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> 43 #include <linux/ramfs.h>
44 #include <linux/page_idle.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> 45 #include <asm/pgalloc.h>
51 #include <asm/tlbflush.h> 46 #include <asm/tlbflush.h>
52 #include "internal.h" 47 #include "internal.h"
53 48
54 #define CREATE_TRACE_POINTS 49 #define CREATE_TRACE_POINTS
55 #include <trace/events/filemap.h> 50 #include <trace/events/filemap.h>
56 51
57 /* 52 /*
58 * FIXME: remove all knowledge of the buffer l 53 * FIXME: remove all knowledge of the buffer layer from the core VM
59 */ 54 */
60 #include <linux/buffer_head.h> /* for try_to_f 55 #include <linux/buffer_head.h> /* for try_to_free_buffers */
61 56
62 #include <asm/mman.h> 57 #include <asm/mman.h>
63 58
64 #include "swap.h" <<
65 <<
66 /* 59 /*
67 * Shared mappings implemented 30.11.1994. It' 60 * Shared mappings implemented 30.11.1994. It's not fully working yet,
68 * though. 61 * though.
69 * 62 *
70 * Shared mappings now work. 15.8.1995 Bruno. 63 * Shared mappings now work. 15.8.1995 Bruno.
71 * 64 *
72 * finished 'unifying' the page and buffer cac 65 * finished 'unifying' the page and buffer cache and SMP-threaded the
73 * page-cache, 21.05.1999, Ingo Molnar <mingo@ 66 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
74 * 67 *
75 * SMP-threaded pagemap-LRU 1999, Andrea Arcan 68 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
76 */ 69 */
77 70
78 /* 71 /*
79 * Lock ordering: 72 * Lock ordering:
80 * 73 *
81 * ->i_mmap_rwsem (truncate_page 74 * ->i_mmap_rwsem (truncate_pagecache)
82 * ->private_lock (__free_pte->b !! 75 * ->private_lock (__free_pte->__set_page_dirty_buffers)
83 * ->swap_lock (exclusive_swa 76 * ->swap_lock (exclusive_swap_page, others)
84 * ->i_pages lock 77 * ->i_pages lock
85 * 78 *
86 * ->i_rwsem !! 79 * ->i_mutex
87 * ->invalidate_lock (acquired by f !! 80 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
88 * ->i_mmap_rwsem (truncate->unm <<
89 * 81 *
90 * ->mmap_lock 82 * ->mmap_lock
91 * ->i_mmap_rwsem 83 * ->i_mmap_rwsem
92 * ->page_table_lock or pte_lock (vario 84 * ->page_table_lock or pte_lock (various, mainly in memory.c)
93 * ->i_pages lock (arch-dependen 85 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
94 * 86 *
95 * ->mmap_lock 87 * ->mmap_lock
96 * ->invalidate_lock (filemap_fault !! 88 * ->lock_page (access_process_vm)
97 * ->lock_page (filemap_fault <<
98 * 89 *
99 * ->i_rwsem (generic_perfo !! 90 * ->i_mutex (generic_perform_write)
100 * ->mmap_lock (fault_in_read !! 91 * ->mmap_lock (fault_in_pages_readable->do_page_fault)
101 * 92 *
102 * bdi->wb.list_lock 93 * bdi->wb.list_lock
103 * sb_lock (fs/fs-writeba 94 * sb_lock (fs/fs-writeback.c)
104 * ->i_pages lock (__sync_single 95 * ->i_pages lock (__sync_single_inode)
105 * 96 *
106 * ->i_mmap_rwsem 97 * ->i_mmap_rwsem
107 * ->anon_vma.lock (vma_merge) !! 98 * ->anon_vma.lock (vma_adjust)
108 * 99 *
109 * ->anon_vma.lock 100 * ->anon_vma.lock
110 * ->page_table_lock or pte_lock (anon_ 101 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
111 * 102 *
112 * ->page_table_lock or pte_lock 103 * ->page_table_lock or pte_lock
113 * ->swap_lock (try_to_unmap_ 104 * ->swap_lock (try_to_unmap_one)
114 * ->private_lock (try_to_unmap_ 105 * ->private_lock (try_to_unmap_one)
115 * ->i_pages lock (try_to_unmap_ 106 * ->i_pages lock (try_to_unmap_one)
116 * ->lruvec->lru_lock (follow_page_m !! 107 * ->lruvec->lru_lock (follow_page->mark_page_accessed)
117 * ->lruvec->lru_lock (check_pte_ran !! 108 * ->lruvec->lru_lock (check_pte_range->isolate_lru_page)
118 * ->private_lock (folio_remove_ !! 109 * ->private_lock (page_remove_rmap->set_page_dirty)
119 * ->i_pages lock (folio_remove_ !! 110 * ->i_pages lock (page_remove_rmap->set_page_dirty)
120 * bdi.wb->list_lock (folio_remove_ !! 111 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
121 * ->inode->i_lock (folio_remove_ !! 112 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
122 * ->memcg->move_lock (folio_remove_ !! 113 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
123 * bdi.wb->list_lock (zap_pte_range 114 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
124 * ->inode->i_lock (zap_pte_range 115 * ->inode->i_lock (zap_pte_range->set_page_dirty)
125 * ->private_lock (zap_pte_range !! 116 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
>> 117 *
>> 118 * ->i_mmap_rwsem
>> 119 * ->tasklist_lock (memory_failure, collect_procs_ao)
126 */ 120 */
127 121
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 122 static void page_cache_delete(struct address_space *mapping,
138 struct foli !! 123 struct page *page, void *shadow)
139 { 124 {
140 XA_STATE(xas, &mapping->i_pages, folio !! 125 XA_STATE(xas, &mapping->i_pages, page->index);
141 long nr = 1; !! 126 unsigned int nr = 1;
142 127
143 mapping_set_update(&xas, mapping); 128 mapping_set_update(&xas, mapping);
144 129
145 xas_set_order(&xas, folio->index, foli !! 130 /* hugetlb pages are represented by a single entry in the xarray */
146 nr = folio_nr_pages(folio); !! 131 if (!PageHuge(page)) {
>> 132 xas_set_order(&xas, page->index, compound_order(page));
>> 133 nr = compound_nr(page);
>> 134 }
147 135
148 VM_BUG_ON_FOLIO(!folio_test_locked(fol !! 136 VM_BUG_ON_PAGE(!PageLocked(page), page);
>> 137 VM_BUG_ON_PAGE(PageTail(page), page);
>> 138 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
149 139
150 xas_store(&xas, shadow); 140 xas_store(&xas, shadow);
151 xas_init_marks(&xas); 141 xas_init_marks(&xas);
152 142
153 folio->mapping = NULL; !! 143 page->mapping = NULL;
154 /* Leave page->index set: truncation l 144 /* Leave page->index set: truncation lookup relies upon it */
155 mapping->nrpages -= nr; 145 mapping->nrpages -= nr;
156 } 146 }
157 147
158 static void filemap_unaccount_folio(struct add !! 148 static void unaccount_page_cache_page(struct address_space *mapping,
159 struct folio *folio) !! 149 struct page *page)
160 { 150 {
161 long nr; !! 151 int nr;
>> 152
>> 153 /*
>> 154 * if we're uptodate, flush out into the cleancache, otherwise
>> 155 * invalidate any existing cleancache entries. We can't leave
>> 156 * stale data around in the cleancache once our page is gone
>> 157 */
>> 158 if (PageUptodate(page) && PageMappedToDisk(page))
>> 159 cleancache_put_page(page);
>> 160 else
>> 161 cleancache_invalidate_page(mapping, page);
>> 162
>> 163 VM_BUG_ON_PAGE(PageTail(page), page);
>> 164 VM_BUG_ON_PAGE(page_mapped(page), page);
>> 165 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
>> 166 int mapcount;
162 167
163 VM_BUG_ON_FOLIO(folio_mapped(folio), f <<
164 if (!IS_ENABLED(CONFIG_DEBUG_VM) && un <<
165 pr_alert("BUG: Bad page cache 168 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
166 current->comm, folio_ !! 169 current->comm, page_to_pfn(page));
167 dump_page(&folio->page, "still !! 170 dump_page(page, "still mapped when deleted");
168 dump_stack(); 171 dump_stack();
169 add_taint(TAINT_BAD_PAGE, LOCK 172 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
170 173
171 if (mapping_exiting(mapping) & !! 174 mapcount = page_mapcount(page);
172 int mapcount = folio_m !! 175 if (mapping_exiting(mapping) &&
173 !! 176 page_count(page) >= mapcount + 2) {
174 if (folio_ref_count(fo !! 177 /*
175 /* !! 178 * All vmas have already been torn down, so it's
176 * All vmas ha !! 179 * a good bet that actually the page is unmapped,
177 * a good bet !! 180 * and we'd prefer not to leak it: if we're wrong,
178 * and we'd ra !! 181 * some other bad page check should catch it later.
179 * another bad !! 182 */
180 */ !! 183 page_mapcount_reset(page);
181 atomic_set(&fo !! 184 page_ref_sub(page, mapcount);
182 folio_ref_sub( <<
183 } <<
184 } 185 }
185 } 186 }
186 187
187 /* hugetlb folios do not participate i !! 188 /* hugetlb pages do not participate in page cache accounting. */
188 if (folio_test_hugetlb(folio)) !! 189 if (PageHuge(page))
189 return; 190 return;
190 191
191 nr = folio_nr_pages(folio); !! 192 nr = thp_nr_pages(page);
192 193
193 __lruvec_stat_mod_folio(folio, NR_FILE !! 194 __mod_lruvec_page_state(page, NR_FILE_PAGES, -nr);
194 if (folio_test_swapbacked(folio)) { !! 195 if (PageSwapBacked(page)) {
195 __lruvec_stat_mod_folio(folio, !! 196 __mod_lruvec_page_state(page, NR_SHMEM, -nr);
196 if (folio_test_pmd_mappable(fo !! 197 if (PageTransHuge(page))
197 __lruvec_stat_mod_foli !! 198 __mod_lruvec_page_state(page, NR_SHMEM_THPS, -nr);
198 } else if (folio_test_pmd_mappable(fol !! 199 } else if (PageTransHuge(page)) {
199 __lruvec_stat_mod_folio(folio, !! 200 __mod_lruvec_page_state(page, NR_FILE_THPS, -nr);
200 filemap_nr_thps_dec(mapping); 201 filemap_nr_thps_dec(mapping);
201 } 202 }
202 203
203 /* 204 /*
204 * At this point folio must be either !! 205 * At this point page must be either written or cleaned by
205 * truncate. Dirty folio here signals !! 206 * truncate. Dirty page here signals a bug and loss of
206 * unwritten data - on ordinary filesy !! 207 * 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 * 208 *
212 * Below fixes dirty accounting after !! 209 * This fixes dirty accounting after removing the page entirely
213 * but leaves the dirty flag set: it h !! 210 * but leaves PageDirty set: it has no effect for truncated
214 * folio and anyway will be cleared be !! 211 * page and anyway will be cleared before returning page into
215 * buddy allocator. 212 * buddy allocator.
216 */ 213 */
217 if (WARN_ON_ONCE(folio_test_dirty(foli !! 214 if (WARN_ON_ONCE(PageDirty(page)))
218 mapping_can_writeback !! 215 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
219 folio_account_cleaned(folio, i <<
220 } 216 }
221 217
222 /* 218 /*
223 * Delete a page from the page cache and free 219 * Delete a page from the page cache and free it. Caller has to make
224 * sure the page is locked and that nobody els 220 * sure the page is locked and that nobody else uses it - or that usage
225 * is safe. The caller must hold the i_pages 221 * is safe. The caller must hold the i_pages lock.
226 */ 222 */
227 void __filemap_remove_folio(struct folio *foli !! 223 void __delete_from_page_cache(struct page *page, void *shadow)
228 { 224 {
229 struct address_space *mapping = folio- !! 225 struct address_space *mapping = page->mapping;
>> 226
>> 227 trace_mm_filemap_delete_from_page_cache(page);
230 228
231 trace_mm_filemap_delete_from_page_cach !! 229 unaccount_page_cache_page(mapping, page);
232 filemap_unaccount_folio(mapping, folio !! 230 page_cache_delete(mapping, page, shadow);
233 page_cache_delete(mapping, folio, shad <<
234 } 231 }
235 232
236 void filemap_free_folio(struct address_space * !! 233 static void page_cache_free_page(struct address_space *mapping,
>> 234 struct page *page)
237 { 235 {
238 void (*free_folio)(struct folio *); !! 236 void (*freepage)(struct page *);
239 int refs = 1; <<
240 237
241 free_folio = mapping->a_ops->free_foli !! 238 freepage = mapping->a_ops->freepage;
242 if (free_folio) !! 239 if (freepage)
243 free_folio(folio); !! 240 freepage(page);
244 241
245 if (folio_test_large(folio)) !! 242 if (PageTransHuge(page) && !PageHuge(page)) {
246 refs = folio_nr_pages(folio); !! 243 page_ref_sub(page, thp_nr_pages(page));
247 folio_put_refs(folio, refs); !! 244 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
>> 245 } else {
>> 246 put_page(page);
>> 247 }
248 } 248 }
249 249
250 /** 250 /**
251 * filemap_remove_folio - Remove folio from pa !! 251 * delete_from_page_cache - delete page from page cache
252 * @folio: The folio. !! 252 * @page: the page which the kernel is trying to remove from page cache
253 * 253 *
254 * This must be called only on folios that are !! 254 * 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 !! 255 * 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 !! 256 * has a reference on the page.
257 */ !! 257 */
258 void filemap_remove_folio(struct folio *folio) !! 258 void delete_from_page_cache(struct page *page)
259 { !! 259 {
260 struct address_space *mapping = folio- !! 260 struct address_space *mapping = page_mapping(page);
261 !! 261 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 } <<
273 262
274 /* !! 263 BUG_ON(!PageLocked(page));
275 * page_cache_delete_batch - delete several fo !! 264 xa_lock_irqsave(&mapping->i_pages, flags);
276 * @mapping: the mapping to which folios belon !! 265 __delete_from_page_cache(page, NULL);
277 * @fbatch: batch of folios to delete !! 266 xa_unlock_irqrestore(&mapping->i_pages, flags);
278 * !! 267
279 * The function walks over mapping->i_pages an !! 268 page_cache_free_page(mapping, page);
280 * @fbatch from the mapping. The function expe !! 269 }
281 * by page index and is optimised for it to be !! 270 EXPORT_SYMBOL(delete_from_page_cache);
282 * It tolerates holes in @fbatch (mapping entr !! 271
283 * modified). !! 272 /*
>> 273 * page_cache_delete_batch - delete several pages from page cache
>> 274 * @mapping: the mapping to which pages belong
>> 275 * @pvec: pagevec with pages to delete
>> 276 *
>> 277 * The function walks over mapping->i_pages and removes pages passed in @pvec
>> 278 * from the mapping. The function expects @pvec to be sorted by page index
>> 279 * and is optimised for it to be dense.
>> 280 * It tolerates holes in @pvec (mapping entries at those indices are not
>> 281 * modified). The function expects only THP head pages to be present in the
>> 282 * @pvec.
284 * 283 *
285 * The function expects the i_pages lock to be 284 * The function expects the i_pages lock to be held.
286 */ 285 */
287 static void page_cache_delete_batch(struct add 286 static void page_cache_delete_batch(struct address_space *mapping,
288 struct folio_batc !! 287 struct pagevec *pvec)
289 { 288 {
290 XA_STATE(xas, &mapping->i_pages, fbatc !! 289 XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
291 long total_pages = 0; !! 290 int total_pages = 0;
292 int i = 0; 291 int i = 0;
293 struct folio *folio; !! 292 struct page *page;
294 293
295 mapping_set_update(&xas, mapping); 294 mapping_set_update(&xas, mapping);
296 xas_for_each(&xas, folio, ULONG_MAX) { !! 295 xas_for_each(&xas, page, ULONG_MAX) {
297 if (i >= folio_batch_count(fba !! 296 if (i >= pagevec_count(pvec))
298 break; 297 break;
299 298
300 /* A swap/dax/shadow entry got 299 /* A swap/dax/shadow entry got inserted? Skip it. */
301 if (xa_is_value(folio)) !! 300 if (xa_is_value(page))
302 continue; 301 continue;
303 /* 302 /*
304 * A page got inserted in our 303 * A page got inserted in our range? Skip it. We have our
305 * pages locked so they are pr 304 * pages locked so they are protected from being removed.
306 * If we see a page whose inde 305 * If we see a page whose index is higher than ours, it
307 * means our page has been rem 306 * means our page has been removed, which shouldn't be
308 * possible because we're hold 307 * possible because we're holding the PageLock.
309 */ 308 */
310 if (folio != fbatch->folios[i] !! 309 if (page != pvec->pages[i]) {
311 VM_BUG_ON_FOLIO(folio- !! 310 VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index,
312 fbatch !! 311 page);
313 continue; 312 continue;
314 } 313 }
315 314
316 WARN_ON_ONCE(!folio_test_locke !! 315 WARN_ON_ONCE(!PageLocked(page));
317 316
318 folio->mapping = NULL; !! 317 if (page->index == xas.xa_index)
319 /* Leave folio->index set: tru !! 318 page->mapping = NULL;
>> 319 /* Leave page->index set: truncation lookup relies on it */
320 320
321 i++; !! 321 /*
>> 322 * Move to the next page in the vector if this is a regular
>> 323 * page or the index is of the last sub-page of this compound
>> 324 * page.
>> 325 */
>> 326 if (page->index + compound_nr(page) - 1 == xas.xa_index)
>> 327 i++;
322 xas_store(&xas, NULL); 328 xas_store(&xas, NULL);
323 total_pages += folio_nr_pages( !! 329 total_pages++;
324 } 330 }
325 mapping->nrpages -= total_pages; 331 mapping->nrpages -= total_pages;
326 } 332 }
327 333
328 void delete_from_page_cache_batch(struct addre 334 void delete_from_page_cache_batch(struct address_space *mapping,
329 struct folio !! 335 struct pagevec *pvec)
330 { 336 {
331 int i; 337 int i;
>> 338 unsigned long flags;
332 339
333 if (!folio_batch_count(fbatch)) !! 340 if (!pagevec_count(pvec))
334 return; 341 return;
335 342
336 spin_lock(&mapping->host->i_lock); !! 343 xa_lock_irqsave(&mapping->i_pages, flags);
337 xa_lock_irq(&mapping->i_pages); !! 344 for (i = 0; i < pagevec_count(pvec); i++) {
338 for (i = 0; i < folio_batch_count(fbat !! 345 trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
339 struct folio *folio = fbatch-> <<
340 <<
341 trace_mm_filemap_delete_from_p <<
342 filemap_unaccount_folio(mappin <<
343 } <<
344 page_cache_delete_batch(mapping, fbatc <<
345 xa_unlock_irq(&mapping->i_pages); <<
346 if (mapping_shrinkable(mapping)) <<
347 inode_add_lru(mapping->host); <<
348 spin_unlock(&mapping->host->i_lock); <<
349 346
350 for (i = 0; i < folio_batch_count(fbat !! 347 unaccount_page_cache_page(mapping, pvec->pages[i]);
351 filemap_free_folio(mapping, fb !! 348 }
>> 349 page_cache_delete_batch(mapping, pvec);
>> 350 xa_unlock_irqrestore(&mapping->i_pages, flags);
>> 351
>> 352 for (i = 0; i < pagevec_count(pvec); i++)
>> 353 page_cache_free_page(mapping, pvec->pages[i]);
352 } 354 }
353 355
354 int filemap_check_errors(struct address_space 356 int filemap_check_errors(struct address_space *mapping)
355 { 357 {
356 int ret = 0; 358 int ret = 0;
357 /* Check for outstanding write errors 359 /* Check for outstanding write errors */
358 if (test_bit(AS_ENOSPC, &mapping->flag 360 if (test_bit(AS_ENOSPC, &mapping->flags) &&
359 test_and_clear_bit(AS_ENOSPC, &map 361 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
360 ret = -ENOSPC; 362 ret = -ENOSPC;
361 if (test_bit(AS_EIO, &mapping->flags) 363 if (test_bit(AS_EIO, &mapping->flags) &&
362 test_and_clear_bit(AS_EIO, &mappin 364 test_and_clear_bit(AS_EIO, &mapping->flags))
363 ret = -EIO; 365 ret = -EIO;
364 return ret; 366 return ret;
365 } 367 }
366 EXPORT_SYMBOL(filemap_check_errors); 368 EXPORT_SYMBOL(filemap_check_errors);
367 369
368 static int filemap_check_and_keep_errors(struc 370 static int filemap_check_and_keep_errors(struct address_space *mapping)
369 { 371 {
370 /* Check for outstanding write errors 372 /* Check for outstanding write errors */
371 if (test_bit(AS_EIO, &mapping->flags)) 373 if (test_bit(AS_EIO, &mapping->flags))
372 return -EIO; 374 return -EIO;
373 if (test_bit(AS_ENOSPC, &mapping->flag 375 if (test_bit(AS_ENOSPC, &mapping->flags))
374 return -ENOSPC; 376 return -ENOSPC;
375 return 0; 377 return 0;
376 } 378 }
377 379
378 /** 380 /**
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 381 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
406 * @mapping: address space structure to wri 382 * @mapping: address space structure to write
407 * @start: offset in bytes where the rang 383 * @start: offset in bytes where the range starts
408 * @end: offset in bytes where the rang 384 * @end: offset in bytes where the range ends (inclusive)
409 * @sync_mode: enable synchronous operation 385 * @sync_mode: enable synchronous operation
410 * 386 *
411 * Start writeback against all of a mapping's 387 * Start writeback against all of a mapping's dirty pages that lie
412 * within the byte offsets <start, end> inclus 388 * within the byte offsets <start, end> inclusive.
413 * 389 *
414 * If sync_mode is WB_SYNC_ALL then this is a 390 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
415 * opposed to a regular memory cleansing write 391 * opposed to a regular memory cleansing writeback. The difference between
416 * these two operations is that if a dirty pag 392 * these two operations is that if a dirty page/buffer is encountered, it must
417 * be waited upon, and not just skipped over. 393 * be waited upon, and not just skipped over.
418 * 394 *
419 * Return: %0 on success, negative error code 395 * Return: %0 on success, negative error code otherwise.
420 */ 396 */
421 int __filemap_fdatawrite_range(struct address_ 397 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
422 loff_t end, in 398 loff_t end, int sync_mode)
423 { 399 {
>> 400 int ret;
424 struct writeback_control wbc = { 401 struct writeback_control wbc = {
425 .sync_mode = sync_mode, 402 .sync_mode = sync_mode,
426 .nr_to_write = LONG_MAX, 403 .nr_to_write = LONG_MAX,
427 .range_start = start, 404 .range_start = start,
428 .range_end = end, 405 .range_end = end,
429 }; 406 };
430 407
431 return filemap_fdatawrite_wbc(mapping, !! 408 if (!mapping_can_writeback(mapping) ||
>> 409 !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
>> 410 return 0;
>> 411
>> 412 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
>> 413 ret = do_writepages(mapping, &wbc);
>> 414 wbc_detach_inode(&wbc);
>> 415 return ret;
432 } 416 }
433 417
434 static inline int __filemap_fdatawrite(struct 418 static inline int __filemap_fdatawrite(struct address_space *mapping,
435 int sync_mode) 419 int sync_mode)
436 { 420 {
437 return __filemap_fdatawrite_range(mapp 421 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
438 } 422 }
439 423
440 int filemap_fdatawrite(struct address_space *m 424 int filemap_fdatawrite(struct address_space *mapping)
441 { 425 {
442 return __filemap_fdatawrite(mapping, W 426 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
443 } 427 }
444 EXPORT_SYMBOL(filemap_fdatawrite); 428 EXPORT_SYMBOL(filemap_fdatawrite);
445 429
446 int filemap_fdatawrite_range(struct address_sp 430 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
447 loff_t end) 431 loff_t end)
448 { 432 {
449 return __filemap_fdatawrite_range(mapp 433 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
450 } 434 }
451 EXPORT_SYMBOL(filemap_fdatawrite_range); 435 EXPORT_SYMBOL(filemap_fdatawrite_range);
452 436
453 /** 437 /**
454 * filemap_flush - mostly a non-blocking flush 438 * filemap_flush - mostly a non-blocking flush
455 * @mapping: target address_space 439 * @mapping: target address_space
456 * 440 *
457 * This is a mostly non-blocking flush. Not s 441 * This is a mostly non-blocking flush. Not suitable for data-integrity
458 * purposes - I/O may not be started against a 442 * purposes - I/O may not be started against all dirty pages.
459 * 443 *
460 * Return: %0 on success, negative error code 444 * Return: %0 on success, negative error code otherwise.
461 */ 445 */
462 int filemap_flush(struct address_space *mappin 446 int filemap_flush(struct address_space *mapping)
463 { 447 {
464 return __filemap_fdatawrite(mapping, W 448 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
465 } 449 }
466 EXPORT_SYMBOL(filemap_flush); 450 EXPORT_SYMBOL(filemap_flush);
467 451
468 /** 452 /**
469 * filemap_range_has_page - check if a page ex 453 * filemap_range_has_page - check if a page exists in range.
470 * @mapping: address space within wh 454 * @mapping: address space within which to check
471 * @start_byte: offset in bytes where t 455 * @start_byte: offset in bytes where the range starts
472 * @end_byte: offset in bytes where t 456 * @end_byte: offset in bytes where the range ends (inclusive)
473 * 457 *
474 * Find at least one page in the range supplie 458 * Find at least one page in the range supplied, usually used to check if
475 * direct writing in this range will trigger a 459 * direct writing in this range will trigger a writeback.
476 * 460 *
477 * Return: %true if at least one page exists i 461 * Return: %true if at least one page exists in the specified range,
478 * %false otherwise. 462 * %false otherwise.
479 */ 463 */
480 bool filemap_range_has_page(struct address_spa 464 bool filemap_range_has_page(struct address_space *mapping,
481 loff_t start_byte, 465 loff_t start_byte, loff_t end_byte)
482 { 466 {
483 struct folio *folio; !! 467 struct page *page;
484 XA_STATE(xas, &mapping->i_pages, start 468 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
485 pgoff_t max = end_byte >> PAGE_SHIFT; 469 pgoff_t max = end_byte >> PAGE_SHIFT;
486 470
487 if (end_byte < start_byte) 471 if (end_byte < start_byte)
488 return false; 472 return false;
489 473
490 rcu_read_lock(); 474 rcu_read_lock();
491 for (;;) { 475 for (;;) {
492 folio = xas_find(&xas, max); !! 476 page = xas_find(&xas, max);
493 if (xas_retry(&xas, folio)) !! 477 if (xas_retry(&xas, page))
494 continue; 478 continue;
495 /* Shadow entries don't count 479 /* Shadow entries don't count */
496 if (xa_is_value(folio)) !! 480 if (xa_is_value(page))
497 continue; 481 continue;
498 /* 482 /*
499 * We don't need to try to pin 483 * We don't need to try to pin this page; we're about to
500 * release the RCU lock anyway 484 * release the RCU lock anyway. It is enough to know that
501 * there was a page here recen 485 * there was a page here recently.
502 */ 486 */
503 break; 487 break;
504 } 488 }
505 rcu_read_unlock(); 489 rcu_read_unlock();
506 490
507 return folio != NULL; !! 491 return page != NULL;
508 } 492 }
509 EXPORT_SYMBOL(filemap_range_has_page); 493 EXPORT_SYMBOL(filemap_range_has_page);
510 494
511 static void __filemap_fdatawait_range(struct a 495 static void __filemap_fdatawait_range(struct address_space *mapping,
512 loff_t st 496 loff_t start_byte, loff_t end_byte)
513 { 497 {
514 pgoff_t index = start_byte >> PAGE_SHI 498 pgoff_t index = start_byte >> PAGE_SHIFT;
515 pgoff_t end = end_byte >> PAGE_SHIFT; 499 pgoff_t end = end_byte >> PAGE_SHIFT;
516 struct folio_batch fbatch; !! 500 struct pagevec pvec;
517 unsigned nr_folios; !! 501 int nr_pages;
518 502
519 folio_batch_init(&fbatch); !! 503 if (end_byte < start_byte)
>> 504 return;
520 505
>> 506 pagevec_init(&pvec);
521 while (index <= end) { 507 while (index <= end) {
522 unsigned i; 508 unsigned i;
523 509
524 nr_folios = filemap_get_folios !! 510 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
525 PAGECACHE_TAG_ !! 511 end, PAGECACHE_TAG_WRITEBACK);
526 !! 512 if (!nr_pages)
527 if (!nr_folios) <<
528 break; 513 break;
529 514
530 for (i = 0; i < nr_folios; i++ !! 515 for (i = 0; i < nr_pages; i++) {
531 struct folio *folio = !! 516 struct page *page = pvec.pages[i];
532 517
533 folio_wait_writeback(f !! 518 wait_on_page_writeback(page);
>> 519 ClearPageError(page);
534 } 520 }
535 folio_batch_release(&fbatch); !! 521 pagevec_release(&pvec);
536 cond_resched(); 522 cond_resched();
537 } 523 }
538 } 524 }
539 525
540 /** 526 /**
541 * filemap_fdatawait_range - wait for writebac 527 * filemap_fdatawait_range - wait for writeback to complete
542 * @mapping: address space structur 528 * @mapping: address space structure to wait for
543 * @start_byte: offset in bytes where 529 * @start_byte: offset in bytes where the range starts
544 * @end_byte: offset in bytes where 530 * @end_byte: offset in bytes where the range ends (inclusive)
545 * 531 *
546 * Walk the list of under-writeback pages of t 532 * Walk the list of under-writeback pages of the given address space
547 * in the given range and wait for all of them 533 * in the given range and wait for all of them. Check error status of
548 * the address space and return it. 534 * the address space and return it.
549 * 535 *
550 * Since the error status of the address space 536 * Since the error status of the address space is cleared by this function,
551 * callers are responsible for checking the re 537 * callers are responsible for checking the return value and handling and/or
552 * reporting the error. 538 * reporting the error.
553 * 539 *
554 * Return: error status of the address space. 540 * Return: error status of the address space.
555 */ 541 */
556 int filemap_fdatawait_range(struct address_spa 542 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
557 loff_t end_byte) 543 loff_t end_byte)
558 { 544 {
559 __filemap_fdatawait_range(mapping, sta 545 __filemap_fdatawait_range(mapping, start_byte, end_byte);
560 return filemap_check_errors(mapping); 546 return filemap_check_errors(mapping);
561 } 547 }
562 EXPORT_SYMBOL(filemap_fdatawait_range); 548 EXPORT_SYMBOL(filemap_fdatawait_range);
563 549
564 /** 550 /**
565 * filemap_fdatawait_range_keep_errors - wait 551 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
566 * @mapping: address space structur 552 * @mapping: address space structure to wait for
567 * @start_byte: offset in bytes where 553 * @start_byte: offset in bytes where the range starts
568 * @end_byte: offset in bytes where 554 * @end_byte: offset in bytes where the range ends (inclusive)
569 * 555 *
570 * Walk the list of under-writeback pages of t 556 * Walk the list of under-writeback pages of the given address space in the
571 * given range and wait for all of them. Unli 557 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
572 * this function does not clear error status o 558 * this function does not clear error status of the address space.
