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> <<
45 #include <linux/migrate.h> <<
46 #include <linux/pipe_fs_i.h> <<
47 #include <linux/splice.h> <<
48 #include <linux/rcupdate_wait.h> <<
49 #include <linux/sched/mm.h> <<
50 #include <asm/pgalloc.h> <<
51 #include <asm/tlbflush.h> <<
52 #include "internal.h" 44 #include "internal.h"
53 45
54 #define CREATE_TRACE_POINTS 46 #define CREATE_TRACE_POINTS
55 #include <trace/events/filemap.h> 47 #include <trace/events/filemap.h>
56 48
57 /* 49 /*
58 * FIXME: remove all knowledge of the buffer l 50 * FIXME: remove all knowledge of the buffer layer from the core VM
59 */ 51 */
60 #include <linux/buffer_head.h> /* for try_to_f 52 #include <linux/buffer_head.h> /* for try_to_free_buffers */
61 53
62 #include <asm/mman.h> 54 #include <asm/mman.h>
63 55
64 #include "swap.h" <<
65 <<
66 /* 56 /*
67 * Shared mappings implemented 30.11.1994. It' 57 * Shared mappings implemented 30.11.1994. It's not fully working yet,
68 * though. 58 * though.
69 * 59 *
70 * Shared mappings now work. 15.8.1995 Bruno. 60 * Shared mappings now work. 15.8.1995 Bruno.
71 * 61 *
72 * finished 'unifying' the page and buffer cac 62 * finished 'unifying' the page and buffer cache and SMP-threaded the
73 * page-cache, 21.05.1999, Ingo Molnar <mingo@ 63 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
74 * 64 *
75 * SMP-threaded pagemap-LRU 1999, Andrea Arcan 65 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
76 */ 66 */
77 67
78 /* 68 /*
79 * Lock ordering: 69 * Lock ordering:
80 * 70 *
81 * ->i_mmap_rwsem (truncate_page 71 * ->i_mmap_rwsem (truncate_pagecache)
82 * ->private_lock (__free_pte->b !! 72 * ->private_lock (__free_pte->__set_page_dirty_buffers)
83 * ->swap_lock (exclusive_swa 73 * ->swap_lock (exclusive_swap_page, others)
84 * ->i_pages lock 74 * ->i_pages lock
85 * 75 *
86 * ->i_rwsem !! 76 * ->i_mutex
87 * ->invalidate_lock (acquired by f !! 77 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
88 * ->i_mmap_rwsem (truncate->unm <<
89 * 78 *
90 * ->mmap_lock !! 79 * ->mmap_sem
91 * ->i_mmap_rwsem 80 * ->i_mmap_rwsem
92 * ->page_table_lock or pte_lock (vario 81 * ->page_table_lock or pte_lock (various, mainly in memory.c)
93 * ->i_pages lock (arch-dependen 82 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
94 * 83 *
95 * ->mmap_lock !! 84 * ->mmap_sem
96 * ->invalidate_lock (filemap_fault !! 85 * ->lock_page (access_process_vm)
97 * ->lock_page (filemap_fault <<
98 * 86 *
99 * ->i_rwsem (generic_perfo !! 87 * ->i_mutex (generic_perform_write)
100 * ->mmap_lock (fault_in_read !! 88 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
101 * 89 *
102 * bdi->wb.list_lock 90 * bdi->wb.list_lock
103 * sb_lock (fs/fs-writeba 91 * sb_lock (fs/fs-writeback.c)
104 * ->i_pages lock (__sync_single 92 * ->i_pages lock (__sync_single_inode)
105 * 93 *
106 * ->i_mmap_rwsem 94 * ->i_mmap_rwsem
107 * ->anon_vma.lock (vma_merge) !! 95 * ->anon_vma.lock (vma_adjust)
108 * 96 *
109 * ->anon_vma.lock 97 * ->anon_vma.lock
110 * ->page_table_lock or pte_lock (anon_ 98 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
111 * 99 *
112 * ->page_table_lock or pte_lock 100 * ->page_table_lock or pte_lock
113 * ->swap_lock (try_to_unmap_ 101 * ->swap_lock (try_to_unmap_one)
114 * ->private_lock (try_to_unmap_ 102 * ->private_lock (try_to_unmap_one)
115 * ->i_pages lock (try_to_unmap_ 103 * ->i_pages lock (try_to_unmap_one)
116 * ->lruvec->lru_lock (follow_page_m !! 104 * ->pgdat->lru_lock (follow_page->mark_page_accessed)
117 * ->lruvec->lru_lock (check_pte_ran !! 105 * ->pgdat->lru_lock (check_pte_range->isolate_lru_page)
118 * ->private_lock (folio_remove_ !! 106 * ->private_lock (page_remove_rmap->set_page_dirty)
119 * ->i_pages lock (folio_remove_ !! 107 * ->i_pages lock (page_remove_rmap->set_page_dirty)
120 * bdi.wb->list_lock (folio_remove_ !! 108 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
121 * ->inode->i_lock (folio_remove_ !! 109 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
122 * ->memcg->move_lock (folio_remove_ !! 110 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
123 * bdi.wb->list_lock (zap_pte_range 111 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
124 * ->inode->i_lock (zap_pte_range 112 * ->inode->i_lock (zap_pte_range->set_page_dirty)
125 * ->private_lock (zap_pte_range !! 113 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
>> 114 *
>> 115 * ->i_mmap_rwsem
>> 116 * ->tasklist_lock (memory_failure, collect_procs_ao)
126 */ 117 */
127 118
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 119 static void page_cache_delete(struct address_space *mapping,
138 struct foli !! 120 struct page *page, void *shadow)
139 { 121 {
140 XA_STATE(xas, &mapping->i_pages, folio !! 122 XA_STATE(xas, &mapping->i_pages, page->index);
141 long nr = 1; !! 123 unsigned int nr = 1;
142 124
143 mapping_set_update(&xas, mapping); 125 mapping_set_update(&xas, mapping);
144 126
145 xas_set_order(&xas, folio->index, foli !! 127 /* hugetlb pages are represented by a single entry in the xarray */
146 nr = folio_nr_pages(folio); !! 128 if (!PageHuge(page)) {
>> 129 xas_set_order(&xas, page->index, compound_order(page));
>> 130 nr = compound_nr(page);
>> 131 }
147 132
148 VM_BUG_ON_FOLIO(!folio_test_locked(fol !! 133 VM_BUG_ON_PAGE(!PageLocked(page), page);
>> 134 VM_BUG_ON_PAGE(PageTail(page), page);
>> 135 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
149 136
150 xas_store(&xas, shadow); 137 xas_store(&xas, shadow);
151 xas_init_marks(&xas); 138 xas_init_marks(&xas);
152 139
153 folio->mapping = NULL; !! 140 page->mapping = NULL;
154 /* Leave page->index set: truncation l 141 /* Leave page->index set: truncation lookup relies upon it */
>> 142
>> 143 if (shadow) {
>> 144 mapping->nrexceptional += nr;
>> 145 /*
>> 146 * Make sure the nrexceptional update is committed before
>> 147 * the nrpages update so that final truncate racing
>> 148 * with reclaim does not see both counters 0 at the
>> 149 * same time and miss a shadow entry.
>> 150 */
>> 151 smp_wmb();
>> 152 }
155 mapping->nrpages -= nr; 153 mapping->nrpages -= nr;
156 } 154 }
157 155
158 static void filemap_unaccount_folio(struct add !! 156 static void unaccount_page_cache_page(struct address_space *mapping,
159 struct folio *folio) !! 157 struct page *page)
160 { 158 {
161 long nr; !! 159 int nr;
>> 160
>> 161 /*
>> 162 * if we're uptodate, flush out into the cleancache, otherwise
>> 163 * invalidate any existing cleancache entries. We can't leave
>> 164 * stale data around in the cleancache once our page is gone
>> 165 */
>> 166 if (PageUptodate(page) && PageMappedToDisk(page))
>> 167 cleancache_put_page(page);
>> 168 else
>> 169 cleancache_invalidate_page(mapping, page);
>> 170
>> 171 VM_BUG_ON_PAGE(PageTail(page), page);
>> 172 VM_BUG_ON_PAGE(page_mapped(page), page);
>> 173 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
>> 174 int mapcount;
162 175
163 VM_BUG_ON_FOLIO(folio_mapped(folio), f <<
164 if (!IS_ENABLED(CONFIG_DEBUG_VM) && un <<
165 pr_alert("BUG: Bad page cache 176 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
166 current->comm, folio_ !! 177 current->comm, page_to_pfn(page));
167 dump_page(&folio->page, "still !! 178 dump_page(page, "still mapped when deleted");
168 dump_stack(); 179 dump_stack();
169 add_taint(TAINT_BAD_PAGE, LOCK 180 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
170 181
171 if (mapping_exiting(mapping) & !! 182 mapcount = page_mapcount(page);
172 int mapcount = folio_m !! 183 if (mapping_exiting(mapping) &&
173 !! 184 page_count(page) >= mapcount + 2) {
174 if (folio_ref_count(fo !! 185 /*
175 /* !! 186 * All vmas have already been torn down, so it's
176 * All vmas ha !! 187 * a good bet that actually the page is unmapped,
177 * a good bet !! 188 * and we'd prefer not to leak it: if we're wrong,
178 * and we'd ra !! 189 * some other bad page check should catch it later.
179 * another bad !! 190 */
180 */ !! 191 page_mapcount_reset(page);
181 atomic_set(&fo !! 192 page_ref_sub(page, mapcount);
182 folio_ref_sub( <<
183 } <<
184 } 193 }
185 } 194 }
186 195
187 /* hugetlb folios do not participate i !! 196 /* hugetlb pages do not participate in page cache accounting. */
188 if (folio_test_hugetlb(folio)) !! 197 if (PageHuge(page))
189 return; 198 return;
190 199
191 nr = folio_nr_pages(folio); !! 200 nr = hpage_nr_pages(page);
192 201
193 __lruvec_stat_mod_folio(folio, NR_FILE !! 202 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
194 if (folio_test_swapbacked(folio)) { !! 203 if (PageSwapBacked(page)) {
195 __lruvec_stat_mod_folio(folio, !! 204 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
196 if (folio_test_pmd_mappable(fo !! 205 if (PageTransHuge(page))
197 __lruvec_stat_mod_foli !! 206 __dec_node_page_state(page, NR_SHMEM_THPS);
198 } else if (folio_test_pmd_mappable(fol !! 207 } else if (PageTransHuge(page)) {
199 __lruvec_stat_mod_folio(folio, !! 208 __dec_node_page_state(page, NR_FILE_THPS);
200 filemap_nr_thps_dec(mapping); 209 filemap_nr_thps_dec(mapping);
201 } 210 }
202 211
203 /* 212 /*
204 * At this point folio must be either !! 213 * At this point page must be either written or cleaned by
205 * truncate. Dirty folio here signals !! 214 * truncate. Dirty page here signals a bug and loss of
206 * unwritten data - on ordinary filesy !! 215 * 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 * 216 *
212 * Below fixes dirty accounting after !! 217 * This fixes dirty accounting after removing the page entirely
213 * but leaves the dirty flag set: it h !! 218 * but leaves PageDirty set: it has no effect for truncated
214 * folio and anyway will be cleared be !! 219 * page and anyway will be cleared before returning page into
215 * buddy allocator. 220 * buddy allocator.
216 */ 221 */
217 if (WARN_ON_ONCE(folio_test_dirty(foli !! 222 if (WARN_ON_ONCE(PageDirty(page)))
218 mapping_can_writeback !! 223 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
219 folio_account_cleaned(folio, i <<
220 } 224 }
221 225
222 /* 226 /*
223 * Delete a page from the page cache and free 227 * Delete a page from the page cache and free it. Caller has to make
224 * sure the page is locked and that nobody els 228 * sure the page is locked and that nobody else uses it - or that usage
225 * is safe. The caller must hold the i_pages 229 * is safe. The caller must hold the i_pages lock.
226 */ 230 */
227 void __filemap_remove_folio(struct folio *foli !! 231 void __delete_from_page_cache(struct page *page, void *shadow)
228 { 232 {
229 struct address_space *mapping = folio- !! 233 struct address_space *mapping = page->mapping;
>> 234
>> 235 trace_mm_filemap_delete_from_page_cache(page);
230 236
231 trace_mm_filemap_delete_from_page_cach !! 237 unaccount_page_cache_page(mapping, page);
232 filemap_unaccount_folio(mapping, folio !! 238 page_cache_delete(mapping, page, shadow);
233 page_cache_delete(mapping, folio, shad <<
234 } 239 }
235 240
236 void filemap_free_folio(struct address_space * !! 241 static void page_cache_free_page(struct address_space *mapping,
>> 242 struct page *page)
237 { 243 {
238 void (*free_folio)(struct folio *); !! 244 void (*freepage)(struct page *);
239 int refs = 1; <<
240 245
241 free_folio = mapping->a_ops->free_foli !! 246 freepage = mapping->a_ops->freepage;
242 if (free_folio) !! 247 if (freepage)
243 free_folio(folio); !! 248 freepage(page);
244 249
245 if (folio_test_large(folio)) !! 250 if (PageTransHuge(page) && !PageHuge(page)) {
246 refs = folio_nr_pages(folio); !! 251 page_ref_sub(page, HPAGE_PMD_NR);
247 folio_put_refs(folio, refs); !! 252 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
>> 253 } else {
>> 254 put_page(page);
>> 255 }
248 } 256 }
249 257
250 /** 258 /**
251 * filemap_remove_folio - Remove folio from pa !! 259 * delete_from_page_cache - delete page from page cache
252 * @folio: The folio. !! 260 * @page: the page which the kernel is trying to remove from page cache
253 * 261 *
254 * This must be called only on folios that are !! 262 * 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 !! 263 * 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 !! 264 * has a reference on the page.
257 */ !! 265 */
258 void filemap_remove_folio(struct folio *folio) !! 266 void delete_from_page_cache(struct page *page)
259 { !! 267 {
260 struct address_space *mapping = folio- !! 268 struct address_space *mapping = page_mapping(page);
261 !! 269 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 } <<
273 270
274 /* !! 271 BUG_ON(!PageLocked(page));
275 * page_cache_delete_batch - delete several fo !! 272 xa_lock_irqsave(&mapping->i_pages, flags);
276 * @mapping: the mapping to which folios belon !! 273 __delete_from_page_cache(page, NULL);
277 * @fbatch: batch of folios to delete !! 274 xa_unlock_irqrestore(&mapping->i_pages, flags);
278 * !! 275
279 * The function walks over mapping->i_pages an !! 276 page_cache_free_page(mapping, page);
280 * @fbatch from the mapping. The function expe !! 277 }
281 * by page index and is optimised for it to be !! 278 EXPORT_SYMBOL(delete_from_page_cache);
282 * It tolerates holes in @fbatch (mapping entr !! 279
283 * modified). !! 280 /*
>> 281 * page_cache_delete_batch - delete several pages from page cache
>> 282 * @mapping: the mapping to which pages belong
>> 283 * @pvec: pagevec with pages to delete
>> 284 *
>> 285 * The function walks over mapping->i_pages and removes pages passed in @pvec
>> 286 * from the mapping. The function expects @pvec to be sorted by page index
>> 287 * and is optimised for it to be dense.
>> 288 * It tolerates holes in @pvec (mapping entries at those indices are not
>> 289 * modified). The function expects only THP head pages to be present in the
>> 290 * @pvec.
284 * 291 *
285 * The function expects the i_pages lock to be 292 * The function expects the i_pages lock to be held.
286 */ 293 */
287 static void page_cache_delete_batch(struct add 294 static void page_cache_delete_batch(struct address_space *mapping,
288 struct folio_batc !! 295 struct pagevec *pvec)
289 { 296 {
290 XA_STATE(xas, &mapping->i_pages, fbatc !! 297 XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
291 long total_pages = 0; !! 298 int total_pages = 0;
292 int i = 0; 299 int i = 0;
293 struct folio *folio; !! 300 struct page *page;
294 301
295 mapping_set_update(&xas, mapping); 302 mapping_set_update(&xas, mapping);
296 xas_for_each(&xas, folio, ULONG_MAX) { !! 303 xas_for_each(&xas, page, ULONG_MAX) {
297 if (i >= folio_batch_count(fba !! 304 if (i >= pagevec_count(pvec))
298 break; 305 break;
299 306
300 /* A swap/dax/shadow entry got 307 /* A swap/dax/shadow entry got inserted? Skip it. */
301 if (xa_is_value(folio)) !! 308 if (xa_is_value(page))
302 continue; 309 continue;
303 /* 310 /*
304 * A page got inserted in our 311 * A page got inserted in our range? Skip it. We have our
305 * pages locked so they are pr 312 * pages locked so they are protected from being removed.
306 * If we see a page whose inde 313 * If we see a page whose index is higher than ours, it
307 * means our page has been rem 314 * means our page has been removed, which shouldn't be
308 * possible because we're hold 315 * possible because we're holding the PageLock.
309 */ 316 */
310 if (folio != fbatch->folios[i] !! 317 if (page != pvec->pages[i]) {
311 VM_BUG_ON_FOLIO(folio- !! 318 VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index,
312 fbatch !! 319 page);
313 continue; 320 continue;
314 } 321 }
315 322
316 WARN_ON_ONCE(!folio_test_locke !! 323 WARN_ON_ONCE(!PageLocked(page));
317 324
318 folio->mapping = NULL; !! 325 if (page->index == xas.xa_index)
319 /* Leave folio->index set: tru !! 326 page->mapping = NULL;
>> 327 /* Leave page->index set: truncation lookup relies on it */
320 328
321 i++; !! 329 /*
>> 330 * Move to the next page in the vector if this is a regular
>> 331 * page or the index is of the last sub-page of this compound
>> 332 * page.
>> 333 */
>> 334 if (page->index + compound_nr(page) - 1 == xas.xa_index)
>> 335 i++;
322 xas_store(&xas, NULL); 336 xas_store(&xas, NULL);
323 total_pages += folio_nr_pages( !! 337 total_pages++;
324 } 338 }
325 mapping->nrpages -= total_pages; 339 mapping->nrpages -= total_pages;
326 } 340 }
327 341
328 void delete_from_page_cache_batch(struct addre 342 void delete_from_page_cache_batch(struct address_space *mapping,
329 struct folio !! 343 struct pagevec *pvec)
330 { 344 {
331 int i; 345 int i;
>> 346 unsigned long flags;
332 347
333 if (!folio_batch_count(fbatch)) !! 348 if (!pagevec_count(pvec))
334 return; 349 return;
335 350
336 spin_lock(&mapping->host->i_lock); !! 351 xa_lock_irqsave(&mapping->i_pages, flags);
337 xa_lock_irq(&mapping->i_pages); !! 352 for (i = 0; i < pagevec_count(pvec); i++) {
338 for (i = 0; i < folio_batch_count(fbat !! 353 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); <<
349 354
350 for (i = 0; i < folio_batch_count(fbat !! 355 unaccount_page_cache_page(mapping, pvec->pages[i]);
351 filemap_free_folio(mapping, fb !! 356 }
>> 357 page_cache_delete_batch(mapping, pvec);
>> 358 xa_unlock_irqrestore(&mapping->i_pages, flags);
>> 359
>> 360 for (i = 0; i < pagevec_count(pvec); i++)
>> 361 page_cache_free_page(mapping, pvec->pages[i]);
352 } 362 }
353 363
354 int filemap_check_errors(struct address_space 364 int filemap_check_errors(struct address_space *mapping)
355 { 365 {
356 int ret = 0; 366 int ret = 0;
357 /* Check for outstanding write errors 367 /* Check for outstanding write errors */
358 if (test_bit(AS_ENOSPC, &mapping->flag 368 if (test_bit(AS_ENOSPC, &mapping->flags) &&
359 test_and_clear_bit(AS_ENOSPC, &map 369 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
360 ret = -ENOSPC; 370 ret = -ENOSPC;
361 if (test_bit(AS_EIO, &mapping->flags) 371 if (test_bit(AS_EIO, &mapping->flags) &&
362 test_and_clear_bit(AS_EIO, &mappin 372 test_and_clear_bit(AS_EIO, &mapping->flags))
363 ret = -EIO; 373 ret = -EIO;
364 return ret; 374 return ret;
365 } 375 }
366 EXPORT_SYMBOL(filemap_check_errors); 376 EXPORT_SYMBOL(filemap_check_errors);
367 377
368 static int filemap_check_and_keep_errors(struc 378 static int filemap_check_and_keep_errors(struct address_space *mapping)
369 { 379 {
370 /* Check for outstanding write errors 380 /* Check for outstanding write errors */
371 if (test_bit(AS_EIO, &mapping->flags)) 381 if (test_bit(AS_EIO, &mapping->flags))
372 return -EIO; 382 return -EIO;
373 if (test_bit(AS_ENOSPC, &mapping->flag 383 if (test_bit(AS_ENOSPC, &mapping->flags))
374 return -ENOSPC; 384 return -ENOSPC;
375 return 0; 385 return 0;
376 } 386 }
377 387
378 /** 388 /**
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 389 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
406 * @mapping: address space structure to wri 390 * @mapping: address space structure to write
407 * @start: offset in bytes where the rang 391 * @start: offset in bytes where the range starts
408 * @end: offset in bytes where the rang 392 * @end: offset in bytes where the range ends (inclusive)
409 * @sync_mode: enable synchronous operation 393 * @sync_mode: enable synchronous operation
410 * 394 *
411 * Start writeback against all of a mapping's 395 * Start writeback against all of a mapping's dirty pages that lie
412 * within the byte offsets <start, end> inclus 396 * within the byte offsets <start, end> inclusive.
413 * 397 *
414 * If sync_mode is WB_SYNC_ALL then this is a 398 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
415 * opposed to a regular memory cleansing write 399 * opposed to a regular memory cleansing writeback. The difference between
416 * these two operations is that if a dirty pag 400 * these two operations is that if a dirty page/buffer is encountered, it must
417 * be waited upon, and not just skipped over. 401 * be waited upon, and not just skipped over.
418 * 402 *
419 * Return: %0 on success, negative error code 403 * Return: %0 on success, negative error code otherwise.
420 */ 404 */
421 int __filemap_fdatawrite_range(struct address_ 405 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
422 loff_t end, in 406 loff_t end, int sync_mode)
423 { 407 {
>> 408 int ret;
424 struct writeback_control wbc = { 409 struct writeback_control wbc = {
425 .sync_mode = sync_mode, 410 .sync_mode = sync_mode,
426 .nr_to_write = LONG_MAX, 411 .nr_to_write = LONG_MAX,
427 .range_start = start, 412 .range_start = start,
428 .range_end = end, 413 .range_end = end,
429 }; 414 };
430 415
431 return filemap_fdatawrite_wbc(mapping, !! 416 if (!mapping_cap_writeback_dirty(mapping) ||
>> 417 !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
>> 418 return 0;
>> 419
>> 420 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
>> 421 ret = do_writepages(mapping, &wbc);
>> 422 wbc_detach_inode(&wbc);
>> 423 return ret;
432 } 424 }
433 425
434 static inline int __filemap_fdatawrite(struct 426 static inline int __filemap_fdatawrite(struct address_space *mapping,
435 int sync_mode) 427 int sync_mode)
436 { 428 {
437 return __filemap_fdatawrite_range(mapp 429 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
438 } 430 }
439 431
440 int filemap_fdatawrite(struct address_space *m 432 int filemap_fdatawrite(struct address_space *mapping)
441 { 433 {
442 return __filemap_fdatawrite(mapping, W 434 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
443 } 435 }
444 EXPORT_SYMBOL(filemap_fdatawrite); 436 EXPORT_SYMBOL(filemap_fdatawrite);
445 437
446 int filemap_fdatawrite_range(struct address_sp 438 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
447 loff_t end) 439 loff_t end)
448 { 440 {
449 return __filemap_fdatawrite_range(mapp 441 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
450 } 442 }
451 EXPORT_SYMBOL(filemap_fdatawrite_range); 443 EXPORT_SYMBOL(filemap_fdatawrite_range);
452 444
453 /** 445 /**
454 * filemap_flush - mostly a non-blocking flush 446 * filemap_flush - mostly a non-blocking flush
455 * @mapping: target address_space 447 * @mapping: target address_space
456 * 448 *
457 * This is a mostly non-blocking flush. Not s 449 * This is a mostly non-blocking flush. Not suitable for data-integrity
458 * purposes - I/O may not be started against a 450 * purposes - I/O may not be started against all dirty pages.
459 * 451 *
460 * Return: %0 on success, negative error code 452 * Return: %0 on success, negative error code otherwise.
461 */ 453 */
462 int filemap_flush(struct address_space *mappin 454 int filemap_flush(struct address_space *mapping)
463 { 455 {
464 return __filemap_fdatawrite(mapping, W 456 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
465 } 457 }
466 EXPORT_SYMBOL(filemap_flush); 458 EXPORT_SYMBOL(filemap_flush);
467 459
468 /** 460 /**
469 * filemap_range_has_page - check if a page ex 461 * filemap_range_has_page - check if a page exists in range.
470 * @mapping: address space within wh 462 * @mapping: address space within which to check
471 * @start_byte: offset in bytes where t 463 * @start_byte: offset in bytes where the range starts
472 * @end_byte: offset in bytes where t 464 * @end_byte: offset in bytes where the range ends (inclusive)
473 * 465 *
474 * Find at least one page in the range supplie 466 * Find at least one page in the range supplied, usually used to check if
475 * direct writing in this range will trigger a 467 * direct writing in this range will trigger a writeback.
476 * 468 *
477 * Return: %true if at least one page exists i 469 * Return: %true if at least one page exists in the specified range,
478 * %false otherwise. 470 * %false otherwise.
479 */ 471 */
480 bool filemap_range_has_page(struct address_spa 472 bool filemap_range_has_page(struct address_space *mapping,
481 loff_t start_byte, 473 loff_t start_byte, loff_t end_byte)
482 { 474 {
483 struct folio *folio; !! 475 struct page *page;
484 XA_STATE(xas, &mapping->i_pages, start 476 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
485 pgoff_t max = end_byte >> PAGE_SHIFT; 477 pgoff_t max = end_byte >> PAGE_SHIFT;
486 478
487 if (end_byte < start_byte) 479 if (end_byte < start_byte)
488 return false; 480 return false;
489 481
490 rcu_read_lock(); 482 rcu_read_lock();
491 for (;;) { 483 for (;;) {
492 folio = xas_find(&xas, max); !! 484 page = xas_find(&xas, max);
493 if (xas_retry(&xas, folio)) !! 485 if (xas_retry(&xas, page))
494 continue; 486 continue;
495 /* Shadow entries don't count 487 /* Shadow entries don't count */
496 if (xa_is_value(folio)) !! 488 if (xa_is_value(page))
497 continue; 489 continue;
498 /* 490 /*
499 * We don't need to try to pin 491 * We don't need to try to pin this page; we're about to
500 * release the RCU lock anyway 492 * release the RCU lock anyway. It is enough to know that
501 * there was a page here recen 493 * there was a page here recently.
502 */ 494 */
503 break; 495 break;
504 } 496 }
505 rcu_read_unlock(); 497 rcu_read_unlock();
506 498
507 return folio != NULL; !! 499 return page != NULL;
508 } 500 }
509 EXPORT_SYMBOL(filemap_range_has_page); 501 EXPORT_SYMBOL(filemap_range_has_page);
510 502
511 static void __filemap_fdatawait_range(struct a 503 static void __filemap_fdatawait_range(struct address_space *mapping,
512 loff_t st 504 loff_t start_byte, loff_t end_byte)
513 { 505 {
514 pgoff_t index = start_byte >> PAGE_SHI 506 pgoff_t index = start_byte >> PAGE_SHIFT;
515 pgoff_t end = end_byte >> PAGE_SHIFT; 507 pgoff_t end = end_byte >> PAGE_SHIFT;
516 struct folio_batch fbatch; !! 508 struct pagevec pvec;
517 unsigned nr_folios; !! 509 int nr_pages;
518 510
519 folio_batch_init(&fbatch); !! 511 if (end_byte < start_byte)
>> 512 return;
520 513
>> 514 pagevec_init(&pvec);
521 while (index <= end) { 515 while (index <= end) {
522 unsigned i; 516 unsigned i;
523 517
524 nr_folios = filemap_get_folios !! 518 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
525 PAGECACHE_TAG_ !! 519 end, PAGECACHE_TAG_WRITEBACK);
526 !! 520 if (!nr_pages)
527 if (!nr_folios) <<
528 break; 521 break;
529 522
530 for (i = 0; i < nr_folios; i++ !! 523 for (i = 0; i < nr_pages; i++) {
531 struct folio *folio = !! 524 struct page *page = pvec.pages[i];
532 525
533 folio_wait_writeback(f !! 526 wait_on_page_writeback(page);
>> 527 ClearPageError(page);
534 } 528 }
535 folio_batch_release(&fbatch); !! 529 pagevec_release(&pvec);
536 cond_resched(); 530 cond_resched();
537 } 531 }
538 } 532 }
539 533
540 /** 534 /**
541 * filemap_fdatawait_range - wait for writebac 535 * filemap_fdatawait_range - wait for writeback to complete
542 * @mapping: address space structur 536 * @mapping: address space structure to wait for
543 * @start_byte: offset in bytes where 537 * @start_byte: offset in bytes where the range starts
544 * @end_byte: offset in bytes where 538 * @end_byte: offset in bytes where the range ends (inclusive)
545 * 539 *
546 * Walk the list of under-writeback pages of t 540 * Walk the list of under-writeback pages of the given address space
547 * in the given range and wait for all of them 541 * in the given range and wait for all of them. Check error status of
548 * the address space and return it. 542 * the address space and return it.
549 * 543 *
550 * Since the error status of the address space 544 * Since the error status of the address space is cleared by this function,
551 * callers are responsible for checking the re 545 * callers are responsible for checking the return value and handling and/or
552 * reporting the error. 546 * reporting the error.
553 * 547 *
554 * Return: error status of the address space. 548 * Return: error status of the address space.
555 */ 549 */
556 int filemap_fdatawait_range(struct address_spa 550 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
557 loff_t end_byte) 551 loff_t end_byte)
558 { 552 {
559 __filemap_fdatawait_range(mapping, sta 553 __filemap_fdatawait_range(mapping, start_byte, end_byte);
560 return filemap_check_errors(mapping); 554 return filemap_check_errors(mapping);
561 } 555 }
562 EXPORT_SYMBOL(filemap_fdatawait_range); 556 EXPORT_SYMBOL(filemap_fdatawait_range);
563 557
564 /** 558 /**
565 * filemap_fdatawait_range_keep_errors - wait 559 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
566 * @mapping: address space structur 560 * @mapping: address space structure to wait for
567 * @start_byte: offset in bytes where 561 * @start_byte: offset in bytes where the range starts
568 * @end_byte: offset in bytes where 562 * @end_byte: offset in bytes where the range ends (inclusive)
569 * 563 *
570 * Walk the list of under-writeback pages of t 564 * Walk the list of under-writeback pages of the given address space in the
571 * given range and wait for all of them. Unli 565 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
572 * this function does not clear error status o 566 * this function does not clear error status of the address space.
