <<
3155 fpin = maybe_unlock_mmap_for_ <<
3156 ractl._index &= ~((unsigned l <<
3157 ra->size = HPAGE_PMD_NR; <<
3158 /* <<
3159 * Fetch two PMD folios, so w <<
3160 * readahead, unless we've be <<
3161 */ <<
3162 if (!(vm_flags & VM_RAND_READ <<
3163 ra->size *= 2; <<
3164 ra->async_size = HPAGE_PMD_NR <<
3165 page_cache_ra_order(&ractl, r <<
3166 return fpin; <<
3167 } <<
3168 #endif <<
3169 2494
3170 /* If we don't want any read-ahead, d 2495 /* If we don't want any read-ahead, don't bother */
3171 if (vm_flags & VM_RAND_READ) !! 2496 if (vmf->vma->vm_flags & VM_RAND_READ)
3172 return fpin; 2497 return fpin;
3173 if (!ra->ra_pages) 2498 if (!ra->ra_pages)
3174 return fpin; 2499 return fpin;
3175 2500
3176 if (vm_flags & VM_SEQ_READ) { !! 2501 if (vmf->vma->vm_flags & VM_SEQ_READ) {
3177 fpin = maybe_unlock_mmap_for_ 2502 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3178 page_cache_sync_ra(&ractl, ra !! 2503 page_cache_sync_readahead(mapping, ra, file, offset,
>> 2504 ra->ra_pages);
3179 return fpin; 2505 return fpin;
3180 } 2506 }
3181 2507
3182 /* Avoid banging the cache line if no 2508 /* Avoid banging the cache line if not needed */
3183 mmap_miss = READ_ONCE(ra->mmap_miss); !! 2509 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
3184 if (mmap_miss < MMAP_LOTSAMISS * 10) !! 2510 ra->mmap_miss++;
3185 WRITE_ONCE(ra->mmap_miss, ++m <<
3186 2511
3187 /* 2512 /*
3188 * Do we miss much more than hit in t 2513 * Do we miss much more than hit in this file? If so,
3189 * stop bothering with read-ahead. It 2514 * stop bothering with read-ahead. It will only hurt.
3190 */ 2515 */
3191 if (mmap_miss > MMAP_LOTSAMISS) !! 2516 if (ra->mmap_miss > MMAP_LOTSAMISS)
3192 return fpin; 2517 return fpin;
3193 2518
3194 /* 2519 /*
3195 * mmap read-around 2520 * mmap read-around
3196 */ 2521 */
3197 fpin = maybe_unlock_mmap_for_io(vmf, 2522 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3198 ra->start = max_t(long, 0, vmf->pgoff !! 2523 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
3199 ra->size = ra->ra_pages; 2524 ra->size = ra->ra_pages;
3200 ra->async_size = ra->ra_pages / 4; 2525 ra->async_size = ra->ra_pages / 4;
3201 ractl._index = ra->start; !! 2526 ra_submit(ra, mapping, file);
3202 page_cache_ra_order(&ractl, ra, 0); <<
3203 return fpin; 2527 return fpin;
3204 } 2528 }
3205 2529
3206 /* 2530 /*
3207 * Asynchronous readahead happens when we fin 2531 * Asynchronous readahead happens when we find the page and PG_readahead,
3208 * so we want to possibly extend the readahea 2532 * 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 !! 2533 * was pinned if we have to drop the mmap_sem in order to do IO.
3210 */ 2534 */
3211 static struct file *do_async_mmap_readahead(s 2535 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
3212 s !! 2536 struct page *page)
3213 { 2537 {
3214 struct file *file = vmf->vma->vm_file 2538 struct file *file = vmf->vma->vm_file;
3215 struct file_ra_state *ra = &file->f_r 2539 struct file_ra_state *ra = &file->f_ra;
3216 DEFINE_READAHEAD(ractl, file, ra, fil !! 2540 struct address_space *mapping = file->f_mapping;
3217 struct file *fpin = NULL; 2541 struct file *fpin = NULL;
3218 unsigned int mmap_miss; !! 2542 pgoff_t offset = vmf->pgoff;
3219 2543
3220 /* If we don't want any read-ahead, d 2544 /* If we don't want any read-ahead, don't bother */
3221 if (vmf->vma->vm_flags & VM_RAND_READ !! 2545 if (vmf->vma->vm_flags & VM_RAND_READ)
3222 return fpin; 2546 return fpin;
3223 !! 2547 if (ra->mmap_miss > 0)
3224 mmap_miss = READ_ONCE(ra->mmap_miss); !! 2548 ra->mmap_miss--;
3225 if (mmap_miss) !! 2549 if (PageReadahead(page)) {
3226 WRITE_ONCE(ra->mmap_miss, --m <<
3227 <<
3228 if (folio_test_readahead(folio)) { <<
3229 fpin = maybe_unlock_mmap_for_ 2550 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3230 page_cache_async_ra(&ractl, f !! 2551 page_cache_async_readahead(mapping, ra, file,
>> 2552 page, offset, ra->ra_pages);
3231 } 2553 }
3232 return fpin; 2554 return fpin;
3233 } 2555 }
3234 2556
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 /** 2557 /**
3279 * filemap_fault - read in file data for page 2558 * filemap_fault - read in file data for page fault handling
3280 * @vmf: struct vm_fault containing de 2559 * @vmf: struct vm_fault containing details of the fault
3281 * 2560 *
3282 * filemap_fault() is invoked via the vma ope 2561 * filemap_fault() is invoked via the vma operations vector for a
3283 * mapped memory region to read in file data 2562 * mapped memory region to read in file data during a page fault.
3284 * 2563 *
3285 * The goto's are kind of ugly, but this stre 2564 * 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 2565 * it in the page cache, and handles the special cases reasonably without
3287 * having a lot of duplicated code. 2566 * having a lot of duplicated code.
3288 * 2567 *
3289 * vma->vm_mm->mmap_lock must be held on entr !! 2568 * vma->vm_mm->mmap_sem must be held on entry.
3290 * 2569 *
3291 * If our return value has VM_FAULT_RETRY set !! 2570 * If our return value has VM_FAULT_RETRY set, it's because
3292 * may be dropped before doing I/O or by lock !! 2571 * lock_page_or_retry() returned 0.
>> 2572 * The mmap_sem has usually been released in this case.
>> 2573 * See __lock_page_or_retry() for the exception.
3293 * 2574 *
3294 * If our return value does not have VM_FAULT !! 2575 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
3295 * has not been released. 2576 * has not been released.
3296 * 2577 *
3297 * We never return with VM_FAULT_RETRY and a 2578 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3298 * 2579 *
3299 * Return: bitwise-OR of %VM_FAULT_ codes. 2580 * Return: bitwise-OR of %VM_FAULT_ codes.
3300 */ 2581 */
3301 vm_fault_t filemap_fault(struct vm_fault *vmf 2582 vm_fault_t filemap_fault(struct vm_fault *vmf)
3302 { 2583 {
3303 int error; 2584 int error;
3304 struct file *file = vmf->vma->vm_file 2585 struct file *file = vmf->vma->vm_file;
3305 struct file *fpin = NULL; 2586 struct file *fpin = NULL;
3306 struct address_space *mapping = file- 2587 struct address_space *mapping = file->f_mapping;
>> 2588 struct file_ra_state *ra = &file->f_ra;
3307 struct inode *inode = mapping->host; 2589 struct inode *inode = mapping->host;
3308 pgoff_t max_idx, index = vmf->pgoff; !! 2590 pgoff_t offset = vmf->pgoff;
3309 struct folio *folio; !! 2591 pgoff_t max_off;
>> 2592 struct page *page;
3310 vm_fault_t ret = 0; 2593 vm_fault_t ret = 0;
3311 bool mapping_locked = false; <<
3312 2594
3313 max_idx = DIV_ROUND_UP(i_size_read(in !! 2595 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3314 if (unlikely(index >= max_idx)) !! 2596 if (unlikely(offset >= max_off))
3315 return VM_FAULT_SIGBUS; 2597 return VM_FAULT_SIGBUS;
3316 2598
3317 trace_mm_filemap_fault(mapping, index <<
3318 <<
3319 /* 2599 /*
3320 * Do we have something in the page c 2600 * Do we have something in the page cache already?
3321 */ 2601 */
3322 folio = filemap_get_folio(mapping, in !! 2602 page = find_get_page(mapping, offset);
3323 if (likely(!IS_ERR(folio))) { !! 2603 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
3324 /* 2604 /*
3325 * We found the page, so try !! 2605 * We found the page, so try async readahead before
3326 * the lock. !! 2606 * waiting for the lock.