573 * 559 *
574 * Use this function if callers don't handle e 560 * Use this function if callers don't handle errors themselves. Expected
575 * call sites are system-wide / filesystem-wid 561 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
576 * fsfreeze(8) 562 * fsfreeze(8)
577 */ 563 */
578 int filemap_fdatawait_range_keep_errors(struct 564 int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
579 loff_t start_byte, loff_t end_ 565 loff_t start_byte, loff_t end_byte)
580 { 566 {
581 __filemap_fdatawait_range(mapping, sta 567 __filemap_fdatawait_range(mapping, start_byte, end_byte);
582 return filemap_check_and_keep_errors(m 568 return filemap_check_and_keep_errors(mapping);
583 } 569 }
584 EXPORT_SYMBOL(filemap_fdatawait_range_keep_err 570 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
585 571
586 /** 572 /**
587 * file_fdatawait_range - wait for writeback t 573 * file_fdatawait_range - wait for writeback to complete
588 * @file: file pointing to addre 574 * @file: file pointing to address space structure to wait for
589 * @start_byte: offset in bytes where 575 * @start_byte: offset in bytes where the range starts
590 * @end_byte: offset in bytes where 576 * @end_byte: offset in bytes where the range ends (inclusive)
591 * 577 *
592 * Walk the list of under-writeback pages of t 578 * Walk the list of under-writeback pages of the address space that file
593 * refers to, in the given range and wait for 579 * refers to, in the given range and wait for all of them. Check error
594 * status of the address space vs. the file->f 580 * status of the address space vs. the file->f_wb_err cursor and return it.
595 * 581 *
596 * Since the error status of the file is advan 582 * Since the error status of the file is advanced by this function,
597 * callers are responsible for checking the re 583 * callers are responsible for checking the return value and handling and/or
598 * reporting the error. 584 * reporting the error.
599 * 585 *
600 * Return: error status of the address space v 586 * Return: error status of the address space vs. the file->f_wb_err cursor.
601 */ 587 */
602 int file_fdatawait_range(struct file *file, lo 588 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
603 { 589 {
604 struct address_space *mapping = file-> 590 struct address_space *mapping = file->f_mapping;
605 591
606 __filemap_fdatawait_range(mapping, sta 592 __filemap_fdatawait_range(mapping, start_byte, end_byte);
607 return file_check_and_advance_wb_err(f 593 return file_check_and_advance_wb_err(file);
608 } 594 }
609 EXPORT_SYMBOL(file_fdatawait_range); 595 EXPORT_SYMBOL(file_fdatawait_range);
610 596
611 /** 597 /**
612 * filemap_fdatawait_keep_errors - wait for wr 598 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
613 * @mapping: address space structure to wait f 599 * @mapping: address space structure to wait for
614 * 600 *
615 * Walk the list of under-writeback pages of t 601 * Walk the list of under-writeback pages of the given address space
616 * and wait for all of them. Unlike filemap_f 602 * and wait for all of them. Unlike filemap_fdatawait(), this function
617 * does not clear error status of the address 603 * does not clear error status of the address space.
618 * 604 *
619 * Use this function if callers don't handle e 605 * Use this function if callers don't handle errors themselves. Expected
620 * call sites are system-wide / filesystem-wid 606 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
621 * fsfreeze(8) 607 * fsfreeze(8)
622 * 608 *
623 * Return: error status of the address space. 609 * Return: error status of the address space.
624 */ 610 */
625 int filemap_fdatawait_keep_errors(struct addre 611 int filemap_fdatawait_keep_errors(struct address_space *mapping)
626 { 612 {
627 __filemap_fdatawait_range(mapping, 0, 613 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
628 return filemap_check_and_keep_errors(m 614 return filemap_check_and_keep_errors(mapping);
629 } 615 }
630 EXPORT_SYMBOL(filemap_fdatawait_keep_errors); 616 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
631 617
632 /* Returns true if writeback might be needed o 618 /* Returns true if writeback might be needed or already in progress. */
633 static bool mapping_needs_writeback(struct add 619 static bool mapping_needs_writeback(struct address_space *mapping)
634 { 620 {
635 return mapping->nrpages; 621 return mapping->nrpages;
636 } 622 }
637 623
638 bool filemap_range_has_writeback(struct addres !! 624 /**
639 loff_t start_ !! 625 * filemap_range_needs_writeback - check if range potentially needs writeback
>> 626 * @mapping: address space within which to check
>> 627 * @start_byte: offset in bytes where the range starts
>> 628 * @end_byte: offset in bytes where the range ends (inclusive)
>> 629 *
>> 630 * Find at least one page in the range supplied, usually used to check if
>> 631 * direct writing in this range will trigger a writeback. Used by O_DIRECT
>> 632 * read/write with IOCB_NOWAIT, to see if the caller needs to do
>> 633 * filemap_write_and_wait_range() before proceeding.
>> 634 *
>> 635 * Return: %true if the caller should do filemap_write_and_wait_range() before
>> 636 * doing O_DIRECT to a page in this range, %false otherwise.
>> 637 */
>> 638 bool filemap_range_needs_writeback(struct address_space *mapping,
>> 639 loff_t start_byte, loff_t end_byte)
640 { 640 {
641 XA_STATE(xas, &mapping->i_pages, start 641 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
642 pgoff_t max = end_byte >> PAGE_SHIFT; 642 pgoff_t max = end_byte >> PAGE_SHIFT;
643 struct folio *folio; !! 643 struct page *page;
644 644
>> 645 if (!mapping_needs_writeback(mapping))
>> 646 return false;
>> 647 if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
>> 648 !mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK))
>> 649 return false;
645 if (end_byte < start_byte) 650 if (end_byte < start_byte)
646 return false; 651 return false;
647 652
648 rcu_read_lock(); 653 rcu_read_lock();
649 xas_for_each(&xas, folio, max) { !! 654 xas_for_each(&xas, page, max) {
650 if (xas_retry(&xas, folio)) !! 655 if (xas_retry(&xas, page))
651 continue; 656 continue;
652 if (xa_is_value(folio)) !! 657 if (xa_is_value(page))
653 continue; 658 continue;
654 if (folio_test_dirty(folio) || !! 659 if (PageDirty(page) || PageLocked(page) || PageWriteback(page))
655 folio_test_wri <<
656 break; 660 break;
657 } 661 }
658 rcu_read_unlock(); 662 rcu_read_unlock();
659 return folio != NULL; !! 663 return page != NULL;
660 } 664 }
661 EXPORT_SYMBOL_GPL(filemap_range_has_writeback) !! 665 EXPORT_SYMBOL_GPL(filemap_range_needs_writeback);
662 666
663 /** 667 /**
664 * filemap_write_and_wait_range - write out & 668 * filemap_write_and_wait_range - write out & wait on a file range
665 * @mapping: the address_space for the page 669 * @mapping: the address_space for the pages
666 * @lstart: offset in bytes where the rang 670 * @lstart: offset in bytes where the range starts
667 * @lend: offset in bytes where the rang 671 * @lend: offset in bytes where the range ends (inclusive)
668 * 672 *
669 * Write out and wait upon file offsets lstart 673 * Write out and wait upon file offsets lstart->lend, inclusive.
670 * 674 *
671 * Note that @lend is inclusive (describes the 675 * Note that @lend is inclusive (describes the last byte to be written) so
672 * that this function can be used to write to 676 * that this function can be used to write to the very end-of-file (end = -1).
673 * 677 *
674 * Return: error status of the address space. 678 * Return: error status of the address space.
675 */ 679 */
676 int filemap_write_and_wait_range(struct addres 680 int filemap_write_and_wait_range(struct address_space *mapping,
677 loff_t lstart 681 loff_t lstart, loff_t lend)
678 { 682 {
679 int err = 0, err2; !! 683 int err = 0;
680 <<
681 if (lend < lstart) <<
682 return 0; <<
683 684
684 if (mapping_needs_writeback(mapping)) 685 if (mapping_needs_writeback(mapping)) {
685 err = __filemap_fdatawrite_ran 686 err = __filemap_fdatawrite_range(mapping, lstart, lend,
686 687 WB_SYNC_ALL);
687 /* 688 /*
688 * Even if the above returned 689 * Even if the above returned error, the pages may be
689 * written partially (e.g. -EN 690 * written partially (e.g. -ENOSPC), so we wait for it.
690 * But the -EIO is special cas 691 * But the -EIO is special case, it may indicate the worst
691 * thing (e.g. bug) happened, 692 * thing (e.g. bug) happened, so we avoid waiting for it.
692 */ 693 */
693 if (err != -EIO) !! 694 if (err != -EIO) {
694 __filemap_fdatawait_ra !! 695 int err2 = filemap_fdatawait_range(mapping,
>> 696 lstart, lend);
>> 697 if (!err)
>> 698 err = err2;
>> 699 } else {
>> 700 /* Clear any previously stored errors */
>> 701 filemap_check_errors(mapping);
>> 702 }
>> 703 } else {
>> 704 err = filemap_check_errors(mapping);
695 } 705 }
696 err2 = filemap_check_errors(mapping); <<
697 if (!err) <<
698 err = err2; <<
699 return err; 706 return err;
700 } 707 }
701 EXPORT_SYMBOL(filemap_write_and_wait_range); 708 EXPORT_SYMBOL(filemap_write_and_wait_range);
702 709
703 void __filemap_set_wb_err(struct address_space 710 void __filemap_set_wb_err(struct address_space *mapping, int err)
704 { 711 {
705 errseq_t eseq = errseq_set(&mapping->w 712 errseq_t eseq = errseq_set(&mapping->wb_err, err);
706 713
707 trace_filemap_set_wb_err(mapping, eseq 714 trace_filemap_set_wb_err(mapping, eseq);
708 } 715 }
709 EXPORT_SYMBOL(__filemap_set_wb_err); 716 EXPORT_SYMBOL(__filemap_set_wb_err);
710 717
711 /** 718 /**
712 * file_check_and_advance_wb_err - report wb e 719 * file_check_and_advance_wb_err - report wb error (if any) that was previously
713 * and advance 720 * and advance wb_err to current one
714 * @file: struct file on which the error is be 721 * @file: struct file on which the error is being reported
715 * 722 *
716 * When userland calls fsync (or something lik 723 * When userland calls fsync (or something like nfsd does the equivalent), we
717 * want to report any writeback errors that oc 724 * want to report any writeback errors that occurred since the last fsync (or
718 * since the file was opened if there haven't 725 * since the file was opened if there haven't been any).
719 * 726 *
720 * Grab the wb_err from the mapping. If it mat 727 * 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 728 * then just quickly return 0. The file is all caught up.
722 * 729 *
723 * If it doesn't match, then take the mapping 730 * 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 731 * 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 732 * 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 733 * portion. The error at this point must be reported via proper channels
727 * (a'la fsync, or NFS COMMIT operation, etc.) 734 * (a'la fsync, or NFS COMMIT operation, etc.).
728 * 735 *
729 * While we handle mapping->wb_err with atomic 736 * While we handle mapping->wb_err with atomic operations, the f_wb_err
730 * value is protected by the f_lock since we m 737 * value is protected by the f_lock since we must ensure that it reflects
731 * the latest value swapped in for this file d 738 * the latest value swapped in for this file descriptor.
732 * 739 *
733 * Return: %0 on success, negative error code 740 * Return: %0 on success, negative error code otherwise.
734 */ 741 */
735 int file_check_and_advance_wb_err(struct file 742 int file_check_and_advance_wb_err(struct file *file)
736 { 743 {
737 int err = 0; 744 int err = 0;
738 errseq_t old = READ_ONCE(file->f_wb_er 745 errseq_t old = READ_ONCE(file->f_wb_err);
739 struct address_space *mapping = file-> 746 struct address_space *mapping = file->f_mapping;
740 747
741 /* Locklessly handle the common case w 748 /* Locklessly handle the common case where nothing has changed */
742 if (errseq_check(&mapping->wb_err, old 749 if (errseq_check(&mapping->wb_err, old)) {
743 /* Something changed, must use 750 /* Something changed, must use slow path */
744 spin_lock(&file->f_lock); 751 spin_lock(&file->f_lock);
745 old = file->f_wb_err; 752 old = file->f_wb_err;
746 err = errseq_check_and_advance 753 err = errseq_check_and_advance(&mapping->wb_err,
747 754 &file->f_wb_err);
748 trace_file_check_and_advance_w 755 trace_file_check_and_advance_wb_err(file, old);
749 spin_unlock(&file->f_lock); 756 spin_unlock(&file->f_lock);
750 } 757 }
751 758
752 /* 759 /*
753 * We're mostly using this function as 760 * We're mostly using this function as a drop in replacement for
754 * filemap_check_errors. Clear AS_EIO/ 761 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
755 * that the legacy code would have had 762 * that the legacy code would have had on these flags.
756 */ 763 */
757 clear_bit(AS_EIO, &mapping->flags); 764 clear_bit(AS_EIO, &mapping->flags);
758 clear_bit(AS_ENOSPC, &mapping->flags); 765 clear_bit(AS_ENOSPC, &mapping->flags);
759 return err; 766 return err;
760 } 767 }
761 EXPORT_SYMBOL(file_check_and_advance_wb_err); 768 EXPORT_SYMBOL(file_check_and_advance_wb_err);
762 769
763 /** 770 /**
764 * file_write_and_wait_range - write out & wai 771 * file_write_and_wait_range - write out & wait on a file range
765 * @file: file pointing to address_space 772 * @file: file pointing to address_space with pages
766 * @lstart: offset in bytes where the rang 773 * @lstart: offset in bytes where the range starts
767 * @lend: offset in bytes where the rang 774 * @lend: offset in bytes where the range ends (inclusive)
768 * 775 *
769 * Write out and wait upon file offsets lstart 776 * Write out and wait upon file offsets lstart->lend, inclusive.
770 * 777 *
771 * Note that @lend is inclusive (describes the 778 * Note that @lend is inclusive (describes the last byte to be written) so
772 * that this function can be used to write to 779 * that this function can be used to write to the very end-of-file (end = -1).
773 * 780 *
774 * After writing out and waiting on the data, 781 * After writing out and waiting on the data, we check and advance the
775 * f_wb_err cursor to the latest value, and re 782 * f_wb_err cursor to the latest value, and return any errors detected there.
776 * 783 *
777 * Return: %0 on success, negative error code 784 * Return: %0 on success, negative error code otherwise.
778 */ 785 */
779 int file_write_and_wait_range(struct file *fil 786 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
780 { 787 {
781 int err = 0, err2; 788 int err = 0, err2;
782 struct address_space *mapping = file-> 789 struct address_space *mapping = file->f_mapping;
783 790
784 if (lend < lstart) <<
785 return 0; <<
786 <<
787 if (mapping_needs_writeback(mapping)) 791 if (mapping_needs_writeback(mapping)) {
788 err = __filemap_fdatawrite_ran 792 err = __filemap_fdatawrite_range(mapping, lstart, lend,
789 793 WB_SYNC_ALL);
790 /* See comment of filemap_writ 794 /* See comment of filemap_write_and_wait() */
791 if (err != -EIO) 795 if (err != -EIO)
792 __filemap_fdatawait_ra 796 __filemap_fdatawait_range(mapping, lstart, lend);
793 } 797 }
794 err2 = file_check_and_advance_wb_err(f 798 err2 = file_check_and_advance_wb_err(file);
795 if (!err) 799 if (!err)
796 err = err2; 800 err = err2;
797 return err; 801 return err;
798 } 802 }
799 EXPORT_SYMBOL(file_write_and_wait_range); 803 EXPORT_SYMBOL(file_write_and_wait_range);
800 804
801 /** 805 /**
802 * replace_page_cache_folio - replace a pageca !! 806 * replace_page_cache_page - replace a pagecache page with a new one
803 * @old: folio to be replaced !! 807 * @old: page to be replaced
804 * @new: folio to replace with !! 808 * @new: page to replace with
805 * !! 809 *
806 * This function replaces a folio in the pagec !! 810 * This function replaces a page in the pagecache with a new one. On
807 * success it acquires the pagecache reference !! 811 * success it acquires the pagecache reference for the new page and
808 * drops it for the old folio. Both the old a !! 812 * drops it for the old page. Both the old and new pages must be
809 * locked. This function does not add the new !! 813 * locked. This function does not add the new page to the LRU, the
810 * caller must do that. 814 * caller must do that.
811 * 815 *
812 * The remove + add is atomic. This function 816 * The remove + add is atomic. This function cannot fail.
813 */ 817 */
814 void replace_page_cache_folio(struct folio *ol !! 818 void replace_page_cache_page(struct page *old, struct page *new)
815 { 819 {
816 struct address_space *mapping = old->m 820 struct address_space *mapping = old->mapping;
817 void (*free_folio)(struct folio *) = m !! 821 void (*freepage)(struct page *) = mapping->a_ops->freepage;
818 pgoff_t offset = old->index; 822 pgoff_t offset = old->index;
819 XA_STATE(xas, &mapping->i_pages, offse 823 XA_STATE(xas, &mapping->i_pages, offset);
>> 824 unsigned long flags;
820 825
821 VM_BUG_ON_FOLIO(!folio_test_locked(old !! 826 VM_BUG_ON_PAGE(!PageLocked(old), old);
822 VM_BUG_ON_FOLIO(!folio_test_locked(new !! 827 VM_BUG_ON_PAGE(!PageLocked(new), new);
823 VM_BUG_ON_FOLIO(new->mapping, new); !! 828 VM_BUG_ON_PAGE(new->mapping, new);
824 829
825 folio_get(new); !! 830 get_page(new);
826 new->mapping = mapping; 831 new->mapping = mapping;
827 new->index = offset; 832 new->index = offset;
828 833
829 mem_cgroup_replace_folio(old, new); !! 834 mem_cgroup_migrate(old, new);
830 835
831 xas_lock_irq(&xas); !! 836 xas_lock_irqsave(&xas, flags);
832 xas_store(&xas, new); 837 xas_store(&xas, new);
833 838
834 old->mapping = NULL; 839 old->mapping = NULL;
835 /* hugetlb pages do not participate in 840 /* hugetlb pages do not participate in page cache accounting. */
836 if (!folio_test_hugetlb(old)) !! 841 if (!PageHuge(old))
837 __lruvec_stat_sub_folio(old, N !! 842 __dec_lruvec_page_state(old, NR_FILE_PAGES);
838 if (!folio_test_hugetlb(new)) !! 843 if (!PageHuge(new))
839 __lruvec_stat_add_folio(new, N !! 844 __inc_lruvec_page_state(new, NR_FILE_PAGES);
840 if (folio_test_swapbacked(old)) !! 845 if (PageSwapBacked(old))
841 __lruvec_stat_sub_folio(old, N !! 846 __dec_lruvec_page_state(old, NR_SHMEM);
842 if (folio_test_swapbacked(new)) !! 847 if (PageSwapBacked(new))
843 __lruvec_stat_add_folio(new, N !! 848 __inc_lruvec_page_state(new, NR_SHMEM);
844 xas_unlock_irq(&xas); !! 849 xas_unlock_irqrestore(&xas, flags);
845 if (free_folio) !! 850 if (freepage)
846 free_folio(old); !! 851 freepage(old);
847 folio_put(old); !! 852 put_page(old);
848 } !! 853 }
849 EXPORT_SYMBOL_GPL(replace_page_cache_folio); !! 854 EXPORT_SYMBOL_GPL(replace_page_cache_page);
850 !! 855
851 noinline int __filemap_add_folio(struct addres !! 856 noinline int __add_to_page_cache_locked(struct page *page,
852 struct folio *folio, pgoff_t i !! 857 struct address_space *mapping,
>> 858 pgoff_t offset, gfp_t gfp,
>> 859 void **shadowp)
853 { 860 {
854 XA_STATE(xas, &mapping->i_pages, index !! 861 XA_STATE(xas, &mapping->i_pages, offset);
855 void *alloced_shadow = NULL; !! 862 int huge = PageHuge(page);
856 int alloced_order = 0; !! 863 int error;
857 bool huge; !! 864 bool charged = false;
858 long nr; !! 865
859 !! 866 VM_BUG_ON_PAGE(!PageLocked(page), page);
860 VM_BUG_ON_FOLIO(!folio_test_locked(fol !! 867 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
861 VM_BUG_ON_FOLIO(folio_test_swapbacked( <<
862 VM_BUG_ON_FOLIO(folio_order(folio) < m <<
863 folio); <<
864 mapping_set_update(&xas, mapping); 868 mapping_set_update(&xas, mapping);
865 869
866 VM_BUG_ON_FOLIO(index & (folio_nr_page !! 870 get_page(page);
867 xas_set_order(&xas, index, folio_order !! 871 page->mapping = mapping;
868 huge = folio_test_hugetlb(folio); !! 872 page->index = offset;
869 nr = folio_nr_pages(folio); !! 873
>> 874 if (!huge) {
>> 875 error = mem_cgroup_charge(page, current->mm, gfp);
>> 876 if (error)
>> 877 goto error;
>> 878 charged = true;
>> 879 }
870 880
871 gfp &= GFP_RECLAIM_MASK; 881 gfp &= GFP_RECLAIM_MASK;
872 folio_ref_add(folio, nr); <<
873 folio->mapping = mapping; <<
874 folio->index = xas.xa_index; <<
875 882
876 for (;;) { !! 883 do {
877 int order = -1, split_order = !! 884 unsigned int order = xa_get_order(xas.xa, xas.xa_index);
878 void *entry, *old = NULL; 885 void *entry, *old = NULL;
879 886
>> 887 if (order > thp_order(page))
>> 888 xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
>> 889 order, gfp);
880 xas_lock_irq(&xas); 890 xas_lock_irq(&xas);
881 xas_for_each_conflict(&xas, en 891 xas_for_each_conflict(&xas, entry) {
882 old = entry; 892 old = entry;
883 if (!xa_is_value(entry 893 if (!xa_is_value(entry)) {
884 xas_set_err(&x 894 xas_set_err(&xas, -EEXIST);
885 goto unlock; 895 goto unlock;
886 } 896 }
887 /* <<
888 * If a larger entry e <<
889 * it will be the firs <<
890 */ <<
891 if (order == -1) <<
892 order = xas_ge <<
893 } <<
894 <<
895 /* entry may have changed befo <<
896 if (alloced_order && (old != a <<
897 xas_destroy(&xas); <<
898 alloced_order = 0; <<
899 } 897 }
900 898
901 if (old) { 899 if (old) {
902 if (order > 0 && order !! 900 if (shadowp)
903 /* How to hand !! 901 *shadowp = old;
904 BUG_ON(shmem_m !! 902 /* entry may have been split before we acquired lock */
905 if (!alloced_o !! 903 order = xa_get_order(xas.xa, xas.xa_index);
906 split_ !! 904 if (order > thp_order(page)) {
907 goto u <<
908 } <<
909 xas_split(&xas 905 xas_split(&xas, old, order);
910 xas_reset(&xas 906 xas_reset(&xas);
911 } 907 }
912 if (shadowp) <<
913 *shadowp = old <<
914 } 908 }
915 909
916 xas_store(&xas, folio); !! 910 xas_store(&xas, page);
917 if (xas_error(&xas)) 911 if (xas_error(&xas))
918 goto unlock; 912 goto unlock;
919 913
920 mapping->nrpages += nr; !! 914 mapping->nrpages++;
921 915
922 /* hugetlb pages do not partic 916 /* hugetlb pages do not participate in page cache accounting */
923 if (!huge) { !! 917 if (!huge)
924 __lruvec_stat_mod_foli !! 918 __inc_lruvec_page_state(page, NR_FILE_PAGES);
925 if (folio_test_pmd_map <<
926 __lruvec_stat_ <<
927 <<
928 } <<
929 <<
930 unlock: 919 unlock:
931 xas_unlock_irq(&xas); 920 xas_unlock_irq(&xas);
>> 921 } while (xas_nomem(&xas, gfp));
932 922
933 /* split needed, alloc here an !! 923 if (xas_error(&xas)) {
934 if (split_order) { !! 924 error = xas_error(&xas);
935 xas_split_alloc(&xas, !! 925 if (charged)
936 if (xas_error(&xas)) !! 926 mem_cgroup_uncharge(page);
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 <<
948 if (xas_error(&xas)) <<
949 goto error; 927 goto error;
>> 928 }
950 929
951 trace_mm_filemap_add_to_page_cache(fol !! 930 trace_mm_filemap_add_to_page_cache(page);
952 return 0; 931 return 0;
953 error: 932 error:
954 folio->mapping = NULL; !! 933 page->mapping = NULL;
955 /* Leave page->index set: truncation r 934 /* Leave page->index set: truncation relies upon it */
956 folio_put_refs(folio, nr); !! 935 put_page(page);
957 return xas_error(&xas); !! 936 return error;
>> 937 }
>> 938 ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO);
>> 939
>> 940 /**
>> 941 * add_to_page_cache_locked - add a locked page to the pagecache
>> 942 * @page: page to add
>> 943 * @mapping: the page's address_space
>> 944 * @offset: page index
>> 945 * @gfp_mask: page allocation mode
>> 946 *
>> 947 * This function is used to add a page to the pagecache. It must be locked.
>> 948 * This function does not add the page to the LRU. The caller must do that.
>> 949 *
>> 950 * Return: %0 on success, negative error code otherwise.
>> 951 */
>> 952 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
>> 953 pgoff_t offset, gfp_t gfp_mask)
>> 954 {
>> 955 return __add_to_page_cache_locked(page, mapping, offset,
>> 956 gfp_mask, NULL);
958 } 957 }
959 ALLOW_ERROR_INJECTION(__filemap_add_folio, ERR !! 958 EXPORT_SYMBOL(add_to_page_cache_locked);
960 959
961 int filemap_add_folio(struct address_space *ma !! 960 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
962 pgoff_t index, !! 961 pgoff_t offset, gfp_t gfp_mask)
963 { 962 {
964 void *shadow = NULL; 963 void *shadow = NULL;
965 int ret; 964 int ret;
966 965
967 ret = mem_cgroup_charge(folio, NULL, g !! 966 __SetPageLocked(page);
968 if (ret) !! 967 ret = __add_to_page_cache_locked(page, mapping, offset,
969 return ret; !! 968 gfp_mask, &shadow);
970 !! 969 if (unlikely(ret))
971 __folio_set_locked(folio); !! 970 __ClearPageLocked(page);
972 ret = __filemap_add_folio(mapping, fol !! 971 else {
973 if (unlikely(ret)) { <<
974 mem_cgroup_uncharge(folio); <<
975 __folio_clear_locked(folio); <<
976 } else { <<
977 /* 972 /*
978 * The folio might have been e !! 973 * The page might have been evicted from cache only
979 * recently, in which case it 974 * recently, in which case it should be activated like
980 * any other repeatedly access !! 975 * any other repeatedly accessed page.
981 * The exception is folios get !! 976 * The exception is pages getting rewritten; evicting other
982 * data from the working set, 977 * data from the working set, only to cache data that will
983 * get overwritten with someth 978 * get overwritten with something else, is a waste of memory.
984 */ 979 */
985 WARN_ON_ONCE(folio_test_active !! 980 WARN_ON_ONCE(PageActive(page));
986 if (!(gfp & __GFP_WRITE) && sh !! 981 if (!(gfp_mask & __GFP_WRITE) && shadow)
987 workingset_refault(fol !! 982 workingset_refault(page, shadow);
988 folio_add_lru(folio); !! 983 lru_cache_add(page);
989 } 984 }
990 return ret; 985 return ret;
991 } 986 }
992 EXPORT_SYMBOL_GPL(filemap_add_folio); !! 987 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
993 988
994 #ifdef CONFIG_NUMA 989 #ifdef CONFIG_NUMA
995 struct folio *filemap_alloc_folio_noprof(gfp_t !! 990 struct page *__page_cache_alloc(gfp_t gfp)
996 { 991 {
997 int n; 992 int n;
998 struct folio *folio; !! 993 struct page *page;
999 994
1000 if (cpuset_do_page_mem_spread()) { 995 if (cpuset_do_page_mem_spread()) {
1001 unsigned int cpuset_mems_cook 996 unsigned int cpuset_mems_cookie;
1002 do { 997 do {
1003 cpuset_mems_cookie = 998 cpuset_mems_cookie = read_mems_allowed_begin();
1004 n = cpuset_mem_spread 999 n = cpuset_mem_spread_node();
1005 folio = __folio_alloc !! 1000 page = __alloc_pages_node(n, gfp, 0);
1006 } while (!folio && read_mems_ !! 1001 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
1007 1002
1008 return folio; !! 1003 return page;
1009 } 1004 }
1010 return folio_alloc_noprof(gfp, order) !! 1005 return alloc_pages(gfp, 0);
1011 } 1006 }
1012 EXPORT_SYMBOL(filemap_alloc_folio_noprof); !! 1007 EXPORT_SYMBOL(__page_cache_alloc);
1013 #endif 1008 #endif
1014 1009
1015 /* 1010 /*
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 1011 * In order to wait for pages to become available there must be
1055 * waitqueues associated with pages. By using 1012 * waitqueues associated with pages. By using a hash table of
1056 * waitqueues where the bucket discipline is 1013 * waitqueues where the bucket discipline is to maintain all
1057 * waiters on the same queue and wake all whe 1014 * waiters on the same queue and wake all when any of the pages
1058 * become available, and for the woken contex 1015 * become available, and for the woken contexts to check to be
1059 * sure the appropriate page became available 1016 * sure the appropriate page became available, this saves space
1060 * at a cost of "thundering herd" phenomena d 1017 * at a cost of "thundering herd" phenomena during rare hash
1061 * collisions. 1018 * collisions.
1062 */ 1019 */
1063 #define PAGE_WAIT_TABLE_BITS 8 1020 #define PAGE_WAIT_TABLE_BITS 8
1064 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_ 1021 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1065 static wait_queue_head_t folio_wait_table[PAG !! 1022 static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1066 1023
1067 static wait_queue_head_t *folio_waitqueue(str !! 1024 static wait_queue_head_t *page_waitqueue(struct page *page)
1068 { 1025 {
1069 return &folio_wait_table[hash_ptr(fol !! 1026 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
1070 } 1027 }
1071 1028
1072 void __init pagecache_init(void) 1029 void __init pagecache_init(void)
1073 { 1030 {
1074 int i; 1031 int i;
1075 1032
1076 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; 1033 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1077 init_waitqueue_head(&folio_wa !! 1034 init_waitqueue_head(&page_wait_table[i]);
1078 1035
1079 page_writeback_init(); 1036 page_writeback_init();
1080 } 1037 }
1081 1038
1082 /* 1039 /*
1083 * The page wait code treats the "wait->flags 1040 * The page wait code treats the "wait->flags" somewhat unusually, because
1084 * we have multiple different kinds of waits, 1041 * we have multiple different kinds of waits, not just the usual "exclusive"
1085 * one. 1042 * one.
1086 * 1043 *
1087 * We have: 1044 * We have:
1088 * 1045 *
1089 * (a) no special bits set: 1046 * (a) no special bits set:
1090 * 1047 *
1091 * We're just waiting for the bit to be 1048 * We're just waiting for the bit to be released, and when a waker
1092 * calls the wakeup function, we set WQ_ 1049 * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1093 * and remove it from the wait queue. 1050 * and remove it from the wait queue.
1094 * 1051 *
1095 * Simple and straightforward. 1052 * Simple and straightforward.
