1 // SPDX-License-Identifier: GPL-2.0-only 1 // SPDX-License-Identifier: GPL-2.0-only
2 /* 2 /*
3 * linux/mm/filemap.c 3 * linux/mm/filemap.c
4 * 4 *
5 * Copyright (C) 1994-1999 Linus Torvalds 5 * Copyright (C) 1994-1999 Linus Torvalds
6 */ 6 */
7 7
8 /* 8 /*
9 * This file handles the generic file mmap sem 9 * This file handles the generic file mmap semantics used by
10 * most "normal" filesystems (but you don't /h 10 * most "normal" filesystems (but you don't /have/ to use this:
11 * the NFS filesystem used to do this differen 11 * the NFS filesystem used to do this differently, for example)
12 */ 12 */
13 #include <linux/export.h> 13 #include <linux/export.h>
14 #include <linux/compiler.h> 14 #include <linux/compiler.h>
15 #include <linux/dax.h> 15 #include <linux/dax.h>
16 #include <linux/fs.h> 16 #include <linux/fs.h>
17 #include <linux/sched/signal.h> 17 #include <linux/sched/signal.h>
18 #include <linux/uaccess.h> 18 #include <linux/uaccess.h>
19 #include <linux/capability.h> 19 #include <linux/capability.h>
20 #include <linux/kernel_stat.h> 20 #include <linux/kernel_stat.h>
21 #include <linux/gfp.h> 21 #include <linux/gfp.h>
22 #include <linux/mm.h> 22 #include <linux/mm.h>
23 #include <linux/swap.h> 23 #include <linux/swap.h>
24 #include <linux/swapops.h> <<
25 #include <linux/syscalls.h> <<
26 #include <linux/mman.h> 24 #include <linux/mman.h>
27 #include <linux/pagemap.h> 25 #include <linux/pagemap.h>
28 #include <linux/file.h> 26 #include <linux/file.h>
29 #include <linux/uio.h> 27 #include <linux/uio.h>
30 #include <linux/error-injection.h> 28 #include <linux/error-injection.h>
31 #include <linux/hash.h> 29 #include <linux/hash.h>
32 #include <linux/writeback.h> 30 #include <linux/writeback.h>
33 #include <linux/backing-dev.h> 31 #include <linux/backing-dev.h>
34 #include <linux/pagevec.h> 32 #include <linux/pagevec.h>
>> 33 #include <linux/blkdev.h>
35 #include <linux/security.h> 34 #include <linux/security.h>
36 #include <linux/cpuset.h> 35 #include <linux/cpuset.h>
37 #include <linux/hugetlb.h> 36 #include <linux/hugetlb.h>
38 #include <linux/memcontrol.h> 37 #include <linux/memcontrol.h>
>> 38 #include <linux/cleancache.h>
39 #include <linux/shmem_fs.h> 39 #include <linux/shmem_fs.h>
40 #include <linux/rmap.h> 40 #include <linux/rmap.h>
41 #include <linux/delayacct.h> 41 #include <linux/delayacct.h>
42 #include <linux/psi.h> 42 #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" 43 #include "internal.h"
53 44
54 #define CREATE_TRACE_POINTS 45 #define CREATE_TRACE_POINTS
55 #include <trace/events/filemap.h> 46 #include <trace/events/filemap.h>
56 47
57 /* 48 /*
58 * FIXME: remove all knowledge of the buffer l 49 * FIXME: remove all knowledge of the buffer layer from the core VM
59 */ 50 */
60 #include <linux/buffer_head.h> /* for try_to_f 51 #include <linux/buffer_head.h> /* for try_to_free_buffers */
61 52
62 #include <asm/mman.h> 53 #include <asm/mman.h>
63 54
64 #include "swap.h" <<
65 <<
66 /* 55 /*
67 * Shared mappings implemented 30.11.1994. It' 56 * Shared mappings implemented 30.11.1994. It's not fully working yet,
68 * though. 57 * though.
69 * 58 *
70 * Shared mappings now work. 15.8.1995 Bruno. 59 * Shared mappings now work. 15.8.1995 Bruno.
71 * 60 *
72 * finished 'unifying' the page and buffer cac 61 * finished 'unifying' the page and buffer cache and SMP-threaded the
73 * page-cache, 21.05.1999, Ingo Molnar <mingo@ 62 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
74 * 63 *
75 * SMP-threaded pagemap-LRU 1999, Andrea Arcan 64 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
76 */ 65 */
77 66
78 /* 67 /*
79 * Lock ordering: 68 * Lock ordering:
80 * 69 *
81 * ->i_mmap_rwsem (truncate_page 70 * ->i_mmap_rwsem (truncate_pagecache)
82 * ->private_lock (__free_pte->b !! 71 * ->private_lock (__free_pte->__set_page_dirty_buffers)
83 * ->swap_lock (exclusive_swa 72 * ->swap_lock (exclusive_swap_page, others)
84 * ->i_pages lock 73 * ->i_pages lock
85 * 74 *
86 * ->i_rwsem !! 75 * ->i_mutex
87 * ->invalidate_lock (acquired by f !! 76 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
88 * ->i_mmap_rwsem (truncate->unm <<
89 * 77 *
90 * ->mmap_lock !! 78 * ->mmap_sem
91 * ->i_mmap_rwsem 79 * ->i_mmap_rwsem
92 * ->page_table_lock or pte_lock (vario 80 * ->page_table_lock or pte_lock (various, mainly in memory.c)
93 * ->i_pages lock (arch-dependen 81 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
94 * 82 *
95 * ->mmap_lock !! 83 * ->mmap_sem
96 * ->invalidate_lock (filemap_fault !! 84 * ->lock_page (access_process_vm)
97 * ->lock_page (filemap_fault <<
98 * 85 *
99 * ->i_rwsem (generic_perfo !! 86 * ->i_mutex (generic_perform_write)
100 * ->mmap_lock (fault_in_read !! 87 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
101 * 88 *
102 * bdi->wb.list_lock 89 * bdi->wb.list_lock
103 * sb_lock (fs/fs-writeba 90 * sb_lock (fs/fs-writeback.c)
104 * ->i_pages lock (__sync_single 91 * ->i_pages lock (__sync_single_inode)
105 * 92 *
106 * ->i_mmap_rwsem 93 * ->i_mmap_rwsem
107 * ->anon_vma.lock (vma_merge) !! 94 * ->anon_vma.lock (vma_adjust)
108 * 95 *
109 * ->anon_vma.lock 96 * ->anon_vma.lock
110 * ->page_table_lock or pte_lock (anon_ 97 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
111 * 98 *
112 * ->page_table_lock or pte_lock 99 * ->page_table_lock or pte_lock
113 * ->swap_lock (try_to_unmap_ 100 * ->swap_lock (try_to_unmap_one)
114 * ->private_lock (try_to_unmap_ 101 * ->private_lock (try_to_unmap_one)
115 * ->i_pages lock (try_to_unmap_ 102 * ->i_pages lock (try_to_unmap_one)
116 * ->lruvec->lru_lock (follow_page_m !! 103 * ->pgdat->lru_lock (follow_page->mark_page_accessed)
117 * ->lruvec->lru_lock (check_pte_ran !! 104 * ->pgdat->lru_lock (check_pte_range->isolate_lru_page)
118 * ->private_lock (folio_remove_ !! 105 * ->private_lock (page_remove_rmap->set_page_dirty)
119 * ->i_pages lock (folio_remove_ !! 106 * ->i_pages lock (page_remove_rmap->set_page_dirty)
120 * bdi.wb->list_lock (folio_remove_ !! 107 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
121 * ->inode->i_lock (folio_remove_ !! 108 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
122 * ->memcg->move_lock (folio_remove_ !! 109 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
123 * bdi.wb->list_lock (zap_pte_range 110 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
124 * ->inode->i_lock (zap_pte_range 111 * ->inode->i_lock (zap_pte_range->set_page_dirty)
125 * ->private_lock (zap_pte_range !! 112 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
>> 113 *
>> 114 * ->i_mmap_rwsem
>> 115 * ->tasklist_lock (memory_failure, collect_procs_ao)
126 */ 116 */
127 117
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 118 static void page_cache_delete(struct address_space *mapping,
138 struct foli !! 119 struct page *page, void *shadow)
139 { 120 {
140 XA_STATE(xas, &mapping->i_pages, folio !! 121 XA_STATE(xas, &mapping->i_pages, page->index);
141 long nr = 1; !! 122 unsigned int nr = 1;
142 123
143 mapping_set_update(&xas, mapping); 124 mapping_set_update(&xas, mapping);
144 125
145 xas_set_order(&xas, folio->index, foli !! 126 /* hugetlb pages are represented by a single entry in the xarray */
146 nr = folio_nr_pages(folio); !! 127 if (!PageHuge(page)) {
>> 128 xas_set_order(&xas, page->index, compound_order(page));
>> 129 nr = 1U << compound_order(page);
>> 130 }
147 131
148 VM_BUG_ON_FOLIO(!folio_test_locked(fol !! 132 VM_BUG_ON_PAGE(!PageLocked(page), page);
>> 133 VM_BUG_ON_PAGE(PageTail(page), page);
>> 134 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
149 135
150 xas_store(&xas, shadow); 136 xas_store(&xas, shadow);
151 xas_init_marks(&xas); 137 xas_init_marks(&xas);
152 138
153 folio->mapping = NULL; !! 139 page->mapping = NULL;
154 /* Leave page->index set: truncation l 140 /* Leave page->index set: truncation lookup relies upon it */
>> 141
>> 142 if (shadow) {
>> 143 mapping->nrexceptional += nr;
>> 144 /*
>> 145 * Make sure the nrexceptional update is committed before
>> 146 * the nrpages update so that final truncate racing
>> 147 * with reclaim does not see both counters 0 at the
>> 148 * same time and miss a shadow entry.
>> 149 */
>> 150 smp_wmb();
>> 151 }
155 mapping->nrpages -= nr; 152 mapping->nrpages -= nr;
156 } 153 }
157 154
158 static void filemap_unaccount_folio(struct add !! 155 static void unaccount_page_cache_page(struct address_space *mapping,
159 struct folio *folio) !! 156 struct page *page)
160 { 157 {
161 long nr; !! 158 int nr;
>> 159
>> 160 /*
>> 161 * if we're uptodate, flush out into the cleancache, otherwise
>> 162 * invalidate any existing cleancache entries. We can't leave
>> 163 * stale data around in the cleancache once our page is gone
>> 164 */
>> 165 if (PageUptodate(page) && PageMappedToDisk(page))
>> 166 cleancache_put_page(page);
>> 167 else
>> 168 cleancache_invalidate_page(mapping, page);
>> 169
>> 170 VM_BUG_ON_PAGE(PageTail(page), page);
>> 171 VM_BUG_ON_PAGE(page_mapped(page), page);
>> 172 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
>> 173 int mapcount;
162 174
163 VM_BUG_ON_FOLIO(folio_mapped(folio), f <<
164 if (!IS_ENABLED(CONFIG_DEBUG_VM) && un <<
165 pr_alert("BUG: Bad page cache 175 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
166 current->comm, folio_ !! 176 current->comm, page_to_pfn(page));
167 dump_page(&folio->page, "still !! 177 dump_page(page, "still mapped when deleted");
168 dump_stack(); 178 dump_stack();
169 add_taint(TAINT_BAD_PAGE, LOCK 179 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
170 180
171 if (mapping_exiting(mapping) & !! 181 mapcount = page_mapcount(page);
172 int mapcount = folio_m !! 182 if (mapping_exiting(mapping) &&
173 !! 183 page_count(page) >= mapcount + 2) {
174 if (folio_ref_count(fo !! 184 /*
175 /* !! 185 * All vmas have already been torn down, so it's
176 * All vmas ha !! 186 * a good bet that actually the page is unmapped,
177 * a good bet !! 187 * and we'd prefer not to leak it: if we're wrong,
178 * and we'd ra !! 188 * some other bad page check should catch it later.
179 * another bad !! 189 */
180 */ !! 190 page_mapcount_reset(page);
181 atomic_set(&fo !! 191 page_ref_sub(page, mapcount);
182 folio_ref_sub( <<
183 } <<
184 } 192 }
185 } 193 }
186 194
187 /* hugetlb folios do not participate i !! 195 /* hugetlb pages do not participate in page cache accounting. */
188 if (folio_test_hugetlb(folio)) !! 196 if (PageHuge(page))
189 return; 197 return;
190 198
191 nr = folio_nr_pages(folio); !! 199 nr = hpage_nr_pages(page);
192 200
193 __lruvec_stat_mod_folio(folio, NR_FILE !! 201 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
194 if (folio_test_swapbacked(folio)) { !! 202 if (PageSwapBacked(page)) {
195 __lruvec_stat_mod_folio(folio, !! 203 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
196 if (folio_test_pmd_mappable(fo !! 204 if (PageTransHuge(page))
197 __lruvec_stat_mod_foli !! 205 __dec_node_page_state(page, NR_SHMEM_THPS);
198 } else if (folio_test_pmd_mappable(fol !! 206 } else {
199 __lruvec_stat_mod_folio(folio, !! 207 VM_BUG_ON_PAGE(PageTransHuge(page), page);
200 filemap_nr_thps_dec(mapping); <<
201 } 208 }
202 209
203 /* 210 /*
204 * At this point folio must be either !! 211 * At this point page must be either written or cleaned by
205 * truncate. Dirty folio here signals !! 212 * truncate. Dirty page here signals a bug and loss of
206 * unwritten data - on ordinary filesy !! 213 * 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 * 214 *
212 * Below fixes dirty accounting after !! 215 * This fixes dirty accounting after removing the page entirely
213 * but leaves the dirty flag set: it h !! 216 * but leaves PageDirty set: it has no effect for truncated
214 * folio and anyway will be cleared be !! 217 * page and anyway will be cleared before returning page into
215 * buddy allocator. 218 * buddy allocator.
216 */ 219 */
217 if (WARN_ON_ONCE(folio_test_dirty(foli !! 220 if (WARN_ON_ONCE(PageDirty(page)))
218 mapping_can_writeback !! 221 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
219 folio_account_cleaned(folio, i <<
220 } 222 }
221 223
222 /* 224 /*
223 * Delete a page from the page cache and free 225 * Delete a page from the page cache and free it. Caller has to make
224 * sure the page is locked and that nobody els 226 * sure the page is locked and that nobody else uses it - or that usage
225 * is safe. The caller must hold the i_pages 227 * is safe. The caller must hold the i_pages lock.
226 */ 228 */
227 void __filemap_remove_folio(struct folio *foli !! 229 void __delete_from_page_cache(struct page *page, void *shadow)
228 { 230 {
229 struct address_space *mapping = folio- !! 231 struct address_space *mapping = page->mapping;
230 232
231 trace_mm_filemap_delete_from_page_cach !! 233 trace_mm_filemap_delete_from_page_cache(page);
232 filemap_unaccount_folio(mapping, folio !! 234
233 page_cache_delete(mapping, folio, shad !! 235 unaccount_page_cache_page(mapping, page);
>> 236 page_cache_delete(mapping, page, shadow);
234 } 237 }
235 238
236 void filemap_free_folio(struct address_space * !! 239 static void page_cache_free_page(struct address_space *mapping,
>> 240 struct page *page)
237 { 241 {
238 void (*free_folio)(struct folio *); !! 242 void (*freepage)(struct page *);
239 int refs = 1; <<
240 243
241 free_folio = mapping->a_ops->free_foli !! 244 freepage = mapping->a_ops->freepage;
242 if (free_folio) !! 245 if (freepage)
243 free_folio(folio); !! 246 freepage(page);
244 247
245 if (folio_test_large(folio)) !! 248 if (PageTransHuge(page) && !PageHuge(page)) {
246 refs = folio_nr_pages(folio); !! 249 page_ref_sub(page, HPAGE_PMD_NR);
247 folio_put_refs(folio, refs); !! 250 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
>> 251 } else {
>> 252 put_page(page);
>> 253 }
248 } 254 }
249 255
250 /** 256 /**
251 * filemap_remove_folio - Remove folio from pa !! 257 * delete_from_page_cache - delete page from page cache
252 * @folio: The folio. !! 258 * @page: the page which the kernel is trying to remove from page cache
253 * 259 *
254 * This must be called only on folios that are !! 260 * 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 !! 261 * 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 !! 262 * has a reference on the page.
257 */ !! 263 */
258 void filemap_remove_folio(struct folio *folio) !! 264 void delete_from_page_cache(struct page *page)
259 { !! 265 {
260 struct address_space *mapping = folio- !! 266 struct address_space *mapping = page_mapping(page);
261 !! 267 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 } <<
273 268
274 /* !! 269 BUG_ON(!PageLocked(page));
275 * page_cache_delete_batch - delete several fo !! 270 xa_lock_irqsave(&mapping->i_pages, flags);
276 * @mapping: the mapping to which folios belon !! 271 __delete_from_page_cache(page, NULL);
277 * @fbatch: batch of folios to delete !! 272 xa_unlock_irqrestore(&mapping->i_pages, flags);
278 * !! 273
279 * The function walks over mapping->i_pages an !! 274 page_cache_free_page(mapping, page);
280 * @fbatch from the mapping. The function expe !! 275 }
281 * by page index and is optimised for it to be !! 276 EXPORT_SYMBOL(delete_from_page_cache);
282 * It tolerates holes in @fbatch (mapping entr !! 277
283 * modified). !! 278 /*
>> 279 * page_cache_delete_batch - delete several pages from page cache
>> 280 * @mapping: the mapping to which pages belong
>> 281 * @pvec: pagevec with pages to delete
>> 282 *
>> 283 * The function walks over mapping->i_pages and removes pages passed in @pvec
>> 284 * from the mapping. The function expects @pvec to be sorted by page index.
>> 285 * It tolerates holes in @pvec (mapping entries at those indices are not
>> 286 * modified). The function expects only THP head pages to be present in the
>> 287 * @pvec and takes care to delete all corresponding tail pages from the
>> 288 * mapping as well.
284 * 289 *
285 * The function expects the i_pages lock to be 290 * The function expects the i_pages lock to be held.
286 */ 291 */
287 static void page_cache_delete_batch(struct add 292 static void page_cache_delete_batch(struct address_space *mapping,
288 struct folio_batc !! 293 struct pagevec *pvec)
289 { 294 {
290 XA_STATE(xas, &mapping->i_pages, fbatc !! 295 XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
291 long total_pages = 0; !! 296 int total_pages = 0;
292 int i = 0; !! 297 int i = 0, tail_pages = 0;
293 struct folio *folio; !! 298 struct page *page;
294 299
295 mapping_set_update(&xas, mapping); 300 mapping_set_update(&xas, mapping);
296 xas_for_each(&xas, folio, ULONG_MAX) { !! 301 xas_for_each(&xas, page, ULONG_MAX) {
297 if (i >= folio_batch_count(fba !! 302 if (i >= pagevec_count(pvec) && !tail_pages)
298 break; 303 break;
299 !! 304 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; 305 continue;
>> 306 if (!tail_pages) {
>> 307 /*
>> 308 * Some page got inserted in our range? Skip it. We
>> 309 * have our pages locked so they are protected from
>> 310 * being removed.
>> 311 */
>> 312 if (page != pvec->pages[i]) {
>> 313 VM_BUG_ON_PAGE(page->index >
>> 314 pvec->pages[i]->index, page);
>> 315 continue;
>> 316 }
>> 317 WARN_ON_ONCE(!PageLocked(page));
>> 318 if (PageTransHuge(page) && !PageHuge(page))
>> 319 tail_pages = HPAGE_PMD_NR - 1;
>> 320 page->mapping = NULL;
>> 321 /*
>> 322 * Leave page->index set: truncation lookup relies
>> 323 * upon it
>> 324 */
>> 325 i++;
>> 326 } else {
>> 327 VM_BUG_ON_PAGE(page->index + HPAGE_PMD_NR - tail_pages
>> 328 != pvec->pages[i]->index, page);
>> 329 tail_pages--;
314 } 330 }
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); 331 xas_store(&xas, NULL);
323 total_pages += folio_nr_pages( !! 332 total_pages++;
324 } 333 }
325 mapping->nrpages -= total_pages; 334 mapping->nrpages -= total_pages;
326 } 335 }
327 336
328 void delete_from_page_cache_batch(struct addre 337 void delete_from_page_cache_batch(struct address_space *mapping,
329 struct folio !! 338 struct pagevec *pvec)
330 { 339 {
331 int i; 340 int i;
>> 341 unsigned long flags;
332 342
333 if (!folio_batch_count(fbatch)) !! 343 if (!pagevec_count(pvec))
334 return; 344 return;
335 345
336 spin_lock(&mapping->host->i_lock); !! 346 xa_lock_irqsave(&mapping->i_pages, flags);
337 xa_lock_irq(&mapping->i_pages); !! 347 for (i = 0; i < pagevec_count(pvec); i++) {
338 for (i = 0; i < folio_batch_count(fbat !! 348 trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
339 struct folio *folio = fbatch-> !! 349
340 !! 350 unaccount_page_cache_page(mapping, pvec->pages[i]);
341 trace_mm_filemap_delete_from_p !! 351 }
342 filemap_unaccount_folio(mappin !! 352 page_cache_delete_batch(mapping, pvec);
343 } !! 353 xa_unlock_irqrestore(&mapping->i_pages, flags);
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); <<
349 354
350 for (i = 0; i < folio_batch_count(fbat !! 355 for (i = 0; i < pagevec_count(pvec); i++)
351 filemap_free_folio(mapping, fb !! 356 page_cache_free_page(mapping, pvec->pages[i]);
352 } 357 }
353 358
354 int filemap_check_errors(struct address_space 359 int filemap_check_errors(struct address_space *mapping)
355 { 360 {
356 int ret = 0; 361 int ret = 0;
357 /* Check for outstanding write errors 362 /* Check for outstanding write errors */
358 if (test_bit(AS_ENOSPC, &mapping->flag 363 if (test_bit(AS_ENOSPC, &mapping->flags) &&
359 test_and_clear_bit(AS_ENOSPC, &map 364 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
360 ret = -ENOSPC; 365 ret = -ENOSPC;
361 if (test_bit(AS_EIO, &mapping->flags) 366 if (test_bit(AS_EIO, &mapping->flags) &&
362 test_and_clear_bit(AS_EIO, &mappin 367 test_and_clear_bit(AS_EIO, &mapping->flags))
363 ret = -EIO; 368 ret = -EIO;
364 return ret; 369 return ret;
365 } 370 }
366 EXPORT_SYMBOL(filemap_check_errors); 371 EXPORT_SYMBOL(filemap_check_errors);
367 372
368 static int filemap_check_and_keep_errors(struc 373 static int filemap_check_and_keep_errors(struct address_space *mapping)
369 { 374 {
370 /* Check for outstanding write errors 375 /* Check for outstanding write errors */
371 if (test_bit(AS_EIO, &mapping->flags)) 376 if (test_bit(AS_EIO, &mapping->flags))
372 return -EIO; 377 return -EIO;
373 if (test_bit(AS_ENOSPC, &mapping->flag 378 if (test_bit(AS_ENOSPC, &mapping->flags))
374 return -ENOSPC; 379 return -ENOSPC;
375 return 0; 380 return 0;
376 } 381 }
377 382
378 /** 383 /**
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 384 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
406 * @mapping: address space structure to wri 385 * @mapping: address space structure to write
407 * @start: offset in bytes where the rang 386 * @start: offset in bytes where the range starts
408 * @end: offset in bytes where the rang 387 * @end: offset in bytes where the range ends (inclusive)
409 * @sync_mode: enable synchronous operation 388 * @sync_mode: enable synchronous operation
410 * 389 *
411 * Start writeback against all of a mapping's 390 * Start writeback against all of a mapping's dirty pages that lie
412 * within the byte offsets <start, end> inclus 391 * within the byte offsets <start, end> inclusive.
413 * 392 *
414 * If sync_mode is WB_SYNC_ALL then this is a 393 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
415 * opposed to a regular memory cleansing write 394 * opposed to a regular memory cleansing writeback. The difference between
416 * these two operations is that if a dirty pag 395 * these two operations is that if a dirty page/buffer is encountered, it must
417 * be waited upon, and not just skipped over. 396 * be waited upon, and not just skipped over.
418 * 397 *
419 * Return: %0 on success, negative error code 398 * Return: %0 on success, negative error code otherwise.
420 */ 399 */
421 int __filemap_fdatawrite_range(struct address_ 400 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
422 loff_t end, in 401 loff_t end, int sync_mode)
423 { 402 {
>> 403 int ret;
424 struct writeback_control wbc = { 404 struct writeback_control wbc = {
425 .sync_mode = sync_mode, 405 .sync_mode = sync_mode,
426 .nr_to_write = LONG_MAX, 406 .nr_to_write = LONG_MAX,
427 .range_start = start, 407 .range_start = start,
428 .range_end = end, 408 .range_end = end,
429 }; 409 };
430 410
431 return filemap_fdatawrite_wbc(mapping, !! 411 if (!mapping_cap_writeback_dirty(mapping) ||
>> 412 !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
>> 413 return 0;
>> 414
>> 415 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
>> 416 ret = do_writepages(mapping, &wbc);
>> 417 wbc_detach_inode(&wbc);
>> 418 return ret;
432 } 419 }
433 420
434 static inline int __filemap_fdatawrite(struct 421 static inline int __filemap_fdatawrite(struct address_space *mapping,
435 int sync_mode) 422 int sync_mode)
436 { 423 {
437 return __filemap_fdatawrite_range(mapp 424 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
438 } 425 }
439 426
440 int filemap_fdatawrite(struct address_space *m 427 int filemap_fdatawrite(struct address_space *mapping)
441 { 428 {
442 return __filemap_fdatawrite(mapping, W 429 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
443 } 430 }
444 EXPORT_SYMBOL(filemap_fdatawrite); 431 EXPORT_SYMBOL(filemap_fdatawrite);
445 432
446 int filemap_fdatawrite_range(struct address_sp 433 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
447 loff_t end) 434 loff_t end)
448 { 435 {
449 return __filemap_fdatawrite_range(mapp 436 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
450 } 437 }
451 EXPORT_SYMBOL(filemap_fdatawrite_range); 438 EXPORT_SYMBOL(filemap_fdatawrite_range);
452 439
453 /** 440 /**
454 * filemap_flush - mostly a non-blocking flush 441 * filemap_flush - mostly a non-blocking flush
455 * @mapping: target address_space 442 * @mapping: target address_space
456 * 443 *
457 * This is a mostly non-blocking flush. Not s 444 * This is a mostly non-blocking flush. Not suitable for data-integrity
458 * purposes - I/O may not be started against a 445 * purposes - I/O may not be started against all dirty pages.
459 * 446 *
460 * Return: %0 on success, negative error code 447 * Return: %0 on success, negative error code otherwise.
461 */ 448 */
462 int filemap_flush(struct address_space *mappin 449 int filemap_flush(struct address_space *mapping)
463 { 450 {
464 return __filemap_fdatawrite(mapping, W 451 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
465 } 452 }
466 EXPORT_SYMBOL(filemap_flush); 453 EXPORT_SYMBOL(filemap_flush);
467 454
468 /** 455 /**
469 * filemap_range_has_page - check if a page ex 456 * filemap_range_has_page - check if a page exists in range.