573 * 567 *
574 * Use this function if callers don't handle e 568 * Use this function if callers don't handle errors themselves. Expected
575 * call sites are system-wide / filesystem-wid 569 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
576 * fsfreeze(8) 570 * fsfreeze(8)
577 */ 571 */
578 int filemap_fdatawait_range_keep_errors(struct 572 int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
579 loff_t start_byte, loff_t end_ 573 loff_t start_byte, loff_t end_byte)
580 { 574 {
581 __filemap_fdatawait_range(mapping, sta 575 __filemap_fdatawait_range(mapping, start_byte, end_byte);
582 return filemap_check_and_keep_errors(m 576 return filemap_check_and_keep_errors(mapping);
583 } 577 }
584 EXPORT_SYMBOL(filemap_fdatawait_range_keep_err 578 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
585 579
586 /** 580 /**
587 * file_fdatawait_range - wait for writeback t 581 * file_fdatawait_range - wait for writeback to complete
588 * @file: file pointing to addre 582 * @file: file pointing to address space structure to wait for
589 * @start_byte: offset in bytes where 583 * @start_byte: offset in bytes where the range starts
590 * @end_byte: offset in bytes where 584 * @end_byte: offset in bytes where the range ends (inclusive)
591 * 585 *
592 * Walk the list of under-writeback pages of t 586 * Walk the list of under-writeback pages of the address space that file
593 * refers to, in the given range and wait for 587 * refers to, in the given range and wait for all of them. Check error
594 * status of the address space vs. the file->f 588 * status of the address space vs. the file->f_wb_err cursor and return it.
595 * 589 *
596 * Since the error status of the file is advan 590 * Since the error status of the file is advanced by this function,
597 * callers are responsible for checking the re 591 * callers are responsible for checking the return value and handling and/or
598 * reporting the error. 592 * reporting the error.
599 * 593 *
600 * Return: error status of the address space v 594 * Return: error status of the address space vs. the file->f_wb_err cursor.
601 */ 595 */
602 int file_fdatawait_range(struct file *file, lo 596 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
603 { 597 {
604 struct address_space *mapping = file-> 598 struct address_space *mapping = file->f_mapping;
605 599
606 __filemap_fdatawait_range(mapping, sta 600 __filemap_fdatawait_range(mapping, start_byte, end_byte);
607 return file_check_and_advance_wb_err(f 601 return file_check_and_advance_wb_err(file);
608 } 602 }
609 EXPORT_SYMBOL(file_fdatawait_range); 603 EXPORT_SYMBOL(file_fdatawait_range);
610 604
611 /** 605 /**
612 * filemap_fdatawait_keep_errors - wait for wr 606 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
613 * @mapping: address space structure to wait f 607 * @mapping: address space structure to wait for
614 * 608 *
615 * Walk the list of under-writeback pages of t 609 * Walk the list of under-writeback pages of the given address space
616 * and wait for all of them. Unlike filemap_f 610 * and wait for all of them. Unlike filemap_fdatawait(), this function
617 * does not clear error status of the address 611 * does not clear error status of the address space.
618 * 612 *
619 * Use this function if callers don't handle e 613 * Use this function if callers don't handle errors themselves. Expected
620 * call sites are system-wide / filesystem-wid 614 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
621 * fsfreeze(8) 615 * fsfreeze(8)
622 * 616 *
623 * Return: error status of the address space. 617 * Return: error status of the address space.
624 */ 618 */
625 int filemap_fdatawait_keep_errors(struct addre 619 int filemap_fdatawait_keep_errors(struct address_space *mapping)
626 { 620 {
627 __filemap_fdatawait_range(mapping, 0, 621 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
628 return filemap_check_and_keep_errors(m 622 return filemap_check_and_keep_errors(mapping);
629 } 623 }
630 EXPORT_SYMBOL(filemap_fdatawait_keep_errors); 624 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
631 625
632 /* Returns true if writeback might be needed o 626 /* Returns true if writeback might be needed or already in progress. */
633 static bool mapping_needs_writeback(struct add 627 static bool mapping_needs_writeback(struct address_space *mapping)
634 { 628 {
>> 629 if (dax_mapping(mapping))
>> 630 return mapping->nrexceptional;
>> 631
635 return mapping->nrpages; 632 return mapping->nrpages;
636 } 633 }
637 634
638 bool filemap_range_has_writeback(struct addres !! 635 int filemap_write_and_wait(struct address_space *mapping)
639 loff_t start_ <<
640 { 636 {
641 XA_STATE(xas, &mapping->i_pages, start !! 637 int err = 0;
642 pgoff_t max = end_byte >> PAGE_SHIFT; <<
643 struct folio *folio; <<
644 <<
645 if (end_byte < start_byte) <<
646 return false; <<
647 638
648 rcu_read_lock(); !! 639 if (mapping_needs_writeback(mapping)) {
649 xas_for_each(&xas, folio, max) { !! 640 err = filemap_fdatawrite(mapping);
650 if (xas_retry(&xas, folio)) !! 641 /*
651 continue; !! 642 * Even if the above returned error, the pages may be
652 if (xa_is_value(folio)) !! 643 * written partially (e.g. -ENOSPC), so we wait for it.
653 continue; !! 644 * But the -EIO is special case, it may indicate the worst
654 if (folio_test_dirty(folio) || !! 645 * thing (e.g. bug) happened, so we avoid waiting for it.
655 folio_test_wri !! 646 */
656 break; !! 647 if (err != -EIO) {
>> 648 int err2 = filemap_fdatawait(mapping);
>> 649 if (!err)
>> 650 err = err2;
>> 651 } else {
>> 652 /* Clear any previously stored errors */
>> 653 filemap_check_errors(mapping);
>> 654 }
>> 655 } else {
>> 656 err = filemap_check_errors(mapping);
657 } 657 }
658 rcu_read_unlock(); !! 658 return err;
659 return folio != NULL; <<
660 } 659 }
661 EXPORT_SYMBOL_GPL(filemap_range_has_writeback) !! 660 EXPORT_SYMBOL(filemap_write_and_wait);
662 661
663 /** 662 /**
664 * filemap_write_and_wait_range - write out & 663 * filemap_write_and_wait_range - write out & wait on a file range
665 * @mapping: the address_space for the page 664 * @mapping: the address_space for the pages
666 * @lstart: offset in bytes where the rang 665 * @lstart: offset in bytes where the range starts
667 * @lend: offset in bytes where the rang 666 * @lend: offset in bytes where the range ends (inclusive)
668 * 667 *
669 * Write out and wait upon file offsets lstart 668 * Write out and wait upon file offsets lstart->lend, inclusive.
670 * 669 *
671 * Note that @lend is inclusive (describes the 670 * Note that @lend is inclusive (describes the last byte to be written) so
672 * that this function can be used to write to 671 * that this function can be used to write to the very end-of-file (end = -1).
673 * 672 *
674 * Return: error status of the address space. 673 * Return: error status of the address space.
675 */ 674 */
676 int filemap_write_and_wait_range(struct addres 675 int filemap_write_and_wait_range(struct address_space *mapping,
677 loff_t lstart 676 loff_t lstart, loff_t lend)
678 { 677 {
679 int err = 0, err2; !! 678 int err = 0;
680 <<
681 if (lend < lstart) <<
682 return 0; <<
683 679
684 if (mapping_needs_writeback(mapping)) 680 if (mapping_needs_writeback(mapping)) {
685 err = __filemap_fdatawrite_ran 681 err = __filemap_fdatawrite_range(mapping, lstart, lend,
686 682 WB_SYNC_ALL);
687 /* !! 683 /* See comment of filemap_write_and_wait() */
688 * Even if the above returned !! 684 if (err != -EIO) {
689 * written partially (e.g. -EN !! 685 int err2 = filemap_fdatawait_range(mapping,
690 * But the -EIO is special cas !! 686 lstart, lend);
691 * thing (e.g. bug) happened, !! 687 if (!err)
692 */ !! 688 err = err2;
693 if (err != -EIO) !! 689 } else {
694 __filemap_fdatawait_ra !! 690 /* Clear any previously stored errors */
>> 691 filemap_check_errors(mapping);
>> 692 }
>> 693 } else {
>> 694 err = filemap_check_errors(mapping);
695 } 695 }
696 err2 = filemap_check_errors(mapping); <<
697 if (!err) <<
698 err = err2; <<
699 return err; 696 return err;
700 } 697 }
701 EXPORT_SYMBOL(filemap_write_and_wait_range); 698 EXPORT_SYMBOL(filemap_write_and_wait_range);
702 699
703 void __filemap_set_wb_err(struct address_space 700 void __filemap_set_wb_err(struct address_space *mapping, int err)
704 { 701 {
705 errseq_t eseq = errseq_set(&mapping->w 702 errseq_t eseq = errseq_set(&mapping->wb_err, err);
706 703
707 trace_filemap_set_wb_err(mapping, eseq 704 trace_filemap_set_wb_err(mapping, eseq);
708 } 705 }
709 EXPORT_SYMBOL(__filemap_set_wb_err); 706 EXPORT_SYMBOL(__filemap_set_wb_err);
710 707
711 /** 708 /**
712 * file_check_and_advance_wb_err - report wb e 709 * file_check_and_advance_wb_err - report wb error (if any) that was previously
713 * and advance 710 * and advance wb_err to current one
714 * @file: struct file on which the error is be 711 * @file: struct file on which the error is being reported
715 * 712 *
716 * When userland calls fsync (or something lik 713 * When userland calls fsync (or something like nfsd does the equivalent), we
717 * want to report any writeback errors that oc 714 * want to report any writeback errors that occurred since the last fsync (or
718 * since the file was opened if there haven't 715 * since the file was opened if there haven't been any).
719 * 716 *
720 * Grab the wb_err from the mapping. If it mat 717 * 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 718 * then just quickly return 0. The file is all caught up.
722 * 719 *
723 * If it doesn't match, then take the mapping 720 * 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 721 * 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 722 * 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 723 * portion. The error at this point must be reported via proper channels
727 * (a'la fsync, or NFS COMMIT operation, etc.) 724 * (a'la fsync, or NFS COMMIT operation, etc.).
728 * 725 *
729 * While we handle mapping->wb_err with atomic 726 * While we handle mapping->wb_err with atomic operations, the f_wb_err
730 * value is protected by the f_lock since we m 727 * value is protected by the f_lock since we must ensure that it reflects
731 * the latest value swapped in for this file d 728 * the latest value swapped in for this file descriptor.
732 * 729 *
733 * Return: %0 on success, negative error code 730 * Return: %0 on success, negative error code otherwise.
734 */ 731 */
735 int file_check_and_advance_wb_err(struct file 732 int file_check_and_advance_wb_err(struct file *file)
736 { 733 {
737 int err = 0; 734 int err = 0;
738 errseq_t old = READ_ONCE(file->f_wb_er 735 errseq_t old = READ_ONCE(file->f_wb_err);
739 struct address_space *mapping = file-> 736 struct address_space *mapping = file->f_mapping;
740 737
741 /* Locklessly handle the common case w 738 /* Locklessly handle the common case where nothing has changed */
742 if (errseq_check(&mapping->wb_err, old 739 if (errseq_check(&mapping->wb_err, old)) {
743 /* Something changed, must use 740 /* Something changed, must use slow path */
744 spin_lock(&file->f_lock); 741 spin_lock(&file->f_lock);
745 old = file->f_wb_err; 742 old = file->f_wb_err;
746 err = errseq_check_and_advance 743 err = errseq_check_and_advance(&mapping->wb_err,
747 744 &file->f_wb_err);
748 trace_file_check_and_advance_w 745 trace_file_check_and_advance_wb_err(file, old);
749 spin_unlock(&file->f_lock); 746 spin_unlock(&file->f_lock);
750 } 747 }
751 748
752 /* 749 /*
753 * We're mostly using this function as 750 * We're mostly using this function as a drop in replacement for
754 * filemap_check_errors. Clear AS_EIO/ 751 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
755 * that the legacy code would have had 752 * that the legacy code would have had on these flags.
756 */ 753 */
757 clear_bit(AS_EIO, &mapping->flags); 754 clear_bit(AS_EIO, &mapping->flags);
758 clear_bit(AS_ENOSPC, &mapping->flags); 755 clear_bit(AS_ENOSPC, &mapping->flags);
759 return err; 756 return err;
760 } 757 }
761 EXPORT_SYMBOL(file_check_and_advance_wb_err); 758 EXPORT_SYMBOL(file_check_and_advance_wb_err);
762 759
763 /** 760 /**
764 * file_write_and_wait_range - write out & wai 761 * file_write_and_wait_range - write out & wait on a file range
765 * @file: file pointing to address_space 762 * @file: file pointing to address_space with pages
766 * @lstart: offset in bytes where the rang 763 * @lstart: offset in bytes where the range starts
767 * @lend: offset in bytes where the rang 764 * @lend: offset in bytes where the range ends (inclusive)
768 * 765 *
769 * Write out and wait upon file offsets lstart 766 * Write out and wait upon file offsets lstart->lend, inclusive.
770 * 767 *
771 * Note that @lend is inclusive (describes the 768 * Note that @lend is inclusive (describes the last byte to be written) so
772 * that this function can be used to write to 769 * that this function can be used to write to the very end-of-file (end = -1).
773 * 770 *
774 * After writing out and waiting on the data, 771 * After writing out and waiting on the data, we check and advance the
775 * f_wb_err cursor to the latest value, and re 772 * f_wb_err cursor to the latest value, and return any errors detected there.
776 * 773 *
777 * Return: %0 on success, negative error code 774 * Return: %0 on success, negative error code otherwise.
778 */ 775 */
779 int file_write_and_wait_range(struct file *fil 776 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
780 { 777 {
781 int err = 0, err2; 778 int err = 0, err2;
782 struct address_space *mapping = file-> 779 struct address_space *mapping = file->f_mapping;
783 780
784 if (lend < lstart) <<
785 return 0; <<
786 <<
787 if (mapping_needs_writeback(mapping)) 781 if (mapping_needs_writeback(mapping)) {
788 err = __filemap_fdatawrite_ran 782 err = __filemap_fdatawrite_range(mapping, lstart, lend,
789 783 WB_SYNC_ALL);
790 /* See comment of filemap_writ 784 /* See comment of filemap_write_and_wait() */
791 if (err != -EIO) 785 if (err != -EIO)
792 __filemap_fdatawait_ra 786 __filemap_fdatawait_range(mapping, lstart, lend);
793 } 787 }
794 err2 = file_check_and_advance_wb_err(f 788 err2 = file_check_and_advance_wb_err(file);
795 if (!err) 789 if (!err)
796 err = err2; 790 err = err2;
797 return err; 791 return err;
798 } 792 }
799 EXPORT_SYMBOL(file_write_and_wait_range); 793 EXPORT_SYMBOL(file_write_and_wait_range);
800 794
801 /** 795 /**
802 * replace_page_cache_folio - replace a pageca !! 796 * replace_page_cache_page - replace a pagecache page with a new one
803 * @old: folio to be replaced !! 797 * @old: page to be replaced
804 * @new: folio to replace with !! 798 * @new: page to replace with
805 * !! 799 * @gfp_mask: allocation mode
806 * This function replaces a folio in the pagec !! 800 *
807 * success it acquires the pagecache reference !! 801 * This function replaces a page in the pagecache with a new one. On
808 * drops it for the old folio. Both the old a !! 802 * success it acquires the pagecache reference for the new page and
809 * locked. This function does not add the new !! 803 * drops it for the old page. Both the old and new pages must be
>> 804 * locked. This function does not add the new page to the LRU, the
810 * caller must do that. 805 * caller must do that.
811 * 806 *
812 * The remove + add is atomic. This function 807 * The remove + add is atomic. This function cannot fail.
>> 808 *
>> 809 * Return: %0
813 */ 810 */
814 void replace_page_cache_folio(struct folio *ol !! 811 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
815 { 812 {
816 struct address_space *mapping = old->m 813 struct address_space *mapping = old->mapping;
817 void (*free_folio)(struct folio *) = m !! 814 void (*freepage)(struct page *) = mapping->a_ops->freepage;
818 pgoff_t offset = old->index; 815 pgoff_t offset = old->index;
819 XA_STATE(xas, &mapping->i_pages, offse 816 XA_STATE(xas, &mapping->i_pages, offset);
>> 817 unsigned long flags;
820 818
821 VM_BUG_ON_FOLIO(!folio_test_locked(old !! 819 VM_BUG_ON_PAGE(!PageLocked(old), old);
822 VM_BUG_ON_FOLIO(!folio_test_locked(new !! 820 VM_BUG_ON_PAGE(!PageLocked(new), new);
823 VM_BUG_ON_FOLIO(new->mapping, new); !! 821 VM_BUG_ON_PAGE(new->mapping, new);
824 822
825 folio_get(new); !! 823 get_page(new);
826 new->mapping = mapping; 824 new->mapping = mapping;
827 new->index = offset; 825 new->index = offset;
828 826
829 mem_cgroup_replace_folio(old, new); !! 827 xas_lock_irqsave(&xas, flags);
830 <<
831 xas_lock_irq(&xas); <<
832 xas_store(&xas, new); 828 xas_store(&xas, new);
833 829
834 old->mapping = NULL; 830 old->mapping = NULL;
835 /* hugetlb pages do not participate in 831 /* hugetlb pages do not participate in page cache accounting. */
836 if (!folio_test_hugetlb(old)) !! 832 if (!PageHuge(old))
837 __lruvec_stat_sub_folio(old, N !! 833 __dec_node_page_state(new, NR_FILE_PAGES);
838 if (!folio_test_hugetlb(new)) !! 834 if (!PageHuge(new))
839 __lruvec_stat_add_folio(new, N !! 835 __inc_node_page_state(new, NR_FILE_PAGES);
840 if (folio_test_swapbacked(old)) !! 836 if (PageSwapBacked(old))
841 __lruvec_stat_sub_folio(old, N !! 837 __dec_node_page_state(new, NR_SHMEM);
842 if (folio_test_swapbacked(new)) !! 838 if (PageSwapBacked(new))
843 __lruvec_stat_add_folio(new, N !! 839 __inc_node_page_state(new, NR_SHMEM);
844 xas_unlock_irq(&xas); !! 840 xas_unlock_irqrestore(&xas, flags);
845 if (free_folio) !! 841 mem_cgroup_migrate(old, new);
846 free_folio(old); !! 842 if (freepage)
847 folio_put(old); !! 843 freepage(old);
>> 844 put_page(old);
>> 845
>> 846 return 0;
848 } 847 }
849 EXPORT_SYMBOL_GPL(replace_page_cache_folio); !! 848 EXPORT_SYMBOL_GPL(replace_page_cache_page);
850 849
851 noinline int __filemap_add_folio(struct addres !! 850 static int __add_to_page_cache_locked(struct page *page,
852 struct folio *folio, pgoff_t i !! 851 struct address_space *mapping,
>> 852 pgoff_t offset, gfp_t gfp_mask,
>> 853 void **shadowp)
853 { 854 {
854 XA_STATE(xas, &mapping->i_pages, index !! 855 XA_STATE(xas, &mapping->i_pages, offset);
855 void *alloced_shadow = NULL; !! 856 int huge = PageHuge(page);
856 int alloced_order = 0; !! 857 struct mem_cgroup *memcg;
857 bool huge; !! 858 int error;
858 long nr; !! 859 void *old;
859 !! 860
860 VM_BUG_ON_FOLIO(!folio_test_locked(fol !! 861 VM_BUG_ON_PAGE(!PageLocked(page), page);
861 VM_BUG_ON_FOLIO(folio_test_swapbacked( !! 862 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
862 VM_BUG_ON_FOLIO(folio_order(folio) < m <<
863 folio); <<
864 mapping_set_update(&xas, mapping); 863 mapping_set_update(&xas, mapping);
865 864
866 VM_BUG_ON_FOLIO(index & (folio_nr_page !! 865 if (!huge) {
867 xas_set_order(&xas, index, folio_order !! 866 error = mem_cgroup_try_charge(page, current->mm,
868 huge = folio_test_hugetlb(folio); !! 867 gfp_mask, &memcg, false);
869 nr = folio_nr_pages(folio); !! 868 if (error)
870 !! 869 return error;
871 gfp &= GFP_RECLAIM_MASK; !! 870 }
872 folio_ref_add(folio, nr); <<
873 folio->mapping = mapping; <<
874 folio->index = xas.xa_index; <<
875 871
876 for (;;) { !! 872 get_page(page);
877 int order = -1, split_order = !! 873 page->mapping = mapping;
878 void *entry, *old = NULL; !! 874 page->index = offset;
879 875
>> 876 do {
880 xas_lock_irq(&xas); 877 xas_lock_irq(&xas);
881 xas_for_each_conflict(&xas, en !! 878 old = xas_load(&xas);
882 old = entry; !! 879 if (old && !xa_is_value(old))
883 if (!xa_is_value(entry !! 880 xas_set_err(&xas, -EEXIST);
884 xas_set_err(&x !! 881 xas_store(&xas, page);
885 goto unlock; !! 882 if (xas_error(&xas))
886 } !! 883 goto unlock;
887 /* <<
888 * If a larger entry e <<
889 * it will be the firs <<
890 */ <<
891 if (order == -1) <<
892 order = xas_ge <<
893 } <<
894 884
895 /* entry may have changed befo !! 885 if (xa_is_value(old)) {
896 if (alloced_order && (old != a !! 886 mapping->nrexceptional--;
897 xas_destroy(&xas); <<
898 alloced_order = 0; <<
899 } <<
900 <<
901 if (old) { <<
902 if (order > 0 && order <<
903 /* How to hand <<
904 BUG_ON(shmem_m <<
905 if (!alloced_o <<
906 split_ <<
907 goto u <<
908 } <<
909 xas_split(&xas <<
910 xas_reset(&xas <<
911 } <<
912 if (shadowp) 887 if (shadowp)
913 *shadowp = old 888 *shadowp = old;
914 } 889 }
915 !! 890 mapping->nrpages++;
916 xas_store(&xas, folio); <<
917 if (xas_error(&xas)) <<
918 goto unlock; <<
919 <<
920 mapping->nrpages += nr; <<
921 891
922 /* hugetlb pages do not partic 892 /* hugetlb pages do not participate in page cache accounting */
923 if (!huge) { !! 893 if (!huge)
924 __lruvec_stat_mod_foli !! 894 __inc_node_page_state(page, NR_FILE_PAGES);
925 if (folio_test_pmd_map <<
926 __lruvec_stat_ <<
927 <<
928 } <<
929 <<
930 unlock: 895 unlock:
931 xas_unlock_irq(&xas); 896 xas_unlock_irq(&xas);
932 !! 897 } while (xas_nomem(&xas, gfp_mask & GFP_RECLAIM_MASK));
933 /* split needed, alloc here an <<
934 if (split_order) { <<
935 xas_split_alloc(&xas, <<
936 if (xas_error(&xas)) <<
937 goto error; <<
938 alloced_shadow = old; <<
939 alloced_order = split_ <<
940 xas_reset(&xas); <<
941 continue; <<
942 } <<
943 <<
944 if (!xas_nomem(&xas, gfp)) <<
945 break; <<
946 } <<
947 898
948 if (xas_error(&xas)) 899 if (xas_error(&xas))
949 goto error; 900 goto error;
950 901
951 trace_mm_filemap_add_to_page_cache(fol !! 902 if (!huge)
>> 903 mem_cgroup_commit_charge(page, memcg, false, false);
>> 904 trace_mm_filemap_add_to_page_cache(page);
952 return 0; 905 return 0;
953 error: 906 error:
954 folio->mapping = NULL; !! 907 page->mapping = NULL;
955 /* Leave page->index set: truncation r 908 /* Leave page->index set: truncation relies upon it */
956 folio_put_refs(folio, nr); !! 909 if (!huge)
>> 910 mem_cgroup_cancel_charge(page, memcg, false);
>> 911 put_page(page);
957 return xas_error(&xas); 912 return xas_error(&xas);
958 } 913 }
959 ALLOW_ERROR_INJECTION(__filemap_add_folio, ERR !! 914 ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO);
>> 915
>> 916 /**
>> 917 * add_to_page_cache_locked - add a locked page to the pagecache
>> 918 * @page: page to add
>> 919 * @mapping: the page's address_space
>> 920 * @offset: page index
>> 921 * @gfp_mask: page allocation mode
>> 922 *
>> 923 * This function is used to add a page to the pagecache. It must be locked.
>> 924 * This function does not add the page to the LRU. The caller must do that.
>> 925 *
>> 926 * Return: %0 on success, negative error code otherwise.
>> 927 */
>> 928 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
>> 929 pgoff_t offset, gfp_t gfp_mask)
>> 930 {
>> 931 return __add_to_page_cache_locked(page, mapping, offset,
>> 932 gfp_mask, NULL);
>> 933 }
>> 934 EXPORT_SYMBOL(add_to_page_cache_locked);
960 935
961 int filemap_add_folio(struct address_space *ma !! 936 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
962 pgoff_t index, !! 937 pgoff_t offset, gfp_t gfp_mask)
963 { 938 {
964 void *shadow = NULL; 939 void *shadow = NULL;
965 int ret; 940 int ret;
966 941
967 ret = mem_cgroup_charge(folio, NULL, g !! 942 __SetPageLocked(page);
968 if (ret) !! 943 ret = __add_to_page_cache_locked(page, mapping, offset,
969 return ret; !! 944 gfp_mask, &shadow);
970 !! 945 if (unlikely(ret))
971 __folio_set_locked(folio); !! 946 __ClearPageLocked(page);
972 ret = __filemap_add_folio(mapping, fol !! 947 else {
973 if (unlikely(ret)) { <<
974 mem_cgroup_uncharge(folio); <<
975 __folio_clear_locked(folio); <<
976 } else { <<
977 /* 948 /*
978 * The folio might have been e !! 949 * The page might have been evicted from cache only
979 * recently, in which case it 950 * recently, in which case it should be activated like
980 * any other repeatedly access !! 951 * any other repeatedly accessed page.
981 * The exception is folios get !! 952 * The exception is pages getting rewritten; evicting other
982 * data from the working set, 953 * data from the working set, only to cache data that will
983 * get overwritten with someth 954 * get overwritten with something else, is a waste of memory.
984 */ 955 */
985 WARN_ON_ONCE(folio_test_active !! 956 WARN_ON_ONCE(PageActive(page));
986 if (!(gfp & __GFP_WRITE) && sh !! 957 if (!(gfp_mask & __GFP_WRITE) && shadow)
987 workingset_refault(fol !! 958 workingset_refault(page, shadow);
988 folio_add_lru(folio); !! 959 lru_cache_add(page);
989 } 960 }
990 return ret; 961 return ret;
991 } 962 }
992 EXPORT_SYMBOL_GPL(filemap_add_folio); !! 963 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
993 964
994 #ifdef CONFIG_NUMA 965 #ifdef CONFIG_NUMA
995 struct folio *filemap_alloc_folio_noprof(gfp_t !! 966 struct page *__page_cache_alloc(gfp_t gfp)
996 { 967 {
997 int n; 968 int n;
998 struct folio *folio; !! 969 struct page *page;
999 970
1000 if (cpuset_do_page_mem_spread()) { 971 if (cpuset_do_page_mem_spread()) {
1001 unsigned int cpuset_mems_cook 972 unsigned int cpuset_mems_cookie;
1002 do { 973 do {
1003 cpuset_mems_cookie = 974 cpuset_mems_cookie = read_mems_allowed_begin();
1004 n = cpuset_mem_spread 975 n = cpuset_mem_spread_node();
1005 folio = __folio_alloc !! 976 page = __alloc_pages_node(n, gfp, 0);
1006 } while (!folio && read_mems_ !! 977 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
1007 978
1008 return folio; !! 979 return page;
1009 } 980 }
1010 return folio_alloc_noprof(gfp, order) !! 981 return alloc_pages(gfp, 0);
1011 } 982 }
1012 EXPORT_SYMBOL(filemap_alloc_folio_noprof); !! 983 EXPORT_SYMBOL(__page_cache_alloc);
1013 #endif 984 #endif
1014 985
1015 /* 986 /*
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 987 * In order to wait for pages to become available there must be
1055 * waitqueues associated with pages. By using 988 * waitqueues associated with pages. By using a hash table of
1056 * waitqueues where the bucket discipline is 989 * waitqueues where the bucket discipline is to maintain all
1057 * waiters on the same queue and wake all whe 990 * waiters on the same queue and wake all when any of the pages
1058 * become available, and for the woken contex 991 * become available, and for the woken contexts to check to be
1059 * sure the appropriate page became available 992 * sure the appropriate page became available, this saves space
1060 * at a cost of "thundering herd" phenomena d 993 * at a cost of "thundering herd" phenomena during rare hash
1061 * collisions. 994 * collisions.
1062 */ 995 */
1063 #define PAGE_WAIT_TABLE_BITS 8 996 #define PAGE_WAIT_TABLE_BITS 8
1064 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_ 997 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1065 static wait_queue_head_t folio_wait_table[PAG !! 998 static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1066 999
1067 static wait_queue_head_t *folio_waitqueue(str !! 1000 static wait_queue_head_t *page_waitqueue(struct page *page)
1068 { 1001 {
1069 return &folio_wait_table[hash_ptr(fol !! 1002 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
1070 } 1003 }
1071 1004
1072 void __init pagecache_init(void) 1005 void __init pagecache_init(void)
1073 { 1006 {
1074 int i; 1007 int i;
1075 1008
1076 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; 1009 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1077 init_waitqueue_head(&folio_wa !! 1010 init_waitqueue_head(&page_wait_table[i]);
1078 1011
1079 page_writeback_init(); 1012 page_writeback_init();
1080 } 1013 }
1081 1014
1082 /* !! 1015 /* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */
1083 * The page wait code treats the "wait->flags !! 1016 struct wait_page_key {
1084 * we have multiple different kinds of waits, !! 1017 struct page *page;
1085 * one. !! 1018 int bit_nr;
1086 * !! 1019 int page_match;
1087 * We have: !! 1020 };
1088 * !! 1021
1089 * (a) no special bits set: !! 1022 struct wait_page_queue {
1090 * !! 1023 struct page *page;
1091 * We're just waiting for the bit to be !! 1024 int bit_nr;
1092 * calls the wakeup function, we set WQ_ !! 1025 wait_queue_entry_t wait;
1093 * and remove it from the wait queue. !! 1026 };
1094 * !! 1027
1095 * Simple and straightforward. <<
1096 * <<
1097 * (b) WQ_FLAG_EXCLUSIVE: <<
1098 * <<
1099 * The waiter is waiting to get the lock <<
1100 * be woken up to avoid any thundering h <<
1101 * WQ_FLAG_WOKEN bit, wake it up, and re <<
1102 * <<
1103 * This is the traditional exclusive wai <<
1104 * <<
1105 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM: <<
1106 * <<
1107 * The waiter is waiting to get the bit, <<
1108 * lock to be transferred to it for fair <<
1109 * cannot be taken, we stop walking the <<
1110 * the waiter. <<
1111 * <<
1112 * This is the "fair lock handoff" case, <<
1113 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to <<
1114 * that it now has the lock. <<
1115 */ <<
1116 static int wake_page_function(wait_queue_entr 1028 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1117 { 1029 {
1118 unsigned int flags; <<
1119 struct wait_page_key *key = arg; 1030 struct wait_page_key *key = arg;
1120 struct wait_page_queue *wait_page 1031 struct wait_page_queue *wait_page
1121 = container_of(wait, struct w 1032 = container_of(wait, struct wait_page_queue, wait);
1122 1033
1123 if (!wake_page_match(wait_page, key)) !! 1034 if (wait_page->page != key->page)
1124 return 0; !! 1035 return 0;
>> 1036 key->page_match = 1;
1125 1037
1126 /* !! 1038 if (wait_page->bit_nr != key->bit_nr)
1127 * If it's a lock handoff wait, we ge !! 1039 return 0;
1128 * stop walking (and do not wake it u <<
1129 */ <<
1130 flags = wait->flags; <<
1131 if (flags & WQ_FLAG_EXCLUSIVE) { <<
1132 if (test_bit(key->bit_nr, &ke <<
1133 return -1; <<
1134 if (flags & WQ_FLAG_CUSTOM) { <<
1135 if (test_and_set_bit( <<
1136 return -1; <<
1137 flags |= WQ_FLAG_DONE <<
1138 } <<
1139 } <<
1140 1040
1141 /* 1041 /*
1142 * We are holding the wait-queue lock !! 1042 * Stop walking if it's locked.