1096 * 1053 *
1097 * (b) WQ_FLAG_EXCLUSIVE: 1054 * (b) WQ_FLAG_EXCLUSIVE:
1098 * 1055 *
1099 * The waiter is waiting to get the lock 1056 * The waiter is waiting to get the lock, and only one waiter should
1100 * be woken up to avoid any thundering h 1057 * be woken up to avoid any thundering herd behavior. We'll set the
1101 * WQ_FLAG_WOKEN bit, wake it up, and re 1058 * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1102 * 1059 *
1103 * This is the traditional exclusive wai 1060 * This is the traditional exclusive wait.
1104 * 1061 *
1105 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM: 1062 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1106 * 1063 *
1107 * The waiter is waiting to get the bit, 1064 * The waiter is waiting to get the bit, and additionally wants the
1108 * lock to be transferred to it for fair 1065 * lock to be transferred to it for fair lock behavior. If the lock
1109 * cannot be taken, we stop walking the 1066 * cannot be taken, we stop walking the wait queue without waking
1110 * the waiter. 1067 * the waiter.
1111 * 1068 *
1112 * This is the "fair lock handoff" case, 1069 * This is the "fair lock handoff" case, and in addition to setting
1113 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to 1070 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1114 * that it now has the lock. 1071 * that it now has the lock.
1115 */ 1072 */
1116 static int wake_page_function(wait_queue_entr 1073 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1117 { 1074 {
1118 unsigned int flags; 1075 unsigned int flags;
1119 struct wait_page_key *key = arg; 1076 struct wait_page_key *key = arg;
1120 struct wait_page_queue *wait_page 1077 struct wait_page_queue *wait_page
1121 = container_of(wait, struct w 1078 = container_of(wait, struct wait_page_queue, wait);
1122 1079
1123 if (!wake_page_match(wait_page, key)) 1080 if (!wake_page_match(wait_page, key))
1124 return 0; 1081 return 0;
1125 1082
1126 /* 1083 /*
1127 * If it's a lock handoff wait, we ge 1084 * If it's a lock handoff wait, we get the bit for it, and
1128 * stop walking (and do not wake it u 1085 * stop walking (and do not wake it up) if we can't.
1129 */ 1086 */
1130 flags = wait->flags; 1087 flags = wait->flags;
1131 if (flags & WQ_FLAG_EXCLUSIVE) { 1088 if (flags & WQ_FLAG_EXCLUSIVE) {
1132 if (test_bit(key->bit_nr, &ke !! 1089 if (test_bit(key->bit_nr, &key->page->flags))
1133 return -1; 1090 return -1;
1134 if (flags & WQ_FLAG_CUSTOM) { 1091 if (flags & WQ_FLAG_CUSTOM) {
1135 if (test_and_set_bit( !! 1092 if (test_and_set_bit(key->bit_nr, &key->page->flags))
1136 return -1; 1093 return -1;
1137 flags |= WQ_FLAG_DONE 1094 flags |= WQ_FLAG_DONE;
1138 } 1095 }
1139 } 1096 }
1140 1097
1141 /* 1098 /*
1142 * We are holding the wait-queue lock 1099 * We are holding the wait-queue lock, but the waiter that
1143 * is waiting for this will be checki 1100 * is waiting for this will be checking the flags without
1144 * any locking. 1101 * any locking.
1145 * 1102 *
1146 * So update the flags atomically, an 1103 * So update the flags atomically, and wake up the waiter
1147 * afterwards to avoid any races. Thi 1104 * afterwards to avoid any races. This store-release pairs
1148 * with the load-acquire in folio_wai !! 1105 * with the load-acquire in wait_on_page_bit_common().
1149 */ 1106 */
1150 smp_store_release(&wait->flags, flags 1107 smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1151 wake_up_state(wait->private, mode); 1108 wake_up_state(wait->private, mode);
1152 1109
1153 /* 1110 /*
1154 * Ok, we have successfully done what 1111 * Ok, we have successfully done what we're waiting for,
1155 * and we can unconditionally remove 1112 * and we can unconditionally remove the wait entry.
1156 * 1113 *
1157 * Note that this pairs with the "fin 1114 * Note that this pairs with the "finish_wait()" in the
1158 * waiter, and has to be the absolute 1115 * waiter, and has to be the absolute last thing we do.
1159 * After this list_del_init(&wait->en 1116 * After this list_del_init(&wait->entry) the wait entry
1160 * might be de-allocated and the proc 1117 * might be de-allocated and the process might even have
1161 * exited. 1118 * exited.
1162 */ 1119 */
1163 list_del_init_careful(&wait->entry); 1120 list_del_init_careful(&wait->entry);
1164 return (flags & WQ_FLAG_EXCLUSIVE) != 1121 return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1165 } 1122 }
1166 1123
1167 static void folio_wake_bit(struct folio *foli !! 1124 static void wake_up_page_bit(struct page *page, int bit_nr)
1168 { 1125 {
1169 wait_queue_head_t *q = folio_waitqueu !! 1126 wait_queue_head_t *q = page_waitqueue(page);
1170 struct wait_page_key key; 1127 struct wait_page_key key;
1171 unsigned long flags; 1128 unsigned long flags;
>> 1129 wait_queue_entry_t bookmark;
1172 1130
1173 key.folio = folio; !! 1131 key.page = page;
1174 key.bit_nr = bit_nr; 1132 key.bit_nr = bit_nr;
1175 key.page_match = 0; 1133 key.page_match = 0;
1176 1134
>> 1135 bookmark.flags = 0;
>> 1136 bookmark.private = NULL;
>> 1137 bookmark.func = NULL;
>> 1138 INIT_LIST_HEAD(&bookmark.entry);
>> 1139
1177 spin_lock_irqsave(&q->lock, flags); 1140 spin_lock_irqsave(&q->lock, flags);
1178 __wake_up_locked_key(q, TASK_NORMAL, !! 1141 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
>> 1142
>> 1143 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
>> 1144 /*
>> 1145 * Take a breather from holding the lock,
>> 1146 * allow pages that finish wake up asynchronously
>> 1147 * to acquire the lock and remove themselves
>> 1148 * from wait queue
>> 1149 */
>> 1150 spin_unlock_irqrestore(&q->lock, flags);
>> 1151 cpu_relax();
>> 1152 spin_lock_irqsave(&q->lock, flags);
>> 1153 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
>> 1154 }
1179 1155
1180 /* 1156 /*
1181 * It's possible to miss clearing wai !! 1157 * It is possible for other pages to have collided on the waitqueue
1182 * waiters, but the hashed waitqueue !! 1158 * hash, so in that case check for a page match. That prevents a long-
1183 * That's okay, it's a rare case. The !! 1159 * term waiter
1184 * 1160 *
1185 * Note that, depending on the page p !! 1161 * It is still possible to miss a case here, when we woke page waiters
1186 * other), the flag may be cleared in !! 1162 * and removed them from the waitqueue, but there are still other
1187 * but that is not required for corre !! 1163 * page waiters.
1188 */ !! 1164 */
1189 if (!waitqueue_active(q) || !key.page !! 1165 if (!waitqueue_active(q) || !key.page_match) {
1190 folio_clear_waiters(folio); !! 1166 ClearPageWaiters(page);
1191 !! 1167 /*
>> 1168 * It's possible to miss clearing Waiters here, when we woke
>> 1169 * our page waiters, but the hashed waitqueue has waiters for
>> 1170 * other pages on it.
>> 1171 *
>> 1172 * That's okay, it's a rare case. The next waker will clear it.
>> 1173 */
>> 1174 }
1192 spin_unlock_irqrestore(&q->lock, flag 1175 spin_unlock_irqrestore(&q->lock, flags);
1193 } 1176 }
1194 1177
>> 1178 static void wake_up_page(struct page *page, int bit)
>> 1179 {
>> 1180 if (!PageWaiters(page))
>> 1181 return;
>> 1182 wake_up_page_bit(page, bit);
>> 1183 }
>> 1184
1195 /* 1185 /*
1196 * A choice of three behaviors for folio_wait !! 1186 * A choice of three behaviors for wait_on_page_bit_common():
1197 */ 1187 */
1198 enum behavior { 1188 enum behavior {
1199 EXCLUSIVE, /* Hold ref to page a 1189 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1200 * __folio_lock() wai !! 1190 * __lock_page() waiting on then setting PG_locked.
1201 */ 1191 */
1202 SHARED, /* Hold ref to page a 1192 SHARED, /* Hold ref to page and check the bit when woken, like
1203 * folio_wait_writeba !! 1193 * wait_on_page_writeback() waiting on PG_writeback.
1204 */ 1194 */
1205 DROP, /* Drop ref to page b 1195 DROP, /* Drop ref to page before wait, no check when woken,
1206 * like folio_put_wai !! 1196 * like put_and_wait_on_page_locked() on PG_locked.
1207 */ 1197 */
1208 }; 1198 };
1209 1199
1210 /* 1200 /*
1211 * Attempt to check (or get) the folio flag, !! 1201 * Attempt to check (or get) the page bit, and mark us done
1212 * if successful. 1202 * if successful.
1213 */ 1203 */
1214 static inline bool folio_trylock_flag(struct !! 1204 static inline bool trylock_page_bit_common(struct page *page, int bit_nr,
1215 struc 1205 struct wait_queue_entry *wait)
1216 { 1206 {
1217 if (wait->flags & WQ_FLAG_EXCLUSIVE) 1207 if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1218 if (test_and_set_bit(bit_nr, !! 1208 if (test_and_set_bit(bit_nr, &page->flags))
1219 return false; 1209 return false;
1220 } else if (test_bit(bit_nr, &folio->f !! 1210 } else if (test_bit(bit_nr, &page->flags))
1221 return false; 1211 return false;
1222 1212
1223 wait->flags |= WQ_FLAG_WOKEN | WQ_FLA 1213 wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1224 return true; 1214 return true;
1225 } 1215 }
1226 1216
1227 /* How many times do we accept lock stealing 1217 /* How many times do we accept lock stealing from under a waiter? */
1228 int sysctl_page_lock_unfairness = 5; 1218 int sysctl_page_lock_unfairness = 5;
1229 1219
1230 static inline int folio_wait_bit_common(struc !! 1220 static inline int wait_on_page_bit_common(wait_queue_head_t *q,
1231 int state, enum behavior beha !! 1221 struct page *page, int bit_nr, int state, enum behavior behavior)
1232 { 1222 {
1233 wait_queue_head_t *q = folio_waitqueu <<
1234 int unfairness = sysctl_page_lock_unf 1223 int unfairness = sysctl_page_lock_unfairness;
1235 struct wait_page_queue wait_page; 1224 struct wait_page_queue wait_page;
1236 wait_queue_entry_t *wait = &wait_page 1225 wait_queue_entry_t *wait = &wait_page.wait;
1237 bool thrashing = false; 1226 bool thrashing = false;
>> 1227 bool delayacct = false;
1238 unsigned long pflags; 1228 unsigned long pflags;
1239 bool in_thrashing; <<
1240 1229
1241 if (bit_nr == PG_locked && 1230 if (bit_nr == PG_locked &&
1242 !folio_test_uptodate(folio) && fo !! 1231 !PageUptodate(page) && PageWorkingset(page)) {
1243 delayacct_thrashing_start(&in !! 1232 if (!PageSwapBacked(page)) {
>> 1233 delayacct_thrashing_start();
>> 1234 delayacct = true;
>> 1235 }
1244 psi_memstall_enter(&pflags); 1236 psi_memstall_enter(&pflags);
1245 thrashing = true; 1237 thrashing = true;
1246 } 1238 }
1247 1239
1248 init_wait(wait); 1240 init_wait(wait);
1249 wait->func = wake_page_function; 1241 wait->func = wake_page_function;
1250 wait_page.folio = folio; !! 1242 wait_page.page = page;
1251 wait_page.bit_nr = bit_nr; 1243 wait_page.bit_nr = bit_nr;
1252 1244
1253 repeat: 1245 repeat:
1254 wait->flags = 0; 1246 wait->flags = 0;
1255 if (behavior == EXCLUSIVE) { 1247 if (behavior == EXCLUSIVE) {
1256 wait->flags = WQ_FLAG_EXCLUSI 1248 wait->flags = WQ_FLAG_EXCLUSIVE;
1257 if (--unfairness < 0) 1249 if (--unfairness < 0)
1258 wait->flags |= WQ_FLA 1250 wait->flags |= WQ_FLAG_CUSTOM;
1259 } 1251 }
1260 1252
1261 /* 1253 /*
1262 * Do one last check whether we can g 1254 * Do one last check whether we can get the
1263 * page bit synchronously. 1255 * page bit synchronously.
1264 * 1256 *
1265 * Do the folio_set_waiters() marking !! 1257 * Do the SetPageWaiters() marking before that
1266 * to let any waker we _just_ missed 1258 * to let any waker we _just_ missed know they
1267 * need to wake us up (otherwise they 1259 * need to wake us up (otherwise they'll never
1268 * even go to the slow case that look 1260 * even go to the slow case that looks at the
1269 * page queue), and add ourselves to 1261 * page queue), and add ourselves to the wait
1270 * queue if we need to sleep. 1262 * queue if we need to sleep.
1271 * 1263 *
1272 * This part needs to be done under t 1264 * This part needs to be done under the queue
1273 * lock to avoid races. 1265 * lock to avoid races.
1274 */ 1266 */
1275 spin_lock_irq(&q->lock); 1267 spin_lock_irq(&q->lock);
1276 folio_set_waiters(folio); !! 1268 SetPageWaiters(page);
1277 if (!folio_trylock_flag(folio, bit_nr !! 1269 if (!trylock_page_bit_common(page, bit_nr, wait))
1278 __add_wait_queue_entry_tail(q 1270 __add_wait_queue_entry_tail(q, wait);
1279 spin_unlock_irq(&q->lock); 1271 spin_unlock_irq(&q->lock);
1280 1272
1281 /* 1273 /*
1282 * From now on, all the logic will be 1274 * From now on, all the logic will be based on
1283 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE 1275 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1284 * see whether the page bit testing h 1276 * see whether the page bit testing has already
1285 * been done by the wake function. 1277 * been done by the wake function.
1286 * 1278 *
1287 * We can drop our reference to the f !! 1279 * We can drop our reference to the page.
1288 */ 1280 */
1289 if (behavior == DROP) 1281 if (behavior == DROP)
1290 folio_put(folio); !! 1282 put_page(page);
1291 1283
1292 /* 1284 /*
1293 * Note that until the "finish_wait() 1285 * Note that until the "finish_wait()", or until
1294 * we see the WQ_FLAG_WOKEN flag, we 1286 * we see the WQ_FLAG_WOKEN flag, we need to
1295 * be very careful with the 'wait->fl 1287 * be very careful with the 'wait->flags', because
1296 * we may race with a waker that sets 1288 * we may race with a waker that sets them.
1297 */ 1289 */
1298 for (;;) { 1290 for (;;) {
1299 unsigned int flags; 1291 unsigned int flags;
1300 1292
1301 set_current_state(state); 1293 set_current_state(state);
1302 1294
1303 /* Loop until we've been woke 1295 /* Loop until we've been woken or interrupted */
1304 flags = smp_load_acquire(&wai 1296 flags = smp_load_acquire(&wait->flags);
1305 if (!(flags & WQ_FLAG_WOKEN)) 1297 if (!(flags & WQ_FLAG_WOKEN)) {
1306 if (signal_pending_st 1298 if (signal_pending_state(state, current))
1307 break; 1299 break;
1308 1300
1309 io_schedule(); 1301 io_schedule();
1310 continue; 1302 continue;
1311 } 1303 }
1312 1304
1313 /* If we were non-exclusive, 1305 /* If we were non-exclusive, we're done */
1314 if (behavior != EXCLUSIVE) 1306 if (behavior != EXCLUSIVE)
1315 break; 1307 break;
1316 1308
1317 /* If the waker got the lock 1309 /* If the waker got the lock for us, we're done */
1318 if (flags & WQ_FLAG_DONE) 1310 if (flags & WQ_FLAG_DONE)
1319 break; 1311 break;
1320 1312
1321 /* 1313 /*
1322 * Otherwise, if we're gettin 1314 * Otherwise, if we're getting the lock, we need to
1323 * try to get it ourselves. 1315 * try to get it ourselves.
1324 * 1316 *
1325 * And if that fails, we'll h 1317 * And if that fails, we'll have to retry this all.
1326 */ 1318 */
1327 if (unlikely(test_and_set_bit !! 1319 if (unlikely(test_and_set_bit(bit_nr, &page->flags)))
1328 goto repeat; 1320 goto repeat;
1329 1321
1330 wait->flags |= WQ_FLAG_DONE; 1322 wait->flags |= WQ_FLAG_DONE;
1331 break; 1323 break;
1332 } 1324 }
1333 1325
1334 /* 1326 /*
1335 * If a signal happened, this 'finish 1327 * If a signal happened, this 'finish_wait()' may remove the last
1336 * waiter from the wait-queues, but t !! 1328 * waiter from the wait-queues, but the PageWaiters bit will remain
1337 * set. That's ok. The next wakeup wi 1329 * set. That's ok. The next wakeup will take care of it, and trying
1338 * to do it here would be difficult a 1330 * to do it here would be difficult and prone to races.
1339 */ 1331 */
1340 finish_wait(q, wait); 1332 finish_wait(q, wait);
1341 1333
1342 if (thrashing) { 1334 if (thrashing) {
1343 delayacct_thrashing_end(&in_t !! 1335 if (delayacct)
>> 1336 delayacct_thrashing_end();
1344 psi_memstall_leave(&pflags); 1337 psi_memstall_leave(&pflags);
1345 } 1338 }
1346 1339
1347 /* 1340 /*
1348 * NOTE! The wait->flags weren't stab 1341 * NOTE! The wait->flags weren't stable until we've done the
1349 * 'finish_wait()', and we could have 1342 * 'finish_wait()', and we could have exited the loop above due
1350 * to a signal, and had a wakeup even 1343 * to a signal, and had a wakeup event happen after the signal
1351 * test but before the 'finish_wait() 1344 * test but before the 'finish_wait()'.
1352 * 1345 *
1353 * So only after the finish_wait() ca 1346 * So only after the finish_wait() can we reliably determine
1354 * if we got woken up or not, so we c 1347 * if we got woken up or not, so we can now figure out the final
1355 * return value based on that state w 1348 * return value based on that state without races.
1356 * 1349 *
1357 * Also note that WQ_FLAG_WOKEN is su 1350 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1358 * waiter, but an exclusive one requi 1351 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1359 */ 1352 */
1360 if (behavior == EXCLUSIVE) 1353 if (behavior == EXCLUSIVE)
1361 return wait->flags & WQ_FLAG_ 1354 return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1362 1355
1363 return wait->flags & WQ_FLAG_WOKEN ? 1356 return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1364 } 1357 }
1365 1358
1366 #ifdef CONFIG_MIGRATION !! 1359 void wait_on_page_bit(struct page *page, int bit_nr)
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 } <<
1444 #endif <<
1445 <<
1446 void folio_wait_bit(struct folio *folio, int <<
1447 { 1360 {
1448 folio_wait_bit_common(folio, bit_nr, !! 1361 wait_queue_head_t *q = page_waitqueue(page);
>> 1362 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1449 } 1363 }
1450 EXPORT_SYMBOL(folio_wait_bit); !! 1364 EXPORT_SYMBOL(wait_on_page_bit);
1451 1365
1452 int folio_wait_bit_killable(struct folio *fol !! 1366 int wait_on_page_bit_killable(struct page *page, int bit_nr)
1453 { 1367 {
1454 return folio_wait_bit_common(folio, b !! 1368 wait_queue_head_t *q = page_waitqueue(page);
>> 1369 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED);
1455 } 1370 }
1456 EXPORT_SYMBOL(folio_wait_bit_killable); !! 1371 EXPORT_SYMBOL(wait_on_page_bit_killable);
1457 1372
1458 /** 1373 /**
1459 * folio_put_wait_locked - Drop a reference a !! 1374 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1460 * @folio: The folio to wait for. !! 1375 * @page: The page to wait for.
1461 * @state: The sleep state (TASK_KILLABLE, TA 1376 * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1462 * 1377 *
1463 * The caller should hold a reference on @fol !! 1378 * The caller should hold a reference on @page. They expect the page to
1464 * become unlocked relatively soon, but do no 1379 * become unlocked relatively soon, but do not wish to hold up migration
1465 * (for example) by holding the reference whi !! 1380 * (for example) by holding the reference while waiting for the page to
1466 * come unlocked. After this function return 1381 * come unlocked. After this function returns, the caller should not
1467 * dereference @folio. !! 1382 * dereference @page.
1468 * 1383 *
1469 * Return: 0 if the folio was unlocked or -EI !! 1384 * Return: 0 if the page was unlocked or -EINTR if interrupted by a signal.
1470 */ 1385 */
1471 static int folio_put_wait_locked(struct folio !! 1386 int put_and_wait_on_page_locked(struct page *page, int state)
1472 { 1387 {
1473 return folio_wait_bit_common(folio, P !! 1388 wait_queue_head_t *q;
>> 1389
>> 1390 page = compound_head(page);
>> 1391 q = page_waitqueue(page);
>> 1392 return wait_on_page_bit_common(q, page, PG_locked, state, DROP);
1474 } 1393 }
1475 1394
1476 /** 1395 /**
1477 * folio_add_wait_queue - Add an arbitrary wa !! 1396 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1478 * @folio: Folio defining the wait queue of i !! 1397 * @page: Page defining the wait queue of interest
1479 * @waiter: Waiter to add to the queue 1398 * @waiter: Waiter to add to the queue
1480 * 1399 *
1481 * Add an arbitrary @waiter to the wait queue !! 1400 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1482 */ 1401 */
1483 void folio_add_wait_queue(struct folio *folio !! 1402 void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1484 { 1403 {
1485 wait_queue_head_t *q = folio_waitqueu !! 1404 wait_queue_head_t *q = page_waitqueue(page);
1486 unsigned long flags; 1405 unsigned long flags;
1487 1406
1488 spin_lock_irqsave(&q->lock, flags); 1407 spin_lock_irqsave(&q->lock, flags);
1489 __add_wait_queue_entry_tail(q, waiter 1408 __add_wait_queue_entry_tail(q, waiter);
1490 folio_set_waiters(folio); !! 1409 SetPageWaiters(page);
1491 spin_unlock_irqrestore(&q->lock, flag 1410 spin_unlock_irqrestore(&q->lock, flags);
1492 } 1411 }
1493 EXPORT_SYMBOL_GPL(folio_add_wait_queue); !! 1412 EXPORT_SYMBOL_GPL(add_page_wait_queue);
1494 1413
1495 /** !! 1414 #ifndef clear_bit_unlock_is_negative_byte
1496 * folio_unlock - Unlock a locked folio. !! 1415
1497 * @folio: The folio. !! 1416 /*
>> 1417 * PG_waiters is the high bit in the same byte as PG_lock.
1498 * 1418 *
1499 * Unlocks the folio and wakes up any thread !! 1419 * On x86 (and on many other architectures), we can clear PG_lock and
>> 1420 * test the sign bit at the same time. But if the architecture does
>> 1421 * not support that special operation, we just do this all by hand
>> 1422 * instead.
1500 * 1423 *
1501 * Context: May be called from interrupt or p !! 1424 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1502 * called from NMI context. !! 1425 * being cleared, but a memory barrier should be unnecessary since it is
>> 1426 * in the same byte as PG_locked.
1503 */ 1427 */
1504 void folio_unlock(struct folio *folio) !! 1428 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1505 { 1429 {
1506 /* Bit 7 allows x86 to check the byte !! 1430 clear_bit_unlock(nr, mem);
1507 BUILD_BUG_ON(PG_waiters != 7); !! 1431 /* smp_mb__after_atomic(); */
1508 BUILD_BUG_ON(PG_locked > 7); !! 1432 return test_bit(PG_waiters, mem);
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 } 1433 }
1513 EXPORT_SYMBOL(folio_unlock); !! 1434
>> 1435 #endif
1514 1436
1515 /** 1437 /**
1516 * folio_end_read - End read on a folio. !! 1438 * unlock_page - unlock a locked page
1517 * @folio: The folio. !! 1439 * @page: the page
1518 * @success: True if all reads completed succ !! 1440 *
1519 * !! 1441 * Unlocks the page and wakes up sleepers in wait_on_page_locked().
1520 * When all reads against a folio have comple !! 1442 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1521 * call this function to let the pagecache kn !! 1443 * mechanism between PageLocked pages and PageWriteback pages is shared.
1522 * are outstanding. This will unlock the fol !! 1444 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1523 * sleeping on the lock. The folio will also !! 1445 *
1524 * reads succeeded. !! 1446 * Note that this depends on PG_waiters being the sign bit in the byte
1525 * !! 1447 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1526 * Context: May be called from interrupt or p !! 1448 * clear the PG_locked bit and test PG_waiters at the same time fairly
1527 * called from NMI context. !! 1449 * portably (architectures that do LL/SC can test any bit, while x86 can
1528 */ !! 1450 * test the sign bit).
1529 void folio_end_read(struct folio *folio, bool !! 1451 */
1530 { !! 1452 void unlock_page(struct page *page)
1531 unsigned long mask = 1 << PG_locked; !! 1453 {
1532 !! 1454 BUILD_BUG_ON(PG_waiters != 7);
1533 /* Must be in bottom byte for x86 to !! 1455 page = compound_head(page);
1534 BUILD_BUG_ON(PG_uptodate > 7); !! 1456 VM_BUG_ON_PAGE(!PageLocked(page), page);
1535 VM_BUG_ON_FOLIO(!folio_test_locked(fo !! 1457 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1536 VM_BUG_ON_FOLIO(folio_test_uptodate(f !! 1458 wake_up_page_bit(page, PG_locked);
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 } 1459 }
1543 EXPORT_SYMBOL(folio_end_read); !! 1460 EXPORT_SYMBOL(unlock_page);
1544 1461
1545 /** 1462 /**
1546 * folio_end_private_2 - Clear PG_private_2 a !! 1463 * end_page_private_2 - Clear PG_private_2 and release any waiters
1547 * @folio: The folio. !! 1464 * @page: The page
1548 * 1465 *
1549 * Clear the PG_private_2 bit on a folio and !! 1466 * Clear the PG_private_2 bit on a page and wake up any sleepers waiting for
1550 * it. The folio reference held for PG_priva !! 1467 * this. The page ref held for PG_private_2 being set is released.
1551 * 1468 *
1552 * This is, for example, used when a netfs fo !! 1469 * This is, for example, used when a netfs page is being written to a local
1553 * disk cache, thereby allowing writes to the !! 1470 * disk cache, thereby allowing writes to the cache for the same page to be
1554 * serialised. 1471 * serialised.
1555 */ 1472 */
1556 void folio_end_private_2(struct folio *folio) !! 1473 void end_page_private_2(struct page *page)
1557 { 1474 {
1558 VM_BUG_ON_FOLIO(!folio_test_private_2 !! 1475 page = compound_head(page);
1559 clear_bit_unlock(PG_private_2, folio_ !! 1476 VM_BUG_ON_PAGE(!PagePrivate2(page), page);
1560 folio_wake_bit(folio, PG_private_2); !! 1477 clear_bit_unlock(PG_private_2, &page->flags);
1561 folio_put(folio); !! 1478 wake_up_page_bit(page, PG_private_2);
>> 1479 put_page(page);
1562 } 1480 }
1563 EXPORT_SYMBOL(folio_end_private_2); !! 1481 EXPORT_SYMBOL(end_page_private_2);
1564 1482
1565 /** 1483 /**
1566 * folio_wait_private_2 - Wait for PG_private !! 1484 * wait_on_page_private_2 - Wait for PG_private_2 to be cleared on a page
1567 * @folio: The folio to wait on. !! 1485 * @page: The page to wait on
1568 * 1486 *
1569 * Wait for PG_private_2 to be cleared on a f !! 1487 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a page.
1570 */ 1488 */
1571 void folio_wait_private_2(struct folio *folio !! 1489 void wait_on_page_private_2(struct page *page)
1572 { 1490 {
1573 while (folio_test_private_2(folio)) !! 1491 page = compound_head(page);
1574 folio_wait_bit(folio, PG_priv !! 1492 while (PagePrivate2(page))
>> 1493 wait_on_page_bit(page, PG_private_2);
1575 } 1494 }
1576 EXPORT_SYMBOL(folio_wait_private_2); !! 1495 EXPORT_SYMBOL(wait_on_page_private_2);
1577 1496
1578 /** 1497 /**
1579 * folio_wait_private_2_killable - Wait for P !! 1498 * wait_on_page_private_2_killable - Wait for PG_private_2 to be cleared on a page
1580 * @folio: The folio to wait on. !! 1499 * @page: The page to wait on
1581 * 1500 *
1582 * Wait for PG_private_2 to be cleared on a f !! 1501 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a page or until a
1583 * received by the calling task. !! 1502 * fatal signal is received by the calling task.
1584 * 1503 *
1585 * Return: 1504 * Return:
1586 * - 0 if successful. 1505 * - 0 if successful.
1587 * - -EINTR if a fatal signal was encountered 1506 * - -EINTR if a fatal signal was encountered.
1588 */ 1507 */
1589 int folio_wait_private_2_killable(struct foli !! 1508 int wait_on_page_private_2_killable(struct page *page)
1590 { 1509 {
1591 int ret = 0; 1510 int ret = 0;
1592 1511
1593 while (folio_test_private_2(folio)) { !! 1512 page = compound_head(page);
1594 ret = folio_wait_bit_killable !! 1513 while (PagePrivate2(page)) {
>> 1514 ret = wait_on_page_bit_killable(page, PG_private_2);
1595 if (ret < 0) 1515 if (ret < 0)
1596 break; 1516 break;
1597 } 1517 }
1598 1518
1599 return ret; 1519 return ret;
1600 } 1520 }
1601 EXPORT_SYMBOL(folio_wait_private_2_killable); !! 1521 EXPORT_SYMBOL(wait_on_page_private_2_killable);
1602 1522
1603 /** 1523 /**
1604 * folio_end_writeback - End writeback agains !! 1524 * end_page_writeback - end writeback against a page
1605 * @folio: The folio. !! 1525 * @page: the page
1606 * <<
1607 * The folio must actually be under writeback <<
1608 * <<
1609 * Context: May be called from process or int <<
1610 */ 1526 */
1611 void folio_end_writeback(struct folio *folio) !! 1527 void end_page_writeback(struct page *page)
1612 { 1528 {
1613 VM_BUG_ON_FOLIO(!folio_test_writeback <<
1614 <<
1615 /* 1529 /*
1616 * folio_test_clear_reclaim() could b !! 1530 * TestClearPageReclaim could be used here but it is an atomic
1617 * atomic operation and overkill in t !! 1531 * operation and overkill in this particular case. Failing to
1618 * to shuffle a folio marked for imme !! 1532 * shuffle a page marked for immediate reclaim is too mild to
1619 * a gain to justify taking an atomic !! 1533 * justify taking an atomic operation penalty at the end of
1620 * end of every folio writeback. !! 1534 * ever page writeback.
1621 */ 1535 */
1622 if (folio_test_reclaim(folio)) { !! 1536 if (PageReclaim(page)) {
1623 folio_clear_reclaim(folio); !! 1537 ClearPageReclaim(page);
1624 folio_rotate_reclaimable(foli !! 1538 rotate_reclaimable_page(page);
1625 } 1539 }
1626 1540
1627 /* 1541 /*
1628 * Writeback does not hold a folio re !! 1542 * Writeback does not hold a page reference of its own, relying
1629 * on truncation to wait for the clea 1543 * on truncation to wait for the clearing of PG_writeback.
