1 // SPDX-License-Identifier: GPL-2.0-only <<
2 /* 1 /*
3 * linux/mm/filemap.c 2 * linux/mm/filemap.c
4 * 3 *
5 * Copyright (C) 1994-1999 Linus Torvalds 4 * Copyright (C) 1994-1999 Linus Torvalds
6 */ 5 */
7 6
8 /* 7 /*
9 * This file handles the generic file mmap sem 8 * This file handles the generic file mmap semantics used by
10 * most "normal" filesystems (but you don't /h 9 * most "normal" filesystems (but you don't /have/ to use this:
11 * the NFS filesystem used to do this differen 10 * the NFS filesystem used to do this differently, for example)
12 */ 11 */
13 #include <linux/export.h> 12 #include <linux/export.h>
14 #include <linux/compiler.h> 13 #include <linux/compiler.h>
15 #include <linux/dax.h> 14 #include <linux/dax.h>
16 #include <linux/fs.h> 15 #include <linux/fs.h>
17 #include <linux/sched/signal.h> 16 #include <linux/sched/signal.h>
18 #include <linux/uaccess.h> 17 #include <linux/uaccess.h>
19 #include <linux/capability.h> 18 #include <linux/capability.h>
20 #include <linux/kernel_stat.h> 19 #include <linux/kernel_stat.h>
21 #include <linux/gfp.h> 20 #include <linux/gfp.h>
22 #include <linux/mm.h> 21 #include <linux/mm.h>
23 #include <linux/swap.h> 22 #include <linux/swap.h>
24 #include <linux/swapops.h> <<
25 #include <linux/syscalls.h> <<
26 #include <linux/mman.h> 23 #include <linux/mman.h>
27 #include <linux/pagemap.h> 24 #include <linux/pagemap.h>
28 #include <linux/file.h> 25 #include <linux/file.h>
29 #include <linux/uio.h> 26 #include <linux/uio.h>
30 #include <linux/error-injection.h> <<
31 #include <linux/hash.h> 27 #include <linux/hash.h>
32 #include <linux/writeback.h> 28 #include <linux/writeback.h>
33 #include <linux/backing-dev.h> 29 #include <linux/backing-dev.h>
34 #include <linux/pagevec.h> 30 #include <linux/pagevec.h>
>> 31 #include <linux/blkdev.h>
35 #include <linux/security.h> 32 #include <linux/security.h>
36 #include <linux/cpuset.h> 33 #include <linux/cpuset.h>
>> 34 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
37 #include <linux/hugetlb.h> 35 #include <linux/hugetlb.h>
38 #include <linux/memcontrol.h> 36 #include <linux/memcontrol.h>
39 #include <linux/shmem_fs.h> !! 37 #include <linux/cleancache.h>
40 #include <linux/rmap.h> 38 #include <linux/rmap.h>
41 #include <linux/delayacct.h> <<
42 #include <linux/psi.h> <<
43 #include <linux/ramfs.h> <<
44 #include <linux/page_idle.h> <<
45 #include <linux/migrate.h> <<
46 #include <linux/pipe_fs_i.h> <<
47 #include <linux/splice.h> <<
48 #include <linux/rcupdate_wait.h> <<
49 #include <linux/sched/mm.h> <<
50 #include <asm/pgalloc.h> <<
51 #include <asm/tlbflush.h> <<
52 #include "internal.h" 39 #include "internal.h"
53 40
54 #define CREATE_TRACE_POINTS 41 #define CREATE_TRACE_POINTS
55 #include <trace/events/filemap.h> 42 #include <trace/events/filemap.h>
56 43
57 /* 44 /*
58 * FIXME: remove all knowledge of the buffer l 45 * FIXME: remove all knowledge of the buffer layer from the core VM
59 */ 46 */
60 #include <linux/buffer_head.h> /* for try_to_f 47 #include <linux/buffer_head.h> /* for try_to_free_buffers */
61 48
62 #include <asm/mman.h> 49 #include <asm/mman.h>
63 50
64 #include "swap.h" <<
65 <<
66 /* 51 /*
67 * Shared mappings implemented 30.11.1994. It' 52 * Shared mappings implemented 30.11.1994. It's not fully working yet,
68 * though. 53 * though.
69 * 54 *
70 * Shared mappings now work. 15.8.1995 Bruno. 55 * Shared mappings now work. 15.8.1995 Bruno.
71 * 56 *
72 * finished 'unifying' the page and buffer cac 57 * finished 'unifying' the page and buffer cache and SMP-threaded the
73 * page-cache, 21.05.1999, Ingo Molnar <mingo@ 58 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
74 * 59 *
75 * SMP-threaded pagemap-LRU 1999, Andrea Arcan 60 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
76 */ 61 */
77 62
78 /* 63 /*
79 * Lock ordering: 64 * Lock ordering:
80 * 65 *
81 * ->i_mmap_rwsem (truncate_page 66 * ->i_mmap_rwsem (truncate_pagecache)
82 * ->private_lock (__free_pte->b !! 67 * ->private_lock (__free_pte->__set_page_dirty_buffers)
83 * ->swap_lock (exclusive_swa 68 * ->swap_lock (exclusive_swap_page, others)
84 * ->i_pages lock !! 69 * ->mapping->tree_lock
85 * 70 *
86 * ->i_rwsem !! 71 * ->i_mutex
87 * ->invalidate_lock (acquired by f !! 72 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
88 * ->i_mmap_rwsem (truncate->unm <<
89 * 73 *
90 * ->mmap_lock !! 74 * ->mmap_sem
91 * ->i_mmap_rwsem 75 * ->i_mmap_rwsem
92 * ->page_table_lock or pte_lock (vario 76 * ->page_table_lock or pte_lock (various, mainly in memory.c)
93 * ->i_pages lock (arch-dependen !! 77 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
94 * 78 *
95 * ->mmap_lock !! 79 * ->mmap_sem
96 * ->invalidate_lock (filemap_fault !! 80 * ->lock_page (access_process_vm)
97 * ->lock_page (filemap_fault <<
98 * 81 *
99 * ->i_rwsem (generic_perfo !! 82 * ->i_mutex (generic_perform_write)
100 * ->mmap_lock (fault_in_read !! 83 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
101 * 84 *
102 * bdi->wb.list_lock 85 * bdi->wb.list_lock
103 * sb_lock (fs/fs-writeba 86 * sb_lock (fs/fs-writeback.c)
104 * ->i_pages lock (__sync_single !! 87 * ->mapping->tree_lock (__sync_single_inode)
105 * 88 *
106 * ->i_mmap_rwsem 89 * ->i_mmap_rwsem
107 * ->anon_vma.lock (vma_merge) !! 90 * ->anon_vma.lock (vma_adjust)
108 * 91 *
109 * ->anon_vma.lock 92 * ->anon_vma.lock
110 * ->page_table_lock or pte_lock (anon_ 93 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
111 * 94 *
112 * ->page_table_lock or pte_lock 95 * ->page_table_lock or pte_lock
113 * ->swap_lock (try_to_unmap_ 96 * ->swap_lock (try_to_unmap_one)
114 * ->private_lock (try_to_unmap_ 97 * ->private_lock (try_to_unmap_one)
115 * ->i_pages lock (try_to_unmap_ !! 98 * ->tree_lock (try_to_unmap_one)
116 * ->lruvec->lru_lock (follow_page_m !! 99 * ->zone_lru_lock(zone) (follow_page->mark_page_accessed)
117 * ->lruvec->lru_lock (check_pte_ran !! 100 * ->zone_lru_lock(zone) (check_pte_range->isolate_lru_page)
118 * ->private_lock (folio_remove_ !! 101 * ->private_lock (page_remove_rmap->set_page_dirty)
119 * ->i_pages lock (folio_remove_ !! 102 * ->tree_lock (page_remove_rmap->set_page_dirty)
120 * bdi.wb->list_lock (folio_remove_ !! 103 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
121 * ->inode->i_lock (folio_remove_ !! 104 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
122 * ->memcg->move_lock (folio_remove_ !! 105 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
123 * bdi.wb->list_lock (zap_pte_range 106 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
124 * ->inode->i_lock (zap_pte_range 107 * ->inode->i_lock (zap_pte_range->set_page_dirty)
125 * ->private_lock (zap_pte_range !! 108 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
>> 109 *
>> 110 * ->i_mmap_rwsem
>> 111 * ->tasklist_lock (memory_failure, collect_procs_ao)
126 */ 112 */
127 113
128 static void mapping_set_update(struct xa_state !! 114 static int page_cache_tree_insert(struct address_space *mapping,
129 struct address_space *mapping) !! 115 struct page *page, void **shadowp)
130 { 116 {
131 if (dax_mapping(mapping) || shmem_mapp !! 117 struct radix_tree_node *node;
132 return; !! 118 void **slot;
133 xas_set_update(xas, workingset_update_ !! 119 int error;
134 xas_set_lru(xas, &shadow_nodes); !! 120
>> 121 error = __radix_tree_create(&mapping->page_tree, page->index, 0,
>> 122 &node, &slot);
>> 123 if (error)
>> 124 return error;
>> 125 if (*slot) {
>> 126 void *p;
>> 127
>> 128 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
>> 129 if (!radix_tree_exceptional_entry(p))
>> 130 return -EEXIST;
>> 131
>> 132 mapping->nrexceptional--;
>> 133 if (!dax_mapping(mapping)) {
>> 134 if (shadowp)
>> 135 *shadowp = p;
>> 136 } else {
>> 137 /* DAX can replace empty locked entry with a hole */
>> 138 WARN_ON_ONCE(p !=
>> 139 dax_radix_locked_entry(0, RADIX_DAX_EMPTY));
>> 140 /* Wakeup waiters for exceptional entry lock */
>> 141 dax_wake_mapping_entry_waiter(mapping, page->index, p,
>> 142 true);
>> 143 }
>> 144 }
>> 145 __radix_tree_replace(&mapping->page_tree, node, slot, page,
>> 146 workingset_update_node, mapping);
>> 147 mapping->nrpages++;
>> 148 return 0;
135 } 149 }
136 150
137 static void page_cache_delete(struct address_s !! 151 static void page_cache_tree_delete(struct address_space *mapping,
138 struct foli !! 152 struct page *page, void *shadow)
139 { 153 {
140 XA_STATE(xas, &mapping->i_pages, folio !! 154 int i, nr;
141 long nr = 1; <<
142 155
143 mapping_set_update(&xas, mapping); !! 156 /* hugetlb pages are represented by one entry in the radix tree */
>> 157 nr = PageHuge(page) ? 1 : hpage_nr_pages(page);
144 158
145 xas_set_order(&xas, folio->index, foli !! 159 VM_BUG_ON_PAGE(!PageLocked(page), page);
146 nr = folio_nr_pages(folio); !! 160 VM_BUG_ON_PAGE(PageTail(page), page);
>> 161 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
147 162
148 VM_BUG_ON_FOLIO(!folio_test_locked(fol !! 163 for (i = 0; i < nr; i++) {
>> 164 struct radix_tree_node *node;
>> 165 void **slot;
149 166
150 xas_store(&xas, shadow); !! 167 __radix_tree_lookup(&mapping->page_tree, page->index + i,
151 xas_init_marks(&xas); !! 168 &node, &slot);
152 169
153 folio->mapping = NULL; !! 170 VM_BUG_ON_PAGE(!node && nr != 1, page);
154 /* Leave page->index set: truncation l !! 171
>> 172 radix_tree_clear_tags(&mapping->page_tree, node, slot);
>> 173 __radix_tree_replace(&mapping->page_tree, node, slot, shadow,
>> 174 workingset_update_node, mapping);
>> 175 }
>> 176
>> 177 if (shadow) {
>> 178 mapping->nrexceptional += nr;
>> 179 /*
>> 180 * Make sure the nrexceptional update is committed before
>> 181 * the nrpages update so that final truncate racing
>> 182 * with reclaim does not see both counters 0 at the
>> 183 * same time and miss a shadow entry.
>> 184 */
>> 185 smp_wmb();
>> 186 }
155 mapping->nrpages -= nr; 187 mapping->nrpages -= nr;
156 } 188 }
157 189
158 static void filemap_unaccount_folio(struct add !! 190 /*
159 struct folio *folio) !! 191 * Delete a page from the page cache and free it. Caller has to make
>> 192 * sure the page is locked and that nobody else uses it - or that usage
>> 193 * is safe. The caller must hold the mapping's tree_lock.
>> 194 */
>> 195 void __delete_from_page_cache(struct page *page, void *shadow)
160 { 196 {
161 long nr; !! 197 struct address_space *mapping = page->mapping;
>> 198 int nr = hpage_nr_pages(page);
>> 199
>> 200 trace_mm_filemap_delete_from_page_cache(page);
>> 201 /*
>> 202 * if we're uptodate, flush out into the cleancache, otherwise
>> 203 * invalidate any existing cleancache entries. We can't leave
>> 204 * stale data around in the cleancache once our page is gone
>> 205 */
>> 206 if (PageUptodate(page) && PageMappedToDisk(page))
>> 207 cleancache_put_page(page);
>> 208 else
>> 209 cleancache_invalidate_page(mapping, page);
>> 210
>> 211 VM_BUG_ON_PAGE(PageTail(page), page);
>> 212 VM_BUG_ON_PAGE(page_mapped(page), page);
>> 213 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
>> 214 int mapcount;
162 215
163 VM_BUG_ON_FOLIO(folio_mapped(folio), f <<
164 if (!IS_ENABLED(CONFIG_DEBUG_VM) && un <<
165 pr_alert("BUG: Bad page cache 216 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
166 current->comm, folio_ !! 217 current->comm, page_to_pfn(page));
167 dump_page(&folio->page, "still !! 218 dump_page(page, "still mapped when deleted");
168 dump_stack(); 219 dump_stack();
169 add_taint(TAINT_BAD_PAGE, LOCK 220 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
170 221
171 if (mapping_exiting(mapping) & !! 222 mapcount = page_mapcount(page);
172 int mapcount = folio_m !! 223 if (mapping_exiting(mapping) &&
173 !! 224 page_count(page) >= mapcount + 2) {
174 if (folio_ref_count(fo !! 225 /*
175 /* !! 226 * All vmas have already been torn down, so it's
176 * All vmas ha !! 227 * a good bet that actually the page is unmapped,
177 * a good bet !! 228 * and we'd prefer not to leak it: if we're wrong,
178 * and we'd ra !! 229 * some other bad page check should catch it later.
179 * another bad !! 230 */
180 */ !! 231 page_mapcount_reset(page);
181 atomic_set(&fo !! 232 page_ref_sub(page, mapcount);
182 folio_ref_sub( <<
183 } <<
184 } 233 }
185 } 234 }
186 235
187 /* hugetlb folios do not participate i !! 236 page_cache_tree_delete(mapping, page, shadow);
188 if (folio_test_hugetlb(folio)) <<
189 return; <<
190 237
191 nr = folio_nr_pages(folio); !! 238 page->mapping = NULL;
>> 239 /* Leave page->index set: truncation lookup relies upon it */
192 240
193 __lruvec_stat_mod_folio(folio, NR_FILE !! 241 /* hugetlb pages do not participate in page cache accounting. */
194 if (folio_test_swapbacked(folio)) { !! 242 if (!PageHuge(page))
195 __lruvec_stat_mod_folio(folio, !! 243 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
196 if (folio_test_pmd_mappable(fo !! 244 if (PageSwapBacked(page)) {
197 __lruvec_stat_mod_foli !! 245 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
198 } else if (folio_test_pmd_mappable(fol !! 246 if (PageTransHuge(page))
199 __lruvec_stat_mod_folio(folio, !! 247 __dec_node_page_state(page, NR_SHMEM_THPS);
200 filemap_nr_thps_dec(mapping); !! 248 } else {
>> 249 VM_BUG_ON_PAGE(PageTransHuge(page) && !PageHuge(page), page);
201 } 250 }
202 251
203 /* 252 /*
204 * At this point folio must be either !! 253 * At this point page must be either written or cleaned by truncate.
205 * truncate. Dirty folio here signals !! 254 * Dirty page here signals a bug and loss of unwritten data.
206 * unwritten data - on ordinary filesy <<
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 * 255 *
212 * Below fixes dirty accounting after !! 256 * This fixes dirty accounting after removing the page entirely but
213 * but leaves the dirty flag set: it h !! 257 * leaves PageDirty set: it has no effect for truncated page and
214 * folio and anyway will be cleared be !! 258 * anyway will be cleared before returning page into buddy allocator.
215 * buddy allocator. <<
216 */ 259 */
217 if (WARN_ON_ONCE(folio_test_dirty(foli !! 260 if (WARN_ON_ONCE(PageDirty(page)))
218 mapping_can_writeback !! 261 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
219 folio_account_cleaned(folio, i <<
220 } <<
221 <<
222 /* <<
223 * Delete a page from the page cache and free <<
224 * sure the page is locked and that nobody els <<
225 * is safe. The caller must hold the i_pages <<
226 */ <<
227 void __filemap_remove_folio(struct folio *foli <<
228 { <<
229 struct address_space *mapping = folio- <<
230 <<
231 trace_mm_filemap_delete_from_page_cach <<
232 filemap_unaccount_folio(mapping, folio <<
233 page_cache_delete(mapping, folio, shad <<
234 } <<
235 <<
236 void filemap_free_folio(struct address_space * <<
237 { <<
238 void (*free_folio)(struct folio *); <<
239 int refs = 1; <<
240 <<
241 free_folio = mapping->a_ops->free_foli <<
242 if (free_folio) <<
243 free_folio(folio); <<
244 <<
245 if (folio_test_large(folio)) <<
246 refs = folio_nr_pages(folio); <<
247 folio_put_refs(folio, refs); <<
248 } 262 }
249 263
250 /** 264 /**
251 * filemap_remove_folio - Remove folio from pa !! 265 * delete_from_page_cache - delete page from page cache
252 * @folio: The folio. !! 266 * @page: the page which the kernel is trying to remove from page cache
253 * 267 *
254 * This must be called only on folios that are !! 268 * 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 !! 269 * 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 !! 270 * has a reference on the page.