1143 * is waiting for this will be checki !! 1043 * Is this safe if put_and_wait_on_page_locked() is in use?
1144 * any locking. !! 1044 * Yes: the waker must hold a reference to this page, and if PG_locked
1145 * !! 1045 * has now already been set by another task, that task must also hold
1146 * So update the flags atomically, an !! 1046 * a reference to the *same usage* of this page; so there is no need
1147 * afterwards to avoid any races. Thi !! 1047 * to walk on to wake even the put_and_wait_on_page_locked() callers.
1148 * with the load-acquire in folio_wai <<
1149 */ 1048 */
1150 smp_store_release(&wait->flags, flags !! 1049 if (test_bit(key->bit_nr, &key->page->flags))
1151 wake_up_state(wait->private, mode); !! 1050 return -1;
1152 1051
1153 /* !! 1052 return autoremove_wake_function(wait, mode, sync, key);
1154 * Ok, we have successfully done what <<
1155 * and we can unconditionally remove <<
1156 * <<
1157 * Note that this pairs with the "fin <<
1158 * waiter, and has to be the absolute <<
1159 * After this list_del_init(&wait->en <<
1160 * might be de-allocated and the proc <<
1161 * exited. <<
1162 */ <<
1163 list_del_init_careful(&wait->entry); <<
1164 return (flags & WQ_FLAG_EXCLUSIVE) != <<
1165 } 1053 }
1166 1054
1167 static void folio_wake_bit(struct folio *foli !! 1055 static void wake_up_page_bit(struct page *page, int bit_nr)
1168 { 1056 {
1169 wait_queue_head_t *q = folio_waitqueu !! 1057 wait_queue_head_t *q = page_waitqueue(page);
1170 struct wait_page_key key; 1058 struct wait_page_key key;
1171 unsigned long flags; 1059 unsigned long flags;
>> 1060 wait_queue_entry_t bookmark;
1172 1061
1173 key.folio = folio; !! 1062 key.page = page;
1174 key.bit_nr = bit_nr; 1063 key.bit_nr = bit_nr;
1175 key.page_match = 0; 1064 key.page_match = 0;
1176 1065
>> 1066 bookmark.flags = 0;
>> 1067 bookmark.private = NULL;
>> 1068 bookmark.func = NULL;
>> 1069 INIT_LIST_HEAD(&bookmark.entry);
>> 1070
1177 spin_lock_irqsave(&q->lock, flags); 1071 spin_lock_irqsave(&q->lock, flags);
1178 __wake_up_locked_key(q, TASK_NORMAL, !! 1072 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
>> 1073
>> 1074 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
>> 1075 /*
>> 1076 * Take a breather from holding the lock,
>> 1077 * allow pages that finish wake up asynchronously
>> 1078 * to acquire the lock and remove themselves
>> 1079 * from wait queue
>> 1080 */
>> 1081 spin_unlock_irqrestore(&q->lock, flags);
>> 1082 cpu_relax();
>> 1083 spin_lock_irqsave(&q->lock, flags);
>> 1084 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
>> 1085 }
1179 1086
1180 /* 1087 /*
1181 * It's possible to miss clearing wai !! 1088 * It is possible for other pages to have collided on the waitqueue
1182 * waiters, but the hashed waitqueue !! 1089 * hash, so in that case check for a page match. That prevents a long-
1183 * That's okay, it's a rare case. The !! 1090 * term waiter
1184 * 1091 *
1185 * Note that, depending on the page p !! 1092 * It is still possible to miss a case here, when we woke page waiters
1186 * other), the flag may be cleared in !! 1093 * and removed them from the waitqueue, but there are still other
1187 * but that is not required for corre !! 1094 * page waiters.
1188 */ !! 1095 */
1189 if (!waitqueue_active(q) || !key.page !! 1096 if (!waitqueue_active(q) || !key.page_match) {
1190 folio_clear_waiters(folio); !! 1097 ClearPageWaiters(page);
1191 !! 1098 /*
>> 1099 * It's possible to miss clearing Waiters here, when we woke
>> 1100 * our page waiters, but the hashed waitqueue has waiters for
>> 1101 * other pages on it.
>> 1102 *
>> 1103 * That's okay, it's a rare case. The next waker will clear it.
>> 1104 */
>> 1105 }
1192 spin_unlock_irqrestore(&q->lock, flag 1106 spin_unlock_irqrestore(&q->lock, flags);
1193 } 1107 }
1194 1108
>> 1109 static void wake_up_page(struct page *page, int bit)
>> 1110 {
>> 1111 if (!PageWaiters(page))
>> 1112 return;
>> 1113 wake_up_page_bit(page, bit);
>> 1114 }
>> 1115
1195 /* 1116 /*
1196 * A choice of three behaviors for folio_wait !! 1117 * A choice of three behaviors for wait_on_page_bit_common():
1197 */ 1118 */
1198 enum behavior { 1119 enum behavior {
1199 EXCLUSIVE, /* Hold ref to page a 1120 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1200 * __folio_lock() wai !! 1121 * __lock_page() waiting on then setting PG_locked.
1201 */ 1122 */
1202 SHARED, /* Hold ref to page a 1123 SHARED, /* Hold ref to page and check the bit when woken, like
1203 * folio_wait_writeba !! 1124 * wait_on_page_writeback() waiting on PG_writeback.
1204 */ 1125 */
1205 DROP, /* Drop ref to page b 1126 DROP, /* Drop ref to page before wait, no check when woken,
1206 * like folio_put_wai !! 1127 * like put_and_wait_on_page_locked() on PG_locked.
1207 */ 1128 */
1208 }; 1129 };
1209 1130
1210 /* !! 1131 static inline int wait_on_page_bit_common(wait_queue_head_t *q,
1211 * Attempt to check (or get) the folio flag, !! 1132 struct page *page, int bit_nr, int state, enum behavior behavior)
1212 * if successful. <<
1213 */ <<
1214 static inline bool folio_trylock_flag(struct <<
1215 struc <<
1216 { 1133 {
1217 if (wait->flags & WQ_FLAG_EXCLUSIVE) <<
1218 if (test_and_set_bit(bit_nr, <<
1219 return false; <<
1220 } else if (test_bit(bit_nr, &folio->f <<
1221 return false; <<
1222 <<
1223 wait->flags |= WQ_FLAG_WOKEN | WQ_FLA <<
1224 return true; <<
1225 } <<
1226 <<
1227 /* How many times do we accept lock stealing <<
1228 int sysctl_page_lock_unfairness = 5; <<
1229 <<
1230 static inline int folio_wait_bit_common(struc <<
1231 int state, enum behavior beha <<
1232 { <<
1233 wait_queue_head_t *q = folio_waitqueu <<
1234 int unfairness = sysctl_page_lock_unf <<
1235 struct wait_page_queue wait_page; 1134 struct wait_page_queue wait_page;
1236 wait_queue_entry_t *wait = &wait_page 1135 wait_queue_entry_t *wait = &wait_page.wait;
>> 1136 bool bit_is_set;
1237 bool thrashing = false; 1137 bool thrashing = false;
>> 1138 bool delayacct = false;
1238 unsigned long pflags; 1139 unsigned long pflags;
1239 bool in_thrashing; !! 1140 int ret = 0;
1240 1141
1241 if (bit_nr == PG_locked && 1142 if (bit_nr == PG_locked &&
1242 !folio_test_uptodate(folio) && fo !! 1143 !PageUptodate(page) && PageWorkingset(page)) {
1243 delayacct_thrashing_start(&in !! 1144 if (!PageSwapBacked(page)) {
>> 1145 delayacct_thrashing_start();
>> 1146 delayacct = true;
>> 1147 }
1244 psi_memstall_enter(&pflags); 1148 psi_memstall_enter(&pflags);
1245 thrashing = true; 1149 thrashing = true;
1246 } 1150 }
1247 1151
1248 init_wait(wait); 1152 init_wait(wait);
>> 1153 wait->flags = behavior == EXCLUSIVE ? WQ_FLAG_EXCLUSIVE : 0;
1249 wait->func = wake_page_function; 1154 wait->func = wake_page_function;
1250 wait_page.folio = folio; !! 1155 wait_page.page = page;
1251 wait_page.bit_nr = bit_nr; 1156 wait_page.bit_nr = bit_nr;
1252 1157
1253 repeat: <<
1254 wait->flags = 0; <<
1255 if (behavior == EXCLUSIVE) { <<
1256 wait->flags = WQ_FLAG_EXCLUSI <<
1257 if (--unfairness < 0) <<
1258 wait->flags |= WQ_FLA <<
1259 } <<
1260 <<
1261 /* <<
1262 * Do one last check whether we can g <<
1263 * page bit synchronously. <<
1264 * <<
1265 * Do the folio_set_waiters() marking <<
1266 * to let any waker we _just_ missed <<
1267 * need to wake us up (otherwise they <<
1268 * even go to the slow case that look <<
1269 * page queue), and add ourselves to <<
1270 * queue if we need to sleep. <<
1271 * <<
1272 * This part needs to be done under t <<
1273 * lock to avoid races. <<
1274 */ <<
1275 spin_lock_irq(&q->lock); <<
1276 folio_set_waiters(folio); <<
1277 if (!folio_trylock_flag(folio, bit_nr <<
1278 __add_wait_queue_entry_tail(q <<
1279 spin_unlock_irq(&q->lock); <<
1280 <<
1281 /* <<
1282 * From now on, all the logic will be <<
1283 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE <<
1284 * see whether the page bit testing h <<
1285 * been done by the wake function. <<
1286 * <<
1287 * We can drop our reference to the f <<
1288 */ <<
1289 if (behavior == DROP) <<
1290 folio_put(folio); <<
1291 <<
1292 /* <<
1293 * Note that until the "finish_wait() <<
1294 * we see the WQ_FLAG_WOKEN flag, we <<
1295 * be very careful with the 'wait->fl <<
1296 * we may race with a waker that sets <<
1297 */ <<
1298 for (;;) { 1158 for (;;) {
1299 unsigned int flags; !! 1159 spin_lock_irq(&q->lock);
>> 1160
>> 1161 if (likely(list_empty(&wait->entry))) {
>> 1162 __add_wait_queue_entry_tail(q, wait);
>> 1163 SetPageWaiters(page);
>> 1164 }
1300 1165
1301 set_current_state(state); 1166 set_current_state(state);
1302 1167
1303 /* Loop until we've been woke !! 1168 spin_unlock_irq(&q->lock);
1304 flags = smp_load_acquire(&wai !! 1169
1305 if (!(flags & WQ_FLAG_WOKEN)) !! 1170 bit_is_set = test_bit(bit_nr, &page->flags);
1306 if (signal_pending_st !! 1171 if (behavior == DROP)
1307 break; !! 1172 put_page(page);
1308 1173
>> 1174 if (likely(bit_is_set))
1309 io_schedule(); 1175 io_schedule();
1310 continue; !! 1176
>> 1177 if (behavior == EXCLUSIVE) {
>> 1178 if (!test_and_set_bit_lock(bit_nr, &page->flags))
>> 1179 break;
>> 1180 } else if (behavior == SHARED) {
>> 1181 if (!test_bit(bit_nr, &page->flags))
>> 1182 break;
1311 } 1183 }
1312 1184
1313 /* If we were non-exclusive, !! 1185 if (signal_pending_state(state, current)) {
1314 if (behavior != EXCLUSIVE) !! 1186 ret = -EINTR;
1315 break; 1187 break;
>> 1188 }
1316 1189
1317 /* If the waker got the lock !! 1190 if (behavior == DROP) {
1318 if (flags & WQ_FLAG_DONE) !! 1191 /*
>> 1192 * We can no longer safely access page->flags:
>> 1193 * even if CONFIG_MEMORY_HOTREMOVE is not enabled,
>> 1194 * there is a risk of waiting forever on a page reused
>> 1195 * for something that keeps it locked indefinitely.
>> 1196 * But best check for -EINTR above before breaking.
>> 1197 */
1319 break; 1198 break;
1320 !! 1199 }
1321 /* <<
1322 * Otherwise, if we're gettin <<
1323 * try to get it ourselves. <<
1324 * <<
1325 * And if that fails, we'll h <<
1326 */ <<
1327 if (unlikely(test_and_set_bit <<
1328 goto repeat; <<
1329 <<
1330 wait->flags |= WQ_FLAG_DONE; <<
1331 break; <<
1332 } 1200 }
1333 1201
1334 /* <<
1335 * If a signal happened, this 'finish <<
1336 * waiter from the wait-queues, but t <<
1337 * set. That's ok. The next wakeup wi <<
1338 * to do it here would be difficult a <<
1339 */ <<
1340 finish_wait(q, wait); 1202 finish_wait(q, wait);
1341 1203
1342 if (thrashing) { 1204 if (thrashing) {
1343 delayacct_thrashing_end(&in_t !! 1205 if (delayacct)
>> 1206 delayacct_thrashing_end();
1344 psi_memstall_leave(&pflags); 1207 psi_memstall_leave(&pflags);
1345 } 1208 }
1346 1209
1347 /* 1210 /*
1348 * NOTE! The wait->flags weren't stab !! 1211 * A signal could leave PageWaiters set. Clearing it here if
1349 * 'finish_wait()', and we could have !! 1212 * !waitqueue_active would be possible (by open-coding finish_wait),
1350 * to a signal, and had a wakeup even !! 1213 * but still fail to catch it in the case of wait hash collision. We
1351 * test but before the 'finish_wait() !! 1214 * already can fail to clear wait hash collision cases, so don't
1352 * !! 1215 * bother with signals either.
1353 * So only after the finish_wait() ca <<
1354 * if we got woken up or not, so we c <<
1355 * return value based on that state w <<
1356 * <<
1357 * Also note that WQ_FLAG_WOKEN is su <<
1358 * waiter, but an exclusive one requi <<
1359 */ <<
1360 if (behavior == EXCLUSIVE) <<
1361 return wait->flags & WQ_FLAG_ <<
1362 <<
1363 return wait->flags & WQ_FLAG_WOKEN ? <<
1364 } <<
1365 <<
1366 #ifdef CONFIG_MIGRATION <<
1367 /** <<
1368 * migration_entry_wait_on_locked - Wait for <<
1369 * @entry: migration swap entry. <<
1370 * @ptl: already locked ptl. This function wi <<
1371 * <<
1372 * Wait for a migration entry referencing the <<
1373 * equivalent to put_and_wait_on_page_locked( <<
1374 * this can be called without taking a refere <<
1375 * should be called while holding the ptl for <<
1376 * the page. <<
1377 * <<
1378 * Returns after unlocking the ptl. <<
1379 * <<
1380 * This follows the same logic as folio_wait_ <<
1381 * there. <<
1382 */ <<
1383 void migration_entry_wait_on_locked(swp_entry <<
1384 __releases(ptl) <<
1385 { <<
1386 struct wait_page_queue wait_page; <<
1387 wait_queue_entry_t *wait = &wait_page <<
1388 bool thrashing = false; <<
1389 unsigned long pflags; <<
1390 bool in_thrashing; <<
1391 wait_queue_head_t *q; <<
1392 struct folio *folio = pfn_swap_entry_ <<
1393 <<
1394 q = folio_waitqueue(folio); <<
1395 if (!folio_test_uptodate(folio) && fo <<
1396 delayacct_thrashing_start(&in <<
1397 psi_memstall_enter(&pflags); <<
1398 thrashing = true; <<
1399 } <<
1400 <<
1401 init_wait(wait); <<
1402 wait->func = wake_page_function; <<
1403 wait_page.folio = folio; <<
1404 wait_page.bit_nr = PG_locked; <<
1405 wait->flags = 0; <<
1406 <<
1407 spin_lock_irq(&q->lock); <<
1408 folio_set_waiters(folio); <<
1409 if (!folio_trylock_flag(folio, PG_loc <<
1410 __add_wait_queue_entry_tail(q <<
1411 spin_unlock_irq(&q->lock); <<
1412 <<
1413 /* <<
1414 * If a migration entry exists for th <<
1415 * a valid reference to the page, and <<
1416 * migration entry. So the page is va <<
1417 */ 1216 */
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 1217
1431 io_schedule(); !! 1218 return ret;
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 } 1219 }
1444 #endif <<
1445 1220
1446 void folio_wait_bit(struct folio *folio, int !! 1221 void wait_on_page_bit(struct page *page, int bit_nr)
1447 { 1222 {
1448 folio_wait_bit_common(folio, bit_nr, !! 1223 wait_queue_head_t *q = page_waitqueue(page);
>> 1224 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1449 } 1225 }
1450 EXPORT_SYMBOL(folio_wait_bit); !! 1226 EXPORT_SYMBOL(wait_on_page_bit);
1451 1227
1452 int folio_wait_bit_killable(struct folio *fol !! 1228 int wait_on_page_bit_killable(struct page *page, int bit_nr)
1453 { 1229 {
1454 return folio_wait_bit_common(folio, b !! 1230 wait_queue_head_t *q = page_waitqueue(page);
>> 1231 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED);
1455 } 1232 }
1456 EXPORT_SYMBOL(folio_wait_bit_killable); !! 1233 EXPORT_SYMBOL(wait_on_page_bit_killable);
1457 1234
1458 /** 1235 /**
1459 * folio_put_wait_locked - Drop a reference a !! 1236 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1460 * @folio: The folio to wait for. !! 1237 * @page: The page to wait for.
1461 * @state: The sleep state (TASK_KILLABLE, TA <<
1462 * 1238 *
1463 * The caller should hold a reference on @fol !! 1239 * The caller should hold a reference on @page. They expect the page to
1464 * become unlocked relatively soon, but do no 1240 * become unlocked relatively soon, but do not wish to hold up migration
1465 * (for example) by holding the reference whi !! 1241 * (for example) by holding the reference while waiting for the page to
1466 * come unlocked. After this function return 1242 * come unlocked. After this function returns, the caller should not
1467 * dereference @folio. !! 1243 * dereference @page.
1468 * <<
1469 * Return: 0 if the folio was unlocked or -EI <<
1470 */ 1244 */
1471 static int folio_put_wait_locked(struct folio !! 1245 void put_and_wait_on_page_locked(struct page *page)
1472 { 1246 {
1473 return folio_wait_bit_common(folio, P !! 1247 wait_queue_head_t *q;
>> 1248
>> 1249 page = compound_head(page);
>> 1250 q = page_waitqueue(page);
>> 1251 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, DROP);
1474 } 1252 }
1475 1253
1476 /** 1254 /**
1477 * folio_add_wait_queue - Add an arbitrary wa !! 1255 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1478 * @folio: Folio defining the wait queue of i !! 1256 * @page: Page defining the wait queue of interest
1479 * @waiter: Waiter to add to the queue 1257 * @waiter: Waiter to add to the queue
1480 * 1258 *
1481 * Add an arbitrary @waiter to the wait queue !! 1259 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1482 */ 1260 */
1483 void folio_add_wait_queue(struct folio *folio !! 1261 void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1484 { 1262 {
1485 wait_queue_head_t *q = folio_waitqueu !! 1263 wait_queue_head_t *q = page_waitqueue(page);
1486 unsigned long flags; 1264 unsigned long flags;
1487 1265
1488 spin_lock_irqsave(&q->lock, flags); 1266 spin_lock_irqsave(&q->lock, flags);
1489 __add_wait_queue_entry_tail(q, waiter 1267 __add_wait_queue_entry_tail(q, waiter);
1490 folio_set_waiters(folio); !! 1268 SetPageWaiters(page);
1491 spin_unlock_irqrestore(&q->lock, flag 1269 spin_unlock_irqrestore(&q->lock, flags);
1492 } 1270 }
1493 EXPORT_SYMBOL_GPL(folio_add_wait_queue); !! 1271 EXPORT_SYMBOL_GPL(add_page_wait_queue);
1494 1272
1495 /** !! 1273 #ifndef clear_bit_unlock_is_negative_byte
1496 * folio_unlock - Unlock a locked folio. <<
1497 * @folio: The folio. <<
1498 * <<
1499 * Unlocks the folio and wakes up any thread <<
1500 * <<
1501 * Context: May be called from interrupt or p <<
1502 * called from NMI context. <<
1503 */ <<
1504 void folio_unlock(struct folio *folio) <<
1505 { <<
1506 /* Bit 7 allows x86 to check the byte <<
1507 BUILD_BUG_ON(PG_waiters != 7); <<
1508 BUILD_BUG_ON(PG_locked > 7); <<
1509 VM_BUG_ON_FOLIO(!folio_test_locked(fo <<
1510 if (folio_xor_flags_has_waiters(folio <<
1511 folio_wake_bit(folio, PG_lock <<
1512 } <<
1513 EXPORT_SYMBOL(folio_unlock); <<
1514 1274
1515 /** !! 1275 /*
1516 * folio_end_read - End read on a folio. !! 1276 * PG_waiters is the high bit in the same byte as PG_lock.
1517 * @folio: The folio. <<
1518 * @success: True if all reads completed succ <<
1519 * <<
1520 * When all reads against a folio have comple <<
1521 * call this function to let the pagecache kn <<
1522 * are outstanding. This will unlock the fol <<
1523 * sleeping on the lock. The folio will also <<
1524 * reads succeeded. <<
1525 * <<
1526 * Context: May be called from interrupt or p <<
1527 * called from NMI context. <<
1528 */ <<
1529 void folio_end_read(struct folio *folio, bool <<
1530 { <<
1531 unsigned long mask = 1 << PG_locked; <<
1532 <<
1533 /* Must be in bottom byte for x86 to <<
1534 BUILD_BUG_ON(PG_uptodate > 7); <<
1535 VM_BUG_ON_FOLIO(!folio_test_locked(fo <<
1536 VM_BUG_ON_FOLIO(folio_test_uptodate(f <<
1537 <<
1538 if (likely(success)) <<
1539 mask |= 1 << PG_uptodate; <<
1540 if (folio_xor_flags_has_waiters(folio <<
1541 folio_wake_bit(folio, PG_lock <<
1542 } <<
1543 EXPORT_SYMBOL(folio_end_read); <<
1544 <<
1545 /** <<
1546 * folio_end_private_2 - Clear PG_private_2 a <<
1547 * @folio: The folio. <<
1548 * 1277 *
1549 * Clear the PG_private_2 bit on a folio and !! 1278 * On x86 (and on many other architectures), we can clear PG_lock and
1550 * it. The folio reference held for PG_priva !! 1279 * test the sign bit at the same time. But if the architecture does
>> 1280 * not support that special operation, we just do this all by hand
>> 1281 * instead.
1551 * 1282 *
1552 * This is, for example, used when a netfs fo !! 1283 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1553 * disk cache, thereby allowing writes to the !! 1284 * being cleared, but a memory barrier should be unneccssary since it is
1554 * serialised. !! 1285 * in the same byte as PG_locked.
1555 */ 1286 */
1556 void folio_end_private_2(struct folio *folio) !! 1287 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1557 { 1288 {
1558 VM_BUG_ON_FOLIO(!folio_test_private_2 !! 1289 clear_bit_unlock(nr, mem);
1559 clear_bit_unlock(PG_private_2, folio_ !! 1290 /* smp_mb__after_atomic(); */
1560 folio_wake_bit(folio, PG_private_2); !! 1291 return test_bit(PG_waiters, mem);
1561 folio_put(folio); <<
1562 } 1292 }
1563 EXPORT_SYMBOL(folio_end_private_2); !! 1293
>> 1294 #endif
1564 1295
1565 /** 1296 /**
1566 * folio_wait_private_2 - Wait for PG_private !! 1297 * unlock_page - unlock a locked page
1567 * @folio: The folio to wait on. !! 1298 * @page: the page
1568 * 1299 *
1569 * Wait for PG_private_2 to be cleared on a f !! 1300 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
>> 1301 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
>> 1302 * mechanism between PageLocked pages and PageWriteback pages is shared.
>> 1303 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
>> 1304 *
>> 1305 * Note that this depends on PG_waiters being the sign bit in the byte
>> 1306 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
>> 1307 * clear the PG_locked bit and test PG_waiters at the same time fairly
>> 1308 * portably (architectures that do LL/SC can test any bit, while x86 can
>> 1309 * test the sign bit).
1570 */ 1310 */
1571 void folio_wait_private_2(struct folio *folio !! 1311 void unlock_page(struct page *page)
1572 { 1312 {
1573 while (folio_test_private_2(folio)) !! 1313 BUILD_BUG_ON(PG_waiters != 7);
1574 folio_wait_bit(folio, PG_priv !! 1314 page = compound_head(page);
>> 1315 VM_BUG_ON_PAGE(!PageLocked(page), page);
>> 1316 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
>> 1317 wake_up_page_bit(page, PG_locked);
1575 } 1318 }
1576 EXPORT_SYMBOL(folio_wait_private_2); !! 1319 EXPORT_SYMBOL(unlock_page);
1577 1320
1578 /** 1321 /**
1579 * folio_wait_private_2_killable - Wait for P !! 1322 * end_page_writeback - end writeback against a page
1580 * @folio: The folio to wait on. !! 1323 * @page: the page
1581 * <<
1582 * Wait for PG_private_2 to be cleared on a f <<
1583 * received by the calling task. <<
1584 * <<
1585 * Return: <<
1586 * - 0 if successful. <<
1587 * - -EINTR if a fatal signal was encountered <<
1588 */ 1324 */
1589 int folio_wait_private_2_killable(struct foli !! 1325 void end_page_writeback(struct page *page)
1590 { 1326 {
1591 int ret = 0; !! 1327 /*
1592 !! 1328 * TestClearPageReclaim could be used here but it is an atomic
1593 while (folio_test_private_2(folio)) { !! 1329 * operation and overkill in this particular case. Failing to
1594 ret = folio_wait_bit_killable !! 1330 * shuffle a page marked for immediate reclaim is too mild to
1595 if (ret < 0) !! 1331 * justify taking an atomic operation penalty at the end of
1596 break; !! 1332 * ever page writeback.
>> 1333 */
>> 1334 if (PageReclaim(page)) {
>> 1335 ClearPageReclaim(page);
>> 1336 rotate_reclaimable_page(page);
1597 } 1337 }
1598 1338
1599 return ret; !! 1339 if (!test_clear_page_writeback(page))
>> 1340 BUG();
>> 1341
>> 1342 smp_mb__after_atomic();
>> 1343 wake_up_page(page, PG_writeback);
1600 } 1344 }
1601 EXPORT_SYMBOL(folio_wait_private_2_killable); !! 1345 EXPORT_SYMBOL(end_page_writeback);
1602 1346
1603 /** !! 1347 /*
1604 * folio_end_writeback - End writeback agains !! 1348 * After completing I/O on a page, call this routine to update the page
1605 * @folio: The folio. !! 1349 * flags appropriately
1606 * <<
1607 * The folio must actually be under writeback <<
1608 * <<
1609 * Context: May be called from process or int <<
1610 */ 1350 */
1611 void folio_end_writeback(struct folio *folio) !! 1351 void page_endio(struct page *page, bool is_write, int err)
1612 { 1352 {
1613 VM_BUG_ON_FOLIO(!folio_test_writeback !! 1353 if (!is_write) {
>> 1354 if (!err) {
>> 1355 SetPageUptodate(page);
>> 1356 } else {
>> 1357 ClearPageUptodate(page);
>> 1358 SetPageError(page);
>> 1359 }
>> 1360 unlock_page(page);
>> 1361 } else {
>> 1362 if (err) {
>> 1363 struct address_space *mapping;
1614 1364
1615 /* !! 1365 SetPageError(page);
1616 * folio_test_clear_reclaim() could b !! 1366 mapping = page_mapping(page);
1617 * atomic operation and overkill in t !! 1367 if (mapping)
1618 * to shuffle a folio marked for imme !! 1368 mapping_set_error(mapping, err);
1619 * a gain to justify taking an atomic !! 1369 }
1620 * end of every folio writeback. !! 1370 end_page_writeback(page);
1621 */ <<
1622 if (folio_test_reclaim(folio)) { <<
1623 folio_clear_reclaim(folio); <<
1624 folio_rotate_reclaimable(foli <<
1625 } 1371 }
1626 <<
1627 /* <<
1628 * Writeback does not hold a folio re <<
1629 * on truncation to wait for the clea <<
1630 * But here we must make sure that th <<
1631 * reused before the folio_wake_bit() <<
1632 */ <<
1633 folio_get(folio); <<
1634 if (__folio_end_writeback(folio)) <<
1635 folio_wake_bit(folio, PG_writ <<
1636 acct_reclaim_writeback(folio); <<
1637 folio_put(folio); <<
1638 } 1372 }
1639 EXPORT_SYMBOL(folio_end_writeback); !! 1373 EXPORT_SYMBOL_GPL(page_endio);
1640 1374
1641 /** 1375 /**
1642 * __folio_lock - Get a lock on the folio, as !! 1376 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1643 * @folio: The folio to lock !! 1377 * @__page: the page to lock
1644 */ 1378 */
1645 void __folio_lock(struct folio *folio) !! 1379 void __lock_page(struct page *__page)
1646 { 1380 {
1647 folio_wait_bit_common(folio, PG_locke !! 1381 struct page *page = compound_head(__page);
>> 1382 wait_queue_head_t *q = page_waitqueue(page);
>> 1383 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE,
1648 EXCLUSIVE); 1384 EXCLUSIVE);
1649 } 1385 }
1650 EXPORT_SYMBOL(__folio_lock); !! 1386 EXPORT_SYMBOL(__lock_page);
1651 1387
1652 int __folio_lock_killable(struct folio *folio !! 1388 int __lock_page_killable(struct page *__page)
1653 { 1389 {
1654 return folio_wait_bit_common(folio, P !! 1390 struct page *page = compound_head(__page);
>> 1391 wait_queue_head_t *q = page_waitqueue(page);
>> 1392 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE,
1655 EXCLU 1393 EXCLUSIVE);
1656 } 1394 }
1657 EXPORT_SYMBOL_GPL(__folio_lock_killable); !! 1395 EXPORT_SYMBOL_GPL(__lock_page_killable);
1658 <<
1659 static int __folio_lock_async(struct folio *f <<
1660 { <<
1661 struct wait_queue_head *q = folio_wai <<
1662 int ret; <<
1663 <<
1664 wait->folio = folio; <<
1665 wait->bit_nr = PG_locked; <<
1666 <<
1667 spin_lock_irq(&q->lock); <<
1668 __add_wait_queue_entry_tail(q, &wait- <<
1669 folio_set_waiters(folio); <<
1670 ret = !folio_trylock(folio); <<
1671 /* <<
1672 * If we were successful now, we know <<
1673 * waitqueue as we're still under the <<
1674 * safe to remove and return success, <<
1675 * isn't going to trigger. <<
1676 */ <<
1677 if (!ret) <<
1678 __remove_wait_queue(q, &wait- <<
1679 else <<
1680 ret = -EIOCBQUEUED; <<
1681 spin_unlock_irq(&q->lock); <<
1682 return ret; <<
1683 } <<
1684 1396
1685 /* 1397 /*
1686 * Return values: 1398 * Return values:
1687 * 0 - folio is locked. !! 1399 * 1 - page is locked; mmap_sem is still held.