1630 * But here we must make sure that th !! 1544 * But here we must make sure that the page is not freed and
1631 * reused before the folio_wake_bit() !! 1545 * reused before the wake_up_page().
1632 */ 1546 */
1633 folio_get(folio); !! 1547 get_page(page);
1634 if (__folio_end_writeback(folio)) !! 1548 if (!test_clear_page_writeback(page))
1635 folio_wake_bit(folio, PG_writ !! 1549 BUG();
1636 acct_reclaim_writeback(folio); !! 1550
1637 folio_put(folio); !! 1551 smp_mb__after_atomic();
>> 1552 wake_up_page(page, PG_writeback);
>> 1553 put_page(page);
>> 1554 }
>> 1555 EXPORT_SYMBOL(end_page_writeback);
>> 1556
>> 1557 /*
>> 1558 * After completing I/O on a page, call this routine to update the page
>> 1559 * flags appropriately
>> 1560 */
>> 1561 void page_endio(struct page *page, bool is_write, int err)
>> 1562 {
>> 1563 if (!is_write) {
>> 1564 if (!err) {
>> 1565 SetPageUptodate(page);
>> 1566 } else {
>> 1567 ClearPageUptodate(page);
>> 1568 SetPageError(page);
>> 1569 }
>> 1570 unlock_page(page);
>> 1571 } else {
>> 1572 if (err) {
>> 1573 struct address_space *mapping;
>> 1574
>> 1575 SetPageError(page);
>> 1576 mapping = page_mapping(page);
>> 1577 if (mapping)
>> 1578 mapping_set_error(mapping, err);
>> 1579 }
>> 1580 end_page_writeback(page);
>> 1581 }
1638 } 1582 }
1639 EXPORT_SYMBOL(folio_end_writeback); !! 1583 EXPORT_SYMBOL_GPL(page_endio);
1640 1584
1641 /** 1585 /**
1642 * __folio_lock - Get a lock on the folio, as !! 1586 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1643 * @folio: The folio to lock !! 1587 * @__page: the page to lock
1644 */ 1588 */
1645 void __folio_lock(struct folio *folio) !! 1589 void __lock_page(struct page *__page)
1646 { 1590 {
1647 folio_wait_bit_common(folio, PG_locke !! 1591 struct page *page = compound_head(__page);
>> 1592 wait_queue_head_t *q = page_waitqueue(page);
>> 1593 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE,
1648 EXCLUSIVE); 1594 EXCLUSIVE);
1649 } 1595 }
1650 EXPORT_SYMBOL(__folio_lock); !! 1596 EXPORT_SYMBOL(__lock_page);
1651 1597
1652 int __folio_lock_killable(struct folio *folio !! 1598 int __lock_page_killable(struct page *__page)
1653 { 1599 {
1654 return folio_wait_bit_common(folio, P !! 1600 struct page *page = compound_head(__page);
>> 1601 wait_queue_head_t *q = page_waitqueue(page);
>> 1602 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE,
1655 EXCLU 1603 EXCLUSIVE);
1656 } 1604 }
1657 EXPORT_SYMBOL_GPL(__folio_lock_killable); !! 1605 EXPORT_SYMBOL_GPL(__lock_page_killable);
1658 1606
1659 static int __folio_lock_async(struct folio *f !! 1607 int __lock_page_async(struct page *page, struct wait_page_queue *wait)
1660 { 1608 {
1661 struct wait_queue_head *q = folio_wai !! 1609 struct wait_queue_head *q = page_waitqueue(page);
1662 int ret; !! 1610 int ret = 0;
1663 1611
1664 wait->folio = folio; !! 1612 wait->page = page;
1665 wait->bit_nr = PG_locked; 1613 wait->bit_nr = PG_locked;
1666 1614
1667 spin_lock_irq(&q->lock); 1615 spin_lock_irq(&q->lock);
1668 __add_wait_queue_entry_tail(q, &wait- 1616 __add_wait_queue_entry_tail(q, &wait->wait);
1669 folio_set_waiters(folio); !! 1617 SetPageWaiters(page);
1670 ret = !folio_trylock(folio); !! 1618 ret = !trylock_page(page);
1671 /* 1619 /*
1672 * If we were successful now, we know 1620 * If we were successful now, we know we're still on the
1673 * waitqueue as we're still under the 1621 * waitqueue as we're still under the lock. This means it's
1674 * safe to remove and return success, 1622 * safe to remove and return success, we know the callback
1675 * isn't going to trigger. 1623 * isn't going to trigger.
1676 */ 1624 */
1677 if (!ret) 1625 if (!ret)
1678 __remove_wait_queue(q, &wait- 1626 __remove_wait_queue(q, &wait->wait);
1679 else 1627 else
1680 ret = -EIOCBQUEUED; 1628 ret = -EIOCBQUEUED;
1681 spin_unlock_irq(&q->lock); 1629 spin_unlock_irq(&q->lock);
1682 return ret; 1630 return ret;
1683 } 1631 }
1684 1632
1685 /* 1633 /*
1686 * Return values: 1634 * Return values:
1687 * 0 - folio is locked. !! 1635 * 1 - page is locked; mmap_lock is still held.
1688 * non-zero - folio is not locked. !! 1636 * 0 - page is not locked.
1689 * mmap_lock or per-VMA lock has been rel !! 1637 * mmap_lock has been released (mmap_read_unlock(), unless flags had both
1690 * vma_end_read()), unless flags had both !! 1638 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1691 * FAULT_FLAG_RETRY_NOWAIT set, in which !! 1639 * which case mmap_lock is still held.
1692 * 1640 *
1693 * If neither ALLOW_RETRY nor KILLABLE are se !! 1641 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1694 * with the folio locked and the mmap_lock/pe !! 1642 * with the page locked and the mmap_lock unperturbed.
1695 */ 1643 */
1696 vm_fault_t __folio_lock_or_retry(struct folio !! 1644 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
>> 1645 unsigned int flags)
1697 { 1646 {
1698 unsigned int flags = vmf->flags; <<
1699 <<
1700 if (fault_flag_allow_retry_first(flag 1647 if (fault_flag_allow_retry_first(flags)) {
1701 /* 1648 /*
1702 * CAUTION! In this case, mma !! 1649 * CAUTION! In this case, mmap_lock is not released
1703 * released even though retur !! 1650 * even though return 0.
1704 */ 1651 */
1705 if (flags & FAULT_FLAG_RETRY_ 1652 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1706 return VM_FAULT_RETRY !! 1653 return 0;
1707 1654
1708 release_fault_lock(vmf); !! 1655 mmap_read_unlock(mm);
1709 if (flags & FAULT_FLAG_KILLAB 1656 if (flags & FAULT_FLAG_KILLABLE)
1710 folio_wait_locked_kil !! 1657 wait_on_page_locked_killable(page);
1711 else 1658 else
1712 folio_wait_locked(fol !! 1659 wait_on_page_locked(page);
1713 return VM_FAULT_RETRY; !! 1660 return 0;
1714 } 1661 }
1715 if (flags & FAULT_FLAG_KILLABLE) { 1662 if (flags & FAULT_FLAG_KILLABLE) {
1716 bool ret; !! 1663 int ret;
1717 1664
1718 ret = __folio_lock_killable(f !! 1665 ret = __lock_page_killable(page);
1719 if (ret) { 1666 if (ret) {
1720 release_fault_lock(vm !! 1667 mmap_read_unlock(mm);
1721 return VM_FAULT_RETRY !! 1668 return 0;
1722 } 1669 }
1723 } else { 1670 } else {
1724 __folio_lock(folio); !! 1671 __lock_page(page);
1725 } 1672 }
>> 1673 return 1;
1726 1674
1727 return 0; <<
1728 } 1675 }
1729 1676
1730 /** 1677 /**
1731 * page_cache_next_miss() - Find the next gap 1678 * page_cache_next_miss() - Find the next gap in the page cache.
1732 * @mapping: Mapping. 1679 * @mapping: Mapping.
1733 * @index: Index. 1680 * @index: Index.
1734 * @max_scan: Maximum range to search. 1681 * @max_scan: Maximum range to search.
1735 * 1682 *
1736 * Search the range [index, min(index + max_s 1683 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1737 * gap with the lowest index. 1684 * gap with the lowest index.
1738 * 1685 *
1739 * This function may be called under the rcu_ 1686 * This function may be called under the rcu_read_lock. However, this will
1740 * not atomically search a snapshot of the ca 1687 * not atomically search a snapshot of the cache at a single point in time.
1741 * For example, if a gap is created at index 1688 * For example, if a gap is created at index 5, then subsequently a gap is
1742 * created at index 10, page_cache_next_miss 1689 * created at index 10, page_cache_next_miss covering both indices may
1743 * return 10 if called under the rcu_read_loc 1690 * return 10 if called under the rcu_read_lock.
1744 * 1691 *
1745 * Return: The index of the gap if found, oth 1692 * Return: The index of the gap if found, otherwise an index outside the
1746 * range specified (in which case 'return - i 1693 * range specified (in which case 'return - index >= max_scan' will be true).
1747 * In the rare case of index wrap-around, 0 w 1694 * In the rare case of index wrap-around, 0 will be returned.
1748 */ 1695 */
1749 pgoff_t page_cache_next_miss(struct address_s 1696 pgoff_t page_cache_next_miss(struct address_space *mapping,
1750 pgoff_t index, u 1697 pgoff_t index, unsigned long max_scan)
1751 { 1698 {
1752 XA_STATE(xas, &mapping->i_pages, inde 1699 XA_STATE(xas, &mapping->i_pages, index);
1753 1700
1754 while (max_scan--) { 1701 while (max_scan--) {
1755 void *entry = xas_next(&xas); 1702 void *entry = xas_next(&xas);
1756 if (!entry || xa_is_value(ent 1703 if (!entry || xa_is_value(entry))
1757 return xas.xa_index; !! 1704 break;
1758 if (xas.xa_index == 0) 1705 if (xas.xa_index == 0)
1759 return 0; !! 1706 break;
1760 } 1707 }
1761 1708
1762 return index + max_scan; !! 1709 return xas.xa_index;
1763 } 1710 }
1764 EXPORT_SYMBOL(page_cache_next_miss); 1711 EXPORT_SYMBOL(page_cache_next_miss);
1765 1712
1766 /** 1713 /**
1767 * page_cache_prev_miss() - Find the previous 1714 * page_cache_prev_miss() - Find the previous gap in the page cache.
1768 * @mapping: Mapping. 1715 * @mapping: Mapping.
1769 * @index: Index. 1716 * @index: Index.
1770 * @max_scan: Maximum range to search. 1717 * @max_scan: Maximum range to search.
1771 * 1718 *
1772 * Search the range [max(index - max_scan + 1 1719 * Search the range [max(index - max_scan + 1, 0), index] for the
1773 * gap with the highest index. 1720 * gap with the highest index.
1774 * 1721 *
1775 * This function may be called under the rcu_ 1722 * This function may be called under the rcu_read_lock. However, this will
1776 * not atomically search a snapshot of the ca 1723 * not atomically search a snapshot of the cache at a single point in time.
1777 * For example, if a gap is created at index 1724 * For example, if a gap is created at index 10, then subsequently a gap is
1778 * created at index 5, page_cache_prev_miss() 1725 * created at index 5, page_cache_prev_miss() covering both indices may
1779 * return 5 if called under the rcu_read_lock 1726 * return 5 if called under the rcu_read_lock.
1780 * 1727 *
1781 * Return: The index of the gap if found, oth 1728 * Return: The index of the gap if found, otherwise an index outside the
1782 * range specified (in which case 'index - re 1729 * range specified (in which case 'index - return >= max_scan' will be true).
1783 * In the rare case of wrap-around, ULONG_MAX 1730 * In the rare case of wrap-around, ULONG_MAX will be returned.
1784 */ 1731 */
1785 pgoff_t page_cache_prev_miss(struct address_s 1732 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1786 pgoff_t index, u 1733 pgoff_t index, unsigned long max_scan)
1787 { 1734 {
1788 XA_STATE(xas, &mapping->i_pages, inde 1735 XA_STATE(xas, &mapping->i_pages, index);
1789 1736
1790 while (max_scan--) { 1737 while (max_scan--) {
1791 void *entry = xas_prev(&xas); 1738 void *entry = xas_prev(&xas);
1792 if (!entry || xa_is_value(ent 1739 if (!entry || xa_is_value(entry))
1793 break; 1740 break;
1794 if (xas.xa_index == ULONG_MAX 1741 if (xas.xa_index == ULONG_MAX)
1795 break; 1742 break;
1796 } 1743 }
1797 1744
1798 return xas.xa_index; 1745 return xas.xa_index;
1799 } 1746 }
1800 EXPORT_SYMBOL(page_cache_prev_miss); 1747 EXPORT_SYMBOL(page_cache_prev_miss);
1801 1748
1802 /* 1749 /*
1803 * Lockless page cache protocol: !! 1750 * mapping_get_entry - 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 1751 * @mapping: the address_space to search
1825 * @index: The page cache index. 1752 * @index: The page cache index.
1826 * 1753 *
1827 * Looks up the page cache entry at @mapping !! 1754 * Looks up the page cache slot at @mapping & @index. If there is a
1828 * it is returned with an increased refcount. !! 1755 * page cache page, the head page is returned with an increased refcount.
1829 * of a previously evicted folio, or a swap e <<
1830 * it is returned without further action. <<
1831 * 1756 *
1832 * Return: The folio, swap or shadow entry, % !! 1757 * If the slot holds a shadow entry of a previously evicted page, or a
>> 1758 * swap entry from shmem/tmpfs, it is returned.
>> 1759 *
>> 1760 * Return: The head page or shadow entry, %NULL if nothing is found.
1833 */ 1761 */
1834 void *filemap_get_entry(struct address_space !! 1762 static struct page *mapping_get_entry(struct address_space *mapping,
>> 1763 pgoff_t index)
1835 { 1764 {
1836 XA_STATE(xas, &mapping->i_pages, inde 1765 XA_STATE(xas, &mapping->i_pages, index);
1837 struct folio *folio; !! 1766 struct page *page;
1838 1767
1839 rcu_read_lock(); 1768 rcu_read_lock();
1840 repeat: 1769 repeat:
1841 xas_reset(&xas); 1770 xas_reset(&xas);
1842 folio = xas_load(&xas); !! 1771 page = xas_load(&xas);
1843 if (xas_retry(&xas, folio)) !! 1772 if (xas_retry(&xas, page))
1844 goto repeat; 1773 goto repeat;
1845 /* 1774 /*
1846 * A shadow entry of a recently evict 1775 * A shadow entry of a recently evicted page, or a swap entry from
1847 * shmem/tmpfs. Return it without at 1776 * shmem/tmpfs. Return it without attempting to raise page count.
1848 */ 1777 */
1849 if (!folio || xa_is_value(folio)) !! 1778 if (!page || xa_is_value(page))
1850 goto out; 1779 goto out;
1851 1780
1852 if (!folio_try_get(folio)) !! 1781 if (!page_cache_get_speculative(page))
1853 goto repeat; 1782 goto repeat;
1854 1783
1855 if (unlikely(folio != xas_reload(&xas !! 1784 /*
1856 folio_put(folio); !! 1785 * Has the page moved or been split?
>> 1786 * This is part of the lockless pagecache protocol. See
>> 1787 * include/linux/pagemap.h for details.
>> 1788 */
>> 1789 if (unlikely(page != xas_reload(&xas))) {
>> 1790 put_page(page);
1857 goto repeat; 1791 goto repeat;
1858 } 1792 }
1859 out: 1793 out:
1860 rcu_read_unlock(); 1794 rcu_read_unlock();
1861 1795
1862 return folio; !! 1796 return page;
1863 } 1797 }
1864 1798
1865 /** 1799 /**
1866 * __filemap_get_folio - Find and get a refer !! 1800 * pagecache_get_page - Find and get a reference to a page.
1867 * @mapping: The address_space to search. 1801 * @mapping: The address_space to search.
1868 * @index: The page index. 1802 * @index: The page index.
1869 * @fgp_flags: %FGP flags modify how the foli !! 1803 * @fgp_flags: %FGP flags modify how the page is returned.
1870 * @gfp: Memory allocation flags to use if %F !! 1804 * @gfp_mask: Memory allocation flags to use if %FGP_CREAT is specified.
1871 * 1805 *
1872 * Looks up the page cache entry at @mapping 1806 * Looks up the page cache entry at @mapping & @index.
1873 * 1807 *
>> 1808 * @fgp_flags can be zero or more of these flags:
>> 1809 *
>> 1810 * * %FGP_ACCESSED - The page will be marked accessed.
>> 1811 * * %FGP_LOCK - The page is returned locked.
>> 1812 * * %FGP_HEAD - If the page is present and a THP, return the head page
>> 1813 * rather than the exact page specified by the index.
>> 1814 * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it
>> 1815 * instead of allocating a new page to replace it.
>> 1816 * * %FGP_CREAT - If no page is present then a new page is allocated using
>> 1817 * @gfp_mask and added to the page cache and the VM's LRU list.
>> 1818 * The page is returned locked and with an increased refcount.
>> 1819 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
>> 1820 * page is already in cache. If the page was allocated, unlock it before
>> 1821 * returning so the caller can do the same dance.
>> 1822 * * %FGP_WRITE - The page will be written
>> 1823 * * %FGP_NOFS - __GFP_FS will get cleared in gfp mask
>> 1824 * * %FGP_NOWAIT - Don't get blocked by page lock
>> 1825 *
1874 * If %FGP_LOCK or %FGP_CREAT are specified t 1826 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1875 * if the %GFP flags specified for %FGP_CREAT 1827 * if the %GFP flags specified for %FGP_CREAT are atomic.
1876 * 1828 *
1877 * If this function returns a folio, it is re !! 1829 * If there is a page cache page, it is returned with an increased refcount.
1878 * 1830 *
1879 * Return: The found folio or an ERR_PTR() ot !! 1831 * Return: The found page or %NULL otherwise.
1880 */ 1832 */
1881 struct folio *__filemap_get_folio(struct addr !! 1833 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index,
1882 fgf_t fgp_flags, gfp_t gfp) !! 1834 int fgp_flags, gfp_t gfp_mask)
1883 { 1835 {
1884 struct folio *folio; !! 1836 struct page *page;
1885 1837
1886 repeat: 1838 repeat:
1887 folio = filemap_get_entry(mapping, in !! 1839 page = mapping_get_entry(mapping, index);
1888 if (xa_is_value(folio)) !! 1840 if (xa_is_value(page)) {
1889 folio = NULL; !! 1841 if (fgp_flags & FGP_ENTRY)
1890 if (!folio) !! 1842 return page;
>> 1843 page = NULL;
>> 1844 }
>> 1845 if (!page)
1891 goto no_page; 1846 goto no_page;
1892 1847
1893 if (fgp_flags & FGP_LOCK) { 1848 if (fgp_flags & FGP_LOCK) {
1894 if (fgp_flags & FGP_NOWAIT) { 1849 if (fgp_flags & FGP_NOWAIT) {
1895 if (!folio_trylock(fo !! 1850 if (!trylock_page(page)) {
1896 folio_put(fol !! 1851 put_page(page);
1897 return ERR_PT !! 1852 return NULL;
1898 } 1853 }
1899 } else { 1854 } else {
1900 folio_lock(folio); !! 1855 lock_page(page);
1901 } 1856 }
1902 1857
1903 /* Has the page been truncate 1858 /* Has the page been truncated? */
1904 if (unlikely(folio->mapping ! !! 1859 if (unlikely(page->mapping != mapping)) {
1905 folio_unlock(folio); !! 1860 unlock_page(page);
1906 folio_put(folio); !! 1861 put_page(page);
1907 goto repeat; 1862 goto repeat;
1908 } 1863 }
1909 VM_BUG_ON_FOLIO(!folio_contai !! 1864 VM_BUG_ON_PAGE(!thp_contains(page, index), page);
1910 } 1865 }
1911 1866
1912 if (fgp_flags & FGP_ACCESSED) 1867 if (fgp_flags & FGP_ACCESSED)
1913 folio_mark_accessed(folio); !! 1868 mark_page_accessed(page);
1914 else if (fgp_flags & FGP_WRITE) { 1869 else if (fgp_flags & FGP_WRITE) {
1915 /* Clear idle flag for buffer 1870 /* Clear idle flag for buffer write */
1916 if (folio_test_idle(folio)) !! 1871 if (page_is_idle(page))
1917 folio_clear_idle(foli !! 1872 clear_page_idle(page);
1918 } 1873 }
>> 1874 if (!(fgp_flags & FGP_HEAD))
>> 1875 page = find_subpage(page, index);
1919 1876
1920 if (fgp_flags & FGP_STABLE) <<
1921 folio_wait_stable(folio); <<
1922 no_page: 1877 no_page:
1923 if (!folio && (fgp_flags & FGP_CREAT) !! 1878 if (!page && (fgp_flags & FGP_CREAT)) {
1924 unsigned int min_order = mapp <<
1925 unsigned int order = max(min_ <<
1926 int err; 1879 int err;
1927 index = mapping_align_index(m <<
1928 <<
1929 if ((fgp_flags & FGP_WRITE) & 1880 if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1930 gfp |= __GFP_WRITE; !! 1881 gfp_mask |= __GFP_WRITE;
1931 if (fgp_flags & FGP_NOFS) 1882 if (fgp_flags & FGP_NOFS)
1932 gfp &= ~__GFP_FS; !! 1883 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 1884
1940 if (order > mapping_max_folio !! 1885 page = __page_cache_alloc(gfp_mask);
1941 order = mapping_max_f !! 1886 if (!page)
1942 /* If we're not aligned, allo !! 1887 return NULL;
1943 if (index & ((1UL << order) - <<
1944 order = __ffs(index); <<
1945 1888
1946 do { !! 1889 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1947 gfp_t alloc_gfp = gfp !! 1890 fgp_flags |= FGP_LOCK;
1948 1891
1949 err = -ENOMEM; !! 1892 /* Init accessed so avoid atomic mark_page_accessed later */
1950 if (order > min_order !! 1893 if (fgp_flags & FGP_ACCESSED)
1951 alloc_gfp |= !! 1894 __SetPageReferenced(page);
1952 folio = filemap_alloc <<
1953 if (!folio) <<
1954 continue; <<
1955 <<
1956 /* Init accessed so a <<
1957 if (fgp_flags & FGP_A <<
1958 __folio_set_r <<
1959 1895
1960 err = filemap_add_fol !! 1896 err = add_to_page_cache_lru(page, mapping, index, gfp_mask);
1961 if (!err) !! 1897 if (unlikely(err)) {
1962 break; !! 1898 put_page(page);
1963 folio_put(folio); !! 1899 page = NULL;
1964 folio = NULL; !! 1900 if (err == -EEXIST)
1965 } while (order-- > min_order) !! 1901 goto repeat;
>> 1902 }
1966 1903
1967 if (err == -EEXIST) <<
1968 goto repeat; <<
1969 if (err) <<
1970 return ERR_PTR(err); <<
1971 /* 1904 /*
1972 * filemap_add_folio locks th !! 1905 * add_to_page_cache_lru locks the page, and for mmap we expect
1973 * we expect an unlocked page !! 1906 * an unlocked page.
1974 */ 1907 */
1975 if (folio && (fgp_flags & FGP !! 1908 if (page && (fgp_flags & FGP_FOR_MMAP))
1976 folio_unlock(folio); !! 1909 unlock_page(page);
1977 } 1910 }
1978 1911
1979 if (!folio) !! 1912 return page;
1980 return ERR_PTR(-ENOENT); <<
1981 return folio; <<
1982 } 1913 }
1983 EXPORT_SYMBOL(__filemap_get_folio); !! 1914 EXPORT_SYMBOL(pagecache_get_page);
1984 1915
1985 static inline struct folio *find_get_entry(st !! 1916 static inline struct page *find_get_entry(struct xa_state *xas, pgoff_t max,
1986 xa_mark_t mark) 1917 xa_mark_t mark)
1987 { 1918 {
1988 struct folio *folio; !! 1919 struct page *page;
1989 1920
1990 retry: 1921 retry:
1991 if (mark == XA_PRESENT) 1922 if (mark == XA_PRESENT)
1992 folio = xas_find(xas, max); !! 1923 page = xas_find(xas, max);
1993 else 1924 else
1994 folio = xas_find_marked(xas, !! 1925 page = xas_find_marked(xas, max, mark);
1995 1926
1996 if (xas_retry(xas, folio)) !! 1927 if (xas_retry(xas, page))
1997 goto retry; 1928 goto retry;
1998 /* 1929 /*
1999 * A shadow entry of a recently evict 1930 * A shadow entry of a recently evicted page, a swap
2000 * entry from shmem/tmpfs or a DAX en 1931 * entry from shmem/tmpfs or a DAX entry. Return it
2001 * without attempting to raise page c 1932 * without attempting to raise page count.
2002 */ 1933 */
2003 if (!folio || xa_is_value(folio)) !! 1934 if (!page || xa_is_value(page))
2004 return folio; !! 1935 return page;
2005 1936
2006 if (!folio_try_get(folio)) !! 1937 if (!page_cache_get_speculative(page))
2007 goto reset; 1938 goto reset;
2008 1939
2009 if (unlikely(folio != xas_reload(xas) !! 1940 /* Has the page moved or been split? */
2010 folio_put(folio); !! 1941 if (unlikely(page != xas_reload(xas))) {
>> 1942 put_page(page);
2011 goto reset; 1943 goto reset;
2012 } 1944 }
2013 1945
2014 return folio; !! 1946 return page;
2015 reset: 1947 reset:
2016 xas_reset(xas); 1948 xas_reset(xas);
2017 goto retry; 1949 goto retry;
2018 } 1950 }
2019 1951
2020 /** 1952 /**
2021 * find_get_entries - gang pagecache lookup 1953 * find_get_entries - gang pagecache lookup
2022 * @mapping: The address_space to search 1954 * @mapping: The address_space to search
2023 * @start: The starting page cache index 1955 * @start: The starting page cache index
2024 * @end: The final page index (inclusi 1956 * @end: The final page index (inclusive).
2025 * @fbatch: Where the resulting entries a !! 1957 * @pvec: Where the resulting entries are placed.
2026 * @indices: The cache indices correspondi 1958 * @indices: The cache indices corresponding to the entries in @entries
2027 * 1959 *
2028 * find_get_entries() will search for and ret 1960 * find_get_entries() will search for and return a batch of entries in
2029 * the mapping. The entries are placed in @f !! 1961 * the mapping. The entries are placed in @pvec. find_get_entries()
2030 * takes a reference on any actual folios it !! 1962 * takes a reference on any actual pages it returns.
2031 * 1963 *
2032 * The entries have ascending indexes. The i !! 1964 * The search returns a group of mapping-contiguous page cache entries
2033 * due to not-present entries or large folios !! 1965 * with ascending indexes. There may be holes in the indices due to
>> 1966 * not-present pages.
2034 * 1967 *
2035 * Any shadow entries of evicted folios, or s !! 1968 * Any shadow entries of evicted pages, or swap entries from
2036 * shmem/tmpfs, are included in the returned 1969 * shmem/tmpfs, are included in the returned array.
2037 * 1970 *
2038 * Return: The number of entries which were f !! 1971 * If it finds a Transparent Huge Page, head or tail, find_get_entries()
>> 1972 * stops at that page: the caller is likely to have a better way to handle
>> 1973 * the compound page as a whole, and then skip its extent, than repeatedly
>> 1974 * calling find_get_entries() to return all its tails.
>> 1975 *
>> 1976 * Return: the number of pages and shadow entries which were found.
2039 */ 1977 */
2040 unsigned find_get_entries(struct address_spac !! 1978 unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
2041 pgoff_t end, struct folio_bat !! 1979 pgoff_t end, struct pagevec *pvec, pgoff_t *indices)
2042 { 1980 {
2043 XA_STATE(xas, &mapping->i_pages, *sta !! 1981 XA_STATE(xas, &mapping->i_pages, start);
2044 struct folio *folio; !! 1982 struct page *page;
>> 1983 unsigned int ret = 0;
>> 1984 unsigned nr_entries = PAGEVEC_SIZE;
2045 1985
2046 rcu_read_lock(); 1986 rcu_read_lock();
2047 while ((folio = find_get_entry(&xas, !! 1987 while ((page = find_get_entry(&xas, end, XA_PRESENT))) {
2048 indices[fbatch->nr] = xas.xa_ !! 1988 /*
2049 if (!folio_batch_add(fbatch, !! 1989 * Terminate early on finding a THP, to allow the caller to
2050 break; !! 1990 * handle it all at once; but continue if this is hugetlbfs.
2051 } !! 1991 */
>> 1992 if (!xa_is_value(page) && PageTransHuge(page) &&
>> 1993 !PageHuge(page)) {
>> 1994 page = find_subpage(page, xas.xa_index);
>> 1995 nr_entries = ret + 1;
>> 1996 }
2052 1997
2053 if (folio_batch_count(fbatch)) { !! 1998 indices[ret] = xas.xa_index;
2054 unsigned long nr; !! 1999 pvec->pages[ret] = page;
2055 int idx = folio_batch_count(f !! 2000 if (++ret == nr_entries)
2056 !! 2001 break;
2057 folio = fbatch->folios[idx]; <<
2058 if (!xa_is_value(folio)) <<
2059 nr = folio_nr_pages(f <<
2060 else <<
2061 nr = 1 << xa_get_orde <<
2062 *start = round_down(indices[i <<
2063 } 2002 }
2064 rcu_read_unlock(); 2003 rcu_read_unlock();
2065 2004
2066 return folio_batch_count(fbatch); !! 2005 pvec->nr = ret;
>> 2006 return ret;
2067 } 2007 }
2068 2008
2069 /** 2009 /**
2070 * find_lock_entries - Find a batch of pageca 2010 * find_lock_entries - Find a batch of pagecache entries.
2071 * @mapping: The address_space to search. 2011 * @mapping: The address_space to search.
2072 * @start: The starting page cache index 2012 * @start: The starting page cache index.
2073 * @end: The final page index (inclusi 2013 * @end: The final page index (inclusive).
2074 * @fbatch: Where the resulting entries a !! 2014 * @pvec: Where the resulting entries are placed.
2075 * @indices: The cache indices of the entr !! 2015 * @indices: The cache indices of the entries in @pvec.
2076 * 2016 *
2077 * find_lock_entries() will return a batch of 2017 * find_lock_entries() will return a batch of entries from @mapping.
2078 * Swap, shadow and DAX entries are included. !! 2018 * Swap, shadow and DAX entries are included. Pages are returned
2079 * locked and with an incremented refcount. !! 2019 * locked and with an incremented refcount. Pages which are locked by
2080 * by somebody else or under writeback are sk !! 2020 * somebody else or under writeback are skipped. Only the head page of
2081 * partially outside the range are not return !! 2021 * a THP is returned. Pages which are partially outside the range are
>> 2022 * not returned.
2082 * 2023 *
2083 * The entries have ascending indexes. The i 2024 * The entries have ascending indexes. The indices may not be consecutive
2084 * due to not-present entries, large folios, !! 2025 * due to not-present entries, THP pages, pages which could not be locked
2085 * locked or folios under writeback. !! 2026 * or pages under writeback.