470 * @mapping: address space within wh 457 * @mapping: address space within which to check
471 * @start_byte: offset in bytes where t 458 * @start_byte: offset in bytes where the range starts
472 * @end_byte: offset in bytes where t 459 * @end_byte: offset in bytes where the range ends (inclusive)
473 * 460 *
474 * Find at least one page in the range supplie 461 * Find at least one page in the range supplied, usually used to check if
475 * direct writing in this range will trigger a 462 * direct writing in this range will trigger a writeback.
476 * 463 *
477 * Return: %true if at least one page exists i 464 * Return: %true if at least one page exists in the specified range,
478 * %false otherwise. 465 * %false otherwise.
479 */ 466 */
480 bool filemap_range_has_page(struct address_spa 467 bool filemap_range_has_page(struct address_space *mapping,
481 loff_t start_byte, 468 loff_t start_byte, loff_t end_byte)
482 { 469 {
483 struct folio *folio; !! 470 struct page *page;
484 XA_STATE(xas, &mapping->i_pages, start 471 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
485 pgoff_t max = end_byte >> PAGE_SHIFT; 472 pgoff_t max = end_byte >> PAGE_SHIFT;
486 473
487 if (end_byte < start_byte) 474 if (end_byte < start_byte)
488 return false; 475 return false;
489 476
490 rcu_read_lock(); 477 rcu_read_lock();
491 for (;;) { 478 for (;;) {
492 folio = xas_find(&xas, max); !! 479 page = xas_find(&xas, max);
493 if (xas_retry(&xas, folio)) !! 480 if (xas_retry(&xas, page))
494 continue; 481 continue;
495 /* Shadow entries don't count 482 /* Shadow entries don't count */
496 if (xa_is_value(folio)) !! 483 if (xa_is_value(page))
497 continue; 484 continue;
498 /* 485 /*
499 * We don't need to try to pin 486 * We don't need to try to pin this page; we're about to
500 * release the RCU lock anyway 487 * release the RCU lock anyway. It is enough to know that
501 * there was a page here recen 488 * there was a page here recently.
502 */ 489 */
503 break; 490 break;
504 } 491 }
505 rcu_read_unlock(); 492 rcu_read_unlock();
506 493
507 return folio != NULL; !! 494 return page != NULL;
508 } 495 }
509 EXPORT_SYMBOL(filemap_range_has_page); 496 EXPORT_SYMBOL(filemap_range_has_page);
510 497
511 static void __filemap_fdatawait_range(struct a 498 static void __filemap_fdatawait_range(struct address_space *mapping,
512 loff_t st 499 loff_t start_byte, loff_t end_byte)
513 { 500 {
514 pgoff_t index = start_byte >> PAGE_SHI 501 pgoff_t index = start_byte >> PAGE_SHIFT;
515 pgoff_t end = end_byte >> PAGE_SHIFT; 502 pgoff_t end = end_byte >> PAGE_SHIFT;
516 struct folio_batch fbatch; !! 503 struct pagevec pvec;
517 unsigned nr_folios; !! 504 int nr_pages;
518 505
519 folio_batch_init(&fbatch); !! 506 if (end_byte < start_byte)
>> 507 return;
520 508
>> 509 pagevec_init(&pvec);
521 while (index <= end) { 510 while (index <= end) {
522 unsigned i; 511 unsigned i;
523 512
524 nr_folios = filemap_get_folios !! 513 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
525 PAGECACHE_TAG_ !! 514 end, PAGECACHE_TAG_WRITEBACK);
526 !! 515 if (!nr_pages)
527 if (!nr_folios) <<
528 break; 516 break;
529 517
530 for (i = 0; i < nr_folios; i++ !! 518 for (i = 0; i < nr_pages; i++) {
531 struct folio *folio = !! 519 struct page *page = pvec.pages[i];
532 520
533 folio_wait_writeback(f !! 521 wait_on_page_writeback(page);
>> 522 ClearPageError(page);
534 } 523 }
535 folio_batch_release(&fbatch); !! 524 pagevec_release(&pvec);
536 cond_resched(); 525 cond_resched();
537 } 526 }
538 } 527 }
539 528
540 /** 529 /**
541 * filemap_fdatawait_range - wait for writebac 530 * filemap_fdatawait_range - wait for writeback to complete
542 * @mapping: address space structur 531 * @mapping: address space structure to wait for
543 * @start_byte: offset in bytes where 532 * @start_byte: offset in bytes where the range starts
544 * @end_byte: offset in bytes where 533 * @end_byte: offset in bytes where the range ends (inclusive)
545 * 534 *
546 * Walk the list of under-writeback pages of t 535 * Walk the list of under-writeback pages of the given address space
547 * in the given range and wait for all of them 536 * in the given range and wait for all of them. Check error status of
548 * the address space and return it. 537 * the address space and return it.
549 * 538 *
550 * Since the error status of the address space 539 * Since the error status of the address space is cleared by this function,
551 * callers are responsible for checking the re 540 * callers are responsible for checking the return value and handling and/or
552 * reporting the error. 541 * reporting the error.
553 * 542 *
554 * Return: error status of the address space. 543 * Return: error status of the address space.
555 */ 544 */
556 int filemap_fdatawait_range(struct address_spa 545 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
557 loff_t end_byte) 546 loff_t end_byte)
558 { 547 {
559 __filemap_fdatawait_range(mapping, sta 548 __filemap_fdatawait_range(mapping, start_byte, end_byte);
560 return filemap_check_errors(mapping); 549 return filemap_check_errors(mapping);
561 } 550 }
562 EXPORT_SYMBOL(filemap_fdatawait_range); 551 EXPORT_SYMBOL(filemap_fdatawait_range);
563 552
564 /** 553 /**
565 * filemap_fdatawait_range_keep_errors - wait 554 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
566 * @mapping: address space structur 555 * @mapping: address space structure to wait for
567 * @start_byte: offset in bytes where 556 * @start_byte: offset in bytes where the range starts
568 * @end_byte: offset in bytes where 557 * @end_byte: offset in bytes where the range ends (inclusive)
569 * 558 *
570 * Walk the list of under-writeback pages of t 559 * Walk the list of under-writeback pages of the given address space in the
571 * given range and wait for all of them. Unli 560 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
572 * this function does not clear error status o 561 * this function does not clear error status of the address space.
573 * 562 *
574 * Use this function if callers don't handle e 563 * Use this function if callers don't handle errors themselves. Expected
575 * call sites are system-wide / filesystem-wid 564 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
576 * fsfreeze(8) 565 * fsfreeze(8)
577 */ 566 */
578 int filemap_fdatawait_range_keep_errors(struct 567 int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
579 loff_t start_byte, loff_t end_ 568 loff_t start_byte, loff_t end_byte)
580 { 569 {
581 __filemap_fdatawait_range(mapping, sta 570 __filemap_fdatawait_range(mapping, start_byte, end_byte);
582 return filemap_check_and_keep_errors(m 571 return filemap_check_and_keep_errors(mapping);
583 } 572 }
584 EXPORT_SYMBOL(filemap_fdatawait_range_keep_err 573 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
585 574
586 /** 575 /**
587 * file_fdatawait_range - wait for writeback t 576 * file_fdatawait_range - wait for writeback to complete
588 * @file: file pointing to addre 577 * @file: file pointing to address space structure to wait for
589 * @start_byte: offset in bytes where 578 * @start_byte: offset in bytes where the range starts
590 * @end_byte: offset in bytes where 579 * @end_byte: offset in bytes where the range ends (inclusive)
591 * 580 *
592 * Walk the list of under-writeback pages of t 581 * Walk the list of under-writeback pages of the address space that file
593 * refers to, in the given range and wait for 582 * refers to, in the given range and wait for all of them. Check error
594 * status of the address space vs. the file->f 583 * status of the address space vs. the file->f_wb_err cursor and return it.
595 * 584 *
596 * Since the error status of the file is advan 585 * Since the error status of the file is advanced by this function,
597 * callers are responsible for checking the re 586 * callers are responsible for checking the return value and handling and/or
598 * reporting the error. 587 * reporting the error.
599 * 588 *
600 * Return: error status of the address space v 589 * Return: error status of the address space vs. the file->f_wb_err cursor.
601 */ 590 */
602 int file_fdatawait_range(struct file *file, lo 591 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
603 { 592 {
604 struct address_space *mapping = file-> 593 struct address_space *mapping = file->f_mapping;
605 594
606 __filemap_fdatawait_range(mapping, sta 595 __filemap_fdatawait_range(mapping, start_byte, end_byte);
607 return file_check_and_advance_wb_err(f 596 return file_check_and_advance_wb_err(file);
608 } 597 }
609 EXPORT_SYMBOL(file_fdatawait_range); 598 EXPORT_SYMBOL(file_fdatawait_range);
610 599
611 /** 600 /**
612 * filemap_fdatawait_keep_errors - wait for wr 601 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
613 * @mapping: address space structure to wait f 602 * @mapping: address space structure to wait for
614 * 603 *
615 * Walk the list of under-writeback pages of t 604 * Walk the list of under-writeback pages of the given address space
616 * and wait for all of them. Unlike filemap_f 605 * and wait for all of them. Unlike filemap_fdatawait(), this function
617 * does not clear error status of the address 606 * does not clear error status of the address space.
618 * 607 *
619 * Use this function if callers don't handle e 608 * Use this function if callers don't handle errors themselves. Expected
620 * call sites are system-wide / filesystem-wid 609 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
621 * fsfreeze(8) 610 * fsfreeze(8)
622 * 611 *
623 * Return: error status of the address space. 612 * Return: error status of the address space.
624 */ 613 */
625 int filemap_fdatawait_keep_errors(struct addre 614 int filemap_fdatawait_keep_errors(struct address_space *mapping)
626 { 615 {
627 __filemap_fdatawait_range(mapping, 0, 616 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
628 return filemap_check_and_keep_errors(m 617 return filemap_check_and_keep_errors(mapping);
629 } 618 }
630 EXPORT_SYMBOL(filemap_fdatawait_keep_errors); 619 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
631 620
632 /* Returns true if writeback might be needed o <<
633 static bool mapping_needs_writeback(struct add 621 static bool mapping_needs_writeback(struct address_space *mapping)
634 { 622 {
635 return mapping->nrpages; !! 623 return (!dax_mapping(mapping) && mapping->nrpages) ||
>> 624 (dax_mapping(mapping) && mapping->nrexceptional);
636 } 625 }
637 626
638 bool filemap_range_has_writeback(struct addres !! 627 int filemap_write_and_wait(struct address_space *mapping)
639 loff_t start_ <<
640 { 628 {
641 XA_STATE(xas, &mapping->i_pages, start !! 629 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; <<
647 630
648 rcu_read_lock(); !! 631 if (mapping_needs_writeback(mapping)) {
649 xas_for_each(&xas, folio, max) { !! 632 err = filemap_fdatawrite(mapping);
650 if (xas_retry(&xas, folio)) !! 633 /*
651 continue; !! 634 * Even if the above returned error, the pages may be
652 if (xa_is_value(folio)) !! 635 * written partially (e.g. -ENOSPC), so we wait for it.
653 continue; !! 636 * But the -EIO is special case, it may indicate the worst
654 if (folio_test_dirty(folio) || !! 637 * thing (e.g. bug) happened, so we avoid waiting for it.
655 folio_test_wri !! 638 */
656 break; !! 639 if (err != -EIO) {
>> 640 int err2 = filemap_fdatawait(mapping);
>> 641 if (!err)
>> 642 err = err2;
>> 643 } else {
>> 644 /* Clear any previously stored errors */
>> 645 filemap_check_errors(mapping);
>> 646 }
>> 647 } else {
>> 648 err = filemap_check_errors(mapping);
657 } 649 }
658 rcu_read_unlock(); !! 650 return err;
659 return folio != NULL; <<
660 } 651 }
661 EXPORT_SYMBOL_GPL(filemap_range_has_writeback) !! 652 EXPORT_SYMBOL(filemap_write_and_wait);
662 653
663 /** 654 /**
664 * filemap_write_and_wait_range - write out & 655 * filemap_write_and_wait_range - write out & wait on a file range
665 * @mapping: the address_space for the page 656 * @mapping: the address_space for the pages
666 * @lstart: offset in bytes where the rang 657 * @lstart: offset in bytes where the range starts
667 * @lend: offset in bytes where the rang 658 * @lend: offset in bytes where the range ends (inclusive)
668 * 659 *
669 * Write out and wait upon file offsets lstart 660 * Write out and wait upon file offsets lstart->lend, inclusive.
670 * 661 *
671 * Note that @lend is inclusive (describes the 662 * Note that @lend is inclusive (describes the last byte to be written) so
672 * that this function can be used to write to 663 * that this function can be used to write to the very end-of-file (end = -1).
673 * 664 *
674 * Return: error status of the address space. 665 * Return: error status of the address space.
675 */ 666 */
676 int filemap_write_and_wait_range(struct addres 667 int filemap_write_and_wait_range(struct address_space *mapping,
677 loff_t lstart 668 loff_t lstart, loff_t lend)
678 { 669 {
679 int err = 0, err2; !! 670 int err = 0;
680 <<
681 if (lend < lstart) <<
682 return 0; <<
683 671
684 if (mapping_needs_writeback(mapping)) 672 if (mapping_needs_writeback(mapping)) {
685 err = __filemap_fdatawrite_ran 673 err = __filemap_fdatawrite_range(mapping, lstart, lend,
686 674 WB_SYNC_ALL);
687 /* !! 675 /* See comment of filemap_write_and_wait() */
688 * Even if the above returned !! 676 if (err != -EIO) {
689 * written partially (e.g. -EN !! 677 int err2 = filemap_fdatawait_range(mapping,
690 * But the -EIO is special cas !! 678 lstart, lend);
691 * thing (e.g. bug) happened, !! 679 if (!err)
692 */ !! 680 err = err2;
693 if (err != -EIO) !! 681 } else {
694 __filemap_fdatawait_ra !! 682 /* Clear any previously stored errors */
>> 683 filemap_check_errors(mapping);
>> 684 }
>> 685 } else {
>> 686 err = filemap_check_errors(mapping);
695 } 687 }
696 err2 = filemap_check_errors(mapping); <<
697 if (!err) <<
698 err = err2; <<
699 return err; 688 return err;
700 } 689 }
701 EXPORT_SYMBOL(filemap_write_and_wait_range); 690 EXPORT_SYMBOL(filemap_write_and_wait_range);
702 691
703 void __filemap_set_wb_err(struct address_space 692 void __filemap_set_wb_err(struct address_space *mapping, int err)
704 { 693 {
705 errseq_t eseq = errseq_set(&mapping->w 694 errseq_t eseq = errseq_set(&mapping->wb_err, err);
706 695
707 trace_filemap_set_wb_err(mapping, eseq 696 trace_filemap_set_wb_err(mapping, eseq);
708 } 697 }
709 EXPORT_SYMBOL(__filemap_set_wb_err); 698 EXPORT_SYMBOL(__filemap_set_wb_err);
710 699
711 /** 700 /**
712 * file_check_and_advance_wb_err - report wb e 701 * file_check_and_advance_wb_err - report wb error (if any) that was previously
713 * and advance 702 * and advance wb_err to current one
714 * @file: struct file on which the error is be 703 * @file: struct file on which the error is being reported
715 * 704 *
716 * When userland calls fsync (or something lik 705 * When userland calls fsync (or something like nfsd does the equivalent), we
717 * want to report any writeback errors that oc 706 * want to report any writeback errors that occurred since the last fsync (or
718 * since the file was opened if there haven't 707 * since the file was opened if there haven't been any).
719 * 708 *
720 * Grab the wb_err from the mapping. If it mat 709 * 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 710 * then just quickly return 0. The file is all caught up.
722 * 711 *
723 * If it doesn't match, then take the mapping 712 * 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 713 * 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 714 * 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 715 * portion. The error at this point must be reported via proper channels
727 * (a'la fsync, or NFS COMMIT operation, etc.) 716 * (a'la fsync, or NFS COMMIT operation, etc.).
728 * 717 *
729 * While we handle mapping->wb_err with atomic 718 * While we handle mapping->wb_err with atomic operations, the f_wb_err
730 * value is protected by the f_lock since we m 719 * value is protected by the f_lock since we must ensure that it reflects
731 * the latest value swapped in for this file d 720 * the latest value swapped in for this file descriptor.
732 * 721 *
733 * Return: %0 on success, negative error code 722 * Return: %0 on success, negative error code otherwise.
734 */ 723 */
735 int file_check_and_advance_wb_err(struct file 724 int file_check_and_advance_wb_err(struct file *file)
736 { 725 {
737 int err = 0; 726 int err = 0;
738 errseq_t old = READ_ONCE(file->f_wb_er 727 errseq_t old = READ_ONCE(file->f_wb_err);
739 struct address_space *mapping = file-> 728 struct address_space *mapping = file->f_mapping;
740 729
741 /* Locklessly handle the common case w 730 /* Locklessly handle the common case where nothing has changed */
742 if (errseq_check(&mapping->wb_err, old 731 if (errseq_check(&mapping->wb_err, old)) {
743 /* Something changed, must use 732 /* Something changed, must use slow path */
744 spin_lock(&file->f_lock); 733 spin_lock(&file->f_lock);
745 old = file->f_wb_err; 734 old = file->f_wb_err;
746 err = errseq_check_and_advance 735 err = errseq_check_and_advance(&mapping->wb_err,
747 736 &file->f_wb_err);
748 trace_file_check_and_advance_w 737 trace_file_check_and_advance_wb_err(file, old);
749 spin_unlock(&file->f_lock); 738 spin_unlock(&file->f_lock);
750 } 739 }
751 740
752 /* 741 /*
753 * We're mostly using this function as 742 * We're mostly using this function as a drop in replacement for
754 * filemap_check_errors. Clear AS_EIO/ 743 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
755 * that the legacy code would have had 744 * that the legacy code would have had on these flags.
756 */ 745 */
757 clear_bit(AS_EIO, &mapping->flags); 746 clear_bit(AS_EIO, &mapping->flags);
758 clear_bit(AS_ENOSPC, &mapping->flags); 747 clear_bit(AS_ENOSPC, &mapping->flags);
759 return err; 748 return err;
760 } 749 }
761 EXPORT_SYMBOL(file_check_and_advance_wb_err); 750 EXPORT_SYMBOL(file_check_and_advance_wb_err);
762 751
763 /** 752 /**
764 * file_write_and_wait_range - write out & wai 753 * file_write_and_wait_range - write out & wait on a file range
765 * @file: file pointing to address_space 754 * @file: file pointing to address_space with pages
766 * @lstart: offset in bytes where the rang 755 * @lstart: offset in bytes where the range starts
767 * @lend: offset in bytes where the rang 756 * @lend: offset in bytes where the range ends (inclusive)
768 * 757 *
769 * Write out and wait upon file offsets lstart 758 * Write out and wait upon file offsets lstart->lend, inclusive.
770 * 759 *
771 * Note that @lend is inclusive (describes the 760 * Note that @lend is inclusive (describes the last byte to be written) so
772 * that this function can be used to write to 761 * that this function can be used to write to the very end-of-file (end = -1).
773 * 762 *
774 * After writing out and waiting on the data, 763 * After writing out and waiting on the data, we check and advance the
775 * f_wb_err cursor to the latest value, and re 764 * f_wb_err cursor to the latest value, and return any errors detected there.
776 * 765 *
777 * Return: %0 on success, negative error code 766 * Return: %0 on success, negative error code otherwise.
778 */ 767 */
779 int file_write_and_wait_range(struct file *fil 768 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
780 { 769 {
781 int err = 0, err2; 770 int err = 0, err2;
782 struct address_space *mapping = file-> 771 struct address_space *mapping = file->f_mapping;
783 772
784 if (lend < lstart) <<
785 return 0; <<
786 <<
787 if (mapping_needs_writeback(mapping)) 773 if (mapping_needs_writeback(mapping)) {
788 err = __filemap_fdatawrite_ran 774 err = __filemap_fdatawrite_range(mapping, lstart, lend,
789 775 WB_SYNC_ALL);
790 /* See comment of filemap_writ 776 /* See comment of filemap_write_and_wait() */
791 if (err != -EIO) 777 if (err != -EIO)
792 __filemap_fdatawait_ra 778 __filemap_fdatawait_range(mapping, lstart, lend);
793 } 779 }
794 err2 = file_check_and_advance_wb_err(f 780 err2 = file_check_and_advance_wb_err(file);
795 if (!err) 781 if (!err)
796 err = err2; 782 err = err2;
797 return err; 783 return err;
798 } 784 }
799 EXPORT_SYMBOL(file_write_and_wait_range); 785 EXPORT_SYMBOL(file_write_and_wait_range);
800 786
801 /** 787 /**
802 * replace_page_cache_folio - replace a pageca !! 788 * replace_page_cache_page - replace a pagecache page with a new one
803 * @old: folio to be replaced !! 789 * @old: page to be replaced
804 * @new: folio to replace with !! 790 * @new: page to replace with
805 * !! 791 * @gfp_mask: allocation mode
806 * This function replaces a folio in the pagec !! 792 *
807 * success it acquires the pagecache reference !! 793 * This function replaces a page in the pagecache with a new one. On
808 * drops it for the old folio. Both the old a !! 794 * success it acquires the pagecache reference for the new page and
809 * locked. This function does not add the new !! 795 * drops it for the old page. Both the old and new pages must be
>> 796 * locked. This function does not add the new page to the LRU, the
810 * caller must do that. 797 * caller must do that.
811 * 798 *
812 * The remove + add is atomic. This function 799 * The remove + add is atomic. This function cannot fail.
>> 800 *
>> 801 * Return: %0
813 */ 802 */
814 void replace_page_cache_folio(struct folio *ol !! 803 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
815 { 804 {
816 struct address_space *mapping = old->m 805 struct address_space *mapping = old->mapping;
817 void (*free_folio)(struct folio *) = m !! 806 void (*freepage)(struct page *) = mapping->a_ops->freepage;
818 pgoff_t offset = old->index; 807 pgoff_t offset = old->index;
819 XA_STATE(xas, &mapping->i_pages, offse 808 XA_STATE(xas, &mapping->i_pages, offset);
>> 809 unsigned long flags;
820 810
821 VM_BUG_ON_FOLIO(!folio_test_locked(old !! 811 VM_BUG_ON_PAGE(!PageLocked(old), old);
822 VM_BUG_ON_FOLIO(!folio_test_locked(new !! 812 VM_BUG_ON_PAGE(!PageLocked(new), new);
823 VM_BUG_ON_FOLIO(new->mapping, new); !! 813 VM_BUG_ON_PAGE(new->mapping, new);
824 814
825 folio_get(new); !! 815 get_page(new);
826 new->mapping = mapping; 816 new->mapping = mapping;
827 new->index = offset; 817 new->index = offset;
828 818
829 mem_cgroup_replace_folio(old, new); !! 819 xas_lock_irqsave(&xas, flags);
830 <<
831 xas_lock_irq(&xas); <<
832 xas_store(&xas, new); 820 xas_store(&xas, new);
833 821
834 old->mapping = NULL; 822 old->mapping = NULL;
835 /* hugetlb pages do not participate in 823 /* hugetlb pages do not participate in page cache accounting. */
836 if (!folio_test_hugetlb(old)) !! 824 if (!PageHuge(old))
837 __lruvec_stat_sub_folio(old, N !! 825 __dec_node_page_state(new, NR_FILE_PAGES);
838 if (!folio_test_hugetlb(new)) !! 826 if (!PageHuge(new))
839 __lruvec_stat_add_folio(new, N !! 827 __inc_node_page_state(new, NR_FILE_PAGES);
840 if (folio_test_swapbacked(old)) !! 828 if (PageSwapBacked(old))
841 __lruvec_stat_sub_folio(old, N !! 829 __dec_node_page_state(new, NR_SHMEM);
842 if (folio_test_swapbacked(new)) !! 830 if (PageSwapBacked(new))
843 __lruvec_stat_add_folio(new, N !! 831 __inc_node_page_state(new, NR_SHMEM);
844 xas_unlock_irq(&xas); !! 832 xas_unlock_irqrestore(&xas, flags);
845 if (free_folio) !! 833 mem_cgroup_migrate(old, new);
846 free_folio(old); !! 834 if (freepage)
847 folio_put(old); !! 835 freepage(old);
>> 836 put_page(old);
>> 837
>> 838 return 0;
848 } 839 }
849 EXPORT_SYMBOL_GPL(replace_page_cache_folio); !! 840 EXPORT_SYMBOL_GPL(replace_page_cache_page);
850 841
851 noinline int __filemap_add_folio(struct addres !! 842 static int __add_to_page_cache_locked(struct page *page,
852 struct folio *folio, pgoff_t i !! 843 struct address_space *mapping,
>> 844 pgoff_t offset, gfp_t gfp_mask,
>> 845 void **shadowp)
853 { 846 {
854 XA_STATE(xas, &mapping->i_pages, index !! 847 XA_STATE(xas, &mapping->i_pages, offset);
855 void *alloced_shadow = NULL; !! 848 int huge = PageHuge(page);
856 int alloced_order = 0; !! 849 struct mem_cgroup *memcg;
857 bool huge; !! 850 int error;
858 long nr; !! 851 void *old;
859 !! 852
860 VM_BUG_ON_FOLIO(!folio_test_locked(fol !! 853 VM_BUG_ON_PAGE(!PageLocked(page), page);
861 VM_BUG_ON_FOLIO(folio_test_swapbacked( !! 854 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
862 VM_BUG_ON_FOLIO(folio_order(folio) < m <<
863 folio); <<
864 mapping_set_update(&xas, mapping); 855 mapping_set_update(&xas, mapping);
865 856
866 VM_BUG_ON_FOLIO(index & (folio_nr_page !! 857 if (!huge) {
867 xas_set_order(&xas, index, folio_order !! 858 error = mem_cgroup_try_charge(page, current->mm,
868 huge = folio_test_hugetlb(folio); !! 859 gfp_mask, &memcg, false);
869 nr = folio_nr_pages(folio); !! 860 if (error)
870 !! 861 return error;
871 gfp &= GFP_RECLAIM_MASK; !! 862 }
872 folio_ref_add(folio, nr); <<
873 folio->mapping = mapping; <<
874 folio->index = xas.xa_index; <<
875 863
876 for (;;) { !! 864 get_page(page);
877 int order = -1, split_order = !! 865 page->mapping = mapping;
878 void *entry, *old = NULL; !! 866 page->index = offset;
879 867
>> 868 do {
880 xas_lock_irq(&xas); 869 xas_lock_irq(&xas);
881 xas_for_each_conflict(&xas, en !! 870 old = xas_load(&xas);
882 old = entry; !! 871 if (old && !xa_is_value(old))
883 if (!xa_is_value(entry !! 872 xas_set_err(&xas, -EEXIST);
884 xas_set_err(&x !! 873 xas_store(&xas, page);
885 goto unlock; !! 874 if (xas_error(&xas))
886 } !! 875 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 } <<
894 876
895 /* entry may have changed befo !! 877 if (xa_is_value(old)) {
896 if (alloced_order && (old != a !! 878 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) 879 if (shadowp)
913 *shadowp = old 880 *shadowp = old;
914 } 881 }
915 !! 882 mapping->nrpages++;
916 xas_store(&xas, folio); <<
917 if (xas_error(&xas)) <<
918 goto unlock; <<
919 <<
920 mapping->nrpages += nr; <<
921 883
922 /* hugetlb pages do not partic 884 /* hugetlb pages do not participate in page cache accounting */
923 if (!huge) { !! 885 if (!huge)
924 __lruvec_stat_mod_foli !! 886 __inc_node_page_state(page, NR_FILE_PAGES);
925 if (folio_test_pmd_map <<
926 __lruvec_stat_ <<
927 <<
928 } <<
929 <<
930 unlock: 887 unlock:
931 xas_unlock_irq(&xas); 888 xas_unlock_irq(&xas);
932 !! 889 } 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 } <<
947 890
948 if (xas_error(&xas)) 891 if (xas_error(&xas))
949 goto error; 892 goto error;
950 893
951 trace_mm_filemap_add_to_page_cache(fol !! 894 if (!huge)
>> 895 mem_cgroup_commit_charge(page, memcg, false, false);
>> 896 trace_mm_filemap_add_to_page_cache(page);
952 return 0; 897 return 0;
953 error: 898 error:
954 folio->mapping = NULL; !! 899 page->mapping = NULL;
955 /* Leave page->index set: truncation r 900 /* Leave page->index set: truncation relies upon it */
956 folio_put_refs(folio, nr); !! 901 if (!huge)
>> 902 mem_cgroup_cancel_charge(page, memcg, false);
>> 903 put_page(page);
957 return xas_error(&xas); 904 return xas_error(&xas);
958 } 905 }
959 ALLOW_ERROR_INJECTION(__filemap_add_folio, ERR !! 906 ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO);
960 907
961 int filemap_add_folio(struct address_space *ma !! 908 /**
962 pgoff_t index, !! 909 * add_to_page_cache_locked - add a locked page to the pagecache
>> 910 * @page: page to add
>> 911 * @mapping: the page's address_space
>> 912 * @offset: page index
>> 913 * @gfp_mask: page allocation mode
>> 914 *
>> 915 * This function is used to add a page to the pagecache. It must be locked.