3327 */ 2607 */
3328 if (!(vmf->flags & FAULT_FLAG !! 2608 fpin = do_async_mmap_readahead(vmf, page);
3329 fpin = do_async_mmap_ !! 2609 } else if (!page) {
3330 if (unlikely(!folio_test_upto <<
3331 filemap_invalidate_lo <<
3332 mapping_locked = true <<
3333 } <<
3334 } else { <<
3335 ret = filemap_fault_recheck_p <<
3336 if (unlikely(ret)) <<
3337 return ret; <<
3338 <<
3339 /* No page in the page cache 2610 /* No page in the page cache at all */
3340 count_vm_event(PGMAJFAULT); 2611 count_vm_event(PGMAJFAULT);
3341 count_memcg_event_mm(vmf->vma 2612 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3342 ret = VM_FAULT_MAJOR; 2613 ret = VM_FAULT_MAJOR;
3343 fpin = do_sync_mmap_readahead 2614 fpin = do_sync_mmap_readahead(vmf);
3344 retry_find: 2615 retry_find:
3345 /* !! 2616 page = pagecache_get_page(mapping, offset,
3346 * See comment in filemap_cre <<
3347 * invalidate_lock <<
3348 */ <<
3349 if (!mapping_locked) { <<
3350 filemap_invalidate_lo <<
3351 mapping_locked = true <<
3352 } <<
3353 folio = __filemap_get_folio(m <<
3354 FGP 2617 FGP_CREAT|FGP_FOR_MMAP,
3355 vmf 2618 vmf->gfp_mask);
3356 if (IS_ERR(folio)) { !! 2619 if (!page) {
3357 if (fpin) 2620 if (fpin)
3358 goto out_retr 2621 goto out_retry;
3359 filemap_invalidate_un !! 2622 return vmf_error(-ENOMEM);
3360 return VM_FAULT_OOM; <<
3361 } 2623 }
3362 } 2624 }
3363 2625
3364 if (!lock_folio_maybe_drop_mmap(vmf, !! 2626 if (!lock_page_maybe_drop_mmap(vmf, page, &fpin))
3365 goto out_retry; 2627 goto out_retry;
3366 2628
3367 /* Did it get truncated? */ 2629 /* Did it get truncated? */
3368 if (unlikely(folio->mapping != mappin !! 2630 if (unlikely(page->mapping != mapping)) {
3369 folio_unlock(folio); !! 2631 unlock_page(page);
3370 folio_put(folio); !! 2632 put_page(page);
3371 goto retry_find; 2633 goto retry_find;
3372 } 2634 }
3373 VM_BUG_ON_FOLIO(!folio_contains(folio !! 2635 VM_BUG_ON_PAGE(page->index != offset, page);
3374 2636
3375 /* 2637 /*
3376 * We have a locked folio in the page !! 2638 * 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 !! 2639 * that it's up-to-date. If not, it is going to be due to an error.
3378 * or because readahead was otherwise <<
3379 */ 2640 */
3380 if (unlikely(!folio_test_uptodate(fol !! 2641 if (unlikely(!PageUptodate(page)))
3381 /* <<
3382 * If the invalidate lock is <<
3383 * and uptodate and now it is <<
3384 * didn't hold the page lock <<
3385 * everything, get the invali <<
3386 */ <<
3387 if (!mapping_locked) { <<
3388 folio_unlock(folio); <<
3389 folio_put(folio); <<
3390 goto retry_find; <<
3391 } <<
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; 2642 goto page_not_uptodate;
3399 } <<
3400 2643
3401 /* 2644 /*
3402 * We've made it this far and we had !! 2645 * We've made it this far and we had to drop our mmap_sem, now is the
3403 * time to return to the upper layer 2646 * time to return to the upper layer and have it re-find the vma and
3404 * redo the fault. 2647 * redo the fault.
3405 */ 2648 */
3406 if (fpin) { 2649 if (fpin) {
3407 folio_unlock(folio); !! 2650 unlock_page(page);
3408 goto out_retry; 2651 goto out_retry;
3409 } 2652 }
3410 if (mapping_locked) <<
3411 filemap_invalidate_unlock_sha <<
3412 2653
3413 /* 2654 /*
3414 * Found the page and have a referenc 2655 * Found the page and have a reference on it.
3415 * We must recheck i_size under page 2656 * We must recheck i_size under page lock.
3416 */ 2657 */
3417 max_idx = DIV_ROUND_UP(i_size_read(in !! 2658 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3418 if (unlikely(index >= max_idx)) { !! 2659 if (unlikely(offset >= max_off)) {
3419 folio_unlock(folio); !! 2660 unlock_page(page);
3420 folio_put(folio); !! 2661 put_page(page);
3421 return VM_FAULT_SIGBUS; 2662 return VM_FAULT_SIGBUS;
3422 } 2663 }
3423 2664
3424 vmf->page = folio_file_page(folio, in !! 2665 vmf->page = page;
3425 return ret | VM_FAULT_LOCKED; 2666 return ret | VM_FAULT_LOCKED;
3426 2667
3427 page_not_uptodate: 2668 page_not_uptodate:
3428 /* 2669 /*
3429 * Umm, take care of errors if the pa 2670 * Umm, take care of errors if the page isn't up-to-date.
3430 * Try to re-read it _once_. We do th 2671 * Try to re-read it _once_. We do this synchronously,
3431 * because there really aren't any pe 2672 * because there really aren't any performance issues here
3432 * and we need to check for errors. 2673 * and we need to check for errors.
3433 */ 2674 */
>> 2675 ClearPageError(page);
3434 fpin = maybe_unlock_mmap_for_io(vmf, 2676 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3435 error = filemap_read_folio(file, mapp !! 2677 error = mapping->a_ops->readpage(file, page);
>> 2678 if (!error) {
>> 2679 wait_on_page_locked(page);
>> 2680 if (!PageUptodate(page))
>> 2681 error = -EIO;
>> 2682 }
3436 if (fpin) 2683 if (fpin)
3437 goto out_retry; 2684 goto out_retry;
3438 folio_put(folio); !! 2685 put_page(page);
3439 2686
3440 if (!error || error == AOP_TRUNCATED_ 2687 if (!error || error == AOP_TRUNCATED_PAGE)
3441 goto retry_find; 2688 goto retry_find;
3442 filemap_invalidate_unlock_shared(mapp <<
3443 2689
>> 2690 /* Things didn't work out. Return zero to tell the mm layer so. */
>> 2691 shrink_readahead_size_eio(file, ra);
3444 return VM_FAULT_SIGBUS; 2692 return VM_FAULT_SIGBUS;
3445 2693
3446 out_retry: 2694 out_retry:
3447 /* 2695 /*
3448 * We dropped the mmap_lock, we need !! 2696 * We dropped the mmap_sem, we need to return to the fault handler to
3449 * re-find the vma and come back and 2697 * re-find the vma and come back and find our hopefully still populated
3450 * page. 2698 * page.