2086 * 2027 *
2087 * Return: The number of entries which were f 2028 * Return: The number of entries which were found.
2088 */ 2029 */
2089 unsigned find_lock_entries(struct address_spa !! 2030 unsigned find_lock_entries(struct address_space *mapping, pgoff_t start,
2090 pgoff_t end, struct folio_bat !! 2031 pgoff_t end, struct pagevec *pvec, pgoff_t *indices)
2091 { 2032 {
2092 XA_STATE(xas, &mapping->i_pages, *sta !! 2033 XA_STATE(xas, &mapping->i_pages, start);
2093 struct folio *folio; !! 2034 struct page *page;
2094 2035
2095 rcu_read_lock(); 2036 rcu_read_lock();
2096 while ((folio = find_get_entry(&xas, !! 2037 while ((page = find_get_entry(&xas, end, XA_PRESENT))) {
2097 unsigned long base; !! 2038 if (!xa_is_value(page)) {
2098 unsigned long nr; !! 2039 if (page->index < start)
2099 <<
2100 if (!xa_is_value(folio)) { <<
2101 nr = folio_nr_pages(f <<
2102 base = folio->index; <<
2103 /* Omit large folio w <<
2104 if (base < *start) <<
2105 goto put; 2040 goto put;
2106 /* Omit large folio w !! 2041 VM_BUG_ON_PAGE(page->index != xas.xa_index, page);
2107 if (base + nr - 1 > e !! 2042 if (page->index + thp_nr_pages(page) - 1 > end)
2108 goto put; 2043 goto put;
2109 if (!folio_trylock(fo !! 2044 if (!trylock_page(page))
2110 goto put; 2045 goto put;
2111 if (folio->mapping != !! 2046 if (page->mapping != mapping || PageWriteback(page))
2112 folio_test_writeb <<
2113 goto unlock; 2047 goto unlock;
2114 VM_BUG_ON_FOLIO(!foli !! 2048 VM_BUG_ON_PAGE(!thp_contains(page, xas.xa_index),
2115 folio !! 2049 page);
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 } 2050 }
2126 !! 2051 indices[pvec->nr] = xas.xa_index;
2127 /* Update start now so that l !! 2052 if (!pagevec_add(pvec, page))
2128 *start = base + nr; <<
2129 indices[fbatch->nr] = xas.xa_ <<
2130 if (!folio_batch_add(fbatch, <<
2131 break; 2053 break;
2132 continue; !! 2054 goto next;
2133 unlock: 2055 unlock:
2134 folio_unlock(folio); !! 2056 unlock_page(page);
2135 put: 2057 put:
2136 folio_put(folio); !! 2058 put_page(page);
>> 2059 next:
>> 2060 if (!xa_is_value(page) && PageTransHuge(page)) {
>> 2061 unsigned int nr_pages = thp_nr_pages(page);
>> 2062
>> 2063 /* Final THP may cross MAX_LFS_FILESIZE on 32-bit */
>> 2064 xas_set(&xas, page->index + nr_pages);
>> 2065 if (xas.xa_index < nr_pages)
>> 2066 break;
>> 2067 }
2137 } 2068 }
2138 rcu_read_unlock(); 2069 rcu_read_unlock();
2139 2070
2140 return folio_batch_count(fbatch); !! 2071 return pagevec_count(pvec);
2141 } 2072 }
2142 2073
2143 /** 2074 /**
2144 * filemap_get_folios - Get a batch of folios !! 2075 * find_get_pages_range - gang pagecache lookup
2145 * @mapping: The address_space to search 2076 * @mapping: The address_space to search
2146 * @start: The starting page index 2077 * @start: The starting page index
2147 * @end: The final page index (inclusi 2078 * @end: The final page index (inclusive)
2148 * @fbatch: The batch to fill. !! 2079 * @nr_pages: The maximum number of pages
>> 2080 * @pages: Where the resulting pages are placed
2149 * 2081 *
2150 * Search for and return a batch of folios in !! 2082 * find_get_pages_range() will search for and return a group of up to @nr_pages
2151 * index @start and up to index @end (inclusi !! 2083 * pages in the mapping starting at index @start and up to index @end
2152 * in @fbatch with an elevated reference coun !! 2084 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
2153 * !! 2085 * a reference against the returned pages.
2154 * Return: The number of folios which were fo !! 2086 *
2155 * We also update @start to index the next fo !! 2087 * The search returns a group of mapping-contiguous pages with ascending
2156 */ !! 2088 * indexes. There may be holes in the indices due to not-present pages.
2157 unsigned filemap_get_folios(struct address_sp !! 2089 * We also update @start to index the next page for the traversal.
2158 pgoff_t end, struct folio_bat !! 2090 *
>> 2091 * Return: the number of pages which were found. If this number is
>> 2092 * smaller than @nr_pages, the end of specified range has been
>> 2093 * reached.
>> 2094 */
>> 2095 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
>> 2096 pgoff_t end, unsigned int nr_pages,
>> 2097 struct page **pages)
2159 { 2098 {
2160 return filemap_get_folios_tag(mapping !! 2099 XA_STATE(xas, &mapping->i_pages, *start);
>> 2100 struct page *page;
>> 2101 unsigned ret = 0;
>> 2102
>> 2103 if (unlikely(!nr_pages))
>> 2104 return 0;
>> 2105
>> 2106 rcu_read_lock();
>> 2107 while ((page = find_get_entry(&xas, end, XA_PRESENT))) {
>> 2108 /* Skip over shadow, swap and DAX entries */
>> 2109 if (xa_is_value(page))
>> 2110 continue;
>> 2111
>> 2112 pages[ret] = find_subpage(page, xas.xa_index);
>> 2113 if (++ret == nr_pages) {
>> 2114 *start = xas.xa_index + 1;
>> 2115 goto out;
>> 2116 }
>> 2117 }
>> 2118
>> 2119 /*
>> 2120 * We come here when there is no page beyond @end. We take care to not
>> 2121 * overflow the index @start as it confuses some of the callers. This
>> 2122 * breaks the iteration when there is a page at index -1 but that is
>> 2123 * already broken anyway.
>> 2124 */
>> 2125 if (end == (pgoff_t)-1)
>> 2126 *start = (pgoff_t)-1;
>> 2127 else
>> 2128 *start = end + 1;
>> 2129 out:
>> 2130 rcu_read_unlock();
>> 2131
>> 2132 return ret;
2161 } 2133 }
2162 EXPORT_SYMBOL(filemap_get_folios); <<
2163 2134
2164 /** 2135 /**
2165 * filemap_get_folios_contig - Get a batch of !! 2136 * find_get_pages_contig - gang contiguous pagecache lookup
2166 * @mapping: The address_space to search 2137 * @mapping: The address_space to search
2167 * @start: The starting page index !! 2138 * @index: The starting page index
2168 * @end: The final page index (inclusi !! 2139 * @nr_pages: The maximum number of pages
2169 * @fbatch: The batch to fill !! 2140 * @pages: Where the resulting pages are placed
2170 * 2141 *
2171 * filemap_get_folios_contig() works exactly !! 2142 * find_get_pages_contig() works exactly like find_get_pages(), except
2172 * except the returned folios are guaranteed !! 2143 * that the returned number of pages are guaranteed to be contiguous.
2173 * not return all contiguous folios if the ba <<
2174 * 2144 *
2175 * Return: The number of folios found. !! 2145 * Return: the number of pages which were found.
2176 * Also update @start to be positioned for tr <<
2177 */ 2146 */
2178 !! 2147 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
2179 unsigned filemap_get_folios_contig(struct add !! 2148 unsigned int nr_pages, struct page **pages)
2180 pgoff_t *start, pgoff_t end, <<
2181 { 2149 {
2182 XA_STATE(xas, &mapping->i_pages, *sta !! 2150 XA_STATE(xas, &mapping->i_pages, index);
2183 unsigned long nr; !! 2151 struct page *page;
2184 struct folio *folio; !! 2152 unsigned int ret = 0;
2185 2153
2186 rcu_read_lock(); !! 2154 if (unlikely(!nr_pages))
>> 2155 return 0;
2187 2156
2188 for (folio = xas_load(&xas); folio && !! 2157 rcu_read_lock();
2189 folio = xas_next(&xas !! 2158 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
2190 if (xas_retry(&xas, folio)) !! 2159 if (xas_retry(&xas, page))
2191 continue; 2160 continue;
2192 /* 2161 /*
2193 * If the entry has been swap 2162 * If the entry has been swapped out, we can stop looking.
2194 * No current caller is looki 2163 * No current caller is looking for DAX entries.
2195 */ 2164 */
2196 if (xa_is_value(folio)) !! 2165 if (xa_is_value(page))
2197 goto update_start; !! 2166 break;
2198 <<
2199 /* If we landed in the middle <<
2200 if (xa_is_sibling(folio)) <<
2201 goto update_start; <<
2202 2167
2203 if (!folio_try_get(folio)) !! 2168 if (!page_cache_get_speculative(page))
2204 goto retry; 2169 goto retry;
2205 2170
2206 if (unlikely(folio != xas_rel !! 2171 /* Has the page moved or been split? */
2207 goto put_folio; !! 2172 if (unlikely(page != xas_reload(&xas)))
>> 2173 goto put_page;
2208 2174
2209 if (!folio_batch_add(fbatch, !! 2175 pages[ret] = find_subpage(page, xas.xa_index);
2210 nr = folio_nr_pages(f !! 2176 if (++ret == nr_pages)
2211 *start = folio->index !! 2177 break;
2212 goto out; <<
2213 } <<
2214 continue; 2178 continue;
2215 put_folio: !! 2179 put_page:
2216 folio_put(folio); !! 2180 put_page(page);
2217 <<
2218 retry: 2181 retry:
2219 xas_reset(&xas); 2182 xas_reset(&xas);
2220 } 2183 }
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(); 2184 rcu_read_unlock();
2231 return folio_batch_count(fbatch); !! 2185 return ret;
2232 } 2186 }
2233 EXPORT_SYMBOL(filemap_get_folios_contig); !! 2187 EXPORT_SYMBOL(find_get_pages_contig);
2234 2188
2235 /** 2189 /**
2236 * filemap_get_folios_tag - Get a batch of fo !! 2190 * find_get_pages_range_tag - Find and return head pages matching @tag.
2237 * @mapping: The address_space to search !! 2191 * @mapping: the address_space to search
2238 * @start: The starting page index !! 2192 * @index: the starting page index
2239 * @end: The final page index (inclusi !! 2193 * @end: The final page index (inclusive)
2240 * @tag: The tag index !! 2194 * @tag: the tag index
2241 * @fbatch: The batch to fill !! 2195 * @nr_pages: the maximum number of pages
2242 * !! 2196 * @pages: where the resulting pages are placed
2243 * The first folio may start before @start; i !! 2197 *
2244 * @start. The final folio may extend beyond !! 2198 * Like find_get_pages(), except we only return head pages which are tagged
2245 * contain @end. The folios have ascending i !! 2199 * with @tag. @index is updated to the index immediately after the last
2246 * between the folios if there are indices wh !! 2200 * page we return, ready for the next iteration.
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 * 2201 *
2251 * Return: The number of folios found. !! 2202 * Return: the number of pages which were found.
2252 * Also update @start to index the next folio <<
2253 */ 2203 */
2254 unsigned filemap_get_folios_tag(struct addres !! 2204 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
2255 pgoff_t end, xa_mark_ !! 2205 pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
>> 2206 struct page **pages)
2256 { 2207 {
2257 XA_STATE(xas, &mapping->i_pages, *sta !! 2208 XA_STATE(xas, &mapping->i_pages, *index);
2258 struct folio *folio; !! 2209 struct page *page;
>> 2210 unsigned ret = 0;
>> 2211
>> 2212 if (unlikely(!nr_pages))
>> 2213 return 0;
2259 2214
2260 rcu_read_lock(); 2215 rcu_read_lock();
2261 while ((folio = find_get_entry(&xas, !! 2216 while ((page = find_get_entry(&xas, end, tag))) {
2262 /* 2217 /*
2263 * Shadow entries should neve 2218 * Shadow entries should never be tagged, but this iteration
2264 * is lockless so there is a 2219 * is lockless so there is a window for page reclaim to evict
2265 * a page we saw tagged. Skip !! 2220 * a page we saw tagged. Skip over it.
2266 */ 2221 */
2267 if (xa_is_value(folio)) !! 2222 if (xa_is_value(page))
2268 continue; 2223 continue;
2269 if (!folio_batch_add(fbatch, !! 2224
2270 unsigned long nr = fo !! 2225 pages[ret] = page;
2271 *start = folio->index !! 2226 if (++ret == nr_pages) {
>> 2227 *index = page->index + thp_nr_pages(page);
2272 goto out; 2228 goto out;
2273 } 2229 }
2274 } 2230 }
>> 2231
2275 /* 2232 /*
2276 * We come here when there is no page !! 2233 * We come here when we got to @end. We take care to not overflow the
2277 * overflow the index @start as it co !! 2234 * index @index as it confuses some of the callers. This breaks the
2278 * breaks the iteration when there is !! 2235 * iteration when there is a page at index -1 but that is already
2279 * already broke anyway. !! 2236 * broken anyway.
2280 */ 2237 */
2281 if (end == (pgoff_t)-1) 2238 if (end == (pgoff_t)-1)
2282 *start = (pgoff_t)-1; !! 2239 *index = (pgoff_t)-1;
2283 else 2240 else
2284 *start = end + 1; !! 2241 *index = end + 1;
2285 out: 2242 out:
2286 rcu_read_unlock(); 2243 rcu_read_unlock();
2287 2244
2288 return folio_batch_count(fbatch); !! 2245 return ret;
2289 } 2246 }
2290 EXPORT_SYMBOL(filemap_get_folios_tag); !! 2247 EXPORT_SYMBOL(find_get_pages_range_tag);
2291 2248
2292 /* 2249 /*
2293 * CD/DVDs are error prone. When a medium err 2250 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2294 * a _large_ part of the i/o request. Imagine 2251 * a _large_ part of the i/o request. Imagine the worst scenario:
2295 * 2252 *
2296 * ---R_________________________________ 2253 * ---R__________________________________________B__________
2297 * ^ reading here 2254 * ^ reading here ^ bad block(assume 4k)
2298 * 2255 *
2299 * read(R) => miss => readahead(R...B) => med 2256 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2300 * => failing the whole request => read(R) => 2257 * => failing the whole request => read(R) => read(R+1) =>
2301 * readahead(R+1...B+1) => bang => read(R+2) 2258 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2302 * readahead(R+3...B+2) => bang => read(R+3) 2259 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2303 * readahead(R+4...B+3) => bang => read(R+4) 2260 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2304 * 2261 *
2305 * It is going insane. Fix it by quickly scal 2262 * It is going insane. Fix it by quickly scaling down the readahead size.
2306 */ 2263 */
2307 static void shrink_readahead_size_eio(struct 2264 static void shrink_readahead_size_eio(struct file_ra_state *ra)
2308 { 2265 {
2309 ra->ra_pages /= 4; 2266 ra->ra_pages /= 4;
2310 } 2267 }
2311 2268
2312 /* 2269 /*
2313 * filemap_get_read_batch - Get a batch of fo !! 2270 * filemap_get_read_batch - Get a batch of pages for read
2314 * 2271 *
2315 * Get a batch of folios which represent a co !! 2272 * Get a batch of pages which represent a contiguous range of bytes
2316 * the file. No exceptional entries will be !! 2273 * in the file. No tail pages will be returned. If @index is in the
2317 * the middle of a folio, the entire folio wi !! 2274 * middle of a THP, the entire THP will be returned. The last page in
2318 * folio in the batch may have the readahead !! 2275 * the batch may have Readahead set or be not Uptodate so that the
2319 * clear so that the caller can take the appr !! 2276 * caller can take the appropriate action.
2320 */ 2277 */
2321 static void filemap_get_read_batch(struct add 2278 static void filemap_get_read_batch(struct address_space *mapping,
2322 pgoff_t index, pgoff_t max, s !! 2279 pgoff_t index, pgoff_t max, struct pagevec *pvec)
2323 { 2280 {
2324 XA_STATE(xas, &mapping->i_pages, inde 2281 XA_STATE(xas, &mapping->i_pages, index);
2325 struct folio *folio; !! 2282 struct page *head;
2326 2283
2327 rcu_read_lock(); 2284 rcu_read_lock();
2328 for (folio = xas_load(&xas); folio; f !! 2285 for (head = xas_load(&xas); head; head = xas_next(&xas)) {
2329 if (xas_retry(&xas, folio)) !! 2286 if (xas_retry(&xas, head))
2330 continue; 2287 continue;
2331 if (xas.xa_index > max || xa_ !! 2288 if (xas.xa_index > max || xa_is_value(head))
2332 break; <<
2333 if (xa_is_sibling(folio)) <<
2334 break; 2289 break;
2335 if (!folio_try_get(folio)) !! 2290 if (!page_cache_get_speculative(head))
2336 goto retry; 2291 goto retry;
2337 2292
2338 if (unlikely(folio != xas_rel !! 2293 /* Has the page moved or been split? */
2339 goto put_folio; !! 2294 if (unlikely(head != xas_reload(&xas)))
>> 2295 goto put_page;
2340 2296
2341 if (!folio_batch_add(fbatch, !! 2297 if (!pagevec_add(pvec, head))
2342 break; 2298 break;
2343 if (!folio_test_uptodate(foli !! 2299 if (!PageUptodate(head))
2344 break; 2300 break;
2345 if (folio_test_readahead(foli !! 2301 if (PageReadahead(head))
2346 break; 2302 break;
2347 xas_advance(&xas, folio_next_ !! 2303 xas.xa_index = head->index + thp_nr_pages(head) - 1;
>> 2304 xas.xa_offset = (xas.xa_index >> xas.xa_shift) & XA_CHUNK_MASK;
2348 continue; 2305 continue;
2349 put_folio: !! 2306 put_page:
2350 folio_put(folio); !! 2307 put_page(head);
2351 retry: 2308 retry:
2352 xas_reset(&xas); 2309 xas_reset(&xas);
2353 } 2310 }
2354 rcu_read_unlock(); 2311 rcu_read_unlock();
2355 } 2312 }
2356 2313
2357 static int filemap_read_folio(struct file *fi !! 2314 static int filemap_read_page(struct file *file, struct address_space *mapping,
2358 struct folio *folio) !! 2315 struct page *page)
2359 { 2316 {
2360 bool workingset = folio_test_workings <<
2361 unsigned long pflags; <<
2362 int error; 2317 int error;
2363 2318
>> 2319 /*
>> 2320 * A previous I/O error may have been due to temporary failures,
>> 2321 * eg. multipath errors. PG_error will be set again if readpage
>> 2322 * fails.
>> 2323 */
>> 2324 ClearPageError(page);
2364 /* Start the actual read. The read wi 2325 /* Start the actual read. The read will unlock the page. */
2365 if (unlikely(workingset)) !! 2326 error = mapping->a_ops->readpage(file, page);
2366 psi_memstall_enter(&pflags); <<
2367 error = filler(file, folio); <<
2368 if (unlikely(workingset)) <<
2369 psi_memstall_leave(&pflags); <<
2370 if (error) 2327 if (error)
2371 return error; 2328 return error;
2372 2329
2373 error = folio_wait_locked_killable(fo !! 2330 error = wait_on_page_locked_killable(page);
2374 if (error) 2331 if (error)
2375 return error; 2332 return error;
2376 if (folio_test_uptodate(folio)) !! 2333 if (PageUptodate(page))
2377 return 0; 2334 return 0;
2378 if (file) !! 2335 shrink_readahead_size_eio(&file->f_ra);
2379 shrink_readahead_size_eio(&fi <<
2380 return -EIO; 2336 return -EIO;
2381 } 2337 }
2382 2338
2383 static bool filemap_range_uptodate(struct add 2339 static bool filemap_range_uptodate(struct address_space *mapping,
2384 loff_t pos, size_t count, str !! 2340 loff_t pos, struct iov_iter *iter, struct page *page)
2385 bool need_uptodate) <<
2386 { 2341 {
2387 if (folio_test_uptodate(folio)) !! 2342 int count;
>> 2343
>> 2344 if (PageUptodate(page))
2388 return true; 2345 return true;
2389 /* pipes can't handle partially uptod 2346 /* pipes can't handle partially uptodate pages */
2390 if (need_uptodate) !! 2347 if (iov_iter_is_pipe(iter))
2391 return false; 2348 return false;
2392 if (!mapping->a_ops->is_partially_upt 2349 if (!mapping->a_ops->is_partially_uptodate)
2393 return false; 2350 return false;
2394 if (mapping->host->i_blkbits >= folio !! 2351 if (mapping->host->i_blkbits >= (PAGE_SHIFT + thp_order(page)))
2395 return false; 2352 return false;
2396 2353
2397 if (folio_pos(folio) > pos) { !! 2354 count = iter->count;
2398 count -= folio_pos(folio) - p !! 2355 if (page_offset(page) > pos) {
>> 2356 count -= page_offset(page) - pos;
2399 pos = 0; 2357 pos = 0;
2400 } else { 2358 } else {
2401 pos -= folio_pos(folio); !! 2359 pos -= page_offset(page);
2402 } 2360 }
2403 2361
2404 return mapping->a_ops->is_partially_u !! 2362 return mapping->a_ops->is_partially_uptodate(page, pos, count);
2405 } 2363 }
2406 2364
2407 static int filemap_update_page(struct kiocb * 2365 static int filemap_update_page(struct kiocb *iocb,
2408 struct address_space *mapping !! 2366 struct address_space *mapping, struct iov_iter *iter,
2409 struct folio *folio, bool nee !! 2367 struct page *page)
2410 { 2368 {
2411 int error; 2369 int error;
2412 2370
2413 if (iocb->ki_flags & IOCB_NOWAIT) { !! 2371 if (!trylock_page(page)) {
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 2372 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2423 goto unlock_mapping; !! 2373 return -EAGAIN;
2424 if (!(iocb->ki_flags & IOCB_W 2374 if (!(iocb->ki_flags & IOCB_WAITQ)) {
2425 filemap_invalidate_un !! 2375 put_and_wait_on_page_locked(page, TASK_KILLABLE);
2426 /* <<
2427 * This is where we u <<
2428 * previously submitt <<
2429 */ <<
2430 folio_put_wait_locked <<
2431 return AOP_TRUNCATED_ 2376 return AOP_TRUNCATED_PAGE;
2432 } 2377 }
2433 error = __folio_lock_async(fo !! 2378 error = __lock_page_async(page, iocb->ki_waitq);
2434 if (error) 2379 if (error)
2435 goto unlock_mapping; !! 2380 return error;
2436 } 2381 }
2437 2382
2438 error = AOP_TRUNCATED_PAGE; !! 2383 if (!page->mapping)
2439 if (!folio->mapping) !! 2384 goto truncated;
2440 goto unlock; <<
2441 2385
2442 error = 0; 2386 error = 0;
2443 if (filemap_range_uptodate(mapping, i !! 2387 if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, page))
2444 need_uptod <<
2445 goto unlock; 2388 goto unlock;
2446 2389
2447 error = -EAGAIN; 2390 error = -EAGAIN;
2448 if (iocb->ki_flags & (IOCB_NOIO | IOC 2391 if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2449 goto unlock; 2392 goto unlock;
2450 2393
2451 error = filemap_read_folio(iocb->ki_f !! 2394 error = filemap_read_page(iocb->ki_filp, mapping, page);
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) 2395 if (error == AOP_TRUNCATED_PAGE)
2459 folio_put(folio); !! 2396 put_page(page);
>> 2397 return error;
>> 2398 truncated:
>> 2399 unlock_page(page);
>> 2400 put_page(page);
>> 2401 return AOP_TRUNCATED_PAGE;
>> 2402 unlock:
>> 2403 unlock_page(page);
2460 return error; 2404 return error;
2461 } 2405 }
2462 2406
2463 static int filemap_create_folio(struct file * !! 2407 static int filemap_create_page(struct file *file,
2464 struct address_space *mapping !! 2408 struct address_space *mapping, pgoff_t index,
2465 struct folio_batch *fbatch) !! 2409 struct pagevec *pvec)
2466 { 2410 {
2467 struct folio *folio; !! 2411 struct page *page;
2468 int error; 2412 int error;
2469 unsigned int min_order = mapping_min_ <<
2470 pgoff_t index; <<
2471 2413
2472 folio = filemap_alloc_folio(mapping_g !! 2414 page = page_cache_alloc(mapping);
2473 if (!folio) !! 2415 if (!page)
2474 return -ENOMEM; 2416 return -ENOMEM;
2475 2417
2476 /* !! 2418 error = add_to_page_cache_lru(page, mapping, index,
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 2419 mapping_gfp_constraint(mapping, GFP_KERNEL));
2493 if (error == -EEXIST) 2420 if (error == -EEXIST)
2494 error = AOP_TRUNCATED_PAGE; 2421 error = AOP_TRUNCATED_PAGE;
2495 if (error) 2422 if (error)
2496 goto error; 2423 goto error;
2497 2424
2498 error = filemap_read_folio(file, mapp !! 2425 error = filemap_read_page(file, mapping, page);
2499 if (error) 2426 if (error)
2500 goto error; 2427 goto error;
2501 2428
2502 filemap_invalidate_unlock_shared(mapp !! 2429 pagevec_add(pvec, page);
2503 folio_batch_add(fbatch, folio); <<
2504 return 0; 2430 return 0;
2505 error: 2431 error:
2506 filemap_invalidate_unlock_shared(mapp !! 2432 put_page(page);
2507 folio_put(folio); <<
2508 return error; 2433 return error;
2509 } 2434 }
2510 2435
2511 static int filemap_readahead(struct kiocb *io 2436 static int filemap_readahead(struct kiocb *iocb, struct file *file,
2512 struct address_space *mapping !! 2437 struct address_space *mapping, struct page *page,
2513 pgoff_t last_index) 2438 pgoff_t last_index)
2514 { 2439 {
2515 DEFINE_READAHEAD(ractl, file, &file-> <<
2516 <<
2517 if (iocb->ki_flags & IOCB_NOIO) 2440 if (iocb->ki_flags & IOCB_NOIO)
2518 return -EAGAIN; 2441 return -EAGAIN;
2519 page_cache_async_ra(&ractl, folio, la !! 2442 page_cache_async_readahead(mapping, &file->f_ra, file, page,
>> 2443 page->index, last_index - page->index);
2520 return 0; 2444 return 0;
2521 } 2445 }
2522 2446
2523 static int filemap_get_pages(struct kiocb *io !! 2447 static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter,
2524 struct folio_batch *fbatch, b !! 2448 struct pagevec *pvec)
2525 { 2449 {
2526 struct file *filp = iocb->ki_filp; 2450 struct file *filp = iocb->ki_filp;
2527 struct address_space *mapping = filp- 2451 struct address_space *mapping = filp->f_mapping;
2528 struct file_ra_state *ra = &filp->f_r 2452 struct file_ra_state *ra = &filp->f_ra;
2529 pgoff_t index = iocb->ki_pos >> PAGE_ 2453 pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2530 pgoff_t last_index; 2454 pgoff_t last_index;
2531 struct folio *folio; !! 2455 struct page *page;
2532 unsigned int flags; <<
2533 int err = 0; 2456 int err = 0;
2534 2457
2535 /* "last_index" is the index of the p !! 2458 last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE);
2536 last_index = DIV_ROUND_UP(iocb->ki_po <<
2537 retry: 2459 retry:
2538 if (fatal_signal_pending(current)) 2460 if (fatal_signal_pending(current))
2539 return -EINTR; 2461 return -EINTR;
2540 2462
2541 filemap_get_read_batch(mapping, index !! 2463 filemap_get_read_batch(mapping, index, last_index, pvec);
2542 if (!folio_batch_count(fbatch)) { !! 2464 if (!pagevec_count(pvec)) {
2543 if (iocb->ki_flags & IOCB_NOI 2465 if (iocb->ki_flags & IOCB_NOIO)
2544 return -EAGAIN; 2466 return -EAGAIN;
2545 if (iocb->ki_flags & IOCB_NOW <<
2546 flags = memalloc_noio <<
2547 page_cache_sync_readahead(map 2467 page_cache_sync_readahead(mapping, ra, filp, index,
2548 last_index - 2468 last_index - index);
2549 if (iocb->ki_flags & IOCB_NOW !! 2469 filemap_get_read_batch(mapping, index, last_index, pvec);
2550 memalloc_noio_restore <<
2551 filemap_get_read_batch(mappin <<
2552 } 2470 }
2553 if (!folio_batch_count(fbatch)) { !! 2471 if (!pagevec_count(pvec)) {
2554 if (iocb->ki_flags & (IOCB_NO 2472 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2555 return -EAGAIN; 2473 return -EAGAIN;
2556 err = filemap_create_folio(fi !! 2474 err = filemap_create_page(filp, mapping,
>> 2475 iocb->ki_pos >> PAGE_SHIFT, pvec);
2557 if (err == AOP_TRUNCATED_PAGE 2476 if (err == AOP_TRUNCATED_PAGE)
2558 goto retry; 2477 goto retry;
2559 return err; 2478 return err;
2560 } 2479 }
2561 2480
2562 folio = fbatch->folios[folio_batch_co !! 2481 page = pvec->pages[pagevec_count(pvec) - 1];
2563 if (folio_test_readahead(folio)) { !! 2482 if (PageReadahead(page)) {
2564 err = filemap_readahead(iocb, !! 2483 err = filemap_readahead(iocb, filp, mapping, page, last_index);
2565 if (err) 2484 if (err)
2566 goto err; 2485 goto err;
2567 } 2486 }
2568 if (!folio_test_uptodate(folio)) { !! 2487 if (!PageUptodate(page)) {
2569 if ((iocb->ki_flags & IOCB_WA !! 2488 if ((iocb->ki_flags & IOCB_WAITQ) && pagevec_count(pvec) > 1)
2570 folio_batch_count(fbatch) <<
2571 iocb->ki_flags |= IOC 2489 iocb->ki_flags |= IOCB_NOWAIT;
2572 err = filemap_update_page(ioc !! 2490 err = filemap_update_page(iocb, mapping, iter, page);
2573 nee <<
2574 if (err) 2491 if (err)
2575 goto err; 2492 goto err;
2576 } 2493 }
2577 2494
2578 trace_mm_filemap_get_pages(mapping, i <<
2579 return 0; 2495 return 0;
2580 err: 2496 err:
2581 if (err < 0) 2497 if (err < 0)
2582 folio_put(folio); !! 2498 put_page(page);
2583 if (likely(--fbatch->nr)) !! 2499 if (likely(--pvec->nr))
2584 return 0; 2500 return 0;
2585 if (err == AOP_TRUNCATED_PAGE) 2501 if (err == AOP_TRUNCATED_PAGE)
2586 goto retry; 2502 goto retry;
2587 return err; 2503 return err;
2588 } 2504 }
2589 2505
2590 static inline bool pos_same_folio(loff_t pos1 <<
2591 { <<
2592 unsigned int shift = folio_shift(foli <<
2593 <<
2594 return (pos1 >> shift == pos2 >> shif <<
2595 } <<
2596 <<
2597 /** 2506 /**
2598 * filemap_read - Read data from the page cac 2507 * filemap_read - Read data from the page cache.