>> 916 * This function does not add the page to the LRU. The caller must do that.
>> 917 *
>> 918 * Return: %0 on success, negative error code otherwise.
>> 919 */
>> 920 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
>> 921 pgoff_t offset, gfp_t gfp_mask)
>> 922 {
>> 923 return __add_to_page_cache_locked(page, mapping, offset,
>> 924 gfp_mask, NULL);
>> 925 }
>> 926 EXPORT_SYMBOL(add_to_page_cache_locked);
>> 927
>> 928 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
>> 929 pgoff_t offset, gfp_t gfp_mask)
963 { 930 {
964 void *shadow = NULL; 931 void *shadow = NULL;
965 int ret; 932 int ret;
966 933
967 ret = mem_cgroup_charge(folio, NULL, g !! 934 __SetPageLocked(page);
968 if (ret) !! 935 ret = __add_to_page_cache_locked(page, mapping, offset,
969 return ret; !! 936 gfp_mask, &shadow);
970 !! 937 if (unlikely(ret))
971 __folio_set_locked(folio); !! 938 __ClearPageLocked(page);
972 ret = __filemap_add_folio(mapping, fol !! 939 else {
973 if (unlikely(ret)) { <<
974 mem_cgroup_uncharge(folio); <<
975 __folio_clear_locked(folio); <<
976 } else { <<
977 /* 940 /*
978 * The folio might have been e !! 941 * The page might have been evicted from cache only
979 * recently, in which case it 942 * recently, in which case it should be activated like
980 * any other repeatedly access !! 943 * any other repeatedly accessed page.
981 * The exception is folios get !! 944 * The exception is pages getting rewritten; evicting other
982 * data from the working set, 945 * data from the working set, only to cache data that will
983 * get overwritten with someth 946 * get overwritten with something else, is a waste of memory.
984 */ 947 */
985 WARN_ON_ONCE(folio_test_active !! 948 WARN_ON_ONCE(PageActive(page));
986 if (!(gfp & __GFP_WRITE) && sh !! 949 if (!(gfp_mask & __GFP_WRITE) && shadow)
987 workingset_refault(fol !! 950 workingset_refault(page, shadow);
988 folio_add_lru(folio); !! 951 lru_cache_add(page);
989 } 952 }
990 return ret; 953 return ret;
991 } 954 }
992 EXPORT_SYMBOL_GPL(filemap_add_folio); !! 955 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
993 956
994 #ifdef CONFIG_NUMA 957 #ifdef CONFIG_NUMA
995 struct folio *filemap_alloc_folio_noprof(gfp_t !! 958 struct page *__page_cache_alloc(gfp_t gfp)
996 { 959 {
997 int n; 960 int n;
998 struct folio *folio; !! 961 struct page *page;
999 962
1000 if (cpuset_do_page_mem_spread()) { 963 if (cpuset_do_page_mem_spread()) {
1001 unsigned int cpuset_mems_cook 964 unsigned int cpuset_mems_cookie;
1002 do { 965 do {
1003 cpuset_mems_cookie = 966 cpuset_mems_cookie = read_mems_allowed_begin();
1004 n = cpuset_mem_spread 967 n = cpuset_mem_spread_node();
1005 folio = __folio_alloc !! 968 page = __alloc_pages_node(n, gfp, 0);
1006 } while (!folio && read_mems_ !! 969 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
1007 970
1008 return folio; !! 971 return page;
1009 } 972 }
1010 return folio_alloc_noprof(gfp, order) !! 973 return alloc_pages(gfp, 0);
1011 } 974 }
1012 EXPORT_SYMBOL(filemap_alloc_folio_noprof); !! 975 EXPORT_SYMBOL(__page_cache_alloc);
1013 #endif 976 #endif
1014 977
1015 /* 978 /*
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 979 * In order to wait for pages to become available there must be
1055 * waitqueues associated with pages. By using 980 * waitqueues associated with pages. By using a hash table of
1056 * waitqueues where the bucket discipline is 981 * waitqueues where the bucket discipline is to maintain all
1057 * waiters on the same queue and wake all whe 982 * waiters on the same queue and wake all when any of the pages
1058 * become available, and for the woken contex 983 * become available, and for the woken contexts to check to be
1059 * sure the appropriate page became available 984 * sure the appropriate page became available, this saves space
1060 * at a cost of "thundering herd" phenomena d 985 * at a cost of "thundering herd" phenomena during rare hash
1061 * collisions. 986 * collisions.
1062 */ 987 */
1063 #define PAGE_WAIT_TABLE_BITS 8 988 #define PAGE_WAIT_TABLE_BITS 8
1064 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_ 989 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1065 static wait_queue_head_t folio_wait_table[PAG !! 990 static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1066 991
1067 static wait_queue_head_t *folio_waitqueue(str !! 992 static wait_queue_head_t *page_waitqueue(struct page *page)
1068 { 993 {
1069 return &folio_wait_table[hash_ptr(fol !! 994 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
1070 } 995 }
1071 996
1072 void __init pagecache_init(void) 997 void __init pagecache_init(void)
1073 { 998 {
1074 int i; 999 int i;
1075 1000
1076 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; 1001 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1077 init_waitqueue_head(&folio_wa !! 1002 init_waitqueue_head(&page_wait_table[i]);
1078 1003
1079 page_writeback_init(); 1004 page_writeback_init();
1080 } 1005 }
1081 1006
1082 /* !! 1007 /* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */
1083 * The page wait code treats the "wait->flags !! 1008 struct wait_page_key {
1084 * we have multiple different kinds of waits, !! 1009 struct page *page;
1085 * one. !! 1010 int bit_nr;
1086 * !! 1011 int page_match;
1087 * We have: !! 1012 };
1088 * !! 1013
1089 * (a) no special bits set: !! 1014 struct wait_page_queue {
1090 * !! 1015 struct page *page;
1091 * We're just waiting for the bit to be !! 1016 int bit_nr;
1092 * calls the wakeup function, we set WQ_ !! 1017 wait_queue_entry_t wait;
1093 * and remove it from the wait queue. !! 1018 };
1094 * !! 1019
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 1020 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1117 { 1021 {
1118 unsigned int flags; <<
1119 struct wait_page_key *key = arg; 1022 struct wait_page_key *key = arg;
1120 struct wait_page_queue *wait_page 1023 struct wait_page_queue *wait_page
1121 = container_of(wait, struct w 1024 = container_of(wait, struct wait_page_queue, wait);
1122 1025
1123 if (!wake_page_match(wait_page, key)) !! 1026 if (wait_page->page != key->page)
1124 return 0; !! 1027 return 0;
>> 1028 key->page_match = 1;
1125 1029
1126 /* !! 1030 if (wait_page->bit_nr != key->bit_nr)
1127 * If it's a lock handoff wait, we ge !! 1031 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 1032
1141 /* 1033 /*
1142 * We are holding the wait-queue lock !! 1034 * Stop walking if it's locked.
1143 * is waiting for this will be checki !! 1035 * Is this safe if put_and_wait_on_page_locked() is in use?
1144 * any locking. !! 1036 * Yes: the waker must hold a reference to this page, and if PG_locked
1145 * !! 1037 * has now already been set by another task, that task must also hold
1146 * So update the flags atomically, an !! 1038 * a reference to the *same usage* of this page; so there is no need
1147 * afterwards to avoid any races. Thi !! 1039 * to walk on to wake even the put_and_wait_on_page_locked() callers.
1148 * with the load-acquire in folio_wai <<
1149 */ 1040 */
1150 smp_store_release(&wait->flags, flags !! 1041 if (test_bit(key->bit_nr, &key->page->flags))
1151 wake_up_state(wait->private, mode); !! 1042 return -1;
1152 1043
1153 /* !! 1044 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 } 1045 }
1166 1046
1167 static void folio_wake_bit(struct folio *foli !! 1047 static void wake_up_page_bit(struct page *page, int bit_nr)
1168 { 1048 {
1169 wait_queue_head_t *q = folio_waitqueu !! 1049 wait_queue_head_t *q = page_waitqueue(page);
1170 struct wait_page_key key; 1050 struct wait_page_key key;
1171 unsigned long flags; 1051 unsigned long flags;
>> 1052 wait_queue_entry_t bookmark;
1172 1053
1173 key.folio = folio; !! 1054 key.page = page;
1174 key.bit_nr = bit_nr; 1055 key.bit_nr = bit_nr;
1175 key.page_match = 0; 1056 key.page_match = 0;
1176 1057
>> 1058 bookmark.flags = 0;
>> 1059 bookmark.private = NULL;
>> 1060 bookmark.func = NULL;
>> 1061 INIT_LIST_HEAD(&bookmark.entry);
>> 1062
1177 spin_lock_irqsave(&q->lock, flags); 1063 spin_lock_irqsave(&q->lock, flags);
1178 __wake_up_locked_key(q, TASK_NORMAL, !! 1064 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
>> 1065
>> 1066 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
>> 1067 /*
>> 1068 * Take a breather from holding the lock,
>> 1069 * allow pages that finish wake up asynchronously
>> 1070 * to acquire the lock and remove themselves
>> 1071 * from wait queue
>> 1072 */
>> 1073 spin_unlock_irqrestore(&q->lock, flags);
>> 1074 cpu_relax();
>> 1075 spin_lock_irqsave(&q->lock, flags);
>> 1076 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
>> 1077 }
1179 1078
1180 /* 1079 /*
1181 * It's possible to miss clearing wai !! 1080 * It is possible for other pages to have collided on the waitqueue
1182 * waiters, but the hashed waitqueue !! 1081 * hash, so in that case check for a page match. That prevents a long-
1183 * That's okay, it's a rare case. The !! 1082 * term waiter
1184 * 1083 *
1185 * Note that, depending on the page p !! 1084 * It is still possible to miss a case here, when we woke page waiters
1186 * other), the flag may be cleared in !! 1085 * and removed them from the waitqueue, but there are still other
1187 * but that is not required for corre !! 1086 * page waiters.
1188 */ !! 1087 */
1189 if (!waitqueue_active(q) || !key.page !! 1088 if (!waitqueue_active(q) || !key.page_match) {
1190 folio_clear_waiters(folio); !! 1089 ClearPageWaiters(page);
1191 !! 1090 /*
>> 1091 * It's possible to miss clearing Waiters here, when we woke
>> 1092 * our page waiters, but the hashed waitqueue has waiters for
>> 1093 * other pages on it.
>> 1094 *
>> 1095 * That's okay, it's a rare case. The next waker will clear it.
>> 1096 */
>> 1097 }
1192 spin_unlock_irqrestore(&q->lock, flag 1098 spin_unlock_irqrestore(&q->lock, flags);
1193 } 1099 }
1194 1100
>> 1101 static void wake_up_page(struct page *page, int bit)
>> 1102 {
>> 1103 if (!PageWaiters(page))
>> 1104 return;
>> 1105 wake_up_page_bit(page, bit);
>> 1106 }
>> 1107
1195 /* 1108 /*
1196 * A choice of three behaviors for folio_wait !! 1109 * A choice of three behaviors for wait_on_page_bit_common():
1197 */ 1110 */
1198 enum behavior { 1111 enum behavior {
1199 EXCLUSIVE, /* Hold ref to page a 1112 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1200 * __folio_lock() wai !! 1113 * __lock_page() waiting on then setting PG_locked.
1201 */ 1114 */
1202 SHARED, /* Hold ref to page a 1115 SHARED, /* Hold ref to page and check the bit when woken, like
1203 * folio_wait_writeba !! 1116 * wait_on_page_writeback() waiting on PG_writeback.
1204 */ 1117 */
1205 DROP, /* Drop ref to page b 1118 DROP, /* Drop ref to page before wait, no check when woken,
1206 * like folio_put_wai !! 1119 * like put_and_wait_on_page_locked() on PG_locked.
1207 */ 1120 */
1208 }; 1121 };
1209 1122
1210 /* !! 1123 static inline int wait_on_page_bit_common(wait_queue_head_t *q,
1211 * Attempt to check (or get) the folio flag, !! 1124 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 { 1125 {
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; 1126 struct wait_page_queue wait_page;
1236 wait_queue_entry_t *wait = &wait_page 1127 wait_queue_entry_t *wait = &wait_page.wait;
>> 1128 bool bit_is_set;
1237 bool thrashing = false; 1129 bool thrashing = false;
>> 1130 bool delayacct = false;
1238 unsigned long pflags; 1131 unsigned long pflags;
1239 bool in_thrashing; !! 1132 int ret = 0;
1240 1133
1241 if (bit_nr == PG_locked && 1134 if (bit_nr == PG_locked &&
1242 !folio_test_uptodate(folio) && fo !! 1135 !PageUptodate(page) && PageWorkingset(page)) {
1243 delayacct_thrashing_start(&in !! 1136 if (!PageSwapBacked(page)) {
>> 1137 delayacct_thrashing_start();
>> 1138 delayacct = true;
>> 1139 }
1244 psi_memstall_enter(&pflags); 1140 psi_memstall_enter(&pflags);
1245 thrashing = true; 1141 thrashing = true;
1246 } 1142 }
1247 1143
1248 init_wait(wait); 1144 init_wait(wait);
>> 1145 wait->flags = behavior == EXCLUSIVE ? WQ_FLAG_EXCLUSIVE : 0;
1249 wait->func = wake_page_function; 1146 wait->func = wake_page_function;
1250 wait_page.folio = folio; !! 1147 wait_page.page = page;
1251 wait_page.bit_nr = bit_nr; 1148 wait_page.bit_nr = bit_nr;
1252 1149
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 (;;) { 1150 for (;;) {
1299 unsigned int flags; !! 1151 spin_lock_irq(&q->lock);
>> 1152
>> 1153 if (likely(list_empty(&wait->entry))) {
>> 1154 __add_wait_queue_entry_tail(q, wait);
>> 1155 SetPageWaiters(page);
>> 1156 }
1300 1157
1301 set_current_state(state); 1158 set_current_state(state);
1302 1159
1303 /* Loop until we've been woke !! 1160 spin_unlock_irq(&q->lock);
1304 flags = smp_load_acquire(&wai !! 1161
1305 if (!(flags & WQ_FLAG_WOKEN)) !! 1162 bit_is_set = test_bit(bit_nr, &page->flags);
1306 if (signal_pending_st !! 1163 if (behavior == DROP)
1307 break; !! 1164 put_page(page);
1308 1165
>> 1166 if (likely(bit_is_set))
1309 io_schedule(); 1167 io_schedule();
1310 continue; !! 1168
>> 1169 if (behavior == EXCLUSIVE) {
>> 1170 if (!test_and_set_bit_lock(bit_nr, &page->flags))
>> 1171 break;
>> 1172 } else if (behavior == SHARED) {
>> 1173 if (!test_bit(bit_nr, &page->flags))
>> 1174 break;
1311 } 1175 }
1312 1176
1313 /* If we were non-exclusive, !! 1177 if (signal_pending_state(state, current)) {
1314 if (behavior != EXCLUSIVE) !! 1178 ret = -EINTR;
1315 break; 1179 break;
>> 1180 }
1316 1181
1317 /* If the waker got the lock !! 1182 if (behavior == DROP) {
1318 if (flags & WQ_FLAG_DONE) !! 1183 /*
>> 1184 * We can no longer safely access page->flags:
>> 1185 * even if CONFIG_MEMORY_HOTREMOVE is not enabled,
>> 1186 * there is a risk of waiting forever on a page reused
>> 1187 * for something that keeps it locked indefinitely.
>> 1188 * But best check for -EINTR above before breaking.
>> 1189 */
1319 break; 1190 break;
1320 !! 1191 }
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 } 1192 }
1333 1193
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); 1194 finish_wait(q, wait);
1341 1195
1342 if (thrashing) { 1196 if (thrashing) {
1343 delayacct_thrashing_end(&in_t !! 1197 if (delayacct)
>> 1198 delayacct_thrashing_end();
1344 psi_memstall_leave(&pflags); 1199 psi_memstall_leave(&pflags);
1345 } 1200 }
1346 1201
1347 /* 1202 /*
1348 * NOTE! The wait->flags weren't stab !! 1203 * A signal could leave PageWaiters set. Clearing it here if
1349 * 'finish_wait()', and we could have !! 1204 * !waitqueue_active would be possible (by open-coding finish_wait),
1350 * to a signal, and had a wakeup even !! 1205 * but still fail to catch it in the case of wait hash collision. We
1351 * test but before the 'finish_wait() !! 1206 * already can fail to clear wait hash collision cases, so don't
1352 * !! 1207 * 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 */ 1208 */
1360 if (behavior == EXCLUSIVE) <<
1361 return wait->flags & WQ_FLAG_ <<
1362 1209
1363 return wait->flags & WQ_FLAG_WOKEN ? !! 1210 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 } 1211 }
1444 #endif <<
1445 1212
1446 void folio_wait_bit(struct folio *folio, int !! 1213 void wait_on_page_bit(struct page *page, int bit_nr)
1447 { 1214 {
1448 folio_wait_bit_common(folio, bit_nr, !! 1215 wait_queue_head_t *q = page_waitqueue(page);
>> 1216 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1449 } 1217 }
1450 EXPORT_SYMBOL(folio_wait_bit); !! 1218 EXPORT_SYMBOL(wait_on_page_bit);
1451 1219
1452 int folio_wait_bit_killable(struct folio *fol !! 1220 int wait_on_page_bit_killable(struct page *page, int bit_nr)
1453 { 1221 {
1454 return folio_wait_bit_common(folio, b !! 1222 wait_queue_head_t *q = page_waitqueue(page);
>> 1223 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED);
1455 } 1224 }
1456 EXPORT_SYMBOL(folio_wait_bit_killable); !! 1225 EXPORT_SYMBOL(wait_on_page_bit_killable);
1457 1226
1458 /** 1227 /**
1459 * folio_put_wait_locked - Drop a reference a !! 1228 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1460 * @folio: The folio to wait for. !! 1229 * @page: The page to wait for.
1461 * @state: The sleep state (TASK_KILLABLE, TA <<
1462 * 1230 *
1463 * The caller should hold a reference on @fol !! 1231 * The caller should hold a reference on @page. They expect the page to
1464 * become unlocked relatively soon, but do no 1232 * become unlocked relatively soon, but do not wish to hold up migration
1465 * (for example) by holding the reference whi !! 1233 * (for example) by holding the reference while waiting for the page to
1466 * come unlocked. After this function return 1234 * come unlocked. After this function returns, the caller should not
1467 * dereference @folio. !! 1235 * dereference @page.
1468 * <<
1469 * Return: 0 if the folio was unlocked or -EI <<
1470 */ 1236 */
1471 static int folio_put_wait_locked(struct folio !! 1237 void put_and_wait_on_page_locked(struct page *page)
1472 { 1238 {
1473 return folio_wait_bit_common(folio, P !! 1239 wait_queue_head_t *q;
>> 1240
>> 1241 page = compound_head(page);
>> 1242 q = page_waitqueue(page);
>> 1243 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, DROP);
1474 } 1244 }
1475 1245
1476 /** 1246 /**
1477 * folio_add_wait_queue - Add an arbitrary wa !! 1247 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1478 * @folio: Folio defining the wait queue of i !! 1248 * @page: Page defining the wait queue of interest
1479 * @waiter: Waiter to add to the queue 1249 * @waiter: Waiter to add to the queue
1480 * 1250 *
1481 * Add an arbitrary @waiter to the wait queue !! 1251 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1482 */ 1252 */
1483 void folio_add_wait_queue(struct folio *folio !! 1253 void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1484 { 1254 {
1485 wait_queue_head_t *q = folio_waitqueu !! 1255 wait_queue_head_t *q = page_waitqueue(page);
1486 unsigned long flags; 1256 unsigned long flags;
1487 1257
1488 spin_lock_irqsave(&q->lock, flags); 1258 spin_lock_irqsave(&q->lock, flags);
1489 __add_wait_queue_entry_tail(q, waiter 1259 __add_wait_queue_entry_tail(q, waiter);
1490 folio_set_waiters(folio); !! 1260 SetPageWaiters(page);
1491 spin_unlock_irqrestore(&q->lock, flag 1261 spin_unlock_irqrestore(&q->lock, flags);
1492 } 1262 }
1493 EXPORT_SYMBOL_GPL(folio_add_wait_queue); !! 1263 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 1264
1515 /** !! 1265 #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 1266
1545 /** !! 1267 /*
1546 * folio_end_private_2 - Clear PG_private_2 a !! 1268 * PG_waiters is the high bit in the same byte as PG_lock.
1547 * @folio: The folio. <<
1548 * 1269 *
1549 * Clear the PG_private_2 bit on a folio and !! 1270 * On x86 (and on many other architectures), we can clear PG_lock and
1550 * it. The folio reference held for PG_priva !! 1271 * test the sign bit at the same time. But if the architecture does
>> 1272 * not support that special operation, we just do this all by hand
>> 1273 * instead.
1551 * 1274 *
1552 * This is, for example, used when a netfs fo !! 1275 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1553 * disk cache, thereby allowing writes to the !! 1276 * being cleared, but a memory barrier should be unneccssary since it is
1554 * serialised. !! 1277 * in the same byte as PG_locked.
1555 */ 1278 */
1556 void folio_end_private_2(struct folio *folio) !! 1279 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1557 { 1280 {
1558 VM_BUG_ON_FOLIO(!folio_test_private_2 !! 1281 clear_bit_unlock(nr, mem);
1559 clear_bit_unlock(PG_private_2, folio_ !! 1282 /* smp_mb__after_atomic(); */
1560 folio_wake_bit(folio, PG_private_2); !! 1283 return test_bit(PG_waiters, mem);
1561 folio_put(folio); <<
1562 } 1284 }
1563 EXPORT_SYMBOL(folio_end_private_2); !! 1285
>> 1286 #endif
1564 1287
1565 /** 1288 /**
1566 * folio_wait_private_2 - Wait for PG_private !! 1289 * unlock_page - unlock a locked page
1567 * @folio: The folio to wait on. !! 1290 * @page: the page
1568 * 1291 *
1569 * Wait for PG_private_2 to be cleared on a f !! 1292 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
>> 1293 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
>> 1294 * mechanism between PageLocked pages and PageWriteback pages is shared.
>> 1295 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
>> 1296 *
>> 1297 * Note that this depends on PG_waiters being the sign bit in the byte
>> 1298 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
>> 1299 * clear the PG_locked bit and test PG_waiters at the same time fairly
>> 1300 * portably (architectures that do LL/SC can test any bit, while x86 can
>> 1301 * test the sign bit).
1570 */ 1302 */
1571 void folio_wait_private_2(struct folio *folio !! 1303 void unlock_page(struct page *page)
1572 { 1304 {
1573 while (folio_test_private_2(folio)) !! 1305 BUILD_BUG_ON(PG_waiters != 7);
1574 folio_wait_bit(folio, PG_priv !! 1306 page = compound_head(page);
>> 1307 VM_BUG_ON_PAGE(!PageLocked(page), page);
>> 1308 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
>> 1309 wake_up_page_bit(page, PG_locked);
1575 } 1310 }
1576 EXPORT_SYMBOL(folio_wait_private_2); !! 1311 EXPORT_SYMBOL(unlock_page);
1577 1312
1578 /** 1313 /**
1579 * folio_wait_private_2_killable - Wait for P !! 1314 * end_page_writeback - end writeback against a page
1580 * @folio: The folio to wait on. !! 1315 * @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 */ 1316 */
1589 int folio_wait_private_2_killable(struct foli !! 1317 void end_page_writeback(struct page *page)
1590 { 1318 {
1591 int ret = 0; !! 1319 /*
1592 !! 1320 * TestClearPageReclaim could be used here but it is an atomic
1593 while (folio_test_private_2(folio)) { !! 1321 * operation and overkill in this particular case. Failing to
1594 ret = folio_wait_bit_killable !! 1322 * shuffle a page marked for immediate reclaim is too mild to
1595 if (ret < 0) !! 1323 * justify taking an atomic operation penalty at the end of
1596 break; !! 1324 * ever page writeback.