257 */ !! 271 */
258 void filemap_remove_folio(struct folio *folio) !! 272 void delete_from_page_cache(struct page *page)
259 { !! 273 {
260 struct address_space *mapping = folio- !! 274 struct address_space *mapping = page_mapping(page);
261 !! 275 unsigned long flags;
262 BUG_ON(!folio_test_locked(folio)); !! 276 void (*freepage)(struct page *);
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); <<
270 277
271 filemap_free_folio(mapping, folio); !! 278 BUG_ON(!PageLocked(page));
272 } <<
273 279
274 /* !! 280 freepage = mapping->a_ops->freepage;
275 * page_cache_delete_batch - delete several fo <<
276 * @mapping: the mapping to which folios belon <<
277 * @fbatch: batch of folios to delete <<
278 * <<
279 * The function walks over mapping->i_pages an <<
280 * @fbatch from the mapping. The function expe <<
281 * by page index and is optimised for it to be <<
282 * It tolerates holes in @fbatch (mapping entr <<
283 * modified). <<
284 * <<
285 * The function expects the i_pages lock to be <<
286 */ <<
287 static void page_cache_delete_batch(struct add <<
288 struct folio_batc <<
289 { <<
290 XA_STATE(xas, &mapping->i_pages, fbatc <<
291 long total_pages = 0; <<
292 int i = 0; <<
293 struct folio *folio; <<
294 <<
295 mapping_set_update(&xas, mapping); <<
296 xas_for_each(&xas, folio, ULONG_MAX) { <<
297 if (i >= folio_batch_count(fba <<
298 break; <<
299 281
300 /* A swap/dax/shadow entry got !! 282 spin_lock_irqsave(&mapping->tree_lock, flags);
301 if (xa_is_value(folio)) !! 283 __delete_from_page_cache(page, NULL);
302 continue; !! 284 spin_unlock_irqrestore(&mapping->tree_lock, flags);
303 /* <<
304 * A page got inserted in our <<
305 * pages locked so they are pr <<
306 * If we see a page whose inde <<
307 * means our page has been rem <<
308 * possible because we're hold <<
309 */ <<
310 if (folio != fbatch->folios[i] <<
311 VM_BUG_ON_FOLIO(folio- <<
312 fbatch <<
313 continue; <<
314 } <<
315 285
316 WARN_ON_ONCE(!folio_test_locke !! 286 if (freepage)
>> 287 freepage(page);
317 288
318 folio->mapping = NULL; !! 289 if (PageTransHuge(page) && !PageHuge(page)) {
319 /* Leave folio->index set: tru !! 290 page_ref_sub(page, HPAGE_PMD_NR);
320 !! 291 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
321 i++; !! 292 } else {
322 xas_store(&xas, NULL); !! 293 put_page(page);
323 total_pages += folio_nr_pages( <<
324 } 294 }
325 mapping->nrpages -= total_pages; <<
326 } <<
327 <<
328 void delete_from_page_cache_batch(struct addre <<
329 struct folio <<
330 { <<
331 int i; <<
332 <<
333 if (!folio_batch_count(fbatch)) <<
334 return; <<
335 <<
336 spin_lock(&mapping->host->i_lock); <<
337 xa_lock_irq(&mapping->i_pages); <<
338 for (i = 0; i < folio_batch_count(fbat <<
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 <<
350 for (i = 0; i < folio_batch_count(fbat <<
351 filemap_free_folio(mapping, fb <<
352 } 295 }
>> 296 EXPORT_SYMBOL(delete_from_page_cache);
353 297
354 int filemap_check_errors(struct address_space 298 int filemap_check_errors(struct address_space *mapping)
355 { 299 {
356 int ret = 0; 300 int ret = 0;
357 /* Check for outstanding write errors 301 /* Check for outstanding write errors */
358 if (test_bit(AS_ENOSPC, &mapping->flag 302 if (test_bit(AS_ENOSPC, &mapping->flags) &&
359 test_and_clear_bit(AS_ENOSPC, &map 303 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
360 ret = -ENOSPC; 304 ret = -ENOSPC;
361 if (test_bit(AS_EIO, &mapping->flags) 305 if (test_bit(AS_EIO, &mapping->flags) &&
362 test_and_clear_bit(AS_EIO, &mappin 306 test_and_clear_bit(AS_EIO, &mapping->flags))
363 ret = -EIO; 307 ret = -EIO;
364 return ret; 308 return ret;
365 } 309 }
366 EXPORT_SYMBOL(filemap_check_errors); 310 EXPORT_SYMBOL(filemap_check_errors);
367 311
368 static int filemap_check_and_keep_errors(struc <<
369 { <<
370 /* Check for outstanding write errors <<
371 if (test_bit(AS_EIO, &mapping->flags)) <<
372 return -EIO; <<
373 if (test_bit(AS_ENOSPC, &mapping->flag <<
374 return -ENOSPC; <<
375 return 0; <<
376 } <<
377 <<
378 /** <<
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 /** 312 /**
405 * __filemap_fdatawrite_range - start writebac 313 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
406 * @mapping: address space structure to wri 314 * @mapping: address space structure to write
407 * @start: offset in bytes where the rang 315 * @start: offset in bytes where the range starts
408 * @end: offset in bytes where the rang 316 * @end: offset in bytes where the range ends (inclusive)
409 * @sync_mode: enable synchronous operation 317 * @sync_mode: enable synchronous operation
410 * 318 *
411 * Start writeback against all of a mapping's 319 * Start writeback against all of a mapping's dirty pages that lie
412 * within the byte offsets <start, end> inclus 320 * within the byte offsets <start, end> inclusive.
413 * 321 *
414 * If sync_mode is WB_SYNC_ALL then this is a 322 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
415 * opposed to a regular memory cleansing write 323 * opposed to a regular memory cleansing writeback. The difference between
416 * these two operations is that if a dirty pag 324 * these two operations is that if a dirty page/buffer is encountered, it must
417 * be waited upon, and not just skipped over. 325 * be waited upon, and not just skipped over.
418 * <<
419 * Return: %0 on success, negative error code <<
420 */ 326 */
421 int __filemap_fdatawrite_range(struct address_ 327 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
422 loff_t end, in 328 loff_t end, int sync_mode)
423 { 329 {
>> 330 int ret;
424 struct writeback_control wbc = { 331 struct writeback_control wbc = {
425 .sync_mode = sync_mode, 332 .sync_mode = sync_mode,
426 .nr_to_write = LONG_MAX, 333 .nr_to_write = LONG_MAX,
427 .range_start = start, 334 .range_start = start,
428 .range_end = end, 335 .range_end = end,
429 }; 336 };
430 337
431 return filemap_fdatawrite_wbc(mapping, !! 338 if (!mapping_cap_writeback_dirty(mapping))
>> 339 return 0;
>> 340
>> 341 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
>> 342 ret = do_writepages(mapping, &wbc);
>> 343 wbc_detach_inode(&wbc);
>> 344 return ret;
432 } 345 }
433 346
434 static inline int __filemap_fdatawrite(struct 347 static inline int __filemap_fdatawrite(struct address_space *mapping,
435 int sync_mode) 348 int sync_mode)
436 { 349 {
437 return __filemap_fdatawrite_range(mapp 350 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
438 } 351 }
439 352
440 int filemap_fdatawrite(struct address_space *m 353 int filemap_fdatawrite(struct address_space *mapping)
441 { 354 {
442 return __filemap_fdatawrite(mapping, W 355 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
443 } 356 }
444 EXPORT_SYMBOL(filemap_fdatawrite); 357 EXPORT_SYMBOL(filemap_fdatawrite);
445 358
446 int filemap_fdatawrite_range(struct address_sp 359 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
447 loff_t end) 360 loff_t end)
448 { 361 {
449 return __filemap_fdatawrite_range(mapp 362 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
450 } 363 }
451 EXPORT_SYMBOL(filemap_fdatawrite_range); 364 EXPORT_SYMBOL(filemap_fdatawrite_range);
452 365
453 /** 366 /**
454 * filemap_flush - mostly a non-blocking flush 367 * filemap_flush - mostly a non-blocking flush
455 * @mapping: target address_space 368 * @mapping: target address_space
456 * 369 *
457 * This is a mostly non-blocking flush. Not s 370 * This is a mostly non-blocking flush. Not suitable for data-integrity
458 * purposes - I/O may not be started against a 371 * purposes - I/O may not be started against all dirty pages.
459 * <<
460 * Return: %0 on success, negative error code <<
461 */ 372 */
462 int filemap_flush(struct address_space *mappin 373 int filemap_flush(struct address_space *mapping)
463 { 374 {
464 return __filemap_fdatawrite(mapping, W 375 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
465 } 376 }
466 EXPORT_SYMBOL(filemap_flush); 377 EXPORT_SYMBOL(filemap_flush);
467 378
468 /** !! 379 static int __filemap_fdatawait_range(struct address_space *mapping,
469 * filemap_range_has_page - check if a page ex <<
470 * @mapping: address space within wh <<
471 * @start_byte: offset in bytes where t <<
472 * @end_byte: offset in bytes where t <<
473 * <<
474 * Find at least one page in the range supplie <<
475 * direct writing in this range will trigger a <<
476 * <<
477 * Return: %true if at least one page exists i <<
478 * %false otherwise. <<
479 */ <<
480 bool filemap_range_has_page(struct address_spa <<
481 loff_t start_byte, <<
482 { <<
483 struct folio *folio; <<
484 XA_STATE(xas, &mapping->i_pages, start <<
485 pgoff_t max = end_byte >> PAGE_SHIFT; <<
486 <<
487 if (end_byte < start_byte) <<
488 return false; <<
489 <<
490 rcu_read_lock(); <<
491 for (;;) { <<
492 folio = xas_find(&xas, max); <<
493 if (xas_retry(&xas, folio)) <<
494 continue; <<
495 /* Shadow entries don't count <<
496 if (xa_is_value(folio)) <<
497 continue; <<
498 /* <<
499 * We don't need to try to pin <<
500 * release the RCU lock anyway <<
501 * there was a page here recen <<
502 */ <<
503 break; <<
504 } <<
505 rcu_read_unlock(); <<
506 <<
507 return folio != NULL; <<
508 } <<
509 EXPORT_SYMBOL(filemap_range_has_page); <<
510 <<
511 static void __filemap_fdatawait_range(struct a <<
512 loff_t st 380 loff_t start_byte, loff_t end_byte)
513 { 381 {
514 pgoff_t index = start_byte >> PAGE_SHI 382 pgoff_t index = start_byte >> PAGE_SHIFT;
515 pgoff_t end = end_byte >> PAGE_SHIFT; 383 pgoff_t end = end_byte >> PAGE_SHIFT;
516 struct folio_batch fbatch; !! 384 struct pagevec pvec;
517 unsigned nr_folios; !! 385 int nr_pages;
>> 386 int ret = 0;
518 387
519 folio_batch_init(&fbatch); !! 388 if (end_byte < start_byte)
>> 389 goto out;
520 390
521 while (index <= end) { !! 391 pagevec_init(&pvec, 0);
>> 392 while ((index <= end) &&
>> 393 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
>> 394 PAGECACHE_TAG_WRITEBACK,
>> 395 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
522 unsigned i; 396 unsigned i;
523 397
524 nr_folios = filemap_get_folios !! 398 for (i = 0; i < nr_pages; i++) {
525 PAGECACHE_TAG_ !! 399 struct page *page = pvec.pages[i];
526 400
527 if (!nr_folios) !! 401 /* until radix tree lookup accepts end_index */
528 break; !! 402 if (page->index > end)
529 !! 403 continue;
530 for (i = 0; i < nr_folios; i++ <<
531 struct folio *folio = <<
532 404
533 folio_wait_writeback(f !! 405 wait_on_page_writeback(page);
>> 406 if (TestClearPageError(page))
>> 407 ret = -EIO;
534 } 408 }
535 folio_batch_release(&fbatch); !! 409 pagevec_release(&pvec);
536 cond_resched(); 410 cond_resched();
537 } 411 }
>> 412 out:
>> 413 return ret;
538 } 414 }
539 415
540 /** 416 /**
541 * filemap_fdatawait_range - wait for writebac 417 * filemap_fdatawait_range - wait for writeback to complete
542 * @mapping: address space structur 418 * @mapping: address space structure to wait for
543 * @start_byte: offset in bytes where 419 * @start_byte: offset in bytes where the range starts
544 * @end_byte: offset in bytes where 420 * @end_byte: offset in bytes where the range ends (inclusive)
545 * 421 *
546 * Walk the list of under-writeback pages of t 422 * Walk the list of under-writeback pages of the given address space
547 * in the given range and wait for all of them 423 * in the given range and wait for all of them. Check error status of
548 * the address space and return it. 424 * the address space and return it.
549 * 425 *
550 * Since the error status of the address space 426 * Since the error status of the address space is cleared by this function,
551 * callers are responsible for checking the re 427 * callers are responsible for checking the return value and handling and/or
552 * reporting the error. 428 * reporting the error.
553 * <<
554 * Return: error status of the address space. <<
555 */ 429 */
556 int filemap_fdatawait_range(struct address_spa 430 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
557 loff_t end_byte) 431 loff_t end_byte)
558 { 432 {
559 __filemap_fdatawait_range(mapping, sta !! 433 int ret, ret2;
560 return filemap_check_errors(mapping); <<
561 } <<
562 EXPORT_SYMBOL(filemap_fdatawait_range); <<
563 434
564 /** !! 435 ret = __filemap_fdatawait_range(mapping, start_byte, end_byte);
565 * filemap_fdatawait_range_keep_errors - wait !! 436 ret2 = filemap_check_errors(mapping);
566 * @mapping: address space structur !! 437 if (!ret)
567 * @start_byte: offset in bytes where !! 438 ret = ret2;
568 * @end_byte: offset in bytes where <<
569 * <<
570 * Walk the list of under-writeback pages of t <<
571 * given range and wait for all of them. Unli <<
572 * this function does not clear error status o <<
573 * <<
574 * Use this function if callers don't handle e <<
575 * call sites are system-wide / filesystem-wid <<
576 * fsfreeze(8) <<
577 */ <<
578 int filemap_fdatawait_range_keep_errors(struct <<
579 loff_t start_byte, loff_t end_ <<
580 { <<
581 __filemap_fdatawait_range(mapping, sta <<
582 return filemap_check_and_keep_errors(m <<
583 } <<
584 EXPORT_SYMBOL(filemap_fdatawait_range_keep_err <<
585 <<
586 /** <<
587 * file_fdatawait_range - wait for writeback t <<
588 * @file: file pointing to addre <<
589 * @start_byte: offset in bytes where <<
590 * @end_byte: offset in bytes where <<
591 * <<
592 * Walk the list of under-writeback pages of t <<
593 * refers to, in the given range and wait for <<
594 * status of the address space vs. the file->f <<
595 * <<
596 * Since the error status of the file is advan <<
597 * callers are responsible for checking the re <<
598 * reporting the error. <<
599 * <<
600 * Return: error status of the address space v <<
601 */ <<
602 int file_fdatawait_range(struct file *file, lo <<
603 { <<
604 struct address_space *mapping = file-> <<
605 439
606 __filemap_fdatawait_range(mapping, sta !! 440 return ret;
607 return file_check_and_advance_wb_err(f <<
608 } 441 }
609 EXPORT_SYMBOL(file_fdatawait_range); !! 442 EXPORT_SYMBOL(filemap_fdatawait_range);
610 443
611 /** 444 /**
612 * filemap_fdatawait_keep_errors - wait for wr 445 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
613 * @mapping: address space structure to wait f 446 * @mapping: address space structure to wait for
614 * 447 *
615 * Walk the list of under-writeback pages of t 448 * Walk the list of under-writeback pages of the given address space
616 * and wait for all of them. Unlike filemap_f 449 * and wait for all of them. Unlike filemap_fdatawait(), this function
617 * does not clear error status of the address 450 * does not clear error status of the address space.
618 * 451 *
619 * Use this function if callers don't handle e 452 * Use this function if callers don't handle errors themselves. Expected
620 * call sites are system-wide / filesystem-wid 453 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
621 * fsfreeze(8) 454 * fsfreeze(8)
622 * <<
623 * Return: error status of the address space. <<
624 */ 455 */
625 int filemap_fdatawait_keep_errors(struct addre !! 456 void filemap_fdatawait_keep_errors(struct address_space *mapping)
626 { <<
627 __filemap_fdatawait_range(mapping, 0, <<
628 return filemap_check_and_keep_errors(m <<
629 } <<
630 EXPORT_SYMBOL(filemap_fdatawait_keep_errors); <<
631 <<
632 /* Returns true if writeback might be needed o <<
633 static bool mapping_needs_writeback(struct add <<
634 { <<
635 return mapping->nrpages; <<
636 } <<
637 <<
638 bool filemap_range_has_writeback(struct addres <<
639 loff_t start_ <<
640 { 457 {
641 XA_STATE(xas, &mapping->i_pages, start !! 458 loff_t i_size = i_size_read(mapping->host);
642 pgoff_t max = end_byte >> PAGE_SHIFT; <<
643 struct folio *folio; <<
644 459
645 if (end_byte < start_byte) !! 460 if (i_size == 0)
646 return false; !! 461 return;
647 462
648 rcu_read_lock(); !! 463 __filemap_fdatawait_range(mapping, 0, i_size - 1);
649 xas_for_each(&xas, folio, max) { <<
650 if (xas_retry(&xas, folio)) <<
651 continue; <<
652 if (xa_is_value(folio)) <<
653 continue; <<
654 if (folio_test_dirty(folio) || <<
655 folio_test_wri <<
656 break; <<
657 } <<
658 rcu_read_unlock(); <<
659 return folio != NULL; <<
660 } 464 }
661 EXPORT_SYMBOL_GPL(filemap_range_has_writeback) <<
662 465
663 /** 466 /**
664 * filemap_write_and_wait_range - write out & !! 467 * filemap_fdatawait - wait for all under-writeback pages to complete
665 * @mapping: the address_space for the page !! 468 * @mapping: address space structure to wait for
666 * @lstart: offset in bytes where the rang <<
667 * @lend: offset in bytes where the rang <<
668 * <<
669 * Write out and wait upon file offsets lstart <<
670 * 469 *
671 * Note that @lend is inclusive (describes the !! 470 * Walk the list of under-writeback pages of the given address space
672 * that this function can be used to write to !! 471 * and wait for all of them. Check error status of the address space
>> 472 * and return it.
673 * 473 *
674 * Return: error status of the address space. !! 474 * Since the error status of the address space is cleared by this function,
>> 475 * callers are responsible for checking the return value and handling and/or
>> 476 * reporting the error.
675 */ 477 */
676 int filemap_write_and_wait_range(struct addres !! 478 int filemap_fdatawait(struct address_space *mapping)
677 loff_t lstart <<
678 { 479 {
679 int err = 0, err2; !! 480 loff_t i_size = i_size_read(mapping->host);
680 481
681 if (lend < lstart) !! 482 if (i_size == 0)
682 return 0; 483 return 0;
683 484
684 if (mapping_needs_writeback(mapping)) !! 485 return filemap_fdatawait_range(mapping, 0, i_size - 1);
685 err = __filemap_fdatawrite_ran !! 486 }
686 !! 487 EXPORT_SYMBOL(filemap_fdatawait);
>> 488
>> 489 int filemap_write_and_wait(struct address_space *mapping)
>> 490 {
>> 491 int err = 0;
>> 492
>> 493 if ((!dax_mapping(mapping) && mapping->nrpages) ||
>> 494 (dax_mapping(mapping) && mapping->nrexceptional)) {
>> 495 err = filemap_fdatawrite(mapping);
687 /* 496 /*
688 * Even if the above returned 497 * Even if the above returned error, the pages may be
689 * written partially (e.g. -EN 498 * written partially (e.g. -ENOSPC), so we wait for it.
690 * But the -EIO is special cas 499 * But the -EIO is special case, it may indicate the worst
691 * thing (e.g. bug) happened, 500 * thing (e.g. bug) happened, so we avoid waiting for it.