2599 * @iocb: The iocb to read. 2508 * @iocb: The iocb to read.
2600 * @iter: Destination for the data. 2509 * @iter: Destination for the data.
2601 * @already_read: Number of bytes already rea 2510 * @already_read: Number of bytes already read by the caller.
2602 * 2511 *
2603 * Copies data from the page cache. If the d 2512 * Copies data from the page cache. If the data is not currently present,
2604 * uses the readahead and read_folio address_ !! 2513 * uses the readahead and readpage address_space operations to fetch it.
2605 * 2514 *
2606 * Return: Total number of bytes copied, incl 2515 * Return: Total number of bytes copied, including those already read by
2607 * the caller. If an error happens before an 2516 * the caller. If an error happens before any bytes are copied, returns
2608 * a negative error number. 2517 * a negative error number.
2609 */ 2518 */
2610 ssize_t filemap_read(struct kiocb *iocb, stru 2519 ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2611 ssize_t already_read) 2520 ssize_t already_read)
2612 { 2521 {
2613 struct file *filp = iocb->ki_filp; 2522 struct file *filp = iocb->ki_filp;
2614 struct file_ra_state *ra = &filp->f_r 2523 struct file_ra_state *ra = &filp->f_ra;
2615 struct address_space *mapping = filp- 2524 struct address_space *mapping = filp->f_mapping;
2616 struct inode *inode = mapping->host; 2525 struct inode *inode = mapping->host;
2617 struct folio_batch fbatch; !! 2526 struct pagevec pvec;
2618 int i, error = 0; 2527 int i, error = 0;
2619 bool writably_mapped; 2528 bool writably_mapped;
2620 loff_t isize, end_offset; 2529 loff_t isize, end_offset;
2621 loff_t last_pos = ra->prev_pos; <<
2622 2530
2623 if (unlikely(iocb->ki_pos >= inode->i 2531 if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2624 return 0; 2532 return 0;
2625 if (unlikely(!iov_iter_count(iter))) 2533 if (unlikely(!iov_iter_count(iter)))
2626 return 0; 2534 return 0;
2627 2535
2628 iov_iter_truncate(iter, inode->i_sb-> !! 2536 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2629 folio_batch_init(&fbatch); !! 2537 pagevec_init(&pvec);
2630 2538
2631 do { 2539 do {
2632 cond_resched(); 2540 cond_resched();
2633 2541
2634 /* 2542 /*
2635 * If we've already successfu 2543 * If we've already successfully copied some data, then we
2636 * can no longer safely retur 2544 * can no longer safely return -EIOCBQUEUED. Hence mark
2637 * an async read NOWAIT at th 2545 * an async read NOWAIT at that point.
2638 */ 2546 */
2639 if ((iocb->ki_flags & IOCB_WA 2547 if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
2640 iocb->ki_flags |= IOC 2548 iocb->ki_flags |= IOCB_NOWAIT;
2641 2549
2642 if (unlikely(iocb->ki_pos >= !! 2550 error = filemap_get_pages(iocb, iter, &pvec);
2643 break; <<
2644 <<
2645 error = filemap_get_pages(ioc <<
2646 if (error < 0) 2551 if (error < 0)
2647 break; 2552 break;
2648 2553
2649 /* 2554 /*
2650 * i_size must be checked aft 2555 * i_size must be checked after we know the pages are Uptodate.
2651 * 2556 *
2652 * Checking i_size after the 2557 * Checking i_size after the check allows us to calculate
2653 * the correct value for "nr" 2558 * the correct value for "nr", which means the zero-filled
2654 * part of the page is not co 2559 * part of the page is not copied back to userspace (unless
2655 * another truncate extends t 2560 * another truncate extends the file - this is desired though).
2656 */ 2561 */
2657 isize = i_size_read(inode); 2562 isize = i_size_read(inode);
2658 if (unlikely(iocb->ki_pos >= 2563 if (unlikely(iocb->ki_pos >= isize))
2659 goto put_folios; !! 2564 goto put_pages;
2660 end_offset = min_t(loff_t, is 2565 end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2661 2566
2662 /* 2567 /*
2663 * Once we start copying data 2568 * Once we start copying data, we don't want to be touching any
2664 * cachelines that might be c 2569 * cachelines that might be contended:
2665 */ 2570 */
2666 writably_mapped = mapping_wri 2571 writably_mapped = mapping_writably_mapped(mapping);
2667 2572
2668 /* 2573 /*
2669 * When a read accesses the s !! 2574 * When a sequential read accesses a page several times, only
2670 * mark it as accessed the fi 2575 * mark it as accessed the first time.
2671 */ 2576 */
2672 if (!pos_same_folio(iocb->ki_ !! 2577 if (iocb->ki_pos >> PAGE_SHIFT !=
2673 fbatch.fo !! 2578 ra->prev_pos >> PAGE_SHIFT)
2674 folio_mark_accessed(f !! 2579 mark_page_accessed(pvec.pages[0]);
2675 !! 2580
2676 for (i = 0; i < folio_batch_c !! 2581 for (i = 0; i < pagevec_count(&pvec); i++) {
2677 struct folio *folio = !! 2582 struct page *page = pvec.pages[i];
2678 size_t fsize = folio_ !! 2583 size_t page_size = thp_size(page);
2679 size_t offset = iocb- !! 2584 size_t offset = iocb->ki_pos & (page_size - 1);
2680 size_t bytes = min_t( 2585 size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2681 !! 2586 page_size - offset);
2682 size_t copied; 2587 size_t copied;
2683 2588
2684 if (end_offset < foli !! 2589 if (end_offset < page_offset(page))
2685 break; 2590 break;
2686 if (i > 0) 2591 if (i > 0)
2687 folio_mark_ac !! 2592 mark_page_accessed(page);
2688 /* 2593 /*
2689 * If users can be wr !! 2594 * If users can be writing to this page using arbitrary
2690 * virtual addresses, !! 2595 * virtual addresses, take care about potential aliasing
2691 * before reading the !! 2596 * before reading the page on the kernel side.
2692 */ 2597 */
2693 if (writably_mapped) !! 2598 if (writably_mapped) {
2694 flush_dcache_ !! 2599 int j;
>> 2600
>> 2601 for (j = 0; j < thp_nr_pages(page); j++)
>> 2602 flush_dcache_page(page + j);
>> 2603 }
2695 2604
2696 copied = copy_folio_t !! 2605 copied = copy_page_to_iter(page, offset, bytes, iter);
2697 2606
2698 already_read += copie 2607 already_read += copied;
2699 iocb->ki_pos += copie 2608 iocb->ki_pos += copied;
2700 last_pos = iocb->ki_p !! 2609 ra->prev_pos = iocb->ki_pos;
2701 2610
2702 if (copied < bytes) { 2611 if (copied < bytes) {
2703 error = -EFAU 2612 error = -EFAULT;
2704 break; 2613 break;
2705 } 2614 }
2706 } 2615 }
2707 put_folios: !! 2616 put_pages:
2708 for (i = 0; i < folio_batch_c !! 2617 for (i = 0; i < pagevec_count(&pvec); i++)
2709 folio_put(fbatch.foli !! 2618 put_page(pvec.pages[i]);
2710 folio_batch_init(&fbatch); !! 2619 pagevec_reinit(&pvec);
2711 } while (iov_iter_count(iter) && iocb 2620 } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2712 2621
2713 file_accessed(filp); 2622 file_accessed(filp);
2714 ra->prev_pos = last_pos; !! 2623
2715 return already_read ? already_read : 2624 return already_read ? already_read : error;
2716 } 2625 }
2717 EXPORT_SYMBOL_GPL(filemap_read); 2626 EXPORT_SYMBOL_GPL(filemap_read);
2718 2627
2719 int kiocb_write_and_wait(struct kiocb *iocb, <<
2720 { <<
2721 struct address_space *mapping = iocb- <<
2722 loff_t pos = iocb->ki_pos; <<
2723 loff_t end = pos + count - 1; <<
2724 <<
2725 if (iocb->ki_flags & IOCB_NOWAIT) { <<
2726 if (filemap_range_needs_write <<
2727 return -EAGAIN; <<
2728 return 0; <<
2729 } <<
2730 <<
2731 return filemap_write_and_wait_range(m <<
2732 } <<
2733 EXPORT_SYMBOL_GPL(kiocb_write_and_wait); <<
2734 <<
2735 int filemap_invalidate_pages(struct address_s <<
2736 loff_t pos, loff <<
2737 { <<
2738 int ret; <<
2739 <<
2740 if (nowait) { <<
2741 /* we could block if there ar <<
2742 if (filemap_range_has_page(ma <<
2743 return -EAGAIN; <<
2744 } else { <<
2745 ret = filemap_write_and_wait_ <<
2746 if (ret) <<
2747 return ret; <<
2748 } <<
2749 <<
2750 /* <<
2751 * After a write we want buffered rea <<
2752 * the new data. We invalidate clean <<
2753 * about to write. We do this *befor <<
2754 * without clobbering -EIOCBQUEUED fr <<
2755 */ <<
2756 return invalidate_inode_pages2_range( <<
2757 <<
2758 } <<
2759 <<
2760 int kiocb_invalidate_pages(struct kiocb *iocb <<
2761 { <<
2762 struct address_space *mapping = iocb- <<
2763 <<
2764 return filemap_invalidate_pages(mappi <<
2765 iocb- <<
2766 iocb- <<
2767 } <<
2768 EXPORT_SYMBOL_GPL(kiocb_invalidate_pages); <<
2769 <<
2770 /** 2628 /**
2771 * generic_file_read_iter - generic filesyste 2629 * generic_file_read_iter - generic filesystem read routine
2772 * @iocb: kernel I/O control block 2630 * @iocb: kernel I/O control block
2773 * @iter: destination for the data read 2631 * @iter: destination for the data read
2774 * 2632 *
2775 * This is the "read_iter()" routine for all 2633 * This is the "read_iter()" routine for all filesystems
2776 * that can use the page cache directly. 2634 * that can use the page cache directly.
2777 * 2635 *
2778 * The IOCB_NOWAIT flag in iocb->ki_flags ind 2636 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2779 * be returned when no data can be read witho 2637 * be returned when no data can be read without waiting for I/O requests
2780 * to complete; it doesn't prevent readahead. 2638 * to complete; it doesn't prevent readahead.
2781 * 2639 *
2782 * The IOCB_NOIO flag in iocb->ki_flags indic 2640 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2783 * requests shall be made for the read or for 2641 * requests shall be made for the read or for readahead. When no data
2784 * can be read, -EAGAIN shall be returned. W 2642 * can be read, -EAGAIN shall be returned. When readahead would be
2785 * triggered, a partial, possibly empty read 2643 * triggered, a partial, possibly empty read shall be returned.
2786 * 2644 *
2787 * Return: 2645 * Return:
2788 * * number of bytes copied, even for partial 2646 * * number of bytes copied, even for partial reads
2789 * * negative error code (or 0 if IOCB_NOIO) 2647 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2790 */ 2648 */
2791 ssize_t 2649 ssize_t
2792 generic_file_read_iter(struct kiocb *iocb, st 2650 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2793 { 2651 {
2794 size_t count = iov_iter_count(iter); 2652 size_t count = iov_iter_count(iter);
2795 ssize_t retval = 0; 2653 ssize_t retval = 0;
2796 2654
2797 if (!count) 2655 if (!count)
2798 return 0; /* skip atime */ 2656 return 0; /* skip atime */
2799 2657
2800 if (iocb->ki_flags & IOCB_DIRECT) { 2658 if (iocb->ki_flags & IOCB_DIRECT) {
2801 struct file *file = iocb->ki_ 2659 struct file *file = iocb->ki_filp;
2802 struct address_space *mapping 2660 struct address_space *mapping = file->f_mapping;
2803 struct inode *inode = mapping 2661 struct inode *inode = mapping->host;
>> 2662 loff_t size;
>> 2663
>> 2664 size = i_size_read(inode);
>> 2665 if (iocb->ki_flags & IOCB_NOWAIT) {
>> 2666 if (filemap_range_needs_writeback(mapping, iocb->ki_pos,
>> 2667 iocb->ki_pos + count - 1))
>> 2668 return -EAGAIN;
>> 2669 } else {
>> 2670 retval = filemap_write_and_wait_range(mapping,
>> 2671 iocb->ki_pos,
>> 2672 iocb->ki_pos + count - 1);
>> 2673 if (retval < 0)
>> 2674 return retval;
>> 2675 }
2804 2676
2805 retval = kiocb_write_and_wait <<
2806 if (retval < 0) <<
2807 return retval; <<
2808 file_accessed(file); 2677 file_accessed(file);
2809 2678
2810 retval = mapping->a_ops->dire 2679 retval = mapping->a_ops->direct_IO(iocb, iter);
2811 if (retval >= 0) { 2680 if (retval >= 0) {
2812 iocb->ki_pos += retva 2681 iocb->ki_pos += retval;
2813 count -= retval; 2682 count -= retval;
2814 } 2683 }
2815 if (retval != -EIOCBQUEUED) 2684 if (retval != -EIOCBQUEUED)
2816 iov_iter_revert(iter, 2685 iov_iter_revert(iter, count - iov_iter_count(iter));
2817 2686
2818 /* 2687 /*
2819 * Btrfs can have a short DIO 2688 * Btrfs can have a short DIO read if we encounter
2820 * compressed extents, so if 2689 * compressed extents, so if there was an error, or if
2821 * we've already read everyth 2690 * we've already read everything we wanted to, or if
2822 * there was a short read bec 2691 * there was a short read because we hit EOF, go ahead
2823 * and return. Otherwise fal 2692 * and return. Otherwise fallthrough to buffered io for
2824 * the rest of the read. Buf 2693 * the rest of the read. Buffered reads will not work for
2825 * DAX files, so don't bother 2694 * DAX files, so don't bother trying.
2826 */ 2695 */
2827 if (retval < 0 || !count || I !! 2696 if (retval < 0 || !count || iocb->ki_pos >= size ||
2828 return retval; !! 2697 IS_DAX(inode))
2829 if (iocb->ki_pos >= i_size_re <<
2830 return retval; 2698 return retval;
2831 } 2699 }
2832 2700
2833 return filemap_read(iocb, iter, retva 2701 return filemap_read(iocb, iter, retval);
2834 } 2702 }
2835 EXPORT_SYMBOL(generic_file_read_iter); 2703 EXPORT_SYMBOL(generic_file_read_iter);
2836 2704
2837 /* !! 2705 static inline loff_t page_seek_hole_data(struct xa_state *xas,
2838 * Splice subpages from a folio into a pipe. !! 2706 struct address_space *mapping, struct page *page,
2839 */ <<
2840 size_t splice_folio_into_pipe(struct pipe_ino <<
2841 struct folio *f <<
2842 { <<
2843 struct page *page; <<
2844 size_t spliced = 0, offset = offset_i <<
2845 <<
2846 page = folio_page(folio, offset / PAG <<
2847 size = min(size, folio_size(folio) - <<
2848 offset %= PAGE_SIZE; <<
2849 <<
2850 while (spliced < size && <<
2851 !pipe_full(pipe->head, pipe->t <<
2852 struct pipe_buffer *buf = pip <<
2853 size_t part = min_t(size_t, P <<
2854 <<
2855 *buf = (struct pipe_buffer) { <<
2856 .ops = &page_cache <<
2857 .page = page, <<
2858 .offset = offset, <<
2859 .len = part, <<
2860 }; <<
2861 folio_get(folio); <<
2862 pipe->head++; <<
2863 page++; <<
2864 spliced += part; <<
2865 offset = 0; <<
2866 } <<
2867 <<
2868 return spliced; <<
2869 } <<
2870 <<
2871 /** <<
2872 * filemap_splice_read - Splice data from a <<
2873 * @in: The file to read from <<
2874 * @ppos: Pointer to the file position to rea <<
2875 * @pipe: The pipe to splice into <<
2876 * @len: The amount to splice <<
2877 * @flags: The SPLICE_F_* flags <<
2878 * <<
2879 * This function gets folios from a file's pa <<
2880 * pipe. Readahead will be called as necessa <<
2881 * be used for blockdevs also. <<
2882 * <<
2883 * Return: On success, the number of bytes re <<
2884 * will be updated if appropriate; 0 will be <<
2885 * to be read; -EAGAIN will be returned if th <<
2886 * other negative error code will be returned <<
2887 * if the pipe has insufficient space, we rea <<
2888 * hole. <<
2889 */ <<
2890 ssize_t filemap_splice_read(struct file *in, <<
2891 struct pipe_inode <<
2892 size_t len, unsig <<
2893 { <<
2894 struct folio_batch fbatch; <<
2895 struct kiocb iocb; <<
2896 size_t total_spliced = 0, used, npage <<
2897 loff_t isize, end_offset; <<
2898 bool writably_mapped; <<
2899 int i, error = 0; <<
2900 <<
2901 if (unlikely(*ppos >= in->f_mapping-> <<
2902 return 0; <<
2903 <<
2904 init_sync_kiocb(&iocb, in); <<
2905 iocb.ki_pos = *ppos; <<
2906 <<
2907 /* Work out how much data we can actu <<
2908 used = pipe_occupancy(pipe->head, pip <<
2909 npages = max_t(ssize_t, pipe->max_usa <<
2910 len = min_t(size_t, len, npages * PAG <<
2911 <<
2912 folio_batch_init(&fbatch); <<
2913 <<
2914 do { <<
2915 cond_resched(); <<
2916 <<
2917 if (*ppos >= i_size_read(in-> <<
2918 break; <<
2919 <<
2920 iocb.ki_pos = *ppos; <<
2921 error = filemap_get_pages(&io <<
2922 if (error < 0) <<
2923 break; <<
2924 <<
2925 /* <<
2926 * i_size must be checked aft <<
2927 * <<
2928 * Checking i_size after the <<
2929 * the correct value for "nr" <<
2930 * part of the page is not co <<
2931 * another truncate extends t <<
2932 */ <<
2933 isize = i_size_read(in->f_map <<
2934 if (unlikely(*ppos >= isize)) <<
2935 break; <<
2936 end_offset = min_t(loff_t, is <<
2937 <<
2938 /* <<
2939 * Once we start copying data <<
2940 * cachelines that might be c <<
2941 */ <<
2942 writably_mapped = mapping_wri <<
2943 <<
2944 for (i = 0; i < folio_batch_c <<
2945 struct folio *folio = <<
2946 size_t n; <<
2947 <<
2948 if (folio_pos(folio) <<
2949 goto out; <<
2950 folio_mark_accessed(f <<
2951 <<
2952 /* <<
2953 * If users can be wr <<
2954 * virtual addresses, <<
2955 * before reading the <<
2956 */ <<
2957 if (writably_mapped) <<
2958 flush_dcache_ <<
2959 <<
2960 n = min_t(loff_t, len <<
2961 n = splice_folio_into <<
2962 if (!n) <<
2963 goto out; <<
2964 len -= n; <<
2965 total_spliced += n; <<
2966 *ppos += n; <<
2967 in->f_ra.prev_pos = * <<
2968 if (pipe_full(pipe->h <<
2969 goto out; <<
2970 } <<
2971 <<
2972 folio_batch_release(&fbatch); <<
2973 } while (len); <<
2974 <<
2975 out: <<
2976 folio_batch_release(&fbatch); <<
2977 file_accessed(in); <<
2978 <<
2979 return total_spliced ? total_spliced <<
2980 } <<
2981 EXPORT_SYMBOL(filemap_splice_read); <<
2982 <<
2983 static inline loff_t folio_seek_hole_data(str <<
2984 struct address_space *mapping <<
2985 loff_t start, loff_t end, boo 2707 loff_t start, loff_t end, bool seek_data)
2986 { 2708 {
2987 const struct address_space_operations 2709 const struct address_space_operations *ops = mapping->a_ops;
2988 size_t offset, bsz = i_blocksize(mapp 2710 size_t offset, bsz = i_blocksize(mapping->host);
2989 2711
2990 if (xa_is_value(folio) || folio_test_ !! 2712 if (xa_is_value(page) || PageUptodate(page))
2991 return seek_data ? start : en 2713 return seek_data ? start : end;
2992 if (!ops->is_partially_uptodate) 2714 if (!ops->is_partially_uptodate)
2993 return seek_data ? end : star 2715 return seek_data ? end : start;
2994 2716
2995 xas_pause(xas); 2717 xas_pause(xas);
2996 rcu_read_unlock(); 2718 rcu_read_unlock();
2997 folio_lock(folio); !! 2719 lock_page(page);
2998 if (unlikely(folio->mapping != mappin !! 2720 if (unlikely(page->mapping != mapping))
2999 goto unlock; 2721 goto unlock;
3000 2722
3001 offset = offset_in_folio(folio, start !! 2723 offset = offset_in_thp(page, start) & ~(bsz - 1);
3002 2724
3003 do { 2725 do {
3004 if (ops->is_partially_uptodat !! 2726 if (ops->is_partially_uptodate(page, offset, bsz) == seek_data)
3005 <<
3006 break; 2727 break;
3007 start = (start + bsz) & ~(bsz 2728 start = (start + bsz) & ~(bsz - 1);
3008 offset += bsz; 2729 offset += bsz;
3009 } while (offset < folio_size(folio)); !! 2730 } while (offset < thp_size(page));
3010 unlock: 2731 unlock:
3011 folio_unlock(folio); !! 2732 unlock_page(page);
3012 rcu_read_lock(); 2733 rcu_read_lock();
3013 return start; 2734 return start;
3014 } 2735 }
3015 2736
3016 static inline size_t seek_folio_size(struct x !! 2737 static inline
>> 2738 unsigned int seek_page_size(struct xa_state *xas, struct page *page)
3017 { 2739 {
3018 if (xa_is_value(folio)) !! 2740 if (xa_is_value(page))
3019 return PAGE_SIZE << xas_get_o !! 2741 return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
3020 return folio_size(folio); !! 2742 return thp_size(page);
3021 } 2743 }
3022 2744
3023 /** 2745 /**
3024 * mapping_seek_hole_data - Seek for SEEK_DAT 2746 * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
3025 * @mapping: Address space to search. 2747 * @mapping: Address space to search.
3026 * @start: First byte to consider. 2748 * @start: First byte to consider.
3027 * @end: Limit of search (exclusive). 2749 * @end: Limit of search (exclusive).
3028 * @whence: Either SEEK_HOLE or SEEK_DATA. 2750 * @whence: Either SEEK_HOLE or SEEK_DATA.
3029 * 2751 *
3030 * If the page cache knows which blocks conta 2752 * If the page cache knows which blocks contain holes and which blocks
3031 * contain data, your filesystem can use this 2753 * contain data, your filesystem can use this function to implement
3032 * SEEK_HOLE and SEEK_DATA. This is useful f 2754 * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are
3033 * entirely memory-based such as tmpfs, and f 2755 * entirely memory-based such as tmpfs, and filesystems which support
3034 * unwritten extents. 2756 * unwritten extents.
3035 * 2757 *
3036 * Return: The requested offset on success, o 2758 * Return: The requested offset on success, or -ENXIO if @whence specifies
3037 * SEEK_DATA and there is no data after @star 2759 * SEEK_DATA and there is no data after @start. There is an implicit hole
3038 * after @end - 1, so SEEK_HOLE returns @end 2760 * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
3039 * and @end contain data. 2761 * and @end contain data.
3040 */ 2762 */
3041 loff_t mapping_seek_hole_data(struct address_ 2763 loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
3042 loff_t end, int whence) 2764 loff_t end, int whence)
3043 { 2765 {
3044 XA_STATE(xas, &mapping->i_pages, star 2766 XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
3045 pgoff_t max = (end - 1) >> PAGE_SHIFT 2767 pgoff_t max = (end - 1) >> PAGE_SHIFT;
3046 bool seek_data = (whence == SEEK_DATA 2768 bool seek_data = (whence == SEEK_DATA);
3047 struct folio *folio; !! 2769 struct page *page;
3048 2770
3049 if (end <= start) 2771 if (end <= start)
3050 return -ENXIO; 2772 return -ENXIO;
3051 2773
3052 rcu_read_lock(); 2774 rcu_read_lock();
3053 while ((folio = find_get_entry(&xas, !! 2775 while ((page = find_get_entry(&xas, max, XA_PRESENT))) {
3054 loff_t pos = (u64)xas.xa_inde 2776 loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
3055 size_t seek_size; !! 2777 unsigned int seek_size;
3056 2778
3057 if (start < pos) { 2779 if (start < pos) {
3058 if (!seek_data) 2780 if (!seek_data)
3059 goto unlock; 2781 goto unlock;
3060 start = pos; 2782 start = pos;
3061 } 2783 }
3062 2784
3063 seek_size = seek_folio_size(& !! 2785 seek_size = seek_page_size(&xas, page);
3064 pos = round_up((u64)pos + 1, !! 2786 pos = round_up(pos + 1, seek_size);
3065 start = folio_seek_hole_data( !! 2787 start = page_seek_hole_data(&xas, mapping, page, start, pos,
3066 seek_data); 2788 seek_data);
3067 if (start < pos) 2789 if (start < pos)
3068 goto unlock; 2790 goto unlock;
3069 if (start >= end) 2791 if (start >= end)
3070 break; 2792 break;
3071 if (seek_size > PAGE_SIZE) 2793 if (seek_size > PAGE_SIZE)
3072 xas_set(&xas, pos >> 2794 xas_set(&xas, pos >> PAGE_SHIFT);
3073 if (!xa_is_value(folio)) !! 2795 if (!xa_is_value(page))
3074 folio_put(folio); !! 2796 put_page(page);
3075 } 2797 }
3076 if (seek_data) 2798 if (seek_data)
3077 start = -ENXIO; 2799 start = -ENXIO;
3078 unlock: 2800 unlock:
3079 rcu_read_unlock(); 2801 rcu_read_unlock();
3080 if (folio && !xa_is_value(folio)) !! 2802 if (page && !xa_is_value(page))
3081 folio_put(folio); !! 2803 put_page(page);
3082 if (start > end) 2804 if (start > end)
3083 return end; 2805 return end;
3084 return start; 2806 return start;
3085 } 2807 }
3086 2808
3087 #ifdef CONFIG_MMU 2809 #ifdef CONFIG_MMU
3088 #define MMAP_LOTSAMISS (100) 2810 #define MMAP_LOTSAMISS (100)
3089 /* 2811 /*
3090 * lock_folio_maybe_drop_mmap - lock the page !! 2812 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
3091 * @vmf - the vm_fault for this fault. 2813 * @vmf - the vm_fault for this fault.
3092 * @folio - the folio to lock. !! 2814 * @page - the page to lock.
3093 * @fpin - the pointer to the file we may pin 2815 * @fpin - the pointer to the file we may pin (or is already pinned).
3094 * 2816 *
3095 * This works similar to lock_folio_or_retry !! 2817 * This works similar to lock_page_or_retry in that it can drop the mmap_lock.
3096 * mmap_lock. It differs in that it actually !! 2818 * 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 !! 2819 * if it couldn't lock the page. If we did have to drop the mmap_lock then fpin
3098 * to drop the mmap_lock then fpin will point !! 2820 * 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 */ 2821 */
3101 static int lock_folio_maybe_drop_mmap(struct !! 2822 static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page,
3102 struct f 2823 struct file **fpin)
3103 { 2824 {
3104 if (folio_trylock(folio)) !! 2825 if (trylock_page(page))
3105 return 1; 2826 return 1;
3106 2827
3107 /* 2828 /*
3108 * NOTE! This will make us return wit 2829 * NOTE! This will make us return with VM_FAULT_RETRY, but with
3109 * the fault lock still held. That's !! 2830 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
3110 * is supposed to work. We have way t 2831 * is supposed to work. We have way too many special cases..
3111 */ 2832 */
3112 if (vmf->flags & FAULT_FLAG_RETRY_NOW 2833 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
3113 return 0; 2834 return 0;
3114 2835
3115 *fpin = maybe_unlock_mmap_for_io(vmf, 2836 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
3116 if (vmf->flags & FAULT_FLAG_KILLABLE) 2837 if (vmf->flags & FAULT_FLAG_KILLABLE) {
3117 if (__folio_lock_killable(fol !! 2838 if (__lock_page_killable(page)) {
3118 /* 2839 /*
3119 * We didn't have the !! 2840 * We didn't have the right flags to drop the mmap_lock,
3120 * fault lock, but al !! 2841 * but all fault_handlers only check for fatal signals
3121 * for fatal signals !! 2842 * if we return VM_FAULT_RETRY, so we need to drop the
3122 * so we need to drop !! 2843 * mmap_lock here and return 0 if we don't have a fpin.
3123 * return 0 if we don <<
3124 */ 2844 */
3125 if (*fpin == NULL) 2845 if (*fpin == NULL)
3126 release_fault !! 2846 mmap_read_unlock(vmf->vma->vm_mm);
3127 return 0; 2847 return 0;
3128 } 2848 }
3129 } else 2849 } else
3130 __folio_lock(folio); !! 2850 __lock_page(page);
3131 <<
3132 return 1; 2851 return 1;
3133 } 2852 }
3134 2853
>> 2854
3135 /* 2855 /*
3136 * Synchronous readahead happens when we don' 2856 * 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 2857 * 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 2858 * 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 2859 * 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 2860 * returned needs to be fput()'ed when we're done with it.
3141 */ 2861 */
3142 static struct file *do_sync_mmap_readahead(st 2862 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
3143 { 2863 {
3144 struct file *file = vmf->vma->vm_file 2864 struct file *file = vmf->vma->vm_file;
3145 struct file_ra_state *ra = &file->f_r 2865 struct file_ra_state *ra = &file->f_ra;
3146 struct address_space *mapping = file- 2866 struct address_space *mapping = file->f_mapping;
3147 DEFINE_READAHEAD(ractl, file, ra, map 2867 DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
3148 struct file *fpin = NULL; 2868 struct file *fpin = NULL;
3149 unsigned long vm_flags = vmf->vma->vm <<
3150 unsigned int mmap_miss; 2869 unsigned int mmap_miss;
3151 2870
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 <<
3170 /* If we don't want any read-ahead, d 2871 /* If we don't want any read-ahead, don't bother */
3171 if (vm_flags & VM_RAND_READ) !! 2872 if (vmf->vma->vm_flags & VM_RAND_READ)
3172 return fpin; 2873 return fpin;
3173 if (!ra->ra_pages) 2874 if (!ra->ra_pages)
3174 return fpin; 2875 return fpin;
3175 2876
3176 if (vm_flags & VM_SEQ_READ) { !! 2877 if (vmf->vma->vm_flags & VM_SEQ_READ) {
3177 fpin = maybe_unlock_mmap_for_ 2878 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3178 page_cache_sync_ra(&ractl, ra 2879 page_cache_sync_ra(&ractl, ra->ra_pages);
3179 return fpin; 2880 return fpin;
3180 } 2881 }
3181 2882
3182 /* Avoid banging the cache line if no 2883 /* Avoid banging the cache line if not needed */
3183 mmap_miss = READ_ONCE(ra->mmap_miss); 2884 mmap_miss = READ_ONCE(ra->mmap_miss);
3184 if (mmap_miss < MMAP_LOTSAMISS * 10) 2885 if (mmap_miss < MMAP_LOTSAMISS * 10)
3185 WRITE_ONCE(ra->mmap_miss, ++m 2886 WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
3186 2887
3187 /* 2888 /*
3188 * Do we miss much more than hit in t 2889 * Do we miss much more than hit in this file? If so,
3189 * stop bothering with read-ahead. It 2890 * stop bothering with read-ahead. It will only hurt.