>> 1325 */
>> 1326 if (PageReclaim(page)) {
>> 1327 ClearPageReclaim(page);
>> 1328 rotate_reclaimable_page(page);
1597 } 1329 }
1598 1330
1599 return ret; !! 1331 if (!test_clear_page_writeback(page))
>> 1332 BUG();
>> 1333
>> 1334 smp_mb__after_atomic();
>> 1335 wake_up_page(page, PG_writeback);
1600 } 1336 }
1601 EXPORT_SYMBOL(folio_wait_private_2_killable); !! 1337 EXPORT_SYMBOL(end_page_writeback);
1602 1338
1603 /** !! 1339 /*
1604 * folio_end_writeback - End writeback agains !! 1340 * After completing I/O on a page, call this routine to update the page
1605 * @folio: The folio. !! 1341 * flags appropriately
1606 * <<
1607 * The folio must actually be under writeback <<
1608 * <<
1609 * Context: May be called from process or int <<
1610 */ 1342 */
1611 void folio_end_writeback(struct folio *folio) !! 1343 void page_endio(struct page *page, bool is_write, int err)
1612 { 1344 {
1613 VM_BUG_ON_FOLIO(!folio_test_writeback !! 1345 if (!is_write) {
>> 1346 if (!err) {
>> 1347 SetPageUptodate(page);
>> 1348 } else {
>> 1349 ClearPageUptodate(page);
>> 1350 SetPageError(page);
>> 1351 }
>> 1352 unlock_page(page);
>> 1353 } else {
>> 1354 if (err) {
>> 1355 struct address_space *mapping;
1614 1356
1615 /* !! 1357 SetPageError(page);
1616 * folio_test_clear_reclaim() could b !! 1358 mapping = page_mapping(page);
1617 * atomic operation and overkill in t !! 1359 if (mapping)
1618 * to shuffle a folio marked for imme !! 1360 mapping_set_error(mapping, err);
1619 * a gain to justify taking an atomic !! 1361 }
1620 * end of every folio writeback. !! 1362 end_page_writeback(page);
1621 */ <<
1622 if (folio_test_reclaim(folio)) { <<
1623 folio_clear_reclaim(folio); <<
1624 folio_rotate_reclaimable(foli <<
1625 } 1363 }
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 } 1364 }
1639 EXPORT_SYMBOL(folio_end_writeback); !! 1365 EXPORT_SYMBOL_GPL(page_endio);
1640 1366
1641 /** 1367 /**
1642 * __folio_lock - Get a lock on the folio, as !! 1368 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1643 * @folio: The folio to lock !! 1369 * @__page: the page to lock
1644 */ 1370 */
1645 void __folio_lock(struct folio *folio) !! 1371 void __lock_page(struct page *__page)
1646 { 1372 {
1647 folio_wait_bit_common(folio, PG_locke !! 1373 struct page *page = compound_head(__page);
>> 1374 wait_queue_head_t *q = page_waitqueue(page);
>> 1375 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE,
1648 EXCLUSIVE); 1376 EXCLUSIVE);
1649 } 1377 }
1650 EXPORT_SYMBOL(__folio_lock); !! 1378 EXPORT_SYMBOL(__lock_page);
1651 1379
1652 int __folio_lock_killable(struct folio *folio !! 1380 int __lock_page_killable(struct page *__page)
1653 { 1381 {
1654 return folio_wait_bit_common(folio, P !! 1382 struct page *page = compound_head(__page);
>> 1383 wait_queue_head_t *q = page_waitqueue(page);
>> 1384 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE,
1655 EXCLU 1385 EXCLUSIVE);
1656 } 1386 }
1657 EXPORT_SYMBOL_GPL(__folio_lock_killable); !! 1387 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 1388
1685 /* 1389 /*
1686 * Return values: 1390 * Return values:
1687 * 0 - folio is locked. !! 1391 * 1 - page is locked; mmap_sem is still held.
1688 * non-zero - folio is not locked. !! 1392 * 0 - page is not locked.
1689 * mmap_lock or per-VMA lock has been rel !! 1393 * mmap_sem has been released (up_read()), unless flags had both
1690 * vma_end_read()), unless flags had both !! 1394 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1691 * FAULT_FLAG_RETRY_NOWAIT set, in which !! 1395 * which case mmap_sem is still held.
1692 * !! 1396 *
1693 * If neither ALLOW_RETRY nor KILLABLE are se !! 1397 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1694 * with the folio locked and the mmap_lock/pe !! 1398 * with the page locked and the mmap_sem unperturbed.
1695 */ 1399 */
1696 vm_fault_t __folio_lock_or_retry(struct folio !! 1400 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
>> 1401 unsigned int flags)
1697 { 1402 {
1698 unsigned int flags = vmf->flags; !! 1403 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1699 <<
1700 if (fault_flag_allow_retry_first(flag <<
1701 /* 1404 /*
1702 * CAUTION! In this case, mma !! 1405 * CAUTION! In this case, mmap_sem is not released
1703 * released even though retur !! 1406 * even though return 0.
1704 */ 1407 */
1705 if (flags & FAULT_FLAG_RETRY_ 1408 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1706 return VM_FAULT_RETRY !! 1409 return 0;
1707 1410
1708 release_fault_lock(vmf); !! 1411 up_read(&mm->mmap_sem);
1709 if (flags & FAULT_FLAG_KILLAB 1412 if (flags & FAULT_FLAG_KILLABLE)
1710 folio_wait_locked_kil !! 1413 wait_on_page_locked_killable(page);
1711 else 1414 else
1712 folio_wait_locked(fol !! 1415 wait_on_page_locked(page);
1713 return VM_FAULT_RETRY; !! 1416 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 { 1417 } else {
1724 __folio_lock(folio); !! 1418 if (flags & FAULT_FLAG_KILLABLE) {
1725 } !! 1419 int ret;
1726 1420
1727 return 0; !! 1421 ret = __lock_page_killable(page);
>> 1422 if (ret) {
>> 1423 up_read(&mm->mmap_sem);
>> 1424 return 0;
>> 1425 }
>> 1426 } else
>> 1427 __lock_page(page);
>> 1428 return 1;
>> 1429 }
1728 } 1430 }
1729 1431
1730 /** 1432 /**
1731 * page_cache_next_miss() - Find the next gap 1433 * page_cache_next_miss() - Find the next gap in the page cache.
1732 * @mapping: Mapping. 1434 * @mapping: Mapping.
1733 * @index: Index. 1435 * @index: Index.
1734 * @max_scan: Maximum range to search. 1436 * @max_scan: Maximum range to search.
1735 * 1437 *
1736 * Search the range [index, min(index + max_s 1438 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1737 * gap with the lowest index. 1439 * gap with the lowest index.
1738 * 1440 *
1739 * This function may be called under the rcu_ 1441 * This function may be called under the rcu_read_lock. However, this will
1740 * not atomically search a snapshot of the ca 1442 * not atomically search a snapshot of the cache at a single point in time.
1741 * For example, if a gap is created at index 1443 * For example, if a gap is created at index 5, then subsequently a gap is
1742 * created at index 10, page_cache_next_miss 1444 * created at index 10, page_cache_next_miss covering both indices may
1743 * return 10 if called under the rcu_read_loc 1445 * return 10 if called under the rcu_read_lock.
1744 * 1446 *
1745 * Return: The index of the gap if found, oth 1447 * Return: The index of the gap if found, otherwise an index outside the
1746 * range specified (in which case 'return - i 1448 * range specified (in which case 'return - index >= max_scan' will be true).
1747 * In the rare case of index wrap-around, 0 w 1449 * In the rare case of index wrap-around, 0 will be returned.
1748 */ 1450 */
1749 pgoff_t page_cache_next_miss(struct address_s 1451 pgoff_t page_cache_next_miss(struct address_space *mapping,
1750 pgoff_t index, u 1452 pgoff_t index, unsigned long max_scan)
1751 { 1453 {
1752 XA_STATE(xas, &mapping->i_pages, inde 1454 XA_STATE(xas, &mapping->i_pages, index);
1753 1455
1754 while (max_scan--) { 1456 while (max_scan--) {
1755 void *entry = xas_next(&xas); 1457 void *entry = xas_next(&xas);
1756 if (!entry || xa_is_value(ent 1458 if (!entry || xa_is_value(entry))
1757 return xas.xa_index; !! 1459 break;
1758 if (xas.xa_index == 0) 1460 if (xas.xa_index == 0)
1759 return 0; !! 1461 break;
1760 } 1462 }
1761 1463
1762 return index + max_scan; !! 1464 return xas.xa_index;
1763 } 1465 }
1764 EXPORT_SYMBOL(page_cache_next_miss); 1466 EXPORT_SYMBOL(page_cache_next_miss);
1765 1467
1766 /** 1468 /**
1767 * page_cache_prev_miss() - Find the previous 1469 * page_cache_prev_miss() - Find the previous gap in the page cache.
1768 * @mapping: Mapping. 1470 * @mapping: Mapping.
1769 * @index: Index. 1471 * @index: Index.
1770 * @max_scan: Maximum range to search. 1472 * @max_scan: Maximum range to search.
1771 * 1473 *
1772 * Search the range [max(index - max_scan + 1 1474 * Search the range [max(index - max_scan + 1, 0), index] for the
1773 * gap with the highest index. 1475 * gap with the highest index.
1774 * 1476 *
1775 * This function may be called under the rcu_ 1477 * This function may be called under the rcu_read_lock. However, this will
1776 * not atomically search a snapshot of the ca 1478 * not atomically search a snapshot of the cache at a single point in time.
1777 * For example, if a gap is created at index 1479 * For example, if a gap is created at index 10, then subsequently a gap is
1778 * created at index 5, page_cache_prev_miss() 1480 * created at index 5, page_cache_prev_miss() covering both indices may
1779 * return 5 if called under the rcu_read_lock 1481 * return 5 if called under the rcu_read_lock.
1780 * 1482 *
1781 * Return: The index of the gap if found, oth 1483 * Return: The index of the gap if found, otherwise an index outside the
1782 * range specified (in which case 'index - re 1484 * range specified (in which case 'index - return >= max_scan' will be true).
1783 * In the rare case of wrap-around, ULONG_MAX 1485 * In the rare case of wrap-around, ULONG_MAX will be returned.
1784 */ 1486 */
1785 pgoff_t page_cache_prev_miss(struct address_s 1487 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1786 pgoff_t index, u 1488 pgoff_t index, unsigned long max_scan)
1787 { 1489 {
1788 XA_STATE(xas, &mapping->i_pages, inde 1490 XA_STATE(xas, &mapping->i_pages, index);
1789 1491
1790 while (max_scan--) { 1492 while (max_scan--) {
1791 void *entry = xas_prev(&xas); 1493 void *entry = xas_prev(&xas);
1792 if (!entry || xa_is_value(ent 1494 if (!entry || xa_is_value(entry))
1793 break; 1495 break;
1794 if (xas.xa_index == ULONG_MAX 1496 if (xas.xa_index == ULONG_MAX)
1795 break; 1497 break;
1796 } 1498 }
1797 1499
1798 return xas.xa_index; 1500 return xas.xa_index;
1799 } 1501 }
1800 EXPORT_SYMBOL(page_cache_prev_miss); 1502 EXPORT_SYMBOL(page_cache_prev_miss);
1801 1503
1802 /* !! 1504 /**
1803 * Lockless page cache protocol: !! 1505 * 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 1506 * @mapping: the address_space to search
1825 * @index: The page cache index. !! 1507 * @offset: the page cache index
>> 1508 *
>> 1509 * Looks up the page cache slot at @mapping & @offset. If there is a
>> 1510 * page cache page, it is returned with an increased refcount.
1826 * 1511 *
1827 * Looks up the page cache entry at @mapping !! 1512 * If the slot holds a shadow entry of a previously evicted page, or a
1828 * it is returned with an increased refcount. !! 1513 * swap entry from shmem/tmpfs, it is returned.
1829 * of a previously evicted folio, or a swap e <<
1830 * it is returned without further action. <<
1831 * 1514 *
1832 * Return: The folio, swap or shadow entry, % !! 1515 * Return: the found page or shadow entry, %NULL if nothing is found.
1833 */ 1516 */
1834 void *filemap_get_entry(struct address_space !! 1517 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1835 { 1518 {
1836 XA_STATE(xas, &mapping->i_pages, inde !! 1519 XA_STATE(xas, &mapping->i_pages, offset);
1837 struct folio *folio; !! 1520 struct page *head, *page;
1838 1521
1839 rcu_read_lock(); 1522 rcu_read_lock();
1840 repeat: 1523 repeat:
1841 xas_reset(&xas); 1524 xas_reset(&xas);
1842 folio = xas_load(&xas); !! 1525 page = xas_load(&xas);
1843 if (xas_retry(&xas, folio)) !! 1526 if (xas_retry(&xas, page))
1844 goto repeat; 1527 goto repeat;
1845 /* 1528 /*
1846 * A shadow entry of a recently evict 1529 * A shadow entry of a recently evicted page, or a swap entry from
1847 * shmem/tmpfs. Return it without at 1530 * shmem/tmpfs. Return it without attempting to raise page count.
1848 */ 1531 */
1849 if (!folio || xa_is_value(folio)) !! 1532 if (!page || xa_is_value(page))
1850 goto out; 1533 goto out;
1851 1534
1852 if (!folio_try_get(folio)) !! 1535 head = compound_head(page);
>> 1536 if (!page_cache_get_speculative(head))
1853 goto repeat; 1537 goto repeat;
1854 1538
1855 if (unlikely(folio != xas_reload(&xas !! 1539 /* The page was split under us? */
1856 folio_put(folio); !! 1540 if (compound_head(page) != head) {
>> 1541 put_page(head);
>> 1542 goto repeat;
>> 1543 }
>> 1544
>> 1545 /*
>> 1546 * Has the page moved?
>> 1547 * This is part of the lockless pagecache protocol. See
>> 1548 * include/linux/pagemap.h for details.
>> 1549 */
>> 1550 if (unlikely(page != xas_reload(&xas))) {
>> 1551 put_page(head);
1857 goto repeat; 1552 goto repeat;
1858 } 1553 }
1859 out: 1554 out:
1860 rcu_read_unlock(); 1555 rcu_read_unlock();
1861 1556
1862 return folio; !! 1557 return page;
1863 } 1558 }
>> 1559 EXPORT_SYMBOL(find_get_entry);
1864 1560
1865 /** 1561 /**
1866 * __filemap_get_folio - Find and get a refer !! 1562 * find_lock_entry - locate, pin and lock a page cache entry
1867 * @mapping: The address_space to search. !! 1563 * @mapping: the address_space to search
1868 * @index: The page index. !! 1564 * @offset: the page cache index
1869 * @fgp_flags: %FGP flags modify how the foli <<
1870 * @gfp: Memory allocation flags to use if %F <<
1871 * 1565 *
1872 * Looks up the page cache entry at @mapping !! 1566 * Looks up the page cache slot at @mapping & @offset. If there is a
>> 1567 * page cache page, it is returned locked and with an increased
>> 1568 * refcount.
1873 * 1569 *
1874 * If %FGP_LOCK or %FGP_CREAT are specified t !! 1570 * If the slot holds a shadow entry of a previously evicted page, or a
1875 * if the %GFP flags specified for %FGP_CREAT !! 1571 * swap entry from shmem/tmpfs, it is returned.
1876 * 1572 *
1877 * If this function returns a folio, it is re !! 1573 * find_lock_entry() may sleep.
1878 * 1574 *
1879 * Return: The found folio or an ERR_PTR() ot !! 1575 * Return: the found page or shadow entry, %NULL if nothing is found.
1880 */ 1576 */
1881 struct folio *__filemap_get_folio(struct addr !! 1577 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1882 fgf_t fgp_flags, gfp_t gfp) <<
1883 { 1578 {
1884 struct folio *folio; !! 1579 struct page *page;
1885 1580
1886 repeat: 1581 repeat:
1887 folio = filemap_get_entry(mapping, in !! 1582 page = find_get_entry(mapping, offset);
1888 if (xa_is_value(folio)) !! 1583 if (page && !xa_is_value(page)) {
1889 folio = NULL; !! 1584 lock_page(page);
1890 if (!folio) !! 1585 /* Has the page been truncated? */
>> 1586 if (unlikely(page_mapping(page) != mapping)) {
>> 1587 unlock_page(page);
>> 1588 put_page(page);
>> 1589 goto repeat;
>> 1590 }
>> 1591 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
>> 1592 }
>> 1593 return page;
>> 1594 }
>> 1595 EXPORT_SYMBOL(find_lock_entry);
>> 1596
>> 1597 /**
>> 1598 * pagecache_get_page - find and get a page reference
>> 1599 * @mapping: the address_space to search
>> 1600 * @offset: the page index
>> 1601 * @fgp_flags: PCG flags
>> 1602 * @gfp_mask: gfp mask to use for the page cache data page allocation
>> 1603 *
>> 1604 * Looks up the page cache slot at @mapping & @offset.
>> 1605 *
>> 1606 * PCG flags modify how the page is returned.
>> 1607 *
>> 1608 * @fgp_flags can be:
>> 1609 *
>> 1610 * - FGP_ACCESSED: the page will be marked accessed
>> 1611 * - FGP_LOCK: Page is return locked
>> 1612 * - FGP_CREAT: If page is not present then a new page is allocated using
>> 1613 * @gfp_mask and added to the page cache and the VM's LRU
>> 1614 * list. The page is returned locked and with an increased
>> 1615 * refcount.
>> 1616 * - FGP_FOR_MMAP: Similar to FGP_CREAT, only we want to allow the caller to do
>> 1617 * its own locking dance if the page is already in cache, or unlock the page
>> 1618 * before returning if we had to add the page to pagecache.
>> 1619 *
>> 1620 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
>> 1621 * if the GFP flags specified for FGP_CREAT are atomic.
>> 1622 *
>> 1623 * If there is a page cache page, it is returned with an increased refcount.
>> 1624 *
>> 1625 * Return: the found page or %NULL otherwise.
>> 1626 */
>> 1627 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
>> 1628 int fgp_flags, gfp_t gfp_mask)
>> 1629 {
>> 1630 struct page *page;
>> 1631
>> 1632 repeat:
>> 1633 page = find_get_entry(mapping, offset);
>> 1634 if (xa_is_value(page))
>> 1635 page = NULL;
>> 1636 if (!page)
1891 goto no_page; 1637 goto no_page;
1892 1638
1893 if (fgp_flags & FGP_LOCK) { 1639 if (fgp_flags & FGP_LOCK) {
1894 if (fgp_flags & FGP_NOWAIT) { 1640 if (fgp_flags & FGP_NOWAIT) {
1895 if (!folio_trylock(fo !! 1641 if (!trylock_page(page)) {
1896 folio_put(fol !! 1642 put_page(page);
1897 return ERR_PT !! 1643 return NULL;
1898 } 1644 }
1899 } else { 1645 } else {
1900 folio_lock(folio); !! 1646 lock_page(page);
1901 } 1647 }
1902 1648
1903 /* Has the page been truncate 1649 /* Has the page been truncated? */
1904 if (unlikely(folio->mapping ! !! 1650 if (unlikely(page->mapping != mapping)) {
1905 folio_unlock(folio); !! 1651 unlock_page(page);
1906 folio_put(folio); !! 1652 put_page(page);
1907 goto repeat; 1653 goto repeat;
1908 } 1654 }
1909 VM_BUG_ON_FOLIO(!folio_contai !! 1655 VM_BUG_ON_PAGE(page->index != offset, page);
1910 } 1656 }
1911 1657
1912 if (fgp_flags & FGP_ACCESSED) 1658 if (fgp_flags & FGP_ACCESSED)
1913 folio_mark_accessed(folio); !! 1659 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 1660
1920 if (fgp_flags & FGP_STABLE) <<
1921 folio_wait_stable(folio); <<
1922 no_page: 1661 no_page:
1923 if (!folio && (fgp_flags & FGP_CREAT) !! 1662 if (!page && (fgp_flags & FGP_CREAT)) {
1924 unsigned int min_order = mapp <<
1925 unsigned int order = max(min_ <<
1926 int err; 1663 int err;
1927 index = mapping_align_index(m !! 1664 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1928 !! 1665 gfp_mask |= __GFP_WRITE;
1929 if ((fgp_flags & FGP_WRITE) & <<
1930 gfp |= __GFP_WRITE; <<
1931 if (fgp_flags & FGP_NOFS) 1666 if (fgp_flags & FGP_NOFS)
1932 gfp &= ~__GFP_FS; !! 1667 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 1668
1940 if (order > mapping_max_folio !! 1669 page = __page_cache_alloc(gfp_mask);
1941 order = mapping_max_f !! 1670 if (!page)
1942 /* If we're not aligned, allo !! 1671 return NULL;
1943 if (index & ((1UL << order) - <<
1944 order = __ffs(index); <<
1945 <<
1946 do { <<
1947 gfp_t alloc_gfp = gfp <<
1948 1672
1949 err = -ENOMEM; !! 1673 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1950 if (order > min_order !! 1674 fgp_flags |= FGP_LOCK;
1951 alloc_gfp |= <<
1952 folio = filemap_alloc <<
1953 if (!folio) <<
1954 continue; <<
1955 1675
1956 /* Init accessed so a !! 1676 /* Init accessed so avoid atomic mark_page_accessed later */
1957 if (fgp_flags & FGP_A !! 1677 if (fgp_flags & FGP_ACCESSED)
1958 __folio_set_r !! 1678 __SetPageReferenced(page);
1959 1679
1960 err = filemap_add_fol !! 1680 err = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
1961 if (!err) !! 1681 if (unlikely(err)) {
1962 break; !! 1682 put_page(page);
1963 folio_put(folio); !! 1683 page = NULL;
1964 folio = NULL; !! 1684 if (err == -EEXIST)
1965 } while (order-- > min_order) !! 1685 goto repeat;
>> 1686 }
1966 1687
1967 if (err == -EEXIST) <<
1968 goto repeat; <<
1969 if (err) <<
1970 return ERR_PTR(err); <<
1971 /* 1688 /*
1972 * filemap_add_folio locks th !! 1689 * add_to_page_cache_lru locks the page, and for mmap we expect
1973 * we expect an unlocked page !! 1690 * an unlocked page.
1974 */ 1691 */
1975 if (folio && (fgp_flags & FGP !! 1692 if (page && (fgp_flags & FGP_FOR_MMAP))
1976 folio_unlock(folio); !! 1693 unlock_page(page);
1977 } <<
1978 <<
1979 if (!folio) <<
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 } 1694 }
2013 1695
2014 return folio; !! 1696 return page;
2015 reset: <<
2016 xas_reset(xas); <<
2017 goto retry; <<
2018 } 1697 }
>> 1698 EXPORT_SYMBOL(pagecache_get_page);
2019 1699
2020 /** 1700 /**
2021 * find_get_entries - gang pagecache lookup 1701 * find_get_entries - gang pagecache lookup
2022 * @mapping: The address_space to search 1702 * @mapping: The address_space to search
2023 * @start: The starting page cache index 1703 * @start: The starting page cache index
2024 * @end: The final page index (inclusi !! 1704 * @nr_entries: The maximum number of entries
2025 * @fbatch: Where the resulting entries a !! 1705 * @entries: Where the resulting entries are placed
2026 * @indices: The cache indices correspondi 1706 * @indices: The cache indices corresponding to the entries in @entries
2027 * 1707 *
2028 * find_get_entries() will search for and ret !! 1708 * find_get_entries() will search for and return a group of up to
2029 * the mapping. The entries are placed in @f !! 1709 * @nr_entries entries in the mapping. The entries are placed at
2030 * takes a reference on any actual folios it !! 1710 * @entries. find_get_entries() takes a reference against any actual
2031 * !! 1711 * pages it returns.
2032 * The entries have ascending indexes. The i !! 1712 *
2033 * due to not-present entries or large folios !! 1713 * The search returns a group of mapping-contiguous page cache entries
>> 1714 * with ascending indexes. There may be holes in the indices due to
>> 1715 * not-present pages.
2034 * 1716 *
2035 * Any shadow entries of evicted folios, or s !! 1717 * Any shadow entries of evicted pages, or swap entries from
2036 * shmem/tmpfs, are included in the returned 1718 * shmem/tmpfs, are included in the returned array.
2037 * 1719 *
2038 * Return: The number of entries which were f !! 1720 * Return: the number of pages and shadow entries which were found.
2039 */ 1721 */
2040 unsigned find_get_entries(struct address_spac !! 1722 unsigned find_get_entries(struct address_space *mapping,
2041 pgoff_t end, struct folio_bat !! 1723 pgoff_t start, unsigned int nr_entries,
>> 1724 struct page **entries, pgoff_t *indices)
2042 { 1725 {
2043 XA_STATE(xas, &mapping->i_pages, *sta !! 1726 XA_STATE(xas, &mapping->i_pages, start);
2044 struct folio *folio; !! 1727 struct page *page;
>> 1728 unsigned int ret = 0;
>> 1729
>> 1730 if (!nr_entries)
>> 1731 return 0;
2045 1732
2046 rcu_read_lock(); 1733 rcu_read_lock();
2047 while ((folio = find_get_entry(&xas, !! 1734 xas_for_each(&xas, page, ULONG_MAX) {
2048 indices[fbatch->nr] = xas.xa_ !! 1735 struct page *head;
2049 if (!folio_batch_add(fbatch, !! 1736 if (xas_retry(&xas, page))
2050 break; !! 1737 continue;
2051 } !! 1738 /*
>> 1739 * A shadow entry of a recently evicted page, a swap
>> 1740 * entry from shmem/tmpfs or a DAX entry. Return it
>> 1741 * without attempting to raise page count.
>> 1742 */
>> 1743 if (xa_is_value(page))
>> 1744 goto export;
2052 1745
2053 if (folio_batch_count(fbatch)) { !! 1746 head = compound_head(page);
2054 unsigned long nr; !! 1747 if (!page_cache_get_speculative(head))
2055 int idx = folio_batch_count(f !! 1748 goto retry;
2056 !! 1749
2057 folio = fbatch->folios[idx]; !! 1750 /* The page was split under us? */
2058 if (!xa_is_value(folio)) !! 1751 if (compound_head(page) != head)
2059 nr = folio_nr_pages(f !! 1752 goto put_page;
2060 else !! 1753
2061 nr = 1 << xa_get_orde !! 1754 /* Has the page moved? */
2062 *start = round_down(indices[i !! 1755 if (unlikely(page != xas_reload(&xas)))
>> 1756 goto put_page;
>> 1757
>> 1758 export:
>> 1759 indices[ret] = xas.xa_index;
>> 1760 entries[ret] = page;
>> 1761 if (++ret == nr_entries)
>> 1762 break;
>> 1763 continue;
>> 1764 put_page:
>> 1765 put_page(head);
>> 1766 retry:
>> 1767 xas_reset(&xas);
2063 } 1768 }
2064 rcu_read_unlock(); 1769 rcu_read_unlock();
2065 !! 1770 return ret;
2066 return folio_batch_count(fbatch); <<
2067 } 1771 }
2068 1772
2069 /** 1773 /**
2070 * find_lock_entries - Find a batch of pageca !! 1774 * find_get_pages_range - gang pagecache lookup
2071 * @mapping: The address_space to search. !! 1775 * @mapping: The address_space to search
2072 * @start: The starting page cache index !! 1776 * @start: The starting page index
2073 * @end: The final page index (inclusi !! 1777 * @end: The final page index (inclusive)
2074 * @fbatch: Where the resulting entries a !! 1778 * @nr_pages: The maximum number of pages
2075 * @indices: The cache indices of the entr !! 1779 * @pages: Where the resulting pages are placed
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 * 1780 *
2087 * Return: The number of entries which were f !! 1781 * find_get_pages_range() will search for and return a group of up to @nr_pages
2088 */ !! 1782 * pages in the mapping starting at index @start and up to index @end
2089 unsigned find_lock_entries(struct address_spa !! 1783 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
2090 pgoff_t end, struct folio_bat !! 1784 * a reference against the returned pages.
>> 1785 *
>> 1786 * The search returns a group of mapping-contiguous pages with ascending
>> 1787 * indexes. There may be holes in the indices due to not-present pages.
>> 1788 * We also update @start to index the next page for the traversal.