3451 */ 2699 */
3452 if (!IS_ERR(folio)) !! 2700 if (page)
3453 folio_put(folio); !! 2701 put_page(page);
3454 if (mapping_locked) <<
3455 filemap_invalidate_unlock_sha <<
3456 if (fpin) 2702 if (fpin)
3457 fput(fpin); 2703 fput(fpin);
3458 return ret | VM_FAULT_RETRY; 2704 return ret | VM_FAULT_RETRY;
3459 } 2705 }
3460 EXPORT_SYMBOL(filemap_fault); 2706 EXPORT_SYMBOL(filemap_fault);
3461 2707
3462 static bool filemap_map_pmd(struct vm_fault * !! 2708 void filemap_map_pages(struct vm_fault *vmf,
3463 pgoff_t start) !! 2709 pgoff_t start_pgoff, pgoff_t end_pgoff)
3464 { 2710 {
3465 struct mm_struct *mm = vmf->vma->vm_m !! 2711 struct file *file = vmf->vma->vm_file;
3466 !! 2712 struct address_space *mapping = file->f_mapping;
3467 /* Huge page is mapped? No need to pr !! 2713 pgoff_t last_pgoff = start_pgoff;
3468 if (pmd_trans_huge(*vmf->pmd)) { <<
3469 folio_unlock(folio); <<
3470 folio_put(folio); <<
3471 return true; <<
3472 } <<
3473 <<
3474 if (pmd_none(*vmf->pmd) && folio_test <<
3475 struct page *page = folio_fil <<
3476 vm_fault_t ret = do_set_pmd(v <<
3477 if (!ret) { <<
3478 /* The page is mapped <<
3479 folio_unlock(folio); <<
3480 return true; <<
3481 } <<
3482 } <<
3483 <<
3484 if (pmd_none(*vmf->pmd) && vmf->preal <<
3485 pmd_install(mm, vmf->pmd, &vm <<
3486 <<
3487 return false; <<
3488 } <<
3489 <<
3490 static struct folio *next_uptodate_folio(stru <<
3491 struct address_space *mapping <<
3492 { <<
3493 struct folio *folio = xas_next_entry( <<
3494 unsigned long max_idx; 2714 unsigned long max_idx;
>> 2715 XA_STATE(xas, &mapping->i_pages, start_pgoff);
>> 2716 struct page *head, *page;
3495 2717
3496 do { !! 2718 rcu_read_lock();
3497 if (!folio) !! 2719 xas_for_each(&xas, page, end_pgoff) {
3498 return NULL; !! 2720 if (xas_retry(&xas, page))
3499 if (xas_retry(xas, folio)) <<
3500 continue; <<
3501 if (xa_is_value(folio)) <<
3502 continue; <<
3503 if (folio_test_locked(folio)) <<
3504 continue; <<
3505 if (!folio_try_get(folio)) <<
3506 continue; 2721 continue;
3507 /* Has the page moved or been !! 2722 if (xa_is_value(page))
3508 if (unlikely(folio != xas_rel !! 2723 goto next;
3509 goto skip; <<
3510 if (!folio_test_uptodate(foli <<
3511 goto skip; <<
3512 if (!folio_trylock(folio)) <<
3513 goto skip; <<
3514 if (folio->mapping != mapping <<
3515 goto unlock; <<
3516 if (!folio_test_uptodate(foli <<
3517 goto unlock; <<
3518 max_idx = DIV_ROUND_UP(i_size <<
3519 if (xas->xa_index >= max_idx) <<
3520 goto unlock; <<
3521 return folio; <<
3522 unlock: <<
3523 folio_unlock(folio); <<
3524 skip: <<
3525 folio_put(folio); <<
3526 } while ((folio = xas_next_entry(xas, <<
3527 2724
3528 return NULL; !! 2725 head = compound_head(page);
3529 } <<
3530 <<
3531 /* <<
3532 * Map page range [start_page, start_page + n <<
3533 * start_page is gotten from start by folio_p <<
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 { <<
3540 vm_fault_t ret = 0; <<
3541 struct page *page = folio_page(folio, <<
3542 unsigned int count = 0; <<
3543 pte_t *old_ptep = vmf->pte; <<
3544 <<
3545 do { <<
3546 if (PageHWPoison(page + count <<
3547 goto skip; <<
3548 2726
3549 /* 2727 /*
3550 * If there are too many foli !! 2728 * Check for a locked page first, as a speculative
3551 * in a file, they will proba !! 2729 * reference may adversely influence page migration.
3552 * In such situation, read-ah <<
3553 * Don't decrease mmap_miss i <<
3554 * we can stop read-ahead. <<
3555 */ 2730 */
3556 if (!folio_test_workingset(fo !! 2731 if (PageLocked(head))
3557 (*mmap_miss)++; !! 2732 goto next;
>> 2733 if (!page_cache_get_speculative(head))
>> 2734 goto next;
3558 2735
3559 /* !! 2736 /* The page was split under us? */
3560 * NOTE: If there're PTE mark !! 2737 if (compound_head(page) != head)
3561 * handled in the specific fa <<
3562 * fault-around logic. <<
3563 */ <<
3564 if (!pte_none(ptep_get(&vmf-> <<
3565 goto skip; 2738 goto skip;
3566 2739
3567 count++; !! 2740 /* Has the page moved? */
3568 continue; !! 2741 if (unlikely(page != xas_reload(&xas)))
3569 skip: !! 2742 goto 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 } <<
3597 <<
3598 static vm_fault_t filemap_map_order0_folio(st <<
3599 struct folio *folio, unsigned <<
3600 unsigned long *rss, unsigned <<
3601 { <<
3602 vm_fault_t ret = 0; <<
3603 struct page *page = &folio->page; <<
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 } <<
3629 <<
3630 vm_fault_t filemap_map_pages(struct vm_fault <<
3631 pgoff_t start_pg <<
3632 { <<
3633 struct vm_area_struct *vma = vmf->vma <<
3634 struct file *file = vma->vm_file; <<
3635 struct address_space *mapping = file- <<
3636 pgoff_t file_end, last_pgoff = start_ <<
3637 unsigned long addr; <<
3638 XA_STATE(xas, &mapping->i_pages, star <<
3639 struct folio *folio; <<
3640 vm_fault_t ret = 0; <<
3641 unsigned long rss = 0; <<
3642 unsigned int nr_pages = 0, mmap_miss <<
3643 <<
3644 rcu_read_lock(); <<
3645 folio = next_uptodate_folio(&xas, map <<
3646 if (!folio) <<
3647 goto out; <<
3648 2743
3649 if (filemap_map_pmd(vmf, folio, start !! 2744 if (!PageUptodate(page) ||
3650 ret = VM_FAULT_NOPAGE; !! 2745 PageReadahead(page) ||
3651 goto out; !! 2746 PageHWPoison(page))
3652 } !! 2747 goto skip;
>> 2748 if (!trylock_page(page))
>> 2749 goto skip;
3653 2750
3654 addr = vma->vm_start + ((start_pgoff !! 2751 if (page->mapping != mapping || !PageUptodate(page))
3655 vmf->pte = pte_offset_map_lock(vma->v !! 2752 goto unlock;
3656 if (!vmf->pte) { <<
3657 folio_unlock(folio); <<
3658 folio_put(folio); <<
3659 goto out; <<
3660 } <<
3661 2753
3662 file_end = DIV_ROUND_UP(i_size_read(m !! 2754 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3663 if (end_pgoff > file_end) !! 2755 if (page->index >= max_idx)
3664 end_pgoff = file_end; !! 2756 goto unlock;
3665 2757
3666 folio_type = mm_counter_file(folio); !! 2758 if (file->f_ra.mmap_miss > 0)
3667 do { !! 2759 file->f_ra.mmap_miss--;
3668 unsigned long end; <<
3669 2760
3670 addr += (xas.xa_index - last_ !! 2761 vmf->address += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3671 vmf->pte += xas.xa_index - la !! 2762 if (vmf->pte)
>> 2763 vmf->pte += xas.xa_index - last_pgoff;
3672 last_pgoff = xas.xa_index; 2764 last_pgoff = xas.xa_index;
3673 end = folio_next_index(folio) !! 2765 if (alloc_set_pte(vmf, NULL, page))
3674 nr_pages = min(end, end_pgoff !! 2766 goto unlock;
3675 !! 2767 unlock_page(page);
3676 if (!folio_test_large(folio)) !! 2768 goto next;
3677 ret |= filemap_map_or !! 2769 unlock:
3678 folio !! 2770 unlock_page(page);
3679 else !! 2771 skip:
3680 ret |= filemap_map_fo !! 2772 put_page(page);
3681 xas.x !! 2773 next:
3682 nr_pa !! 2774 /* Huge page is mapped? No need to proceed. */
3683 !! 2775 if (pmd_trans_huge(*vmf->pmd))
3684 folio_unlock(folio); !! 2776 break;
3685 folio_put(folio); !! 2777 }
3686 } while ((folio = next_uptodate_folio <<
3687 add_mm_counter(vma->vm_mm, folio_type <<
3688 pte_unmap_unlock(vmf->pte, vmf->ptl); <<
3689 trace_mm_filemap_map_pages(mapping, s <<
3690 out: <<
3691 rcu_read_unlock(); 2778 rcu_read_unlock();
3692 <<
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; <<
3700 } 2779 }
3701 EXPORT_SYMBOL(filemap_map_pages); 2780 EXPORT_SYMBOL(filemap_map_pages);
3702 2781
3703 vm_fault_t filemap_page_mkwrite(struct vm_fau 2782 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3704 { 2783 {
3705 struct address_space *mapping = vmf-> !! 2784 struct page *page = vmf->page;
3706 struct folio *folio = page_folio(vmf- !! 2785 struct inode *inode = file_inode(vmf->vma->vm_file);
3707 vm_fault_t ret = VM_FAULT_LOCKED; 2786 vm_fault_t ret = VM_FAULT_LOCKED;
3708 2787
3709 sb_start_pagefault(mapping->host->i_s !! 2788 sb_start_pagefault(inode->i_sb);
3710 file_update_time(vmf->vma->vm_file); 2789 file_update_time(vmf->vma->vm_file);
3711 folio_lock(folio); !! 2790 lock_page(page);
3712 if (folio->mapping != mapping) { !! 2791 if (page->mapping != inode->i_mapping) {
3713 folio_unlock(folio); !! 2792 unlock_page(page);
3714 ret = VM_FAULT_NOPAGE; 2793 ret = VM_FAULT_NOPAGE;
3715 goto out; 2794 goto out;
3716 } 2795 }
3717 /* 2796 /*
3718 * We mark the folio dirty already he !! 2797 * We mark the page dirty already here so that when freeze is in
3719 * progress, we are guaranteed that w 2798 * progress, we are guaranteed that writeback during freezing will
3720 * see the dirty folio and writeprote !! 2799 * see the dirty page and writeprotect it again.