692 */ 501 */
693 if (err != -EIO) !! 502 if (err != -EIO) {
694 __filemap_fdatawait_ra !! 503 int err2 = filemap_fdatawait(mapping);
>> 504 if (!err)
>> 505 err = err2;
>> 506 }
>> 507 } else {
>> 508 err = filemap_check_errors(mapping);
695 } 509 }
696 err2 = filemap_check_errors(mapping); <<
697 if (!err) <<
698 err = err2; <<
699 return err; 510 return err;
700 } 511 }
701 EXPORT_SYMBOL(filemap_write_and_wait_range); !! 512 EXPORT_SYMBOL(filemap_write_and_wait);
702 <<
703 void __filemap_set_wb_err(struct address_space <<
704 { <<
705 errseq_t eseq = errseq_set(&mapping->w <<
706 <<
707 trace_filemap_set_wb_err(mapping, eseq <<
708 } <<
709 EXPORT_SYMBOL(__filemap_set_wb_err); <<
710 513
711 /** 514 /**
712 * file_check_and_advance_wb_err - report wb e !! 515 * filemap_write_and_wait_range - write out & wait on a file range
713 * and advance !! 516 * @mapping: the address_space for the pages
714 * @file: struct file on which the error is be <<
715 * <<
716 * When userland calls fsync (or something lik <<
717 * want to report any writeback errors that oc <<
718 * since the file was opened if there haven't <<
719 * <<
720 * Grab the wb_err from the mapping. If it mat <<
721 * then just quickly return 0. The file is all <<
722 * <<
723 * If it doesn't match, then take the mapping <<
724 * it and try to swap it into place. If it wor <<
725 * to it with the new value, then update the f <<
726 * portion. The error at this point must be re <<
727 * (a'la fsync, or NFS COMMIT operation, etc.) <<
728 * <<
729 * While we handle mapping->wb_err with atomic <<
730 * value is protected by the f_lock since we m <<
731 * the latest value swapped in for this file d <<
732 * <<
733 * Return: %0 on success, negative error code <<
734 */ <<
735 int file_check_and_advance_wb_err(struct file <<
736 { <<
737 int err = 0; <<
738 errseq_t old = READ_ONCE(file->f_wb_er <<
739 struct address_space *mapping = file-> <<
740 <<
741 /* Locklessly handle the common case w <<
742 if (errseq_check(&mapping->wb_err, old <<
743 /* Something changed, must use <<
744 spin_lock(&file->f_lock); <<
745 old = file->f_wb_err; <<
746 err = errseq_check_and_advance <<
747 <<
748 trace_file_check_and_advance_w <<
749 spin_unlock(&file->f_lock); <<
750 } <<
751 <<
752 /* <<
753 * We're mostly using this function as <<
754 * filemap_check_errors. Clear AS_EIO/ <<
755 * that the legacy code would have had <<
756 */ <<
757 clear_bit(AS_EIO, &mapping->flags); <<
758 clear_bit(AS_ENOSPC, &mapping->flags); <<
759 return err; <<
760 } <<
761 EXPORT_SYMBOL(file_check_and_advance_wb_err); <<
762 <<
763 /** <<
764 * file_write_and_wait_range - write out & wai <<
765 * @file: file pointing to address_space <<
766 * @lstart: offset in bytes where the rang 517 * @lstart: offset in bytes where the range starts
767 * @lend: offset in bytes where the rang 518 * @lend: offset in bytes where the range ends (inclusive)
768 * 519 *
769 * Write out and wait upon file offsets lstart 520 * Write out and wait upon file offsets lstart->lend, inclusive.
770 * 521 *
771 * Note that @lend is inclusive (describes the 522 * Note that @lend is inclusive (describes the last byte to be written) so
772 * that this function can be used to write to 523 * that this function can be used to write to the very end-of-file (end = -1).
773 * <<
774 * After writing out and waiting on the data, <<
775 * f_wb_err cursor to the latest value, and re <<
776 * <<
777 * Return: %0 on success, negative error code <<
778 */ 524 */
779 int file_write_and_wait_range(struct file *fil !! 525 int filemap_write_and_wait_range(struct address_space *mapping,
>> 526 loff_t lstart, loff_t lend)
780 { 527 {
781 int err = 0, err2; !! 528 int err = 0;
782 struct address_space *mapping = file-> <<
783 <<
784 if (lend < lstart) <<
785 return 0; <<
786 529
787 if (mapping_needs_writeback(mapping)) !! 530 if ((!dax_mapping(mapping) && mapping->nrpages) ||
>> 531 (dax_mapping(mapping) && mapping->nrexceptional)) {
788 err = __filemap_fdatawrite_ran 532 err = __filemap_fdatawrite_range(mapping, lstart, lend,
789 533 WB_SYNC_ALL);
790 /* See comment of filemap_writ 534 /* See comment of filemap_write_and_wait() */
791 if (err != -EIO) !! 535 if (err != -EIO) {
792 __filemap_fdatawait_ra !! 536 int err2 = filemap_fdatawait_range(mapping,
>> 537 lstart, lend);
>> 538 if (!err)
>> 539 err = err2;
>> 540 }
>> 541 } else {
>> 542 err = filemap_check_errors(mapping);
793 } 543 }
794 err2 = file_check_and_advance_wb_err(f <<
795 if (!err) <<
796 err = err2; <<
797 return err; 544 return err;
798 } 545 }
799 EXPORT_SYMBOL(file_write_and_wait_range); !! 546 EXPORT_SYMBOL(filemap_write_and_wait_range);
800 547
801 /** 548 /**
802 * replace_page_cache_folio - replace a pageca !! 549 * replace_page_cache_page - replace a pagecache page with a new one
803 * @old: folio to be replaced !! 550 * @old: page to be replaced
804 * @new: folio to replace with !! 551 * @new: page to replace with
805 * !! 552 * @gfp_mask: allocation mode
806 * This function replaces a folio in the pagec !! 553 *
807 * success it acquires the pagecache reference !! 554 * This function replaces a page in the pagecache with a new one. On
808 * drops it for the old folio. Both the old a !! 555 * success it acquires the pagecache reference for the new page and
809 * locked. This function does not add the new !! 556 * drops it for the old page. Both the old and new pages must be
>> 557 * locked. This function does not add the new page to the LRU, the
810 * caller must do that. 558 * caller must do that.
811 * 559 *
812 * The remove + add is atomic. This function !! 560 * The remove + add is atomic. The only way this function can fail is
>> 561 * memory allocation failure.
813 */ 562 */
814 void replace_page_cache_folio(struct folio *ol !! 563 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
815 { 564 {
816 struct address_space *mapping = old->m !! 565 int error;
817 void (*free_folio)(struct folio *) = m <<
818 pgoff_t offset = old->index; <<
819 XA_STATE(xas, &mapping->i_pages, offse <<
820 <<
821 VM_BUG_ON_FOLIO(!folio_test_locked(old <<
822 VM_BUG_ON_FOLIO(!folio_test_locked(new <<
823 VM_BUG_ON_FOLIO(new->mapping, new); <<
824 <<
825 folio_get(new); <<
826 new->mapping = mapping; <<
827 new->index = offset; <<
828 <<
829 mem_cgroup_replace_folio(old, new); <<
830 <<
831 xas_lock_irq(&xas); <<
832 xas_store(&xas, new); <<
833 <<
834 old->mapping = NULL; <<
835 /* hugetlb pages do not participate in <<
836 if (!folio_test_hugetlb(old)) <<
837 __lruvec_stat_sub_folio(old, N <<
838 if (!folio_test_hugetlb(new)) <<
839 __lruvec_stat_add_folio(new, N <<
840 if (folio_test_swapbacked(old)) <<
841 __lruvec_stat_sub_folio(old, N <<
842 if (folio_test_swapbacked(new)) <<
843 __lruvec_stat_add_folio(new, N <<
844 xas_unlock_irq(&xas); <<
845 if (free_folio) <<
846 free_folio(old); <<
847 folio_put(old); <<
848 } <<
849 EXPORT_SYMBOL_GPL(replace_page_cache_folio); <<
850 <<
851 noinline int __filemap_add_folio(struct addres <<
852 struct folio *folio, pgoff_t i <<
853 { <<
854 XA_STATE(xas, &mapping->i_pages, index <<
855 void *alloced_shadow = NULL; <<
856 int alloced_order = 0; <<
857 bool huge; <<
858 long nr; <<
859 <<
860 VM_BUG_ON_FOLIO(!folio_test_locked(fol <<
861 VM_BUG_ON_FOLIO(folio_test_swapbacked( <<
862 VM_BUG_ON_FOLIO(folio_order(folio) < m <<
863 folio); <<
864 mapping_set_update(&xas, mapping); <<
865 <<
866 VM_BUG_ON_FOLIO(index & (folio_nr_page <<
867 xas_set_order(&xas, index, folio_order <<
868 huge = folio_test_hugetlb(folio); <<
869 nr = folio_nr_pages(folio); <<
870 <<
871 gfp &= GFP_RECLAIM_MASK; <<
872 folio_ref_add(folio, nr); <<
873 folio->mapping = mapping; <<
874 folio->index = xas.xa_index; <<
875 <<
876 for (;;) { <<
877 int order = -1, split_order = <<
878 void *entry, *old = NULL; <<
879 <<
880 xas_lock_irq(&xas); <<
881 xas_for_each_conflict(&xas, en <<
882 old = entry; <<
883 if (!xa_is_value(entry <<
884 xas_set_err(&x <<
885 goto unlock; <<
886 } <<
887 /* <<
888 * If a larger entry e <<
889 * it will be the firs <<
890 */ <<
891 if (order == -1) <<
892 order = xas_ge <<
893 } <<
894 566
895 /* entry may have changed befo !! 567 VM_BUG_ON_PAGE(!PageLocked(old), old);
896 if (alloced_order && (old != a !! 568 VM_BUG_ON_PAGE(!PageLocked(new), new);
897 xas_destroy(&xas); !! 569 VM_BUG_ON_PAGE(new->mapping, new);
898 alloced_order = 0; !! 570
899 } !! 571 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
900 !! 572 if (!error) {
901 if (old) { !! 573 struct address_space *mapping = old->mapping;
902 if (order > 0 && order !! 574 void (*freepage)(struct page *);
903 /* How to hand !! 575 unsigned long flags;
904 BUG_ON(shmem_m !! 576
905 if (!alloced_o !! 577 pgoff_t offset = old->index;
906 split_ !! 578 freepage = mapping->a_ops->freepage;
907 goto u !! 579
908 } !! 580 get_page(new);
909 xas_split(&xas !! 581 new->mapping = mapping;
910 xas_reset(&xas !! 582 new->index = offset;
911 } !! 583
912 if (shadowp) !! 584 spin_lock_irqsave(&mapping->tree_lock, flags);
913 *shadowp = old !! 585 __delete_from_page_cache(old, NULL);
914 } !! 586 error = page_cache_tree_insert(mapping, new, NULL);
>> 587 BUG_ON(error);
>> 588
>> 589 /*
>> 590 * hugetlb pages do not participate in page cache accounting.
>> 591 */
>> 592 if (!PageHuge(new))
>> 593 __inc_node_page_state(new, NR_FILE_PAGES);
>> 594 if (PageSwapBacked(new))
>> 595 __inc_node_page_state(new, NR_SHMEM);
>> 596 spin_unlock_irqrestore(&mapping->tree_lock, flags);
>> 597 mem_cgroup_migrate(old, new);
>> 598 radix_tree_preload_end();
>> 599 if (freepage)
>> 600 freepage(old);
>> 601 put_page(old);
>> 602 }
915 603
916 xas_store(&xas, folio); !! 604 return error;
917 if (xas_error(&xas)) !! 605 }
918 goto unlock; !! 606 EXPORT_SYMBOL_GPL(replace_page_cache_page);
919 607
920 mapping->nrpages += nr; !! 608 static int __add_to_page_cache_locked(struct page *page,
>> 609 struct address_space *mapping,
>> 610 pgoff_t offset, gfp_t gfp_mask,
>> 611 void **shadowp)
>> 612 {
>> 613 int huge = PageHuge(page);
>> 614 struct mem_cgroup *memcg;
>> 615 int error;
921 616
922 /* hugetlb pages do not partic !! 617 VM_BUG_ON_PAGE(!PageLocked(page), page);
923 if (!huge) { !! 618 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
924 __lruvec_stat_mod_foli <<
925 if (folio_test_pmd_map <<
926 __lruvec_stat_ <<
927 <<
928 } <<
929 619
930 unlock: !! 620 if (!huge) {
931 xas_unlock_irq(&xas); !! 621 error = mem_cgroup_try_charge(page, current->mm,
932 !! 622 gfp_mask, &memcg, false);
933 /* split needed, alloc here an !! 623 if (error)
934 if (split_order) { !! 624 return error;
935 xas_split_alloc(&xas, !! 625 }
936 if (xas_error(&xas)) <<
937 goto error; <<
938 alloced_shadow = old; <<
939 alloced_order = split_ <<
940 xas_reset(&xas); <<
941 continue; <<
942 } <<
943 626
944 if (!xas_nomem(&xas, gfp)) !! 627 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
945 break; !! 628 if (error) {
>> 629 if (!huge)
>> 630 mem_cgroup_cancel_charge(page, memcg, false);
>> 631 return error;
946 } 632 }
947 633
948 if (xas_error(&xas)) !! 634 get_page(page);
949 goto error; !! 635 page->mapping = mapping;
>> 636 page->index = offset;
>> 637
>> 638 spin_lock_irq(&mapping->tree_lock);
>> 639 error = page_cache_tree_insert(mapping, page, shadowp);
>> 640 radix_tree_preload_end();
>> 641 if (unlikely(error))
>> 642 goto err_insert;
950 643
951 trace_mm_filemap_add_to_page_cache(fol !! 644 /* hugetlb pages do not participate in page cache accounting. */
>> 645 if (!huge)
>> 646 __inc_node_page_state(page, NR_FILE_PAGES);
>> 647 spin_unlock_irq(&mapping->tree_lock);
>> 648 if (!huge)
>> 649 mem_cgroup_commit_charge(page, memcg, false, false);
>> 650 trace_mm_filemap_add_to_page_cache(page);
952 return 0; 651 return 0;
953 error: !! 652 err_insert:
954 folio->mapping = NULL; !! 653 page->mapping = NULL;
955 /* Leave page->index set: truncation r 654 /* Leave page->index set: truncation relies upon it */
956 folio_put_refs(folio, nr); !! 655 spin_unlock_irq(&mapping->tree_lock);
957 return xas_error(&xas); !! 656 if (!huge)
>> 657 mem_cgroup_cancel_charge(page, memcg, false);
>> 658 put_page(page);
>> 659 return error;
958 } 660 }
959 ALLOW_ERROR_INJECTION(__filemap_add_folio, ERR <<
960 661
961 int filemap_add_folio(struct address_space *ma !! 662 /**
962 pgoff_t index, !! 663 * add_to_page_cache_locked - add a locked page to the pagecache
>> 664 * @page: page to add
>> 665 * @mapping: the page's address_space
>> 666 * @offset: page index
>> 667 * @gfp_mask: page allocation mode
>> 668 *
>> 669 * This function is used to add a page to the pagecache. It must be locked.
>> 670 * This function does not add the page to the LRU. The caller must do that.
>> 671 */
>> 672 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
>> 673 pgoff_t offset, gfp_t gfp_mask)
>> 674 {
>> 675 return __add_to_page_cache_locked(page, mapping, offset,
>> 676 gfp_mask, NULL);
>> 677 }
>> 678 EXPORT_SYMBOL(add_to_page_cache_locked);
>> 679
>> 680 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
>> 681 pgoff_t offset, gfp_t gfp_mask)
963 { 682 {
964 void *shadow = NULL; 683 void *shadow = NULL;
965 int ret; 684 int ret;
966 685
967 ret = mem_cgroup_charge(folio, NULL, g !! 686 __SetPageLocked(page);
968 if (ret) !! 687 ret = __add_to_page_cache_locked(page, mapping, offset,
969 return ret; !! 688 gfp_mask, &shadow);
970 !! 689 if (unlikely(ret))
971 __folio_set_locked(folio); !! 690 __ClearPageLocked(page);
972 ret = __filemap_add_folio(mapping, fol !! 691 else {
973 if (unlikely(ret)) { <<
974 mem_cgroup_uncharge(folio); <<
975 __folio_clear_locked(folio); <<
976 } else { <<
977 /* 692 /*
978 * The folio might have been e !! 693 * The page might have been evicted from cache only
979 * recently, in which case it 694 * recently, in which case it should be activated like
980 * any other repeatedly access !! 695 * any other repeatedly accessed page.
981 * The exception is folios get !! 696 * The exception is pages getting rewritten; evicting other
982 * data from the working set, 697 * data from the working set, only to cache data that will
983 * get overwritten with someth 698 * get overwritten with something else, is a waste of memory.
984 */ 699 */
985 WARN_ON_ONCE(folio_test_active !! 700 if (!(gfp_mask & __GFP_WRITE) &&
986 if (!(gfp & __GFP_WRITE) && sh !! 701 shadow && workingset_refault(shadow)) {
987 workingset_refault(fol !! 702 SetPageActive(page);
988 folio_add_lru(folio); !! 703 workingset_activation(page);
>> 704 } else
>> 705 ClearPageActive(page);
>> 706 lru_cache_add(page);
989 } 707 }
990 return ret; 708 return ret;
991 } 709 }
992 EXPORT_SYMBOL_GPL(filemap_add_folio); !! 710 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
993 711
994 #ifdef CONFIG_NUMA 712 #ifdef CONFIG_NUMA
995 struct folio *filemap_alloc_folio_noprof(gfp_t !! 713 struct page *__page_cache_alloc(gfp_t gfp)
996 { 714 {
997 int n; 715 int n;
998 struct folio *folio; !! 716 struct page *page;
999 717
1000 if (cpuset_do_page_mem_spread()) { 718 if (cpuset_do_page_mem_spread()) {
1001 unsigned int cpuset_mems_cook 719 unsigned int cpuset_mems_cookie;
1002 do { 720 do {
1003 cpuset_mems_cookie = 721 cpuset_mems_cookie = read_mems_allowed_begin();
1004 n = cpuset_mem_spread 722 n = cpuset_mem_spread_node();
1005 folio = __folio_alloc !! 723 page = __alloc_pages_node(n, gfp, 0);
1006 } while (!folio && read_mems_ !! 724 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
1007 725
1008 return folio; !! 726 return page;
1009 } 727 }
1010 return folio_alloc_noprof(gfp, order) !! 728 return alloc_pages(gfp, 0);
1011 } 729 }
1012 EXPORT_SYMBOL(filemap_alloc_folio_noprof); !! 730 EXPORT_SYMBOL(__page_cache_alloc);
1013 #endif 731 #endif
1014 732
1015 /* 733 /*
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 734 * In order to wait for pages to become available there must be
1055 * waitqueues associated with pages. By using 735 * waitqueues associated with pages. By using a hash table of
1056 * waitqueues where the bucket discipline is 736 * waitqueues where the bucket discipline is to maintain all
1057 * waiters on the same queue and wake all whe 737 * waiters on the same queue and wake all when any of the pages
1058 * become available, and for the woken contex 738 * become available, and for the woken contexts to check to be
1059 * sure the appropriate page became available 739 * sure the appropriate page became available, this saves space
1060 * at a cost of "thundering herd" phenomena d 740 * at a cost of "thundering herd" phenomena during rare hash
1061 * collisions. 741 * collisions.