1688 * non-zero - folio is not locked. !! 1400 * 0 - page is not locked.
1689 * mmap_lock or per-VMA lock has been rel !! 1401 * mmap_sem has been released (up_read()), unless flags had both
1690 * vma_end_read()), unless flags had both !! 1402 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1691 * FAULT_FLAG_RETRY_NOWAIT set, in which !! 1403 * which case mmap_sem is still held.
1692 * !! 1404 *
1693 * If neither ALLOW_RETRY nor KILLABLE are se !! 1405 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1694 * with the folio locked and the mmap_lock/pe !! 1406 * with the page locked and the mmap_sem unperturbed.
1695 */ 1407 */
1696 vm_fault_t __folio_lock_or_retry(struct folio !! 1408 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
>> 1409 unsigned int flags)
1697 { 1410 {
1698 unsigned int flags = vmf->flags; !! 1411 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1699 <<
1700 if (fault_flag_allow_retry_first(flag <<
1701 /* 1412 /*
1702 * CAUTION! In this case, mma !! 1413 * CAUTION! In this case, mmap_sem is not released
1703 * released even though retur !! 1414 * even though return 0.
1704 */ 1415 */
1705 if (flags & FAULT_FLAG_RETRY_ 1416 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1706 return VM_FAULT_RETRY !! 1417 return 0;
1707 1418
1708 release_fault_lock(vmf); !! 1419 up_read(&mm->mmap_sem);
1709 if (flags & FAULT_FLAG_KILLAB 1420 if (flags & FAULT_FLAG_KILLABLE)
1710 folio_wait_locked_kil !! 1421 wait_on_page_locked_killable(page);
1711 else 1422 else
1712 folio_wait_locked(fol !! 1423 wait_on_page_locked(page);
1713 return VM_FAULT_RETRY; !! 1424 return 0;
1714 } <<
1715 if (flags & FAULT_FLAG_KILLABLE) { <<
1716 bool ret; <<
1717 <<
1718 ret = __folio_lock_killable(f <<
1719 if (ret) { <<
1720 release_fault_lock(vm <<
1721 return VM_FAULT_RETRY <<
1722 } <<
1723 } else { 1425 } else {
1724 __folio_lock(folio); !! 1426 if (flags & FAULT_FLAG_KILLABLE) {
1725 } !! 1427 int ret;
1726 1428
1727 return 0; !! 1429 ret = __lock_page_killable(page);
>> 1430 if (ret) {
>> 1431 up_read(&mm->mmap_sem);
>> 1432 return 0;
>> 1433 }
>> 1434 } else
>> 1435 __lock_page(page);
>> 1436 return 1;
>> 1437 }
1728 } 1438 }
1729 1439
1730 /** 1440 /**
1731 * page_cache_next_miss() - Find the next gap 1441 * page_cache_next_miss() - Find the next gap in the page cache.
1732 * @mapping: Mapping. 1442 * @mapping: Mapping.
1733 * @index: Index. 1443 * @index: Index.
1734 * @max_scan: Maximum range to search. 1444 * @max_scan: Maximum range to search.
1735 * 1445 *
1736 * Search the range [index, min(index + max_s 1446 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1737 * gap with the lowest index. 1447 * gap with the lowest index.
1738 * 1448 *
1739 * This function may be called under the rcu_ 1449 * This function may be called under the rcu_read_lock. However, this will
1740 * not atomically search a snapshot of the ca 1450 * not atomically search a snapshot of the cache at a single point in time.
1741 * For example, if a gap is created at index 1451 * For example, if a gap is created at index 5, then subsequently a gap is
1742 * created at index 10, page_cache_next_miss 1452 * created at index 10, page_cache_next_miss covering both indices may
1743 * return 10 if called under the rcu_read_loc 1453 * return 10 if called under the rcu_read_lock.
1744 * 1454 *
1745 * Return: The index of the gap if found, oth 1455 * Return: The index of the gap if found, otherwise an index outside the
1746 * range specified (in which case 'return - i 1456 * range specified (in which case 'return - index >= max_scan' will be true).
1747 * In the rare case of index wrap-around, 0 w 1457 * In the rare case of index wrap-around, 0 will be returned.
1748 */ 1458 */
1749 pgoff_t page_cache_next_miss(struct address_s 1459 pgoff_t page_cache_next_miss(struct address_space *mapping,
1750 pgoff_t index, u 1460 pgoff_t index, unsigned long max_scan)
1751 { 1461 {
1752 XA_STATE(xas, &mapping->i_pages, inde 1462 XA_STATE(xas, &mapping->i_pages, index);
1753 1463
1754 while (max_scan--) { 1464 while (max_scan--) {
1755 void *entry = xas_next(&xas); 1465 void *entry = xas_next(&xas);
1756 if (!entry || xa_is_value(ent 1466 if (!entry || xa_is_value(entry))
1757 return xas.xa_index; !! 1467 break;
1758 if (xas.xa_index == 0) 1468 if (xas.xa_index == 0)
1759 return 0; !! 1469 break;
1760 } 1470 }
1761 1471
1762 return index + max_scan; !! 1472 return xas.xa_index;
1763 } 1473 }
1764 EXPORT_SYMBOL(page_cache_next_miss); 1474 EXPORT_SYMBOL(page_cache_next_miss);
1765 1475
1766 /** 1476 /**
1767 * page_cache_prev_miss() - Find the previous 1477 * page_cache_prev_miss() - Find the previous gap in the page cache.
1768 * @mapping: Mapping. 1478 * @mapping: Mapping.
1769 * @index: Index. 1479 * @index: Index.
1770 * @max_scan: Maximum range to search. 1480 * @max_scan: Maximum range to search.
1771 * 1481 *
1772 * Search the range [max(index - max_scan + 1 1482 * Search the range [max(index - max_scan + 1, 0), index] for the
1773 * gap with the highest index. 1483 * gap with the highest index.
1774 * 1484 *
1775 * This function may be called under the rcu_ 1485 * This function may be called under the rcu_read_lock. However, this will
1776 * not atomically search a snapshot of the ca 1486 * not atomically search a snapshot of the cache at a single point in time.
1777 * For example, if a gap is created at index 1487 * For example, if a gap is created at index 10, then subsequently a gap is
1778 * created at index 5, page_cache_prev_miss() 1488 * created at index 5, page_cache_prev_miss() covering both indices may
1779 * return 5 if called under the rcu_read_lock 1489 * return 5 if called under the rcu_read_lock.
1780 * 1490 *
1781 * Return: The index of the gap if found, oth 1491 * Return: The index of the gap if found, otherwise an index outside the
1782 * range specified (in which case 'index - re 1492 * range specified (in which case 'index - return >= max_scan' will be true).
1783 * In the rare case of wrap-around, ULONG_MAX 1493 * In the rare case of wrap-around, ULONG_MAX will be returned.
1784 */ 1494 */
1785 pgoff_t page_cache_prev_miss(struct address_s 1495 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1786 pgoff_t index, u 1496 pgoff_t index, unsigned long max_scan)
1787 { 1497 {
1788 XA_STATE(xas, &mapping->i_pages, inde 1498 XA_STATE(xas, &mapping->i_pages, index);
1789 1499
1790 while (max_scan--) { 1500 while (max_scan--) {
1791 void *entry = xas_prev(&xas); 1501 void *entry = xas_prev(&xas);
1792 if (!entry || xa_is_value(ent 1502 if (!entry || xa_is_value(entry))
1793 break; 1503 break;
1794 if (xas.xa_index == ULONG_MAX 1504 if (xas.xa_index == ULONG_MAX)
1795 break; 1505 break;
1796 } 1506 }
1797 1507
1798 return xas.xa_index; 1508 return xas.xa_index;
1799 } 1509 }
1800 EXPORT_SYMBOL(page_cache_prev_miss); 1510 EXPORT_SYMBOL(page_cache_prev_miss);
1801 1511
1802 /* !! 1512 /**
1803 * Lockless page cache protocol: !! 1513 * find_get_entry - find and get a page cache entry
1804 * On the lookup side: <<
1805 * 1. Load the folio from i_pages <<
1806 * 2. Increment the refcount if it's not zero <<
1807 * 3. If the folio is not found by xas_reload <<
1808 * <<
1809 * On the removal side: <<
1810 * A. Freeze the page (by zeroing the refcoun <<
1811 * B. Remove the page from i_pages <<
1812 * C. Return the page to the page allocator <<
1813 * <<
1814 * This means that any page may have its refe <<
1815 * increased by a speculative page cache (or <<
1816 * be allocated by another user before the RC <<
1817 * Because the refcount temporarily acquired <<
1818 * last refcount on the page, any page alloca <<
1819 * folio_put(). <<
1820 */ <<
1821 <<
1822 /* <<
1823 * filemap_get_entry - Get a page cache entry <<
1824 * @mapping: the address_space to search 1514 * @mapping: the address_space to search
1825 * @index: The page cache index. !! 1515 * @offset: the page cache index
>> 1516 *
>> 1517 * Looks up the page cache slot at @mapping & @offset. If there is a
>> 1518 * page cache page, it is returned with an increased refcount.
1826 * 1519 *
1827 * Looks up the page cache entry at @mapping !! 1520 * If the slot holds a shadow entry of a previously evicted page, or a
1828 * it is returned with an increased refcount. !! 1521 * swap entry from shmem/tmpfs, it is returned.
1829 * of a previously evicted folio, or a swap e <<
1830 * it is returned without further action. <<
1831 * 1522 *
1832 * Return: The folio, swap or shadow entry, % !! 1523 * Return: the found page or shadow entry, %NULL if nothing is found.
1833 */ 1524 */
1834 void *filemap_get_entry(struct address_space !! 1525 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1835 { 1526 {
1836 XA_STATE(xas, &mapping->i_pages, inde !! 1527 XA_STATE(xas, &mapping->i_pages, offset);
1837 struct folio *folio; !! 1528 struct page *page;
1838 1529
1839 rcu_read_lock(); 1530 rcu_read_lock();
1840 repeat: 1531 repeat:
1841 xas_reset(&xas); 1532 xas_reset(&xas);
1842 folio = xas_load(&xas); !! 1533 page = xas_load(&xas);
1843 if (xas_retry(&xas, folio)) !! 1534 if (xas_retry(&xas, page))
1844 goto repeat; 1535 goto repeat;
1845 /* 1536 /*
1846 * A shadow entry of a recently evict 1537 * A shadow entry of a recently evicted page, or a swap entry from
1847 * shmem/tmpfs. Return it without at 1538 * shmem/tmpfs. Return it without attempting to raise page count.
1848 */ 1539 */
1849 if (!folio || xa_is_value(folio)) !! 1540 if (!page || xa_is_value(page))
1850 goto out; 1541 goto out;
1851 1542
1852 if (!folio_try_get(folio)) !! 1543 if (!page_cache_get_speculative(page))
1853 goto repeat; 1544 goto repeat;
1854 1545
1855 if (unlikely(folio != xas_reload(&xas !! 1546 /*
1856 folio_put(folio); !! 1547 * Has the page moved or been split?
>> 1548 * This is part of the lockless pagecache protocol. See
>> 1549 * include/linux/pagemap.h for details.
>> 1550 */
>> 1551 if (unlikely(page != xas_reload(&xas))) {
>> 1552 put_page(page);
1857 goto repeat; 1553 goto repeat;
1858 } 1554 }
>> 1555 page = find_subpage(page, offset);
1859 out: 1556 out:
1860 rcu_read_unlock(); 1557 rcu_read_unlock();
1861 1558
1862 return folio; !! 1559 return page;
>> 1560 }
>> 1561 EXPORT_SYMBOL(find_get_entry);
>> 1562
>> 1563 /**
>> 1564 * find_lock_entry - locate, pin and lock a page cache entry
>> 1565 * @mapping: the address_space to search
>> 1566 * @offset: the page cache index
>> 1567 *
>> 1568 * Looks up the page cache slot at @mapping & @offset. If there is a
>> 1569 * page cache page, it is returned locked and with an increased
>> 1570 * refcount.
>> 1571 *
>> 1572 * If the slot holds a shadow entry of a previously evicted page, or a
>> 1573 * swap entry from shmem/tmpfs, it is returned.
>> 1574 *
>> 1575 * find_lock_entry() may sleep.
>> 1576 *
>> 1577 * Return: the found page or shadow entry, %NULL if nothing is found.
>> 1578 */
>> 1579 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
>> 1580 {
>> 1581 struct page *page;
>> 1582
>> 1583 repeat:
>> 1584 page = find_get_entry(mapping, offset);
>> 1585 if (page && !xa_is_value(page)) {
>> 1586 lock_page(page);
>> 1587 /* Has the page been truncated? */
>> 1588 if (unlikely(page_mapping(page) != mapping)) {
>> 1589 unlock_page(page);
>> 1590 put_page(page);
>> 1591 goto repeat;
>> 1592 }
>> 1593 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
>> 1594 }
>> 1595 return page;
1863 } 1596 }
>> 1597 EXPORT_SYMBOL(find_lock_entry);
1864 1598
1865 /** 1599 /**
1866 * __filemap_get_folio - Find and get a refer !! 1600 * pagecache_get_page - find and get a page reference
1867 * @mapping: The address_space to search. !! 1601 * @mapping: the address_space to search
1868 * @index: The page index. !! 1602 * @offset: the page index
1869 * @fgp_flags: %FGP flags modify how the foli !! 1603 * @fgp_flags: PCG flags
1870 * @gfp: Memory allocation flags to use if %F !! 1604 * @gfp_mask: gfp mask to use for the page cache data page allocation
>> 1605 *
>> 1606 * Looks up the page cache slot at @mapping & @offset.
>> 1607 *
>> 1608 * PCG flags modify how the page is returned.
>> 1609 *
>> 1610 * @fgp_flags can be:
1871 * 1611 *
1872 * Looks up the page cache entry at @mapping !! 1612 * - FGP_ACCESSED: the page will be marked accessed
>> 1613 * - FGP_LOCK: Page is return locked
>> 1614 * - FGP_CREAT: If page is not present then a new page is allocated using
>> 1615 * @gfp_mask and added to the page cache and the VM's LRU
>> 1616 * list. The page is returned locked and with an increased
>> 1617 * refcount.
>> 1618 * - FGP_FOR_MMAP: Similar to FGP_CREAT, only we want to allow the caller to do
>> 1619 * its own locking dance if the page is already in cache, or unlock the page
>> 1620 * before returning if we had to add the page to pagecache.
1873 * 1621 *
1874 * If %FGP_LOCK or %FGP_CREAT are specified t !! 1622 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1875 * if the %GFP flags specified for %FGP_CREAT !! 1623 * if the GFP flags specified for FGP_CREAT are atomic.
1876 * 1624 *
1877 * If this function returns a folio, it is re !! 1625 * If there is a page cache page, it is returned with an increased refcount.
1878 * 1626 *
1879 * Return: The found folio or an ERR_PTR() ot !! 1627 * Return: the found page or %NULL otherwise.
1880 */ 1628 */
1881 struct folio *__filemap_get_folio(struct addr !! 1629 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1882 fgf_t fgp_flags, gfp_t gfp) !! 1630 int fgp_flags, gfp_t gfp_mask)
1883 { 1631 {
1884 struct folio *folio; !! 1632 struct page *page;
1885 1633
1886 repeat: 1634 repeat:
1887 folio = filemap_get_entry(mapping, in !! 1635 page = find_get_entry(mapping, offset);
1888 if (xa_is_value(folio)) !! 1636 if (xa_is_value(page))
1889 folio = NULL; !! 1637 page = NULL;
1890 if (!folio) !! 1638 if (!page)
1891 goto no_page; 1639 goto no_page;
1892 1640
1893 if (fgp_flags & FGP_LOCK) { 1641 if (fgp_flags & FGP_LOCK) {
1894 if (fgp_flags & FGP_NOWAIT) { 1642 if (fgp_flags & FGP_NOWAIT) {
1895 if (!folio_trylock(fo !! 1643 if (!trylock_page(page)) {
1896 folio_put(fol !! 1644 put_page(page);
1897 return ERR_PT !! 1645 return NULL;
1898 } 1646 }
1899 } else { 1647 } else {
1900 folio_lock(folio); !! 1648 lock_page(page);
1901 } 1649 }
1902 1650
1903 /* Has the page been truncate 1651 /* Has the page been truncated? */
1904 if (unlikely(folio->mapping ! !! 1652 if (unlikely(compound_head(page)->mapping != mapping)) {
1905 folio_unlock(folio); !! 1653 unlock_page(page);
1906 folio_put(folio); !! 1654 put_page(page);
1907 goto repeat; 1655 goto repeat;
1908 } 1656 }
1909 VM_BUG_ON_FOLIO(!folio_contai !! 1657 VM_BUG_ON_PAGE(page->index != offset, page);
1910 } 1658 }
1911 1659
1912 if (fgp_flags & FGP_ACCESSED) 1660 if (fgp_flags & FGP_ACCESSED)
1913 folio_mark_accessed(folio); !! 1661 mark_page_accessed(page);
1914 else if (fgp_flags & FGP_WRITE) { <<
1915 /* Clear idle flag for buffer <<
1916 if (folio_test_idle(folio)) <<
1917 folio_clear_idle(foli <<
1918 } <<
1919 1662
1920 if (fgp_flags & FGP_STABLE) <<
1921 folio_wait_stable(folio); <<
1922 no_page: 1663 no_page:
1923 if (!folio && (fgp_flags & FGP_CREAT) !! 1664 if (!page && (fgp_flags & FGP_CREAT)) {
1924 unsigned int min_order = mapp <<
1925 unsigned int order = max(min_ <<
1926 int err; 1665 int err;
1927 index = mapping_align_index(m !! 1666 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1928 !! 1667 gfp_mask |= __GFP_WRITE;
1929 if ((fgp_flags & FGP_WRITE) & <<
1930 gfp |= __GFP_WRITE; <<
1931 if (fgp_flags & FGP_NOFS) 1668 if (fgp_flags & FGP_NOFS)
1932 gfp &= ~__GFP_FS; !! 1669 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 1670
1940 if (order > mapping_max_folio !! 1671 page = __page_cache_alloc(gfp_mask);
1941 order = mapping_max_f !! 1672 if (!page)
1942 /* If we're not aligned, allo !! 1673 return NULL;
1943 if (index & ((1UL << order) - <<
1944 order = __ffs(index); <<
1945 1674
1946 do { !! 1675 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1947 gfp_t alloc_gfp = gfp !! 1676 fgp_flags |= FGP_LOCK;
1948 1677
1949 err = -ENOMEM; !! 1678 /* Init accessed so avoid atomic mark_page_accessed later */
1950 if (order > min_order !! 1679 if (fgp_flags & FGP_ACCESSED)
1951 alloc_gfp |= !! 1680 __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 1681
1960 err = filemap_add_fol !! 1682 err = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
1961 if (!err) !! 1683 if (unlikely(err)) {
1962 break; !! 1684 put_page(page);
1963 folio_put(folio); !! 1685 page = NULL;
1964 folio = NULL; !! 1686 if (err == -EEXIST)
1965 } while (order-- > min_order) !! 1687 goto repeat;
>> 1688 }
1966 1689
1967 if (err == -EEXIST) <<
1968 goto repeat; <<
1969 if (err) <<
1970 return ERR_PTR(err); <<
1971 /* 1690 /*
1972 * filemap_add_folio locks th !! 1691 * add_to_page_cache_lru locks the page, and for mmap we expect
1973 * we expect an unlocked page !! 1692 * an unlocked page.
1974 */ 1693 */
1975 if (folio && (fgp_flags & FGP !! 1694 if (page && (fgp_flags & FGP_FOR_MMAP))
1976 folio_unlock(folio); !! 1695 unlock_page(page);
1977 } 1696 }
1978 1697
1979 if (!folio) !! 1698 return page;
1980 return ERR_PTR(-ENOENT); <<
1981 return folio; <<
1982 } <<
1983 EXPORT_SYMBOL(__filemap_get_folio); <<
1984 <<
1985 static inline struct folio *find_get_entry(st <<
1986 xa_mark_t mark) <<
1987 { <<
1988 struct folio *folio; <<
1989 <<
1990 retry: <<
1991 if (mark == XA_PRESENT) <<
1992 folio = xas_find(xas, max); <<
1993 else <<
1994 folio = xas_find_marked(xas, <<
1995 <<
1996 if (xas_retry(xas, folio)) <<
1997 goto retry; <<
1998 /* <<
1999 * A shadow entry of a recently evict <<
2000 * entry from shmem/tmpfs or a DAX en <<
2001 * without attempting to raise page c <<
2002 */ <<
2003 if (!folio || xa_is_value(folio)) <<
2004 return folio; <<
2005 <<
2006 if (!folio_try_get(folio)) <<
2007 goto reset; <<
2008 <<
2009 if (unlikely(folio != xas_reload(xas) <<
2010 folio_put(folio); <<
2011 goto reset; <<
2012 } <<
2013 <<
2014 return folio; <<
2015 reset: <<
2016 xas_reset(xas); <<
2017 goto retry; <<
2018 } 1699 }
>> 1700 EXPORT_SYMBOL(pagecache_get_page);
2019 1701
2020 /** 1702 /**
2021 * find_get_entries - gang pagecache lookup 1703 * find_get_entries - gang pagecache lookup
2022 * @mapping: The address_space to search 1704 * @mapping: The address_space to search
2023 * @start: The starting page cache index 1705 * @start: The starting page cache index
2024 * @end: The final page index (inclusi !! 1706 * @nr_entries: The maximum number of entries
2025 * @fbatch: Where the resulting entries a !! 1707 * @entries: Where the resulting entries are placed
2026 * @indices: The cache indices correspondi 1708 * @indices: The cache indices corresponding to the entries in @entries
2027 * 1709 *
2028 * find_get_entries() will search for and ret !! 1710 * find_get_entries() will search for and return a group of up to
2029 * the mapping. The entries are placed in @f !! 1711 * @nr_entries entries in the mapping. The entries are placed at
2030 * takes a reference on any actual folios it !! 1712 * @entries. find_get_entries() takes a reference against any actual
>> 1713 * pages it returns.
>> 1714 *
>> 1715 * The search returns a group of mapping-contiguous page cache entries
>> 1716 * with ascending indexes. There may be holes in the indices due to
>> 1717 * not-present pages.
2031 * 1718 *
2032 * The entries have ascending indexes. The i !! 1719 * Any shadow entries of evicted pages, or swap entries from
2033 * due to not-present entries or large folios <<
2034 * <<
2035 * Any shadow entries of evicted folios, or s <<
2036 * shmem/tmpfs, are included in the returned 1720 * shmem/tmpfs, are included in the returned array.
2037 * 1721 *
2038 * Return: The number of entries which were f !! 1722 * Return: the number of pages and shadow entries which were found.
2039 */ 1723 */
2040 unsigned find_get_entries(struct address_spac !! 1724 unsigned find_get_entries(struct address_space *mapping,
2041 pgoff_t end, struct folio_bat !! 1725 pgoff_t start, unsigned int nr_entries,
>> 1726 struct page **entries, pgoff_t *indices)
2042 { 1727 {
2043 XA_STATE(xas, &mapping->i_pages, *sta !! 1728 XA_STATE(xas, &mapping->i_pages, start);
2044 struct folio *folio; !! 1729 struct page *page;
>> 1730 unsigned int ret = 0;
>> 1731
>> 1732 if (!nr_entries)
>> 1733 return 0;
2045 1734
2046 rcu_read_lock(); 1735 rcu_read_lock();
2047 while ((folio = find_get_entry(&xas, !! 1736 xas_for_each(&xas, page, ULONG_MAX) {
2048 indices[fbatch->nr] = xas.xa_ !! 1737 if (xas_retry(&xas, page))
2049 if (!folio_batch_add(fbatch, !! 1738 continue;
2050 break; !! 1739 /*
2051 } !! 1740 * A shadow entry of a recently evicted page, a swap
>> 1741 * entry from shmem/tmpfs or a DAX entry. Return it
>> 1742 * without attempting to raise page count.
>> 1743 */
>> 1744 if (xa_is_value(page))
>> 1745 goto export;
2052 1746
2053 if (folio_batch_count(fbatch)) { !! 1747 if (!page_cache_get_speculative(page))
2054 unsigned long nr; !! 1748 goto retry;
2055 int idx = folio_batch_count(f !! 1749
2056 !! 1750 /* Has the page moved or been split? */
2057 folio = fbatch->folios[idx]; !! 1751 if (unlikely(page != xas_reload(&xas)))
2058 if (!xa_is_value(folio)) !! 1752 goto put_page;
2059 nr = folio_nr_pages(f !! 1753 page = find_subpage(page, xas.xa_index);
2060 else !! 1754
2061 nr = 1 << xa_get_orde !! 1755 export:
2062 *start = round_down(indices[i !! 1756 indices[ret] = xas.xa_index;
>> 1757 entries[ret] = page;
>> 1758 if (++ret == nr_entries)
>> 1759 break;
>> 1760 continue;
>> 1761 put_page:
>> 1762 put_page(page);
>> 1763 retry:
>> 1764 xas_reset(&xas);
2063 } 1765 }
2064 rcu_read_unlock(); 1766 rcu_read_unlock();
2065 !! 1767 return ret;
2066 return folio_batch_count(fbatch); <<
2067 } 1768 }
2068 1769
2069 /** 1770 /**
2070 * find_lock_entries - Find a batch of pageca !! 1771 * find_get_pages_range - gang pagecache lookup
2071 * @mapping: The address_space to search. !! 1772 * @mapping: The address_space to search
2072 * @start: The starting page cache index !! 1773 * @start: The starting page index
2073 * @end: The final page index (inclusi !! 1774 * @end: The final page index (inclusive)
2074 * @fbatch: Where the resulting entries a !! 1775 * @nr_pages: The maximum number of pages
2075 * @indices: The cache indices of the entr !! 1776 * @pages: Where the resulting pages are placed
2076 * <<
2077 * find_lock_entries() will return a batch of <<
2078 * Swap, shadow and DAX entries are included. <<
2079 * locked and with an incremented refcount. <<
2080 * by somebody else or under writeback are sk <<
2081 * partially outside the range are not return <<
2082 * <<
2083 * The entries have ascending indexes. The i <<
2084 * due to not-present entries, large folios, <<
2085 * locked or folios under writeback. <<
2086 * 1777 *
2087 * Return: The number of entries which were f !! 1778 * find_get_pages_range() will search for and return a group of up to @nr_pages
2088 */ !! 1779 * pages in the mapping starting at index @start and up to index @end
2089 unsigned find_lock_entries(struct address_spa !! 1780 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
2090 pgoff_t end, struct folio_bat !! 1781 * a reference against the returned pages.
>> 1782 *
>> 1783 * The search returns a group of mapping-contiguous pages with ascending
>> 1784 * indexes. There may be holes in the indices due to not-present pages.
>> 1785 * We also update @start to index the next page for the traversal.
>> 1786 *
>> 1787 * Return: the number of pages which were found. If this number is
>> 1788 * smaller than @nr_pages, the end of specified range has been
>> 1789 * reached.
>> 1790 */
>> 1791 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
>> 1792 pgoff_t end, unsigned int nr_pages,
>> 1793 struct page **pages)
2091 { 1794 {
2092 XA_STATE(xas, &mapping->i_pages, *sta 1795 XA_STATE(xas, &mapping->i_pages, *start);
2093 struct folio *folio; !! 1796 struct page *page;
>> 1797 unsigned ret = 0;
>> 1798
>> 1799 if (unlikely(!nr_pages))
>> 1800 return 0;
2094 1801
2095 rcu_read_lock(); 1802 rcu_read_lock();
2096 while ((folio = find_get_entry(&xas, !! 1803 xas_for_each(&xas, page, end) {
2097 unsigned long base; !! 1804 if (xas_retry(&xas, page))
2098 unsigned long nr; !! 1805 continue;
2099 !! 1806 /* Skip over shadow, swap and DAX entries */
2100 if (!xa_is_value(folio)) { !! 1807 if (xa_is_value(page))
2101 nr = folio_nr_pages(f !! 1808 continue;
2102 base = folio->index; <<
2103 /* Omit large folio w <<
2104 if (base < *start) <<
2105 goto put; <<
2106 /* Omit large folio w <<
2107 if (base + nr - 1 > e <<
2108 goto put; <<
2109 if (!folio_trylock(fo <<
2110 goto put; <<
2111 if (folio->mapping != <<
2112 folio_test_writeb <<
2113 goto unlock; <<
2114 VM_BUG_ON_FOLIO(!foli <<
2115 folio <<
2116 } else { <<
2117 nr = 1 << xas_get_ord <<
2118 base = xas.xa_index & <<
2119 /* Omit order>0 value <<
2120 if (base < *start) <<
2121 continue; <<
2122 /* Omit order>0 value <<
2123 if (base + nr - 1 > e <<
2124 break; <<
2125 } <<
2126 1809
2127 /* Update start now so that l !! 1810 if (!page_cache_get_speculative(page))
2128 *start = base + nr; !! 1811 goto retry;
2129 indices[fbatch->nr] = xas.xa_ !! 1812
2130 if (!folio_batch_add(fbatch, !! 1813 /* Has the page moved or been split? */
2131 break; !! 1814 if (unlikely(page != xas_reload(&xas)))
>> 1815 goto put_page;
>> 1816
>> 1817 pages[ret] = find_subpage(page, xas.xa_index);
>> 1818 if (++ret == nr_pages) {
>> 1819 *start = xas.xa_index + 1;
>> 1820 goto out;
>> 1821 }
2132 continue; 1822 continue;
2133 unlock: !! 1823 put_page:
2134 folio_unlock(folio); !! 1824 put_page(page);
2135 put: !! 1825 retry:
2136 folio_put(folio); !! 1826 xas_reset(&xas);
2137 } 1827 }
>> 1828
>> 1829 /*
>> 1830 * We come here when there is no page beyond @end. We take care to not
>> 1831 * overflow the index @start as it confuses some of the callers. This
>> 1832 * breaks the iteration when there is a page at index -1 but that is
>> 1833 * already broken anyway.