3721 */ 2800 */
3722 folio_mark_dirty(folio); !! 2801 set_page_dirty(page);
3723 folio_wait_stable(folio); !! 2802 wait_for_stable_page(page);
3724 out: 2803 out:
3725 sb_end_pagefault(mapping->host->i_sb) !! 2804 sb_end_pagefault(inode->i_sb);
3726 return ret; 2805 return ret;
3727 } 2806 }
3728 2807
3729 const struct vm_operations_struct generic_fil 2808 const struct vm_operations_struct generic_file_vm_ops = {
3730 .fault = filemap_fault, 2809 .fault = filemap_fault,
3731 .map_pages = filemap_map_pages, 2810 .map_pages = filemap_map_pages,
3732 .page_mkwrite = filemap_page_mkwrit 2811 .page_mkwrite = filemap_page_mkwrite,
3733 }; 2812 };
3734 2813
3735 /* This is used for a general mmap of a disk 2814 /* This is used for a general mmap of a disk file */
3736 2815
3737 int generic_file_mmap(struct file *file, stru !! 2816 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
3738 { 2817 {
3739 struct address_space *mapping = file- 2818 struct address_space *mapping = file->f_mapping;
3740 2819
3741 if (!mapping->a_ops->read_folio) !! 2820 if (!mapping->a_ops->readpage)
3742 return -ENOEXEC; 2821 return -ENOEXEC;
3743 file_accessed(file); 2822 file_accessed(file);
3744 vma->vm_ops = &generic_file_vm_ops; 2823 vma->vm_ops = &generic_file_vm_ops;
3745 return 0; 2824 return 0;
3746 } 2825 }
3747 2826
3748 /* 2827 /*
3749 * This is for filesystems which do not imple 2828 * This is for filesystems which do not implement ->writepage.
3750 */ 2829 */
3751 int generic_file_readonly_mmap(struct file *f 2830 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3752 { 2831 {
3753 if (vma_is_shared_maywrite(vma)) !! 2832 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3754 return -EINVAL; 2833 return -EINVAL;
3755 return generic_file_mmap(file, vma); 2834 return generic_file_mmap(file, vma);
3756 } 2835 }
3757 #else 2836 #else
3758 vm_fault_t filemap_page_mkwrite(struct vm_fau 2837 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3759 { 2838 {
3760 return VM_FAULT_SIGBUS; 2839 return VM_FAULT_SIGBUS;
3761 } 2840 }
3762 int generic_file_mmap(struct file *file, stru !! 2841 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
3763 { 2842 {
3764 return -ENOSYS; 2843 return -ENOSYS;
3765 } 2844 }
3766 int generic_file_readonly_mmap(struct file *f !! 2845 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
3767 { 2846 {
3768 return -ENOSYS; 2847 return -ENOSYS;
3769 } 2848 }
3770 #endif /* CONFIG_MMU */ 2849 #endif /* CONFIG_MMU */
3771 2850
3772 EXPORT_SYMBOL(filemap_page_mkwrite); 2851 EXPORT_SYMBOL(filemap_page_mkwrite);
3773 EXPORT_SYMBOL(generic_file_mmap); 2852 EXPORT_SYMBOL(generic_file_mmap);
3774 EXPORT_SYMBOL(generic_file_readonly_mmap); 2853 EXPORT_SYMBOL(generic_file_readonly_mmap);
3775 2854
3776 static struct folio *do_read_cache_folio(stru !! 2855 static struct page *wait_on_page_read(struct page *page)
3777 pgoff_t index, filler_t fille <<
3778 { 2856 {
3779 struct folio *folio; !! 2857 if (!IS_ERR(page)) {
3780 int err; !! 2858 wait_on_page_locked(page);
>> 2859 if (!PageUptodate(page)) {
>> 2860 put_page(page);
>> 2861 page = ERR_PTR(-EIO);
>> 2862 }
>> 2863 }
>> 2864 return page;
>> 2865 }
3781 2866
3782 if (!filler) !! 2867 static struct page *do_read_cache_page(struct address_space *mapping,
3783 filler = mapping->a_ops->read !! 2868 pgoff_t index,
>> 2869 int (*filler)(void *, struct page *),
>> 2870 void *data,
>> 2871 gfp_t gfp)
>> 2872 {
>> 2873 struct page *page;
>> 2874 int err;
3784 repeat: 2875 repeat:
3785 folio = filemap_get_folio(mapping, in !! 2876 page = find_get_page(mapping, index);
3786 if (IS_ERR(folio)) { !! 2877 if (!page) {
3787 folio = filemap_alloc_folio(g !! 2878 page = __page_cache_alloc(gfp);
3788 m !! 2879 if (!page)
3789 if (!folio) <<
3790 return ERR_PTR(-ENOME 2880 return ERR_PTR(-ENOMEM);
3791 index = mapping_align_index(m !! 2881 err = add_to_page_cache_lru(page, mapping, index, gfp);
3792 err = filemap_add_folio(mappi <<
3793 if (unlikely(err)) { 2882 if (unlikely(err)) {
3794 folio_put(folio); !! 2883 put_page(page);
3795 if (err == -EEXIST) 2884 if (err == -EEXIST)
3796 goto repeat; 2885 goto repeat;
3797 /* Presumably ENOMEM 2886 /* Presumably ENOMEM for xarray node */
3798 return ERR_PTR(err); 2887 return ERR_PTR(err);
3799 } 2888 }
3800 2889
3801 goto filler; !! 2890 filler:
>> 2891 err = filler(data, page);
>> 2892 if (err < 0) {
>> 2893 put_page(page);
>> 2894 return ERR_PTR(err);
>> 2895 }
>> 2896
>> 2897 page = wait_on_page_read(page);
>> 2898 if (IS_ERR(page))
>> 2899 return page;
>> 2900 goto out;
3802 } 2901 }
3803 if (folio_test_uptodate(folio)) !! 2902 if (PageUptodate(page))
3804 goto out; 2903 goto out;
3805 2904
3806 if (!folio_trylock(folio)) { !! 2905 /*
3807 folio_put_wait_locked(folio, !! 2906 * Page is not up to date and may be locked due one of the following
3808 goto repeat; !! 2907 * case a: Page is being filled and the page lock is held
3809 } !! 2908 * case b: Read/write error clearing the page uptodate status
>> 2909 * case c: Truncation in progress (page locked)
>> 2910 * case d: Reclaim in progress
>> 2911 *
>> 2912 * Case a, the page will be up to date when the page is unlocked.
>> 2913 * There is no need to serialise on the page lock here as the page
>> 2914 * is pinned so the lock gives no additional protection. Even if the
>> 2915 * the page is truncated, the data is still valid if PageUptodate as
>> 2916 * it's a race vs truncate race.
>> 2917 * Case b, the page will not be up to date
>> 2918 * Case c, the page may be truncated but in itself, the data may still
>> 2919 * be valid after IO completes as it's a read vs truncate race. The
>> 2920 * operation must restart if the page is not uptodate on unlock but
>> 2921 * otherwise serialising on page lock to stabilise the mapping gives
>> 2922 * no additional guarantees to the caller as the page lock is
>> 2923 * released before return.
>> 2924 * Case d, similar to truncation. If reclaim holds the page lock, it
>> 2925 * will be a race with remove_mapping that determines if the mapping
>> 2926 * is valid on unlock but otherwise the data is valid and there is
>> 2927 * no need to serialise with page lock.
>> 2928 *
>> 2929 * As the page lock gives no additional guarantee, we optimistically
>> 2930 * wait on the page to be unlocked and check if it's up to date and
>> 2931 * use the page if it is. Otherwise, the page lock is required to
>> 2932 * distinguish between the different cases. The motivation is that we
>> 2933 * avoid spurious serialisations and wakeups when multiple processes
>> 2934 * wait on the same page for IO to complete.