1062 */ 742 */
1063 #define PAGE_WAIT_TABLE_BITS 8 743 #define PAGE_WAIT_TABLE_BITS 8
1064 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_ 744 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1065 static wait_queue_head_t folio_wait_table[PAG !! 745 static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1066 746
1067 static wait_queue_head_t *folio_waitqueue(str !! 747 static wait_queue_head_t *page_waitqueue(struct page *page)
1068 { 748 {
1069 return &folio_wait_table[hash_ptr(fol !! 749 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
1070 } 750 }
1071 751
1072 void __init pagecache_init(void) 752 void __init pagecache_init(void)
1073 { 753 {
1074 int i; 754 int i;
1075 755
1076 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; 756 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1077 init_waitqueue_head(&folio_wa !! 757 init_waitqueue_head(&page_wait_table[i]);
1078 758
1079 page_writeback_init(); 759 page_writeback_init();
1080 } 760 }
1081 761
1082 /* !! 762 struct wait_page_key {
1083 * The page wait code treats the "wait->flags !! 763 struct page *page;
1084 * we have multiple different kinds of waits, !! 764 int bit_nr;
1085 * one. !! 765 int page_match;
1086 * !! 766 };
1087 * We have: !! 767
1088 * !! 768 struct wait_page_queue {
1089 * (a) no special bits set: !! 769 struct page *page;
1090 * !! 770 int bit_nr;
1091 * We're just waiting for the bit to be !! 771 wait_queue_t wait;
1092 * calls the wakeup function, we set WQ_ !! 772 };
1093 * and remove it from the wait queue. !! 773
1094 * !! 774 static int wake_page_function(wait_queue_t *wait, unsigned mode, int sync, void *arg)
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 <<
1117 { 775 {
1118 unsigned int flags; <<
1119 struct wait_page_key *key = arg; 776 struct wait_page_key *key = arg;
1120 struct wait_page_queue *wait_page 777 struct wait_page_queue *wait_page
1121 = container_of(wait, struct w 778 = container_of(wait, struct wait_page_queue, wait);
1122 779
1123 if (!wake_page_match(wait_page, key)) !! 780 if (wait_page->page != key->page)
1124 return 0; !! 781 return 0;
1125 !! 782 key->page_match = 1;
1126 /* <<
1127 * If it's a lock handoff wait, we ge <<
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 783
1141 /* !! 784 if (wait_page->bit_nr != key->bit_nr)
1142 * We are holding the wait-queue lock !! 785 return 0;
1143 * is waiting for this will be checki !! 786 if (test_bit(key->bit_nr, &key->page->flags))
1144 * any locking. !! 787 return 0;
1145 * <<
1146 * So update the flags atomically, an <<
1147 * afterwards to avoid any races. Thi <<
1148 * with the load-acquire in folio_wai <<
1149 */ <<
1150 smp_store_release(&wait->flags, flags <<
1151 wake_up_state(wait->private, mode); <<
1152 788
1153 /* !! 789 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 } 790 }
1166 791
1167 static void folio_wake_bit(struct folio *foli !! 792 static void wake_up_page_bit(struct page *page, int bit_nr)
1168 { 793 {
1169 wait_queue_head_t *q = folio_waitqueu !! 794 wait_queue_head_t *q = page_waitqueue(page);
1170 struct wait_page_key key; 795 struct wait_page_key key;
1171 unsigned long flags; 796 unsigned long flags;
1172 797
1173 key.folio = folio; !! 798 key.page = page;
1174 key.bit_nr = bit_nr; 799 key.bit_nr = bit_nr;
1175 key.page_match = 0; 800 key.page_match = 0;
1176 801
1177 spin_lock_irqsave(&q->lock, flags); 802 spin_lock_irqsave(&q->lock, flags);
1178 __wake_up_locked_key(q, TASK_NORMAL, 803 __wake_up_locked_key(q, TASK_NORMAL, &key);
1179 <<
1180 /* 804 /*
1181 * It's possible to miss clearing wai !! 805 * It is possible for other pages to have collided on the waitqueue
1182 * waiters, but the hashed waitqueue !! 806 * hash, so in that case check for a page match. That prevents a long-
1183 * That's okay, it's a rare case. The !! 807 * term waiter
1184 * 808 *
1185 * Note that, depending on the page p !! 809 * It is still possible to miss a case here, when we woke page waiters
1186 * other), the flag may be cleared in !! 810 * and removed them from the waitqueue, but there are still other
1187 * but that is not required for corre !! 811 * page waiters.
1188 */ !! 812 */
1189 if (!waitqueue_active(q) || !key.page !! 813 if (!waitqueue_active(q) || !key.page_match) {
1190 folio_clear_waiters(folio); !! 814 ClearPageWaiters(page);
1191 !! 815 /*
>> 816 * It's possible to miss clearing Waiters here, when we woke
>> 817 * our page waiters, but the hashed waitqueue has waiters for
>> 818 * other pages on it.
>> 819 *
>> 820 * That's okay, it's a rare case. The next waker will clear it.
>> 821 */
>> 822 }
1192 spin_unlock_irqrestore(&q->lock, flag 823 spin_unlock_irqrestore(&q->lock, flags);
1193 } 824 }
1194 825
1195 /* !! 826 static void wake_up_page(struct page *page, int bit)
1196 * A choice of three behaviors for folio_wait <<
1197 */ <<
1198 enum behavior { <<
1199 EXCLUSIVE, /* Hold ref to page a <<
1200 * __folio_lock() wai <<
1201 */ <<
1202 SHARED, /* Hold ref to page a <<
1203 * folio_wait_writeba <<
1204 */ <<
1205 DROP, /* Drop ref to page b <<
1206 * like folio_put_wai <<
1207 */ <<
1208 }; <<
1209 <<
1210 /* <<
1211 * Attempt to check (or get) the folio flag, <<
1212 * if successful. <<
1213 */ <<
1214 static inline bool folio_trylock_flag(struct <<
1215 struc <<
1216 { 827 {
1217 if (wait->flags & WQ_FLAG_EXCLUSIVE) !! 828 if (!PageWaiters(page))
1218 if (test_and_set_bit(bit_nr, !! 829 return;
1219 return false; !! 830 wake_up_page_bit(page, bit);
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 } 831 }
1226 832
1227 /* How many times do we accept lock stealing !! 833 static inline int wait_on_page_bit_common(wait_queue_head_t *q,
1228 int sysctl_page_lock_unfairness = 5; !! 834 struct page *page, int bit_nr, int state, bool lock)
1229 <<
1230 static inline int folio_wait_bit_common(struc <<
1231 int state, enum behavior beha <<
1232 { 835 {
1233 wait_queue_head_t *q = folio_waitqueu <<
1234 int unfairness = sysctl_page_lock_unf <<
1235 struct wait_page_queue wait_page; 836 struct wait_page_queue wait_page;
1236 wait_queue_entry_t *wait = &wait_page !! 837 wait_queue_t *wait = &wait_page.wait;
1237 bool thrashing = false; !! 838 int ret = 0;
1238 unsigned long pflags; <<
1239 bool in_thrashing; <<
1240 <<
1241 if (bit_nr == PG_locked && <<
1242 !folio_test_uptodate(folio) && fo <<
1243 delayacct_thrashing_start(&in <<
1244 psi_memstall_enter(&pflags); <<
1245 thrashing = true; <<
1246 } <<
1247 839
1248 init_wait(wait); 840 init_wait(wait);
1249 wait->func = wake_page_function; 841 wait->func = wake_page_function;
1250 wait_page.folio = folio; !! 842 wait_page.page = page;
1251 wait_page.bit_nr = bit_nr; 843 wait_page.bit_nr = bit_nr;
1252 844
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 (;;) { 845 for (;;) {
1299 unsigned int flags; !! 846 spin_lock_irq(&q->lock);
>> 847
>> 848 if (likely(list_empty(&wait->task_list))) {
>> 849 if (lock)
>> 850 __add_wait_queue_tail_exclusive(q, wait);
>> 851 else
>> 852 __add_wait_queue(q, wait);
>> 853 SetPageWaiters(page);
>> 854 }
1300 855
1301 set_current_state(state); 856 set_current_state(state);
1302 857
1303 /* Loop until we've been woke !! 858 spin_unlock_irq(&q->lock);
1304 flags = smp_load_acquire(&wai <<
1305 if (!(flags & WQ_FLAG_WOKEN)) <<
1306 if (signal_pending_st <<
1307 break; <<
1308 859
>> 860 if (likely(test_bit(bit_nr, &page->flags))) {
1309 io_schedule(); 861 io_schedule();
1310 continue; !! 862 if (unlikely(signal_pending_state(state, current))) {
>> 863 ret = -EINTR;
>> 864 break;
>> 865 }
1311 } 866 }
1312 867
1313 /* If we were non-exclusive, !! 868 if (lock) {
1314 if (behavior != EXCLUSIVE) !! 869 if (!test_and_set_bit_lock(bit_nr, &page->flags))
1315 break; !! 870 break;
1316 !! 871 } else {
1317 /* If the waker got the lock !! 872 if (!test_bit(bit_nr, &page->flags))
1318 if (flags & WQ_FLAG_DONE) <<
1319 break; <<
1320 <<
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 } <<
1333 <<
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); <<
1341 <<
1342 if (thrashing) { <<
1343 delayacct_thrashing_end(&in_t <<
1344 psi_memstall_leave(&pflags); <<
1345 } <<
1346 <<
1347 /* <<
1348 * NOTE! The wait->flags weren't stab <<
1349 * 'finish_wait()', and we could have <<
1350 * to a signal, and had a wakeup even <<
1351 * test but before the 'finish_wait() <<
1352 * <<
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 */ <<
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; 873 break;
1430 <<
1431 io_schedule(); <<
1432 continue; <<
1433 } 874 }
1434 break; <<
1435 } 875 }
1436 876
1437 finish_wait(q, wait); 877 finish_wait(q, wait);
1438 878
1439 if (thrashing) { !! 879 /*
1440 delayacct_thrashing_end(&in_t !! 880 * A signal could leave PageWaiters set. Clearing it here if
1441 psi_memstall_leave(&pflags); !! 881 * !waitqueue_active would be possible (by open-coding finish_wait),
1442 } !! 882 * but still fail to catch it in the case of wait hash collision. We
1443 } !! 883 * already can fail to clear wait hash collision cases, so don't
1444 #endif !! 884 * bother with signals either.
>> 885 */
1445 886
1446 void folio_wait_bit(struct folio *folio, int !! 887 return ret;
1447 { <<
1448 folio_wait_bit_common(folio, bit_nr, <<
1449 } 888 }
1450 EXPORT_SYMBOL(folio_wait_bit); <<
1451 889
1452 int folio_wait_bit_killable(struct folio *fol !! 890 void wait_on_page_bit(struct page *page, int bit_nr)
1453 { 891 {
1454 return folio_wait_bit_common(folio, b !! 892 wait_queue_head_t *q = page_waitqueue(page);
>> 893 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, false);
1455 } 894 }
1456 EXPORT_SYMBOL(folio_wait_bit_killable); !! 895 EXPORT_SYMBOL(wait_on_page_bit);
1457 896
1458 /** !! 897 int wait_on_page_bit_killable(struct page *page, int bit_nr)
1459 * folio_put_wait_locked - Drop a reference a <<
1460 * @folio: The folio to wait for. <<
1461 * @state: The sleep state (TASK_KILLABLE, TA <<
1462 * <<
1463 * The caller should hold a reference on @fol <<
1464 * become unlocked relatively soon, but do no <<
1465 * (for example) by holding the reference whi <<
1466 * come unlocked. After this function return <<
1467 * dereference @folio. <<
1468 * <<
1469 * Return: 0 if the folio was unlocked or -EI <<
1470 */ <<
1471 static int folio_put_wait_locked(struct folio <<
1472 { 898 {
1473 return folio_wait_bit_common(folio, P !! 899 wait_queue_head_t *q = page_waitqueue(page);
>> 900 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, false);
1474 } 901 }
1475 902
1476 /** 903 /**
1477 * folio_add_wait_queue - Add an arbitrary wa !! 904 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1478 * @folio: Folio defining the wait queue of i !! 905 * @page: Page defining the wait queue of interest
1479 * @waiter: Waiter to add to the queue 906 * @waiter: Waiter to add to the queue
1480 * 907 *
1481 * Add an arbitrary @waiter to the wait queue !! 908 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1482 */ 909 */
1483 void folio_add_wait_queue(struct folio *folio !! 910 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
1484 { 911 {
1485 wait_queue_head_t *q = folio_waitqueu !! 912 wait_queue_head_t *q = page_waitqueue(page);
1486 unsigned long flags; 913 unsigned long flags;
1487 914
1488 spin_lock_irqsave(&q->lock, flags); 915 spin_lock_irqsave(&q->lock, flags);
1489 __add_wait_queue_entry_tail(q, waiter !! 916 __add_wait_queue(q, waiter);
1490 folio_set_waiters(folio); !! 917 SetPageWaiters(page);
1491 spin_unlock_irqrestore(&q->lock, flag 918 spin_unlock_irqrestore(&q->lock, flags);
1492 } 919 }
1493 EXPORT_SYMBOL_GPL(folio_add_wait_queue); !! 920 EXPORT_SYMBOL_GPL(add_page_wait_queue);
1494 921
1495 /** !! 922 #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 923
1515 /** !! 924 /*
1516 * folio_end_read - End read on a folio. !! 925 * 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 * 926 *
1549 * Clear the PG_private_2 bit on a folio and !! 927 * On x86 (and on many other architectures), we can clear PG_lock and
1550 * it. The folio reference held for PG_priva !! 928 * test the sign bit at the same time. But if the architecture does
>> 929 * not support that special operation, we just do this all by hand
>> 930 * instead.
1551 * 931 *
1552 * This is, for example, used when a netfs fo !! 932 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1553 * disk cache, thereby allowing writes to the !! 933 * being cleared, but a memory barrier should be unneccssary since it is
1554 * serialised. !! 934 * in the same byte as PG_locked.
1555 */ 935 */
1556 void folio_end_private_2(struct folio *folio) !! 936 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1557 { 937 {
1558 VM_BUG_ON_FOLIO(!folio_test_private_2 !! 938 clear_bit_unlock(nr, mem);
1559 clear_bit_unlock(PG_private_2, folio_ !! 939 /* smp_mb__after_atomic(); */
1560 folio_wake_bit(folio, PG_private_2); !! 940 return test_bit(PG_waiters, mem);
1561 folio_put(folio); <<
1562 } 941 }
1563 EXPORT_SYMBOL(folio_end_private_2); !! 942
>> 943 #endif
1564 944
1565 /** 945 /**
1566 * folio_wait_private_2 - Wait for PG_private !! 946 * unlock_page - unlock a locked page
1567 * @folio: The folio to wait on. !! 947 * @page: the page
1568 * 948 *
1569 * Wait for PG_private_2 to be cleared on a f !! 949 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
>> 950 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
>> 951 * mechanism between PageLocked pages and PageWriteback pages is shared.
>> 952 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
>> 953 *
>> 954 * Note that this depends on PG_waiters being the sign bit in the byte
>> 955 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
>> 956 * clear the PG_locked bit and test PG_waiters at the same time fairly
>> 957 * portably (architectures that do LL/SC can test any bit, while x86 can
>> 958 * test the sign bit).
1570 */ 959 */
1571 void folio_wait_private_2(struct folio *folio !! 960 void unlock_page(struct page *page)
1572 { 961 {
1573 while (folio_test_private_2(folio)) !! 962 BUILD_BUG_ON(PG_waiters != 7);
1574 folio_wait_bit(folio, PG_priv !! 963 page = compound_head(page);
>> 964 VM_BUG_ON_PAGE(!PageLocked(page), page);
>> 965 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
>> 966 wake_up_page_bit(page, PG_locked);
1575 } 967 }
1576 EXPORT_SYMBOL(folio_wait_private_2); !! 968 EXPORT_SYMBOL(unlock_page);
1577 969
1578 /** 970 /**
1579 * folio_wait_private_2_killable - Wait for P !! 971 * end_page_writeback - end writeback against a page
1580 * @folio: The folio to wait on. !! 972 * @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 */ 973 */
1589 int folio_wait_private_2_killable(struct foli !! 974 void end_page_writeback(struct page *page)
1590 { 975 {
1591 int ret = 0; !! 976 /*
1592 !! 977 * TestClearPageReclaim could be used here but it is an atomic
1593 while (folio_test_private_2(folio)) { !! 978 * operation and overkill in this particular case. Failing to
1594 ret = folio_wait_bit_killable !! 979 * shuffle a page marked for immediate reclaim is too mild to
1595 if (ret < 0) !! 980 * justify taking an atomic operation penalty at the end of
1596 break; !! 981 * ever page writeback.
>> 982 */
>> 983 if (PageReclaim(page)) {
>> 984 ClearPageReclaim(page);
>> 985 rotate_reclaimable_page(page);
1597 } 986 }
1598 987
1599 return ret; !! 988 if (!test_clear_page_writeback(page))
>> 989 BUG();
>> 990
>> 991 smp_mb__after_atomic();
>> 992 wake_up_page(page, PG_writeback);
1600 } 993 }
1601 EXPORT_SYMBOL(folio_wait_private_2_killable); !! 994 EXPORT_SYMBOL(end_page_writeback);
1602 995
1603 /** !! 996 /*
1604 * folio_end_writeback - End writeback agains !! 997 * After completing I/O on a page, call this routine to update the page
1605 * @folio: The folio. !! 998 * flags appropriately
1606 * <<
1607 * The folio must actually be under writeback <<
1608 * <<
1609 * Context: May be called from process or int <<
1610 */ 999 */
1611 void folio_end_writeback(struct folio *folio) !! 1000 void page_endio(struct page *page, bool is_write, int err)
1612 { 1001 {
1613 VM_BUG_ON_FOLIO(!folio_test_writeback !! 1002 if (!is_write) {
>> 1003 if (!err) {
>> 1004 SetPageUptodate(page);
>> 1005 } else {
>> 1006 ClearPageUptodate(page);
>> 1007 SetPageError(page);
>> 1008 }
>> 1009 unlock_page(page);
>> 1010 } else {
>> 1011 if (err) {
>> 1012 struct address_space *mapping;
1614 1013
1615 /* !! 1014 SetPageError(page);
1616 * folio_test_clear_reclaim() could b !! 1015 mapping = page_mapping(page);
1617 * atomic operation and overkill in t !! 1016 if (mapping)
1618 * to shuffle a folio marked for imme !! 1017 mapping_set_error(mapping, err);
1619 * a gain to justify taking an atomic !! 1018 }
1620 * end of every folio writeback. !! 1019 end_page_writeback(page);
1621 */ <<
1622 if (folio_test_reclaim(folio)) { <<
1623 folio_clear_reclaim(folio); <<
1624 folio_rotate_reclaimable(foli <<
1625 } 1020 }
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 } 1021 }
1639 EXPORT_SYMBOL(folio_end_writeback); !! 1022 EXPORT_SYMBOL_GPL(page_endio);
1640 1023
1641 /** 1024 /**
1642 * __folio_lock - Get a lock on the folio, as !! 1025 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1643 * @folio: The folio to lock !! 1026 * @__page: the page to lock
1644 */ 1027 */
1645 void __folio_lock(struct folio *folio) !! 1028 void __lock_page(struct page *__page)
1646 { <<
1647 folio_wait_bit_common(folio, PG_locke <<
1648 EXCLUSIVE); <<
1649 } <<
1650 EXPORT_SYMBOL(__folio_lock); <<
1651 <<
1652 int __folio_lock_killable(struct folio *folio <<
1653 { 1029 {
1654 return folio_wait_bit_common(folio, P !! 1030 struct page *page = compound_head(__page);
1655 EXCLU !! 1031 wait_queue_head_t *q = page_waitqueue(page);
>> 1032 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, true);
1656 } 1033 }
1657 EXPORT_SYMBOL_GPL(__folio_lock_killable); !! 1034 EXPORT_SYMBOL(__lock_page);
1658 1035
1659 static int __folio_lock_async(struct folio *f !! 1036 int __lock_page_killable(struct page *__page)
1660 { 1037 {
1661 struct wait_queue_head *q = folio_wai !! 1038 struct page *page = compound_head(__page);
1662 int ret; !! 1039 wait_queue_head_t *q = page_waitqueue(page);
1663 !! 1040 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, true);
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 } 1041 }
>> 1042 EXPORT_SYMBOL_GPL(__lock_page_killable);
1684 1043
1685 /* 1044 /*
1686 * Return values: 1045 * Return values:
1687 * 0 - folio is locked. !! 1046 * 1 - page is locked; mmap_sem is still held.