>> 1834 */
>> 1835 if (end == (pgoff_t)-1)
>> 1836 *start = (pgoff_t)-1;
>> 1837 else
>> 1838 *start = end + 1;
>> 1839 out:
2138 rcu_read_unlock(); 1840 rcu_read_unlock();
2139 1841
2140 return folio_batch_count(fbatch); !! 1842 return ret;
2141 } 1843 }
2142 1844
2143 /** 1845 /**
2144 * filemap_get_folios - Get a batch of folios !! 1846 * find_get_pages_contig - gang contiguous pagecache lookup
2145 * @mapping: The address_space to search 1847 * @mapping: The address_space to search
2146 * @start: The starting page index !! 1848 * @index: The starting page index
2147 * @end: The final page index (inclusi !! 1849 * @nr_pages: The maximum number of pages
2148 * @fbatch: The batch to fill. !! 1850 * @pages: Where the resulting pages are placed
2149 * 1851 *
2150 * Search for and return a batch of folios in !! 1852 * find_get_pages_contig() works exactly like find_get_pages(), except
2151 * index @start and up to index @end (inclusi !! 1853 * that the returned number of pages are guaranteed to be contiguous.
2152 * in @fbatch with an elevated reference coun <<
2153 * 1854 *
2154 * Return: The number of folios which were fo !! 1855 * Return: the number of pages which were found.
2155 * We also update @start to index the next fo <<
2156 */ 1856 */
2157 unsigned filemap_get_folios(struct address_sp !! 1857 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
2158 pgoff_t end, struct folio_bat !! 1858 unsigned int nr_pages, struct page **pages)
2159 { 1859 {
2160 return filemap_get_folios_tag(mapping !! 1860 XA_STATE(xas, &mapping->i_pages, index);
2161 } !! 1861 struct page *page;
2162 EXPORT_SYMBOL(filemap_get_folios); !! 1862 unsigned int ret = 0;
2163 <<
2164 /** <<
2165 * filemap_get_folios_contig - Get a batch of <<
2166 * @mapping: The address_space to search <<
2167 * @start: The starting page index <<
2168 * @end: The final page index (inclusi <<
2169 * @fbatch: The batch to fill <<
2170 * <<
2171 * filemap_get_folios_contig() works exactly <<
2172 * except the returned folios are guaranteed <<
2173 * not return all contiguous folios if the ba <<
2174 * <<
2175 * Return: The number of folios found. <<
2176 * Also update @start to be positioned for tr <<
2177 */ <<
2178 1863
2179 unsigned filemap_get_folios_contig(struct add !! 1864 if (unlikely(!nr_pages))
2180 pgoff_t *start, pgoff_t end, !! 1865 return 0;
2181 { <<
2182 XA_STATE(xas, &mapping->i_pages, *sta <<
2183 unsigned long nr; <<
2184 struct folio *folio; <<
2185 1866
2186 rcu_read_lock(); 1867 rcu_read_lock();
2187 !! 1868 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
2188 for (folio = xas_load(&xas); folio && !! 1869 if (xas_retry(&xas, page))
2189 folio = xas_next(&xas <<
2190 if (xas_retry(&xas, folio)) <<
2191 continue; 1870 continue;
2192 /* 1871 /*
2193 * If the entry has been swap 1872 * If the entry has been swapped out, we can stop looking.
2194 * No current caller is looki 1873 * No current caller is looking for DAX entries.
2195 */ 1874 */
2196 if (xa_is_value(folio)) !! 1875 if (xa_is_value(page))
2197 goto update_start; !! 1876 break;
2198 <<
2199 /* If we landed in the middle <<
2200 if (xa_is_sibling(folio)) <<
2201 goto update_start; <<
2202 1877
2203 if (!folio_try_get(folio)) !! 1878 if (!page_cache_get_speculative(page))
2204 goto retry; 1879 goto retry;
2205 1880
2206 if (unlikely(folio != xas_rel !! 1881 /* Has the page moved or been split? */
2207 goto put_folio; !! 1882 if (unlikely(page != xas_reload(&xas)))
>> 1883 goto put_page;
2208 1884
2209 if (!folio_batch_add(fbatch, !! 1885 pages[ret] = find_subpage(page, xas.xa_index);
2210 nr = folio_nr_pages(f !! 1886 if (++ret == nr_pages)
2211 *start = folio->index !! 1887 break;
2212 goto out; <<
2213 } <<
2214 continue; 1888 continue;
2215 put_folio: !! 1889 put_page:
2216 folio_put(folio); !! 1890 put_page(page);
2217 <<
2218 retry: 1891 retry:
2219 xas_reset(&xas); 1892 xas_reset(&xas);
2220 } 1893 }
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(); 1894 rcu_read_unlock();
2231 return folio_batch_count(fbatch); !! 1895 return ret;
2232 } 1896 }
2233 EXPORT_SYMBOL(filemap_get_folios_contig); !! 1897 EXPORT_SYMBOL(find_get_pages_contig);
2234 1898
2235 /** 1899 /**
2236 * filemap_get_folios_tag - Get a batch of fo !! 1900 * find_get_pages_range_tag - find and return pages in given range matching @tag
2237 * @mapping: The address_space to search !! 1901 * @mapping: the address_space to search
2238 * @start: The starting page index !! 1902 * @index: the starting page index
2239 * @end: The final page index (inclusi !! 1903 * @end: The final page index (inclusive)
2240 * @tag: The tag index !! 1904 * @tag: the tag index
2241 * @fbatch: The batch to fill !! 1905 * @nr_pages: the maximum number of pages
2242 * !! 1906 * @pages: where the resulting pages are placed
2243 * The first folio may start before @start; i !! 1907 *
2244 * @start. The final folio may extend beyond !! 1908 * Like find_get_pages, except we only return pages which are tagged with
2245 * contain @end. The folios have ascending i !! 1909 * @tag. We update @index to index the next page for the traversal.
2246 * between the folios if there are indices wh <<
2247 * page cache. If folios are added to or rem <<
2248 * while this is running, they may or may not <<
2249 * Only returns folios that are tagged with @ <<
2250 * 1910 *
2251 * Return: The number of folios found. !! 1911 * Return: the number of pages which were found.
2252 * Also update @start to index the next folio <<
2253 */ 1912 */
2254 unsigned filemap_get_folios_tag(struct addres !! 1913 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
2255 pgoff_t end, xa_mark_ !! 1914 pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
>> 1915 struct page **pages)
2256 { 1916 {
2257 XA_STATE(xas, &mapping->i_pages, *sta !! 1917 XA_STATE(xas, &mapping->i_pages, *index);
2258 struct folio *folio; !! 1918 struct page *page;
>> 1919 unsigned ret = 0;
>> 1920
>> 1921 if (unlikely(!nr_pages))
>> 1922 return 0;
2259 1923
2260 rcu_read_lock(); 1924 rcu_read_lock();
2261 while ((folio = find_get_entry(&xas, !! 1925 xas_for_each_marked(&xas, page, end, tag) {
>> 1926 if (xas_retry(&xas, page))
>> 1927 continue;
2262 /* 1928 /*
2263 * Shadow entries should neve 1929 * Shadow entries should never be tagged, but this iteration
2264 * is lockless so there is a 1930 * is lockless so there is a window for page reclaim to evict
2265 * a page we saw tagged. Skip !! 1931 * a page we saw tagged. Skip over it.
2266 */ 1932 */
2267 if (xa_is_value(folio)) !! 1933 if (xa_is_value(page))
2268 continue; 1934 continue;
2269 if (!folio_batch_add(fbatch, !! 1935
2270 unsigned long nr = fo !! 1936 if (!page_cache_get_speculative(page))
2271 *start = folio->index !! 1937 goto retry;
>> 1938
>> 1939 /* Has the page moved or been split? */
>> 1940 if (unlikely(page != xas_reload(&xas)))
>> 1941 goto put_page;
>> 1942
>> 1943 pages[ret] = find_subpage(page, xas.xa_index);
>> 1944 if (++ret == nr_pages) {
>> 1945 *index = xas.xa_index + 1;
2272 goto out; 1946 goto out;
2273 } 1947 }
>> 1948 continue;
>> 1949 put_page:
>> 1950 put_page(page);
>> 1951 retry:
>> 1952 xas_reset(&xas);
2274 } 1953 }
>> 1954
2275 /* 1955 /*
2276 * We come here when there is no page !! 1956 * We come here when we got to @end. We take care to not overflow the
2277 * overflow the index @start as it co !! 1957 * index @index as it confuses some of the callers. This breaks the
2278 * breaks the iteration when there is !! 1958 * iteration when there is a page at index -1 but that is already
2279 * already broke anyway. !! 1959 * broken anyway.
2280 */ 1960 */
2281 if (end == (pgoff_t)-1) 1961 if (end == (pgoff_t)-1)
2282 *start = (pgoff_t)-1; !! 1962 *index = (pgoff_t)-1;
2283 else 1963 else
2284 *start = end + 1; !! 1964 *index = end + 1;
2285 out: 1965 out:
2286 rcu_read_unlock(); 1966 rcu_read_unlock();
2287 1967
2288 return folio_batch_count(fbatch); !! 1968 return ret;
2289 } 1969 }
2290 EXPORT_SYMBOL(filemap_get_folios_tag); !! 1970 EXPORT_SYMBOL(find_get_pages_range_tag);
2291 1971
2292 /* 1972 /*
2293 * CD/DVDs are error prone. When a medium err 1973 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2294 * a _large_ part of the i/o request. Imagine 1974 * a _large_ part of the i/o request. Imagine the worst scenario:
2295 * 1975 *
2296 * ---R_________________________________ 1976 * ---R__________________________________________B__________
2297 * ^ reading here 1977 * ^ reading here ^ bad block(assume 4k)
2298 * 1978 *
2299 * read(R) => miss => readahead(R...B) => med 1979 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2300 * => failing the whole request => read(R) => 1980 * => failing the whole request => read(R) => read(R+1) =>
2301 * readahead(R+1...B+1) => bang => read(R+2) 1981 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2302 * readahead(R+3...B+2) => bang => read(R+3) 1982 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2303 * readahead(R+4...B+3) => bang => read(R+4) 1983 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2304 * 1984 *
2305 * It is going insane. Fix it by quickly scal 1985 * It is going insane. Fix it by quickly scaling down the readahead size.
2306 */ 1986 */
2307 static void shrink_readahead_size_eio(struct !! 1987 static void shrink_readahead_size_eio(struct file *filp,
>> 1988 struct file_ra_state *ra)
2308 { 1989 {
2309 ra->ra_pages /= 4; 1990 ra->ra_pages /= 4;
2310 } 1991 }
2311 1992
2312 /* !! 1993 /**
2313 * filemap_get_read_batch - Get a batch of fo !! 1994 * generic_file_buffered_read - generic file read routine
>> 1995 * @iocb: the iocb to read
>> 1996 * @iter: data destination
>> 1997 * @written: already copied
>> 1998 *
>> 1999 * This is a generic file read routine, and uses the
>> 2000 * mapping->a_ops->readpage() function for the actual low-level stuff.
2314 * 2001 *
2315 * Get a batch of folios which represent a co !! 2002 * This is really ugly. But the goto's actually try to clarify some
2316 * the file. No exceptional entries will be !! 2003 * of the logic when it comes to error handling etc.
2317 * the middle of a folio, the entire folio wi !! 2004 *
2318 * folio in the batch may have the readahead !! 2005 * Return:
2319 * clear so that the caller can take the appr !! 2006 * * total number of bytes copied, including those the were already @written
>> 2007 * * negative error code if nothing was copied
2320 */ 2008 */
2321 static void filemap_get_read_batch(struct add !! 2009 static ssize_t generic_file_buffered_read(struct kiocb *iocb,
2322 pgoff_t index, pgoff_t max, s !! 2010 struct iov_iter *iter, ssize_t written)
2323 { <<
2324 XA_STATE(xas, &mapping->i_pages, inde <<
2325 struct folio *folio; <<
2326 <<
2327 rcu_read_lock(); <<
2328 for (folio = xas_load(&xas); folio; f <<
2329 if (xas_retry(&xas, folio)) <<
2330 continue; <<
2331 if (xas.xa_index > max || xa_ <<
2332 break; <<
2333 if (xa_is_sibling(folio)) <<
2334 break; <<
2335 if (!folio_try_get(folio)) <<
2336 goto retry; <<
2337 <<
2338 if (unlikely(folio != xas_rel <<
2339 goto put_folio; <<
2340 <<
2341 if (!folio_batch_add(fbatch, <<
2342 break; <<
2343 if (!folio_test_uptodate(foli <<
2344 break; <<
2345 if (folio_test_readahead(foli <<
2346 break; <<
2347 xas_advance(&xas, folio_next_ <<
2348 continue; <<
2349 put_folio: <<
2350 folio_put(folio); <<
2351 retry: <<
2352 xas_reset(&xas); <<
2353 } <<
2354 rcu_read_unlock(); <<
2355 } <<
2356 <<
2357 static int filemap_read_folio(struct file *fi <<
2358 struct folio *folio) <<
2359 { <<
2360 bool workingset = folio_test_workings <<
2361 unsigned long pflags; <<
2362 int error; <<
2363 <<
2364 /* Start the actual read. The read wi <<
2365 if (unlikely(workingset)) <<
2366 psi_memstall_enter(&pflags); <<
2367 error = filler(file, folio); <<
2368 if (unlikely(workingset)) <<
2369 psi_memstall_leave(&pflags); <<
2370 if (error) <<
2371 return error; <<
2372 <<
2373 error = folio_wait_locked_killable(fo <<
2374 if (error) <<
2375 return error; <<
2376 if (folio_test_uptodate(folio)) <<
2377 return 0; <<
2378 if (file) <<
2379 shrink_readahead_size_eio(&fi <<
2380 return -EIO; <<
2381 } <<
2382 <<
2383 static bool filemap_range_uptodate(struct add <<
2384 loff_t pos, size_t count, str <<
2385 bool need_uptodate) <<
2386 { <<
2387 if (folio_test_uptodate(folio)) <<
2388 return true; <<
2389 /* pipes can't handle partially uptod <<
2390 if (need_uptodate) <<
2391 return false; <<
2392 if (!mapping->a_ops->is_partially_upt <<
2393 return false; <<
2394 if (mapping->host->i_blkbits >= folio <<
2395 return false; <<
2396 <<
2397 if (folio_pos(folio) > pos) { <<
2398 count -= folio_pos(folio) - p <<
2399 pos = 0; <<
2400 } else { <<
2401 pos -= folio_pos(folio); <<
2402 } <<
2403 <<
2404 return mapping->a_ops->is_partially_u <<
2405 } <<
2406 <<
2407 static int filemap_update_page(struct kiocb * <<
2408 struct address_space *mapping <<
2409 struct folio *folio, bool nee <<
2410 { <<
2411 int error; <<
2412 <<
2413 if (iocb->ki_flags & IOCB_NOWAIT) { <<
2414 if (!filemap_invalidate_trylo <<
2415 return -EAGAIN; <<
2416 } else { <<
2417 filemap_invalidate_lock_share <<
2418 } <<
2419 <<
2420 if (!folio_trylock(folio)) { <<
2421 error = -EAGAIN; <<
2422 if (iocb->ki_flags & (IOCB_NO <<
2423 goto unlock_mapping; <<
2424 if (!(iocb->ki_flags & IOCB_W <<
2425 filemap_invalidate_un <<
2426 /* <<
2427 * This is where we u <<
2428 * previously submitt <<
2429 */ <<
2430 folio_put_wait_locked <<
2431 return AOP_TRUNCATED_ <<
2432 } <<
2433 error = __folio_lock_async(fo <<
2434 if (error) <<
2435 goto unlock_mapping; <<
2436 } <<
2437 <<
2438 error = AOP_TRUNCATED_PAGE; <<
2439 if (!folio->mapping) <<
2440 goto unlock; <<
2441 <<
2442 error = 0; <<
2443 if (filemap_range_uptodate(mapping, i <<
2444 need_uptod <<
2445 goto unlock; <<
2446 <<
2447 error = -EAGAIN; <<
2448 if (iocb->ki_flags & (IOCB_NOIO | IOC <<
2449 goto unlock; <<
2450 <<
2451 error = filemap_read_folio(iocb->ki_f <<
2452 folio); <<
2453 goto unlock_mapping; <<
2454 unlock: <<
2455 folio_unlock(folio); <<
2456 unlock_mapping: <<
2457 filemap_invalidate_unlock_shared(mapp <<
2458 if (error == AOP_TRUNCATED_PAGE) <<
2459 folio_put(folio); <<
2460 return error; <<
2461 } <<
2462 <<
2463 static int filemap_create_folio(struct file * <<
2464 struct address_space *mapping <<
2465 struct folio_batch *fbatch) <<
2466 { <<
2467 struct folio *folio; <<
2468 int error; <<
2469 unsigned int min_order = mapping_min_ <<
2470 pgoff_t index; <<
2471 <<
2472 folio = filemap_alloc_folio(mapping_g <<
2473 if (!folio) <<
2474 return -ENOMEM; <<
2475 <<
2476 /* <<
2477 * Protect against truncate / hole pu <<
2478 * here assures we cannot instantiate <<
2479 * pagecache folios after evicting pa <<
2480 * and before actually freeing blocks <<
2481 * release invalidate_lock after inse <<
2482 * the page cache as the locked folio <<
2483 * synchronize with hole punching. Bu <<
2484 * such as filemap_update_page() fill <<
2485 * pages or ->readahead() that need t <<
2486 * while mapping blocks for IO so let <<
2487 * well to keep locking rules simple. <<
2488 */ <<
2489 filemap_invalidate_lock_shared(mappin <<
2490 index = (pos >> (PAGE_SHIFT + min_ord <<
2491 error = filemap_add_folio(mapping, fo <<
2492 mapping_gfp_constrain <<
2493 if (error == -EEXIST) <<
2494 error = AOP_TRUNCATED_PAGE; <<
2495 if (error) <<
2496 goto error; <<
2497 <<
2498 error = filemap_read_folio(file, mapp <<
2499 if (error) <<
2500 goto error; <<
2501 <<
2502 filemap_invalidate_unlock_shared(mapp <<
2503 folio_batch_add(fbatch, folio); <<
2504 return 0; <<
2505 error: <<
2506 filemap_invalidate_unlock_shared(mapp <<
2507 folio_put(folio); <<
2508 return error; <<
2509 } <<
2510 <<
2511 static int filemap_readahead(struct kiocb *io <<
2512 struct address_space *mapping <<
2513 pgoff_t last_index) <<
2514 { <<
2515 DEFINE_READAHEAD(ractl, file, &file-> <<
2516 <<
2517 if (iocb->ki_flags & IOCB_NOIO) <<
2518 return -EAGAIN; <<
2519 page_cache_async_ra(&ractl, folio, la <<
2520 return 0; <<
2521 } <<
2522 <<
2523 static int filemap_get_pages(struct kiocb *io <<
2524 struct folio_batch *fbatch, b <<
2525 { 2011 {
2526 struct file *filp = iocb->ki_filp; 2012 struct file *filp = iocb->ki_filp;
2527 struct address_space *mapping = filp- 2013 struct address_space *mapping = filp->f_mapping;
>> 2014 struct inode *inode = mapping->host;
2528 struct file_ra_state *ra = &filp->f_r 2015 struct file_ra_state *ra = &filp->f_ra;
2529 pgoff_t index = iocb->ki_pos >> PAGE_ !! 2016 loff_t *ppos = &iocb->ki_pos;
>> 2017 pgoff_t index;
2530 pgoff_t last_index; 2018 pgoff_t last_index;
2531 struct folio *folio; !! 2019 pgoff_t prev_index;
2532 unsigned int flags; !! 2020 unsigned long offset; /* offset into pagecache page */
2533 int err = 0; !! 2021 unsigned int prev_offset;
>> 2022 int error = 0;
2534 2023
2535 /* "last_index" is the index of the p !! 2024 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
2536 last_index = DIV_ROUND_UP(iocb->ki_po <<
2537 retry: <<
2538 if (fatal_signal_pending(current)) <<
2539 return -EINTR; <<
2540 <<
2541 filemap_get_read_batch(mapping, index <<
2542 if (!folio_batch_count(fbatch)) { <<
2543 if (iocb->ki_flags & IOCB_NOI <<
2544 return -EAGAIN; <<
2545 if (iocb->ki_flags & IOCB_NOW <<
2546 flags = memalloc_noio <<
2547 page_cache_sync_readahead(map <<
2548 last_index - <<
2549 if (iocb->ki_flags & IOCB_NOW <<
2550 memalloc_noio_restore <<
2551 filemap_get_read_batch(mappin <<
2552 } <<
2553 if (!folio_batch_count(fbatch)) { <<
2554 if (iocb->ki_flags & (IOCB_NO <<
2555 return -EAGAIN; <<
2556 err = filemap_create_folio(fi <<
2557 if (err == AOP_TRUNCATED_PAGE <<
2558 goto retry; <<
2559 return err; <<
2560 } <<
2561 <<
2562 folio = fbatch->folios[folio_batch_co <<
2563 if (folio_test_readahead(folio)) { <<
2564 err = filemap_readahead(iocb, <<
2565 if (err) <<
2566 goto err; <<
2567 } <<
2568 if (!folio_test_uptodate(folio)) { <<
2569 if ((iocb->ki_flags & IOCB_WA <<
2570 folio_batch_count(fbatch) <<
2571 iocb->ki_flags |= IOC <<
2572 err = filemap_update_page(ioc <<
2573 nee <<
2574 if (err) <<
2575 goto err; <<
2576 } <<
2577 <<
2578 trace_mm_filemap_get_pages(mapping, i <<
2579 return 0; <<
2580 err: <<
2581 if (err < 0) <<
2582 folio_put(folio); <<
2583 if (likely(--fbatch->nr)) <<
2584 return 0; 2025 return 0;
2585 if (err == AOP_TRUNCATED_PAGE) !! 2026 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2586 goto retry; <<
2587 return err; <<
2588 } <<
2589 <<
2590 static inline bool pos_same_folio(loff_t pos1 <<
2591 { <<
2592 unsigned int shift = folio_shift(foli <<
2593 <<
2594 return (pos1 >> shift == pos2 >> shif <<
2595 } <<
2596 2027
2597 /** !! 2028 index = *ppos >> PAGE_SHIFT;
2598 * filemap_read - Read data from the page cac !! 2029 prev_index = ra->prev_pos >> PAGE_SHIFT;
2599 * @iocb: The iocb to read. !! 2030 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
2600 * @iter: Destination for the data. !! 2031 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
2601 * @already_read: Number of bytes already rea !! 2032 offset = *ppos & ~PAGE_MASK;
2602 * <<
2603 * Copies data from the page cache. If the d <<
2604 * uses the readahead and read_folio address_ <<
2605 * <<
2606 * Return: Total number of bytes copied, incl <<
2607 * the caller. If an error happens before an <<
2608 * a negative error number. <<
2609 */ <<
2610 ssize_t filemap_read(struct kiocb *iocb, stru <<
2611 ssize_t already_read) <<
2612 { <<
2613 struct file *filp = iocb->ki_filp; <<
2614 struct file_ra_state *ra = &filp->f_r <<
2615 struct address_space *mapping = filp- <<
2616 struct inode *inode = mapping->host; <<
2617 struct folio_batch fbatch; <<
2618 int i, error = 0; <<
2619 bool writably_mapped; <<
2620 loff_t isize, end_offset; <<
2621 loff_t last_pos = ra->prev_pos; <<
2622 2033
2623 if (unlikely(iocb->ki_pos >= inode->i !! 2034 for (;;) {
2624 return 0; !! 2035 struct page *page;
2625 if (unlikely(!iov_iter_count(iter))) !! 2036 pgoff_t end_index;
2626 return 0; !! 2037 loff_t isize;
2627 !! 2038 unsigned long nr, ret;
2628 iov_iter_truncate(iter, inode->i_sb-> <<
2629 folio_batch_init(&fbatch); <<
2630 2039
2631 do { <<
2632 cond_resched(); 2040 cond_resched();
>> 2041 find_page:
>> 2042 if (fatal_signal_pending(current)) {
>> 2043 error = -EINTR;
>> 2044 goto out;
>> 2045 }
2633 2046
2634 /* !! 2047 page = find_get_page(mapping, index);
2635 * If we've already successfu !! 2048 if (!page) {
2636 * can no longer safely retur !! 2049 if (iocb->ki_flags & IOCB_NOWAIT)
2637 * an async read NOWAIT at th !! 2050 goto would_block;
2638 */ !! 2051 page_cache_sync_readahead(mapping,
2639 if ((iocb->ki_flags & IOCB_WA !! 2052 ra, filp,
2640 iocb->ki_flags |= IOC !! 2053 index, last_index - index);
2641 !! 2054 page = find_get_page(mapping, index);
2642 if (unlikely(iocb->ki_pos >= !! 2055 if (unlikely(page == NULL))
2643 break; !! 2056 goto no_cached_page;
2644 !! 2057 }
2645 error = filemap_get_pages(ioc !! 2058 if (PageReadahead(page)) {
2646 if (error < 0) !! 2059 page_cache_async_readahead(mapping,
2647 break; !! 2060 ra, filp, page,
>> 2061 index, last_index - index);
>> 2062 }
>> 2063 if (!PageUptodate(page)) {
>> 2064 if (iocb->ki_flags & IOCB_NOWAIT) {
>> 2065 put_page(page);
>> 2066 goto would_block;
>> 2067 }
2648 2068
>> 2069 /*
>> 2070 * See comment in do_read_cache_page on why
>> 2071 * wait_on_page_locked is used to avoid unnecessarily
>> 2072 * serialisations and why it's safe.
>> 2073 */
>> 2074 error = wait_on_page_locked_killable(page);
>> 2075 if (unlikely(error))
>> 2076 goto readpage_error;
>> 2077 if (PageUptodate(page))
>> 2078 goto page_ok;
>> 2079
>> 2080 if (inode->i_blkbits == PAGE_SHIFT ||
>> 2081 !mapping->a_ops->is_partially_uptodate)
>> 2082 goto page_not_up_to_date;
>> 2083 /* pipes can't handle partially uptodate pages */
>> 2084 if (unlikely(iov_iter_is_pipe(iter)))
>> 2085 goto page_not_up_to_date;
>> 2086 if (!trylock_page(page))
>> 2087 goto page_not_up_to_date;
>> 2088 /* Did it get truncated before we got the lock? */
>> 2089 if (!page->mapping)
>> 2090 goto page_not_up_to_date_locked;
>> 2091 if (!mapping->a_ops->is_partially_uptodate(page,
>> 2092 offset, iter->count))
>> 2093 goto page_not_up_to_date_locked;
>> 2094 unlock_page(page);
>> 2095 }
>> 2096 page_ok:
2649 /* 2097 /*
2650 * i_size must be checked aft !! 2098 * i_size must be checked after we know the page is Uptodate.
2651 * 2099 *
2652 * Checking i_size after the 2100 * Checking i_size after the check allows us to calculate
2653 * the correct value for "nr" 2101 * the correct value for "nr", which means the zero-filled
2654 * part of the page is not co 2102 * part of the page is not copied back to userspace (unless
2655 * another truncate extends t 2103 * another truncate extends the file - this is desired though).
2656 */ 2104 */
>> 2105
2657 isize = i_size_read(inode); 2106 isize = i_size_read(inode);
2658 if (unlikely(iocb->ki_pos >= !! 2107 end_index = (isize - 1) >> PAGE_SHIFT;
2659 goto put_folios; !! 2108 if (unlikely(!isize || index > end_index)) {
2660 end_offset = min_t(loff_t, is !! 2109 put_page(page);
>> 2110 goto out;
>> 2111 }
>> 2112
>> 2113 /* nr is the maximum number of bytes to copy from this page */
>> 2114 nr = PAGE_SIZE;
>> 2115 if (index == end_index) {
>> 2116 nr = ((isize - 1) & ~PAGE_MASK) + 1;
>> 2117 if (nr <= offset) {
>> 2118 put_page(page);
>> 2119 goto out;
>> 2120 }
>> 2121 }
>> 2122 nr = nr - offset;
>> 2123
>> 2124 /* If users can be writing to this page using arbitrary
>> 2125 * virtual addresses, take care about potential aliasing
>> 2126 * before reading the page on the kernel side.
>> 2127 */
>> 2128 if (mapping_writably_mapped(mapping))
>> 2129 flush_dcache_page(page);
2661 2130
2662 /* 2131 /*
2663 * Once we start copying data !! 2132 * When a sequential read accesses a page several times,
2664 * cachelines that might be c !! 2133 * only mark it as accessed the first time.
2665 */ 2134 */
2666 writably_mapped = mapping_wri !! 2135 if (prev_index != index || offset != prev_offset)
>> 2136 mark_page_accessed(page);
>> 2137 prev_index = index;
2667 2138
2668 /* 2139 /*
2669 * When a read accesses the s !! 2140 * Ok, we have the page, and it's up-to-date, so
2670 * mark it as accessed the fi !! 2141 * now we can copy it to user space...
2671 */ 2142 */
2672 if (!pos_same_folio(iocb->ki_ <<
2673 fbatch.fo <<
2674 folio_mark_accessed(f <<
2675 <<
2676 for (i = 0; i < folio_batch_c <<
2677 struct folio *folio = <<
2678 size_t fsize = folio_ <<
2679 size_t offset = iocb- <<
2680 size_t bytes = min_t( <<
2681 <<
2682 size_t copied; <<
2683 2143
2684 if (end_offset < foli !! 2144 ret = copy_page_to_iter(page, offset, nr, iter);
2685 break; !! 2145 offset += ret;
2686 if (i > 0) !! 2146 index += offset >> PAGE_SHIFT;
2687 folio_mark_ac !! 2147 offset &= ~PAGE_MASK;
2688 /* !! 2148 prev_offset = offset;
2689 * If users can be wr <<
2690 * virtual addresses, <<
2691 * before reading the <<
2692 */ <<
2693 if (writably_mapped) <<
2694 flush_dcache_ <<
2695 2149
2696 copied = copy_folio_t !! 2150 put_page(page);
>> 2151 written += ret;
>> 2152 if (!iov_iter_count(iter))
>> 2153 goto out;
>> 2154 if (ret < nr) {
>> 2155 error = -EFAULT;
>> 2156 goto out;
>> 2157 }
>> 2158 continue;
2697 2159
2698 already_read += copie !! 2160 page_not_up_to_date:
2699 iocb->ki_pos += copie !! 2161 /* Get exclusive access to the page ... */
2700 last_pos = iocb->ki_p !! 2162 error = lock_page_killable(page);
>> 2163 if (unlikely(error))
>> 2164 goto readpage_error;
>> 2165
>> 2166 page_not_up_to_date_locked:
>> 2167 /* Did it get truncated before we got the lock? */
>> 2168 if (!page->mapping) {
>> 2169 unlock_page(page);
>> 2170 put_page(page);
>> 2171 continue;
>> 2172 }
2701 2173
2702 if (copied < bytes) { !! 2174 /* Did somebody else fill it already? */
2703 error = -EFAU !! 2175 if (PageUptodate(page)) {
2704 break; !! 2176 unlock_page(page);
>> 2177 goto page_ok;
>> 2178 }
>> 2179
>> 2180 readpage:
>> 2181 /*
>> 2182 * A previous I/O error may have been due to temporary
>> 2183 * failures, eg. multipath errors.