3190 */ 2891 */
3191 if (mmap_miss > MMAP_LOTSAMISS) 2892 if (mmap_miss > MMAP_LOTSAMISS)
3192 return fpin; 2893 return fpin;
3193 2894
3194 /* 2895 /*
3195 * mmap read-around 2896 * mmap read-around
3196 */ 2897 */
3197 fpin = maybe_unlock_mmap_for_io(vmf, 2898 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3198 ra->start = max_t(long, 0, vmf->pgoff 2899 ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
3199 ra->size = ra->ra_pages; 2900 ra->size = ra->ra_pages;
3200 ra->async_size = ra->ra_pages / 4; 2901 ra->async_size = ra->ra_pages / 4;
3201 ractl._index = ra->start; 2902 ractl._index = ra->start;
3202 page_cache_ra_order(&ractl, ra, 0); !! 2903 do_page_cache_ra(&ractl, ra->size, ra->async_size);
3203 return fpin; 2904 return fpin;
3204 } 2905 }
3205 2906
3206 /* 2907 /*
3207 * Asynchronous readahead happens when we fin 2908 * Asynchronous readahead happens when we find the page and PG_readahead,
3208 * so we want to possibly extend the readahea 2909 * 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 2910 * was pinned if we have to drop the mmap_lock in order to do IO.
3210 */ 2911 */
3211 static struct file *do_async_mmap_readahead(s 2912 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
3212 s !! 2913 struct page *page)
3213 { 2914 {
3214 struct file *file = vmf->vma->vm_file 2915 struct file *file = vmf->vma->vm_file;
3215 struct file_ra_state *ra = &file->f_r 2916 struct file_ra_state *ra = &file->f_ra;
3216 DEFINE_READAHEAD(ractl, file, ra, fil !! 2917 struct address_space *mapping = file->f_mapping;
3217 struct file *fpin = NULL; 2918 struct file *fpin = NULL;
3218 unsigned int mmap_miss; 2919 unsigned int mmap_miss;
>> 2920 pgoff_t offset = vmf->pgoff;
3219 2921
3220 /* If we don't want any read-ahead, d 2922 /* If we don't want any read-ahead, don't bother */
3221 if (vmf->vma->vm_flags & VM_RAND_READ 2923 if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
3222 return fpin; 2924 return fpin;
3223 <<
3224 mmap_miss = READ_ONCE(ra->mmap_miss); 2925 mmap_miss = READ_ONCE(ra->mmap_miss);
3225 if (mmap_miss) 2926 if (mmap_miss)
3226 WRITE_ONCE(ra->mmap_miss, --m 2927 WRITE_ONCE(ra->mmap_miss, --mmap_miss);
3227 !! 2928 if (PageReadahead(page)) {
3228 if (folio_test_readahead(folio)) { <<
3229 fpin = maybe_unlock_mmap_for_ 2929 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3230 page_cache_async_ra(&ractl, f !! 2930 page_cache_async_readahead(mapping, ra, file,
>> 2931 page, offset, ra->ra_pages);
3231 } 2932 }
3232 return fpin; 2933 return fpin;
3233 } 2934 }
3234 2935
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 /** 2936 /**
3279 * filemap_fault - read in file data for page 2937 * filemap_fault - read in file data for page fault handling
3280 * @vmf: struct vm_fault containing de 2938 * @vmf: struct vm_fault containing details of the fault
3281 * 2939 *
3282 * filemap_fault() is invoked via the vma ope 2940 * filemap_fault() is invoked via the vma operations vector for a
3283 * mapped memory region to read in file data 2941 * mapped memory region to read in file data during a page fault.
3284 * 2942 *
3285 * The goto's are kind of ugly, but this stre 2943 * 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 2944 * it in the page cache, and handles the special cases reasonably without
3287 * having a lot of duplicated code. 2945 * having a lot of duplicated code.
3288 * 2946 *
3289 * vma->vm_mm->mmap_lock must be held on entr 2947 * vma->vm_mm->mmap_lock must be held on entry.
3290 * 2948 *
3291 * If our return value has VM_FAULT_RETRY set 2949 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
3292 * may be dropped before doing I/O or by lock !! 2950 * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
3293 * 2951 *
3294 * If our return value does not have VM_FAULT 2952 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
3295 * has not been released. 2953 * has not been released.
3296 * 2954 *
3297 * We never return with VM_FAULT_RETRY and a 2955 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3298 * 2956 *
3299 * Return: bitwise-OR of %VM_FAULT_ codes. 2957 * Return: bitwise-OR of %VM_FAULT_ codes.
3300 */ 2958 */
3301 vm_fault_t filemap_fault(struct vm_fault *vmf 2959 vm_fault_t filemap_fault(struct vm_fault *vmf)
3302 { 2960 {
3303 int error; 2961 int error;
3304 struct file *file = vmf->vma->vm_file 2962 struct file *file = vmf->vma->vm_file;
3305 struct file *fpin = NULL; 2963 struct file *fpin = NULL;
3306 struct address_space *mapping = file- 2964 struct address_space *mapping = file->f_mapping;
3307 struct inode *inode = mapping->host; 2965 struct inode *inode = mapping->host;
3308 pgoff_t max_idx, index = vmf->pgoff; !! 2966 pgoff_t offset = vmf->pgoff;
3309 struct folio *folio; !! 2967 pgoff_t max_off;
>> 2968 struct page *page;
3310 vm_fault_t ret = 0; 2969 vm_fault_t ret = 0;
3311 bool mapping_locked = false; <<
3312 2970
3313 max_idx = DIV_ROUND_UP(i_size_read(in !! 2971 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3314 if (unlikely(index >= max_idx)) !! 2972 if (unlikely(offset >= max_off))
3315 return VM_FAULT_SIGBUS; 2973 return VM_FAULT_SIGBUS;
3316 2974
3317 trace_mm_filemap_fault(mapping, index <<
3318 <<
3319 /* 2975 /*
3320 * Do we have something in the page c 2976 * Do we have something in the page cache already?
3321 */ 2977 */
3322 folio = filemap_get_folio(mapping, in !! 2978 page = find_get_page(mapping, offset);
3323 if (likely(!IS_ERR(folio))) { !! 2979 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
3324 /* 2980 /*
3325 * We found the page, so try !! 2981 * We found the page, so try async readahead before
3326 * the lock. !! 2982 * waiting for the lock.
3327 */ 2983 */
3328 if (!(vmf->flags & FAULT_FLAG !! 2984 fpin = do_async_mmap_readahead(vmf, page);
3329 fpin = do_async_mmap_ !! 2985 } 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 2986 /* No page in the page cache at all */
3340 count_vm_event(PGMAJFAULT); 2987 count_vm_event(PGMAJFAULT);
3341 count_memcg_event_mm(vmf->vma 2988 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3342 ret = VM_FAULT_MAJOR; 2989 ret = VM_FAULT_MAJOR;
3343 fpin = do_sync_mmap_readahead 2990 fpin = do_sync_mmap_readahead(vmf);
3344 retry_find: 2991 retry_find:
3345 /* !! 2992 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 2993 FGP_CREAT|FGP_FOR_MMAP,
3355 vmf 2994 vmf->gfp_mask);
3356 if (IS_ERR(folio)) { !! 2995 if (!page) {
3357 if (fpin) 2996 if (fpin)
3358 goto out_retr 2997 goto out_retry;
3359 filemap_invalidate_un <<
3360 return VM_FAULT_OOM; 2998 return VM_FAULT_OOM;
3361 } 2999 }
3362 } 3000 }
3363 3001
3364 if (!lock_folio_maybe_drop_mmap(vmf, !! 3002 if (!lock_page_maybe_drop_mmap(vmf, page, &fpin))
3365 goto out_retry; 3003 goto out_retry;
3366 3004
3367 /* Did it get truncated? */ 3005 /* Did it get truncated? */
3368 if (unlikely(folio->mapping != mappin !! 3006 if (unlikely(compound_head(page)->mapping != mapping)) {
3369 folio_unlock(folio); !! 3007 unlock_page(page);
3370 folio_put(folio); !! 3008 put_page(page);
3371 goto retry_find; 3009 goto retry_find;
3372 } 3010 }
3373 VM_BUG_ON_FOLIO(!folio_contains(folio !! 3011 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
3374 3012
3375 /* 3013 /*
3376 * We have a locked folio in the page !! 3014 * 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 !! 3015 * that it's up-to-date. If not, it is going to be due to an error.
3378 * or because readahead was otherwise <<
3379 */ 3016 */
3380 if (unlikely(!folio_test_uptodate(fol !! 3017 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; 3018 goto page_not_uptodate;
3399 } <<
3400 3019
3401 /* 3020 /*
3402 * We've made it this far and we had 3021 * We've made it this far and we had to drop our mmap_lock, now is the
3403 * time to return to the upper layer 3022 * time to return to the upper layer and have it re-find the vma and
3404 * redo the fault. 3023 * redo the fault.
3405 */ 3024 */
3406 if (fpin) { 3025 if (fpin) {
3407 folio_unlock(folio); !! 3026 unlock_page(page);
3408 goto out_retry; 3027 goto out_retry;
3409 } 3028 }
3410 if (mapping_locked) <<
3411 filemap_invalidate_unlock_sha <<
3412 3029
3413 /* 3030 /*
3414 * Found the page and have a referenc 3031 * Found the page and have a reference on it.
3415 * We must recheck i_size under page 3032 * We must recheck i_size under page lock.
3416 */ 3033 */
3417 max_idx = DIV_ROUND_UP(i_size_read(in !! 3034 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3418 if (unlikely(index >= max_idx)) { !! 3035 if (unlikely(offset >= max_off)) {
3419 folio_unlock(folio); !! 3036 unlock_page(page);
3420 folio_put(folio); !! 3037 put_page(page);
3421 return VM_FAULT_SIGBUS; 3038 return VM_FAULT_SIGBUS;
3422 } 3039 }
3423 3040
3424 vmf->page = folio_file_page(folio, in !! 3041 vmf->page = page;
3425 return ret | VM_FAULT_LOCKED; 3042 return ret | VM_FAULT_LOCKED;
3426 3043
3427 page_not_uptodate: 3044 page_not_uptodate:
3428 /* 3045 /*
3429 * Umm, take care of errors if the pa 3046 * Umm, take care of errors if the page isn't up-to-date.
3430 * Try to re-read it _once_. We do th 3047 * Try to re-read it _once_. We do this synchronously,
3431 * because there really aren't any pe 3048 * because there really aren't any performance issues here
3432 * and we need to check for errors. 3049 * and we need to check for errors.
3433 */ 3050 */
3434 fpin = maybe_unlock_mmap_for_io(vmf, 3051 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3435 error = filemap_read_folio(file, mapp !! 3052 error = filemap_read_page(file, mapping, page);
3436 if (fpin) 3053 if (fpin)
3437 goto out_retry; 3054 goto out_retry;
3438 folio_put(folio); !! 3055 put_page(page);
3439 3056
3440 if (!error || error == AOP_TRUNCATED_ 3057 if (!error || error == AOP_TRUNCATED_PAGE)
3441 goto retry_find; 3058 goto retry_find;
3442 filemap_invalidate_unlock_shared(mapp <<
3443 3059
3444 return VM_FAULT_SIGBUS; 3060 return VM_FAULT_SIGBUS;
3445 3061
3446 out_retry: 3062 out_retry:
3447 /* 3063 /*
3448 * We dropped the mmap_lock, we need 3064 * We dropped the mmap_lock, we need to return to the fault handler to
3449 * re-find the vma and come back and 3065 * re-find the vma and come back and find our hopefully still populated
3450 * page. 3066 * page.
3451 */ 3067 */
3452 if (!IS_ERR(folio)) !! 3068 if (page)
3453 folio_put(folio); !! 3069 put_page(page);
3454 if (mapping_locked) <<
3455 filemap_invalidate_unlock_sha <<
3456 if (fpin) 3070 if (fpin)
3457 fput(fpin); 3071 fput(fpin);
3458 return ret | VM_FAULT_RETRY; 3072 return ret | VM_FAULT_RETRY;
3459 } 3073 }
3460 EXPORT_SYMBOL(filemap_fault); 3074 EXPORT_SYMBOL(filemap_fault);
3461 3075
3462 static bool filemap_map_pmd(struct vm_fault * !! 3076 static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page)
3463 pgoff_t start) <<
3464 { 3077 {
3465 struct mm_struct *mm = vmf->vma->vm_m 3078 struct mm_struct *mm = vmf->vma->vm_mm;
3466 3079
3467 /* Huge page is mapped? No need to pr 3080 /* Huge page is mapped? No need to proceed. */
3468 if (pmd_trans_huge(*vmf->pmd)) { 3081 if (pmd_trans_huge(*vmf->pmd)) {
3469 folio_unlock(folio); !! 3082 unlock_page(page);
3470 folio_put(folio); !! 3083 put_page(page);
3471 return true; 3084 return true;
3472 } 3085 }
3473 3086
3474 if (pmd_none(*vmf->pmd) && folio_test !! 3087 if (pmd_none(*vmf->pmd) && PageTransHuge(page)) {
3475 struct page *page = folio_fil !! 3088 vm_fault_t ret = do_set_pmd(vmf, page);
3476 vm_fault_t ret = do_set_pmd(v !! 3089 if (!ret) {
3477 if (!ret) { !! 3090 /* The page is mapped successfully, reference consumed. */
3478 /* The page is mapped !! 3091 unlock_page(page);
3479 folio_unlock(folio); !! 3092 return true;
3480 return true; !! 3093 }
>> 3094 }
>> 3095
>> 3096 if (pmd_none(*vmf->pmd)) {
>> 3097 vmf->ptl = pmd_lock(mm, vmf->pmd);
>> 3098 if (likely(pmd_none(*vmf->pmd))) {
>> 3099 mm_inc_nr_ptes(mm);
>> 3100 pmd_populate(mm, vmf->pmd, vmf->prealloc_pte);
>> 3101 vmf->prealloc_pte = NULL;
3481 } 3102 }
>> 3103 spin_unlock(vmf->ptl);
3482 } 3104 }
3483 3105
3484 if (pmd_none(*vmf->pmd) && vmf->preal !! 3106 /* See comment in handle_pte_fault() */
3485 pmd_install(mm, vmf->pmd, &vm !! 3107 if (pmd_devmap_trans_unstable(vmf->pmd)) {
>> 3108 unlock_page(page);
>> 3109 put_page(page);
>> 3110 return true;
>> 3111 }
3486 3112
3487 return false; 3113 return false;
3488 } 3114 }
3489 3115
3490 static struct folio *next_uptodate_folio(stru !! 3116 static struct page *next_uptodate_page(struct page *page,
3491 struct address_space *mapping !! 3117 struct address_space *mapping,
>> 3118 struct xa_state *xas, pgoff_t end_pgoff)
3492 { 3119 {
3493 struct folio *folio = xas_next_entry( <<
3494 unsigned long max_idx; 3120 unsigned long max_idx;
3495 3121
3496 do { 3122 do {
3497 if (!folio) !! 3123 if (!page)
3498 return NULL; 3124 return NULL;
3499 if (xas_retry(xas, folio)) !! 3125 if (xas_retry(xas, page))
3500 continue; 3126 continue;
3501 if (xa_is_value(folio)) !! 3127 if (xa_is_value(page))
3502 continue; 3128 continue;
3503 if (folio_test_locked(folio)) !! 3129 if (PageLocked(page))
3504 continue; 3130 continue;
3505 if (!folio_try_get(folio)) !! 3131 if (!page_cache_get_speculative(page))
3506 continue; 3132 continue;
3507 /* Has the page moved or been 3133 /* Has the page moved or been split? */
3508 if (unlikely(folio != xas_rel !! 3134 if (unlikely(page != xas_reload(xas)))
>> 3135 goto skip;
>> 3136 if (!PageUptodate(page) || PageReadahead(page))
3509 goto skip; 3137 goto skip;
3510 if (!folio_test_uptodate(foli !! 3138 if (PageHWPoison(page))
3511 goto skip; 3139 goto skip;
3512 if (!folio_trylock(folio)) !! 3140 if (!trylock_page(page))
3513 goto skip; 3141 goto skip;
3514 if (folio->mapping != mapping !! 3142 if (page->mapping != mapping)
3515 goto unlock; 3143 goto unlock;
3516 if (!folio_test_uptodate(foli !! 3144 if (!PageUptodate(page))
3517 goto unlock; 3145 goto unlock;
3518 max_idx = DIV_ROUND_UP(i_size 3146 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3519 if (xas->xa_index >= max_idx) 3147 if (xas->xa_index >= max_idx)
3520 goto unlock; 3148 goto unlock;
3521 return folio; !! 3149 return page;
3522 unlock: 3150 unlock:
3523 folio_unlock(folio); !! 3151 unlock_page(page);
3524 skip: 3152 skip:
3525 folio_put(folio); !! 3153 put_page(page);
3526 } while ((folio = xas_next_entry(xas, !! 3154 } while ((page = xas_next_entry(xas, end_pgoff)) != NULL);
3527 3155
3528 return NULL; 3156 return NULL;
3529 } 3157 }
3530 3158
3531 /* !! 3159 static inline struct page *first_map_page(struct address_space *mapping,
3532 * Map page range [start_page, start_page + n !! 3160 struct xa_state *xas,
3533 * start_page is gotten from start by folio_p !! 3161 pgoff_t end_pgoff)
3534 */ <<
3535 static vm_fault_t filemap_map_folio_range(str <<
3536 struct folio *folio, <<
3537 unsigned long addr, u <<
3538 unsigned long *rss, u <<
3539 { 3162 {
3540 vm_fault_t ret = 0; !! 3163 return next_uptodate_page(xas_find(xas, end_pgoff),
3541 struct page *page = folio_page(folio, !! 3164 mapping, xas, end_pgoff);
3542 unsigned int count = 0; <<
3543 pte_t *old_ptep = vmf->pte; <<
3544 <<
3545 do { <<
3546 if (PageHWPoison(page + count <<
3547 goto skip; <<
3548 <<
3549 /* <<
3550 * If there are too many foli <<
3551 * in a file, they will proba <<
3552 * In such situation, read-ah <<
3553 * Don't decrease mmap_miss i <<
3554 * we can stop read-ahead. <<
3555 */ <<
3556 if (!folio_test_workingset(fo <<
3557 (*mmap_miss)++; <<
3558 <<
3559 /* <<
3560 * NOTE: If there're PTE mark <<
3561 * handled in the specific fa <<
3562 * fault-around logic. <<
3563 */ <<
3564 if (!pte_none(ptep_get(&vmf-> <<
3565 goto skip; <<
3566 <<
3567 count++; <<
3568 continue; <<
3569 skip: <<
3570 if (count) { <<
3571 set_pte_range(vmf, fo <<
3572 *rss += count; <<
3573 folio_ref_add(folio, <<
3574 if (in_range(vmf->add <<
3575 ret = VM_FAUL <<
3576 } <<
3577 <<
3578 count++; <<
3579 page += count; <<
3580 vmf->pte += count; <<
3581 addr += count * PAGE_SIZE; <<
3582 count = 0; <<
3583 } while (--nr_pages > 0); <<
3584 <<
3585 if (count) { <<
3586 set_pte_range(vmf, folio, pag <<
3587 *rss += count; <<
3588 folio_ref_add(folio, count); <<
3589 if (in_range(vmf->address, ad <<
3590 ret = VM_FAULT_NOPAGE <<
3591 } <<
3592 <<
3593 vmf->pte = old_ptep; <<
3594 <<
3595 return ret; <<
3596 } 3165 }
3597 3166
3598 static vm_fault_t filemap_map_order0_folio(st !! 3167 static inline struct page *next_map_page(struct address_space *mapping,
3599 struct folio *folio, unsigned !! 3168 struct xa_state *xas,
3600 unsigned long *rss, unsigned !! 3169 pgoff_t end_pgoff)
3601 { 3170 {
3602 vm_fault_t ret = 0; !! 3171 return next_uptodate_page(xas_next_entry(xas, end_pgoff),
3603 struct page *page = &folio->page; !! 3172 mapping, xas, end_pgoff);
3604 <<
3605 if (PageHWPoison(page)) <<
3606 return ret; <<
3607 <<
3608 /* See comment of filemap_map_folio_r <<
3609 if (!folio_test_workingset(folio)) <<
3610 (*mmap_miss)++; <<
3611 <<
3612 /* <<
3613 * NOTE: If there're PTE markers, we' <<
3614 * handled in the specific fault path <<
3615 * the fault-around logic. <<
3616 */ <<
3617 if (!pte_none(ptep_get(vmf->pte))) <<
3618 return ret; <<
3619 <<
3620 if (vmf->address == addr) <<
3621 ret = VM_FAULT_NOPAGE; <<
3622 <<
3623 set_pte_range(vmf, folio, page, 1, ad <<
3624 (*rss)++; <<
3625 folio_ref_inc(folio); <<
3626 <<
3627 return ret; <<
3628 } 3173 }
3629 3174
3630 vm_fault_t filemap_map_pages(struct vm_fault 3175 vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3631 pgoff_t start_pg 3176 pgoff_t start_pgoff, pgoff_t end_pgoff)
3632 { 3177 {
3633 struct vm_area_struct *vma = vmf->vma 3178 struct vm_area_struct *vma = vmf->vma;
3634 struct file *file = vma->vm_file; 3179 struct file *file = vma->vm_file;
3635 struct address_space *mapping = file- 3180 struct address_space *mapping = file->f_mapping;
3636 pgoff_t file_end, last_pgoff = start_ !! 3181 pgoff_t last_pgoff = start_pgoff;
3637 unsigned long addr; 3182 unsigned long addr;
3638 XA_STATE(xas, &mapping->i_pages, star 3183 XA_STATE(xas, &mapping->i_pages, start_pgoff);
3639 struct folio *folio; !! 3184 struct page *head, *page;
>> 3185 unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
3640 vm_fault_t ret = 0; 3186 vm_fault_t ret = 0;
3641 unsigned long rss = 0; <<
3642 unsigned int nr_pages = 0, mmap_miss <<
3643 3187
3644 rcu_read_lock(); 3188 rcu_read_lock();
3645 folio = next_uptodate_folio(&xas, map !! 3189 head = first_map_page(mapping, &xas, end_pgoff);
3646 if (!folio) !! 3190 if (!head)
3647 goto out; 3191 goto out;
3648 3192
3649 if (filemap_map_pmd(vmf, folio, start !! 3193 if (filemap_map_pmd(vmf, head)) {
3650 ret = VM_FAULT_NOPAGE; 3194 ret = VM_FAULT_NOPAGE;
3651 goto out; 3195 goto out;
3652 } 3196 }
3653 3197
3654 addr = vma->vm_start + ((start_pgoff 3198 addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3655 vmf->pte = pte_offset_map_lock(vma->v 3199 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3656 if (!vmf->pte) { <<
3657 folio_unlock(folio); <<
3658 folio_put(folio); <<
3659 goto out; <<
3660 } <<
3661 <<
3662 file_end = DIV_ROUND_UP(i_size_read(m <<
3663 if (end_pgoff > file_end) <<
3664 end_pgoff = file_end; <<
3665 <<
3666 folio_type = mm_counter_file(folio); <<
3667 do { 3200 do {
3668 unsigned long end; !! 3201 page = find_subpage(head, xas.xa_index);
>> 3202 if (PageHWPoison(page))
>> 3203 goto unlock;
>> 3204
>> 3205 if (mmap_miss > 0)
>> 3206 mmap_miss--;
3669 3207
3670 addr += (xas.xa_index - last_ 3208 addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3671 vmf->pte += xas.xa_index - la 3209 vmf->pte += xas.xa_index - last_pgoff;
3672 last_pgoff = xas.xa_index; 3210 last_pgoff = xas.xa_index;
3673 end = folio_next_index(folio) <<
3674 nr_pages = min(end, end_pgoff <<
3675 3211
3676 if (!folio_test_large(folio)) !! 3212 if (!pte_none(*vmf->pte))
3677 ret |= filemap_map_or !! 3213 goto unlock;
3678 folio !! 3214
3679 else !! 3215 /* We're about to handle the fault */
3680 ret |= filemap_map_fo !! 3216 if (vmf->address == addr)
3681 xas.x !! 3217 ret = VM_FAULT_NOPAGE;
3682 nr_pa !! 3218
3683 !! 3219 do_set_pte(vmf, page, addr);
3684 folio_unlock(folio); !! 3220 /* no need to invalidate: a not-present page won't be cached */
3685 folio_put(folio); !! 3221 update_mmu_cache(vma, addr, vmf->pte);
3686 } while ((folio = next_uptodate_folio !! 3222 unlock_page(head);
3687 add_mm_counter(vma->vm_mm, folio_type !! 3223 continue;
>> 3224 unlock:
>> 3225 unlock_page(head);
>> 3226 put_page(head);
>> 3227 } while ((head = next_map_page(mapping, &xas, end_pgoff)) != NULL);
3688 pte_unmap_unlock(vmf->pte, vmf->ptl); 3228 pte_unmap_unlock(vmf->pte, vmf->ptl);
3689 trace_mm_filemap_map_pages(mapping, s <<
3690 out: 3229 out:
3691 rcu_read_unlock(); 3230 rcu_read_unlock();
3692 !! 3231 WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
3693 mmap_miss_saved = READ_ONCE(file->f_r <<
3694 if (mmap_miss >= mmap_miss_saved) <<
3695 WRITE_ONCE(file->f_ra.mmap_mi <<
3696 else <<
3697 WRITE_ONCE(file->f_ra.mmap_mi <<
3698 <<
3699 return ret; 3232 return ret;
3700 } 3233 }
3701 EXPORT_SYMBOL(filemap_map_pages); 3234 EXPORT_SYMBOL(filemap_map_pages);
3702 3235
3703 vm_fault_t filemap_page_mkwrite(struct vm_fau 3236 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3704 { 3237 {
3705 struct address_space *mapping = vmf-> 3238 struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3706 struct folio *folio = page_folio(vmf- !! 3239 struct page *page = vmf->page;
3707 vm_fault_t ret = VM_FAULT_LOCKED; 3240 vm_fault_t ret = VM_FAULT_LOCKED;
3708 3241
3709 sb_start_pagefault(mapping->host->i_s 3242 sb_start_pagefault(mapping->host->i_sb);
3710 file_update_time(vmf->vma->vm_file); 3243 file_update_time(vmf->vma->vm_file);
3711 folio_lock(folio); !! 3244 lock_page(page);
3712 if (folio->mapping != mapping) { !! 3245 if (page->mapping != mapping) {
3713 folio_unlock(folio); !! 3246 unlock_page(page);
3714 ret = VM_FAULT_NOPAGE; 3247 ret = VM_FAULT_NOPAGE;
3715 goto out; 3248 goto out;
3716 } 3249 }
3717 /* 3250 /*
3718 * We mark the folio dirty already he !! 3251 * We mark the page dirty already here so that when freeze is in
3719 * progress, we are guaranteed that w 3252 * progress, we are guaranteed that writeback during freezing will
3720 * see the dirty folio and writeprote !! 3253 * see the dirty page and writeprotect it again.
3721 */ 3254 */
3722 folio_mark_dirty(folio); !! 3255 set_page_dirty(page);
3723 folio_wait_stable(folio); !! 3256 wait_for_stable_page(page);
3724 out: 3257 out:
3725 sb_end_pagefault(mapping->host->i_sb) 3258 sb_end_pagefault(mapping->host->i_sb);
3726 return ret; 3259 return ret;
3727 } 3260 }
3728 3261
3729 const struct vm_operations_struct generic_fil 3262 const struct vm_operations_struct generic_file_vm_ops = {
3730 .fault = filemap_fault, 3263 .fault = filemap_fault,
3731 .map_pages = filemap_map_pages, 3264 .map_pages = filemap_map_pages,
3732 .page_mkwrite = filemap_page_mkwrit 3265 .page_mkwrite = filemap_page_mkwrite,
3733 }; 3266 };
3734 3267
3735 /* This is used for a general mmap of a disk 3268 /* This is used for a general mmap of a disk file */
3736 3269
3737 int generic_file_mmap(struct file *file, stru 3270 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3738 { 3271 {
3739 struct address_space *mapping = file- 3272 struct address_space *mapping = file->f_mapping;
3740 3273
3741 if (!mapping->a_ops->read_folio) !! 3274 if (!mapping->a_ops->readpage)
3742 return -ENOEXEC; 3275 return -ENOEXEC;
3743 file_accessed(file); 3276 file_accessed(file);
3744 vma->vm_ops = &generic_file_vm_ops; 3277 vma->vm_ops = &generic_file_vm_ops;
3745 return 0; 3278 return 0;
3746 } 3279 }
3747 3280
3748 /* 3281 /*
3749 * This is for filesystems which do not imple 3282 * This is for filesystems which do not implement ->writepage.
3750 */ 3283 */
3751 int generic_file_readonly_mmap(struct file *f 3284 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3752 { 3285 {
3753 if (vma_is_shared_maywrite(vma)) !! 3286 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3754 return -EINVAL; 3287 return -EINVAL;
3755 return generic_file_mmap(file, vma); 3288 return generic_file_mmap(file, vma);
3756 } 3289 }
3757 #else 3290 #else
3758 vm_fault_t filemap_page_mkwrite(struct vm_fau 3291 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3759 { 3292 {
3760 return VM_FAULT_SIGBUS; 3293 return VM_FAULT_SIGBUS;
3761 } 3294 }
3762 int generic_file_mmap(struct file *file, stru 3295 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3763 { 3296 {
3764 return -ENOSYS; 3297 return -ENOSYS;
3765 } 3298 }
3766 int generic_file_readonly_mmap(struct file *f 3299 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3767 { 3300 {
3768 return -ENOSYS; 3301 return -ENOSYS;
3769 } 3302 }
3770 #endif /* CONFIG_MMU */ 3303 #endif /* CONFIG_MMU */
3771 3304
3772 EXPORT_SYMBOL(filemap_page_mkwrite); 3305 EXPORT_SYMBOL(filemap_page_mkwrite);
3773 EXPORT_SYMBOL(generic_file_mmap); 3306 EXPORT_SYMBOL(generic_file_mmap);
3774 EXPORT_SYMBOL(generic_file_readonly_mmap); 3307 EXPORT_SYMBOL(generic_file_readonly_mmap);
3775 3308
3776 static struct folio *do_read_cache_folio(stru !! 3309 static struct page *wait_on_page_read(struct page *page)
3777 pgoff_t index, filler_t fille <<
3778 { 3310 {
3779 struct folio *folio; !! 3311 if (!IS_ERR(page)) {
3780 int err; !! 3312 wait_on_page_locked(page);
>> 3313 if (!PageUptodate(page)) {
>> 3314 put_page(page);
>> 3315 page = ERR_PTR(-EIO);
>> 3316 }
>> 3317 }
>> 3318 return page;
>> 3319 }
3781 3320
3782 if (!filler) !! 3321 static struct page *do_read_cache_page(struct address_space *mapping,
3783 filler = mapping->a_ops->read !! 3322 pgoff_t index,
>> 3323 int (*filler)(void *, struct page *),
>> 3324 void *data,
>> 3325 gfp_t gfp)
>> 3326 {
>> 3327 struct page *page;
>> 3328 int err;
3784 repeat: 3329 repeat:
3785 folio = filemap_get_folio(mapping, in !! 3330 page = find_get_page(mapping, index);
3786 if (IS_ERR(folio)) { !! 3331 if (!page) {
3787 folio = filemap_alloc_folio(g !! 3332 page = __page_cache_alloc(gfp);
3788 m !! 3333 if (!page)
3789 if (!folio) <<
3790 return ERR_PTR(-ENOME 3334 return ERR_PTR(-ENOMEM);
3791 index = mapping_align_index(m !! 3335 err = add_to_page_cache_lru(page, mapping, index, gfp);
3792 err = filemap_add_folio(mappi <<
3793 if (unlikely(err)) { 3336 if (unlikely(err)) {
3794 folio_put(folio); !! 3337 put_page(page);
3795 if (err == -EEXIST) 3338 if (err == -EEXIST)
3796 goto repeat; 3339 goto repeat;
3797 /* Presumably ENOMEM 3340 /* Presumably ENOMEM for xarray node */
3798 return ERR_PTR(err); 3341 return ERR_PTR(err);
3799 } 3342 }
3800 3343
3801 goto filler; !! 3344 filler:
>> 3345 if (filler)
>> 3346 err = filler(data, page);
>> 3347 else
>> 3348 err = mapping->a_ops->readpage(data, page);
>> 3349
>> 3350 if (err < 0) {
>> 3351 put_page(page);
>> 3352 return ERR_PTR(err);
>> 3353 }
>> 3354
>> 3355 page = wait_on_page_read(page);
>> 3356 if (IS_ERR(page))
>> 3357 return page;
>> 3358 goto out;
3802 } 3359 }
3803 if (folio_test_uptodate(folio)) !! 3360 if (PageUptodate(page))
3804 goto out; 3361 goto out;
3805 3362
3806 if (!folio_trylock(folio)) { !! 3363 /*
3807 folio_put_wait_locked(folio, !! 3364 * Page is not up to date and may be locked due to one of the following
3808 goto repeat; !! 3365 * case a: Page is being filled and the page lock is held
3809 } !! 3366 * case b: Read/write error clearing the page uptodate status
>> 3367 * case c: Truncation in progress (page locked)
>> 3368 * case d: Reclaim in progress
>> 3369 *
>> 3370 * Case a, the page will be up to date when the page is unlocked.