>> 2935 */
>> 2936 wait_on_page_locked(page);
>> 2937 if (PageUptodate(page))
>> 2938 goto out;
>> 2939
>> 2940 /* Distinguish between all the cases under the safety of the lock */
>> 2941 lock_page(page);
3810 2942
3811 /* Folio was truncated from mapping * !! 2943 /* Case c or d, restart the operation */
3812 if (!folio->mapping) { !! 2944 if (!page->mapping) {
3813 folio_unlock(folio); !! 2945 unlock_page(page);
3814 folio_put(folio); !! 2946 put_page(page);
3815 goto repeat; 2947 goto repeat;
3816 } 2948 }
3817 2949
3818 /* Someone else locked and filled the 2950 /* Someone else locked and filled the page in a very small window */
3819 if (folio_test_uptodate(folio)) { !! 2951 if (PageUptodate(page)) {
3820 folio_unlock(folio); !! 2952 unlock_page(page);
3821 goto out; 2953 goto out;
3822 } 2954 }
3823 !! 2955 goto filler;
3824 filler: <<
3825 err = filemap_read_folio(file, filler <<
3826 if (err) { <<
3827 folio_put(folio); <<
3828 if (err == AOP_TRUNCATED_PAGE <<
3829 goto repeat; <<
3830 return ERR_PTR(err); <<
3831 } <<
3832 2956
3833 out: 2957 out:
3834 folio_mark_accessed(folio); !! 2958 mark_page_accessed(page);
3835 return folio; !! 2959 return page;
3836 } <<
3837 <<
3838 /** <<
3839 * read_cache_folio - Read into page cache, f <<
3840 * @mapping: The address_space to read from. <<
3841 * @index: The index to read. <<
3842 * @filler: Function to perform the read, or <<
3843 * @file: Passed to filler function, may be N <<
3844 * <<
3845 * Read one page into the page cache. If it <<
3846 * will contain @index, but it may not be the <<
3847 * <<
3848 * If the filler function returns an error, i <<
3849 * caller. <<
3850 * <<
3851 * Context: May sleep. Expects mapping->inva <<
3852 * Return: An uptodate folio on success, ERR_ <<
3853 */ <<
3854 struct folio *read_cache_folio(struct address <<
3855 filler_t filler, struct file <<
3856 { <<
3857 return do_read_cache_folio(mapping, i <<
3858 mapping_gfp_mask(mapp <<
3859 } 2960 }
3860 EXPORT_SYMBOL(read_cache_folio); <<
3861 2961
3862 /** 2962 /**
3863 * mapping_read_folio_gfp - Read into page ca !! 2963 * read_cache_page - read into page cache, fill it if needed
3864 * @mapping: The address_space for the fol !! 2964 * @mapping: the page's address_space
3865 * @index: The index that the allocated !! 2965 * @index: the page index
3866 * @gfp: The page allocator flags to u !! 2966 * @filler: function to perform the read
3867 * !! 2967 * @data: first arg to filler(data, page) function, often left as NULL
3868 * This is the same as "read_cache_folio(mapp <<
3869 * any new memory allocations done using the <<
3870 * 2968 *
3871 * The most likely error from this function i !! 2969 * Read into the page cache. If a page already exists, and PageUptodate() is
3872 * possible and so is EINTR. If ->read_folio !! 2970 * not set, try to fill the page and wait for it to become unlocked.
3873 * that will be returned to the caller. <<
3874 * 2971 *
3875 * The function expects mapping->invalidate_l !! 2972 * If the page does not get brought uptodate, return -EIO.
3876 * 2973 *
3877 * Return: Uptodate folio on success, ERR_PTR !! 2974 * Return: up to date page on success, ERR_PTR() on failure.
3878 */ 2975 */
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 <<
3887 pgoff_t index, filler_t *fill <<
3888 { <<
3889 struct folio *folio; <<
3890 <<
3891 folio = do_read_cache_folio(mapping, <<
3892 if (IS_ERR(folio)) <<
3893 return &folio->page; <<
3894 return folio_file_page(folio, index); <<
3895 } <<
3896 <<
3897 struct page *read_cache_page(struct address_s 2976 struct page *read_cache_page(struct address_space *mapping,
3898 pgoff_t index, filler !! 2977 pgoff_t index,
>> 2978 int (*filler)(void *, struct page *),
>> 2979 void *data)
3899 { 2980 {
3900 return do_read_cache_page(mapping, in !! 2981 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
3901 mapping_gfp_mask(mapp <<
3902 } 2982 }
3903 EXPORT_SYMBOL(read_cache_page); 2983 EXPORT_SYMBOL(read_cache_page);
3904 2984
3905 /** 2985 /**
3906 * read_cache_page_gfp - read into page cache 2986 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3907 * @mapping: the page's address_space 2987 * @mapping: the page's address_space
3908 * @index: the page index 2988 * @index: the page index
3909 * @gfp: the page allocator flags to u 2989 * @gfp: the page allocator flags to use if allocating
3910 * 2990 *
3911 * This is the same as "read_mapping_page(map 2991 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3912 * any new page allocations done using the sp 2992 * any new page allocations done using the specified allocation flags.
3913 * 2993 *
3914 * If the page does not get brought uptodate, 2994 * If the page does not get brought uptodate, return -EIO.
3915 * 2995 *
3916 * The function expects mapping->invalidate_l <<
3917 * <<
3918 * Return: up to date page on success, ERR_PT 2996 * Return: up to date page on success, ERR_PTR() on failure.
3919 */ 2997 */
3920 struct page *read_cache_page_gfp(struct addre 2998 struct page *read_cache_page_gfp(struct address_space *mapping,
3921 pgoff_t index 2999 pgoff_t index,
3922 gfp_t gfp) 3000 gfp_t gfp)
3923 { 3001 {
3924 return do_read_cache_page(mapping, in !! 3002 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
>> 3003
>> 3004 return do_read_cache_page(mapping, index, filler, NULL, gfp);
3925 } 3005 }
3926 EXPORT_SYMBOL(read_cache_page_gfp); 3006 EXPORT_SYMBOL(read_cache_page_gfp);
3927 3007
3928 /* 3008 /*
3929 * Warn about a page cache invalidation failu !! 3009 * Don't operate on ranges the page cache doesn't support, and don't exceed the
>> 3010 * LFS limits. If pos is under the limit it becomes a short access. If it
>> 3011 * exceeds the limit we return -EFBIG.
3930 */ 3012 */
3931 static void dio_warn_stale_pagecache(struct f !! 3013 static int generic_access_check_limits(struct file *file, loff_t pos,
>> 3014 loff_t *count)
3932 { 3015 {
3933 static DEFINE_RATELIMIT_STATE(_rs, 86 !! 3016 struct inode *inode = file->f_mapping->host;
3934 char pathname[128]; !! 3017 loff_t max_size = inode->i_sb->s_maxbytes;
3935 char *path; !! 3018
>> 3019 if (!(file->f_flags & O_LARGEFILE))
>> 3020 max_size = MAX_NON_LFS;
3936 3021
3937 errseq_set(&filp->f_mapping->wb_err, !! 3022 if (unlikely(pos >= max_size))
3938 if (__ratelimit(&_rs)) { !! 3023 return -EFBIG;
3939 path = file_path(filp, pathna !! 3024 *count = min(*count, max_size - pos);
3940 if (IS_ERR(path)) !! 3025 return 0;
3941 path = "(unknown)"; !! 3026 }
3942 pr_crit("Page cache invalidat !! 3027
3943 pr_crit("File: %s PID: %d Com !! 3028 static int generic_write_check_limits(struct file *file, loff_t pos,
3944 current->comm); !! 3029 loff_t *count)
>> 3030 {
>> 3031 loff_t limit = rlimit(RLIMIT_FSIZE);
>> 3032
>> 3033 if (limit != RLIM_INFINITY) {
>> 3034 if (pos >= limit) {
>> 3035 send_sig(SIGXFSZ, current, 0);
>> 3036 return -EFBIG;
>> 3037 }
>> 3038 *count = min(*count, limit - pos);
3945 } 3039 }
>> 3040
>> 3041 return generic_access_check_limits(file, pos, count);
3946 } 3042 }
3947 3043
3948 void kiocb_invalidate_post_direct_write(struc !! 3044 /*
>> 3045 * Performs necessary checks before doing a write
>> 3046 *
>> 3047 * Can adjust writing position or amount of bytes to write.
>> 3048 * Returns appropriate error code that caller should return or
>> 3049 * zero in case that write should be allowed.
>> 3050 */
>> 3051 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
3949 { 3052 {
3950 struct address_space *mapping = iocb- !! 3053 struct file *file = iocb->ki_filp;
>> 3054 struct inode *inode = file->f_mapping->host;
>> 3055 loff_t count;
>> 3056 int ret;
3951 3057
3952 if (mapping->nrpages && !! 3058 if (!iov_iter_count(from))
3953 invalidate_inode_pages2_range(map !! 3059 return 0;
3954 iocb->ki_pos >> PAGE_ !! 3060
3955 (iocb->ki_pos + count !! 3061 /* FIXME: this is for backwards compatibility with 2.4 */
3956 dio_warn_stale_pagecache(iocb !! 3062 if (iocb->ki_flags & IOCB_APPEND)
>> 3063 iocb->ki_pos = i_size_read(inode);
>> 3064
>> 3065 if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
>> 3066 return -EINVAL;
>> 3067
>> 3068 count = iov_iter_count(from);
>> 3069 ret = generic_write_check_limits(file, iocb->ki_pos, &count);
>> 3070 if (ret)
>> 3071 return ret;
>> 3072
>> 3073 iov_iter_truncate(from, count);
>> 3074 return iov_iter_count(from);
3957 } 3075 }
>> 3076 EXPORT_SYMBOL(generic_write_checks);
>> 3077
>> 3078 /*
>> 3079 * Performs necessary checks before doing a clone.