>> 2184 * PG_error will be set again if readpage fails.
>> 2185 */
>> 2186 ClearPageError(page);
>> 2187 /* Start the actual read. The read will unlock the page. */
>> 2188 error = mapping->a_ops->readpage(filp, page);
>> 2189
>> 2190 if (unlikely(error)) {
>> 2191 if (error == AOP_TRUNCATED_PAGE) {
>> 2192 put_page(page);
>> 2193 error = 0;
>> 2194 goto find_page;
2705 } 2195 }
>> 2196 goto readpage_error;
2706 } 2197 }
2707 put_folios: <<
2708 for (i = 0; i < folio_batch_c <<
2709 folio_put(fbatch.foli <<
2710 folio_batch_init(&fbatch); <<
2711 } while (iov_iter_count(iter) && iocb <<
2712 2198
2713 file_accessed(filp); !! 2199 if (!PageUptodate(page)) {
2714 ra->prev_pos = last_pos; !! 2200 error = lock_page_killable(page);
2715 return already_read ? already_read : !! 2201 if (unlikely(error))
2716 } !! 2202 goto readpage_error;
2717 EXPORT_SYMBOL_GPL(filemap_read); !! 2203 if (!PageUptodate(page)) {
2718 !! 2204 if (page->mapping == NULL) {
2719 int kiocb_write_and_wait(struct kiocb *iocb, !! 2205 /*
2720 { !! 2206 * invalidate_mapping_pages got it
2721 struct address_space *mapping = iocb- !! 2207 */
2722 loff_t pos = iocb->ki_pos; !! 2208 unlock_page(page);
2723 loff_t end = pos + count - 1; !! 2209 put_page(page);
2724 !! 2210 goto find_page;
2725 if (iocb->ki_flags & IOCB_NOWAIT) { !! 2211 }
2726 if (filemap_range_needs_write !! 2212 unlock_page(page);
2727 return -EAGAIN; !! 2213 shrink_readahead_size_eio(filp, ra);
2728 return 0; !! 2214 error = -EIO;
2729 } !! 2215 goto readpage_error;
>> 2216 }
>> 2217 unlock_page(page);
>> 2218 }
2730 2219
2731 return filemap_write_and_wait_range(m !! 2220 goto page_ok;
2732 } <<
2733 EXPORT_SYMBOL_GPL(kiocb_write_and_wait); <<
2734 2221
2735 int filemap_invalidate_pages(struct address_s !! 2222 readpage_error:
2736 loff_t pos, loff !! 2223 /* UHHUH! A synchronous read error occurred. Report it */
2737 { !! 2224 put_page(page);
2738 int ret; !! 2225 goto out;
2739 2226
2740 if (nowait) { !! 2227 no_cached_page:
2741 /* we could block if there ar !! 2228 /*
2742 if (filemap_range_has_page(ma !! 2229 * Ok, it wasn't cached, so we need to create a new
2743 return -EAGAIN; !! 2230 * page..
2744 } else { !! 2231 */
2745 ret = filemap_write_and_wait_ !! 2232 page = page_cache_alloc(mapping);
2746 if (ret) !! 2233 if (!page) {
2747 return ret; !! 2234 error = -ENOMEM;
>> 2235 goto out;
>> 2236 }
>> 2237 error = add_to_page_cache_lru(page, mapping, index,
>> 2238 mapping_gfp_constraint(mapping, GFP_KERNEL));
>> 2239 if (error) {
>> 2240 put_page(page);
>> 2241 if (error == -EEXIST) {
>> 2242 error = 0;
>> 2243 goto find_page;
>> 2244 }
>> 2245 goto out;
>> 2246 }
>> 2247 goto readpage;
2748 } 2248 }
2749 2249
2750 /* !! 2250 would_block:
2751 * After a write we want buffered rea !! 2251 error = -EAGAIN;
2752 * the new data. We invalidate clean !! 2252 out:
2753 * about to write. We do this *befor !! 2253 ra->prev_pos = prev_index;
2754 * without clobbering -EIOCBQUEUED fr !! 2254 ra->prev_pos <<= PAGE_SHIFT;
2755 */ !! 2255 ra->prev_pos |= prev_offset;
2756 return invalidate_inode_pages2_range( <<
2757 <<
2758 } <<
2759 <<
2760 int kiocb_invalidate_pages(struct kiocb *iocb <<
2761 { <<
2762 struct address_space *mapping = iocb- <<
2763 2256
2764 return filemap_invalidate_pages(mappi !! 2257 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2765 iocb- !! 2258 file_accessed(filp);
2766 iocb- !! 2259 return written ? written : error;
2767 } 2260 }
2768 EXPORT_SYMBOL_GPL(kiocb_invalidate_pages); <<
2769 2261
2770 /** 2262 /**
2771 * generic_file_read_iter - generic filesyste 2263 * generic_file_read_iter - generic filesystem read routine
2772 * @iocb: kernel I/O control block 2264 * @iocb: kernel I/O control block
2773 * @iter: destination for the data read 2265 * @iter: destination for the data read
2774 * 2266 *
2775 * This is the "read_iter()" routine for all 2267 * This is the "read_iter()" routine for all filesystems
2776 * that can use the page cache directly. 2268 * that can use the page cache directly.
2777 * <<
2778 * The IOCB_NOWAIT flag in iocb->ki_flags ind <<
2779 * be returned when no data can be read witho <<
2780 * to complete; it doesn't prevent readahead. <<
2781 * <<
2782 * The IOCB_NOIO flag in iocb->ki_flags indic <<
2783 * requests shall be made for the read or for <<
2784 * can be read, -EAGAIN shall be returned. W <<
2785 * triggered, a partial, possibly empty read <<
2786 * <<
2787 * Return: 2269 * Return:
2788 * * number of bytes copied, even for partial 2270 * * number of bytes copied, even for partial reads
2789 * * negative error code (or 0 if IOCB_NOIO) !! 2271 * * negative error code if nothing was read
2790 */ 2272 */
2791 ssize_t 2273 ssize_t
2792 generic_file_read_iter(struct kiocb *iocb, st 2274 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2793 { 2275 {
2794 size_t count = iov_iter_count(iter); 2276 size_t count = iov_iter_count(iter);
2795 ssize_t retval = 0; 2277 ssize_t retval = 0;
2796 2278
2797 if (!count) 2279 if (!count)
2798 return 0; /* skip atime */ !! 2280 goto out; /* skip atime */
2799 2281
2800 if (iocb->ki_flags & IOCB_DIRECT) { 2282 if (iocb->ki_flags & IOCB_DIRECT) {
2801 struct file *file = iocb->ki_ 2283 struct file *file = iocb->ki_filp;
2802 struct address_space *mapping 2284 struct address_space *mapping = file->f_mapping;
2803 struct inode *inode = mapping 2285 struct inode *inode = mapping->host;
>> 2286 loff_t size;
>> 2287
>> 2288 size = i_size_read(inode);
>> 2289 if (iocb->ki_flags & IOCB_NOWAIT) {
>> 2290 if (filemap_range_has_page(mapping, iocb->ki_pos,
>> 2291 iocb->ki_pos + count - 1))
>> 2292 return -EAGAIN;
>> 2293 } else {
>> 2294 retval = filemap_write_and_wait_range(mapping,
>> 2295 iocb->ki_pos,
>> 2296 iocb->ki_pos + count - 1);
>> 2297 if (retval < 0)
>> 2298 goto out;
>> 2299 }
2804 2300
2805 retval = kiocb_write_and_wait <<
2806 if (retval < 0) <<
2807 return retval; <<
2808 file_accessed(file); 2301 file_accessed(file);
2809 2302
2810 retval = mapping->a_ops->dire 2303 retval = mapping->a_ops->direct_IO(iocb, iter);
2811 if (retval >= 0) { 2304 if (retval >= 0) {
2812 iocb->ki_pos += retva 2305 iocb->ki_pos += retval;
2813 count -= retval; 2306 count -= retval;
2814 } 2307 }
2815 if (retval != -EIOCBQUEUED) !! 2308 iov_iter_revert(iter, count - iov_iter_count(iter));
2816 iov_iter_revert(iter, <<
2817 2309
2818 /* 2310 /*
2819 * Btrfs can have a short DIO 2311 * Btrfs can have a short DIO read if we encounter
2820 * compressed extents, so if 2312 * compressed extents, so if there was an error, or if
2821 * we've already read everyth 2313 * we've already read everything we wanted to, or if
2822 * there was a short read bec 2314 * there was a short read because we hit EOF, go ahead
2823 * and return. Otherwise fal 2315 * and return. Otherwise fallthrough to buffered io for
2824 * the rest of the read. Buf 2316 * the rest of the read. Buffered reads will not work for
2825 * DAX files, so don't bother 2317 * DAX files, so don't bother trying.
2826 */ 2318 */
2827 if (retval < 0 || !count || I !! 2319 if (retval < 0 || !count || iocb->ki_pos >= size ||
2828 return retval; !! 2320 IS_DAX(inode))
2829 if (iocb->ki_pos >= i_size_re !! 2321 goto out;
2830 return retval; <<
2831 } <<
2832 <<
2833 return filemap_read(iocb, iter, retva <<
2834 } <<
2835 EXPORT_SYMBOL(generic_file_read_iter); <<
2836 <<
2837 /* <<
2838 * Splice subpages from a folio into a pipe. <<
2839 */ <<
2840 size_t splice_folio_into_pipe(struct pipe_ino <<
2841 struct folio *f <<
2842 { <<
2843 struct page *page; <<
2844 size_t spliced = 0, offset = offset_i <<
2845 <<
2846 page = folio_page(folio, offset / PAG <<
2847 size = min(size, folio_size(folio) - <<
2848 offset %= PAGE_SIZE; <<
2849 <<
2850 while (spliced < size && <<
2851 !pipe_full(pipe->head, pipe->t <<
2852 struct pipe_buffer *buf = pip <<
2853 size_t part = min_t(size_t, P <<
2854 <<
2855 *buf = (struct pipe_buffer) { <<
2856 .ops = &page_cache <<
2857 .page = page, <<
2858 .offset = offset, <<
2859 .len = part, <<
2860 }; <<
2861 folio_get(folio); <<
2862 pipe->head++; <<
2863 page++; <<
2864 spliced += part; <<
2865 offset = 0; <<
2866 } 2322 }
2867 2323
2868 return spliced; !! 2324 retval = generic_file_buffered_read(iocb, iter, retval);
2869 } <<
2870 <<
2871 /** <<
2872 * filemap_splice_read - Splice data from a <<
2873 * @in: The file to read from <<
2874 * @ppos: Pointer to the file position to rea <<
2875 * @pipe: The pipe to splice into <<
2876 * @len: The amount to splice <<
2877 * @flags: The SPLICE_F_* flags <<
2878 * <<
2879 * This function gets folios from a file's pa <<
2880 * pipe. Readahead will be called as necessa <<
2881 * be used for blockdevs also. <<
2882 * <<
2883 * Return: On success, the number of bytes re <<
2884 * will be updated if appropriate; 0 will be <<
2885 * to be read; -EAGAIN will be returned if th <<
2886 * other negative error code will be returned <<
2887 * if the pipe has insufficient space, we rea <<
2888 * hole. <<
2889 */ <<
2890 ssize_t filemap_splice_read(struct file *in, <<
2891 struct pipe_inode <<
2892 size_t len, unsig <<
2893 { <<
2894 struct folio_batch fbatch; <<
2895 struct kiocb iocb; <<
2896 size_t total_spliced = 0, used, npage <<
2897 loff_t isize, end_offset; <<
2898 bool writably_mapped; <<
2899 int i, error = 0; <<
2900 <<
2901 if (unlikely(*ppos >= in->f_mapping-> <<
2902 return 0; <<
2903 <<
2904 init_sync_kiocb(&iocb, in); <<
2905 iocb.ki_pos = *ppos; <<
2906 <<
2907 /* Work out how much data we can actu <<
2908 used = pipe_occupancy(pipe->head, pip <<
2909 npages = max_t(ssize_t, pipe->max_usa <<
2910 len = min_t(size_t, len, npages * PAG <<
2911 <<
2912 folio_batch_init(&fbatch); <<
2913 <<
2914 do { <<
2915 cond_resched(); <<
2916 <<
2917 if (*ppos >= i_size_read(in-> <<
2918 break; <<
2919 <<
2920 iocb.ki_pos = *ppos; <<
2921 error = filemap_get_pages(&io <<
2922 if (error < 0) <<
2923 break; <<
2924 <<
2925 /* <<
2926 * i_size must be checked aft <<
2927 * <<
2928 * Checking i_size after the <<
2929 * the correct value for "nr" <<
2930 * part of the page is not co <<
2931 * another truncate extends t <<
2932 */ <<
2933 isize = i_size_read(in->f_map <<
2934 if (unlikely(*ppos >= isize)) <<
2935 break; <<
2936 end_offset = min_t(loff_t, is <<
2937 <<
2938 /* <<
2939 * Once we start copying data <<
2940 * cachelines that might be c <<
2941 */ <<
2942 writably_mapped = mapping_wri <<
2943 <<
2944 for (i = 0; i < folio_batch_c <<
2945 struct folio *folio = <<
2946 size_t n; <<
2947 <<
2948 if (folio_pos(folio) <<
2949 goto out; <<
2950 folio_mark_accessed(f <<
2951 <<
2952 /* <<
2953 * If users can be wr <<
2954 * virtual addresses, <<
2955 * before reading the <<
2956 */ <<
2957 if (writably_mapped) <<
2958 flush_dcache_ <<
2959 <<
2960 n = min_t(loff_t, len <<
2961 n = splice_folio_into <<
2962 if (!n) <<
2963 goto out; <<
2964 len -= n; <<
2965 total_spliced += n; <<
2966 *ppos += n; <<
2967 in->f_ra.prev_pos = * <<
2968 if (pipe_full(pipe->h <<
2969 goto out; <<
2970 } <<
2971 <<
2972 folio_batch_release(&fbatch); <<
2973 } while (len); <<
2974 <<
2975 out: 2325 out:
2976 folio_batch_release(&fbatch); !! 2326 return retval;
2977 file_accessed(in); <<
2978 <<
2979 return total_spliced ? total_spliced <<
2980 } <<
2981 EXPORT_SYMBOL(filemap_splice_read); <<
2982 <<
2983 static inline loff_t folio_seek_hole_data(str <<
2984 struct address_space *mapping <<
2985 loff_t start, loff_t end, boo <<
2986 { <<
2987 const struct address_space_operations <<
2988 size_t offset, bsz = i_blocksize(mapp <<
2989 <<
2990 if (xa_is_value(folio) || folio_test_ <<
2991 return seek_data ? start : en <<
2992 if (!ops->is_partially_uptodate) <<
2993 return seek_data ? end : star <<
2994 <<
2995 xas_pause(xas); <<
2996 rcu_read_unlock(); <<
2997 folio_lock(folio); <<
2998 if (unlikely(folio->mapping != mappin <<
2999 goto unlock; <<
3000 <<
3001 offset = offset_in_folio(folio, start <<
3002 <<
3003 do { <<
3004 if (ops->is_partially_uptodat <<
3005 <<
3006 break; <<
3007 start = (start + bsz) & ~(bsz <<
3008 offset += bsz; <<
3009 } while (offset < folio_size(folio)); <<
3010 unlock: <<
3011 folio_unlock(folio); <<
3012 rcu_read_lock(); <<
3013 return start; <<
3014 } <<
3015 <<
3016 static inline size_t seek_folio_size(struct x <<
3017 { <<
3018 if (xa_is_value(folio)) <<
3019 return PAGE_SIZE << xas_get_o <<
3020 return folio_size(folio); <<
3021 } <<
3022 <<
3023 /** <<
3024 * mapping_seek_hole_data - Seek for SEEK_DAT <<
3025 * @mapping: Address space to search. <<
3026 * @start: First byte to consider. <<
3027 * @end: Limit of search (exclusive). <<
3028 * @whence: Either SEEK_HOLE or SEEK_DATA. <<
3029 * <<
3030 * If the page cache knows which blocks conta <<
3031 * contain data, your filesystem can use this <<
3032 * SEEK_HOLE and SEEK_DATA. This is useful f <<
3033 * entirely memory-based such as tmpfs, and f <<
3034 * unwritten extents. <<
3035 * <<
3036 * Return: The requested offset on success, o <<
3037 * SEEK_DATA and there is no data after @star <<
3038 * after @end - 1, so SEEK_HOLE returns @end <<
3039 * and @end contain data. <<
3040 */ <<
3041 loff_t mapping_seek_hole_data(struct address_ <<
3042 loff_t end, int whence) <<
3043 { <<
3044 XA_STATE(xas, &mapping->i_pages, star <<
3045 pgoff_t max = (end - 1) >> PAGE_SHIFT <<
3046 bool seek_data = (whence == SEEK_DATA <<
3047 struct folio *folio; <<
3048 <<
3049 if (end <= start) <<
3050 return -ENXIO; <<
3051 <<
3052 rcu_read_lock(); <<
3053 while ((folio = find_get_entry(&xas, <<
3054 loff_t pos = (u64)xas.xa_inde <<
3055 size_t seek_size; <<
3056 <<
3057 if (start < pos) { <<
3058 if (!seek_data) <<
3059 goto unlock; <<
3060 start = pos; <<
3061 } <<
3062 <<
3063 seek_size = seek_folio_size(& <<
3064 pos = round_up((u64)pos + 1, <<
3065 start = folio_seek_hole_data( <<
3066 seek_data); <<
3067 if (start < pos) <<
3068 goto unlock; <<
3069 if (start >= end) <<
3070 break; <<
3071 if (seek_size > PAGE_SIZE) <<
3072 xas_set(&xas, pos >> <<
3073 if (!xa_is_value(folio)) <<
3074 folio_put(folio); <<
3075 } <<
3076 if (seek_data) <<
3077 start = -ENXIO; <<
3078 unlock: <<
3079 rcu_read_unlock(); <<
3080 if (folio && !xa_is_value(folio)) <<
3081 folio_put(folio); <<
3082 if (start > end) <<
3083 return end; <<
3084 return start; <<
3085 } 2327 }
>> 2328 EXPORT_SYMBOL(generic_file_read_iter);
3086 2329
3087 #ifdef CONFIG_MMU 2330 #ifdef CONFIG_MMU
3088 #define MMAP_LOTSAMISS (100) 2331 #define MMAP_LOTSAMISS (100)
3089 /* 2332 /*
3090 * lock_folio_maybe_drop_mmap - lock the page !! 2333 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_sem
3091 * @vmf - the vm_fault for this fault. 2334 * @vmf - the vm_fault for this fault.
3092 * @folio - the folio to lock. !! 2335 * @page - the page to lock.
3093 * @fpin - the pointer to the file we may pin 2336 * @fpin - the pointer to the file we may pin (or is already pinned).
3094 * 2337 *
3095 * This works similar to lock_folio_or_retry !! 2338 * This works similar to lock_page_or_retry in that it can drop the mmap_sem.
3096 * mmap_lock. It differs in that it actually !! 2339 * 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 !! 2340 * if it couldn't lock the page. If we did have to drop the mmap_sem then fpin
3098 * to drop the mmap_lock then fpin will point !! 2341 * 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 */ 2342 */
3101 static int lock_folio_maybe_drop_mmap(struct !! 2343 static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page,
3102 struct f 2344 struct file **fpin)
3103 { 2345 {
3104 if (folio_trylock(folio)) !! 2346 if (trylock_page(page))
3105 return 1; 2347 return 1;
3106 2348
3107 /* 2349 /*
3108 * NOTE! This will make us return wit 2350 * NOTE! This will make us return with VM_FAULT_RETRY, but with
3109 * the fault lock still held. That's !! 2351 * the mmap_sem still held. That's how FAULT_FLAG_RETRY_NOWAIT
3110 * is supposed to work. We have way t 2352 * is supposed to work. We have way too many special cases..
3111 */ 2353 */
3112 if (vmf->flags & FAULT_FLAG_RETRY_NOW 2354 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
3113 return 0; 2355 return 0;
3114 2356
3115 *fpin = maybe_unlock_mmap_for_io(vmf, 2357 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
3116 if (vmf->flags & FAULT_FLAG_KILLABLE) 2358 if (vmf->flags & FAULT_FLAG_KILLABLE) {
3117 if (__folio_lock_killable(fol !! 2359 if (__lock_page_killable(page)) {
3118 /* 2360 /*
3119 * We didn't have the !! 2361 * We didn't have the right flags to drop the mmap_sem,
3120 * fault lock, but al !! 2362 * but all fault_handlers only check for fatal signals
3121 * for fatal signals !! 2363 * if we return VM_FAULT_RETRY, so we need to drop the
3122 * so we need to drop !! 2364 * mmap_sem here and return 0 if we don't have a fpin.
3123 * return 0 if we don <<
3124 */ 2365 */
3125 if (*fpin == NULL) 2366 if (*fpin == NULL)
3126 release_fault !! 2367 up_read(&vmf->vma->vm_mm->mmap_sem);
3127 return 0; 2368 return 0;
3128 } 2369 }
3129 } else 2370 } else
3130 __folio_lock(folio); !! 2371 __lock_page(page);
3131 <<
3132 return 1; 2372 return 1;
3133 } 2373 }
3134 2374
>> 2375
3135 /* 2376 /*
3136 * Synchronous readahead happens when we don' 2377 * 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 2378 * 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 2379 * 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 2380 * 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 2381 * returned needs to be fput()'ed when we're done with it.
3141 */ 2382 */
3142 static struct file *do_sync_mmap_readahead(st 2383 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
3143 { 2384 {
3144 struct file *file = vmf->vma->vm_file 2385 struct file *file = vmf->vma->vm_file;
3145 struct file_ra_state *ra = &file->f_r 2386 struct file_ra_state *ra = &file->f_ra;
3146 struct address_space *mapping = file- 2387 struct address_space *mapping = file->f_mapping;
3147 DEFINE_READAHEAD(ractl, file, ra, map <<
3148 struct file *fpin = NULL; 2388 struct file *fpin = NULL;
3149 unsigned long vm_flags = vmf->vma->vm !! 2389 pgoff_t offset = vmf->pgoff;
3150 unsigned int mmap_miss; <<
3151 <<
3152 #ifdef CONFIG_TRANSPARENT_HUGEPAGE <<
3153 /* Use the readahead code, even if re <<
3154 if ((vm_flags & VM_HUGEPAGE) && HPAGE <<
3155 fpin = maybe_unlock_mmap_for_ <<
3156 ractl._index &= ~((unsigned l <<
3157 ra->size = HPAGE_PMD_NR; <<
3158 /* <<
3159 * Fetch two PMD folios, so w <<
3160 * readahead, unless we've be <<
3161 */ <<
3162 if (!(vm_flags & VM_RAND_READ <<
3163 ra->size *= 2; <<
3164 ra->async_size = HPAGE_PMD_NR <<
3165 page_cache_ra_order(&ractl, r <<
3166 return fpin; <<
3167 } <<
3168 #endif <<
3169 2390
3170 /* If we don't want any read-ahead, d 2391 /* If we don't want any read-ahead, don't bother */
3171 if (vm_flags & VM_RAND_READ) !! 2392 if (vmf->vma->vm_flags & VM_RAND_READ)
3172 return fpin; 2393 return fpin;
3173 if (!ra->ra_pages) 2394 if (!ra->ra_pages)
3174 return fpin; 2395 return fpin;
3175 2396
3176 if (vm_flags & VM_SEQ_READ) { !! 2397 if (vmf->vma->vm_flags & VM_SEQ_READ) {
3177 fpin = maybe_unlock_mmap_for_ 2398 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3178 page_cache_sync_ra(&ractl, ra !! 2399 page_cache_sync_readahead(mapping, ra, file, offset,
>> 2400 ra->ra_pages);
3179 return fpin; 2401 return fpin;
3180 } 2402 }
3181 2403
3182 /* Avoid banging the cache line if no 2404 /* Avoid banging the cache line if not needed */
3183 mmap_miss = READ_ONCE(ra->mmap_miss); !! 2405 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
3184 if (mmap_miss < MMAP_LOTSAMISS * 10) !! 2406 ra->mmap_miss++;
3185 WRITE_ONCE(ra->mmap_miss, ++m <<
3186 2407
3187 /* 2408 /*
3188 * Do we miss much more than hit in t 2409 * Do we miss much more than hit in this file? If so,
3189 * stop bothering with read-ahead. It 2410 * stop bothering with read-ahead. It will only hurt.
3190 */ 2411 */
3191 if (mmap_miss > MMAP_LOTSAMISS) !! 2412 if (ra->mmap_miss > MMAP_LOTSAMISS)
3192 return fpin; 2413 return fpin;
3193 2414
3194 /* 2415 /*
3195 * mmap read-around 2416 * mmap read-around
3196 */ 2417 */
3197 fpin = maybe_unlock_mmap_for_io(vmf, 2418 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3198 ra->start = max_t(long, 0, vmf->pgoff !! 2419 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
3199 ra->size = ra->ra_pages; 2420 ra->size = ra->ra_pages;
3200 ra->async_size = ra->ra_pages / 4; 2421 ra->async_size = ra->ra_pages / 4;
3201 ractl._index = ra->start; !! 2422 ra_submit(ra, mapping, file);
3202 page_cache_ra_order(&ractl, ra, 0); <<
3203 return fpin; 2423 return fpin;
3204 } 2424 }
3205 2425
3206 /* 2426 /*
3207 * Asynchronous readahead happens when we fin 2427 * Asynchronous readahead happens when we find the page and PG_readahead,
3208 * so we want to possibly extend the readahea 2428 * 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 !! 2429 * was pinned if we have to drop the mmap_sem in order to do IO.
3210 */ 2430 */
3211 static struct file *do_async_mmap_readahead(s 2431 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
3212 s !! 2432 struct page *page)
3213 { 2433 {
3214 struct file *file = vmf->vma->vm_file 2434 struct file *file = vmf->vma->vm_file;
3215 struct file_ra_state *ra = &file->f_r 2435 struct file_ra_state *ra = &file->f_ra;
3216 DEFINE_READAHEAD(ractl, file, ra, fil !! 2436 struct address_space *mapping = file->f_mapping;
3217 struct file *fpin = NULL; 2437 struct file *fpin = NULL;
3218 unsigned int mmap_miss; !! 2438 pgoff_t offset = vmf->pgoff;
3219 2439
3220 /* If we don't want any read-ahead, d 2440 /* If we don't want any read-ahead, don't bother */
3221 if (vmf->vma->vm_flags & VM_RAND_READ !! 2441 if (vmf->vma->vm_flags & VM_RAND_READ)
3222 return fpin; 2442 return fpin;
3223 !! 2443 if (ra->mmap_miss > 0)
3224 mmap_miss = READ_ONCE(ra->mmap_miss); !! 2444 ra->mmap_miss--;
3225 if (mmap_miss) !! 2445 if (PageReadahead(page)) {
3226 WRITE_ONCE(ra->mmap_miss, --m <<
3227 <<
3228 if (folio_test_readahead(folio)) { <<
3229 fpin = maybe_unlock_mmap_for_ 2446 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3230 page_cache_async_ra(&ractl, f !! 2447 page_cache_async_readahead(mapping, ra, file,
>> 2448 page, offset, ra->ra_pages);
3231 } 2449 }
3232 return fpin; 2450 return fpin;
3233 } 2451 }
3234 2452
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 /** 2453 /**
3279 * filemap_fault - read in file data for page 2454 * filemap_fault - read in file data for page fault handling
3280 * @vmf: struct vm_fault containing de 2455 * @vmf: struct vm_fault containing details of the fault
3281 * 2456 *
3282 * filemap_fault() is invoked via the vma ope 2457 * filemap_fault() is invoked via the vma operations vector for a
3283 * mapped memory region to read in file data 2458 * mapped memory region to read in file data during a page fault.
3284 * 2459 *
3285 * The goto's are kind of ugly, but this stre 2460 * 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 2461 * it in the page cache, and handles the special cases reasonably without
3287 * having a lot of duplicated code. 2462 * having a lot of duplicated code.
3288 * 2463 *
3289 * vma->vm_mm->mmap_lock must be held on entr !! 2464 * vma->vm_mm->mmap_sem must be held on entry.
3290 * 2465 *
3291 * If our return value has VM_FAULT_RETRY set !! 2466 * If our return value has VM_FAULT_RETRY set, it's because the mmap_sem
3292 * may be dropped before doing I/O or by lock !! 2467 * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
3293 * 2468 *
3294 * If our return value does not have VM_FAULT !! 2469 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
3295 * has not been released. 2470 * has not been released.
3296 * 2471 *
3297 * We never return with VM_FAULT_RETRY and a 2472 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3298 * 2473 *
3299 * Return: bitwise-OR of %VM_FAULT_ codes. 2474 * Return: bitwise-OR of %VM_FAULT_ codes.
3300 */ 2475 */
3301 vm_fault_t filemap_fault(struct vm_fault *vmf 2476 vm_fault_t filemap_fault(struct vm_fault *vmf)
3302 { 2477 {
3303 int error; 2478 int error;
3304 struct file *file = vmf->vma->vm_file 2479 struct file *file = vmf->vma->vm_file;
3305 struct file *fpin = NULL; 2480 struct file *fpin = NULL;
3306 struct address_space *mapping = file- 2481 struct address_space *mapping = file->f_mapping;
>> 2482 struct file_ra_state *ra = &file->f_ra;
3307 struct inode *inode = mapping->host; 2483 struct inode *inode = mapping->host;
3308 pgoff_t max_idx, index = vmf->pgoff; !! 2484 pgoff_t offset = vmf->pgoff;
3309 struct folio *folio; !! 2485 pgoff_t max_off;
>> 2486 struct page *page;
3310 vm_fault_t ret = 0; 2487 vm_fault_t ret = 0;
3311 bool mapping_locked = false; <<
3312 2488
3313 max_idx = DIV_ROUND_UP(i_size_read(in !! 2489 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3314 if (unlikely(index >= max_idx)) !! 2490 if (unlikely(offset >= max_off))
3315 return VM_FAULT_SIGBUS; 2491 return VM_FAULT_SIGBUS;
3316 2492
3317 trace_mm_filemap_fault(mapping, index <<
3318 <<
3319 /* 2493 /*
3320 * Do we have something in the page c 2494 * Do we have something in the page cache already?
3321 */ 2495 */
3322 folio = filemap_get_folio(mapping, in !! 2496 page = find_get_page(mapping, offset);
3323 if (likely(!IS_ERR(folio))) { !! 2497 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
3324 /* 2498 /*
3325 * We found the page, so try !! 2499 * We found the page, so try async readahead before
3326 * the lock. !! 2500 * waiting for the lock.