1688 * non-zero - folio is not locked. !! 1047 * 0 - page is not locked.
1689 * mmap_lock or per-VMA lock has been rel !! 1048 * mmap_sem has been released (up_read()), unless flags had both
1690 * vma_end_read()), unless flags had both !! 1049 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1691 * FAULT_FLAG_RETRY_NOWAIT set, in which !! 1050 * which case mmap_sem is still held.
1692 * !! 1051 *
1693 * If neither ALLOW_RETRY nor KILLABLE are se !! 1052 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1694 * with the folio locked and the mmap_lock/pe !! 1053 * with the page locked and the mmap_sem unperturbed.
1695 */ 1054 */
1696 vm_fault_t __folio_lock_or_retry(struct folio !! 1055 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
>> 1056 unsigned int flags)
1697 { 1057 {
1698 unsigned int flags = vmf->flags; !! 1058 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1699 <<
1700 if (fault_flag_allow_retry_first(flag <<
1701 /* 1059 /*
1702 * CAUTION! In this case, mma !! 1060 * CAUTION! In this case, mmap_sem is not released
1703 * released even though retur !! 1061 * even though return 0.
1704 */ 1062 */
1705 if (flags & FAULT_FLAG_RETRY_ 1063 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1706 return VM_FAULT_RETRY !! 1064 return 0;
1707 1065
1708 release_fault_lock(vmf); !! 1066 up_read(&mm->mmap_sem);
1709 if (flags & FAULT_FLAG_KILLAB 1067 if (flags & FAULT_FLAG_KILLABLE)
1710 folio_wait_locked_kil !! 1068 wait_on_page_locked_killable(page);
1711 else 1069 else
1712 folio_wait_locked(fol !! 1070 wait_on_page_locked(page);
1713 return VM_FAULT_RETRY; !! 1071 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 { 1072 } else {
1724 __folio_lock(folio); !! 1073 if (flags & FAULT_FLAG_KILLABLE) {
1725 } !! 1074 int ret;
1726 1075
1727 return 0; !! 1076 ret = __lock_page_killable(page);
>> 1077 if (ret) {
>> 1078 up_read(&mm->mmap_sem);
>> 1079 return 0;
>> 1080 }
>> 1081 } else
>> 1082 __lock_page(page);
>> 1083 return 1;
>> 1084 }
1728 } 1085 }
1729 1086
1730 /** 1087 /**
1731 * page_cache_next_miss() - Find the next gap !! 1088 * page_cache_next_hole - find the next hole (not-present entry)
1732 * @mapping: Mapping. !! 1089 * @mapping: mapping
1733 * @index: Index. !! 1090 * @index: index
1734 * @max_scan: Maximum range to search. !! 1091 * @max_scan: maximum range to search
1735 * !! 1092 *
1736 * Search the range [index, min(index + max_s !! 1093 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
1737 * gap with the lowest index. !! 1094 * lowest indexed hole.
1738 * !! 1095 *
1739 * This function may be called under the rcu_ !! 1096 * Returns: the index of the hole if found, otherwise returns an index
1740 * not atomically search a snapshot of the ca !! 1097 * outside of the set specified (in which case 'return - index >=
1741 * For example, if a gap is created at index !! 1098 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1742 * created at index 10, page_cache_next_miss !! 1099 * be returned.
1743 * return 10 if called under the rcu_read_loc !! 1100 *
1744 * !! 1101 * page_cache_next_hole may be called under rcu_read_lock. However,
1745 * Return: The index of the gap if found, oth !! 1102 * like radix_tree_gang_lookup, this will not atomically search a
1746 * range specified (in which case 'return - i !! 1103 * snapshot of the tree at a single point in time. For example, if a
1747 * In the rare case of index wrap-around, 0 w !! 1104 * hole is created at index 5, then subsequently a hole is created at
>> 1105 * index 10, page_cache_next_hole covering both indexes may return 10
>> 1106 * if called under rcu_read_lock.
1748 */ 1107 */
1749 pgoff_t page_cache_next_miss(struct address_s !! 1108 pgoff_t page_cache_next_hole(struct address_space *mapping,
1750 pgoff_t index, u 1109 pgoff_t index, unsigned long max_scan)
1751 { 1110 {
1752 XA_STATE(xas, &mapping->i_pages, inde !! 1111 unsigned long i;
1753 1112
1754 while (max_scan--) { !! 1113 for (i = 0; i < max_scan; i++) {
1755 void *entry = xas_next(&xas); !! 1114 struct page *page;
1756 if (!entry || xa_is_value(ent !! 1115
1757 return xas.xa_index; !! 1116 page = radix_tree_lookup(&mapping->page_tree, index);
1758 if (xas.xa_index == 0) !! 1117 if (!page || radix_tree_exceptional_entry(page))
1759 return 0; !! 1118 break;
>> 1119 index++;
>> 1120 if (index == 0)
>> 1121 break;
1760 } 1122 }
1761 1123
1762 return index + max_scan; !! 1124 return index;
1763 } 1125 }
1764 EXPORT_SYMBOL(page_cache_next_miss); !! 1126 EXPORT_SYMBOL(page_cache_next_hole);
1765 1127
1766 /** 1128 /**
1767 * page_cache_prev_miss() - Find the previous !! 1129 * page_cache_prev_hole - find the prev hole (not-present entry)
1768 * @mapping: Mapping. !! 1130 * @mapping: mapping
1769 * @index: Index. !! 1131 * @index: index
1770 * @max_scan: Maximum range to search. !! 1132 * @max_scan: maximum range to search
1771 * !! 1133 *
1772 * Search the range [max(index - max_scan + 1 !! 1134 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1773 * gap with the highest index. !! 1135 * the first hole.
1774 * !! 1136 *
1775 * This function may be called under the rcu_ !! 1137 * Returns: the index of the hole if found, otherwise returns an index
1776 * not atomically search a snapshot of the ca !! 1138 * outside of the set specified (in which case 'index - return >=
1777 * For example, if a gap is created at index !! 1139 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1778 * created at index 5, page_cache_prev_miss() !! 1140 * will be returned.
1779 * return 5 if called under the rcu_read_lock !! 1141 *
1780 * !! 1142 * page_cache_prev_hole may be called under rcu_read_lock. However,
1781 * Return: The index of the gap if found, oth !! 1143 * like radix_tree_gang_lookup, this will not atomically search a
1782 * range specified (in which case 'index - re !! 1144 * snapshot of the tree at a single point in time. For example, if a
1783 * In the rare case of wrap-around, ULONG_MAX !! 1145 * hole is created at index 10, then subsequently a hole is created at
>> 1146 * index 5, page_cache_prev_hole covering both indexes may return 5 if
>> 1147 * called under rcu_read_lock.
1784 */ 1148 */
1785 pgoff_t page_cache_prev_miss(struct address_s !! 1149 pgoff_t page_cache_prev_hole(struct address_space *mapping,
1786 pgoff_t index, u 1150 pgoff_t index, unsigned long max_scan)
1787 { 1151 {
1788 XA_STATE(xas, &mapping->i_pages, inde !! 1152 unsigned long i;
1789 1153
1790 while (max_scan--) { !! 1154 for (i = 0; i < max_scan; i++) {
1791 void *entry = xas_prev(&xas); !! 1155 struct page *page;
1792 if (!entry || xa_is_value(ent !! 1156
>> 1157 page = radix_tree_lookup(&mapping->page_tree, index);
>> 1158 if (!page || radix_tree_exceptional_entry(page))
1793 break; 1159 break;
1794 if (xas.xa_index == ULONG_MAX !! 1160 index--;
>> 1161 if (index == ULONG_MAX)
1795 break; 1162 break;
1796 } 1163 }
1797 1164
1798 return xas.xa_index; !! 1165 return index;
1799 } 1166 }
1800 EXPORT_SYMBOL(page_cache_prev_miss); !! 1167 EXPORT_SYMBOL(page_cache_prev_hole);
1801 <<
1802 /* <<
1803 * Lockless page cache protocol: <<
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 1168
1822 /* !! 1169 /**
1823 * filemap_get_entry - Get a page cache entry !! 1170 * find_get_entry - find and get a page cache entry
1824 * @mapping: the address_space to search 1171 * @mapping: the address_space to search
1825 * @index: The page cache index. !! 1172 * @offset: the page cache index
>> 1173 *
>> 1174 * Looks up the page cache slot at @mapping & @offset. If there is a
>> 1175 * page cache page, it is returned with an increased refcount.
1826 * 1176 *
1827 * Looks up the page cache entry at @mapping !! 1177 * If the slot holds a shadow entry of a previously evicted page, or a
1828 * it is returned with an increased refcount. !! 1178 * 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 * 1179 *
1832 * Return: The folio, swap or shadow entry, % !! 1180 * Otherwise, %NULL is returned.
1833 */ 1181 */
1834 void *filemap_get_entry(struct address_space !! 1182 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1835 { 1183 {
1836 XA_STATE(xas, &mapping->i_pages, inde !! 1184 void **pagep;
1837 struct folio *folio; !! 1185 struct page *head, *page;
1838 1186
1839 rcu_read_lock(); 1187 rcu_read_lock();
1840 repeat: 1188 repeat:
1841 xas_reset(&xas); !! 1189 page = NULL;
1842 folio = xas_load(&xas); !! 1190 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1843 if (xas_retry(&xas, folio)) !! 1191 if (pagep) {
1844 goto repeat; !! 1192 page = radix_tree_deref_slot(pagep);
1845 /* !! 1193 if (unlikely(!page))
1846 * A shadow entry of a recently evict !! 1194 goto out;
1847 * shmem/tmpfs. Return it without at !! 1195 if (radix_tree_exception(page)) {
1848 */ !! 1196 if (radix_tree_deref_retry(page))
1849 if (!folio || xa_is_value(folio)) !! 1197 goto repeat;
1850 goto out; !! 1198 /*
>> 1199 * A shadow entry of a recently evicted page,
>> 1200 * or a swap entry from shmem/tmpfs. Return
>> 1201 * it without attempting to raise page count.
>> 1202 */
>> 1203 goto out;
>> 1204 }
1851 1205
1852 if (!folio_try_get(folio)) !! 1206 head = compound_head(page);
1853 goto repeat; !! 1207 if (!page_cache_get_speculative(head))
>> 1208 goto repeat;
1854 1209
1855 if (unlikely(folio != xas_reload(&xas !! 1210 /* The page was split under us? */
1856 folio_put(folio); !! 1211 if (compound_head(page) != head) {
1857 goto repeat; !! 1212 put_page(head);
>> 1213 goto repeat;
>> 1214 }
>> 1215
>> 1216 /*
>> 1217 * Has the page moved?
>> 1218 * This is part of the lockless pagecache protocol. See
>> 1219 * include/linux/pagemap.h for details.
>> 1220 */
>> 1221 if (unlikely(page != *pagep)) {
>> 1222 put_page(head);
>> 1223 goto repeat;
>> 1224 }
1858 } 1225 }
1859 out: 1226 out:
1860 rcu_read_unlock(); 1227 rcu_read_unlock();
1861 1228
1862 return folio; !! 1229 return page;
>> 1230 }
>> 1231 EXPORT_SYMBOL(find_get_entry);
>> 1232
>> 1233 /**
>> 1234 * find_lock_entry - locate, pin and lock a page cache entry
>> 1235 * @mapping: the address_space to search
>> 1236 * @offset: the page cache index
>> 1237 *
>> 1238 * Looks up the page cache slot at @mapping & @offset. If there is a
>> 1239 * page cache page, it is returned locked and with an increased
>> 1240 * refcount.
>> 1241 *
>> 1242 * If the slot holds a shadow entry of a previously evicted page, or a
>> 1243 * swap entry from shmem/tmpfs, it is returned.
>> 1244 *
>> 1245 * Otherwise, %NULL is returned.
>> 1246 *
>> 1247 * find_lock_entry() may sleep.
>> 1248 */
>> 1249 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
>> 1250 {
>> 1251 struct page *page;
>> 1252
>> 1253 repeat:
>> 1254 page = find_get_entry(mapping, offset);
>> 1255 if (page && !radix_tree_exception(page)) {
>> 1256 lock_page(page);
>> 1257 /* Has the page been truncated? */
>> 1258 if (unlikely(page_mapping(page) != mapping)) {
>> 1259 unlock_page(page);
>> 1260 put_page(page);
>> 1261 goto repeat;
>> 1262 }
>> 1263 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
>> 1264 }
>> 1265 return page;
1863 } 1266 }
>> 1267 EXPORT_SYMBOL(find_lock_entry);
1864 1268
1865 /** 1269 /**
1866 * __filemap_get_folio - Find and get a refer !! 1270 * pagecache_get_page - find and get a page reference
1867 * @mapping: The address_space to search. !! 1271 * @mapping: the address_space to search
1868 * @index: The page index. !! 1272 * @offset: the page index
1869 * @fgp_flags: %FGP flags modify how the foli !! 1273 * @fgp_flags: PCG flags
1870 * @gfp: Memory allocation flags to use if %F !! 1274 * @gfp_mask: gfp mask to use for the page cache data page allocation
>> 1275 *
>> 1276 * Looks up the page cache slot at @mapping & @offset.
>> 1277 *
>> 1278 * PCG flags modify how the page is returned.
1871 * 1279 *
1872 * Looks up the page cache entry at @mapping !! 1280 * @fgp_flags can be:
1873 * 1281 *
1874 * If %FGP_LOCK or %FGP_CREAT are specified t !! 1282 * - FGP_ACCESSED: the page will be marked accessed
1875 * if the %GFP flags specified for %FGP_CREAT !! 1283 * - FGP_LOCK: Page is return locked
>> 1284 * - FGP_CREAT: If page is not present then a new page is allocated using
>> 1285 * @gfp_mask and added to the page cache and the VM's LRU
>> 1286 * list. The page is returned locked and with an increased
>> 1287 * refcount. Otherwise, NULL is returned.
1876 * 1288 *
1877 * If this function returns a folio, it is re !! 1289 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
>> 1290 * if the GFP flags specified for FGP_CREAT are atomic.
1878 * 1291 *
1879 * Return: The found folio or an ERR_PTR() ot !! 1292 * If there is a page cache page, it is returned with an increased refcount.
1880 */ 1293 */
1881 struct folio *__filemap_get_folio(struct addr !! 1294 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1882 fgf_t fgp_flags, gfp_t gfp) !! 1295 int fgp_flags, gfp_t gfp_mask)
1883 { 1296 {
1884 struct folio *folio; !! 1297 struct page *page;
1885 1298
1886 repeat: 1299 repeat:
1887 folio = filemap_get_entry(mapping, in !! 1300 page = find_get_entry(mapping, offset);
1888 if (xa_is_value(folio)) !! 1301 if (radix_tree_exceptional_entry(page))
1889 folio = NULL; !! 1302 page = NULL;
1890 if (!folio) !! 1303 if (!page)
1891 goto no_page; 1304 goto no_page;
1892 1305
1893 if (fgp_flags & FGP_LOCK) { 1306 if (fgp_flags & FGP_LOCK) {
1894 if (fgp_flags & FGP_NOWAIT) { 1307 if (fgp_flags & FGP_NOWAIT) {
1895 if (!folio_trylock(fo !! 1308 if (!trylock_page(page)) {
1896 folio_put(fol !! 1309 put_page(page);
1897 return ERR_PT !! 1310 return NULL;
1898 } 1311 }
1899 } else { 1312 } else {
1900 folio_lock(folio); !! 1313 lock_page(page);
1901 } 1314 }
1902 1315
1903 /* Has the page been truncate 1316 /* Has the page been truncated? */
1904 if (unlikely(folio->mapping ! !! 1317 if (unlikely(page->mapping != mapping)) {
1905 folio_unlock(folio); !! 1318 unlock_page(page);
1906 folio_put(folio); !! 1319 put_page(page);
1907 goto repeat; 1320 goto repeat;
1908 } 1321 }
1909 VM_BUG_ON_FOLIO(!folio_contai !! 1322 VM_BUG_ON_PAGE(page->index != offset, page);
1910 } 1323 }
1911 1324
1912 if (fgp_flags & FGP_ACCESSED) !! 1325 if (page && (fgp_flags & FGP_ACCESSED))
1913 folio_mark_accessed(folio); !! 1326 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 1327
1920 if (fgp_flags & FGP_STABLE) <<
1921 folio_wait_stable(folio); <<
1922 no_page: 1328 no_page:
1923 if (!folio && (fgp_flags & FGP_CREAT) !! 1329 if (!page && (fgp_flags & FGP_CREAT)) {
1924 unsigned int min_order = mapp <<
1925 unsigned int order = max(min_ <<
1926 int err; 1330 int err;
1927 index = mapping_align_index(m !! 1331 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1928 !! 1332 gfp_mask |= __GFP_WRITE;
1929 if ((fgp_flags & FGP_WRITE) & <<
1930 gfp |= __GFP_WRITE; <<
1931 if (fgp_flags & FGP_NOFS) 1333 if (fgp_flags & FGP_NOFS)
1932 gfp &= ~__GFP_FS; !! 1334 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 <<
1940 if (order > mapping_max_folio <<
1941 order = mapping_max_f <<
1942 /* If we're not aligned, allo <<
1943 if (index & ((1UL << order) - <<
1944 order = __ffs(index); <<
1945 <<
1946 do { <<
1947 gfp_t alloc_gfp = gfp <<
1948 <<
1949 err = -ENOMEM; <<
1950 if (order > min_order <<
1951 alloc_gfp |= <<
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 1335
1960 err = filemap_add_fol !! 1336 page = __page_cache_alloc(gfp_mask);
1961 if (!err) !! 1337 if (!page)
1962 break; !! 1338 return NULL;
1963 folio_put(folio); <<
1964 folio = NULL; <<
1965 } while (order-- > min_order) <<
1966 <<
1967 if (err == -EEXIST) <<
1968 goto repeat; <<
1969 if (err) <<
1970 return ERR_PTR(err); <<
1971 /* <<
1972 * filemap_add_folio locks th <<
1973 * we expect an unlocked page <<
1974 */ <<
1975 if (folio && (fgp_flags & FGP <<
1976 folio_unlock(folio); <<
1977 } <<
1978 <<
1979 if (!folio) <<
1980 return ERR_PTR(-ENOENT); <<
1981 return folio; <<
1982 } <<
1983 EXPORT_SYMBOL(__filemap_get_folio); <<
1984 <<
1985 static inline struct folio *find_get_entry(st <<
1986 xa_mark_t mark) <<
1987 { <<
1988 struct folio *folio; <<
1989 <<
1990 retry: <<
1991 if (mark == XA_PRESENT) <<
1992 folio = xas_find(xas, max); <<
1993 else <<
1994 folio = xas_find_marked(xas, <<
1995 1339
1996 if (xas_retry(xas, folio)) !! 1340 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1997 goto retry; !! 1341 fgp_flags |= FGP_LOCK;
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 1342
2006 if (!folio_try_get(folio)) !! 1343 /* Init accessed so avoid atomic mark_page_accessed later */
2007 goto reset; !! 1344 if (fgp_flags & FGP_ACCESSED)
>> 1345 __SetPageReferenced(page);
2008 1346
2009 if (unlikely(folio != xas_reload(xas) !! 1347 err = add_to_page_cache_lru(page, mapping, offset,
2010 folio_put(folio); !! 1348 gfp_mask & GFP_RECLAIM_MASK);
2011 goto reset; !! 1349 if (unlikely(err)) {
>> 1350 put_page(page);
>> 1351 page = NULL;
>> 1352 if (err == -EEXIST)
>> 1353 goto repeat;
>> 1354 }
2012 } 1355 }
2013 1356
2014 return folio; !! 1357 return page;
2015 reset: <<
2016 xas_reset(xas); <<
2017 goto retry; <<
2018 } 1358 }
>> 1359 EXPORT_SYMBOL(pagecache_get_page);
2019 1360
2020 /** 1361 /**
2021 * find_get_entries - gang pagecache lookup 1362 * find_get_entries - gang pagecache lookup
2022 * @mapping: The address_space to search 1363 * @mapping: The address_space to search
2023 * @start: The starting page cache index 1364 * @start: The starting page cache index
2024 * @end: The final page index (inclusi !! 1365 * @nr_entries: The maximum number of entries
2025 * @fbatch: Where the resulting entries a !! 1366 * @entries: Where the resulting entries are placed
2026 * @indices: The cache indices correspondi 1367 * @indices: The cache indices corresponding to the entries in @entries
2027 * 1368 *
2028 * find_get_entries() will search for and ret !! 1369 * find_get_entries() will search for and return a group of up to
2029 * the mapping. The entries are placed in @f !! 1370 * @nr_entries entries in the mapping. The entries are placed at
2030 * takes a reference on any actual folios it !! 1371 * @entries. find_get_entries() takes a reference against any actual
2031 * !! 1372 * pages it returns.