>> 3371 * There is no need to serialise on the page lock here as the page
>> 3372 * is pinned so the lock gives no additional protection. Even if the
>> 3373 * page is truncated, the data is still valid if PageUptodate as
>> 3374 * it's a race vs truncate race.
>> 3375 * Case b, the page will not be up to date
>> 3376 * Case c, the page may be truncated but in itself, the data may still
>> 3377 * be valid after IO completes as it's a read vs truncate race. The
>> 3378 * operation must restart if the page is not uptodate on unlock but
>> 3379 * otherwise serialising on page lock to stabilise the mapping gives
>> 3380 * no additional guarantees to the caller as the page lock is
>> 3381 * released before return.
>> 3382 * Case d, similar to truncation. If reclaim holds the page lock, it
>> 3383 * will be a race with remove_mapping that determines if the mapping
>> 3384 * is valid on unlock but otherwise the data is valid and there is
>> 3385 * no need to serialise with page lock.
>> 3386 *
>> 3387 * As the page lock gives no additional guarantee, we optimistically
>> 3388 * wait on the page to be unlocked and check if it's up to date and
>> 3389 * use the page if it is. Otherwise, the page lock is required to
>> 3390 * distinguish between the different cases. The motivation is that we
>> 3391 * avoid spurious serialisations and wakeups when multiple processes
>> 3392 * wait on the same page for IO to complete.
>> 3393 */
>> 3394 wait_on_page_locked(page);
>> 3395 if (PageUptodate(page))
>> 3396 goto out;
>> 3397
>> 3398 /* Distinguish between all the cases under the safety of the lock */
>> 3399 lock_page(page);
3810 3400
3811 /* Folio was truncated from mapping * !! 3401 /* Case c or d, restart the operation */
3812 if (!folio->mapping) { !! 3402 if (!page->mapping) {
3813 folio_unlock(folio); !! 3403 unlock_page(page);
3814 folio_put(folio); !! 3404 put_page(page);
3815 goto repeat; 3405 goto repeat;
3816 } 3406 }
3817 3407
3818 /* Someone else locked and filled the 3408 /* Someone else locked and filled the page in a very small window */
3819 if (folio_test_uptodate(folio)) { !! 3409 if (PageUptodate(page)) {
3820 folio_unlock(folio); !! 3410 unlock_page(page);
3821 goto out; 3411 goto out;
3822 } 3412 }
3823 3413
3824 filler: !! 3414 /*
3825 err = filemap_read_folio(file, filler !! 3415 * A previous I/O error may have been due to temporary
3826 if (err) { !! 3416 * failures.
3827 folio_put(folio); !! 3417 * Clear page error before actual read, PG_error will be
3828 if (err == AOP_TRUNCATED_PAGE !! 3418 * set again if read page fails.
3829 goto repeat; !! 3419 */
3830 return ERR_PTR(err); !! 3420 ClearPageError(page);
3831 } !! 3421 goto filler;
3832 3422
3833 out: 3423 out:
3834 folio_mark_accessed(folio); !! 3424 mark_page_accessed(page);
3835 return folio; !! 3425 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 } 3426 }
3860 EXPORT_SYMBOL(read_cache_folio); <<
3861 3427
3862 /** 3428 /**
3863 * mapping_read_folio_gfp - Read into page ca !! 3429 * read_cache_page - read into page cache, fill it if needed
3864 * @mapping: The address_space for the fol !! 3430 * @mapping: the page's address_space
3865 * @index: The index that the allocated !! 3431 * @index: the page index
3866 * @gfp: The page allocator flags to u !! 3432 * @filler: function to perform the read
3867 * !! 3433 * @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 * 3434 *
3871 * The most likely error from this function i !! 3435 * Read into the page cache. If a page already exists, and PageUptodate() is
3872 * possible and so is EINTR. If ->read_folio !! 3436 * not set, try to fill the page and wait for it to become unlocked.
3873 * that will be returned to the caller. <<
3874 * 3437 *
3875 * The function expects mapping->invalidate_l !! 3438 * If the page does not get brought uptodate, return -EIO.
3876 * 3439 *
3877 * Return: Uptodate folio on success, ERR_PTR !! 3440 * Return: up to date page on success, ERR_PTR() on failure.
3878 */ 3441 */
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 3442 struct page *read_cache_page(struct address_space *mapping,
3898 pgoff_t index, filler !! 3443 pgoff_t index,
>> 3444 int (*filler)(void *, struct page *),
>> 3445 void *data)
3899 { 3446 {
3900 return do_read_cache_page(mapping, in !! 3447 return do_read_cache_page(mapping, index, filler, data,
3901 mapping_gfp_mask(mapp 3448 mapping_gfp_mask(mapping));
3902 } 3449 }
3903 EXPORT_SYMBOL(read_cache_page); 3450 EXPORT_SYMBOL(read_cache_page);
3904 3451
3905 /** 3452 /**
3906 * read_cache_page_gfp - read into page cache 3453 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3907 * @mapping: the page's address_space 3454 * @mapping: the page's address_space
3908 * @index: the page index 3455 * @index: the page index
3909 * @gfp: the page allocator flags to u 3456 * @gfp: the page allocator flags to use if allocating
3910 * 3457 *
3911 * This is the same as "read_mapping_page(map 3458 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3912 * any new page allocations done using the sp 3459 * any new page allocations done using the specified allocation flags.
3913 * 3460 *
3914 * If the page does not get brought uptodate, 3461 * If the page does not get brought uptodate, return -EIO.
3915 * 3462 *
3916 * The function expects mapping->invalidate_l <<
3917 * <<
3918 * Return: up to date page on success, ERR_PT 3463 * Return: up to date page on success, ERR_PTR() on failure.
3919 */ 3464 */
3920 struct page *read_cache_page_gfp(struct addre 3465 struct page *read_cache_page_gfp(struct address_space *mapping,
3921 pgoff_t index 3466 pgoff_t index,
3922 gfp_t gfp) 3467 gfp_t gfp)
3923 { 3468 {
3924 return do_read_cache_page(mapping, in 3469 return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3925 } 3470 }
3926 EXPORT_SYMBOL(read_cache_page_gfp); 3471 EXPORT_SYMBOL(read_cache_page_gfp);
3927 3472
>> 3473 int pagecache_write_begin(struct file *file, struct address_space *mapping,
>> 3474 loff_t pos, unsigned len, unsigned flags,
>> 3475 struct page **pagep, void **fsdata)
>> 3476 {
>> 3477 const struct address_space_operations *aops = mapping->a_ops;
>> 3478
>> 3479 return aops->write_begin(file, mapping, pos, len, flags,
>> 3480 pagep, fsdata);
>> 3481 }
>> 3482 EXPORT_SYMBOL(pagecache_write_begin);
>> 3483
>> 3484 int pagecache_write_end(struct file *file, struct address_space *mapping,
>> 3485 loff_t pos, unsigned len, unsigned copied,
>> 3486 struct page *page, void *fsdata)
>> 3487 {
>> 3488 const struct address_space_operations *aops = mapping->a_ops;
>> 3489
>> 3490 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
>> 3491 }
>> 3492 EXPORT_SYMBOL(pagecache_write_end);
>> 3493
3928 /* 3494 /*
3929 * Warn about a page cache invalidation failu 3495 * Warn about a page cache invalidation failure during a direct I/O write.
3930 */ 3496 */
3931 static void dio_warn_stale_pagecache(struct f !! 3497 void dio_warn_stale_pagecache(struct file *filp)
3932 { 3498 {
3933 static DEFINE_RATELIMIT_STATE(_rs, 86 3499 static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3934 char pathname[128]; 3500 char pathname[128];
3935 char *path; 3501 char *path;
3936 3502
3937 errseq_set(&filp->f_mapping->wb_err, 3503 errseq_set(&filp->f_mapping->wb_err, -EIO);
3938 if (__ratelimit(&_rs)) { 3504 if (__ratelimit(&_rs)) {
3939 path = file_path(filp, pathna 3505 path = file_path(filp, pathname, sizeof(pathname));
3940 if (IS_ERR(path)) 3506 if (IS_ERR(path))
3941 path = "(unknown)"; 3507 path = "(unknown)";
3942 pr_crit("Page cache invalidat 3508 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3943 pr_crit("File: %s PID: %d Com 3509 pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3944 current->comm); 3510 current->comm);
3945 } 3511 }
3946 } 3512 }
3947 3513
3948 void kiocb_invalidate_post_direct_write(struc <<
3949 { <<
3950 struct address_space *mapping = iocb- <<
3951 <<
3952 if (mapping->nrpages && <<
3953 invalidate_inode_pages2_range(map <<
3954 iocb->ki_pos >> PAGE_ <<
3955 (iocb->ki_pos + count <<
3956 dio_warn_stale_pagecache(iocb <<
3957 } <<
3958 <<
3959 ssize_t 3514 ssize_t
3960 generic_file_direct_write(struct kiocb *iocb, 3515 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3961 { 3516 {
3962 struct address_space *mapping = iocb- !! 3517 struct file *file = iocb->ki_filp;
3963 size_t write_len = iov_iter_count(fro !! 3518 struct address_space *mapping = file->f_mapping;
3964 ssize_t written; !! 3519 struct inode *inode = mapping->host;
>> 3520 loff_t pos = iocb->ki_pos;
>> 3521 ssize_t written;
>> 3522 size_t write_len;
>> 3523 pgoff_t end;
>> 3524
>> 3525 write_len = iov_iter_count(from);
>> 3526 end = (pos + write_len - 1) >> PAGE_SHIFT;
>> 3527
>> 3528 if (iocb->ki_flags & IOCB_NOWAIT) {
>> 3529 /* If there are pages to writeback, return */
>> 3530 if (filemap_range_has_page(file->f_mapping, pos,
>> 3531 pos + write_len - 1))
>> 3532 return -EAGAIN;
>> 3533 } else {
>> 3534 written = filemap_write_and_wait_range(mapping, pos,
>> 3535 pos + write_len - 1);
>> 3536 if (written)
>> 3537 goto out;
>> 3538 }
3965 3539
3966 /* 3540 /*
>> 3541 * After a write we want buffered reads to be sure to go to disk to get
>> 3542 * the new data. We invalidate clean cached page from the region we're
>> 3543 * about to write. We do this *before* the write so that we can return
>> 3544 * without clobbering -EIOCBQUEUED from ->direct_IO().
>> 3545 */
>> 3546 written = invalidate_inode_pages2_range(mapping,
>> 3547 pos >> PAGE_SHIFT, end);
>> 3548 /*
3967 * If a page can not be invalidated, 3549 * If a page can not be invalidated, return 0 to fall back
3968 * to buffered write. 3550 * to buffered write.
3969 */ 3551 */
3970 written = kiocb_invalidate_pages(iocb <<
3971 if (written) { 3552 if (written) {
3972 if (written == -EBUSY) 3553 if (written == -EBUSY)
3973 return 0; 3554 return 0;
3974 return written; !! 3555 goto out;
3975 } 3556 }
3976 3557
3977 written = mapping->a_ops->direct_IO(i 3558 written = mapping->a_ops->direct_IO(iocb, from);
3978 3559
3979 /* 3560 /*
3980 * Finally, try again to invalidate c 3561 * Finally, try again to invalidate clean pages which might have been
3981 * cached by non-direct readahead, or 3562 * cached by non-direct readahead, or faulted in by get_user_pages()
3982 * if the source of the write was an 3563 * if the source of the write was an mmap'ed region of the file
3983 * we're writing. Either one is a pr 3564 * we're writing. Either one is a pretty crazy thing to do,
3984 * so we don't support it 100%. If t 3565 * so we don't support it 100%. If this invalidation
3985 * fails, tough, the write still work 3566 * fails, tough, the write still worked...
3986 * 3567 *
3987 * Most of the time we do not need th 3568 * Most of the time we do not need this since dio_complete() will do
3988 * the invalidation for us. However t 3569 * the invalidation for us. However there are some file systems that
3989 * do not end up with dio_complete() 3570 * do not end up with dio_complete() being called, so let's not break
3990 * them by removing it completely. 3571 * them by removing it completely.
3991 * 3572 *
3992 * Noticeable example is a blkdev_dir 3573 * Noticeable example is a blkdev_direct_IO().
3993 * 3574 *
3994 * Skip invalidation for async writes 3575 * Skip invalidation for async writes or if mapping has no pages.
3995 */ 3576 */
3996 if (written > 0) { !! 3577 if (written > 0 && mapping->nrpages &&
3997 struct inode *inode = mapping !! 3578 invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3998 loff_t pos = iocb->ki_pos; !! 3579 dio_warn_stale_pagecache(file);
3999 3580
4000 kiocb_invalidate_post_direct_ !! 3581 if (written > 0) {
4001 pos += written; 3582 pos += written;
4002 write_len -= written; 3583 write_len -= written;
4003 if (pos > i_size_read(inode) 3584 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
4004 i_size_write(inode, p 3585 i_size_write(inode, pos);
4005 mark_inode_dirty(inod 3586 mark_inode_dirty(inode);
4006 } 3587 }
4007 iocb->ki_pos = pos; 3588 iocb->ki_pos = pos;
4008 } 3589 }
4009 if (written != -EIOCBQUEUED) 3590 if (written != -EIOCBQUEUED)
4010 iov_iter_revert(from, write_l 3591 iov_iter_revert(from, write_len - iov_iter_count(from));
>> 3592 out:
4011 return written; 3593 return written;
4012 } 3594 }
4013 EXPORT_SYMBOL(generic_file_direct_write); 3595 EXPORT_SYMBOL(generic_file_direct_write);
4014 3596
4015 ssize_t generic_perform_write(struct kiocb *i !! 3597 /*
>> 3598 * Find or create a page at the given pagecache position. Return the locked
>> 3599 * page. This function is specifically for buffered writes.
>> 3600 */
>> 3601 struct page *grab_cache_page_write_begin(struct address_space *mapping,
>> 3602 pgoff_t index, unsigned flags)
>> 3603 {
>> 3604 struct page *page;
>> 3605 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
>> 3606
>> 3607 if (flags & AOP_FLAG_NOFS)
>> 3608 fgp_flags |= FGP_NOFS;
>> 3609
>> 3610 page = pagecache_get_page(mapping, index, fgp_flags,
>> 3611 mapping_gfp_mask(mapping));
>> 3612 if (page)
>> 3613 wait_for_stable_page(page);
>> 3614
>> 3615 return page;
>> 3616 }
>> 3617 EXPORT_SYMBOL(grab_cache_page_write_begin);
>> 3618
>> 3619 ssize_t generic_perform_write(struct file *file,
>> 3620 struct iov_iter *i, loff_t pos)
4016 { 3621 {
4017 struct file *file = iocb->ki_filp; <<
4018 loff_t pos = iocb->ki_pos; <<
4019 struct address_space *mapping = file- 3622 struct address_space *mapping = file->f_mapping;
4020 const struct address_space_operations 3623 const struct address_space_operations *a_ops = mapping->a_ops;
4021 size_t chunk = mapping_max_folio_size <<
4022 long status = 0; 3624 long status = 0;
4023 ssize_t written = 0; 3625 ssize_t written = 0;
>> 3626 unsigned int flags = 0;
4024 3627
4025 do { 3628 do {
4026 struct folio *folio; !! 3629 struct page *page;
4027 size_t offset; /* Of !! 3630 unsigned long offset; /* Offset into pagecache page */
4028 size_t bytes; /* By !! 3631 unsigned long bytes; /* Bytes to write to page */
4029 size_t copied; /* By 3632 size_t copied; /* Bytes copied from user */
4030 void *fsdata = NULL; !! 3633 void *fsdata;
4031 3634
4032 bytes = iov_iter_count(i); !! 3635 offset = (pos & (PAGE_SIZE - 1));
4033 retry: !! 3636 bytes = min_t(unsigned long, PAGE_SIZE - offset,
4034 offset = pos & (chunk - 1); !! 3637 iov_iter_count(i));
4035 bytes = min(chunk - offset, b <<
4036 balance_dirty_pages_ratelimit <<
4037 3638
>> 3639 again:
4038 /* 3640 /*
4039 * Bring in the user page tha 3641 * Bring in the user page that we will copy from _first_.
4040 * Otherwise there's a nasty 3642 * Otherwise there's a nasty deadlock on copying from the
4041 * same page as we're writing 3643 * same page as we're writing to, without it being marked
4042 * up-to-date. 3644 * up-to-date.
>> 3645 *
>> 3646 * Not only is this an optimisation, but it is also required
>> 3647 * to check that the address is actually valid, when atomic
>> 3648 * usercopies are used, below.
4043 */ 3649 */
4044 if (unlikely(fault_in_iov_ite !! 3650 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
4045 status = -EFAULT; 3651 status = -EFAULT;
4046 break; 3652 break;
4047 } 3653 }
4048 3654
4049 if (fatal_signal_pending(curr 3655 if (fatal_signal_pending(current)) {
4050 status = -EINTR; 3656 status = -EINTR;
4051 break; 3657 break;
4052 } 3658 }
4053 3659
4054 status = a_ops->write_begin(f !! 3660 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
4055 !! 3661 &page, &fsdata);
4056 if (unlikely(status < 0)) 3662 if (unlikely(status < 0))
4057 break; 3663 break;
4058 3664
4059 offset = offset_in_folio(foli <<
4060 if (bytes > folio_size(folio) <<
4061 bytes = folio_size(fo <<
4062 <<
4063 if (mapping_writably_mapped(m 3665 if (mapping_writably_mapped(mapping))
4064 flush_dcache_folio(fo !! 3666 flush_dcache_page(page);
4065 3667
4066 copied = copy_folio_from_iter !! 3668 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
4067 flush_dcache_folio(folio); !! 3669 flush_dcache_page(page);
4068 3670
4069 status = a_ops->write_end(fil 3671 status = a_ops->write_end(file, mapping, pos, bytes, copied,
4070 !! 3672 page, fsdata);
4071 if (unlikely(status != copied !! 3673 if (unlikely(status < 0))
4072 iov_iter_revert(i, co !! 3674 break;
4073 if (unlikely(status < !! 3675 copied = status;
4074 break; !! 3676
4075 } <<
4076 cond_resched(); 3677 cond_resched();
4077 3678
4078 if (unlikely(status == 0)) { !! 3679 iov_iter_advance(i, copied);
>> 3680 if (unlikely(copied == 0)) {
4079 /* 3681 /*
4080 * A short copy made !! 3682 * If we were unable to copy any data at all, we must
4081 * thing entirely. M !! 3683 * fall back to a single segment length write.
4082 * halfway through, m !! 3684 *
4083 * might be severe me !! 3685 * If we didn't fallback here, we could livelock
>> 3686 * because not all segments in the iov can be copied at
>> 3687 * once without a pagefault.
4084 */ 3688 */
4085 if (chunk > PAGE_SIZE !! 3689 bytes = min_t(unsigned long, PAGE_SIZE - offset,
4086 chunk /= 2; !! 3690 iov_iter_single_seg_count(i));
4087 if (copied) { !! 3691 goto again;
4088 bytes = copie <<
4089 goto retry; <<
4090 } <<
4091 } else { <<
4092 pos += status; <<
4093 written += status; <<
4094 } 3692 }
>> 3693 pos += copied;
>> 3694 written += copied;
>> 3695
>> 3696 balance_dirty_pages_ratelimited(mapping);
4095 } while (iov_iter_count(i)); 3697 } while (iov_iter_count(i));
4096 3698
4097 if (!written) !! 3699 return written ? written : status;
4098 return status; <<
4099 iocb->ki_pos += written; <<
4100 return written; <<
4101 } 3700 }
4102 EXPORT_SYMBOL(generic_perform_write); 3701 EXPORT_SYMBOL(generic_perform_write);
4103 3702
4104 /** 3703 /**
4105 * __generic_file_write_iter - write data to 3704 * __generic_file_write_iter - write data to a file
4106 * @iocb: IO state structure (file, off 3705 * @iocb: IO state structure (file, offset, etc.)
4107 * @from: iov_iter with data to write 3706 * @from: iov_iter with data to write
4108 * 3707 *
4109 * This function does all the work needed for 3708 * This function does all the work needed for actually writing data to a
4110 * file. It does all basic checks, removes SU 3709 * file. It does all basic checks, removes SUID from the file, updates
4111 * modification times and calls proper subrou 3710 * modification times and calls proper subroutines depending on whether we
4112 * do direct IO or a standard buffered write. 3711 * do direct IO or a standard buffered write.
4113 * 3712 *
4114 * It expects i_rwsem to be grabbed unless we !! 3713 * 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. 3714 * object which does not need locking at all.
4116 * 3715 *
4117 * This function does *not* take care of sync 3716 * 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 3717 * A caller has to handle it. This is mainly due to the fact that we want to
4119 * avoid syncing under i_rwsem. !! 3718 * avoid syncing under i_mutex.
4120 * 3719 *
4121 * Return: 3720 * Return:
4122 * * number of bytes written, even for trunca 3721 * * number of bytes written, even for truncated writes
4123 * * negative error code if no data has been 3722 * * negative error code if no data has been written at all
4124 */ 3723 */
4125 ssize_t __generic_file_write_iter(struct kioc 3724 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4126 { 3725 {
4127 struct file *file = iocb->ki_filp; 3726 struct file *file = iocb->ki_filp;
4128 struct address_space *mapping = file- 3727 struct address_space *mapping = file->f_mapping;
4129 struct inode *inode = mapping->host; !! 3728 struct inode *inode = mapping->host;
4130 ssize_t ret; !! 3729 ssize_t written = 0;
>> 3730 ssize_t err;
>> 3731 ssize_t status;
>> 3732
>> 3733 /* We can write back this queue in page reclaim */
>> 3734 current->backing_dev_info = inode_to_bdi(inode);
>> 3735 err = file_remove_privs(file);
>> 3736 if (err)
>> 3737 goto out;
4131 3738
4132 ret = file_remove_privs(file); !! 3739 err = file_update_time(file);
4133 if (ret) !! 3740 if (err)
4134 return ret; !! 3741 goto out;
4135 <<
4136 ret = file_update_time(file); <<
4137 if (ret) <<
4138 return ret; <<
4139 3742
4140 if (iocb->ki_flags & IOCB_DIRECT) { 3743 if (iocb->ki_flags & IOCB_DIRECT) {
4141 ret = generic_file_direct_wri !! 3744 loff_t pos, endbyte;
>> 3745
>> 3746 written = generic_file_direct_write(iocb, from);
4142 /* 3747 /*
4143 * If the write stopped short 3748 * If the write stopped short of completing, fall back to
4144 * buffered writes. Some fil 3749 * buffered writes. Some filesystems do this for writes to
4145 * holes, for example. For D 3750 * holes, for example. For DAX files, a buffered write will
4146 * not succeed (even if it di 3751 * not succeed (even if it did, DAX does not handle dirty
4147 * page-cache pages correctly 3752 * page-cache pages correctly).
4148 */ 3753 */
4149 if (ret < 0 || !iov_iter_coun !! 3754 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
4150 return ret; !! 3755 goto out;
4151 return direct_write_fallback( <<
4152 generic_perfo <<
4153 } <<
4154 3756
4155 return generic_perform_write(iocb, fr !! 3757 status = generic_perform_write(file, from, pos = iocb->ki_pos);
>> 3758 /*
>> 3759 * If generic_perform_write() returned a synchronous error
>> 3760 * then we want to return the number of bytes which were
>> 3761 * direct-written, or the error code if that was zero. Note
>> 3762 * that this differs from normal direct-io semantics, which
>> 3763 * will return -EFOO even if some bytes were written.
>> 3764 */
>> 3765 if (unlikely(status < 0)) {
>> 3766 err = status;
>> 3767 goto out;
>> 3768 }
>> 3769 /*
>> 3770 * We need to ensure that the page cache pages are written to
>> 3771 * disk and invalidated to preserve the expected O_DIRECT
>> 3772 * semantics.
>> 3773 */
>> 3774 endbyte = pos + status - 1;
>> 3775 err = filemap_write_and_wait_range(mapping, pos, endbyte);
>> 3776 if (err == 0) {
>> 3777 iocb->ki_pos = endbyte + 1;
>> 3778 written += status;
>> 3779 invalidate_mapping_pages(mapping,
>> 3780 pos >> PAGE_SHIFT,
>> 3781 endbyte >> PAGE_SHIFT);
>> 3782 } else {
>> 3783 /*
>> 3784 * We don't know how much we wrote, so just return
>> 3785 * the number of bytes which were direct-written
>> 3786 */
>> 3787 }
>> 3788 } else {
>> 3789 written = generic_perform_write(file, from, iocb->ki_pos);
>> 3790 if (likely(written > 0))
>> 3791 iocb->ki_pos += written;
>> 3792 }
>> 3793 out:
>> 3794 current->backing_dev_info = NULL;
>> 3795 return written ? written : err;
4156 } 3796 }
4157 EXPORT_SYMBOL(__generic_file_write_iter); 3797 EXPORT_SYMBOL(__generic_file_write_iter);
4158 3798
4159 /** 3799 /**
4160 * generic_file_write_iter - write data to a 3800 * generic_file_write_iter - write data to a file
4161 * @iocb: IO state structure 3801 * @iocb: IO state structure
4162 * @from: iov_iter with data to write 3802 * @from: iov_iter with data to write
4163 * 3803 *
4164 * This is a wrapper around __generic_file_wr 3804 * This is a wrapper around __generic_file_write_iter() to be used by most
4165 * filesystems. It takes care of syncing the 3805 * filesystems. It takes care of syncing the file in case of O_SYNC file
4166 * and acquires i_rwsem as needed. !! 3806 * and acquires i_mutex as needed.
4167 * Return: 3807 * Return:
4168 * * negative error code if no data has been 3808 * * negative error code if no data has been written at all of
4169 * vfs_fsync_range() failed for a synchrono 3809 * vfs_fsync_range() failed for a synchronous write
4170 * * number of bytes written, even for trunca 3810 * * number of bytes written, even for truncated writes
4171 */ 3811 */
4172 ssize_t generic_file_write_iter(struct kiocb 3812 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4173 { 3813 {
4174 struct file *file = iocb->ki_filp; 3814 struct file *file = iocb->ki_filp;
4175 struct inode *inode = file->f_mapping 3815 struct inode *inode = file->f_mapping->host;
4176 ssize_t ret; 3816 ssize_t ret;
4177 3817
4178 inode_lock(inode); 3818 inode_lock(inode);
4179 ret = generic_write_checks(iocb, from 3819 ret = generic_write_checks(iocb, from);
4180 if (ret > 0) 3820 if (ret > 0)
4181 ret = __generic_file_write_it 3821 ret = __generic_file_write_iter(iocb, from);
4182 inode_unlock(inode); 3822 inode_unlock(inode);
4183 3823
4184 if (ret > 0) 3824 if (ret > 0)
4185 ret = generic_write_sync(iocb 3825 ret = generic_write_sync(iocb, ret);
4186 return ret; 3826 return ret;
4187 } 3827 }
4188 EXPORT_SYMBOL(generic_file_write_iter); 3828 EXPORT_SYMBOL(generic_file_write_iter);
4189 3829
4190 /** 3830 /**
4191 * filemap_release_folio() - Release fs-speci !! 3831 * 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 * 3832 *
4195 * The address_space is trying to release any !! 3833 * @page: the page which the kernel is trying to free
4196 * (presumably at folio->private). !! 3834 * @gfp_mask: memory allocation flags (and I/O mode)
4197 * 3835 *
4198 * This will also be called if the private_2 !! 3836 * The address_space is to try to release any data against the page
4199 * indicating that the folio has other metada !! 3837 * (presumably at page->private).
4200 * 3838 *
4201 * The @gfp argument specifies whether I/O ma !! 3839 * 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 !! 3840 * page is known to the local caching routines.
4203 * (__GFP_RECLAIM & __GFP_FS). <<
4204 * 3841 *
4205 * Return: %true if the release was successfu !! 3842 * The @gfp_mask argument specifies whether I/O may be performed to release
4206 */ !! 3843 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
4207 bool filemap_release_folio(struct folio *foli !! 3844 *
4208 { !! 3845 * 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 */ 3846 */
4290 static void filemap_cachestat(struct address_ !! 3847 int try_to_release_page(struct page *page, gfp_t gfp_mask)
4291 pgoff_t first_index, pgoff_t <<
4292 { 3848 {
4293 XA_STATE(xas, &mapping->i_pages, firs !! 3849 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 3850
4316 if (xas_retry(&xas, folio)) !! 3851 BUG_ON(!PageLocked(page));
4317 continue; !! 3852 if (PageWriteback(page))
4318 !! 3853 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 3854
4377 resched: !! 3855 if (mapping && mapping->a_ops->releasepage)
4378 if (need_resched()) { !! 3856 return mapping->a_ops->releasepage(page, gfp_mask);
4379 xas_pause(&xas); !! 3857 return try_to_free_buffers(page);
4380 cond_resched_rcu(); <<
4381 } <<
4382 } <<
4383 rcu_read_unlock(); <<
4384 } 3858 }
4385 3859
4386 /* !! 3860 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 3861
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.