>> 3080 *
>> 3081 * Can adjust amount of bytes to clone.
>> 3082 * Returns appropriate error code that caller should return or
>> 3083 * zero in case the clone should be allowed.
>> 3084 */
>> 3085 int generic_remap_checks(struct file *file_in, loff_t pos_in,
>> 3086 struct file *file_out, loff_t pos_out,
>> 3087 loff_t *req_count, unsigned int remap_flags)
>> 3088 {
>> 3089 struct inode *inode_in = file_in->f_mapping->host;
>> 3090 struct inode *inode_out = file_out->f_mapping->host;
>> 3091 uint64_t count = *req_count;
>> 3092 uint64_t bcount;
>> 3093 loff_t size_in, size_out;
>> 3094 loff_t bs = inode_out->i_sb->s_blocksize;
>> 3095 int ret;
>> 3096
>> 3097 /* The start of both ranges must be aligned to an fs block. */
>> 3098 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs))
>> 3099 return -EINVAL;
>> 3100
>> 3101 /* Ensure offsets don't wrap. */
>> 3102 if (pos_in + count < pos_in || pos_out + count < pos_out)
>> 3103 return -EINVAL;
>> 3104
>> 3105 size_in = i_size_read(inode_in);
>> 3106 size_out = i_size_read(inode_out);
>> 3107
>> 3108 /* Dedupe requires both ranges to be within EOF. */
>> 3109 if ((remap_flags & REMAP_FILE_DEDUP) &&
>> 3110 (pos_in >= size_in || pos_in + count > size_in ||
>> 3111 pos_out >= size_out || pos_out + count > size_out))
>> 3112 return -EINVAL;
>> 3113
>> 3114 /* Ensure the infile range is within the infile. */
>> 3115 if (pos_in >= size_in)
>> 3116 return -EINVAL;
>> 3117 count = min(count, size_in - (uint64_t)pos_in);
>> 3118
>> 3119 ret = generic_access_check_limits(file_in, pos_in, &count);
>> 3120 if (ret)
>> 3121 return ret;
>> 3122
>> 3123 ret = generic_write_check_limits(file_out, pos_out, &count);
>> 3124 if (ret)
>> 3125 return ret;
>> 3126
>> 3127 /*
>> 3128 * If the user wanted us to link to the infile's EOF, round up to the
>> 3129 * next block boundary for this check.
>> 3130 *
>> 3131 * Otherwise, make sure the count is also block-aligned, having
>> 3132 * already confirmed the starting offsets' block alignment.
>> 3133 */
>> 3134 if (pos_in + count == size_in) {
>> 3135 bcount = ALIGN(size_in, bs) - pos_in;
>> 3136 } else {
>> 3137 if (!IS_ALIGNED(count, bs))
>> 3138 count = ALIGN_DOWN(count, bs);
>> 3139 bcount = count;
>> 3140 }
>> 3141
>> 3142 /* Don't allow overlapped cloning within the same file. */
>> 3143 if (inode_in == inode_out &&
>> 3144 pos_out + bcount > pos_in &&
>> 3145 pos_out < pos_in + bcount)
>> 3146 return -EINVAL;
>> 3147
>> 3148 /*
>> 3149 * We shortened the request but the caller can't deal with that, so
>> 3150 * bounce the request back to userspace.
>> 3151 */
>> 3152 if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN))
>> 3153 return -EINVAL;
>> 3154
>> 3155 *req_count = count;
>> 3156 return 0;
>> 3157 }
>> 3158
>> 3159 int pagecache_write_begin(struct file *file, struct address_space *mapping,
>> 3160 loff_t pos, unsigned len, unsigned flags,
>> 3161 struct page **pagep, void **fsdata)
>> 3162 {
>> 3163 const struct address_space_operations *aops = mapping->a_ops;
>> 3164
>> 3165 return aops->write_begin(file, mapping, pos, len, flags,
>> 3166 pagep, fsdata);
>> 3167 }
>> 3168 EXPORT_SYMBOL(pagecache_write_begin);
>> 3169
>> 3170 int pagecache_write_end(struct file *file, struct address_space *mapping,
>> 3171 loff_t pos, unsigned len, unsigned copied,
>> 3172 struct page *page, void *fsdata)
>> 3173 {
>> 3174 const struct address_space_operations *aops = mapping->a_ops;
>> 3175
>> 3176 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
>> 3177 }
>> 3178 EXPORT_SYMBOL(pagecache_write_end);
3958 3179
3959 ssize_t 3180 ssize_t
3960 generic_file_direct_write(struct kiocb *iocb, 3181 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3961 { 3182 {
3962 struct address_space *mapping = iocb- !! 3183 struct file *file = iocb->ki_filp;
3963 size_t write_len = iov_iter_count(fro !! 3184 struct address_space *mapping = file->f_mapping;
3964 ssize_t written; !! 3185 struct inode *inode = mapping->host;
>> 3186 loff_t pos = iocb->ki_pos;
>> 3187 ssize_t written;
>> 3188 size_t write_len;
>> 3189 pgoff_t end;
>> 3190
>> 3191 write_len = iov_iter_count(from);
>> 3192 end = (pos + write_len - 1) >> PAGE_SHIFT;
>> 3193
>> 3194 if (iocb->ki_flags & IOCB_NOWAIT) {
>> 3195 /* If there are pages to writeback, return */
>> 3196 if (filemap_range_has_page(inode->i_mapping, pos,
>> 3197 pos + write_len - 1))
>> 3198 return -EAGAIN;
>> 3199 } else {
>> 3200 written = filemap_write_and_wait_range(mapping, pos,
>> 3201 pos + write_len - 1);
>> 3202 if (written)
>> 3203 goto out;
>> 3204 }
3965 3205
3966 /* 3206 /*
>> 3207 * After a write we want buffered reads to be sure to go to disk to get
>> 3208 * the new data. We invalidate clean cached page from the region we're
>> 3209 * about to write. We do this *before* the write so that we can return
>> 3210 * without clobbering -EIOCBQUEUED from ->direct_IO().
>> 3211 */
>> 3212 written = invalidate_inode_pages2_range(mapping,
>> 3213 pos >> PAGE_SHIFT, end);
>> 3214 /*
3967 * If a page can not be invalidated, 3215 * If a page can not be invalidated, return 0 to fall back
3968 * to buffered write. 3216 * to buffered write.
3969 */ 3217 */
3970 written = kiocb_invalidate_pages(iocb <<
3971 if (written) { 3218 if (written) {
3972 if (written == -EBUSY) 3219 if (written == -EBUSY)
3973 return 0; 3220 return 0;
3974 return written; !! 3221 goto out;
3975 } 3222 }
3976 3223
3977 written = mapping->a_ops->direct_IO(i 3224 written = mapping->a_ops->direct_IO(iocb, from);
3978 3225
3979 /* 3226 /*
3980 * Finally, try again to invalidate c 3227 * Finally, try again to invalidate clean pages which might have been
3981 * cached by non-direct readahead, or 3228 * cached by non-direct readahead, or faulted in by get_user_pages()
3982 * if the source of the write was an 3229 * if the source of the write was an mmap'ed region of the file
3983 * we're writing. Either one is a pr 3230 * we're writing. Either one is a pretty crazy thing to do,
3984 * so we don't support it 100%. If t 3231 * so we don't support it 100%. If this invalidation
3985 * fails, tough, the write still work 3232 * fails, tough, the write still worked...
3986 * 3233 *
3987 * Most of the time we do not need th 3234 * Most of the time we do not need this since dio_complete() will do
3988 * the invalidation for us. However t 3235 * the invalidation for us. However there are some file systems that
3989 * do not end up with dio_complete() 3236 * do not end up with dio_complete() being called, so let's not break
3990 * them by removing it completely. !! 3237 * them by removing it completely
3991 * <<
3992 * Noticeable example is a blkdev_dir <<
3993 * <<
3994 * Skip invalidation for async writes <<
3995 */ 3238 */
3996 if (written > 0) { !! 3239 if (mapping->nrpages)
3997 struct inode *inode = mapping !! 3240 invalidate_inode_pages2_range(mapping,
3998 loff_t pos = iocb->ki_pos; !! 3241 pos >> PAGE_SHIFT, end);
3999 3242
4000 kiocb_invalidate_post_direct_ !! 3243 if (written > 0) {
4001 pos += written; 3244 pos += written;
4002 write_len -= written; 3245 write_len -= written;
4003 if (pos > i_size_read(inode) 3246 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
4004 i_size_write(inode, p 3247 i_size_write(inode, pos);
4005 mark_inode_dirty(inod 3248 mark_inode_dirty(inode);
4006 } 3249 }
4007 iocb->ki_pos = pos; 3250 iocb->ki_pos = pos;
4008 } 3251 }
4009 if (written != -EIOCBQUEUED) !! 3252 iov_iter_revert(from, write_len - iov_iter_count(from));
4010 iov_iter_revert(from, write_l !! 3253 out:
4011 return written; 3254 return written;
4012 } 3255 }
4013 EXPORT_SYMBOL(generic_file_direct_write); 3256 EXPORT_SYMBOL(generic_file_direct_write);
4014 3257
4015 ssize_t generic_perform_write(struct kiocb *i !! 3258 /*
>> 3259 * Find or create a page at the given pagecache position. Return the locked
>> 3260 * page. This function is specifically for buffered writes.