2032 * The entries have ascending indexes. The i !! 1373 *
2033 * due to not-present entries or large folios !! 1374 * The search returns a group of mapping-contiguous page cache entries
>> 1375 * with ascending indexes. There may be holes in the indices due to
>> 1376 * not-present pages.
2034 * 1377 *
2035 * Any shadow entries of evicted folios, or s !! 1378 * Any shadow entries of evicted pages, or swap entries from
2036 * shmem/tmpfs, are included in the returned 1379 * shmem/tmpfs, are included in the returned array.
2037 * 1380 *
2038 * Return: The number of entries which were f !! 1381 * find_get_entries() returns the number of pages and shadow entries
>> 1382 * which were found.
2039 */ 1383 */
2040 unsigned find_get_entries(struct address_spac !! 1384 unsigned find_get_entries(struct address_space *mapping,
2041 pgoff_t end, struct folio_bat !! 1385 pgoff_t start, unsigned int nr_entries,
2042 { !! 1386 struct page **entries, pgoff_t *indices)
2043 XA_STATE(xas, &mapping->i_pages, *sta !! 1387 {
2044 struct folio *folio; !! 1388 void **slot;
>> 1389 unsigned int ret = 0;
>> 1390 struct radix_tree_iter iter;
>> 1391
>> 1392 if (!nr_entries)
>> 1393 return 0;
2045 1394
2046 rcu_read_lock(); 1395 rcu_read_lock();
2047 while ((folio = find_get_entry(&xas, !! 1396 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
2048 indices[fbatch->nr] = xas.xa_ !! 1397 struct page *head, *page;
2049 if (!folio_batch_add(fbatch, !! 1398 repeat:
2050 break; !! 1399 page = radix_tree_deref_slot(slot);
2051 } !! 1400 if (unlikely(!page))
>> 1401 continue;
>> 1402 if (radix_tree_exception(page)) {
>> 1403 if (radix_tree_deref_retry(page)) {
>> 1404 slot = radix_tree_iter_retry(&iter);
>> 1405 continue;
>> 1406 }
>> 1407 /*
>> 1408 * A shadow entry of a recently evicted page, a swap
>> 1409 * entry from shmem/tmpfs or a DAX entry. Return it
>> 1410 * without attempting to raise page count.
>> 1411 */
>> 1412 goto export;
>> 1413 }
2052 1414
2053 if (folio_batch_count(fbatch)) { !! 1415 head = compound_head(page);
2054 unsigned long nr; !! 1416 if (!page_cache_get_speculative(head))
2055 int idx = folio_batch_count(f !! 1417 goto repeat;
2056 !! 1418
2057 folio = fbatch->folios[idx]; !! 1419 /* The page was split under us? */
2058 if (!xa_is_value(folio)) !! 1420 if (compound_head(page) != head) {
2059 nr = folio_nr_pages(f !! 1421 put_page(head);
2060 else !! 1422 goto repeat;
2061 nr = 1 << xa_get_orde !! 1423 }
2062 *start = round_down(indices[i !! 1424
>> 1425 /* Has the page moved? */
>> 1426 if (unlikely(page != *slot)) {
>> 1427 put_page(head);
>> 1428 goto repeat;
>> 1429 }
>> 1430 export:
>> 1431 indices[ret] = iter.index;
>> 1432 entries[ret] = page;
>> 1433 if (++ret == nr_entries)
>> 1434 break;
2063 } 1435 }
2064 rcu_read_unlock(); 1436 rcu_read_unlock();
2065 !! 1437 return ret;
2066 return folio_batch_count(fbatch); <<
2067 } 1438 }
2068 1439
2069 /** 1440 /**
2070 * find_lock_entries - Find a batch of pageca !! 1441 * find_get_pages - gang pagecache lookup
2071 * @mapping: The address_space to search. !! 1442 * @mapping: The address_space to search
2072 * @start: The starting page cache index !! 1443 * @start: The starting page index
2073 * @end: The final page index (inclusi !! 1444 * @nr_pages: The maximum number of pages
2074 * @fbatch: Where the resulting entries a !! 1445 * @pages: Where the resulting pages are placed
2075 * @indices: The cache indices of the entr !! 1446 *
2076 * !! 1447 * find_get_pages() will search for and return a group of up to
2077 * find_lock_entries() will return a batch of !! 1448 * @nr_pages pages in the mapping. The pages are placed at @pages.
2078 * Swap, shadow and DAX entries are included. !! 1449 * find_get_pages() takes a reference against the returned pages.
2079 * locked and with an incremented refcount. !! 1450 *
2080 * by somebody else or under writeback are sk !! 1451 * The search returns a group of mapping-contiguous pages with ascending
2081 * partially outside the range are not return !! 1452 * indexes. There may be holes in the indices due to not-present pages.
2082 * <<
2083 * The entries have ascending indexes. The i <<
2084 * due to not-present entries, large folios, <<
2085 * locked or folios under writeback. <<
2086 * 1453 *
2087 * Return: The number of entries which were f !! 1454 * find_get_pages() returns the number of pages which were found.
2088 */ 1455 */
2089 unsigned find_lock_entries(struct address_spa !! 1456 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
2090 pgoff_t end, struct folio_bat !! 1457 unsigned int nr_pages, struct page **pages)
2091 { 1458 {
2092 XA_STATE(xas, &mapping->i_pages, *sta !! 1459 struct radix_tree_iter iter;
2093 struct folio *folio; !! 1460 void **slot;
>> 1461 unsigned ret = 0;
>> 1462
>> 1463 if (unlikely(!nr_pages))
>> 1464 return 0;
2094 1465
2095 rcu_read_lock(); 1466 rcu_read_lock();
2096 while ((folio = find_get_entry(&xas, !! 1467 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
2097 unsigned long base; !! 1468 struct page *head, *page;
2098 unsigned long nr; !! 1469 repeat:
2099 !! 1470 page = radix_tree_deref_slot(slot);
2100 if (!xa_is_value(folio)) { !! 1471 if (unlikely(!page))
2101 nr = folio_nr_pages(f !! 1472 continue;
2102 base = folio->index; !! 1473
2103 /* Omit large folio w !! 1474 if (radix_tree_exception(page)) {
2104 if (base < *start) !! 1475 if (radix_tree_deref_retry(page)) {
2105 goto put; !! 1476 slot = radix_tree_iter_retry(&iter);
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; 1477 continue;
2122 /* Omit order>0 value !! 1478 }
2123 if (base + nr - 1 > e !! 1479 /*
2124 break; !! 1480 * A shadow entry of a recently evicted page,
>> 1481 * or a swap entry from shmem/tmpfs. Skip
>> 1482 * over it.
>> 1483 */
>> 1484 continue;
2125 } 1485 }
2126 1486
2127 /* Update start now so that l !! 1487 head = compound_head(page);
2128 *start = base + nr; !! 1488 if (!page_cache_get_speculative(head))
2129 indices[fbatch->nr] = xas.xa_ !! 1489 goto repeat;
2130 if (!folio_batch_add(fbatch, !! 1490
>> 1491 /* The page was split under us? */
>> 1492 if (compound_head(page) != head) {
>> 1493 put_page(head);
>> 1494 goto repeat;
>> 1495 }
>> 1496
>> 1497 /* Has the page moved? */
>> 1498 if (unlikely(page != *slot)) {
>> 1499 put_page(head);
>> 1500 goto repeat;
>> 1501 }
>> 1502
>> 1503 pages[ret] = page;
>> 1504 if (++ret == nr_pages)
2131 break; 1505 break;
2132 continue; <<
2133 unlock: <<
2134 folio_unlock(folio); <<
2135 put: <<
2136 folio_put(folio); <<
2137 } 1506 }
2138 rcu_read_unlock(); <<
2139 1507
2140 return folio_batch_count(fbatch); !! 1508 rcu_read_unlock();
>> 1509 return ret;
2141 } 1510 }
2142 1511
2143 /** 1512 /**
2144 * filemap_get_folios - Get a batch of folios !! 1513 * find_get_pages_contig - gang contiguous pagecache lookup
2145 * @mapping: The address_space to search 1514 * @mapping: The address_space to search
2146 * @start: The starting page index !! 1515 * @index: The starting page index
2147 * @end: The final page index (inclusi !! 1516 * @nr_pages: The maximum number of pages
2148 * @fbatch: The batch to fill. !! 1517 * @pages: Where the resulting pages are placed
2149 * 1518 *
2150 * Search for and return a batch of folios in !! 1519 * find_get_pages_contig() works exactly like find_get_pages(), except
2151 * index @start and up to index @end (inclusi !! 1520 * that the returned number of pages are guaranteed to be contiguous.
2152 * in @fbatch with an elevated reference coun <<
2153 * 1521 *
2154 * Return: The number of folios which were fo !! 1522 * find_get_pages_contig() returns the number of pages which were found.
2155 * We also update @start to index the next fo <<
2156 */ 1523 */
2157 unsigned filemap_get_folios(struct address_sp !! 1524 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
2158 pgoff_t end, struct folio_bat !! 1525 unsigned int nr_pages, struct page **pages)
2159 { 1526 {
2160 return filemap_get_folios_tag(mapping !! 1527 struct radix_tree_iter iter;
>> 1528 void **slot;
>> 1529 unsigned int ret = 0;
>> 1530
>> 1531 if (unlikely(!nr_pages))
>> 1532 return 0;
>> 1533
>> 1534 rcu_read_lock();
>> 1535 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
>> 1536 struct page *head, *page;
>> 1537 repeat:
>> 1538 page = radix_tree_deref_slot(slot);
>> 1539 /* The hole, there no reason to continue */
>> 1540 if (unlikely(!page))
>> 1541 break;
>> 1542
>> 1543 if (radix_tree_exception(page)) {
>> 1544 if (radix_tree_deref_retry(page)) {
>> 1545 slot = radix_tree_iter_retry(&iter);
>> 1546 continue;
>> 1547 }
>> 1548 /*
>> 1549 * A shadow entry of a recently evicted page,
>> 1550 * or a swap entry from shmem/tmpfs. Stop
>> 1551 * looking for contiguous pages.
>> 1552 */
>> 1553 break;
>> 1554 }
>> 1555
>> 1556 head = compound_head(page);
>> 1557 if (!page_cache_get_speculative(head))
>> 1558 goto repeat;
>> 1559
>> 1560 /* The page was split under us? */
>> 1561 if (compound_head(page) != head) {
>> 1562 put_page(head);
>> 1563 goto repeat;
>> 1564 }
>> 1565
>> 1566 /* Has the page moved? */
>> 1567 if (unlikely(page != *slot)) {
>> 1568 put_page(head);
>> 1569 goto repeat;
>> 1570 }
>> 1571
>> 1572 /*
>> 1573 * must check mapping and index after taking the ref.
>> 1574 * otherwise we can get both false positives and false
>> 1575 * negatives, which is just confusing to the caller.
>> 1576 */
>> 1577 if (page->mapping == NULL || page_to_pgoff(page) != iter.index) {
>> 1578 put_page(page);
>> 1579 break;
>> 1580 }
>> 1581
>> 1582 pages[ret] = page;
>> 1583 if (++ret == nr_pages)
>> 1584 break;
>> 1585 }
>> 1586 rcu_read_unlock();
>> 1587 return ret;
2161 } 1588 }
2162 EXPORT_SYMBOL(filemap_get_folios); !! 1589 EXPORT_SYMBOL(find_get_pages_contig);
2163 1590
2164 /** 1591 /**
2165 * filemap_get_folios_contig - Get a batch of !! 1592 * find_get_pages_tag - find and return pages that match @tag
2166 * @mapping: The address_space to search !! 1593 * @mapping: the address_space to search
2167 * @start: The starting page index !! 1594 * @index: the starting page index
2168 * @end: The final page index (inclusi !! 1595 * @tag: the tag index
2169 * @fbatch: The batch to fill !! 1596 * @nr_pages: the maximum number of pages
>> 1597 * @pages: where the resulting pages are placed
2170 * 1598 *
2171 * filemap_get_folios_contig() works exactly !! 1599 * Like find_get_pages, except we only return pages which are tagged with
2172 * except the returned folios are guaranteed !! 1600 * @tag. We update @index to index the next page for the traversal.
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 */ 1601 */
2178 !! 1602 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
2179 unsigned filemap_get_folios_contig(struct add !! 1603 int tag, unsigned int nr_pages, struct page **pages)
2180 pgoff_t *start, pgoff_t end, <<
2181 { 1604 {
2182 XA_STATE(xas, &mapping->i_pages, *sta !! 1605 struct radix_tree_iter iter;
2183 unsigned long nr; !! 1606 void **slot;
2184 struct folio *folio; !! 1607 unsigned ret = 0;
2185 1608
2186 rcu_read_lock(); !! 1609 if (unlikely(!nr_pages))
>> 1610 return 0;
2187 1611
2188 for (folio = xas_load(&xas); folio && !! 1612 rcu_read_lock();
2189 folio = xas_next(&xas !! 1613 radix_tree_for_each_tagged(slot, &mapping->page_tree,
2190 if (xas_retry(&xas, folio)) !! 1614 &iter, *index, tag) {
>> 1615 struct page *head, *page;
>> 1616 repeat:
>> 1617 page = radix_tree_deref_slot(slot);
>> 1618 if (unlikely(!page))
2191 continue; 1619 continue;
2192 /* <<
2193 * If the entry has been swap <<
2194 * No current caller is looki <<
2195 */ <<
2196 if (xa_is_value(folio)) <<
2197 goto update_start; <<
2198 1620
2199 /* If we landed in the middle !! 1621 if (radix_tree_exception(page)) {
2200 if (xa_is_sibling(folio)) !! 1622 if (radix_tree_deref_retry(page)) {
2201 goto update_start; !! 1623 slot = radix_tree_iter_retry(&iter);
2202 !! 1624 continue;
2203 if (!folio_try_get(folio)) !! 1625 }
2204 goto retry; !! 1626 /*
2205 !! 1627 * A shadow entry of a recently evicted page.
2206 if (unlikely(folio != xas_rel !! 1628 *
2207 goto put_folio; !! 1629 * Those entries should never be tagged, but
2208 !! 1630 * this tree walk is lockless and the tags are
2209 if (!folio_batch_add(fbatch, !! 1631 * looked up in bulk, one radix tree node at a
2210 nr = folio_nr_pages(f !! 1632 * time, so there is a sizable window for page
2211 *start = folio->index !! 1633 * reclaim to evict a page we saw tagged.
2212 goto out; !! 1634 *
>> 1635 * Skip over it.
>> 1636 */
>> 1637 continue;
2213 } 1638 }
2214 continue; <<
2215 put_folio: <<
2216 folio_put(folio); <<
2217 1639
2218 retry: !! 1640 head = compound_head(page);
2219 xas_reset(&xas); !! 1641 if (!page_cache_get_speculative(head))
2220 } !! 1642 goto repeat;
2221 1643
2222 update_start: !! 1644 /* The page was split under us? */
2223 nr = folio_batch_count(fbatch); !! 1645 if (compound_head(page) != head) {
>> 1646 put_page(head);
>> 1647 goto repeat;
>> 1648 }
2224 1649
2225 if (nr) { !! 1650 /* Has the page moved? */
2226 folio = fbatch->folios[nr - 1 !! 1651 if (unlikely(page != *slot)) {
2227 *start = folio_next_index(fol !! 1652 put_page(head);
>> 1653 goto repeat;
>> 1654 }
>> 1655
>> 1656 pages[ret] = page;
>> 1657 if (++ret == nr_pages)
>> 1658 break;
2228 } 1659 }
2229 out: !! 1660
2230 rcu_read_unlock(); 1661 rcu_read_unlock();
2231 return folio_batch_count(fbatch); !! 1662
>> 1663 if (ret)
>> 1664 *index = pages[ret - 1]->index + 1;
>> 1665
>> 1666 return ret;
2232 } 1667 }
2233 EXPORT_SYMBOL(filemap_get_folios_contig); !! 1668 EXPORT_SYMBOL(find_get_pages_tag);
2234 1669
2235 /** 1670 /**
2236 * filemap_get_folios_tag - Get a batch of fo !! 1671 * find_get_entries_tag - find and return entries that match @tag
2237 * @mapping: The address_space to search !! 1672 * @mapping: the address_space to search
2238 * @start: The starting page index !! 1673 * @start: the starting page cache index
2239 * @end: The final page index (inclusi !! 1674 * @tag: the tag index
2240 * @tag: The tag index !! 1675 * @nr_entries: the maximum number of entries
2241 * @fbatch: The batch to fill !! 1676 * @entries: where the resulting entries are placed
2242 * !! 1677 * @indices: the cache indices corresponding to the entries in @entries
2243 * The first folio may start before @start; i <<
2244 * @start. The final folio may extend beyond <<
2245 * contain @end. The folios have ascending i <<
2246 * between the folios if there are indices wh <<
2247 * page cache. If folios are added to or rem <<
2248 * while this is running, they may or may not <<
2249 * Only returns folios that are tagged with @ <<
2250 * 1678 *
2251 * Return: The number of folios found. !! 1679 * Like find_get_entries, except we only return entries which are tagged with
2252 * Also update @start to index the next folio !! 1680 * @tag.