3327 */ 2501 */
3328 if (!(vmf->flags & FAULT_FLAG !! 2502 fpin = do_async_mmap_readahead(vmf, page);
3329 fpin = do_async_mmap_ !! 2503 } 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 2504 /* No page in the page cache at all */
3340 count_vm_event(PGMAJFAULT); 2505 count_vm_event(PGMAJFAULT);
3341 count_memcg_event_mm(vmf->vma 2506 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3342 ret = VM_FAULT_MAJOR; 2507 ret = VM_FAULT_MAJOR;
3343 fpin = do_sync_mmap_readahead 2508 fpin = do_sync_mmap_readahead(vmf);
3344 retry_find: 2509 retry_find:
3345 /* !! 2510 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 2511 FGP_CREAT|FGP_FOR_MMAP,
3355 vmf 2512 vmf->gfp_mask);
3356 if (IS_ERR(folio)) { !! 2513 if (!page) {
3357 if (fpin) 2514 if (fpin)
3358 goto out_retr 2515 goto out_retry;
3359 filemap_invalidate_un !! 2516 return vmf_error(-ENOMEM);
3360 return VM_FAULT_OOM; <<
3361 } 2517 }
3362 } 2518 }
3363 2519
3364 if (!lock_folio_maybe_drop_mmap(vmf, !! 2520 if (!lock_page_maybe_drop_mmap(vmf, page, &fpin))
3365 goto out_retry; 2521 goto out_retry;
3366 2522
3367 /* Did it get truncated? */ 2523 /* Did it get truncated? */
3368 if (unlikely(folio->mapping != mappin !! 2524 if (unlikely(compound_head(page)->mapping != mapping)) {
3369 folio_unlock(folio); !! 2525 unlock_page(page);
3370 folio_put(folio); !! 2526 put_page(page);
3371 goto retry_find; 2527 goto retry_find;
3372 } 2528 }
3373 VM_BUG_ON_FOLIO(!folio_contains(folio !! 2529 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
3374 2530
3375 /* 2531 /*
3376 * We have a locked folio in the page !! 2532 * 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 !! 2533 * that it's up-to-date. If not, it is going to be due to an error.
3378 * or because readahead was otherwise <<
3379 */ 2534 */
3380 if (unlikely(!folio_test_uptodate(fol !! 2535 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; 2536 goto page_not_uptodate;
3399 } <<
3400 2537
3401 /* 2538 /*
3402 * We've made it this far and we had !! 2539 * We've made it this far and we had to drop our mmap_sem, now is the
3403 * time to return to the upper layer 2540 * time to return to the upper layer and have it re-find the vma and
3404 * redo the fault. 2541 * redo the fault.
3405 */ 2542 */
3406 if (fpin) { 2543 if (fpin) {
3407 folio_unlock(folio); !! 2544 unlock_page(page);
3408 goto out_retry; 2545 goto out_retry;
3409 } 2546 }
3410 if (mapping_locked) <<
3411 filemap_invalidate_unlock_sha <<
3412 2547
3413 /* 2548 /*
3414 * Found the page and have a referenc 2549 * Found the page and have a reference on it.
3415 * We must recheck i_size under page 2550 * We must recheck i_size under page lock.
3416 */ 2551 */
3417 max_idx = DIV_ROUND_UP(i_size_read(in !! 2552 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3418 if (unlikely(index >= max_idx)) { !! 2553 if (unlikely(offset >= max_off)) {
3419 folio_unlock(folio); !! 2554 unlock_page(page);
3420 folio_put(folio); !! 2555 put_page(page);
3421 return VM_FAULT_SIGBUS; 2556 return VM_FAULT_SIGBUS;
3422 } 2557 }
3423 2558
3424 vmf->page = folio_file_page(folio, in !! 2559 vmf->page = page;
3425 return ret | VM_FAULT_LOCKED; 2560 return ret | VM_FAULT_LOCKED;
3426 2561
3427 page_not_uptodate: 2562 page_not_uptodate:
3428 /* 2563 /*
3429 * Umm, take care of errors if the pa 2564 * Umm, take care of errors if the page isn't up-to-date.
3430 * Try to re-read it _once_. We do th 2565 * Try to re-read it _once_. We do this synchronously,
3431 * because there really aren't any pe 2566 * because there really aren't any performance issues here
3432 * and we need to check for errors. 2567 * and we need to check for errors.
3433 */ 2568 */
>> 2569 ClearPageError(page);
3434 fpin = maybe_unlock_mmap_for_io(vmf, 2570 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3435 error = filemap_read_folio(file, mapp !! 2571 error = mapping->a_ops->readpage(file, page);
>> 2572 if (!error) {
>> 2573 wait_on_page_locked(page);
>> 2574 if (!PageUptodate(page))
>> 2575 error = -EIO;
>> 2576 }
3436 if (fpin) 2577 if (fpin)
3437 goto out_retry; 2578 goto out_retry;
3438 folio_put(folio); !! 2579 put_page(page);
3439 2580
3440 if (!error || error == AOP_TRUNCATED_ 2581 if (!error || error == AOP_TRUNCATED_PAGE)
3441 goto retry_find; 2582 goto retry_find;
3442 filemap_invalidate_unlock_shared(mapp <<
3443 2583
>> 2584 /* Things didn't work out. Return zero to tell the mm layer so. */
>> 2585 shrink_readahead_size_eio(file, ra);
3444 return VM_FAULT_SIGBUS; 2586 return VM_FAULT_SIGBUS;
3445 2587
3446 out_retry: 2588 out_retry:
3447 /* 2589 /*
3448 * We dropped the mmap_lock, we need !! 2590 * We dropped the mmap_sem, we need to return to the fault handler to
3449 * re-find the vma and come back and 2591 * re-find the vma and come back and find our hopefully still populated
3450 * page. 2592 * page.
3451 */ 2593 */
3452 if (!IS_ERR(folio)) !! 2594 if (page)
3453 folio_put(folio); !! 2595 put_page(page);
3454 if (mapping_locked) <<
3455 filemap_invalidate_unlock_sha <<
3456 if (fpin) 2596 if (fpin)
3457 fput(fpin); 2597 fput(fpin);
3458 return ret | VM_FAULT_RETRY; 2598 return ret | VM_FAULT_RETRY;
3459 } 2599 }
3460 EXPORT_SYMBOL(filemap_fault); 2600 EXPORT_SYMBOL(filemap_fault);
3461 2601
3462 static bool filemap_map_pmd(struct vm_fault * !! 2602 void filemap_map_pages(struct vm_fault *vmf,
3463 pgoff_t start) !! 2603 pgoff_t start_pgoff, pgoff_t end_pgoff)
3464 { <<
3465 struct mm_struct *mm = vmf->vma->vm_m <<
3466 <<
3467 /* Huge page is mapped? No need to pr <<
3468 if (pmd_trans_huge(*vmf->pmd)) { <<
3469 folio_unlock(folio); <<
3470 folio_put(folio); <<
3471 return true; <<
3472 } <<
3473 <<
3474 if (pmd_none(*vmf->pmd) && folio_test <<
3475 struct page *page = folio_fil <<
3476 vm_fault_t ret = do_set_pmd(v <<
3477 if (!ret) { <<
3478 /* The page is mapped <<
3479 folio_unlock(folio); <<
3480 return true; <<
3481 } <<
3482 } <<
3483 <<
3484 if (pmd_none(*vmf->pmd) && vmf->preal <<
3485 pmd_install(mm, vmf->pmd, &vm <<
3486 <<
3487 return false; <<
3488 } <<
3489 <<
3490 static struct folio *next_uptodate_folio(stru <<
3491 struct address_space *mapping <<
3492 { 2604 {
3493 struct folio *folio = xas_next_entry( !! 2605 struct file *file = vmf->vma->vm_file;
>> 2606 struct address_space *mapping = file->f_mapping;
>> 2607 pgoff_t last_pgoff = start_pgoff;
3494 unsigned long max_idx; 2608 unsigned long max_idx;
>> 2609 XA_STATE(xas, &mapping->i_pages, start_pgoff);
>> 2610 struct page *page;
3495 2611
3496 do { !! 2612 rcu_read_lock();
3497 if (!folio) !! 2613 xas_for_each(&xas, page, end_pgoff) {
3498 return NULL; !! 2614 if (xas_retry(&xas, page))
3499 if (xas_retry(xas, folio)) <<
3500 continue; <<
3501 if (xa_is_value(folio)) <<
3502 continue; <<
3503 if (folio_test_locked(folio)) <<
3504 continue; <<
3505 if (!folio_try_get(folio)) <<
3506 continue; 2615 continue;
3507 /* Has the page moved or been !! 2616 if (xa_is_value(page))
3508 if (unlikely(folio != xas_rel !! 2617 goto next;
3509 goto skip; <<
3510 if (!folio_test_uptodate(foli <<
3511 goto skip; <<
3512 if (!folio_trylock(folio)) <<
3513 goto skip; <<
3514 if (folio->mapping != mapping <<
3515 goto unlock; <<
3516 if (!folio_test_uptodate(foli <<
3517 goto unlock; <<
3518 max_idx = DIV_ROUND_UP(i_size <<
3519 if (xas->xa_index >= max_idx) <<
3520 goto unlock; <<
3521 return folio; <<
3522 unlock: <<
3523 folio_unlock(folio); <<
3524 skip: <<
3525 folio_put(folio); <<
3526 } while ((folio = xas_next_entry(xas, <<
3527 <<
3528 return NULL; <<
3529 } <<
3530 <<
3531 /* <<
3532 * Map page range [start_page, start_page + n <<
3533 * start_page is gotten from start by folio_p <<
3534 */ <<
3535 static vm_fault_t filemap_map_folio_range(str <<
3536 struct folio *folio, <<
3537 unsigned long addr, u <<
3538 unsigned long *rss, u <<
3539 { <<
3540 vm_fault_t ret = 0; <<
3541 struct page *page = folio_page(folio, <<
3542 unsigned int count = 0; <<
3543 pte_t *old_ptep = vmf->pte; <<
3544 <<
3545 do { <<
3546 if (PageHWPoison(page + count <<
3547 goto skip; <<
3548 2618
3549 /* 2619 /*
3550 * If there are too many foli !! 2620 * Check for a locked page first, as a speculative
3551 * in a file, they will proba !! 2621 * reference may adversely influence page migration.
3552 * In such situation, read-ah <<
3553 * Don't decrease mmap_miss i <<
3554 * we can stop read-ahead. <<
3555 */ 2622 */
3556 if (!folio_test_workingset(fo !! 2623 if (PageLocked(page))
3557 (*mmap_miss)++; !! 2624 goto next;
>> 2625 if (!page_cache_get_speculative(page))
>> 2626 goto next;
3558 2627
3559 /* !! 2628 /* Has the page moved or been split? */
3560 * NOTE: If there're PTE mark !! 2629 if (unlikely(page != xas_reload(&xas)))
3561 * handled in the specific fa <<
3562 * fault-around logic. <<
3563 */ <<
3564 if (!pte_none(ptep_get(&vmf-> <<
3565 goto skip; 2630 goto skip;
>> 2631 page = find_subpage(page, xas.xa_index);
3566 2632
3567 count++; !! 2633 if (!PageUptodate(page) ||
3568 continue; !! 2634 PageReadahead(page) ||
3569 skip: !! 2635 PageHWPoison(page))
3570 if (count) { !! 2636 goto skip;
3571 set_pte_range(vmf, fo !! 2637 if (!trylock_page(page))
3572 *rss += count; !! 2638 goto skip;
3573 folio_ref_add(folio, <<
3574 if (in_range(vmf->add <<
3575 ret = VM_FAUL <<
3576 } <<
3577 <<
3578 count++; <<
3579 page += count; <<
3580 vmf->pte += count; <<
3581 addr += count * PAGE_SIZE; <<
3582 count = 0; <<
3583 } while (--nr_pages > 0); <<
3584 <<
3585 if (count) { <<
3586 set_pte_range(vmf, folio, pag <<
3587 *rss += count; <<
3588 folio_ref_add(folio, count); <<
3589 if (in_range(vmf->address, ad <<
3590 ret = VM_FAULT_NOPAGE <<
3591 } <<
3592 <<
3593 vmf->pte = old_ptep; <<
3594 <<
3595 return ret; <<
3596 } <<
3597 <<
3598 static vm_fault_t filemap_map_order0_folio(st <<
3599 struct folio *folio, unsigned <<
3600 unsigned long *rss, unsigned <<
3601 { <<
3602 vm_fault_t ret = 0; <<
3603 struct page *page = &folio->page; <<
3604 <<
3605 if (PageHWPoison(page)) <<
3606 return ret; <<
3607 <<
3608 /* See comment of filemap_map_folio_r <<
3609 if (!folio_test_workingset(folio)) <<
3610 (*mmap_miss)++; <<
3611 <<
3612 /* <<
3613 * NOTE: If there're PTE markers, we' <<
3614 * handled in the specific fault path <<
3615 * the fault-around logic. <<
3616 */ <<
3617 if (!pte_none(ptep_get(vmf->pte))) <<
3618 return ret; <<
3619 <<
3620 if (vmf->address == addr) <<
3621 ret = VM_FAULT_NOPAGE; <<
3622 <<
3623 set_pte_range(vmf, folio, page, 1, ad <<
3624 (*rss)++; <<
3625 folio_ref_inc(folio); <<
3626 <<
3627 return ret; <<
3628 } <<
3629 <<
3630 vm_fault_t filemap_map_pages(struct vm_fault <<
3631 pgoff_t start_pg <<
3632 { <<
3633 struct vm_area_struct *vma = vmf->vma <<
3634 struct file *file = vma->vm_file; <<
3635 struct address_space *mapping = file- <<
3636 pgoff_t file_end, last_pgoff = start_ <<
3637 unsigned long addr; <<
3638 XA_STATE(xas, &mapping->i_pages, star <<
3639 struct folio *folio; <<
3640 vm_fault_t ret = 0; <<
3641 unsigned long rss = 0; <<
3642 unsigned int nr_pages = 0, mmap_miss <<
3643 <<
3644 rcu_read_lock(); <<
3645 folio = next_uptodate_folio(&xas, map <<
3646 if (!folio) <<
3647 goto out; <<
3648 <<
3649 if (filemap_map_pmd(vmf, folio, start <<
3650 ret = VM_FAULT_NOPAGE; <<
3651 goto out; <<
3652 } <<
3653 2639
3654 addr = vma->vm_start + ((start_pgoff !! 2640 if (page->mapping != mapping || !PageUptodate(page))
3655 vmf->pte = pte_offset_map_lock(vma->v !! 2641 goto unlock;
3656 if (!vmf->pte) { <<
3657 folio_unlock(folio); <<
3658 folio_put(folio); <<
3659 goto out; <<
3660 } <<
3661 2642
3662 file_end = DIV_ROUND_UP(i_size_read(m !! 2643 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3663 if (end_pgoff > file_end) !! 2644 if (page->index >= max_idx)
3664 end_pgoff = file_end; !! 2645 goto unlock;
3665 2646
3666 folio_type = mm_counter_file(folio); !! 2647 if (file->f_ra.mmap_miss > 0)
3667 do { !! 2648 file->f_ra.mmap_miss--;
3668 unsigned long end; <<
3669 2649
3670 addr += (xas.xa_index - last_ !! 2650 vmf->address += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3671 vmf->pte += xas.xa_index - la !! 2651 if (vmf->pte)
>> 2652 vmf->pte += xas.xa_index - last_pgoff;
3672 last_pgoff = xas.xa_index; 2653 last_pgoff = xas.xa_index;
3673 end = folio_next_index(folio) !! 2654 if (alloc_set_pte(vmf, NULL, page))
3674 nr_pages = min(end, end_pgoff !! 2655 goto unlock;
3675 !! 2656 unlock_page(page);
3676 if (!folio_test_large(folio)) !! 2657 goto next;
3677 ret |= filemap_map_or !! 2658 unlock:
3678 folio !! 2659 unlock_page(page);
3679 else !! 2660 skip:
3680 ret |= filemap_map_fo !! 2661 put_page(page);
3681 xas.x !! 2662 next:
3682 nr_pa !! 2663 /* Huge page is mapped? No need to proceed. */
3683 !! 2664 if (pmd_trans_huge(*vmf->pmd))
3684 folio_unlock(folio); !! 2665 break;
3685 folio_put(folio); !! 2666 }
3686 } while ((folio = next_uptodate_folio <<
3687 add_mm_counter(vma->vm_mm, folio_type <<
3688 pte_unmap_unlock(vmf->pte, vmf->ptl); <<
3689 trace_mm_filemap_map_pages(mapping, s <<
3690 out: <<
3691 rcu_read_unlock(); 2667 rcu_read_unlock();
3692 <<
3693 mmap_miss_saved = READ_ONCE(file->f_r <<
3694 if (mmap_miss >= mmap_miss_saved) <<
3695 WRITE_ONCE(file->f_ra.mmap_mi <<
3696 else <<
3697 WRITE_ONCE(file->f_ra.mmap_mi <<
3698 <<
3699 return ret; <<
3700 } 2668 }
3701 EXPORT_SYMBOL(filemap_map_pages); 2669 EXPORT_SYMBOL(filemap_map_pages);
3702 2670
3703 vm_fault_t filemap_page_mkwrite(struct vm_fau 2671 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3704 { 2672 {
3705 struct address_space *mapping = vmf-> !! 2673 struct page *page = vmf->page;
3706 struct folio *folio = page_folio(vmf- !! 2674 struct inode *inode = file_inode(vmf->vma->vm_file);
3707 vm_fault_t ret = VM_FAULT_LOCKED; 2675 vm_fault_t ret = VM_FAULT_LOCKED;
3708 2676
3709 sb_start_pagefault(mapping->host->i_s !! 2677 sb_start_pagefault(inode->i_sb);
3710 file_update_time(vmf->vma->vm_file); 2678 file_update_time(vmf->vma->vm_file);
3711 folio_lock(folio); !! 2679 lock_page(page);
3712 if (folio->mapping != mapping) { !! 2680 if (page->mapping != inode->i_mapping) {
3713 folio_unlock(folio); !! 2681 unlock_page(page);
3714 ret = VM_FAULT_NOPAGE; 2682 ret = VM_FAULT_NOPAGE;
3715 goto out; 2683 goto out;
3716 } 2684 }
3717 /* 2685 /*
3718 * We mark the folio dirty already he !! 2686 * We mark the page dirty already here so that when freeze is in
3719 * progress, we are guaranteed that w 2687 * progress, we are guaranteed that writeback during freezing will
3720 * see the dirty folio and writeprote !! 2688 * see the dirty page and writeprotect it again.
3721 */ 2689 */
3722 folio_mark_dirty(folio); !! 2690 set_page_dirty(page);
3723 folio_wait_stable(folio); !! 2691 wait_for_stable_page(page);
3724 out: 2692 out:
3725 sb_end_pagefault(mapping->host->i_sb) !! 2693 sb_end_pagefault(inode->i_sb);
3726 return ret; 2694 return ret;
3727 } 2695 }
3728 2696
3729 const struct vm_operations_struct generic_fil 2697 const struct vm_operations_struct generic_file_vm_ops = {
3730 .fault = filemap_fault, 2698 .fault = filemap_fault,
3731 .map_pages = filemap_map_pages, 2699 .map_pages = filemap_map_pages,
3732 .page_mkwrite = filemap_page_mkwrit 2700 .page_mkwrite = filemap_page_mkwrite,
3733 }; 2701 };
3734 2702
3735 /* This is used for a general mmap of a disk 2703 /* This is used for a general mmap of a disk file */
3736 2704
3737 int generic_file_mmap(struct file *file, stru !! 2705 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
3738 { 2706 {
3739 struct address_space *mapping = file- 2707 struct address_space *mapping = file->f_mapping;
3740 2708
3741 if (!mapping->a_ops->read_folio) !! 2709 if (!mapping->a_ops->readpage)
3742 return -ENOEXEC; 2710 return -ENOEXEC;
3743 file_accessed(file); 2711 file_accessed(file);
3744 vma->vm_ops = &generic_file_vm_ops; 2712 vma->vm_ops = &generic_file_vm_ops;
3745 return 0; 2713 return 0;
3746 } 2714 }
3747 2715
3748 /* 2716 /*
3749 * This is for filesystems which do not imple 2717 * This is for filesystems which do not implement ->writepage.
3750 */ 2718 */
3751 int generic_file_readonly_mmap(struct file *f 2719 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3752 { 2720 {
3753 if (vma_is_shared_maywrite(vma)) !! 2721 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3754 return -EINVAL; 2722 return -EINVAL;
3755 return generic_file_mmap(file, vma); 2723 return generic_file_mmap(file, vma);
3756 } 2724 }
3757 #else 2725 #else
3758 vm_fault_t filemap_page_mkwrite(struct vm_fau 2726 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3759 { 2727 {
3760 return VM_FAULT_SIGBUS; 2728 return VM_FAULT_SIGBUS;
3761 } 2729 }
3762 int generic_file_mmap(struct file *file, stru !! 2730 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
3763 { 2731 {
3764 return -ENOSYS; 2732 return -ENOSYS;
3765 } 2733 }
3766 int generic_file_readonly_mmap(struct file *f !! 2734 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
3767 { 2735 {
3768 return -ENOSYS; 2736 return -ENOSYS;
3769 } 2737 }
3770 #endif /* CONFIG_MMU */ 2738 #endif /* CONFIG_MMU */
3771 2739
3772 EXPORT_SYMBOL(filemap_page_mkwrite); 2740 EXPORT_SYMBOL(filemap_page_mkwrite);
3773 EXPORT_SYMBOL(generic_file_mmap); 2741 EXPORT_SYMBOL(generic_file_mmap);
3774 EXPORT_SYMBOL(generic_file_readonly_mmap); 2742 EXPORT_SYMBOL(generic_file_readonly_mmap);
3775 2743
3776 static struct folio *do_read_cache_folio(stru !! 2744 static struct page *wait_on_page_read(struct page *page)
3777 pgoff_t index, filler_t fille <<
3778 { 2745 {
3779 struct folio *folio; !! 2746 if (!IS_ERR(page)) {
3780 int err; !! 2747 wait_on_page_locked(page);
>> 2748 if (!PageUptodate(page)) {
>> 2749 put_page(page);
>> 2750 page = ERR_PTR(-EIO);
>> 2751 }
>> 2752 }
>> 2753 return page;
>> 2754 }
3781 2755
3782 if (!filler) !! 2756 static struct page *do_read_cache_page(struct address_space *mapping,
3783 filler = mapping->a_ops->read !! 2757 pgoff_t index,
>> 2758 int (*filler)(void *, struct page *),
>> 2759 void *data,
>> 2760 gfp_t gfp)
>> 2761 {
>> 2762 struct page *page;
>> 2763 int err;
3784 repeat: 2764 repeat:
3785 folio = filemap_get_folio(mapping, in !! 2765 page = find_get_page(mapping, index);
3786 if (IS_ERR(folio)) { !! 2766 if (!page) {
3787 folio = filemap_alloc_folio(g !! 2767 page = __page_cache_alloc(gfp);
3788 m !! 2768 if (!page)
3789 if (!folio) <<
3790 return ERR_PTR(-ENOME 2769 return ERR_PTR(-ENOMEM);
3791 index = mapping_align_index(m !! 2770 err = add_to_page_cache_lru(page, mapping, index, gfp);
3792 err = filemap_add_folio(mappi <<
3793 if (unlikely(err)) { 2771 if (unlikely(err)) {
3794 folio_put(folio); !! 2772 put_page(page);
3795 if (err == -EEXIST) 2773 if (err == -EEXIST)
3796 goto repeat; 2774 goto repeat;
3797 /* Presumably ENOMEM 2775 /* Presumably ENOMEM for xarray node */
3798 return ERR_PTR(err); 2776 return ERR_PTR(err);
3799 } 2777 }
3800 2778
3801 goto filler; !! 2779 filler:
>> 2780 if (filler)
>> 2781 err = filler(data, page);
>> 2782 else
>> 2783 err = mapping->a_ops->readpage(data, page);
>> 2784
>> 2785 if (err < 0) {
>> 2786 put_page(page);
>> 2787 return ERR_PTR(err);
>> 2788 }
>> 2789
>> 2790 page = wait_on_page_read(page);
>> 2791 if (IS_ERR(page))
>> 2792 return page;
>> 2793 goto out;
3802 } 2794 }
3803 if (folio_test_uptodate(folio)) !! 2795 if (PageUptodate(page))
3804 goto out; 2796 goto out;
3805 2797
3806 if (!folio_trylock(folio)) { !! 2798 /*
3807 folio_put_wait_locked(folio, !! 2799 * Page is not up to date and may be locked due one of the following
3808 goto repeat; !! 2800 * case a: Page is being filled and the page lock is held
3809 } !! 2801 * case b: Read/write error clearing the page uptodate status
>> 2802 * case c: Truncation in progress (page locked)
>> 2803 * case d: Reclaim in progress
>> 2804 *
>> 2805 * Case a, the page will be up to date when the page is unlocked.
>> 2806 * There is no need to serialise on the page lock here as the page
>> 2807 * is pinned so the lock gives no additional protection. Even if the
>> 2808 * the page is truncated, the data is still valid if PageUptodate as
>> 2809 * it's a race vs truncate race.
>> 2810 * Case b, the page will not be up to date
>> 2811 * Case c, the page may be truncated but in itself, the data may still
>> 2812 * be valid after IO completes as it's a read vs truncate race. The
>> 2813 * operation must restart if the page is not uptodate on unlock but
>> 2814 * otherwise serialising on page lock to stabilise the mapping gives
>> 2815 * no additional guarantees to the caller as the page lock is
>> 2816 * released before return.
>> 2817 * Case d, similar to truncation. If reclaim holds the page lock, it
>> 2818 * will be a race with remove_mapping that determines if the mapping
>> 2819 * is valid on unlock but otherwise the data is valid and there is
>> 2820 * no need to serialise with page lock.
>> 2821 *
>> 2822 * As the page lock gives no additional guarantee, we optimistically
>> 2823 * wait on the page to be unlocked and check if it's up to date and
>> 2824 * use the page if it is. Otherwise, the page lock is required to
>> 2825 * distinguish between the different cases. The motivation is that we
>> 2826 * avoid spurious serialisations and wakeups when multiple processes
>> 2827 * wait on the same page for IO to complete.
>> 2828 */
>> 2829 wait_on_page_locked(page);
>> 2830 if (PageUptodate(page))
>> 2831 goto out;
3810 2832
3811 /* Folio was truncated from mapping * !! 2833 /* Distinguish between all the cases under the safety of the lock */
3812 if (!folio->mapping) { !! 2834 lock_page(page);
3813 folio_unlock(folio); !! 2835
3814 folio_put(folio); !! 2836 /* Case c or d, restart the operation */
>> 2837 if (!page->mapping) {
>> 2838 unlock_page(page);
>> 2839 put_page(page);
3815 goto repeat; 2840 goto repeat;
3816 } 2841 }
3817 2842
3818 /* Someone else locked and filled the 2843 /* Someone else locked and filled the page in a very small window */
3819 if (folio_test_uptodate(folio)) { !! 2844 if (PageUptodate(page)) {
3820 folio_unlock(folio); !! 2845 unlock_page(page);
3821 goto out; 2846 goto out;
3822 } 2847 }
3823 !! 2848 goto filler;
3824 filler: <<
3825 err = filemap_read_folio(file, filler <<
3826 if (err) { <<
3827 folio_put(folio); <<
3828 if (err == AOP_TRUNCATED_PAGE <<
3829 goto repeat; <<
3830 return ERR_PTR(err); <<
3831 } <<
3832 2849
3833 out: 2850 out:
3834 folio_mark_accessed(folio); !! 2851 mark_page_accessed(page);
3835 return folio; !! 2852 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 } 2853 }
3860 EXPORT_SYMBOL(read_cache_folio); <<
3861 2854
3862 /** 2855 /**
3863 * mapping_read_folio_gfp - Read into page ca !! 2856 * read_cache_page - read into page cache, fill it if needed
3864 * @mapping: The address_space for the fol !! 2857 * @mapping: the page's address_space
3865 * @index: The index that the allocated !! 2858 * @index: the page index
3866 * @gfp: The page allocator flags to u !! 2859 * @filler: function to perform the read
3867 * !! 2860 * @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 * 2861 *
3871 * The most likely error from this function i !! 2862 * Read into the page cache. If a page already exists, and PageUptodate() is
3872 * possible and so is EINTR. If ->read_folio !! 2863 * not set, try to fill the page and wait for it to become unlocked.
3873 * that will be returned to the caller. <<
3874 * 2864 *
3875 * The function expects mapping->invalidate_l !! 2865 * If the page does not get brought uptodate, return -EIO.
3876 * 2866 *
3877 * Return: Uptodate folio on success, ERR_PTR !! 2867 * Return: up to date page on success, ERR_PTR() on failure.
3878 */ 2868 */
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 2869 struct page *read_cache_page(struct address_space *mapping,
3898 pgoff_t index, filler !! 2870 pgoff_t index,
>> 2871 int (*filler)(void *, struct page *),
>> 2872 void *data)
3899 { 2873 {
3900 return do_read_cache_page(mapping, in !! 2874 return do_read_cache_page(mapping, index, filler, data,
3901 mapping_gfp_mask(mapp 2875 mapping_gfp_mask(mapping));
3902 } 2876 }
3903 EXPORT_SYMBOL(read_cache_page); 2877 EXPORT_SYMBOL(read_cache_page);
3904 2878
3905 /** 2879 /**
3906 * read_cache_page_gfp - read into page cache 2880 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3907 * @mapping: the page's address_space 2881 * @mapping: the page's address_space
3908 * @index: the page index 2882 * @index: the page index
3909 * @gfp: the page allocator flags to u 2883 * @gfp: the page allocator flags to use if allocating
3910 * 2884 *
3911 * This is the same as "read_mapping_page(map 2885 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3912 * any new page allocations done using the sp 2886 * any new page allocations done using the specified allocation flags.
3913 * 2887 *
3914 * If the page does not get brought uptodate, 2888 * If the page does not get brought uptodate, return -EIO.
3915 * 2889 *
3916 * The function expects mapping->invalidate_l <<
3917 * <<
3918 * Return: up to date page on success, ERR_PT 2890 * Return: up to date page on success, ERR_PTR() on failure.
3919 */ 2891 */
3920 struct page *read_cache_page_gfp(struct addre 2892 struct page *read_cache_page_gfp(struct address_space *mapping,
3921 pgoff_t index 2893 pgoff_t index,
3922 gfp_t gfp) 2894 gfp_t gfp)
3923 { 2895 {
3924 return do_read_cache_page(mapping, in 2896 return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3925 } 2897 }
3926 EXPORT_SYMBOL(read_cache_page_gfp); 2898 EXPORT_SYMBOL(read_cache_page_gfp);
3927 2899
3928 /* 2900 /*
>> 2901 * Don't operate on ranges the page cache doesn't support, and don't exceed the
>> 2902 * LFS limits. If pos is under the limit it becomes a short access. If it
>> 2903 * exceeds the limit we return -EFBIG.