>> 3261 */
>> 3262 struct page *grab_cache_page_write_begin(struct address_space *mapping,
>> 3263 pgoff_t index, unsigned flags)
>> 3264 {
>> 3265 struct page *page;
>> 3266 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
>> 3267
>> 3268 if (flags & AOP_FLAG_NOFS)
>> 3269 fgp_flags |= FGP_NOFS;
>> 3270
>> 3271 page = pagecache_get_page(mapping, index, fgp_flags,
>> 3272 mapping_gfp_mask(mapping));
>> 3273 if (page)
>> 3274 wait_for_stable_page(page);
>> 3275
>> 3276 return page;
>> 3277 }
>> 3278 EXPORT_SYMBOL(grab_cache_page_write_begin);
>> 3279
>> 3280 ssize_t generic_perform_write(struct file *file,
>> 3281 struct iov_iter *i, loff_t pos)
4016 { 3282 {
4017 struct file *file = iocb->ki_filp; <<
4018 loff_t pos = iocb->ki_pos; <<
4019 struct address_space *mapping = file- 3283 struct address_space *mapping = file->f_mapping;
4020 const struct address_space_operations 3284 const struct address_space_operations *a_ops = mapping->a_ops;
4021 size_t chunk = mapping_max_folio_size <<
4022 long status = 0; 3285 long status = 0;
4023 ssize_t written = 0; 3286 ssize_t written = 0;
>> 3287 unsigned int flags = 0;
4024 3288
4025 do { 3289 do {
4026 struct folio *folio; !! 3290 struct page *page;
4027 size_t offset; /* Of !! 3291 unsigned long offset; /* Offset into pagecache page */
4028 size_t bytes; /* By !! 3292 unsigned long bytes; /* Bytes to write to page */
4029 size_t copied; /* By 3293 size_t copied; /* Bytes copied from user */
4030 void *fsdata = NULL; !! 3294 void *fsdata;
4031 3295
4032 bytes = iov_iter_count(i); !! 3296 offset = (pos & (PAGE_SIZE - 1));
4033 retry: !! 3297 bytes = min_t(unsigned long, PAGE_SIZE - offset,
4034 offset = pos & (chunk - 1); !! 3298 iov_iter_count(i));
4035 bytes = min(chunk - offset, b <<
4036 balance_dirty_pages_ratelimit <<
4037 3299
>> 3300 again:
4038 /* 3301 /*
4039 * Bring in the user page tha 3302 * Bring in the user page that we will copy from _first_.
4040 * Otherwise there's a nasty 3303 * Otherwise there's a nasty deadlock on copying from the
4041 * same page as we're writing 3304 * same page as we're writing to, without it being marked
4042 * up-to-date. 3305 * up-to-date.
>> 3306 *
>> 3307 * Not only is this an optimisation, but it is also required
>> 3308 * to check that the address is actually valid, when atomic
>> 3309 * usercopies are used, below.
4043 */ 3310 */
4044 if (unlikely(fault_in_iov_ite !! 3311 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
4045 status = -EFAULT; 3312 status = -EFAULT;
4046 break; 3313 break;
4047 } 3314 }
4048 3315
4049 if (fatal_signal_pending(curr 3316 if (fatal_signal_pending(current)) {
4050 status = -EINTR; 3317 status = -EINTR;
4051 break; 3318 break;
4052 } 3319 }
4053 3320
4054 status = a_ops->write_begin(f !! 3321 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
4055 !! 3322 &page, &fsdata);
4056 if (unlikely(status < 0)) 3323 if (unlikely(status < 0))
4057 break; 3324 break;
4058 3325
4059 offset = offset_in_folio(foli <<
4060 if (bytes > folio_size(folio) <<
4061 bytes = folio_size(fo <<
4062 <<
4063 if (mapping_writably_mapped(m 3326 if (mapping_writably_mapped(mapping))
4064 flush_dcache_folio(fo !! 3327 flush_dcache_page(page);
4065 3328
4066 copied = copy_folio_from_iter !! 3329 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
4067 flush_dcache_folio(folio); !! 3330 flush_dcache_page(page);
4068 3331
4069 status = a_ops->write_end(fil 3332 status = a_ops->write_end(file, mapping, pos, bytes, copied,
4070 !! 3333 page, fsdata);
4071 if (unlikely(status != copied !! 3334 if (unlikely(status < 0))
4072 iov_iter_revert(i, co !! 3335 break;
4073 if (unlikely(status < !! 3336 copied = status;
4074 break; !! 3337
4075 } <<
4076 cond_resched(); 3338 cond_resched();
4077 3339
4078 if (unlikely(status == 0)) { !! 3340 iov_iter_advance(i, copied);
>> 3341 if (unlikely(copied == 0)) {
4079 /* 3342 /*
4080 * A short copy made !! 3343 * If we were unable to copy any data at all, we must
4081 * thing entirely. M !! 3344 * fall back to a single segment length write.
4082 * halfway through, m !! 3345 *
4083 * might be severe me !! 3346 * If we didn't fallback here, we could livelock
>> 3347 * because not all segments in the iov can be copied at
>> 3348 * once without a pagefault.
4084 */ 3349 */
4085 if (chunk > PAGE_SIZE !! 3350 bytes = min_t(unsigned long, PAGE_SIZE - offset,
4086 chunk /= 2; !! 3351 iov_iter_single_seg_count(i));
4087 if (copied) { !! 3352 goto again;
4088 bytes = copie <<
4089 goto retry; <<
4090 } <<
4091 } else { <<
4092 pos += status; <<
4093 written += status; <<
4094 } 3353 }
>> 3354 pos += copied;
>> 3355 written += copied;
>> 3356
>> 3357 balance_dirty_pages_ratelimited(mapping);
4095 } while (iov_iter_count(i)); 3358 } while (iov_iter_count(i));
4096 3359
4097 if (!written) !! 3360 return written ? written : status;
4098 return status; <<
4099 iocb->ki_pos += written; <<
4100 return written; <<
4101 } 3361 }
4102 EXPORT_SYMBOL(generic_perform_write); 3362 EXPORT_SYMBOL(generic_perform_write);
4103 3363
4104 /** 3364 /**
4105 * __generic_file_write_iter - write data to 3365 * __generic_file_write_iter - write data to a file
4106 * @iocb: IO state structure (file, off 3366 * @iocb: IO state structure (file, offset, etc.)
4107 * @from: iov_iter with data to write 3367 * @from: iov_iter with data to write
4108 * 3368 *
4109 * This function does all the work needed for 3369 * This function does all the work needed for actually writing data to a
4110 * file. It does all basic checks, removes SU 3370 * file. It does all basic checks, removes SUID from the file, updates
4111 * modification times and calls proper subrou 3371 * modification times and calls proper subroutines depending on whether we
4112 * do direct IO or a standard buffered write. 3372 * do direct IO or a standard buffered write.
4113 * 3373 *
4114 * It expects i_rwsem to be grabbed unless we !! 3374 * It expects i_mutex to be grabbed unless we work on a block device or similar
4115 * object which does not need locking at all. 3375 * object which does not need locking at all.
4116 * 3376 *
4117 * This function does *not* take care of sync 3377 * 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 3378 * A caller has to handle it. This is mainly due to the fact that we want to
4119 * avoid syncing under i_rwsem. !! 3379 * avoid syncing under i_mutex.