2253 */ 1681 */
2254 unsigned filemap_get_folios_tag(struct addres !! 1682 unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
2255 pgoff_t end, xa_mark_ !! 1683 int tag, unsigned int nr_entries,
>> 1684 struct page **entries, pgoff_t *indices)
2256 { 1685 {
2257 XA_STATE(xas, &mapping->i_pages, *sta !! 1686 void **slot;
2258 struct folio *folio; !! 1687 unsigned int ret = 0;
>> 1688 struct radix_tree_iter iter;
>> 1689
>> 1690 if (!nr_entries)
>> 1691 return 0;
2259 1692
2260 rcu_read_lock(); 1693 rcu_read_lock();
2261 while ((folio = find_get_entry(&xas, !! 1694 radix_tree_for_each_tagged(slot, &mapping->page_tree,
2262 /* !! 1695 &iter, start, tag) {
2263 * Shadow entries should neve !! 1696 struct page *head, *page;
2264 * is lockless so there is a !! 1697 repeat:
2265 * a page we saw tagged. Skip !! 1698 page = radix_tree_deref_slot(slot);
2266 */ !! 1699 if (unlikely(!page))
2267 if (xa_is_value(folio)) <<
2268 continue; 1700 continue;
2269 if (!folio_batch_add(fbatch, !! 1701 if (radix_tree_exception(page)) {
2270 unsigned long nr = fo !! 1702 if (radix_tree_deref_retry(page)) {
2271 *start = folio->index !! 1703 slot = radix_tree_iter_retry(&iter);
2272 goto out; !! 1704 continue;
>> 1705 }
>> 1706
>> 1707 /*
>> 1708 * A shadow entry of a recently evicted page, a swap
>> 1709 * entry from shmem/tmpfs or a DAX entry. Return it
>> 1710 * without attempting to raise page count.
>> 1711 */
>> 1712 goto export;
2273 } 1713 }
>> 1714
>> 1715 head = compound_head(page);
>> 1716 if (!page_cache_get_speculative(head))
>> 1717 goto repeat;
>> 1718
>> 1719 /* The page was split under us? */
>> 1720 if (compound_head(page) != head) {
>> 1721 put_page(head);
>> 1722 goto repeat;
>> 1723 }
>> 1724
>> 1725 /* Has the page moved? */
>> 1726 if (unlikely(page != *slot)) {
>> 1727 put_page(head);
>> 1728 goto repeat;
>> 1729 }
>> 1730 export:
>> 1731 indices[ret] = iter.index;
>> 1732 entries[ret] = page;
>> 1733 if (++ret == nr_entries)
>> 1734 break;
2274 } 1735 }
2275 /* <<
2276 * We come here when there is no page <<
2277 * overflow the index @start as it co <<
2278 * breaks the iteration when there is <<
2279 * already broke anyway. <<
2280 */ <<
2281 if (end == (pgoff_t)-1) <<
2282 *start = (pgoff_t)-1; <<
2283 else <<
2284 *start = end + 1; <<
2285 out: <<
2286 rcu_read_unlock(); 1736 rcu_read_unlock();
2287 !! 1737 return ret;
2288 return folio_batch_count(fbatch); <<
2289 } 1738 }
2290 EXPORT_SYMBOL(filemap_get_folios_tag); !! 1739 EXPORT_SYMBOL(find_get_entries_tag);
2291 1740
2292 /* 1741 /*
2293 * CD/DVDs are error prone. When a medium err 1742 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2294 * a _large_ part of the i/o request. Imagine 1743 * a _large_ part of the i/o request. Imagine the worst scenario:
2295 * 1744 *
2296 * ---R_________________________________ 1745 * ---R__________________________________________B__________
2297 * ^ reading here 1746 * ^ reading here ^ bad block(assume 4k)
2298 * 1747 *
2299 * read(R) => miss => readahead(R...B) => med 1748 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2300 * => failing the whole request => read(R) => 1749 * => failing the whole request => read(R) => read(R+1) =>
2301 * readahead(R+1...B+1) => bang => read(R+2) 1750 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2302 * readahead(R+3...B+2) => bang => read(R+3) 1751 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2303 * readahead(R+4...B+3) => bang => read(R+4) 1752 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2304 * 1753 *
2305 * It is going insane. Fix it by quickly scal 1754 * It is going insane. Fix it by quickly scaling down the readahead size.
2306 */ 1755 */
2307 static void shrink_readahead_size_eio(struct !! 1756 static void shrink_readahead_size_eio(struct file *filp,
>> 1757 struct file_ra_state *ra)
2308 { 1758 {
2309 ra->ra_pages /= 4; 1759 ra->ra_pages /= 4;
2310 } 1760 }
2311 1761
2312 /* !! 1762 /**
2313 * filemap_get_read_batch - Get a batch of fo !! 1763 * do_generic_file_read - generic file read routine
>> 1764 * @filp: the file to read
>> 1765 * @ppos: current file position
>> 1766 * @iter: data destination
>> 1767 * @written: already copied
2314 * 1768 *
2315 * Get a batch of folios which represent a co !! 1769 * This is a generic file read routine, and uses the
2316 * the file. No exceptional entries will be !! 1770 * mapping->a_ops->readpage() function for the actual low-level stuff.
2317 * the middle of a folio, the entire folio wi !! 1771 *
2318 * folio in the batch may have the readahead !! 1772 * This is really ugly. But the goto's actually try to clarify some
2319 * clear so that the caller can take the appr !! 1773 * of the logic when it comes to error handling etc.
2320 */ 1774 */
2321 static void filemap_get_read_batch(struct add !! 1775 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
2322 pgoff_t index, pgoff_t max, s !! 1776 struct iov_iter *iter, ssize_t written)
2323 { 1777 {
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 { <<
2526 struct file *filp = iocb->ki_filp; <<
2527 struct address_space *mapping = filp- 1778 struct address_space *mapping = filp->f_mapping;
>> 1779 struct inode *inode = mapping->host;
2528 struct file_ra_state *ra = &filp->f_r 1780 struct file_ra_state *ra = &filp->f_ra;
2529 pgoff_t index = iocb->ki_pos >> PAGE_ !! 1781 pgoff_t index;
2530 pgoff_t last_index; 1782 pgoff_t last_index;
2531 struct folio *folio; !! 1783 pgoff_t prev_index;
2532 unsigned int flags; !! 1784 unsigned long offset; /* offset into pagecache page */
2533 int err = 0; !! 1785 unsigned int prev_offset;
2534 !! 1786 int error = 0;
2535 /* "last_index" is the index of the p <<
2536 last_index = DIV_ROUND_UP(iocb->ki_po <<
2537 retry: <<
2538 if (fatal_signal_pending(current)) <<
2539 return -EINTR; <<
2540 <<
2541 filemap_get_read_batch(mapping, index <<
2542 if (!folio_batch_count(fbatch)) { <<
2543 if (iocb->ki_flags & IOCB_NOI <<
2544 return -EAGAIN; <<
2545 if (iocb->ki_flags & IOCB_NOW <<
2546 flags = memalloc_noio <<
2547 page_cache_sync_readahead(map <<
2548 last_index - <<
2549 if (iocb->ki_flags & IOCB_NOW <<
2550 memalloc_noio_restore <<
2551 filemap_get_read_batch(mappin <<
2552 } <<
2553 if (!folio_batch_count(fbatch)) { <<
2554 if (iocb->ki_flags & (IOCB_NO <<
2555 return -EAGAIN; <<
2556 err = filemap_create_folio(fi <<
2557 if (err == AOP_TRUNCATED_PAGE <<
2558 goto retry; <<
2559 return err; <<
2560 } <<
2561 <<
2562 folio = fbatch->folios[folio_batch_co <<
2563 if (folio_test_readahead(folio)) { <<
2564 err = filemap_readahead(iocb, <<
2565 if (err) <<
2566 goto err; <<
2567 } <<
2568 if (!folio_test_uptodate(folio)) { <<
2569 if ((iocb->ki_flags & IOCB_WA <<
2570 folio_batch_count(fbatch) <<
2571 iocb->ki_flags |= IOC <<
2572 err = filemap_update_page(ioc <<
2573 nee <<
2574 if (err) <<
2575 goto err; <<
2576 } <<
2577 1787
2578 trace_mm_filemap_get_pages(mapping, i !! 1788 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
2579 return 0; <<
2580 err: <<
2581 if (err < 0) <<
2582 folio_put(folio); <<
2583 if (likely(--fbatch->nr)) <<
2584 return 0; 1789 return 0;
2585 if (err == AOP_TRUNCATED_PAGE) !! 1790 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 <<
2597 /** <<
2598 * filemap_read - Read data from the page cac <<
2599 * @iocb: The iocb to read. <<
2600 * @iter: Destination for the data. <<
2601 * @already_read: Number of bytes already rea <<
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 1791
2623 if (unlikely(iocb->ki_pos >= inode->i !! 1792 index = *ppos >> PAGE_SHIFT;
2624 return 0; !! 1793 prev_index = ra->prev_pos >> PAGE_SHIFT;
2625 if (unlikely(!iov_iter_count(iter))) !! 1794 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
2626 return 0; !! 1795 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
>> 1796 offset = *ppos & ~PAGE_MASK;
2627 1797
2628 iov_iter_truncate(iter, inode->i_sb-> !! 1798 for (;;) {
2629 folio_batch_init(&fbatch); !! 1799 struct page *page;
>> 1800 pgoff_t end_index;
>> 1801 loff_t isize;
>> 1802 unsigned long nr, ret;
2630 1803
2631 do { <<
2632 cond_resched(); 1804 cond_resched();
>> 1805 find_page:
>> 1806 if (fatal_signal_pending(current)) {
>> 1807 error = -EINTR;
>> 1808 goto out;
>> 1809 }
2633 1810
>> 1811 page = find_get_page(mapping, index);
>> 1812 if (!page) {
>> 1813 page_cache_sync_readahead(mapping,
>> 1814 ra, filp,
>> 1815 index, last_index - index);
>> 1816 page = find_get_page(mapping, index);
>> 1817 if (unlikely(page == NULL))
>> 1818 goto no_cached_page;
>> 1819 }
>> 1820 if (PageReadahead(page)) {
>> 1821 page_cache_async_readahead(mapping,
>> 1822 ra, filp, page,
>> 1823 index, last_index - index);
>> 1824 }
>> 1825 if (!PageUptodate(page)) {
>> 1826 /*
>> 1827 * See comment in do_read_cache_page on why
>> 1828 * wait_on_page_locked is used to avoid unnecessarily
>> 1829 * serialisations and why it's safe.
>> 1830 */
>> 1831 error = wait_on_page_locked_killable(page);
>> 1832 if (unlikely(error))
>> 1833 goto readpage_error;
>> 1834 if (PageUptodate(page))
>> 1835 goto page_ok;
>> 1836
>> 1837 if (inode->i_blkbits == PAGE_SHIFT ||
>> 1838 !mapping->a_ops->is_partially_uptodate)
>> 1839 goto page_not_up_to_date;
>> 1840 /* pipes can't handle partially uptodate pages */
>> 1841 if (unlikely(iter->type & ITER_PIPE))
>> 1842 goto page_not_up_to_date;
>> 1843 if (!trylock_page(page))
>> 1844 goto page_not_up_to_date;
>> 1845 /* Did it get truncated before we got the lock? */
>> 1846 if (!page->mapping)
>> 1847 goto page_not_up_to_date_locked;
>> 1848 if (!mapping->a_ops->is_partially_uptodate(page,
>> 1849 offset, iter->count))
>> 1850 goto page_not_up_to_date_locked;
>> 1851 unlock_page(page);
>> 1852 }
>> 1853 page_ok:
2634 /* 1854 /*
2635 * If we've already successfu !! 1855 * i_size must be checked after we know the page is Uptodate.
2636 * can no longer safely retur <<
2637 * an async read NOWAIT at th <<
2638 */ <<
2639 if ((iocb->ki_flags & IOCB_WA <<
2640 iocb->ki_flags |= IOC <<
2641 <<
2642 if (unlikely(iocb->ki_pos >= <<
2643 break; <<
2644 <<
2645 error = filemap_get_pages(ioc <<
2646 if (error < 0) <<
2647 break; <<
2648 <<
2649 /* <<
2650 * i_size must be checked aft <<
2651 * 1856 *
2652 * Checking i_size after the 1857 * Checking i_size after the check allows us to calculate
2653 * the correct value for "nr" 1858 * the correct value for "nr", which means the zero-filled
2654 * part of the page is not co 1859 * part of the page is not copied back to userspace (unless
2655 * another truncate extends t 1860 * another truncate extends the file - this is desired though).
2656 */ 1861 */
>> 1862
2657 isize = i_size_read(inode); 1863 isize = i_size_read(inode);
2658 if (unlikely(iocb->ki_pos >= !! 1864 end_index = (isize - 1) >> PAGE_SHIFT;
2659 goto put_folios; !! 1865 if (unlikely(!isize || index > end_index)) {
2660 end_offset = min_t(loff_t, is !! 1866 put_page(page);
>> 1867 goto out;
>> 1868 }
>> 1869
>> 1870 /* nr is the maximum number of bytes to copy from this page */
>> 1871 nr = PAGE_SIZE;
>> 1872 if (index == end_index) {
>> 1873 nr = ((isize - 1) & ~PAGE_MASK) + 1;
>> 1874 if (nr <= offset) {
>> 1875 put_page(page);
>> 1876 goto out;
>> 1877 }
>> 1878 }
>> 1879 nr = nr - offset;
>> 1880
>> 1881 /* If users can be writing to this page using arbitrary
>> 1882 * virtual addresses, take care about potential aliasing
>> 1883 * before reading the page on the kernel side.
>> 1884 */
>> 1885 if (mapping_writably_mapped(mapping))
>> 1886 flush_dcache_page(page);
2661 1887
2662 /* 1888 /*
2663 * Once we start copying data !! 1889 * When a sequential read accesses a page several times,
2664 * cachelines that might be c !! 1890 * only mark it as accessed the first time.
2665 */ 1891 */
2666 writably_mapped = mapping_wri !! 1892 if (prev_index != index || offset != prev_offset)
>> 1893 mark_page_accessed(page);
>> 1894 prev_index = index;
2667 1895
2668 /* 1896 /*
2669 * When a read accesses the s !! 1897 * Ok, we have the page, and it's up-to-date, so
2670 * mark it as accessed the fi !! 1898 * now we can copy it to user space...
2671 */ 1899 */
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 1900
2684 if (end_offset < foli !! 1901 ret = copy_page_to_iter(page, offset, nr, iter);
2685 break; !! 1902 offset += ret;
2686 if (i > 0) !! 1903 index += offset >> PAGE_SHIFT;
2687 folio_mark_ac !! 1904 offset &= ~PAGE_MASK;
2688 /* !! 1905 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 1906
2696 copied = copy_folio_t !! 1907 put_page(page);
>> 1908 written += ret;
>> 1909 if (!iov_iter_count(iter))
>> 1910 goto out;
>> 1911 if (ret < nr) {
>> 1912 error = -EFAULT;
>> 1913 goto out;
>> 1914 }
>> 1915 continue;
2697 1916
2698 already_read += copie !! 1917 page_not_up_to_date:
2699 iocb->ki_pos += copie !! 1918 /* Get exclusive access to the page ... */
2700 last_pos = iocb->ki_p !! 1919 error = lock_page_killable(page);
>> 1920 if (unlikely(error))
>> 1921 goto readpage_error;
>> 1922
>> 1923 page_not_up_to_date_locked:
>> 1924 /* Did it get truncated before we got the lock? */
>> 1925 if (!page->mapping) {
>> 1926 unlock_page(page);
>> 1927 put_page(page);
>> 1928 continue;
>> 1929 }
2701 1930
2702 if (copied < bytes) { !! 1931 /* Did somebody else fill it already? */
2703 error = -EFAU !! 1932 if (PageUptodate(page)) {
2704 break; !! 1933 unlock_page(page);
>> 1934 goto page_ok;
>> 1935 }
>> 1936
>> 1937 readpage:
>> 1938 /*
>> 1939 * A previous I/O error may have been due to temporary
>> 1940 * failures, eg. multipath errors.
>> 1941 * PG_error will be set again if readpage fails.
>> 1942 */
>> 1943 ClearPageError(page);
>> 1944 /* Start the actual read. The read will unlock the page. */
>> 1945 error = mapping->a_ops->readpage(filp, page);
>> 1946
>> 1947 if (unlikely(error)) {
>> 1948 if (error == AOP_TRUNCATED_PAGE) {
>> 1949 put_page(page);
>> 1950 error = 0;
>> 1951 goto find_page;
2705 } 1952 }
>> 1953 goto readpage_error;
2706 } 1954 }
2707 put_folios: <<
2708 for (i = 0; i < folio_batch_c <<
2709 folio_put(fbatch.foli <<
2710 folio_batch_init(&fbatch); <<
2711 } while (iov_iter_count(iter) && iocb <<
2712 <<
2713 file_accessed(filp); <<
2714 ra->prev_pos = last_pos; <<
2715 return already_read ? already_read : <<
2716 } <<
2717 EXPORT_SYMBOL_GPL(filemap_read); <<
2718 1955
2719 int kiocb_write_and_wait(struct kiocb *iocb, !! 1956 if (!PageUptodate(page)) {
2720 { !! 1957 error = lock_page_killable(page);
2721 struct address_space *mapping = iocb- !! 1958 if (unlikely(error))
2722 loff_t pos = iocb->ki_pos; !! 1959 goto readpage_error;
2723 loff_t end = pos + count - 1; !! 1960 if (!PageUptodate(page)) {
2724 !! 1961 if (page->mapping == NULL) {
2725 if (iocb->ki_flags & IOCB_NOWAIT) { !! 1962 /*
2726 if (filemap_range_needs_write !! 1963 * invalidate_mapping_pages got it
2727 return -EAGAIN; !! 1964 */
2728 return 0; !! 1965 unlock_page(page);
2729 } !! 1966 put_page(page);
>> 1967 goto find_page;
>> 1968 }
>> 1969 unlock_page(page);
>> 1970 shrink_readahead_size_eio(filp, ra);
>> 1971 error = -EIO;
>> 1972 goto readpage_error;
>> 1973 }
>> 1974 unlock_page(page);
>> 1975 }
2730 1976
2731 return filemap_write_and_wait_range(m !! 1977 goto page_ok;
2732 } <<
2733 EXPORT_SYMBOL_GPL(kiocb_write_and_wait); <<
2734 1978
2735 int filemap_invalidate_pages(struct address_s !! 1979 readpage_error:
2736 loff_t pos, loff !! 1980 /* UHHUH! A synchronous read error occurred. Report it */
2737 { !! 1981 put_page(page);
2738 int ret; !! 1982 goto out;
2739 1983
2740 if (nowait) { !! 1984 no_cached_page:
2741 /* we could block if there ar !! 1985 /*
2742 if (filemap_range_has_page(ma !! 1986 * Ok, it wasn't cached, so we need to create a new
2743 return -EAGAIN; !! 1987 * page..