>> 2904 */
>> 2905 static int generic_write_check_limits(struct file *file, loff_t pos,
>> 2906 loff_t *count)
>> 2907 {
>> 2908 struct inode *inode = file->f_mapping->host;
>> 2909 loff_t max_size = inode->i_sb->s_maxbytes;
>> 2910 loff_t limit = rlimit(RLIMIT_FSIZE);
>> 2911
>> 2912 if (limit != RLIM_INFINITY) {
>> 2913 if (pos >= limit) {
>> 2914 send_sig(SIGXFSZ, current, 0);
>> 2915 return -EFBIG;
>> 2916 }
>> 2917 *count = min(*count, limit - pos);
>> 2918 }
>> 2919
>> 2920 if (!(file->f_flags & O_LARGEFILE))
>> 2921 max_size = MAX_NON_LFS;
>> 2922
>> 2923 if (unlikely(pos >= max_size))
>> 2924 return -EFBIG;
>> 2925
>> 2926 *count = min(*count, max_size - pos);
>> 2927
>> 2928 return 0;
>> 2929 }
>> 2930
>> 2931 /*
>> 2932 * Performs necessary checks before doing a write
>> 2933 *
>> 2934 * Can adjust writing position or amount of bytes to write.
>> 2935 * Returns appropriate error code that caller should return or
>> 2936 * zero in case that write should be allowed.
>> 2937 */
>> 2938 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
>> 2939 {
>> 2940 struct file *file = iocb->ki_filp;
>> 2941 struct inode *inode = file->f_mapping->host;
>> 2942 loff_t count;
>> 2943 int ret;
>> 2944
>> 2945 if (IS_SWAPFILE(inode))
>> 2946 return -ETXTBSY;
>> 2947
>> 2948 if (!iov_iter_count(from))
>> 2949 return 0;
>> 2950
>> 2951 /* FIXME: this is for backwards compatibility with 2.4 */
>> 2952 if (iocb->ki_flags & IOCB_APPEND)
>> 2953 iocb->ki_pos = i_size_read(inode);
>> 2954
>> 2955 if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
>> 2956 return -EINVAL;
>> 2957
>> 2958 count = iov_iter_count(from);
>> 2959 ret = generic_write_check_limits(file, iocb->ki_pos, &count);
>> 2960 if (ret)
>> 2961 return ret;
>> 2962
>> 2963 iov_iter_truncate(from, count);
>> 2964 return iov_iter_count(from);
>> 2965 }
>> 2966 EXPORT_SYMBOL(generic_write_checks);
>> 2967
>> 2968 /*
>> 2969 * Performs necessary checks before doing a clone.
>> 2970 *
>> 2971 * Can adjust amount of bytes to clone via @req_count argument.
>> 2972 * Returns appropriate error code that caller should return or
>> 2973 * zero in case the clone should be allowed.
>> 2974 */
>> 2975 int generic_remap_checks(struct file *file_in, loff_t pos_in,
>> 2976 struct file *file_out, loff_t pos_out,
>> 2977 loff_t *req_count, unsigned int remap_flags)
>> 2978 {
>> 2979 struct inode *inode_in = file_in->f_mapping->host;
>> 2980 struct inode *inode_out = file_out->f_mapping->host;
>> 2981 uint64_t count = *req_count;
>> 2982 uint64_t bcount;
>> 2983 loff_t size_in, size_out;
>> 2984 loff_t bs = inode_out->i_sb->s_blocksize;
>> 2985 int ret;
>> 2986
>> 2987 /* The start of both ranges must be aligned to an fs block. */
>> 2988 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs))
>> 2989 return -EINVAL;
>> 2990
>> 2991 /* Ensure offsets don't wrap. */
>> 2992 if (pos_in + count < pos_in || pos_out + count < pos_out)
>> 2993 return -EINVAL;
>> 2994
>> 2995 size_in = i_size_read(inode_in);
>> 2996 size_out = i_size_read(inode_out);
>> 2997
>> 2998 /* Dedupe requires both ranges to be within EOF. */
>> 2999 if ((remap_flags & REMAP_FILE_DEDUP) &&
>> 3000 (pos_in >= size_in || pos_in + count > size_in ||
>> 3001 pos_out >= size_out || pos_out + count > size_out))
>> 3002 return -EINVAL;
>> 3003
>> 3004 /* Ensure the infile range is within the infile. */
>> 3005 if (pos_in >= size_in)
>> 3006 return -EINVAL;
>> 3007 count = min(count, size_in - (uint64_t)pos_in);
>> 3008
>> 3009 ret = generic_write_check_limits(file_out, pos_out, &count);
>> 3010 if (ret)
>> 3011 return ret;
>> 3012
>> 3013 /*
>> 3014 * If the user wanted us to link to the infile's EOF, round up to the
>> 3015 * next block boundary for this check.
>> 3016 *
>> 3017 * Otherwise, make sure the count is also block-aligned, having
>> 3018 * already confirmed the starting offsets' block alignment.
>> 3019 */
>> 3020 if (pos_in + count == size_in) {
>> 3021 bcount = ALIGN(size_in, bs) - pos_in;
>> 3022 } else {
>> 3023 if (!IS_ALIGNED(count, bs))
>> 3024 count = ALIGN_DOWN(count, bs);
>> 3025 bcount = count;
>> 3026 }
>> 3027
>> 3028 /* Don't allow overlapped cloning within the same file. */
>> 3029 if (inode_in == inode_out &&
>> 3030 pos_out + bcount > pos_in &&
>> 3031 pos_out < pos_in + bcount)
>> 3032 return -EINVAL;
>> 3033
>> 3034 /*
>> 3035 * We shortened the request but the caller can't deal with that, so
>> 3036 * bounce the request back to userspace.
>> 3037 */
>> 3038 if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN))
>> 3039 return -EINVAL;
>> 3040
>> 3041 *req_count = count;
>> 3042 return 0;
>> 3043 }
>> 3044
>> 3045
>> 3046 /*
>> 3047 * Performs common checks before doing a file copy/clone
>> 3048 * from @file_in to @file_out.
>> 3049 */
>> 3050 int generic_file_rw_checks(struct file *file_in, struct file *file_out)
>> 3051 {
>> 3052 struct inode *inode_in = file_inode(file_in);
>> 3053 struct inode *inode_out = file_inode(file_out);
>> 3054
>> 3055 /* Don't copy dirs, pipes, sockets... */
>> 3056 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
>> 3057 return -EISDIR;
>> 3058 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
>> 3059 return -EINVAL;
>> 3060
>> 3061 if (!(file_in->f_mode & FMODE_READ) ||
>> 3062 !(file_out->f_mode & FMODE_WRITE) ||
>> 3063 (file_out->f_flags & O_APPEND))
>> 3064 return -EBADF;
>> 3065
>> 3066 return 0;
>> 3067 }
>> 3068
>> 3069 /*
>> 3070 * Performs necessary checks before doing a file copy
>> 3071 *
>> 3072 * Can adjust amount of bytes to copy via @req_count argument.
>> 3073 * Returns appropriate error code that caller should return or
>> 3074 * zero in case the copy should be allowed.
>> 3075 */
>> 3076 int generic_copy_file_checks(struct file *file_in, loff_t pos_in,
>> 3077 struct file *file_out, loff_t pos_out,
>> 3078 size_t *req_count, unsigned int flags)
>> 3079 {
>> 3080 struct inode *inode_in = file_inode(file_in);
>> 3081 struct inode *inode_out = file_inode(file_out);
>> 3082 uint64_t count = *req_count;
>> 3083 loff_t size_in;
>> 3084 int ret;
>> 3085
>> 3086 ret = generic_file_rw_checks(file_in, file_out);
>> 3087 if (ret)
>> 3088 return ret;
>> 3089
>> 3090 /* Don't touch certain kinds of inodes */
>> 3091 if (IS_IMMUTABLE(inode_out))
>> 3092 return -EPERM;
>> 3093
>> 3094 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
>> 3095 return -ETXTBSY;
>> 3096
>> 3097 /* Ensure offsets don't wrap. */
>> 3098 if (pos_in + count < pos_in || pos_out + count < pos_out)
>> 3099 return -EOVERFLOW;
>> 3100
>> 3101 /* Shorten the copy to EOF */
>> 3102 size_in = i_size_read(inode_in);
>> 3103 if (pos_in >= size_in)
>> 3104 count = 0;
>> 3105 else
>> 3106 count = min(count, size_in - (uint64_t)pos_in);
>> 3107
>> 3108 ret = generic_write_check_limits(file_out, pos_out, &count);
>> 3109 if (ret)
>> 3110 return ret;
>> 3111
>> 3112 /* Don't allow overlapped copying within the same file. */
>> 3113 if (inode_in == inode_out &&
>> 3114 pos_out + count > pos_in &&
>> 3115 pos_out < pos_in + count)
>> 3116 return -EINVAL;
>> 3117
>> 3118 *req_count = count;
>> 3119 return 0;
>> 3120 }
>> 3121
>> 3122 int pagecache_write_begin(struct file *file, struct address_space *mapping,
>> 3123 loff_t pos, unsigned len, unsigned flags,
>> 3124 struct page **pagep, void **fsdata)
>> 3125 {
>> 3126 const struct address_space_operations *aops = mapping->a_ops;
>> 3127
>> 3128 return aops->write_begin(file, mapping, pos, len, flags,
>> 3129 pagep, fsdata);
>> 3130 }
>> 3131 EXPORT_SYMBOL(pagecache_write_begin);
>> 3132
>> 3133 int pagecache_write_end(struct file *file, struct address_space *mapping,
>> 3134 loff_t pos, unsigned len, unsigned copied,
>> 3135 struct page *page, void *fsdata)
>> 3136 {
>> 3137 const struct address_space_operations *aops = mapping->a_ops;
>> 3138
>> 3139 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
>> 3140 }
>> 3141 EXPORT_SYMBOL(pagecache_write_end);
>> 3142
>> 3143 /*
3929 * Warn about a page cache invalidation failu 3144 * Warn about a page cache invalidation failure during a direct I/O write.
3930 */ 3145 */
3931 static void dio_warn_stale_pagecache(struct f !! 3146 void dio_warn_stale_pagecache(struct file *filp)
3932 { 3147 {
3933 static DEFINE_RATELIMIT_STATE(_rs, 86 3148 static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3934 char pathname[128]; 3149 char pathname[128];
>> 3150 struct inode *inode = file_inode(filp);
3935 char *path; 3151 char *path;
3936 3152
3937 errseq_set(&filp->f_mapping->wb_err, !! 3153 errseq_set(&inode->i_mapping->wb_err, -EIO);
3938 if (__ratelimit(&_rs)) { 3154 if (__ratelimit(&_rs)) {
3939 path = file_path(filp, pathna 3155 path = file_path(filp, pathname, sizeof(pathname));
3940 if (IS_ERR(path)) 3156 if (IS_ERR(path))
3941 path = "(unknown)"; 3157 path = "(unknown)";
3942 pr_crit("Page cache invalidat 3158 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 3159 pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3944 current->comm); 3160 current->comm);
3945 } 3161 }
3946 } 3162 }
3947 3163
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 3164 ssize_t
3960 generic_file_direct_write(struct kiocb *iocb, 3165 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3961 { 3166 {
3962 struct address_space *mapping = iocb- !! 3167 struct file *file = iocb->ki_filp;
3963 size_t write_len = iov_iter_count(fro !! 3168 struct address_space *mapping = file->f_mapping;
3964 ssize_t written; !! 3169 struct inode *inode = mapping->host;
>> 3170 loff_t pos = iocb->ki_pos;
>> 3171 ssize_t written;
>> 3172 size_t write_len;
>> 3173 pgoff_t end;
>> 3174
>> 3175 write_len = iov_iter_count(from);
>> 3176 end = (pos + write_len - 1) >> PAGE_SHIFT;
>> 3177
>> 3178 if (iocb->ki_flags & IOCB_NOWAIT) {
>> 3179 /* If there are pages to writeback, return */
>> 3180 if (filemap_range_has_page(inode->i_mapping, pos,
>> 3181 pos + write_len - 1))
>> 3182 return -EAGAIN;
>> 3183 } else {
>> 3184 written = filemap_write_and_wait_range(mapping, pos,
>> 3185 pos + write_len - 1);
>> 3186 if (written)
>> 3187 goto out;
>> 3188 }
3965 3189
3966 /* 3190 /*
>> 3191 * After a write we want buffered reads to be sure to go to disk to get
>> 3192 * the new data. We invalidate clean cached page from the region we're
>> 3193 * about to write. We do this *before* the write so that we can return
>> 3194 * without clobbering -EIOCBQUEUED from ->direct_IO().
>> 3195 */
>> 3196 written = invalidate_inode_pages2_range(mapping,
>> 3197 pos >> PAGE_SHIFT, end);
>> 3198 /*
3967 * If a page can not be invalidated, 3199 * If a page can not be invalidated, return 0 to fall back
3968 * to buffered write. 3200 * to buffered write.
3969 */ 3201 */
3970 written = kiocb_invalidate_pages(iocb <<
3971 if (written) { 3202 if (written) {
3972 if (written == -EBUSY) 3203 if (written == -EBUSY)
3973 return 0; 3204 return 0;
3974 return written; !! 3205 goto out;
3975 } 3206 }
3976 3207
3977 written = mapping->a_ops->direct_IO(i 3208 written = mapping->a_ops->direct_IO(iocb, from);
3978 3209
3979 /* 3210 /*
3980 * Finally, try again to invalidate c 3211 * Finally, try again to invalidate clean pages which might have been
3981 * cached by non-direct readahead, or 3212 * cached by non-direct readahead, or faulted in by get_user_pages()
3982 * if the source of the write was an 3213 * if the source of the write was an mmap'ed region of the file
3983 * we're writing. Either one is a pr 3214 * we're writing. Either one is a pretty crazy thing to do,
3984 * so we don't support it 100%. If t 3215 * so we don't support it 100%. If this invalidation
3985 * fails, tough, the write still work 3216 * fails, tough, the write still worked...
3986 * 3217 *
3987 * Most of the time we do not need th 3218 * Most of the time we do not need this since dio_complete() will do
3988 * the invalidation for us. However t 3219 * the invalidation for us. However there are some file systems that
3989 * do not end up with dio_complete() 3220 * do not end up with dio_complete() being called, so let's not break
3990 * them by removing it completely. 3221 * them by removing it completely.
3991 * 3222 *
3992 * Noticeable example is a blkdev_dir 3223 * Noticeable example is a blkdev_direct_IO().
3993 * 3224 *
3994 * Skip invalidation for async writes 3225 * Skip invalidation for async writes or if mapping has no pages.
3995 */ 3226 */
3996 if (written > 0) { !! 3227 if (written > 0 && mapping->nrpages &&
3997 struct inode *inode = mapping !! 3228 invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3998 loff_t pos = iocb->ki_pos; !! 3229 dio_warn_stale_pagecache(file);
3999 3230
4000 kiocb_invalidate_post_direct_ !! 3231 if (written > 0) {
4001 pos += written; 3232 pos += written;
4002 write_len -= written; 3233 write_len -= written;
4003 if (pos > i_size_read(inode) 3234 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
4004 i_size_write(inode, p 3235 i_size_write(inode, pos);
4005 mark_inode_dirty(inod 3236 mark_inode_dirty(inode);
4006 } 3237 }
4007 iocb->ki_pos = pos; 3238 iocb->ki_pos = pos;
4008 } 3239 }
4009 if (written != -EIOCBQUEUED) !! 3240 iov_iter_revert(from, write_len - iov_iter_count(from));
4010 iov_iter_revert(from, write_l !! 3241 out:
4011 return written; 3242 return written;
4012 } 3243 }
4013 EXPORT_SYMBOL(generic_file_direct_write); 3244 EXPORT_SYMBOL(generic_file_direct_write);
4014 3245
4015 ssize_t generic_perform_write(struct kiocb *i !! 3246 /*
>> 3247 * Find or create a page at the given pagecache position. Return the locked
>> 3248 * page. This function is specifically for buffered writes.
>> 3249 */
>> 3250 struct page *grab_cache_page_write_begin(struct address_space *mapping,
>> 3251 pgoff_t index, unsigned flags)
>> 3252 {
>> 3253 struct page *page;
>> 3254 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
>> 3255
>> 3256 if (flags & AOP_FLAG_NOFS)
>> 3257 fgp_flags |= FGP_NOFS;
>> 3258
>> 3259 page = pagecache_get_page(mapping, index, fgp_flags,
>> 3260 mapping_gfp_mask(mapping));
>> 3261 if (page)
>> 3262 wait_for_stable_page(page);
>> 3263
>> 3264 return page;
>> 3265 }
>> 3266 EXPORT_SYMBOL(grab_cache_page_write_begin);
>> 3267
>> 3268 ssize_t generic_perform_write(struct file *file,
>> 3269 struct iov_iter *i, loff_t pos)
4016 { 3270 {
4017 struct file *file = iocb->ki_filp; <<
4018 loff_t pos = iocb->ki_pos; <<
4019 struct address_space *mapping = file- 3271 struct address_space *mapping = file->f_mapping;
4020 const struct address_space_operations 3272 const struct address_space_operations *a_ops = mapping->a_ops;
4021 size_t chunk = mapping_max_folio_size <<
4022 long status = 0; 3273 long status = 0;
4023 ssize_t written = 0; 3274 ssize_t written = 0;
>> 3275 unsigned int flags = 0;
4024 3276
4025 do { 3277 do {
4026 struct folio *folio; !! 3278 struct page *page;
4027 size_t offset; /* Of !! 3279 unsigned long offset; /* Offset into pagecache page */
4028 size_t bytes; /* By !! 3280 unsigned long bytes; /* Bytes to write to page */
4029 size_t copied; /* By 3281 size_t copied; /* Bytes copied from user */
4030 void *fsdata = NULL; !! 3282 void *fsdata;
4031 3283
4032 bytes = iov_iter_count(i); !! 3284 offset = (pos & (PAGE_SIZE - 1));
4033 retry: !! 3285 bytes = min_t(unsigned long, PAGE_SIZE - offset,
4034 offset = pos & (chunk - 1); !! 3286 iov_iter_count(i));
4035 bytes = min(chunk - offset, b <<
4036 balance_dirty_pages_ratelimit <<
4037 3287
>> 3288 again:
4038 /* 3289 /*
4039 * Bring in the user page tha 3290 * Bring in the user page that we will copy from _first_.
4040 * Otherwise there's a nasty 3291 * Otherwise there's a nasty deadlock on copying from the
4041 * same page as we're writing 3292 * same page as we're writing to, without it being marked
4042 * up-to-date. 3293 * up-to-date.
>> 3294 *
>> 3295 * Not only is this an optimisation, but it is also required
>> 3296 * to check that the address is actually valid, when atomic
>> 3297 * usercopies are used, below.
4043 */ 3298 */
4044 if (unlikely(fault_in_iov_ite !! 3299 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
4045 status = -EFAULT; 3300 status = -EFAULT;
4046 break; 3301 break;
4047 } 3302 }
4048 3303
4049 if (fatal_signal_pending(curr 3304 if (fatal_signal_pending(current)) {
4050 status = -EINTR; 3305 status = -EINTR;
4051 break; 3306 break;
4052 } 3307 }
4053 3308
4054 status = a_ops->write_begin(f !! 3309 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
4055 !! 3310 &page, &fsdata);
4056 if (unlikely(status < 0)) 3311 if (unlikely(status < 0))
4057 break; 3312 break;
4058 3313
4059 offset = offset_in_folio(foli <<
4060 if (bytes > folio_size(folio) <<
4061 bytes = folio_size(fo <<
4062 <<
4063 if (mapping_writably_mapped(m 3314 if (mapping_writably_mapped(mapping))
4064 flush_dcache_folio(fo !! 3315 flush_dcache_page(page);
4065 3316
4066 copied = copy_folio_from_iter !! 3317 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
4067 flush_dcache_folio(folio); !! 3318 flush_dcache_page(page);
4068 3319
4069 status = a_ops->write_end(fil 3320 status = a_ops->write_end(file, mapping, pos, bytes, copied,
4070 !! 3321 page, fsdata);
4071 if (unlikely(status != copied !! 3322 if (unlikely(status < 0))
4072 iov_iter_revert(i, co !! 3323 break;
4073 if (unlikely(status < !! 3324 copied = status;
4074 break; !! 3325
4075 } <<
4076 cond_resched(); 3326 cond_resched();
4077 3327
4078 if (unlikely(status == 0)) { !! 3328 iov_iter_advance(i, copied);
>> 3329 if (unlikely(copied == 0)) {
4079 /* 3330 /*
4080 * A short copy made !! 3331 * If we were unable to copy any data at all, we must
4081 * thing entirely. M !! 3332 * fall back to a single segment length write.
4082 * halfway through, m !! 3333 *
4083 * might be severe me !! 3334 * If we didn't fallback here, we could livelock
>> 3335 * because not all segments in the iov can be copied at
>> 3336 * once without a pagefault.
4084 */ 3337 */
4085 if (chunk > PAGE_SIZE !! 3338 bytes = min_t(unsigned long, PAGE_SIZE - offset,
4086 chunk /= 2; !! 3339 iov_iter_single_seg_count(i));
4087 if (copied) { !! 3340 goto again;
4088 bytes = copie <<
4089 goto retry; <<
4090 } <<
4091 } else { <<
4092 pos += status; <<
4093 written += status; <<
4094 } 3341 }
>> 3342 pos += copied;
>> 3343 written += copied;
>> 3344
>> 3345 balance_dirty_pages_ratelimited(mapping);
4095 } while (iov_iter_count(i)); 3346 } while (iov_iter_count(i));
4096 3347
4097 if (!written) !! 3348 return written ? written : status;
4098 return status; <<
4099 iocb->ki_pos += written; <<
4100 return written; <<
4101 } 3349 }
4102 EXPORT_SYMBOL(generic_perform_write); 3350 EXPORT_SYMBOL(generic_perform_write);
4103 3351
4104 /** 3352 /**
4105 * __generic_file_write_iter - write data to 3353 * __generic_file_write_iter - write data to a file
4106 * @iocb: IO state structure (file, off 3354 * @iocb: IO state structure (file, offset, etc.)
4107 * @from: iov_iter with data to write 3355 * @from: iov_iter with data to write
4108 * 3356 *
4109 * This function does all the work needed for 3357 * This function does all the work needed for actually writing data to a
4110 * file. It does all basic checks, removes SU 3358 * file. It does all basic checks, removes SUID from the file, updates
4111 * modification times and calls proper subrou 3359 * modification times and calls proper subroutines depending on whether we
4112 * do direct IO or a standard buffered write. 3360 * do direct IO or a standard buffered write.
4113 * 3361 *
4114 * It expects i_rwsem to be grabbed unless we !! 3362 * 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. 3363 * object which does not need locking at all.
4116 * 3364 *
4117 * This function does *not* take care of sync 3365 * 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 3366 * A caller has to handle it. This is mainly due to the fact that we want to
4119 * avoid syncing under i_rwsem. !! 3367 * avoid syncing under i_mutex.
4120 * 3368 *
4121 * Return: 3369 * Return:
4122 * * number of bytes written, even for trunca 3370 * * number of bytes written, even for truncated writes
4123 * * negative error code if no data has been 3371 * * negative error code if no data has been written at all
4124 */ 3372 */
4125 ssize_t __generic_file_write_iter(struct kioc 3373 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4126 { 3374 {
4127 struct file *file = iocb->ki_filp; 3375 struct file *file = iocb->ki_filp;
4128 struct address_space *mapping = file- !! 3376 struct address_space * mapping = file->f_mapping;
4129 struct inode *inode = mapping->host; !! 3377 struct inode *inode = mapping->host;
4130 ssize_t ret; !! 3378 ssize_t written = 0;
4131 !! 3379 ssize_t err;
4132 ret = file_remove_privs(file); !! 3380 ssize_t status;
4133 if (ret) !! 3381
4134 return ret; !! 3382 /* We can write back this queue in page reclaim */
>> 3383 current->backing_dev_info = inode_to_bdi(inode);
>> 3384 err = file_remove_privs(file);
>> 3385 if (err)
>> 3386 goto out;
4135 3387
4136 ret = file_update_time(file); !! 3388 err = file_update_time(file);
4137 if (ret) !! 3389 if (err)
4138 return ret; !! 3390 goto out;
4139 3391
4140 if (iocb->ki_flags & IOCB_DIRECT) { 3392 if (iocb->ki_flags & IOCB_DIRECT) {
4141 ret = generic_file_direct_wri !! 3393 loff_t pos, endbyte;
>> 3394
>> 3395 written = generic_file_direct_write(iocb, from);
4142 /* 3396 /*
4143 * If the write stopped short 3397 * If the write stopped short of completing, fall back to
4144 * buffered writes. Some fil 3398 * buffered writes. Some filesystems do this for writes to
4145 * holes, for example. For D 3399 * holes, for example. For DAX files, a buffered write will
4146 * not succeed (even if it di 3400 * not succeed (even if it did, DAX does not handle dirty
4147 * page-cache pages correctly 3401 * page-cache pages correctly).
4148 */ 3402 */
4149 if (ret < 0 || !iov_iter_coun !! 3403 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
4150 return ret; !! 3404 goto out;
4151 return direct_write_fallback( <<
4152 generic_perfo <<
4153 } <<
4154 3405
4155 return generic_perform_write(iocb, fr !! 3406 status = generic_perform_write(file, from, pos = iocb->ki_pos);
>> 3407 /*
>> 3408 * If generic_perform_write() returned a synchronous error
>> 3409 * then we want to return the number of bytes which were
>> 3410 * direct-written, or the error code if that was zero. Note
>> 3411 * that this differs from normal direct-io semantics, which
>> 3412 * will return -EFOO even if some bytes were written.
>> 3413 */
>> 3414 if (unlikely(status < 0)) {
>> 3415 err = status;
>> 3416 goto out;
>> 3417 }
>> 3418 /*
>> 3419 * We need to ensure that the page cache pages are written to
>> 3420 * disk and invalidated to preserve the expected O_DIRECT
>> 3421 * semantics.
>> 3422 */
>> 3423 endbyte = pos + status - 1;
>> 3424 err = filemap_write_and_wait_range(mapping, pos, endbyte);
>> 3425 if (err == 0) {
>> 3426 iocb->ki_pos = endbyte + 1;
>> 3427 written += status;
>> 3428 invalidate_mapping_pages(mapping,
>> 3429 pos >> PAGE_SHIFT,
>> 3430 endbyte >> PAGE_SHIFT);
>> 3431 } else {
>> 3432 /*
>> 3433 * We don't know how much we wrote, so just return
>> 3434 * the number of bytes which were direct-written
>> 3435 */
>> 3436 }
>> 3437 } else {
>> 3438 written = generic_perform_write(file, from, iocb->ki_pos);
>> 3439 if (likely(written > 0))
>> 3440 iocb->ki_pos += written;
>> 3441 }
>> 3442 out:
>> 3443 current->backing_dev_info = NULL;
>> 3444 return written ? written : err;
4156 } 3445 }
4157 EXPORT_SYMBOL(__generic_file_write_iter); 3446 EXPORT_SYMBOL(__generic_file_write_iter);
4158 3447
4159 /** 3448 /**
4160 * generic_file_write_iter - write data to a 3449 * generic_file_write_iter - write data to a file
4161 * @iocb: IO state structure 3450 * @iocb: IO state structure
4162 * @from: iov_iter with data to write 3451 * @from: iov_iter with data to write
4163 * 3452 *
4164 * This is a wrapper around __generic_file_wr 3453 * This is a wrapper around __generic_file_write_iter() to be used by most
4165 * filesystems. It takes care of syncing the 3454 * filesystems. It takes care of syncing the file in case of O_SYNC file
4166 * and acquires i_rwsem as needed. !! 3455 * and acquires i_mutex as needed.
4167 * Return: 3456 * Return:
4168 * * negative error code if no data has been 3457 * * negative error code if no data has been written at all of
4169 * vfs_fsync_range() failed for a synchrono 3458 * vfs_fsync_range() failed for a synchronous write
4170 * * number of bytes written, even for trunca 3459 * * number of bytes written, even for truncated writes
4171 */ 3460 */
4172 ssize_t generic_file_write_iter(struct kiocb 3461 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4173 { 3462 {
4174 struct file *file = iocb->ki_filp; 3463 struct file *file = iocb->ki_filp;
4175 struct inode *inode = file->f_mapping 3464 struct inode *inode = file->f_mapping->host;
4176 ssize_t ret; 3465 ssize_t ret;
4177 3466
4178 inode_lock(inode); 3467 inode_lock(inode);
4179 ret = generic_write_checks(iocb, from 3468 ret = generic_write_checks(iocb, from);
4180 if (ret > 0) 3469 if (ret > 0)
4181 ret = __generic_file_write_it 3470 ret = __generic_file_write_iter(iocb, from);
4182 inode_unlock(inode); 3471 inode_unlock(inode);
4183 3472
4184 if (ret > 0) 3473 if (ret > 0)
4185 ret = generic_write_sync(iocb 3474 ret = generic_write_sync(iocb, ret);
4186 return ret; 3475 return ret;
4187 } 3476 }
4188 EXPORT_SYMBOL(generic_file_write_iter); 3477 EXPORT_SYMBOL(generic_file_write_iter);
4189 3478
4190 /** 3479 /**
4191 * filemap_release_folio() - Release fs-speci !! 3480 * 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 * 3481 *
4195 * The address_space is trying to release any !! 3482 * @page: the page which the kernel is trying to free
4196 * (presumably at folio->private). !! 3483 * @gfp_mask: memory allocation flags (and I/O mode)
4197 * 3484 *
4198 * This will also be called if the private_2 !! 3485 * The address_space is to try to release any data against the page
4199 * indicating that the folio has other metada !! 3486 * (presumably at page->private).
4200 * 3487 *
4201 * The @gfp argument specifies whether I/O ma !! 3488 * 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 !! 3489 * page is known to the local caching routines.
4203 * (__GFP_RECLAIM & __GFP_FS). <<
4204 * 3490 *
4205 * Return: %true if the release was successfu !! 3491 * The @gfp_mask argument specifies whether I/O may be performed to release
4206 */ !! 3492 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
4207 bool filemap_release_folio(struct folio *foli !! 3493 *
4208 { !! 3494 * 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 */ 3495 */
4290 static void filemap_cachestat(struct address_ !! 3496 int try_to_release_page(struct page *page, gfp_t gfp_mask)
4291 pgoff_t first_index, pgoff_t <<
4292 { 3497 {
4293 XA_STATE(xas, &mapping->i_pages, firs !! 3498 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 3499
4316 if (xas_retry(&xas, folio)) !! 3500 BUG_ON(!PageLocked(page));
4317 continue; !! 3501 if (PageWriteback(page))
4318 !! 3502 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 3503
4377 resched: !! 3504 if (mapping && mapping->a_ops->releasepage)
4378 if (need_resched()) { !! 3505 return mapping->a_ops->releasepage(page, gfp_mask);
4379 xas_pause(&xas); !! 3506 return try_to_free_buffers(page);
4380 cond_resched_rcu(); <<
4381 } <<
4382 } <<
4383 rcu_read_unlock(); <<
4384 } 3507 }
4385 3508
4386 /* !! 3509 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 3510
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.