4120 * 3380 *
4121 * Return: 3381 * Return:
4122 * * number of bytes written, even for trunca 3382 * * number of bytes written, even for truncated writes
4123 * * negative error code if no data has been 3383 * * negative error code if no data has been written at all
4124 */ 3384 */
4125 ssize_t __generic_file_write_iter(struct kioc 3385 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4126 { 3386 {
4127 struct file *file = iocb->ki_filp; 3387 struct file *file = iocb->ki_filp;
4128 struct address_space *mapping = file- !! 3388 struct address_space * mapping = file->f_mapping;
4129 struct inode *inode = mapping->host; !! 3389 struct inode *inode = mapping->host;
4130 ssize_t ret; !! 3390 ssize_t written = 0;
4131 !! 3391 ssize_t err;
4132 ret = file_remove_privs(file); !! 3392 ssize_t status;
4133 if (ret) !! 3393
4134 return ret; !! 3394 /* We can write back this queue in page reclaim */
>> 3395 current->backing_dev_info = inode_to_bdi(inode);
>> 3396 err = file_remove_privs(file);
>> 3397 if (err)
>> 3398 goto out;
4135 3399
4136 ret = file_update_time(file); !! 3400 err = file_update_time(file);
4137 if (ret) !! 3401 if (err)
4138 return ret; !! 3402 goto out;
4139 3403
4140 if (iocb->ki_flags & IOCB_DIRECT) { 3404 if (iocb->ki_flags & IOCB_DIRECT) {
4141 ret = generic_file_direct_wri !! 3405 loff_t pos, endbyte;
>> 3406
>> 3407 written = generic_file_direct_write(iocb, from);
4142 /* 3408 /*
4143 * If the write stopped short 3409 * If the write stopped short of completing, fall back to
4144 * buffered writes. Some fil 3410 * buffered writes. Some filesystems do this for writes to
4145 * holes, for example. For D 3411 * holes, for example. For DAX files, a buffered write will
4146 * not succeed (even if it di 3412 * not succeed (even if it did, DAX does not handle dirty
4147 * page-cache pages correctly 3413 * page-cache pages correctly).
4148 */ 3414 */
4149 if (ret < 0 || !iov_iter_coun !! 3415 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
4150 return ret; !! 3416 goto out;
4151 return direct_write_fallback( <<
4152 generic_perfo <<
4153 } <<
4154 3417
4155 return generic_perform_write(iocb, fr !! 3418 status = generic_perform_write(file, from, pos = iocb->ki_pos);
>> 3419 /*
>> 3420 * If generic_perform_write() returned a synchronous error
>> 3421 * then we want to return the number of bytes which were
>> 3422 * direct-written, or the error code if that was zero. Note
>> 3423 * that this differs from normal direct-io semantics, which
>> 3424 * will return -EFOO even if some bytes were written.
>> 3425 */
>> 3426 if (unlikely(status < 0)) {
>> 3427 err = status;
>> 3428 goto out;
>> 3429 }
>> 3430 /*
>> 3431 * We need to ensure that the page cache pages are written to
>> 3432 * disk and invalidated to preserve the expected O_DIRECT
>> 3433 * semantics.
>> 3434 */
>> 3435 endbyte = pos + status - 1;
>> 3436 err = filemap_write_and_wait_range(mapping, pos, endbyte);
>> 3437 if (err == 0) {
>> 3438 iocb->ki_pos = endbyte + 1;
>> 3439 written += status;
>> 3440 invalidate_mapping_pages(mapping,
>> 3441 pos >> PAGE_SHIFT,
>> 3442 endbyte >> PAGE_SHIFT);
>> 3443 } else {
>> 3444 /*
>> 3445 * We don't know how much we wrote, so just return
>> 3446 * the number of bytes which were direct-written
>> 3447 */
>> 3448 }
>> 3449 } else {
>> 3450 written = generic_perform_write(file, from, iocb->ki_pos);
>> 3451 if (likely(written > 0))
>> 3452 iocb->ki_pos += written;
>> 3453 }
>> 3454 out:
>> 3455 current->backing_dev_info = NULL;
>> 3456 return written ? written : err;
4156 } 3457 }
4157 EXPORT_SYMBOL(__generic_file_write_iter); 3458 EXPORT_SYMBOL(__generic_file_write_iter);
4158 3459
4159 /** 3460 /**
4160 * generic_file_write_iter - write data to a 3461 * generic_file_write_iter - write data to a file
4161 * @iocb: IO state structure 3462 * @iocb: IO state structure
4162 * @from: iov_iter with data to write 3463 * @from: iov_iter with data to write
4163 * 3464 *
4164 * This is a wrapper around __generic_file_wr 3465 * This is a wrapper around __generic_file_write_iter() to be used by most
4165 * filesystems. It takes care of syncing the 3466 * filesystems. It takes care of syncing the file in case of O_SYNC file
4166 * and acquires i_rwsem as needed. !! 3467 * and acquires i_mutex as needed.
4167 * Return: 3468 * Return:
4168 * * negative error code if no data has been 3469 * * negative error code if no data has been written at all of
4169 * vfs_fsync_range() failed for a synchrono 3470 * vfs_fsync_range() failed for a synchronous write
4170 * * number of bytes written, even for trunca 3471 * * number of bytes written, even for truncated writes
4171 */ 3472 */
4172 ssize_t generic_file_write_iter(struct kiocb 3473 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4173 { 3474 {
4174 struct file *file = iocb->ki_filp; 3475 struct file *file = iocb->ki_filp;
4175 struct inode *inode = file->f_mapping 3476 struct inode *inode = file->f_mapping->host;
4176 ssize_t ret; 3477 ssize_t ret;
4177 3478
4178 inode_lock(inode); 3479 inode_lock(inode);
4179 ret = generic_write_checks(iocb, from 3480 ret = generic_write_checks(iocb, from);
4180 if (ret > 0) 3481 if (ret > 0)
4181 ret = __generic_file_write_it 3482 ret = __generic_file_write_iter(iocb, from);
4182 inode_unlock(inode); 3483 inode_unlock(inode);
4183 3484
4184 if (ret > 0) 3485 if (ret > 0)
4185 ret = generic_write_sync(iocb 3486 ret = generic_write_sync(iocb, ret);
4186 return ret; 3487 return ret;
4187 } 3488 }
4188 EXPORT_SYMBOL(generic_file_write_iter); 3489 EXPORT_SYMBOL(generic_file_write_iter);
4189 3490
4190 /** 3491 /**
4191 * filemap_release_folio() - Release fs-speci !! 3492 * try_to_release_page() - release old fs-specific metadata on a page
4192 * @folio: The folio which the kernel is tryi <<
4193 * @gfp: Memory allocation flags (and I/O mod <<
4194 * 3493 *
4195 * The address_space is trying to release any !! 3494 * @page: the page which the kernel is trying to free
4196 * (presumably at folio->private). !! 3495 * @gfp_mask: memory allocation flags (and I/O mode)
4197 * 3496 *
4198 * This will also be called if the private_2 !! 3497 * The address_space is to try to release any data against the page
4199 * indicating that the folio has other metada !! 3498 * (presumably at page->private).
4200 * 3499 *
4201 * The @gfp argument specifies whether I/O ma !! 3500 * This may also be called if PG_fscache is set on a page, indicating that the
4202 * this page (__GFP_IO), and whether the call !! 3501 * page is known to the local caching routines.
4203 * (__GFP_RECLAIM & __GFP_FS). <<
4204 * 3502 *
4205 * Return: %true if the release was successfu !! 3503 * The @gfp_mask argument specifies whether I/O may be performed to release
4206 */ !! 3504 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
4207 bool filemap_release_folio(struct folio *foli !! 3505 *
4208 { !! 3506 * Return: %1 if the release was successful, otherwise return zero.
4209 struct address_space * const mapping <<
4210 <<
4211 BUG_ON(!folio_test_locked(folio)); <<
4212 if (!folio_needs_release(folio)) <<
4213 return true; <<
4214 if (folio_test_writeback(folio)) <<
4215 return false; <<
4216 <<
4217 if (mapping && mapping->a_ops->releas <<
4218 return mapping->a_ops->releas <<
4219 return try_to_free_buffers(folio); <<
4220 } <<
4221 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 */ 3507 */
4290 static void filemap_cachestat(struct address_ !! 3508 int try_to_release_page(struct page *page, gfp_t gfp_mask)
4291 pgoff_t first_index, pgoff_t <<
4292 { 3509 {
4293 XA_STATE(xas, &mapping->i_pages, firs !! 3510 struct address_space * const mapping = page->mapping;
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 3511
4365 goto resched; !! 3512 BUG_ON(!PageLocked(page));
4366 } !! 3513 if (PageWriteback(page))
4367 !! 3514 return 0;
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 3515
4377 resched: !! 3516 if (mapping && mapping->a_ops->releasepage)
4378 if (need_resched()) { !! 3517 return mapping->a_ops->releasepage(page, gfp_mask);
4379 xas_pause(&xas); !! 3518 return try_to_free_buffers(page);
4380 cond_resched_rcu(); <<
4381 } <<
4382 } <<
4383 rcu_read_unlock(); <<
4384 } 3519 }
4385 3520
4386 /* !! 3521 EXPORT_SYMBOL(try_to_release_page);
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 3522
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.