>> 1789 *
>> 1790 * Return: the number of pages which were found. If this number is
>> 1791 * smaller than @nr_pages, the end of specified range has been
>> 1792 * reached.
>> 1793 */
>> 1794 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
>> 1795 pgoff_t end, unsigned int nr_pages,
>> 1796 struct page **pages)
2091 { 1797 {
2092 XA_STATE(xas, &mapping->i_pages, *sta 1798 XA_STATE(xas, &mapping->i_pages, *start);
2093 struct folio *folio; !! 1799 struct page *page;
>> 1800 unsigned ret = 0;
>> 1801
>> 1802 if (unlikely(!nr_pages))
>> 1803 return 0;
2094 1804
2095 rcu_read_lock(); 1805 rcu_read_lock();
2096 while ((folio = find_get_entry(&xas, !! 1806 xas_for_each(&xas, page, end) {
2097 unsigned long base; !! 1807 struct page *head;
2098 unsigned long nr; !! 1808 if (xas_retry(&xas, page))
2099 !! 1809 continue;
2100 if (!xa_is_value(folio)) { !! 1810 /* Skip over shadow, swap and DAX entries */
2101 nr = folio_nr_pages(f !! 1811 if (xa_is_value(page))
2102 base = folio->index; !! 1812 continue;
2103 /* Omit large folio w <<
2104 if (base < *start) <<
2105 goto put; <<
2106 /* Omit large folio w <<
2107 if (base + nr - 1 > e <<
2108 goto put; <<
2109 if (!folio_trylock(fo <<
2110 goto put; <<
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 1813
2127 /* Update start now so that l !! 1814 head = compound_head(page);
2128 *start = base + nr; !! 1815 if (!page_cache_get_speculative(head))
2129 indices[fbatch->nr] = xas.xa_ !! 1816 goto retry;
2130 if (!folio_batch_add(fbatch, !! 1817
2131 break; !! 1818 /* The page was split under us? */
>> 1819 if (compound_head(page) != head)
>> 1820 goto put_page;
>> 1821
>> 1822 /* Has the page moved? */
>> 1823 if (unlikely(page != xas_reload(&xas)))
>> 1824 goto put_page;
>> 1825
>> 1826 pages[ret] = page;
>> 1827 if (++ret == nr_pages) {
>> 1828 *start = xas.xa_index + 1;
>> 1829 goto out;
>> 1830 }
2132 continue; 1831 continue;
2133 unlock: !! 1832 put_page:
2134 folio_unlock(folio); !! 1833 put_page(head);
2135 put: !! 1834 retry:
2136 folio_put(folio); !! 1835 xas_reset(&xas);
2137 } 1836 }
>> 1837
>> 1838 /*
>> 1839 * We come here when there is no page beyond @end. We take care to not
>> 1840 * overflow the index @start as it confuses some of the callers. This
>> 1841 * breaks the iteration when there is a page at index -1 but that is
>> 1842 * already broken anyway.
>> 1843 */
>> 1844 if (end == (pgoff_t)-1)
>> 1845 *start = (pgoff_t)-1;
>> 1846 else
>> 1847 *start = end + 1;
>> 1848 out:
2138 rcu_read_unlock(); 1849 rcu_read_unlock();
2139 1850
2140 return folio_batch_count(fbatch); !! 1851 return ret;
2141 } 1852 }
2142 1853
2143 /** 1854 /**
2144 * filemap_get_folios - Get a batch of folios !! 1855 * find_get_pages_contig - gang contiguous pagecache lookup
2145 * @mapping: The address_space to search 1856 * @mapping: The address_space to search
2146 * @start: The starting page index !! 1857 * @index: The starting page index
2147 * @end: The final page index (inclusi !! 1858 * @nr_pages: The maximum number of pages
2148 * @fbatch: The batch to fill. !! 1859 * @pages: Where the resulting pages are placed
2149 * 1860 *
2150 * Search for and return a batch of folios in !! 1861 * find_get_pages_contig() works exactly like find_get_pages(), except
2151 * index @start and up to index @end (inclusi !! 1862 * that the returned number of pages are guaranteed to be contiguous.
2152 * in @fbatch with an elevated reference coun <<
2153 * 1863 *
2154 * Return: The number of folios which were fo !! 1864 * Return: the number of pages which were found.
2155 * We also update @start to index the next fo <<
2156 */ 1865 */
2157 unsigned filemap_get_folios(struct address_sp !! 1866 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
2158 pgoff_t end, struct folio_bat !! 1867 unsigned int nr_pages, struct page **pages)
2159 { 1868 {
2160 return filemap_get_folios_tag(mapping !! 1869 XA_STATE(xas, &mapping->i_pages, index);
2161 } !! 1870 struct page *page;
2162 EXPORT_SYMBOL(filemap_get_folios); !! 1871 unsigned int ret = 0;
2163 <<
2164 /** <<
2165 * filemap_get_folios_contig - Get a batch of <<
2166 * @mapping: The address_space to search <<
2167 * @start: The starting page index <<
2168 * @end: The final page index (inclusi <<
2169 * @fbatch: The batch to fill <<
2170 * <<
2171 * filemap_get_folios_contig() works exactly <<
2172 * except the returned folios are guaranteed <<
2173 * not return all contiguous folios if the ba <<
2174 * <<
2175 * Return: The number of folios found. <<
2176 * Also update @start to be positioned for tr <<
2177 */ <<
2178 1872
2179 unsigned filemap_get_folios_contig(struct add !! 1873 if (unlikely(!nr_pages))
2180 pgoff_t *start, pgoff_t end, !! 1874 return 0;
2181 { <<
2182 XA_STATE(xas, &mapping->i_pages, *sta <<
2183 unsigned long nr; <<
2184 struct folio *folio; <<
2185 1875
2186 rcu_read_lock(); 1876 rcu_read_lock();
2187 !! 1877 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
2188 for (folio = xas_load(&xas); folio && !! 1878 struct page *head;
2189 folio = xas_next(&xas !! 1879 if (xas_retry(&xas, page))
2190 if (xas_retry(&xas, folio)) <<
2191 continue; 1880 continue;
2192 /* 1881 /*
2193 * If the entry has been swap 1882 * If the entry has been swapped out, we can stop looking.
2194 * No current caller is looki 1883 * No current caller is looking for DAX entries.
2195 */ 1884 */
2196 if (xa_is_value(folio)) !! 1885 if (xa_is_value(page))
2197 goto update_start; !! 1886 break;
2198 <<
2199 /* If we landed in the middle <<
2200 if (xa_is_sibling(folio)) <<
2201 goto update_start; <<
2202 1887
2203 if (!folio_try_get(folio)) !! 1888 head = compound_head(page);
>> 1889 if (!page_cache_get_speculative(head))
2204 goto retry; 1890 goto retry;
2205 1891
2206 if (unlikely(folio != xas_rel !! 1892 /* The page was split under us? */
2207 goto put_folio; !! 1893 if (compound_head(page) != head)
>> 1894 goto put_page;
>> 1895
>> 1896 /* Has the page moved? */
>> 1897 if (unlikely(page != xas_reload(&xas)))
>> 1898 goto put_page;
2208 1899
2209 if (!folio_batch_add(fbatch, !! 1900 pages[ret] = page;
2210 nr = folio_nr_pages(f !! 1901 if (++ret == nr_pages)
2211 *start = folio->index !! 1902 break;
2212 goto out; <<
2213 } <<
2214 continue; 1903 continue;
2215 put_folio: !! 1904 put_page:
2216 folio_put(folio); !! 1905 put_page(head);
2217 <<
2218 retry: 1906 retry:
2219 xas_reset(&xas); 1907 xas_reset(&xas);
2220 } 1908 }
2221 <<
2222 update_start: <<
2223 nr = folio_batch_count(fbatch); <<
2224 <<
2225 if (nr) { <<
2226 folio = fbatch->folios[nr - 1 <<
2227 *start = folio_next_index(fol <<
2228 } <<
2229 out: <<
2230 rcu_read_unlock(); 1909 rcu_read_unlock();
2231 return folio_batch_count(fbatch); !! 1910 return ret;
2232 } 1911 }
2233 EXPORT_SYMBOL(filemap_get_folios_contig); !! 1912 EXPORT_SYMBOL(find_get_pages_contig);
2234 1913
2235 /** 1914 /**
2236 * filemap_get_folios_tag - Get a batch of fo !! 1915 * find_get_pages_range_tag - find and return pages in given range matching @tag
2237 * @mapping: The address_space to search !! 1916 * @mapping: the address_space to search
2238 * @start: The starting page index !! 1917 * @index: the starting page index
2239 * @end: The final page index (inclusi !! 1918 * @end: The final page index (inclusive)
2240 * @tag: The tag index !! 1919 * @tag: the tag index
2241 * @fbatch: The batch to fill !! 1920 * @nr_pages: the maximum number of pages
2242 * !! 1921 * @pages: where the resulting pages are placed
2243 * The first folio may start before @start; i !! 1922 *
2244 * @start. The final folio may extend beyond !! 1923 * Like find_get_pages, except we only return pages which are tagged with
2245 * contain @end. The folios have ascending i !! 1924 * @tag. We update @index to index the next page for the traversal.
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 * 1925 *
2251 * Return: The number of folios found. !! 1926 * Return: the number of pages which were found.
2252 * Also update @start to index the next folio <<
2253 */ 1927 */
2254 unsigned filemap_get_folios_tag(struct addres !! 1928 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
2255 pgoff_t end, xa_mark_ !! 1929 pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
>> 1930 struct page **pages)
2256 { 1931 {
2257 XA_STATE(xas, &mapping->i_pages, *sta !! 1932 XA_STATE(xas, &mapping->i_pages, *index);
2258 struct folio *folio; !! 1933 struct page *page;
>> 1934 unsigned ret = 0;
>> 1935
>> 1936 if (unlikely(!nr_pages))
>> 1937 return 0;
2259 1938
2260 rcu_read_lock(); 1939 rcu_read_lock();
2261 while ((folio = find_get_entry(&xas, !! 1940 xas_for_each_marked(&xas, page, end, tag) {
>> 1941 struct page *head;
>> 1942 if (xas_retry(&xas, page))
>> 1943 continue;
2262 /* 1944 /*
2263 * Shadow entries should neve 1945 * Shadow entries should never be tagged, but this iteration
2264 * is lockless so there is a 1946 * is lockless so there is a window for page reclaim to evict
2265 * a page we saw tagged. Skip !! 1947 * a page we saw tagged. Skip over it.
2266 */ 1948 */
2267 if (xa_is_value(folio)) !! 1949 if (xa_is_value(page))
2268 continue; 1950 continue;
2269 if (!folio_batch_add(fbatch, !! 1951
2270 unsigned long nr = fo !! 1952 head = compound_head(page);
2271 *start = folio->index !! 1953 if (!page_cache_get_speculative(head))
>> 1954 goto retry;
>> 1955
>> 1956 /* The page was split under us? */
>> 1957 if (compound_head(page) != head)
>> 1958 goto put_page;
>> 1959
>> 1960 /* Has the page moved? */
>> 1961 if (unlikely(page != xas_reload(&xas)))
>> 1962 goto put_page;
>> 1963
>> 1964 pages[ret] = page;
>> 1965 if (++ret == nr_pages) {
>> 1966 *index = xas.xa_index + 1;
2272 goto out; 1967 goto out;
2273 } 1968 }
>> 1969 continue;
>> 1970 put_page:
>> 1971 put_page(head);
>> 1972 retry:
>> 1973 xas_reset(&xas);
2274 } 1974 }
>> 1975
2275 /* 1976 /*
2276 * We come here when there is no page !! 1977 * We come here when we got to @end. We take care to not overflow the
2277 * overflow the index @start as it co !! 1978 * index @index as it confuses some of the callers. This breaks the
2278 * breaks the iteration when there is !! 1979 * iteration when there is a page at index -1 but that is already
2279 * already broke anyway. !! 1980 * broken anyway.
2280 */ 1981 */
2281 if (end == (pgoff_t)-1) 1982 if (end == (pgoff_t)-1)
2282 *start = (pgoff_t)-1; !! 1983 *index = (pgoff_t)-1;
2283 else 1984 else
2284 *start = end + 1; !! 1985 *index = end + 1;
2285 out: 1986 out:
2286 rcu_read_unlock(); 1987 rcu_read_unlock();
2287 1988
2288 return folio_batch_count(fbatch); !! 1989 return ret;
2289 } 1990 }
2290 EXPORT_SYMBOL(filemap_get_folios_tag); !! 1991 EXPORT_SYMBOL(find_get_pages_range_tag);
2291 1992
2292 /* 1993 /*
2293 * CD/DVDs are error prone. When a medium err 1994 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2294 * a _large_ part of the i/o request. Imagine 1995 * a _large_ part of the i/o request. Imagine the worst scenario:
2295 * 1996 *
2296 * ---R_________________________________ 1997 * ---R__________________________________________B__________
2297 * ^ reading here 1998 * ^ reading here ^ bad block(assume 4k)
2298 * 1999 *
2299 * read(R) => miss => readahead(R...B) => med 2000 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2300 * => failing the whole request => read(R) => 2001 * => failing the whole request => read(R) => read(R+1) =>
2301 * readahead(R+1...B+1) => bang => read(R+2) 2002 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2302 * readahead(R+3...B+2) => bang => read(R+3) 2003 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2303 * readahead(R+4...B+3) => bang => read(R+4) 2004 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2304 * 2005 *
2305 * It is going insane. Fix it by quickly scal 2006 * It is going insane. Fix it by quickly scaling down the readahead size.
2306 */ 2007 */
2307 static void shrink_readahead_size_eio(struct !! 2008 static void shrink_readahead_size_eio(struct file *filp,
>> 2009 struct file_ra_state *ra)
2308 { 2010 {
2309 ra->ra_pages /= 4; 2011 ra->ra_pages /= 4;
2310 } 2012 }
2311 2013
2312 /* !! 2014 /**
2313 * filemap_get_read_batch - Get a batch of fo !! 2015 * generic_file_buffered_read - generic file read routine
>> 2016 * @iocb: the iocb to read
>> 2017 * @iter: data destination
>> 2018 * @written: already copied
>> 2019 *
>> 2020 * This is a generic file read routine, and uses the
>> 2021 * mapping->a_ops->readpage() function for the actual low-level stuff.
>> 2022 *
>> 2023 * This is really ugly. But the goto's actually try to clarify some
>> 2024 * of the logic when it comes to error handling etc.
2314 * 2025 *
2315 * Get a batch of folios which represent a co !! 2026 * Return:
2316 * the file. No exceptional entries will be !! 2027 * * total number of bytes copied, including those the were already @written
2317 * the middle of a folio, the entire folio wi !! 2028 * * negative error code if nothing was copied
2318 * folio in the batch may have the readahead <<
2319 * clear so that the caller can take the appr <<
2320 */ 2029 */
2321 static void filemap_get_read_batch(struct add !! 2030 static ssize_t generic_file_buffered_read(struct kiocb *iocb,
2322 pgoff_t index, pgoff_t max, s !! 2031 struct iov_iter *iter, ssize_t written)
2323 { <<
2324 XA_STATE(xas, &mapping->i_pages, inde <<
2325 struct folio *folio; <<
2326 <<
2327 rcu_read_lock(); <<
2328 for (folio = xas_load(&xas); folio; f <<
2329 if (xas_retry(&xas, folio)) <<
2330 continue; <<
2331 if (xas.xa_index > max || xa_ <<
2332 break; <<
2333 if (xa_is_sibling(folio)) <<
2334 break; <<
2335 if (!folio_try_get(folio)) <<
2336 goto retry; <<
2337 <<
2338 if (unlikely(folio != xas_rel <<
2339 goto put_folio; <<
2340 <<
2341 if (!folio_batch_add(fbatch, <<
2342 break; <<
2343 if (!folio_test_uptodate(foli <<
2344 break; <<
2345 if (folio_test_readahead(foli <<
2346 break; <<
2347 xas_advance(&xas, folio_next_ <<
2348 continue; <<
2349 put_folio: <<
2350 folio_put(folio); <<
2351 retry: <<
2352 xas_reset(&xas); <<
2353 } <<
2354 rcu_read_unlock(); <<
2355 } <<
2356 <<
2357 static int filemap_read_folio(struct file *fi <<
2358 struct folio *folio) <<
2359 { <<
2360 bool workingset = folio_test_workings <<
2361 unsigned long pflags; <<
2362 int error; <<
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 <<
2373 error = folio_wait_locked_killable(fo <<
2374 if (error) <<
2375 return error; <<
2376 if (folio_test_uptodate(folio)) <<
2377 return 0; <<
2378 if (file) <<
2379 shrink_readahead_size_eio(&fi <<
2380 return -EIO; <<
2381 } <<
2382 <<
2383 static bool filemap_range_uptodate(struct add <<
2384 loff_t pos, size_t count, str <<
2385 bool need_uptodate) <<
2386 { <<
2387 if (folio_test_uptodate(folio)) <<
2388 return true; <<
2389 /* pipes can't handle partially uptod <<
2390 if (need_uptodate) <<
2391 return false; <<
2392 if (!mapping->a_ops->is_partially_upt <<
2393 return false; <<
2394 if (mapping->host->i_blkbits >= folio <<
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 <<
2413 if (iocb->ki_flags & IOCB_NOWAIT) { <<
2414 if (!filemap_invalidate_trylo <<
2415 return -EAGAIN; <<
2416 } else { <<
2417 filemap_invalidate_lock_share <<
2418 } <<
2419 <<
2420 if (!folio_trylock(folio)) { <<
2421 error = -EAGAIN; <<
2422 if (iocb->ki_flags & (IOCB_NO <<
2423 goto unlock_mapping; <<
2424 if (!(iocb->ki_flags & IOCB_W <<
2425 filemap_invalidate_un <<
2426 /* <<
2427 * This is where we u <<
2428 * previously submitt <<
2429 */ <<
2430 folio_put_wait_locked <<
2431 return AOP_TRUNCATED_ <<
2432 } <<
2433 error = __folio_lock_async(fo <<
2434 if (error) <<
2435 goto unlock_mapping; <<
2436 } <<
2437 <<
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 /* <<
2477 * Protect against truncate / hole pu <<
2478 * here assures we cannot instantiate <<
2479 * pagecache folios after evicting pa <<
2480 * and before actually freeing blocks <<
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 */ <<
2489 filemap_invalidate_lock_shared(mappin <<
2490 index = (pos >> (PAGE_SHIFT + min_ord <<
2491 error = filemap_add_folio(mapping, fo <<
2492 mapping_gfp_constrain <<
2493 if (error == -EEXIST) <<
2494 error = AOP_TRUNCATED_PAGE; <<
2495 if (error) <<
2496 goto error; <<
2497 <<
2498 error = filemap_read_folio(file, mapp <<
2499 if (error) <<
2500 goto error; <<
2501 <<
2502 filemap_invalidate_unlock_shared(mapp <<
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 } <<
2510 <<
2511 static int filemap_readahead(struct kiocb *io <<
2512 struct address_space *mapping <<
2513 pgoff_t last_index) <<
2514 { <<
2515 DEFINE_READAHEAD(ractl, file, &file-> <<
2516 <<
2517 if (iocb->ki_flags & IOCB_NOIO) <<
2518 return -EAGAIN; <<
2519 page_cache_async_ra(&ractl, folio, la <<
2520 return 0; <<
2521 } <<
2522 <<
2523 static int filemap_get_pages(struct kiocb *io <<
2524 struct folio_batch *fbatch, b <<
2525 { 2032 {
2526 struct file *filp = iocb->ki_filp; 2033 struct file *filp = iocb->ki_filp;
2527 struct address_space *mapping = filp- 2034 struct address_space *mapping = filp->f_mapping;
>> 2035 struct inode *inode = mapping->host;
2528 struct file_ra_state *ra = &filp->f_r 2036 struct file_ra_state *ra = &filp->f_ra;
2529 pgoff_t index = iocb->ki_pos >> PAGE_ !! 2037 loff_t *ppos = &iocb->ki_pos;
>> 2038 pgoff_t index;
2530 pgoff_t last_index; 2039 pgoff_t last_index;
2531 struct folio *folio; !! 2040 pgoff_t prev_index;
2532 unsigned int flags; !! 2041 unsigned long offset; /* offset into pagecache page */
2533 int err = 0; !! 2042 unsigned int prev_offset;
2534 !! 2043 int error = 0;
2535 /* "last_index" is the index of the p <<
2536 last_index = DIV_ROUND_UP(iocb->ki_po <<
2537 retry: <<
2538 if (fatal_signal_pending(current)) <<
2539 return -EINTR; <<
2540 <<
2541 filemap_get_read_batch(mapping, index <<
2542 if (!folio_batch_count(fbatch)) { <<
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; <<
2559 return err; <<
2560 } <<
2561 <<
2562 folio = fbatch->folios[folio_batch_co <<
2563 if (folio_test_readahead(folio)) { <<
2564 err = filemap_readahead(iocb, <<
2565 if (err) <<
2566 goto err; <<
2567 } <<
2568 if (!folio_test_uptodate(folio)) { <<
2569 if ((iocb->ki_flags & IOCB_WA <<
2570 folio_batch_count(fbatch) <<
2571 iocb->ki_flags |= IOC <<
2572 err = filemap_update_page(ioc <<
2573 nee <<
2574 if (err) <<
2575 goto err; <<
2576 } <<
2577 2044
2578 trace_mm_filemap_get_pages(mapping, i !! 2045 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
2579 return 0; <<
2580 err: <<
2581 if (err < 0) <<
2582 folio_put(folio); <<
2583 if (likely(--fbatch->nr)) <<
2584 return 0; 2046 return 0;
2585 if (err == AOP_TRUNCATED_PAGE) !! 2047 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2586 goto retry; <<
2587 return err; <<
2588 } <<
2589 <<
2590 static inline bool pos_same_folio(loff_t pos1 <<
2591 { <<
2592 unsigned int shift = folio_shift(foli <<
2593 2048
2594 return (pos1 >> shift == pos2 >> shif !! 2049 index = *ppos >> PAGE_SHIFT;
2595 } !! 2050 prev_index = ra->prev_pos >> PAGE_SHIFT;
>> 2051 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
>> 2052 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
>> 2053 offset = *ppos & ~PAGE_MASK;
2596 2054
2597 /** !! 2055 for (;;) {
2598 * filemap_read - Read data from the page cac !! 2056 struct page *page;
2599 * @iocb: The iocb to read. !! 2057 pgoff_t end_index;
2600 * @iter: Destination for the data. !! 2058 loff_t isize;
2601 * @already_read: Number of bytes already rea !! 2059 unsigned long nr, ret;
2602 * <<
2603 * Copies data from the page cache. If the d <<
2604 * uses the readahead and read_folio address_ <<
2605 * <<
2606 * Return: Total number of bytes copied, incl <<
2607 * the caller. If an error happens before an <<
2608 * a negative error number. <<
2609 */ <<
2610 ssize_t filemap_read(struct kiocb *iocb, stru <<
2611 ssize_t already_read) <<
2612 { <<
2613 struct file *filp = iocb->ki_filp; <<
2614 struct file_ra_state *ra = &filp->f_r <<
2615 struct address_space *mapping = filp- <<
2616 struct inode *inode = mapping->host; <<
2617 struct folio_batch fbatch; <<
2618 int i, error = 0; <<
2619 bool writably_mapped; <<
2620 loff_t isize, end_offset; <<
2621 loff_t last_pos = ra->prev_pos; <<
2622 <<
2623 if (unlikely(iocb->ki_pos >= inode->i <<
2624 return 0; <<
2625 if (unlikely(!iov_iter_count(iter))) <<
2626 return 0; <<
2627 <<
2628 iov_iter_truncate(iter, inode->i_sb-> <<
2629 folio_batch_init(&fbatch); <<
2630 2060
2631 do { <<
2632 cond_resched(); 2061 cond_resched();
>> 2062 find_page:
>> 2063 if (fatal_signal_pending(current)) {
>> 2064 error = -EINTR;
>> 2065 goto out;
>> 2066 }
2633 2067
2634 /* !! 2068 page = find_get_page(mapping, index);
2635 * If we've already successfu !! 2069 if (!page) {
2636 * can no longer safely retur !! 2070 if (iocb->ki_flags & IOCB_NOWAIT)
2637 * an async read NOWAIT at th !! 2071 goto would_block;
2638 */ !! 2072 page_cache_sync_readahead(mapping,
2639 if ((iocb->ki_flags & IOCB_WA !! 2073 ra, filp,
2640 iocb->ki_flags |= IOC !! 2074 index, last_index - index);
2641 !! 2075 page = find_get_page(mapping, index);
2642 if (unlikely(iocb->ki_pos >= !! 2076 if (unlikely(page == NULL))
2643 break; !! 2077 goto no_cached_page;
2644 !! 2078 }
2645 error = filemap_get_pages(ioc !! 2079 if (PageReadahead(page)) {
2646 if (error < 0) !! 2080 page_cache_async_readahead(mapping,
2647 break; !! 2081 ra, filp, page,
>> 2082 index, last_index - index);
>> 2083 }
>> 2084 if (!PageUptodate(page)) {
>> 2085 if (iocb->ki_flags & IOCB_NOWAIT) {
>> 2086 put_page(page);
>> 2087 goto would_block;
>> 2088 }
2648 2089
>> 2090 /*
>> 2091 * See comment in do_read_cache_page on why
>> 2092 * wait_on_page_locked is used to avoid unnecessarily
>> 2093 * serialisations and why it's safe.
>> 2094 */
>> 2095 error = wait_on_page_locked_killable(page);
>> 2096 if (unlikely(error))
>> 2097 goto readpage_error;
>> 2098 if (PageUptodate(page))
>> 2099 goto page_ok;
>> 2100
>> 2101 if (inode->i_blkbits == PAGE_SHIFT ||
>> 2102 !mapping->a_ops->is_partially_uptodate)
>> 2103 goto page_not_up_to_date;
>> 2104 /* pipes can't handle partially uptodate pages */
>> 2105 if (unlikely(iov_iter_is_pipe(iter)))
>> 2106 goto page_not_up_to_date;
>> 2107 if (!trylock_page(page))
>> 2108 goto page_not_up_to_date;
>> 2109 /* Did it get truncated before we got the lock? */
>> 2110 if (!page->mapping)
>> 2111 goto page_not_up_to_date_locked;
>> 2112 if (!mapping->a_ops->is_partially_uptodate(page,
>> 2113 offset, iter->count))
>> 2114 goto page_not_up_to_date_locked;
>> 2115 unlock_page(page);
>> 2116 }
>> 2117 page_ok:
2649 /* 2118 /*
2650 * i_size must be checked aft !! 2119 * i_size must be checked after we know the page is Uptodate.
2651 * 2120 *
2652 * Checking i_size after the 2121 * Checking i_size after the check allows us to calculate
2653 * the correct value for "nr" 2122 * the correct value for "nr", which means the zero-filled
2654 * part of the page is not co 2123 * part of the page is not copied back to userspace (unless
2655 * another truncate extends t 2124 * another truncate extends the file - this is desired though).
2656 */ 2125 */
>> 2126
2657 isize = i_size_read(inode); 2127 isize = i_size_read(inode);
2658 if (unlikely(iocb->ki_pos >= !! 2128 end_index = (isize - 1) >> PAGE_SHIFT;
2659 goto put_folios; !! 2129 if (unlikely(!isize || index > end_index)) {
2660 end_offset = min_t(loff_t, is !! 2130 put_page(page);
>> 2131 goto out;
>> 2132 }
>> 2133
>> 2134 /* nr is the maximum number of bytes to copy from this page */
>> 2135 nr = PAGE_SIZE;
>> 2136 if (index == end_index) {
>> 2137 nr = ((isize - 1) & ~PAGE_MASK) + 1;
>> 2138 if (nr <= offset) {
>> 2139 put_page(page);
>> 2140 goto out;
>> 2141 }
>> 2142 }
>> 2143 nr = nr - offset;
>> 2144
>> 2145 /* If users can be writing to this page using arbitrary
>> 2146 * virtual addresses, take care about potential aliasing
>> 2147 * before reading the page on the kernel side.
>> 2148 */
>> 2149 if (mapping_writably_mapped(mapping))
>> 2150 flush_dcache_page(page);
2661 2151
2662 /* 2152 /*
2663 * Once we start copying data !! 2153 * When a sequential read accesses a page several times,
2664 * cachelines that might be c !! 2154 * only mark it as accessed the first time.
2665 */ 2155 */
2666 writably_mapped = mapping_wri !! 2156 if (prev_index != index || offset != prev_offset)
>> 2157 mark_page_accessed(page);
>> 2158 prev_index = index;
2667 2159
2668 /* 2160 /*
2669 * When a read accesses the s !! 2161 * Ok, we have the page, and it's up-to-date, so
2670 * mark it as accessed the fi !! 2162 * now we can copy it to user space...
2671 */ 2163 */
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 2164
2684 if (end_offset < foli !! 2165 ret = copy_page_to_iter(page, offset, nr, iter);
2685 break; !! 2166 offset += ret;
2686 if (i > 0) !! 2167 index += offset >> PAGE_SHIFT;
2687 folio_mark_ac !! 2168 offset &= ~PAGE_MASK;
2688 /* !! 2169 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 2170
2696 copied = copy_folio_t !! 2171 put_page(page);
>> 2172 written += ret;
>> 2173 if (!iov_iter_count(iter))
>> 2174 goto out;
>> 2175 if (ret < nr) {
>> 2176 error = -EFAULT;
>> 2177 goto out;
>> 2178 }
>> 2179 continue;
2697 2180
2698 already_read += copie !! 2181 page_not_up_to_date:
2699 iocb->ki_pos += copie !! 2182 /* Get exclusive access to the page ... */
2700 last_pos = iocb->ki_p !! 2183 error = lock_page_killable(page);
>> 2184 if (unlikely(error))
>> 2185 goto readpage_error;
>> 2186
>> 2187 page_not_up_to_date_locked:
>> 2188 /* Did it get truncated before we got the lock? */
>> 2189 if (!page->mapping) {
>> 2190 unlock_page(page);
>> 2191 put_page(page);
>> 2192 continue;
>> 2193 }
2701 2194
2702 if (copied < bytes) { !! 2195 /* Did somebody else fill it already? */
2703 error = -EFAU !! 2196 if (PageUptodate(page)) {
2704 break; !! 2197 unlock_page(page);
>> 2198 goto page_ok;
>> 2199 }
>> 2200
>> 2201 readpage:
>> 2202 /*
>> 2203 * A previous I/O error may have been due to temporary
>> 2204 * failures, eg. multipath errors.
>> 2205 * PG_error will be set again if readpage fails.
>> 2206 */
>> 2207 ClearPageError(page);
>> 2208 /* Start the actual read. The read will unlock the page. */
>> 2209 error = mapping->a_ops->readpage(filp, page);
>> 2210
>> 2211 if (unlikely(error)) {
>> 2212 if (error == AOP_TRUNCATED_PAGE) {
>> 2213 put_page(page);
>> 2214 error = 0;
>> 2215 goto find_page;
2705 } 2216 }
>> 2217 goto readpage_error;
2706 } 2218 }
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 2219
2713 file_accessed(filp); !! 2220 if (!PageUptodate(page)) {
2714 ra->prev_pos = last_pos; !! 2221 error = lock_page_killable(page);
2715 return already_read ? already_read : !! 2222 if (unlikely(error))
2716 } !! 2223 goto readpage_error;
2717 EXPORT_SYMBOL_GPL(filemap_read); !! 2224 if (!PageUptodate(page)) {
2718 !! 2225 if (page->mapping == NULL) {
2719 int kiocb_write_and_wait(struct kiocb *iocb, !! 2226 /*
2720 { !! 2227 * invalidate_mapping_pages got it
2721 struct address_space *mapping = iocb- !! 2228 */
2722 loff_t pos = iocb->ki_pos; !! 2229 unlock_page(page);
2723 loff_t end = pos + count - 1; !! 2230 put_page(page);
2724 !! 2231 goto find_page;
2725 if (iocb->ki_flags & IOCB_NOWAIT) { !! 2232 }
2726 if (filemap_range_needs_write !! 2233 unlock_page(page);
2727 return -EAGAIN; !! 2234 shrink_readahead_size_eio(filp, ra);
2728 return 0; !! 2235 error = -EIO;
2729 } !! 2236 goto readpage_error;
>> 2237 }
>> 2238 unlock_page(page);
>> 2239 }
2730 2240
2731 return filemap_write_and_wait_range(m !! 2241 goto page_ok;
2732 } <<
2733 EXPORT_SYMBOL_GPL(kiocb_write_and_wait); <<
2734 2242
2735 int filemap_invalidate_pages(struct address_s !! 2243 readpage_error:
2736 loff_t pos, loff !! 2244 /* UHHUH! A synchronous read error occurred. Report it */
2737 { !! 2245 put_page(page);
2738 int ret; !! 2246 goto out;
2739 2247
2740 if (nowait) { !! 2248 no_cached_page:
2741 /* we could block if there ar !! 2249 /*
2742 if (filemap_range_has_page(ma !! 2250 * Ok, it wasn't cached, so we need to create a new
2743 return -EAGAIN; !! 2251 * page..
2744 } else { !! 2252 */
2745 ret = filemap_write_and_wait_ !! 2253 page = page_cache_alloc(mapping);
2746 if (ret) !! 2254 if (!page) {
2747 return ret; !! 2255 error = -ENOMEM;
>> 2256 goto out;
>> 2257 }
>> 2258 error = add_to_page_cache_lru(page, mapping, index,
>> 2259 mapping_gfp_constraint(mapping, GFP_KERNEL));
>> 2260 if (error) {
>> 2261 put_page(page);
>> 2262 if (error == -EEXIST) {
>> 2263 error = 0;
>> 2264 goto find_page;
>> 2265 }
>> 2266 goto out;
>> 2267 }
>> 2268 goto readpage;
2748 } 2269 }
2749 2270
2750 /* !! 2271 would_block:
2751 * After a write we want buffered rea !! 2272 error = -EAGAIN;
2752 * the new data. We invalidate clean !! 2273 out:
2753 * about to write. We do this *befor !! 2274 ra->prev_pos = prev_index;
2754 * without clobbering -EIOCBQUEUED fr !! 2275 ra->prev_pos <<= PAGE_SHIFT;
2755 */ !! 2276 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 2277
2764 return filemap_invalidate_pages(mappi !! 2278 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2765 iocb- !! 2279 file_accessed(filp);
2766 iocb- !! 2280 return written ? written : error;
2767 } 2281 }
2768 EXPORT_SYMBOL_GPL(kiocb_invalidate_pages); <<
2769 2282
2770 /** 2283 /**
2771 * generic_file_read_iter - generic filesyste 2284 * generic_file_read_iter - generic filesystem read routine
2772 * @iocb: kernel I/O control block 2285 * @iocb: kernel I/O control block
2773 * @iter: destination for the data read 2286 * @iter: destination for the data read
2774 * 2287 *
2775 * This is the "read_iter()" routine for all 2288 * This is the "read_iter()" routine for all filesystems
2776 * that can use the page cache directly. 2289 * 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: 2290 * Return:
2788 * * number of bytes copied, even for partial 2291 * * number of bytes copied, even for partial reads
2789 * * negative error code (or 0 if IOCB_NOIO) !! 2292 * * negative error code if nothing was read
2790 */ 2293 */
2791 ssize_t 2294 ssize_t
2792 generic_file_read_iter(struct kiocb *iocb, st 2295 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2793 { 2296 {
2794 size_t count = iov_iter_count(iter); 2297 size_t count = iov_iter_count(iter);
2795 ssize_t retval = 0; 2298 ssize_t retval = 0;
2796 2299
2797 if (!count) 2300 if (!count)
2798 return 0; /* skip atime */ !! 2301 goto out; /* skip atime */
2799 2302
2800 if (iocb->ki_flags & IOCB_DIRECT) { 2303 if (iocb->ki_flags & IOCB_DIRECT) {
2801 struct file *file = iocb->ki_ 2304 struct file *file = iocb->ki_filp;
2802 struct address_space *mapping 2305 struct address_space *mapping = file->f_mapping;
2803 struct inode *inode = mapping 2306 struct inode *inode = mapping->host;
>> 2307 loff_t size;
>> 2308
>> 2309 size = i_size_read(inode);
>> 2310 if (iocb->ki_flags & IOCB_NOWAIT) {
>> 2311 if (filemap_range_has_page(mapping, iocb->ki_pos,
>> 2312 iocb->ki_pos + count - 1))
>> 2313 return -EAGAIN;
>> 2314 } else {
>> 2315 retval = filemap_write_and_wait_range(mapping,
>> 2316 iocb->ki_pos,
>> 2317 iocb->ki_pos + count - 1);
>> 2318 if (retval < 0)
>> 2319 goto out;
>> 2320 }
2804 2321
2805 retval = kiocb_write_and_wait <<
2806 if (retval < 0) <<
2807 return retval; <<
2808 file_accessed(file); 2322 file_accessed(file);
2809 2323
2810 retval = mapping->a_ops->dire 2324 retval = mapping->a_ops->direct_IO(iocb, iter);
2811 if (retval >= 0) { 2325 if (retval >= 0) {
2812 iocb->ki_pos += retva 2326 iocb->ki_pos += retval;
2813 count -= retval; 2327 count -= retval;
2814 } 2328 }
2815 if (retval != -EIOCBQUEUED) !! 2329 iov_iter_revert(iter, count - iov_iter_count(iter));
2816 iov_iter_revert(iter, <<
2817 2330
2818 /* 2331 /*
2819 * Btrfs can have a short DIO 2332 * Btrfs can have a short DIO read if we encounter
2820 * compressed extents, so if 2333 * compressed extents, so if there was an error, or if
2821 * we've already read everyth 2334 * we've already read everything we wanted to, or if
2822 * there was a short read bec 2335 * there was a short read because we hit EOF, go ahead
2823 * and return. Otherwise fal 2336 * and return. Otherwise fallthrough to buffered io for
2824 * the rest of the read. Buf 2337 * the rest of the read. Buffered reads will not work for
2825 * DAX files, so don't bother 2338 * DAX files, so don't bother trying.
2826 */ 2339 */
2827 if (retval < 0 || !count || I !! 2340 if (retval < 0 || !count || iocb->ki_pos >= size ||
2828 return retval; !! 2341 IS_DAX(inode))
2829 if (iocb->ki_pos >= i_size_re !! 2342 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 } 2343 }
2867 2344
2868 return spliced; !! 2345 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: 2346 out:
2976 folio_batch_release(&fbatch); !! 2347 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 } 2348 }
>> 2349 EXPORT_SYMBOL(generic_file_read_iter);
3015 2350
3016 static inline size_t seek_folio_size(struct x !! 2351 #ifdef CONFIG_MMU
>> 2352 #define MMAP_LOTSAMISS (100)
>> 2353 static struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
>> 2354 struct file *fpin)
3017 { 2355 {
3018 if (xa_is_value(folio)) !! 2356 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 2357
3049 if (end <= start) !! 2358 if (fpin)
3050 return -ENXIO; !! 2359 return fpin;
3051 <<
3052 rcu_read_lock(); <<
3053 while ((folio = find_get_entry(&xas, <<
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 2360
3063 seek_size = seek_folio_size(& !! 2361 /*
3064 pos = round_up((u64)pos + 1, !! 2362 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
3065 start = folio_seek_hole_data( !! 2363 * anything, so we only pin the file and drop the mmap_sem if only
3066 seek_data); !! 2364 * FAULT_FLAG_ALLOW_RETRY is set.
3067 if (start < pos) !! 2365 */
3068 goto unlock; !! 2366 if ((flags & (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT)) ==
3069 if (start >= end) !! 2367 FAULT_FLAG_ALLOW_RETRY) {
3070 break; !! 2368 fpin = get_file(vmf->vma->vm_file);
3071 if (seek_size > PAGE_SIZE) !! 2369 up_read(&vmf->vma->vm_mm->mmap_sem);
3072 xas_set(&xas, pos >> <<
3073 if (!xa_is_value(folio)) <<
3074 folio_put(folio); <<
3075 } 2370 }
3076 if (seek_data) !! 2371 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 } 2372 }
3086 2373
3087 #ifdef CONFIG_MMU <<
3088 #define MMAP_LOTSAMISS (100) <<
3089 /* 2374 /*
3090 * lock_folio_maybe_drop_mmap - lock the page !! 2375 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_sem
3091 * @vmf - the vm_fault for this fault. 2376 * @vmf - the vm_fault for this fault.
3092 * @folio - the folio to lock. !! 2377 * @page - the page to lock.
3093 * @fpin - the pointer to the file we may pin 2378 * @fpin - the pointer to the file we may pin (or is already pinned).
3094 * 2379 *
3095 * This works similar to lock_folio_or_retry !! 2380 * 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 !! 2381 * 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 !! 2382 * 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 !! 2383 * 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 */ 2384 */
3101 static int lock_folio_maybe_drop_mmap(struct !! 2385 static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page,
3102 struct f 2386 struct file **fpin)
3103 { 2387 {
3104 if (folio_trylock(folio)) !! 2388 if (trylock_page(page))
3105 return 1; 2389 return 1;
3106 2390
3107 /* 2391 /*
3108 * NOTE! This will make us return wit 2392 * NOTE! This will make us return with VM_FAULT_RETRY, but with
3109 * the fault lock still held. That's !! 2393 * the mmap_sem still held. That's how FAULT_FLAG_RETRY_NOWAIT
3110 * is supposed to work. We have way t 2394 * is supposed to work. We have way too many special cases..
3111 */ 2395 */
3112 if (vmf->flags & FAULT_FLAG_RETRY_NOW 2396 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
3113 return 0; 2397 return 0;
3114 2398
3115 *fpin = maybe_unlock_mmap_for_io(vmf, 2399 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
3116 if (vmf->flags & FAULT_FLAG_KILLABLE) 2400 if (vmf->flags & FAULT_FLAG_KILLABLE) {
3117 if (__folio_lock_killable(fol !! 2401 if (__lock_page_killable(page)) {
3118 /* 2402 /*
3119 * We didn't have the !! 2403 * We didn't have the right flags to drop the mmap_sem,
3120 * fault lock, but al !! 2404 * but all fault_handlers only check for fatal signals
3121 * for fatal signals !! 2405 * if we return VM_FAULT_RETRY, so we need to drop the
3122 * so we need to drop !! 2406 * mmap_sem here and return 0 if we don't have a fpin.
3123 * return 0 if we don <<
3124 */ 2407 */
3125 if (*fpin == NULL) 2408 if (*fpin == NULL)
3126 release_fault !! 2409 up_read(&vmf->vma->vm_mm->mmap_sem);
3127 return 0; 2410 return 0;
3128 } 2411 }
3129 } else 2412 } else
3130 __folio_lock(folio); !! 2413 __lock_page(page);
3131 <<
3132 return 1; 2414 return 1;
3133 } 2415 }
3134 2416
>> 2417
3135 /* 2418 /*
3136 * Synchronous readahead happens when we don' 2419 * 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 2420 * 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 2421 * 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 2422 * 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 2423 * returned needs to be fput()'ed when we're done with it.
3141 */ 2424 */
3142 static struct file *do_sync_mmap_readahead(st 2425 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
3143 { 2426 {
3144 struct file *file = vmf->vma->vm_file 2427 struct file *file = vmf->vma->vm_file;
3145 struct file_ra_state *ra = &file->f_r 2428 struct file_ra_state *ra = &file->f_ra;
3146 struct address_space *mapping = file- 2429 struct address_space *mapping = file->f_mapping;
3147 DEFINE_READAHEAD(ractl, file, ra, map <<
3148 struct file *fpin = NULL; 2430 struct file *fpin = NULL;
3149 unsigned long vm_flags = vmf->vma->vm !! 2431 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 2432
3170 /* If we don't want any read-ahead, d 2433 /* If we don't want any read-ahead, don't bother */
3171 if (vm_flags & VM_RAND_READ) !! 2434 if (vmf->vma->vm_flags & VM_RAND_READ)
3172 return fpin; 2435 return fpin;
3173 if (!ra->ra_pages) 2436 if (!ra->ra_pages)
3174 return fpin; 2437 return fpin;
3175 2438
3176 if (vm_flags & VM_SEQ_READ) { !! 2439 if (vmf->vma->vm_flags & VM_SEQ_READ) {
3177 fpin = maybe_unlock_mmap_for_ 2440 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3178 page_cache_sync_ra(&ractl, ra !! 2441 page_cache_sync_readahead(mapping, ra, file, offset,
>> 2442 ra->ra_pages);
3179 return fpin; 2443 return fpin;
3180 } 2444 }
3181 2445
3182 /* Avoid banging the cache line if no 2446 /* Avoid banging the cache line if not needed */
3183 mmap_miss = READ_ONCE(ra->mmap_miss); !! 2447 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
3184 if (mmap_miss < MMAP_LOTSAMISS * 10) !! 2448 ra->mmap_miss++;
3185 WRITE_ONCE(ra->mmap_miss, ++m <<
3186 2449
3187 /* 2450 /*
3188 * Do we miss much more than hit in t 2451 * Do we miss much more than hit in this file? If so,
3189 * stop bothering with read-ahead. It 2452 * stop bothering with read-ahead. It will only hurt.
3190 */ 2453 */
3191 if (mmap_miss > MMAP_LOTSAMISS) !! 2454 if (ra->mmap_miss > MMAP_LOTSAMISS)
3192 return fpin; 2455 return fpin;
3193 2456
3194 /* 2457 /*
3195 * mmap read-around 2458 * mmap read-around
3196 */ 2459 */
3197 fpin = maybe_unlock_mmap_for_io(vmf, 2460 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3198 ra->start = max_t(long, 0, vmf->pgoff !! 2461 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
3199 ra->size = ra->ra_pages; 2462 ra->size = ra->ra_pages;
3200 ra->async_size = ra->ra_pages / 4; 2463 ra->async_size = ra->ra_pages / 4;
3201 ractl._index = ra->start; !! 2464 ra_submit(ra, mapping, file);
3202 page_cache_ra_order(&ractl, ra, 0); <<
3203 return fpin; 2465 return fpin;
3204 } 2466 }
3205 2467
3206 /* 2468 /*
3207 * Asynchronous readahead happens when we fin 2469 * Asynchronous readahead happens when we find the page and PG_readahead,
3208 * so we want to possibly extend the readahea 2470 * 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 !! 2471 * was pinned if we have to drop the mmap_sem in order to do IO.
3210 */ 2472 */
3211 static struct file *do_async_mmap_readahead(s 2473 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
3212 s !! 2474 struct page *page)
3213 { 2475 {
3214 struct file *file = vmf->vma->vm_file 2476 struct file *file = vmf->vma->vm_file;
3215 struct file_ra_state *ra = &file->f_r 2477 struct file_ra_state *ra = &file->f_ra;
3216 DEFINE_READAHEAD(ractl, file, ra, fil !! 2478 struct address_space *mapping = file->f_mapping;
3217 struct file *fpin = NULL; 2479 struct file *fpin = NULL;
3218 unsigned int mmap_miss; !! 2480 pgoff_t offset = vmf->pgoff;
3219 2481
3220 /* If we don't want any read-ahead, d 2482 /* If we don't want any read-ahead, don't bother */
3221 if (vmf->vma->vm_flags & VM_RAND_READ !! 2483 if (vmf->vma->vm_flags & VM_RAND_READ)
3222 return fpin; 2484 return fpin;
3223 !! 2485 if (ra->mmap_miss > 0)
3224 mmap_miss = READ_ONCE(ra->mmap_miss); !! 2486 ra->mmap_miss--;
3225 if (mmap_miss) !! 2487 if (PageReadahead(page)) {
3226 WRITE_ONCE(ra->mmap_miss, --m <<
3227 <<
3228 if (folio_test_readahead(folio)) { <<
3229 fpin = maybe_unlock_mmap_for_ 2488 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3230 page_cache_async_ra(&ractl, f !! 2489 page_cache_async_readahead(mapping, ra, file,
>> 2490 page, offset, ra->ra_pages);
3231 } 2491 }
3232 return fpin; 2492 return fpin;
3233 } 2493 }
3234 2494
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 /** 2495 /**
3279 * filemap_fault - read in file data for page 2496 * filemap_fault - read in file data for page fault handling
3280 * @vmf: struct vm_fault containing de 2497 * @vmf: struct vm_fault containing details of the fault
3281 * 2498 *
3282 * filemap_fault() is invoked via the vma ope 2499 * filemap_fault() is invoked via the vma operations vector for a
3283 * mapped memory region to read in file data 2500 * mapped memory region to read in file data during a page fault.
3284 * 2501 *
3285 * The goto's are kind of ugly, but this stre 2502 * 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 2503 * it in the page cache, and handles the special cases reasonably without
3287 * having a lot of duplicated code. 2504 * having a lot of duplicated code.
3288 * 2505 *
3289 * vma->vm_mm->mmap_lock must be held on entr !! 2506 * vma->vm_mm->mmap_sem must be held on entry.
3290 * 2507 *
3291 * If our return value has VM_FAULT_RETRY set !! 2508 * If our return value has VM_FAULT_RETRY set, it's because the mmap_sem
3292 * may be dropped before doing I/O or by lock !! 2509 * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
3293 * 2510 *
3294 * If our return value does not have VM_FAULT !! 2511 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
3295 * has not been released. 2512 * has not been released.
3296 * 2513 *
3297 * We never return with VM_FAULT_RETRY and a 2514 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3298 * 2515 *
3299 * Return: bitwise-OR of %VM_FAULT_ codes. 2516 * Return: bitwise-OR of %VM_FAULT_ codes.
3300 */ 2517 */
3301 vm_fault_t filemap_fault(struct vm_fault *vmf 2518 vm_fault_t filemap_fault(struct vm_fault *vmf)
3302 { 2519 {
3303 int error; 2520 int error;
3304 struct file *file = vmf->vma->vm_file 2521 struct file *file = vmf->vma->vm_file;
3305 struct file *fpin = NULL; 2522 struct file *fpin = NULL;
3306 struct address_space *mapping = file- 2523 struct address_space *mapping = file->f_mapping;
>> 2524 struct file_ra_state *ra = &file->f_ra;
3307 struct inode *inode = mapping->host; 2525 struct inode *inode = mapping->host;
3308 pgoff_t max_idx, index = vmf->pgoff; !! 2526 pgoff_t offset = vmf->pgoff;
3309 struct folio *folio; !! 2527 pgoff_t max_off;
>> 2528 struct page *page;
3310 vm_fault_t ret = 0; 2529 vm_fault_t ret = 0;
3311 bool mapping_locked = false; <<
3312 2530
3313 max_idx = DIV_ROUND_UP(i_size_read(in !! 2531 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3314 if (unlikely(index >= max_idx)) !! 2532 if (unlikely(offset >= max_off))
3315 return VM_FAULT_SIGBUS; 2533 return VM_FAULT_SIGBUS;
3316 2534
3317 trace_mm_filemap_fault(mapping, index <<
3318 <<
3319 /* 2535 /*
3320 * Do we have something in the page c 2536 * Do we have something in the page cache already?
3321 */ 2537 */
3322 folio = filemap_get_folio(mapping, in !! 2538 page = find_get_page(mapping, offset);
3323 if (likely(!IS_ERR(folio))) { !! 2539 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
3324 /* 2540 /*
3325 * We found the page, so try !! 2541 * We found the page, so try async readahead before
3326 * the lock. !! 2542 * waiting for the lock.
3327 */ 2543 */
3328 if (!(vmf->flags & FAULT_FLAG !! 2544 fpin = do_async_mmap_readahead(vmf, page);
3329 fpin = do_async_mmap_ !! 2545 } 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 2546 /* No page in the page cache at all */
3340 count_vm_event(PGMAJFAULT); 2547 count_vm_event(PGMAJFAULT);
3341 count_memcg_event_mm(vmf->vma 2548 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3342 ret = VM_FAULT_MAJOR; 2549 ret = VM_FAULT_MAJOR;
3343 fpin = do_sync_mmap_readahead 2550 fpin = do_sync_mmap_readahead(vmf);
3344 retry_find: 2551 retry_find:
3345 /* !! 2552 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 2553 FGP_CREAT|FGP_FOR_MMAP,
3355 vmf 2554 vmf->gfp_mask);
3356 if (IS_ERR(folio)) { !! 2555 if (!page) {
3357 if (fpin) 2556 if (fpin)
3358 goto out_retr 2557 goto out_retry;
3359 filemap_invalidate_un !! 2558 return vmf_error(-ENOMEM);
3360 return VM_FAULT_OOM; <<
3361 } 2559 }
3362 } 2560 }
3363 2561
3364 if (!lock_folio_maybe_drop_mmap(vmf, !! 2562 if (!lock_page_maybe_drop_mmap(vmf, page, &fpin))
3365 goto out_retry; 2563 goto out_retry;
3366 2564
3367 /* Did it get truncated? */ 2565 /* Did it get truncated? */
3368 if (unlikely(folio->mapping != mappin !! 2566 if (unlikely(page->mapping != mapping)) {
3369 folio_unlock(folio); !! 2567 unlock_page(page);
3370 folio_put(folio); !! 2568 put_page(page);
3371 goto retry_find; 2569 goto retry_find;
3372 } 2570 }
3373 VM_BUG_ON_FOLIO(!folio_contains(folio !! 2571 VM_BUG_ON_PAGE(page->index != offset, page);
3374 2572
3375 /* 2573 /*
3376 * We have a locked folio in the page !! 2574 * 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 !! 2575 * that it's up-to-date. If not, it is going to be due to an error.
3378 * or because readahead was otherwise <<
3379 */ 2576 */
3380 if (unlikely(!folio_test_uptodate(fol !! 2577 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; 2578 goto page_not_uptodate;
3399 } <<
3400 2579
3401 /* 2580 /*
3402 * We've made it this far and we had !! 2581 * 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 2582 * time to return to the upper layer and have it re-find the vma and
3404 * redo the fault. 2583 * redo the fault.
3405 */ 2584 */
3406 if (fpin) { 2585 if (fpin) {
3407 folio_unlock(folio); !! 2586 unlock_page(page);
3408 goto out_retry; 2587 goto out_retry;
3409 } 2588 }
3410 if (mapping_locked) <<
3411 filemap_invalidate_unlock_sha <<
3412 2589
3413 /* 2590 /*
3414 * Found the page and have a referenc 2591 * Found the page and have a reference on it.
3415 * We must recheck i_size under page 2592 * We must recheck i_size under page lock.
3416 */ 2593 */
3417 max_idx = DIV_ROUND_UP(i_size_read(in !! 2594 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3418 if (unlikely(index >= max_idx)) { !! 2595 if (unlikely(offset >= max_off)) {
3419 folio_unlock(folio); !! 2596 unlock_page(page);
3420 folio_put(folio); !! 2597 put_page(page);
3421 return VM_FAULT_SIGBUS; 2598 return VM_FAULT_SIGBUS;
3422 } 2599 }
3423 2600
3424 vmf->page = folio_file_page(folio, in !! 2601 vmf->page = page;
3425 return ret | VM_FAULT_LOCKED; 2602 return ret | VM_FAULT_LOCKED;
3426 2603
3427 page_not_uptodate: 2604 page_not_uptodate:
3428 /* 2605 /*
3429 * Umm, take care of errors if the pa 2606 * Umm, take care of errors if the page isn't up-to-date.
3430 * Try to re-read it _once_. We do th 2607 * Try to re-read it _once_. We do this synchronously,
3431 * because there really aren't any pe 2608 * because there really aren't any performance issues here
3432 * and we need to check for errors. 2609 * and we need to check for errors.
3433 */ 2610 */
>> 2611 ClearPageError(page);
3434 fpin = maybe_unlock_mmap_for_io(vmf, 2612 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3435 error = filemap_read_folio(file, mapp !! 2613 error = mapping->a_ops->readpage(file, page);
>> 2614 if (!error) {
>> 2615 wait_on_page_locked(page);
>> 2616 if (!PageUptodate(page))
>> 2617 error = -EIO;
>> 2618 }
3436 if (fpin) 2619 if (fpin)
3437 goto out_retry; 2620 goto out_retry;
3438 folio_put(folio); !! 2621 put_page(page);
3439 2622
3440 if (!error || error == AOP_TRUNCATED_ 2623 if (!error || error == AOP_TRUNCATED_PAGE)
3441 goto retry_find; 2624 goto retry_find;
3442 filemap_invalidate_unlock_shared(mapp <<
3443 2625
>> 2626 /* Things didn't work out. Return zero to tell the mm layer so. */
>> 2627 shrink_readahead_size_eio(file, ra);
3444 return VM_FAULT_SIGBUS; 2628 return VM_FAULT_SIGBUS;
3445 2629
3446 out_retry: 2630 out_retry:
3447 /* 2631 /*
3448 * We dropped the mmap_lock, we need !! 2632 * We dropped the mmap_sem, we need to return to the fault handler to
3449 * re-find the vma and come back and 2633 * re-find the vma and come back and find our hopefully still populated
3450 * page. 2634 * page.
3451 */ 2635 */
3452 if (!IS_ERR(folio)) !! 2636 if (page)
3453 folio_put(folio); !! 2637 put_page(page);
3454 if (mapping_locked) <<
3455 filemap_invalidate_unlock_sha <<
3456 if (fpin) 2638 if (fpin)
3457 fput(fpin); 2639 fput(fpin);
3458 return ret | VM_FAULT_RETRY; 2640 return ret | VM_FAULT_RETRY;
3459 } 2641 }
3460 EXPORT_SYMBOL(filemap_fault); 2642 EXPORT_SYMBOL(filemap_fault);
3461 2643
3462 static bool filemap_map_pmd(struct vm_fault * !! 2644 void filemap_map_pages(struct vm_fault *vmf,
3463 pgoff_t start) !! 2645 pgoff_t start_pgoff, pgoff_t end_pgoff)
3464 { <<
3465 struct mm_struct *mm = vmf->vma->vm_m <<
3466 <<
3467 /* Huge page is mapped? No need to pr <<
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 { 2646 {
3493 struct folio *folio = xas_next_entry( !! 2647 struct file *file = vmf->vma->vm_file;
>> 2648 struct address_space *mapping = file->f_mapping;
>> 2649 pgoff_t last_pgoff = start_pgoff;
3494 unsigned long max_idx; 2650 unsigned long max_idx;
>> 2651 XA_STATE(xas, &mapping->i_pages, start_pgoff);
>> 2652 struct page *head, *page;
3495 2653
3496 do { !! 2654 rcu_read_lock();
3497 if (!folio) !! 2655 xas_for_each(&xas, page, end_pgoff) {
3498 return NULL; !! 2656 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; 2657 continue;
3507 /* Has the page moved or been !! 2658 if (xa_is_value(page))
3508 if (unlikely(folio != xas_rel !! 2659 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 2660
3528 return NULL; !! 2661 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 2662
3549 /* 2663 /*
3550 * If there are too many foli !! 2664 * Check for a locked page first, as a speculative
3551 * in a file, they will proba !! 2665 * 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 */ 2666 */
3556 if (!folio_test_workingset(fo !! 2667 if (PageLocked(head))
3557 (*mmap_miss)++; !! 2668 goto next;
>> 2669 if (!page_cache_get_speculative(head))
>> 2670 goto next;
3558 2671
3559 /* !! 2672 /* The page was split under us? */
3560 * NOTE: If there're PTE mark !! 2673 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; 2674 goto skip;
3566 2675
3567 count++; !! 2676 /* Has the page moved? */
3568 continue; !! 2677 if (unlikely(page != xas_reload(&xas)))
3569 skip: !! 2678 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 2679
3649 if (filemap_map_pmd(vmf, folio, start !! 2680 if (!PageUptodate(page) ||
3650 ret = VM_FAULT_NOPAGE; !! 2681 PageReadahead(page) ||
3651 goto out; !! 2682 PageHWPoison(page))
3652 } !! 2683 goto skip;
>> 2684 if (!trylock_page(page))
>> 2685 goto skip;
3653 2686
3654 addr = vma->vm_start + ((start_pgoff !! 2687 if (page->mapping != mapping || !PageUptodate(page))
3655 vmf->pte = pte_offset_map_lock(vma->v !! 2688 goto unlock;
3656 if (!vmf->pte) { <<
3657 folio_unlock(folio); <<
3658 folio_put(folio); <<
3659 goto out; <<
3660 } <<
3661 2689
3662 file_end = DIV_ROUND_UP(i_size_read(m !! 2690 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3663 if (end_pgoff > file_end) !! 2691 if (page->index >= max_idx)
3664 end_pgoff = file_end; !! 2692 goto unlock;
3665 2693
3666 folio_type = mm_counter_file(folio); !! 2694 if (file->f_ra.mmap_miss > 0)
3667 do { !! 2695 file->f_ra.mmap_miss--;
3668 unsigned long end; <<
3669 2696
3670 addr += (xas.xa_index - last_ !! 2697 vmf->address += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3671 vmf->pte += xas.xa_index - la !! 2698 if (vmf->pte)
>> 2699 vmf->pte += xas.xa_index - last_pgoff;
3672 last_pgoff = xas.xa_index; 2700 last_pgoff = xas.xa_index;
3673 end = folio_next_index(folio) !! 2701 if (alloc_set_pte(vmf, NULL, page))
3674 nr_pages = min(end, end_pgoff !! 2702 goto unlock;
3675 !! 2703 unlock_page(page);
3676 if (!folio_test_large(folio)) !! 2704 goto next;
3677 ret |= filemap_map_or !! 2705 unlock:
3678 folio !! 2706 unlock_page(page);
3679 else !! 2707 skip:
3680 ret |= filemap_map_fo !! 2708 put_page(page);
3681 xas.x !! 2709 next:
3682 nr_pa !! 2710 /* Huge page is mapped? No need to proceed. */
3683 !! 2711 if (pmd_trans_huge(*vmf->pmd))
3684 folio_unlock(folio); !! 2712 break;
3685 folio_put(folio); !! 2713 }
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(); 2714 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 } 2715 }
3701 EXPORT_SYMBOL(filemap_map_pages); 2716 EXPORT_SYMBOL(filemap_map_pages);
3702 2717
3703 vm_fault_t filemap_page_mkwrite(struct vm_fau 2718 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3704 { 2719 {
3705 struct address_space *mapping = vmf-> !! 2720 struct page *page = vmf->page;
3706 struct folio *folio = page_folio(vmf- !! 2721 struct inode *inode = file_inode(vmf->vma->vm_file);
3707 vm_fault_t ret = VM_FAULT_LOCKED; 2722 vm_fault_t ret = VM_FAULT_LOCKED;
3708 2723
3709 sb_start_pagefault(mapping->host->i_s !! 2724 sb_start_pagefault(inode->i_sb);
3710 file_update_time(vmf->vma->vm_file); 2725 file_update_time(vmf->vma->vm_file);
3711 folio_lock(folio); !! 2726 lock_page(page);
3712 if (folio->mapping != mapping) { !! 2727 if (page->mapping != inode->i_mapping) {
3713 folio_unlock(folio); !! 2728 unlock_page(page);
3714 ret = VM_FAULT_NOPAGE; 2729 ret = VM_FAULT_NOPAGE;
3715 goto out; 2730 goto out;
3716 } 2731 }
3717 /* 2732 /*
3718 * We mark the folio dirty already he !! 2733 * We mark the page dirty already here so that when freeze is in
3719 * progress, we are guaranteed that w 2734 * progress, we are guaranteed that writeback during freezing will
3720 * see the dirty folio and writeprote !! 2735 * see the dirty page and writeprotect it again.
3721 */ 2736 */
3722 folio_mark_dirty(folio); !! 2737 set_page_dirty(page);
3723 folio_wait_stable(folio); !! 2738 wait_for_stable_page(page);
3724 out: 2739 out:
3725 sb_end_pagefault(mapping->host->i_sb) !! 2740 sb_end_pagefault(inode->i_sb);
3726 return ret; 2741 return ret;
3727 } 2742 }
3728 2743
3729 const struct vm_operations_struct generic_fil 2744 const struct vm_operations_struct generic_file_vm_ops = {
3730 .fault = filemap_fault, 2745 .fault = filemap_fault,
3731 .map_pages = filemap_map_pages, 2746 .map_pages = filemap_map_pages,
3732 .page_mkwrite = filemap_page_mkwrit 2747 .page_mkwrite = filemap_page_mkwrite,
3733 }; 2748 };
3734 2749
3735 /* This is used for a general mmap of a disk 2750 /* This is used for a general mmap of a disk file */
3736 2751
3737 int generic_file_mmap(struct file *file, stru !! 2752 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
3738 { 2753 {
3739 struct address_space *mapping = file- 2754 struct address_space *mapping = file->f_mapping;
3740 2755
3741 if (!mapping->a_ops->read_folio) !! 2756 if (!mapping->a_ops->readpage)
3742 return -ENOEXEC; 2757 return -ENOEXEC;
3743 file_accessed(file); 2758 file_accessed(file);
3744 vma->vm_ops = &generic_file_vm_ops; 2759 vma->vm_ops = &generic_file_vm_ops;
3745 return 0; 2760 return 0;
3746 } 2761 }
3747 2762
3748 /* 2763 /*
3749 * This is for filesystems which do not imple 2764 * This is for filesystems which do not implement ->writepage.
3750 */ 2765 */
3751 int generic_file_readonly_mmap(struct file *f 2766 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3752 { 2767 {
3753 if (vma_is_shared_maywrite(vma)) !! 2768 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3754 return -EINVAL; 2769 return -EINVAL;
3755 return generic_file_mmap(file, vma); 2770 return generic_file_mmap(file, vma);
3756 } 2771 }
3757 #else 2772 #else
3758 vm_fault_t filemap_page_mkwrite(struct vm_fau 2773 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3759 { 2774 {
3760 return VM_FAULT_SIGBUS; 2775 return VM_FAULT_SIGBUS;
3761 } 2776 }
3762 int generic_file_mmap(struct file *file, stru !! 2777 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
3763 { 2778 {
3764 return -ENOSYS; 2779 return -ENOSYS;
3765 } 2780 }
3766 int generic_file_readonly_mmap(struct file *f !! 2781 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
3767 { 2782 {
3768 return -ENOSYS; 2783 return -ENOSYS;
3769 } 2784 }
3770 #endif /* CONFIG_MMU */ 2785 #endif /* CONFIG_MMU */
3771 2786
3772 EXPORT_SYMBOL(filemap_page_mkwrite); 2787 EXPORT_SYMBOL(filemap_page_mkwrite);
3773 EXPORT_SYMBOL(generic_file_mmap); 2788 EXPORT_SYMBOL(generic_file_mmap);
3774 EXPORT_SYMBOL(generic_file_readonly_mmap); 2789 EXPORT_SYMBOL(generic_file_readonly_mmap);
3775 2790
3776 static struct folio *do_read_cache_folio(stru !! 2791 static struct page *wait_on_page_read(struct page *page)
3777 pgoff_t index, filler_t fille <<
3778 { 2792 {
3779 struct folio *folio; !! 2793 if (!IS_ERR(page)) {
3780 int err; !! 2794 wait_on_page_locked(page);
>> 2795 if (!PageUptodate(page)) {
>> 2796 put_page(page);
>> 2797 page = ERR_PTR(-EIO);
>> 2798 }
>> 2799 }
>> 2800 return page;
>> 2801 }
3781 2802
3782 if (!filler) !! 2803 static struct page *do_read_cache_page(struct address_space *mapping,
3783 filler = mapping->a_ops->read !! 2804 pgoff_t index,
>> 2805 int (*filler)(void *, struct page *),
>> 2806 void *data,
>> 2807 gfp_t gfp)
>> 2808 {
>> 2809 struct page *page;
>> 2810 int err;
3784 repeat: 2811 repeat:
3785 folio = filemap_get_folio(mapping, in !! 2812 page = find_get_page(mapping, index);
3786 if (IS_ERR(folio)) { !! 2813 if (!page) {
3787 folio = filemap_alloc_folio(g !! 2814 page = __page_cache_alloc(gfp);
3788 m !! 2815 if (!page)
3789 if (!folio) <<
3790 return ERR_PTR(-ENOME 2816 return ERR_PTR(-ENOMEM);
3791 index = mapping_align_index(m !! 2817 err = add_to_page_cache_lru(page, mapping, index, gfp);
3792 err = filemap_add_folio(mappi <<
3793 if (unlikely(err)) { 2818 if (unlikely(err)) {
3794 folio_put(folio); !! 2819 put_page(page);
3795 if (err == -EEXIST) 2820 if (err == -EEXIST)
3796 goto repeat; 2821 goto repeat;
3797 /* Presumably ENOMEM 2822 /* Presumably ENOMEM for xarray node */
3798 return ERR_PTR(err); 2823 return ERR_PTR(err);
3799 } 2824 }
3800 2825
3801 goto filler; !! 2826 filler:
>> 2827 if (filler)
>> 2828 err = filler(data, page);
>> 2829 else
>> 2830 err = mapping->a_ops->readpage(data, page);
>> 2831
>> 2832 if (err < 0) {
>> 2833 put_page(page);
>> 2834 return ERR_PTR(err);
>> 2835 }
>> 2836
>> 2837 page = wait_on_page_read(page);
>> 2838 if (IS_ERR(page))
>> 2839 return page;
>> 2840 goto out;
3802 } 2841 }
3803 if (folio_test_uptodate(folio)) !! 2842 if (PageUptodate(page))
3804 goto out; 2843 goto out;
3805 2844
3806 if (!folio_trylock(folio)) { !! 2845 /*
3807 folio_put_wait_locked(folio, !! 2846 * Page is not up to date and may be locked due one of the following
3808 goto repeat; !! 2847 * case a: Page is being filled and the page lock is held
3809 } !! 2848 * case b: Read/write error clearing the page uptodate status
>> 2849 * case c: Truncation in progress (page locked)
>> 2850 * case d: Reclaim in progress
>> 2851 *
>> 2852 * Case a, the page will be up to date when the page is unlocked.
>> 2853 * There is no need to serialise on the page lock here as the page
>> 2854 * is pinned so the lock gives no additional protection. Even if the
>> 2855 * the page is truncated, the data is still valid if PageUptodate as
>> 2856 * it's a race vs truncate race.
>> 2857 * Case b, the page will not be up to date
>> 2858 * Case c, the page may be truncated but in itself, the data may still
>> 2859 * be valid after IO completes as it's a read vs truncate race. The
>> 2860 * operation must restart if the page is not uptodate on unlock but
>> 2861 * otherwise serialising on page lock to stabilise the mapping gives
>> 2862 * no additional guarantees to the caller as the page lock is
>> 2863 * released before return.
>> 2864 * Case d, similar to truncation. If reclaim holds the page lock, it
>> 2865 * will be a race with remove_mapping that determines if the mapping
>> 2866 * is valid on unlock but otherwise the data is valid and there is
>> 2867 * no need to serialise with page lock.
>> 2868 *
>> 2869 * As the page lock gives no additional guarantee, we optimistically
>> 2870 * wait on the page to be unlocked and check if it's up to date and
>> 2871 * use the page if it is. Otherwise, the page lock is required to
>> 2872 * distinguish between the different cases. The motivation is that we
>> 2873 * avoid spurious serialisations and wakeups when multiple processes
>> 2874 * wait on the same page for IO to complete.
>> 2875 */
>> 2876 wait_on_page_locked(page);
>> 2877 if (PageUptodate(page))
>> 2878 goto out;
>> 2879
>> 2880 /* Distinguish between all the cases under the safety of the lock */
>> 2881 lock_page(page);
3810 2882
3811 /* Folio was truncated from mapping * !! 2883 /* Case c or d, restart the operation */
3812 if (!folio->mapping) { !! 2884 if (!page->mapping) {
3813 folio_unlock(folio); !! 2885 unlock_page(page);
3814 folio_put(folio); !! 2886 put_page(page);
3815 goto repeat; 2887 goto repeat;
3816 } 2888 }
3817 2889
3818 /* Someone else locked and filled the 2890 /* Someone else locked and filled the page in a very small window */
3819 if (folio_test_uptodate(folio)) { !! 2891 if (PageUptodate(page)) {
3820 folio_unlock(folio); !! 2892 unlock_page(page);
3821 goto out; 2893 goto out;
3822 } 2894 }
3823 !! 2895 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 2896
3833 out: 2897 out:
3834 folio_mark_accessed(folio); !! 2898 mark_page_accessed(page);
3835 return folio; !! 2899 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 } 2900 }
3860 EXPORT_SYMBOL(read_cache_folio); <<
3861 2901
3862 /** 2902 /**
3863 * mapping_read_folio_gfp - Read into page ca !! 2903 * read_cache_page - read into page cache, fill it if needed
3864 * @mapping: The address_space for the fol !! 2904 * @mapping: the page's address_space
3865 * @index: The index that the allocated !! 2905 * @index: the page index
3866 * @gfp: The page allocator flags to u !! 2906 * @filler: function to perform the read
3867 * !! 2907 * @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 * 2908 *
3871 * The most likely error from this function i !! 2909 * Read into the page cache. If a page already exists, and PageUptodate() is
3872 * possible and so is EINTR. If ->read_folio !! 2910 * not set, try to fill the page and wait for it to become unlocked.
3873 * that will be returned to the caller. <<
3874 * 2911 *
3875 * The function expects mapping->invalidate_l !! 2912 * If the page does not get brought uptodate, return -EIO.
3876 * 2913 *
3877 * Return: Uptodate folio on success, ERR_PTR !! 2914 * Return: up to date page on success, ERR_PTR() on failure.
3878 */ 2915 */
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 2916 struct page *read_cache_page(struct address_space *mapping,
3898 pgoff_t index, filler !! 2917 pgoff_t index,
>> 2918 int (*filler)(void *, struct page *),
>> 2919 void *data)
3899 { 2920 {
3900 return do_read_cache_page(mapping, in !! 2921 return do_read_cache_page(mapping, index, filler, data,
3901 mapping_gfp_mask(mapp 2922 mapping_gfp_mask(mapping));
3902 } 2923 }
3903 EXPORT_SYMBOL(read_cache_page); 2924 EXPORT_SYMBOL(read_cache_page);
3904 2925
3905 /** 2926 /**
3906 * read_cache_page_gfp - read into page cache 2927 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3907 * @mapping: the page's address_space 2928 * @mapping: the page's address_space
3908 * @index: the page index 2929 * @index: the page index
3909 * @gfp: the page allocator flags to u 2930 * @gfp: the page allocator flags to use if allocating
3910 * 2931 *
3911 * This is the same as "read_mapping_page(map 2932 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3912 * any new page allocations done using the sp 2933 * any new page allocations done using the specified allocation flags.
3913 * 2934 *
3914 * If the page does not get brought uptodate, 2935 * If the page does not get brought uptodate, return -EIO.
3915 * 2936 *
3916 * The function expects mapping->invalidate_l <<
3917 * <<
3918 * Return: up to date page on success, ERR_PT 2937 * Return: up to date page on success, ERR_PTR() on failure.
3919 */ 2938 */
3920 struct page *read_cache_page_gfp(struct addre 2939 struct page *read_cache_page_gfp(struct address_space *mapping,
3921 pgoff_t index 2940 pgoff_t index,
3922 gfp_t gfp) 2941 gfp_t gfp)
3923 { 2942 {
3924 return do_read_cache_page(mapping, in 2943 return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3925 } 2944 }
3926 EXPORT_SYMBOL(read_cache_page_gfp); 2945 EXPORT_SYMBOL(read_cache_page_gfp);
3927 2946
3928 /* 2947 /*
3929 * Warn about a page cache invalidation failu !! 2948 * Don't operate on ranges the page cache doesn't support, and don't exceed the
>> 2949 * LFS limits. If pos is under the limit it becomes a short access. If it
>> 2950 * exceeds the limit we return -EFBIG.
>> 2951 */
>> 2952 static int generic_write_check_limits(struct file *file, loff_t pos,
>> 2953 loff_t *count)
>> 2954 {
>> 2955 struct inode *inode = file->f_mapping->host;
>> 2956 loff_t max_size = inode->i_sb->s_maxbytes;
>> 2957 loff_t limit = rlimit(RLIMIT_FSIZE);
>> 2958
>> 2959 if (limit != RLIM_INFINITY) {
>> 2960 if (pos >= limit) {
>> 2961 send_sig(SIGXFSZ, current, 0);
>> 2962 return -EFBIG;
>> 2963 }
>> 2964 *count = min(*count, limit - pos);
>> 2965 }
>> 2966
>> 2967 if (!(file->f_flags & O_LARGEFILE))
>> 2968 max_size = MAX_NON_LFS;
>> 2969
>> 2970 if (unlikely(pos >= max_size))
>> 2971 return -EFBIG;
>> 2972
>> 2973 *count = min(*count, max_size - pos);
>> 2974
>> 2975 return 0;
>> 2976 }
>> 2977
>> 2978 /*
>> 2979 * Performs necessary checks before doing a write
>> 2980 *
>> 2981 * Can adjust writing position or amount of bytes to write.
>> 2982 * Returns appropriate error code that caller should return or
>> 2983 * zero in case that write should be allowed.
>> 2984 */
>> 2985 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
>> 2986 {
>> 2987 struct file *file = iocb->ki_filp;
>> 2988 struct inode *inode = file->f_mapping->host;
>> 2989 loff_t count;
>> 2990 int ret;
>> 2991
>> 2992 if (!iov_iter_count(from))
>> 2993 return 0;
>> 2994
>> 2995 /* FIXME: this is for backwards compatibility with 2.4 */
>> 2996 if (iocb->ki_flags & IOCB_APPEND)
>> 2997 iocb->ki_pos = i_size_read(inode);
>> 2998
>> 2999 if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
>> 3000 return -EINVAL;
>> 3001
>> 3002 count = iov_iter_count(from);
>> 3003 ret = generic_write_check_limits(file, iocb->ki_pos, &count);
>> 3004 if (ret)
>> 3005 return ret;
>> 3006
>> 3007 iov_iter_truncate(from, count);
>> 3008 return iov_iter_count(from);
>> 3009 }
>> 3010 EXPORT_SYMBOL(generic_write_checks);
>> 3011
>> 3012 /*
>> 3013 * Performs necessary checks before doing a clone.
>> 3014 *
>> 3015 * Can adjust amount of bytes to clone via @req_count argument.
>> 3016 * Returns appropriate error code that caller should return or
>> 3017 * zero in case the clone should be allowed.
3930 */ 3018 */
3931 static void dio_warn_stale_pagecache(struct f !! 3019 int generic_remap_checks(struct file *file_in, loff_t pos_in,
>> 3020 struct file *file_out, loff_t pos_out,
>> 3021 loff_t *req_count, unsigned int remap_flags)
3932 { 3022 {
3933 static DEFINE_RATELIMIT_STATE(_rs, 86 !! 3023 struct inode *inode_in = file_in->f_mapping->host;
3934 char pathname[128]; !! 3024 struct inode *inode_out = file_out->f_mapping->host;
3935 char *path; !! 3025 uint64_t count = *req_count;
>> 3026 uint64_t bcount;
>> 3027 loff_t size_in, size_out;
>> 3028 loff_t bs = inode_out->i_sb->s_blocksize;
>> 3029 int ret;
>> 3030
>> 3031 /* The start of both ranges must be aligned to an fs block. */
>> 3032 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs))
>> 3033 return -EINVAL;
>> 3034
>> 3035 /* Ensure offsets don't wrap. */
>> 3036 if (pos_in + count < pos_in || pos_out + count < pos_out)
>> 3037 return -EINVAL;
>> 3038
>> 3039 size_in = i_size_read(inode_in);
>> 3040 size_out = i_size_read(inode_out);
>> 3041
>> 3042 /* Dedupe requires both ranges to be within EOF. */
>> 3043 if ((remap_flags & REMAP_FILE_DEDUP) &&
>> 3044 (pos_in >= size_in || pos_in + count > size_in ||
>> 3045 pos_out >= size_out || pos_out + count > size_out))
>> 3046 return -EINVAL;
>> 3047
>> 3048 /* Ensure the infile range is within the infile. */
>> 3049 if (pos_in >= size_in)
>> 3050 return -EINVAL;
>> 3051 count = min(count, size_in - (uint64_t)pos_in);
3936 3052
3937 errseq_set(&filp->f_mapping->wb_err, !! 3053 ret = generic_write_check_limits(file_out, pos_out, &count);
3938 if (__ratelimit(&_rs)) { !! 3054 if (ret)
3939 path = file_path(filp, pathna !! 3055 return ret;
3940 if (IS_ERR(path)) !! 3056
3941 path = "(unknown)"; !! 3057 /*
3942 pr_crit("Page cache invalidat !! 3058 * If the user wanted us to link to the infile's EOF, round up to the
3943 pr_crit("File: %s PID: %d Com !! 3059 * next block boundary for this check.
3944 current->comm); !! 3060 *
>> 3061 * Otherwise, make sure the count is also block-aligned, having
>> 3062 * already confirmed the starting offsets' block alignment.
>> 3063 */
>> 3064 if (pos_in + count == size_in) {
>> 3065 bcount = ALIGN(size_in, bs) - pos_in;
>> 3066 } else {
>> 3067 if (!IS_ALIGNED(count, bs))
>> 3068 count = ALIGN_DOWN(count, bs);
>> 3069 bcount = count;
3945 } 3070 }
>> 3071
>> 3072 /* Don't allow overlapped cloning within the same file. */
>> 3073 if (inode_in == inode_out &&
>> 3074 pos_out + bcount > pos_in &&
>> 3075 pos_out < pos_in + bcount)
>> 3076 return -EINVAL;
>> 3077
>> 3078 /*
>> 3079 * We shortened the request but the caller can't deal with that, so
>> 3080 * bounce the request back to userspace.
>> 3081 */
>> 3082 if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN))
>> 3083 return -EINVAL;
>> 3084
>> 3085 *req_count = count;
>> 3086 return 0;
3946 } 3087 }
3947 3088
3948 void kiocb_invalidate_post_direct_write(struc !! 3089
>> 3090 /*
>> 3091 * Performs common checks before doing a file copy/clone
>> 3092 * from @file_in to @file_out.
>> 3093 */
>> 3094 int generic_file_rw_checks(struct file *file_in, struct file *file_out)
3949 { 3095 {
3950 struct address_space *mapping = iocb- !! 3096 struct inode *inode_in = file_inode(file_in);
>> 3097 struct inode *inode_out = file_inode(file_out);
>> 3098
>> 3099 /* Don't copy dirs, pipes, sockets... */
>> 3100 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
>> 3101 return -EISDIR;
>> 3102 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
>> 3103 return -EINVAL;
3951 3104
3952 if (mapping->nrpages && !! 3105 if (!(file_in->f_mode & FMODE_READ) ||
3953 invalidate_inode_pages2_range(map !! 3106 !(file_out->f_mode & FMODE_WRITE) ||
3954 iocb->ki_pos >> PAGE_ !! 3107 (file_out->f_flags & O_APPEND))
3955 (iocb->ki_pos + count !! 3108 return -EBADF;
3956 dio_warn_stale_pagecache(iocb !! 3109
>> 3110 return 0;
3957 } 3111 }
3958 3112
>> 3113 /*
>> 3114 * Performs necessary checks before doing a file copy
>> 3115 *
>> 3116 * Can adjust amount of bytes to copy via @req_count argument.
>> 3117 * Returns appropriate error code that caller should return or
>> 3118 * zero in case the copy should be allowed.
>> 3119 */
>> 3120 int generic_copy_file_checks(struct file *file_in, loff_t pos_in,
>> 3121 struct file *file_out, loff_t pos_out,
>> 3122 size_t *req_count, unsigned int flags)
>> 3123 {
>> 3124 struct inode *inode_in = file_inode(file_in);
>> 3125 struct inode *inode_out = file_inode(file_out);
>> 3126 uint64_t count = *req_count;
>> 3127 loff_t size_in;
>> 3128 int ret;
>> 3129
>> 3130 ret = generic_file_rw_checks(file_in, file_out);
>> 3131 if (ret)
>> 3132 return ret;
>> 3133
>> 3134 /* Don't touch certain kinds of inodes */
>> 3135 if (IS_IMMUTABLE(inode_out))
>> 3136 return -EPERM;
>> 3137
>> 3138 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
>> 3139 return -ETXTBSY;
>> 3140
>> 3141 /* Ensure offsets don't wrap. */
>> 3142 if (pos_in + count < pos_in || pos_out + count < pos_out)
>> 3143 return -EOVERFLOW;
>> 3144
>> 3145 /* Shorten the copy to EOF */
>> 3146 size_in = i_size_read(inode_in);
>> 3147 if (pos_in >= size_in)
>> 3148 count = 0;
>> 3149 else
>> 3150 count = min(count, size_in - (uint64_t)pos_in);
>> 3151
>> 3152 ret = generic_write_check_limits(file_out, pos_out, &count);
>> 3153 if (ret)
>> 3154 return ret;
>> 3155
>> 3156 /* Don't allow overlapped copying within the same file. */
>> 3157 if (inode_in == inode_out &&
>> 3158 pos_out + count > pos_in &&
>> 3159 pos_out < pos_in + count)
>> 3160 return -EINVAL;
>> 3161
>> 3162 *req_count = count;
>> 3163 return 0;
>> 3164 }
>> 3165
>> 3166 int pagecache_write_begin(struct file *file, struct address_space *mapping,
>> 3167 loff_t pos, unsigned len, unsigned flags,
>> 3168 struct page **pagep, void **fsdata)
>> 3169 {
>> 3170 const struct address_space_operations *aops = mapping->a_ops;
>> 3171
>> 3172 return aops->write_begin(file, mapping, pos, len, flags,
>> 3173 pagep, fsdata);
>> 3174 }
>> 3175 EXPORT_SYMBOL(pagecache_write_begin);
>> 3176
>> 3177 int pagecache_write_end(struct file *file, struct address_space *mapping,
>> 3178 loff_t pos, unsigned len, unsigned copied,
>> 3179 struct page *page, void *fsdata)
>> 3180 {
>> 3181 const struct address_space_operations *aops = mapping->a_ops;
>> 3182
>> 3183 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
>> 3184 }
>> 3185 EXPORT_SYMBOL(pagecache_write_end);
>> 3186
3959 ssize_t 3187 ssize_t
3960 generic_file_direct_write(struct kiocb *iocb, 3188 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3961 { 3189 {
3962 struct address_space *mapping = iocb- !! 3190 struct file *file = iocb->ki_filp;
3963 size_t write_len = iov_iter_count(fro !! 3191 struct address_space *mapping = file->f_mapping;
3964 ssize_t written; !! 3192 struct inode *inode = mapping->host;
>> 3193 loff_t pos = iocb->ki_pos;
>> 3194 ssize_t written;
>> 3195 size_t write_len;
>> 3196 pgoff_t end;
>> 3197
>> 3198 write_len = iov_iter_count(from);
>> 3199 end = (pos + write_len - 1) >> PAGE_SHIFT;
>> 3200
>> 3201 if (iocb->ki_flags & IOCB_NOWAIT) {
>> 3202 /* If there are pages to writeback, return */
>> 3203 if (filemap_range_has_page(inode->i_mapping, pos,
>> 3204 pos + write_len - 1))
>> 3205 return -EAGAIN;
>> 3206 } else {
>> 3207 written = filemap_write_and_wait_range(mapping, pos,
>> 3208 pos + write_len - 1);
>> 3209 if (written)
>> 3210 goto out;
>> 3211 }
3965 3212
3966 /* 3213 /*
>> 3214 * After a write we want buffered reads to be sure to go to disk to get
>> 3215 * the new data. We invalidate clean cached page from the region we're
>> 3216 * about to write. We do this *before* the write so that we can return
>> 3217 * without clobbering -EIOCBQUEUED from ->direct_IO().
>> 3218 */
>> 3219 written = invalidate_inode_pages2_range(mapping,
>> 3220 pos >> PAGE_SHIFT, end);
>> 3221 /*
3967 * If a page can not be invalidated, 3222 * If a page can not be invalidated, return 0 to fall back
3968 * to buffered write. 3223 * to buffered write.
3969 */ 3224 */
3970 written = kiocb_invalidate_pages(iocb <<
3971 if (written) { 3225 if (written) {
3972 if (written == -EBUSY) 3226 if (written == -EBUSY)
3973 return 0; 3227 return 0;
3974 return written; !! 3228 goto out;
3975 } 3229 }
3976 3230
3977 written = mapping->a_ops->direct_IO(i 3231 written = mapping->a_ops->direct_IO(iocb, from);
3978 3232
3979 /* 3233 /*
3980 * Finally, try again to invalidate c 3234 * Finally, try again to invalidate clean pages which might have been
3981 * cached by non-direct readahead, or 3235 * cached by non-direct readahead, or faulted in by get_user_pages()
3982 * if the source of the write was an 3236 * if the source of the write was an mmap'ed region of the file
3983 * we're writing. Either one is a pr 3237 * we're writing. Either one is a pretty crazy thing to do,
3984 * so we don't support it 100%. If t 3238 * so we don't support it 100%. If this invalidation
3985 * fails, tough, the write still work 3239 * fails, tough, the write still worked...
3986 * 3240 *
3987 * Most of the time we do not need th 3241 * Most of the time we do not need this since dio_complete() will do
3988 * the invalidation for us. However t 3242 * the invalidation for us. However there are some file systems that
3989 * do not end up with dio_complete() 3243 * do not end up with dio_complete() being called, so let's not break
3990 * them by removing it completely. !! 3244 * them by removing it completely
3991 * <<
3992 * Noticeable example is a blkdev_dir <<
3993 * <<
3994 * Skip invalidation for async writes <<
3995 */ 3245 */
3996 if (written > 0) { !! 3246 if (mapping->nrpages)
3997 struct inode *inode = mapping !! 3247 invalidate_inode_pages2_range(mapping,
3998 loff_t pos = iocb->ki_pos; !! 3248 pos >> PAGE_SHIFT, end);
3999 3249
4000 kiocb_invalidate_post_direct_ !! 3250 if (written > 0) {
4001 pos += written; 3251 pos += written;
4002 write_len -= written; 3252 write_len -= written;
4003 if (pos > i_size_read(inode) 3253 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
4004 i_size_write(inode, p 3254 i_size_write(inode, pos);
4005 mark_inode_dirty(inod 3255 mark_inode_dirty(inode);
4006 } 3256 }
4007 iocb->ki_pos = pos; 3257 iocb->ki_pos = pos;
4008 } 3258 }
4009 if (written != -EIOCBQUEUED) !! 3259 iov_iter_revert(from, write_len - iov_iter_count(from));
4010 iov_iter_revert(from, write_l !! 3260 out:
4011 return written; 3261 return written;
4012 } 3262 }
4013 EXPORT_SYMBOL(generic_file_direct_write); 3263 EXPORT_SYMBOL(generic_file_direct_write);
4014 3264
4015 ssize_t generic_perform_write(struct kiocb *i !! 3265 /*
>> 3266 * Find or create a page at the given pagecache position. Return the locked
>> 3267 * page. This function is specifically for buffered writes.
>> 3268 */
>> 3269 struct page *grab_cache_page_write_begin(struct address_space *mapping,
>> 3270 pgoff_t index, unsigned flags)
>> 3271 {
>> 3272 struct page *page;
>> 3273 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
>> 3274
>> 3275 if (flags & AOP_FLAG_NOFS)
>> 3276 fgp_flags |= FGP_NOFS;
>> 3277
>> 3278 page = pagecache_get_page(mapping, index, fgp_flags,
>> 3279 mapping_gfp_mask(mapping));
>> 3280 if (page)
>> 3281 wait_for_stable_page(page);
>> 3282
>> 3283 return page;
>> 3284 }
>> 3285 EXPORT_SYMBOL(grab_cache_page_write_begin);
>> 3286
>> 3287 ssize_t generic_perform_write(struct file *file,
>> 3288 struct iov_iter *i, loff_t pos)
4016 { 3289 {
4017 struct file *file = iocb->ki_filp; <<
4018 loff_t pos = iocb->ki_pos; <<
4019 struct address_space *mapping = file- 3290 struct address_space *mapping = file->f_mapping;
4020 const struct address_space_operations 3291 const struct address_space_operations *a_ops = mapping->a_ops;
4021 size_t chunk = mapping_max_folio_size <<
4022 long status = 0; 3292 long status = 0;
4023 ssize_t written = 0; 3293 ssize_t written = 0;
>> 3294 unsigned int flags = 0;
4024 3295
4025 do { 3296 do {
4026 struct folio *folio; !! 3297 struct page *page;
4027 size_t offset; /* Of !! 3298 unsigned long offset; /* Offset into pagecache page */
4028 size_t bytes; /* By !! 3299 unsigned long bytes; /* Bytes to write to page */
4029 size_t copied; /* By 3300 size_t copied; /* Bytes copied from user */
4030 void *fsdata = NULL; !! 3301 void *fsdata;
4031 3302
4032 bytes = iov_iter_count(i); !! 3303 offset = (pos & (PAGE_SIZE - 1));
4033 retry: !! 3304 bytes = min_t(unsigned long, PAGE_SIZE - offset,
4034 offset = pos & (chunk - 1); !! 3305 iov_iter_count(i));
4035 bytes = min(chunk - offset, b <<
4036 balance_dirty_pages_ratelimit <<
4037 3306
>> 3307 again:
4038 /* 3308 /*
4039 * Bring in the user page tha 3309 * Bring in the user page that we will copy from _first_.
4040 * Otherwise there's a nasty 3310 * Otherwise there's a nasty deadlock on copying from the
4041 * same page as we're writing 3311 * same page as we're writing to, without it being marked
4042 * up-to-date. 3312 * up-to-date.
>> 3313 *
>> 3314 * Not only is this an optimisation, but it is also required
>> 3315 * to check that the address is actually valid, when atomic
>> 3316 * usercopies are used, below.
4043 */ 3317 */
4044 if (unlikely(fault_in_iov_ite !! 3318 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
4045 status = -EFAULT; 3319 status = -EFAULT;
4046 break; 3320 break;
4047 } 3321 }
4048 3322
4049 if (fatal_signal_pending(curr 3323 if (fatal_signal_pending(current)) {
4050 status = -EINTR; 3324 status = -EINTR;
4051 break; 3325 break;
4052 } 3326 }
4053 3327
4054 status = a_ops->write_begin(f !! 3328 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
4055 !! 3329 &page, &fsdata);
4056 if (unlikely(status < 0)) 3330 if (unlikely(status < 0))
4057 break; 3331 break;
4058 3332
4059 offset = offset_in_folio(foli <<
4060 if (bytes > folio_size(folio) <<
4061 bytes = folio_size(fo <<
4062 <<
4063 if (mapping_writably_mapped(m 3333 if (mapping_writably_mapped(mapping))
4064 flush_dcache_folio(fo !! 3334 flush_dcache_page(page);
4065 3335
4066 copied = copy_folio_from_iter !! 3336 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
4067 flush_dcache_folio(folio); !! 3337 flush_dcache_page(page);
4068 3338
4069 status = a_ops->write_end(fil 3339 status = a_ops->write_end(file, mapping, pos, bytes, copied,
4070 !! 3340 page, fsdata);
4071 if (unlikely(status != copied !! 3341 if (unlikely(status < 0))
4072 iov_iter_revert(i, co !! 3342 break;
4073 if (unlikely(status < !! 3343 copied = status;
4074 break; !! 3344
4075 } <<
4076 cond_resched(); 3345 cond_resched();
4077 3346
4078 if (unlikely(status == 0)) { !! 3347 iov_iter_advance(i, copied);
>> 3348 if (unlikely(copied == 0)) {
4079 /* 3349 /*
4080 * A short copy made !! 3350 * If we were unable to copy any data at all, we must
4081 * thing entirely. M !! 3351 * fall back to a single segment length write.
4082 * halfway through, m !! 3352 *
4083 * might be severe me !! 3353 * If we didn't fallback here, we could livelock
>> 3354 * because not all segments in the iov can be copied at
>> 3355 * once without a pagefault.
4084 */ 3356 */
4085 if (chunk > PAGE_SIZE !! 3357 bytes = min_t(unsigned long, PAGE_SIZE - offset,
4086 chunk /= 2; !! 3358 iov_iter_single_seg_count(i));
4087 if (copied) { !! 3359 goto again;
4088 bytes = copie <<
4089 goto retry; <<
4090 } <<
4091 } else { <<
4092 pos += status; <<
4093 written += status; <<
4094 } 3360 }
>> 3361 pos += copied;
>> 3362 written += copied;
>> 3363
>> 3364 balance_dirty_pages_ratelimited(mapping);
4095 } while (iov_iter_count(i)); 3365 } while (iov_iter_count(i));
4096 3366
4097 if (!written) !! 3367 return written ? written : status;
4098 return status; <<
4099 iocb->ki_pos += written; <<
4100 return written; <<
4101 } 3368 }
4102 EXPORT_SYMBOL(generic_perform_write); 3369 EXPORT_SYMBOL(generic_perform_write);
4103 3370
4104 /** 3371 /**
4105 * __generic_file_write_iter - write data to 3372 * __generic_file_write_iter - write data to a file
4106 * @iocb: IO state structure (file, off 3373 * @iocb: IO state structure (file, offset, etc.)
4107 * @from: iov_iter with data to write 3374 * @from: iov_iter with data to write
4108 * 3375 *
4109 * This function does all the work needed for 3376 * This function does all the work needed for actually writing data to a
4110 * file. It does all basic checks, removes SU 3377 * file. It does all basic checks, removes SUID from the file, updates
4111 * modification times and calls proper subrou 3378 * modification times and calls proper subroutines depending on whether we
4112 * do direct IO or a standard buffered write. 3379 * do direct IO or a standard buffered write.
4113 * 3380 *
4114 * It expects i_rwsem to be grabbed unless we !! 3381 * 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. 3382 * object which does not need locking at all.
4116 * 3383 *
4117 * This function does *not* take care of sync 3384 * 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 3385 * A caller has to handle it. This is mainly due to the fact that we want to
4119 * avoid syncing under i_rwsem. !! 3386 * avoid syncing under i_mutex.
4120 * 3387 *
4121 * Return: 3388 * Return:
4122 * * number of bytes written, even for trunca 3389 * * number of bytes written, even for truncated writes
4123 * * negative error code if no data has been 3390 * * negative error code if no data has been written at all
4124 */ 3391 */
4125 ssize_t __generic_file_write_iter(struct kioc 3392 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4126 { 3393 {
4127 struct file *file = iocb->ki_filp; 3394 struct file *file = iocb->ki_filp;
4128 struct address_space *mapping = file- !! 3395 struct address_space * mapping = file->f_mapping;
4129 struct inode *inode = mapping->host; !! 3396 struct inode *inode = mapping->host;
4130 ssize_t ret; !! 3397 ssize_t written = 0;
4131 !! 3398 ssize_t err;
4132 ret = file_remove_privs(file); !! 3399 ssize_t status;
4133 if (ret) !! 3400
4134 return ret; !! 3401 /* We can write back this queue in page reclaim */
>> 3402 current->backing_dev_info = inode_to_bdi(inode);
>> 3403 err = file_remove_privs(file);
>> 3404 if (err)
>> 3405 goto out;
4135 3406
4136 ret = file_update_time(file); !! 3407 err = file_update_time(file);
4137 if (ret) !! 3408 if (err)
4138 return ret; !! 3409 goto out;
4139 3410
4140 if (iocb->ki_flags & IOCB_DIRECT) { 3411 if (iocb->ki_flags & IOCB_DIRECT) {
4141 ret = generic_file_direct_wri !! 3412 loff_t pos, endbyte;
>> 3413
>> 3414 written = generic_file_direct_write(iocb, from);
4142 /* 3415 /*
4143 * If the write stopped short 3416 * If the write stopped short of completing, fall back to
4144 * buffered writes. Some fil 3417 * buffered writes. Some filesystems do this for writes to
4145 * holes, for example. For D 3418 * holes, for example. For DAX files, a buffered write will
4146 * not succeed (even if it di 3419 * not succeed (even if it did, DAX does not handle dirty
4147 * page-cache pages correctly 3420 * page-cache pages correctly).
4148 */ 3421 */
4149 if (ret < 0 || !iov_iter_coun !! 3422 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
4150 return ret; !! 3423 goto out;
4151 return direct_write_fallback( <<
4152 generic_perfo <<
4153 } <<
4154 3424
4155 return generic_perform_write(iocb, fr !! 3425 status = generic_perform_write(file, from, pos = iocb->ki_pos);
>> 3426 /*
>> 3427 * If generic_perform_write() returned a synchronous error
>> 3428 * then we want to return the number of bytes which were
>> 3429 * direct-written, or the error code if that was zero. Note
>> 3430 * that this differs from normal direct-io semantics, which
>> 3431 * will return -EFOO even if some bytes were written.
>> 3432 */
>> 3433 if (unlikely(status < 0)) {
>> 3434 err = status;
>> 3435 goto out;
>> 3436 }
>> 3437 /*
>> 3438 * We need to ensure that the page cache pages are written to
>> 3439 * disk and invalidated to preserve the expected O_DIRECT
>> 3440 * semantics.
>> 3441 */
>> 3442 endbyte = pos + status - 1;
>> 3443 err = filemap_write_and_wait_range(mapping, pos, endbyte);
>> 3444 if (err == 0) {
>> 3445 iocb->ki_pos = endbyte + 1;
>> 3446 written += status;
>> 3447 invalidate_mapping_pages(mapping,
>> 3448 pos >> PAGE_SHIFT,
>> 3449 endbyte >> PAGE_SHIFT);
>> 3450 } else {
>> 3451 /*
>> 3452 * We don't know how much we wrote, so just return
>> 3453 * the number of bytes which were direct-written
>> 3454 */
>> 3455 }
>> 3456 } else {
>> 3457 written = generic_perform_write(file, from, iocb->ki_pos);
>> 3458 if (likely(written > 0))
>> 3459 iocb->ki_pos += written;
>> 3460 }
>> 3461 out:
>> 3462 current->backing_dev_info = NULL;
>> 3463 return written ? written : err;
4156 } 3464 }
4157 EXPORT_SYMBOL(__generic_file_write_iter); 3465 EXPORT_SYMBOL(__generic_file_write_iter);
4158 3466
4159 /** 3467 /**
4160 * generic_file_write_iter - write data to a 3468 * generic_file_write_iter - write data to a file
4161 * @iocb: IO state structure 3469 * @iocb: IO state structure
4162 * @from: iov_iter with data to write 3470 * @from: iov_iter with data to write
4163 * 3471 *
4164 * This is a wrapper around __generic_file_wr 3472 * This is a wrapper around __generic_file_write_iter() to be used by most
4165 * filesystems. It takes care of syncing the 3473 * filesystems. It takes care of syncing the file in case of O_SYNC file
4166 * and acquires i_rwsem as needed. !! 3474 * and acquires i_mutex as needed.
4167 * Return: 3475 * Return:
4168 * * negative error code if no data has been 3476 * * negative error code if no data has been written at all of
4169 * vfs_fsync_range() failed for a synchrono 3477 * vfs_fsync_range() failed for a synchronous write
4170 * * number of bytes written, even for trunca 3478 * * number of bytes written, even for truncated writes
4171 */ 3479 */
4172 ssize_t generic_file_write_iter(struct kiocb 3480 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4173 { 3481 {
4174 struct file *file = iocb->ki_filp; 3482 struct file *file = iocb->ki_filp;
4175 struct inode *inode = file->f_mapping 3483 struct inode *inode = file->f_mapping->host;
4176 ssize_t ret; 3484 ssize_t ret;
4177 3485
4178 inode_lock(inode); 3486 inode_lock(inode);
4179 ret = generic_write_checks(iocb, from 3487 ret = generic_write_checks(iocb, from);
4180 if (ret > 0) 3488 if (ret > 0)
4181 ret = __generic_file_write_it 3489 ret = __generic_file_write_iter(iocb, from);
4182 inode_unlock(inode); 3490 inode_unlock(inode);
4183 3491
4184 if (ret > 0) 3492 if (ret > 0)
4185 ret = generic_write_sync(iocb 3493 ret = generic_write_sync(iocb, ret);
4186 return ret; 3494 return ret;
4187 } 3495 }
4188 EXPORT_SYMBOL(generic_file_write_iter); 3496 EXPORT_SYMBOL(generic_file_write_iter);
4189 3497
4190 /** 3498 /**
4191 * filemap_release_folio() - Release fs-speci !! 3499 * 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 * 3500 *
4195 * The address_space is trying to release any !! 3501 * @page: the page which the kernel is trying to free
4196 * (presumably at folio->private). !! 3502 * @gfp_mask: memory allocation flags (and I/O mode)
4197 * 3503 *
4198 * This will also be called if the private_2 !! 3504 * The address_space is to try to release any data against the page
4199 * indicating that the folio has other metada !! 3505 * (presumably at page->private).
4200 * 3506 *
4201 * The @gfp argument specifies whether I/O ma !! 3507 * 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 !! 3508 * page is known to the local caching routines.
4203 * (__GFP_RECLAIM & __GFP_FS). <<
4204 * 3509 *
4205 * Return: %true if the release was successfu !! 3510 * The @gfp_mask argument specifies whether I/O may be performed to release
4206 */ !! 3511 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
4207 bool filemap_release_folio(struct folio *foli !! 3512 *
4208 { !! 3513 * 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 */ 3514 */
4290 static void filemap_cachestat(struct address_ !! 3515 int try_to_release_page(struct page *page, gfp_t gfp_mask)
4291 pgoff_t first_index, pgoff_t <<
4292 { 3516 {
4293 XA_STATE(xas, &mapping->i_pages, firs !! 3517 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 3518
4316 if (xas_retry(&xas, folio)) !! 3519 BUG_ON(!PageLocked(page));
4317 continue; !! 3520 if (PageWriteback(page))
4318 !! 3521 return 0;
4319 order = xas_get_order(&xas); <<
4320 nr_pages = 1 << order; <<
4321 folio_first_index = round_dow <<
4322 folio_last_index = folio_firs <<
4323 <<
4324 /* Folios might straddle the <<
4325 if (folio_first_index < first <<
4326 nr_pages -= first_ind <<
4327 <<
4328 if (folio_last_index > last_i <<
4329 nr_pages -= folio_las <<
4330 <<
4331 if (xa_is_value(folio)) { <<
4332 /* page is evicted */ <<
4333 void *shadow = (void <<
4334 bool workingset; /* n <<
4335 <<
4336 cs->nr_evicted += nr_ <<
4337 <<
4338 #ifdef CONFIG_SWAP /* implies CONFIG_MMU */ <<
4339 if (shmem_mapping(map <<
4340 /* shmem file <<
4341 swp_entry_t s <<
4342 <<
4343 /* swapin err <<
4344 if (non_swap_ <<
4345 goto <<
4346 <<
4347 /* <<
4348 * Getting a <<
4349 * inode mean <<
4350 * shmem_unus <<
4351 * ensures sw <<
4352 * freeing th <<
4353 * we can rac <<
4354 * invalidati <<
4355 * a shadow i <<
4356 */ <<
4357 shadow = get_ <<
4358 if (!shadow) <<
4359 goto <<
4360 } <<
4361 #endif <<
4362 if (workingset_test_r <<
4363 cs->nr_recent <<
4364 <<
4365 goto resched; <<
4366 } <<
4367 <<
4368 /* page is in cache */ <<
4369 cs->nr_cache += nr_pages; <<
4370 <<
4371 if (xas_get_mark(&xas, PAGECA <<
4372 cs->nr_dirty += nr_pa <<
4373 <<
4374 if (xas_get_mark(&xas, PAGECA <<
4375 cs->nr_writeback += n <<
4376 3522
4377 resched: !! 3523 if (mapping && mapping->a_ops->releasepage)
4378 if (need_resched()) { !! 3524 return mapping->a_ops->releasepage(page, gfp_mask);
4379 xas_pause(&xas); !! 3525 return try_to_free_buffers(page);
4380 cond_resched_rcu(); <<
4381 } <<
4382 } <<
4383 rcu_read_unlock(); <<
4384 } 3526 }
4385 3527
4386 /* !! 3528 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 3529
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.