2744 } else { !! 1988 */
2745 ret = filemap_write_and_wait_ !! 1989 page = page_cache_alloc_cold(mapping);
2746 if (ret) !! 1990 if (!page) {
2747 return ret; !! 1991 error = -ENOMEM;
>> 1992 goto out;
>> 1993 }
>> 1994 error = add_to_page_cache_lru(page, mapping, index,
>> 1995 mapping_gfp_constraint(mapping, GFP_KERNEL));
>> 1996 if (error) {
>> 1997 put_page(page);
>> 1998 if (error == -EEXIST) {
>> 1999 error = 0;
>> 2000 goto find_page;
>> 2001 }
>> 2002 goto out;
>> 2003 }
>> 2004 goto readpage;
2748 } 2005 }
2749 2006
2750 /* !! 2007 out:
2751 * After a write we want buffered rea !! 2008 ra->prev_pos = prev_index;
2752 * the new data. We invalidate clean !! 2009 ra->prev_pos <<= PAGE_SHIFT;
2753 * about to write. We do this *befor !! 2010 ra->prev_pos |= prev_offset;
2754 * without clobbering -EIOCBQUEUED fr <<
2755 */ <<
2756 return invalidate_inode_pages2_range( <<
2757 <<
2758 } <<
2759 <<
2760 int kiocb_invalidate_pages(struct kiocb *iocb <<
2761 { <<
2762 struct address_space *mapping = iocb- <<
2763 2011
2764 return filemap_invalidate_pages(mappi !! 2012 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2765 iocb- !! 2013 file_accessed(filp);
2766 iocb- !! 2014 return written ? written : error;
2767 } 2015 }
2768 EXPORT_SYMBOL_GPL(kiocb_invalidate_pages); <<
2769 2016
2770 /** 2017 /**
2771 * generic_file_read_iter - generic filesyste 2018 * generic_file_read_iter - generic filesystem read routine
2772 * @iocb: kernel I/O control block 2019 * @iocb: kernel I/O control block
2773 * @iter: destination for the data read 2020 * @iter: destination for the data read
2774 * 2021 *
2775 * This is the "read_iter()" routine for all 2022 * This is the "read_iter()" routine for all filesystems
2776 * that can use the page cache directly. 2023 * 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: <<
2788 * * number of bytes copied, even for partial <<
2789 * * negative error code (or 0 if IOCB_NOIO) <<
2790 */ 2024 */
2791 ssize_t 2025 ssize_t
2792 generic_file_read_iter(struct kiocb *iocb, st 2026 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2793 { 2027 {
2794 size_t count = iov_iter_count(iter); !! 2028 struct file *file = iocb->ki_filp;
2795 ssize_t retval = 0; 2029 ssize_t retval = 0;
>> 2030 size_t count = iov_iter_count(iter);
2796 2031
2797 if (!count) 2032 if (!count)
2798 return 0; /* skip atime */ !! 2033 goto out; /* skip atime */
2799 2034
2800 if (iocb->ki_flags & IOCB_DIRECT) { 2035 if (iocb->ki_flags & IOCB_DIRECT) {
2801 struct file *file = iocb->ki_ <<
2802 struct address_space *mapping 2036 struct address_space *mapping = file->f_mapping;
2803 struct inode *inode = mapping 2037 struct inode *inode = mapping->host;
>> 2038 loff_t size;
2804 2039
2805 retval = kiocb_write_and_wait !! 2040 size = i_size_read(inode);
>> 2041 retval = filemap_write_and_wait_range(mapping, iocb->ki_pos,
>> 2042 iocb->ki_pos + count - 1);
2806 if (retval < 0) 2043 if (retval < 0)
2807 return retval; !! 2044 goto out;
>> 2045
2808 file_accessed(file); 2046 file_accessed(file);
2809 2047
2810 retval = mapping->a_ops->dire 2048 retval = mapping->a_ops->direct_IO(iocb, iter);
2811 if (retval >= 0) { 2049 if (retval >= 0) {
2812 iocb->ki_pos += retva 2050 iocb->ki_pos += retval;
2813 count -= retval; 2051 count -= retval;
2814 } 2052 }
2815 if (retval != -EIOCBQUEUED) !! 2053 iov_iter_revert(iter, count - iov_iter_count(iter));
2816 iov_iter_revert(iter, <<
2817 2054
2818 /* 2055 /*
2819 * Btrfs can have a short DIO 2056 * Btrfs can have a short DIO read if we encounter
2820 * compressed extents, so if 2057 * compressed extents, so if there was an error, or if
2821 * we've already read everyth 2058 * we've already read everything we wanted to, or if
2822 * there was a short read bec 2059 * there was a short read because we hit EOF, go ahead
2823 * and return. Otherwise fal 2060 * and return. Otherwise fallthrough to buffered io for
2824 * the rest of the read. Buf 2061 * the rest of the read. Buffered reads will not work for
2825 * DAX files, so don't bother 2062 * DAX files, so don't bother trying.
2826 */ 2063 */
2827 if (retval < 0 || !count || I !! 2064 if (retval < 0 || !count || iocb->ki_pos >= size ||
2828 return retval; !! 2065 IS_DAX(inode))
2829 if (iocb->ki_pos >= i_size_re !! 2066 goto out;
2830 return retval; <<
2831 } 2067 }
2832 2068
2833 return filemap_read(iocb, iter, retva !! 2069 retval = do_generic_file_read(file, &iocb->ki_pos, iter, retval);
>> 2070 out:
>> 2071 return retval;
2834 } 2072 }
2835 EXPORT_SYMBOL(generic_file_read_iter); 2073 EXPORT_SYMBOL(generic_file_read_iter);
2836 2074
2837 /* !! 2075 #ifdef CONFIG_MMU
2838 * Splice subpages from a folio into a pipe. !! 2076 /**
>> 2077 * page_cache_read - adds requested page to the page cache if not already there
>> 2078 * @file: file to read
>> 2079 * @offset: page index
>> 2080 * @gfp_mask: memory allocation flags
>> 2081 *
>> 2082 * This adds the requested page to the page cache if it isn't already there,
>> 2083 * and schedules an I/O to read in its contents from disk.
2839 */ 2084 */
2840 size_t splice_folio_into_pipe(struct pipe_ino !! 2085 static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
2841 struct folio *f <<
2842 { 2086 {
>> 2087 struct address_space *mapping = file->f_mapping;
2843 struct page *page; 2088 struct page *page;
2844 size_t spliced = 0, offset = offset_i !! 2089 int ret;
2845 <<
2846 page = folio_page(folio, offset / PAG <<
2847 size = min(size, folio_size(folio) - <<
2848 offset %= PAGE_SIZE; <<
2849 <<
2850 while (spliced < size && <<
2851 !pipe_full(pipe->head, pipe->t <<
2852 struct pipe_buffer *buf = pip <<
2853 size_t part = min_t(size_t, P <<
2854 <<
2855 *buf = (struct pipe_buffer) { <<
2856 .ops = &page_cache <<
2857 .page = page, <<
2858 .offset = offset, <<
2859 .len = part, <<
2860 }; <<
2861 folio_get(folio); <<
2862 pipe->head++; <<
2863 page++; <<
2864 spliced += part; <<
2865 offset = 0; <<
2866 } <<
2867 <<
2868 return spliced; <<
2869 } <<
2870 <<
2871 /** <<
2872 * filemap_splice_read - Splice data from a <<
2873 * @in: The file to read from <<
2874 * @ppos: Pointer to the file position to rea <<
2875 * @pipe: The pipe to splice into <<
2876 * @len: The amount to splice <<
2877 * @flags: The SPLICE_F_* flags <<
2878 * <<
2879 * This function gets folios from a file's pa <<
2880 * pipe. Readahead will be called as necessa <<
2881 * be used for blockdevs also. <<
2882 * <<
2883 * Return: On success, the number of bytes re <<
2884 * will be updated if appropriate; 0 will be <<
2885 * to be read; -EAGAIN will be returned if th <<
2886 * other negative error code will be returned <<
2887 * if the pipe has insufficient space, we rea <<
2888 * hole. <<
2889 */ <<
2890 ssize_t filemap_splice_read(struct file *in, <<
2891 struct pipe_inode <<
2892 size_t len, unsig <<
2893 { <<
2894 struct folio_batch fbatch; <<
2895 struct kiocb iocb; <<
2896 size_t total_spliced = 0, used, npage <<
2897 loff_t isize, end_offset; <<
2898 bool writably_mapped; <<
2899 int i, error = 0; <<
2900 <<
2901 if (unlikely(*ppos >= in->f_mapping-> <<
2902 return 0; <<
2903 <<
2904 init_sync_kiocb(&iocb, in); <<
2905 iocb.ki_pos = *ppos; <<
2906 <<
2907 /* Work out how much data we can actu <<
2908 used = pipe_occupancy(pipe->head, pip <<
2909 npages = max_t(ssize_t, pipe->max_usa <<
2910 len = min_t(size_t, len, npages * PAG <<
2911 <<
2912 folio_batch_init(&fbatch); <<
2913 <<
2914 do { <<
2915 cond_resched(); <<
2916 <<
2917 if (*ppos >= i_size_read(in-> <<
2918 break; <<
2919 <<
2920 iocb.ki_pos = *ppos; <<
2921 error = filemap_get_pages(&io <<
2922 if (error < 0) <<
2923 break; <<
2924 <<
2925 /* <<
2926 * i_size must be checked aft <<
2927 * <<
2928 * Checking i_size after the <<
2929 * the correct value for "nr" <<
2930 * part of the page is not co <<
2931 * another truncate extends t <<
2932 */ <<
2933 isize = i_size_read(in->f_map <<
2934 if (unlikely(*ppos >= isize)) <<
2935 break; <<
2936 end_offset = min_t(loff_t, is <<
2937 <<
2938 /* <<
2939 * Once we start copying data <<
2940 * cachelines that might be c <<
2941 */ <<
2942 writably_mapped = mapping_wri <<
2943 <<
2944 for (i = 0; i < folio_batch_c <<
2945 struct folio *folio = <<
2946 size_t n; <<
2947 <<
2948 if (folio_pos(folio) <<
2949 goto out; <<
2950 folio_mark_accessed(f <<
2951 <<
2952 /* <<
2953 * If users can be wr <<
2954 * virtual addresses, <<
2955 * before reading the <<
2956 */ <<
2957 if (writably_mapped) <<
2958 flush_dcache_ <<
2959 <<
2960 n = min_t(loff_t, len <<
2961 n = splice_folio_into <<
2962 if (!n) <<
2963 goto out; <<
2964 len -= n; <<
2965 total_spliced += n; <<
2966 *ppos += n; <<
2967 in->f_ra.prev_pos = * <<
2968 if (pipe_full(pipe->h <<
2969 goto out; <<
2970 } <<
2971 <<
2972 folio_batch_release(&fbatch); <<
2973 } while (len); <<
2974 <<
2975 out: <<
2976 folio_batch_release(&fbatch); <<
2977 file_accessed(in); <<
2978 <<
2979 return total_spliced ? total_spliced <<
2980 } <<
2981 EXPORT_SYMBOL(filemap_splice_read); <<
2982 <<
2983 static inline loff_t folio_seek_hole_data(str <<
2984 struct address_space *mapping <<
2985 loff_t start, loff_t end, boo <<
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 2090
3003 do { 2091 do {
3004 if (ops->is_partially_uptodat !! 2092 page = __page_cache_alloc(gfp_mask|__GFP_COLD);
3005 !! 2093 if (!page)
3006 break; !! 2094 return -ENOMEM;
3007 start = (start + bsz) & ~(bsz !! 2095
3008 offset += bsz; !! 2096 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL);
3009 } while (offset < folio_size(folio)); !! 2097 if (ret == 0)
3010 unlock: !! 2098 ret = mapping->a_ops->readpage(file, page);
3011 folio_unlock(folio); !! 2099 else if (ret == -EEXIST)
3012 rcu_read_lock(); !! 2100 ret = 0; /* losing race to add is OK */
3013 return start; <<
3014 } <<
3015 2101
3016 static inline size_t seek_folio_size(struct x !! 2102 put_page(page);
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 2103
3052 rcu_read_lock(); !! 2104 } while (ret == AOP_TRUNCATED_PAGE);
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 2105
3063 seek_size = seek_folio_size(& !! 2106 return ret;
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 } 2107 }
3086 2108
3087 #ifdef CONFIG_MMU <<
3088 #define MMAP_LOTSAMISS (100) 2109 #define MMAP_LOTSAMISS (100)
3089 /* <<
3090 * lock_folio_maybe_drop_mmap - lock the page <<
3091 * @vmf - the vm_fault for this fault. <<
3092 * @folio - the folio to lock. <<
3093 * @fpin - the pointer to the file we may pin <<
3094 * <<
3095 * This works similar to lock_folio_or_retry <<
3096 * mmap_lock. It differs in that it actually <<
3097 * if it returns 1 and 0 if it couldn't lock <<
3098 * to drop the mmap_lock then fpin will point <<
3099 * needs to be fput()'ed at a later point. <<
3100 */ <<
3101 static int lock_folio_maybe_drop_mmap(struct <<
3102 struct f <<
3103 { <<
3104 if (folio_trylock(folio)) <<
3105 return 1; <<
3106 <<
3107 /* <<
3108 * NOTE! This will make us return wit <<
3109 * the fault lock still held. That's <<
3110 * is supposed to work. We have way t <<
3111 */ <<
3112 if (vmf->flags & FAULT_FLAG_RETRY_NOW <<
3113 return 0; <<
3114 <<
3115 *fpin = maybe_unlock_mmap_for_io(vmf, <<
3116 if (vmf->flags & FAULT_FLAG_KILLABLE) <<
3117 if (__folio_lock_killable(fol <<
3118 /* <<
3119 * We didn't have the <<
3120 * fault lock, but al <<
3121 * for fatal signals <<
3122 * so we need to drop <<
3123 * return 0 if we don <<
3124 */ <<
3125 if (*fpin == NULL) <<
3126 release_fault <<
3127 return 0; <<
3128 } <<
3129 } else <<
3130 __folio_lock(folio); <<
3131 <<
3132 return 1; <<
3133 } <<
3134 2110
3135 /* 2111 /*
3136 * Synchronous readahead happens when we don' !! 2112 * Synchronous readahead happens when we don't even find
3137 * cache at all. We don't want to perform IO !! 2113 * a page in the page cache at all.
3138 * to drop the mmap sem we return the file th <<
3139 * that. If we didn't pin a file then we ret <<
3140 * returned needs to be fput()'ed when we're <<
3141 */ 2114 */
3142 static struct file *do_sync_mmap_readahead(st !! 2115 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
>> 2116 struct file_ra_state *ra,
>> 2117 struct file *file,
>> 2118 pgoff_t offset)
3143 { 2119 {
3144 struct file *file = vmf->vma->vm_file <<
3145 struct file_ra_state *ra = &file->f_r <<
3146 struct address_space *mapping = file- 2120 struct address_space *mapping = file->f_mapping;
3147 DEFINE_READAHEAD(ractl, file, ra, map <<
3148 struct file *fpin = NULL; <<
3149 unsigned long vm_flags = vmf->vma->vm <<
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 2121
3170 /* If we don't want any read-ahead, d 2122 /* If we don't want any read-ahead, don't bother */
3171 if (vm_flags & VM_RAND_READ) !! 2123 if (vma->vm_flags & VM_RAND_READ)
3172 return fpin; !! 2124 return;
3173 if (!ra->ra_pages) 2125 if (!ra->ra_pages)
3174 return fpin; !! 2126 return;
3175 2127
3176 if (vm_flags & VM_SEQ_READ) { !! 2128 if (vma->vm_flags & VM_SEQ_READ) {
3177 fpin = maybe_unlock_mmap_for_ !! 2129 page_cache_sync_readahead(mapping, ra, file, offset,
3178 page_cache_sync_ra(&ractl, ra !! 2130 ra->ra_pages);
3179 return fpin; !! 2131 return;
3180 } 2132 }
3181 2133
3182 /* Avoid banging the cache line if no 2134 /* Avoid banging the cache line if not needed */
3183 mmap_miss = READ_ONCE(ra->mmap_miss); !! 2135 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
3184 if (mmap_miss < MMAP_LOTSAMISS * 10) !! 2136 ra->mmap_miss++;
3185 WRITE_ONCE(ra->mmap_miss, ++m <<
3186 2137
3187 /* 2138 /*
3188 * Do we miss much more than hit in t 2139 * Do we miss much more than hit in this file? If so,
3189 * stop bothering with read-ahead. It 2140 * stop bothering with read-ahead. It will only hurt.
3190 */ 2141 */
3191 if (mmap_miss > MMAP_LOTSAMISS) !! 2142 if (ra->mmap_miss > MMAP_LOTSAMISS)
3192 return fpin; !! 2143 return;
3193 2144
3194 /* 2145 /*
3195 * mmap read-around 2146 * mmap read-around
3196 */ 2147 */
3197 fpin = maybe_unlock_mmap_for_io(vmf, !! 2148 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
3198 ra->start = max_t(long, 0, vmf->pgoff <<
3199 ra->size = ra->ra_pages; 2149 ra->size = ra->ra_pages;
3200 ra->async_size = ra->ra_pages / 4; 2150 ra->async_size = ra->ra_pages / 4;
3201 ractl._index = ra->start; !! 2151 ra_submit(ra, mapping, file);
3202 page_cache_ra_order(&ractl, ra, 0); <<
3203 return fpin; <<
3204 } 2152 }
3205 2153
3206 /* 2154 /*
3207 * Asynchronous readahead happens when we fin 2155 * Asynchronous readahead happens when we find the page and PG_readahead,
3208 * so we want to possibly extend the readahea !! 2156 * so we want to possibly extend the readahead further..
3209 * was pinned if we have to drop the mmap_loc <<
3210 */ 2157 */
3211 static struct file *do_async_mmap_readahead(s !! 2158 static void do_async_mmap_readahead(struct vm_area_struct *vma,
3212 s !! 2159 struct file_ra_state *ra,
>> 2160 struct file *file,
>> 2161 struct page *page,
>> 2162 pgoff_t offset)
3213 { 2163 {
3214 struct file *file = vmf->vma->vm_file !! 2164 struct address_space *mapping = file->f_mapping;
3215 struct file_ra_state *ra = &file->f_r <<
3216 DEFINE_READAHEAD(ractl, file, ra, fil <<
3217 struct file *fpin = NULL; <<
3218 unsigned int mmap_miss; <<
3219 2165
3220 /* If we don't want any read-ahead, d 2166 /* If we don't want any read-ahead, don't bother */
3221 if (vmf->vma->vm_flags & VM_RAND_READ !! 2167 if (vma->vm_flags & VM_RAND_READ)
3222 return fpin; !! 2168 return;
3223 !! 2169 if (ra->mmap_miss > 0)
3224 mmap_miss = READ_ONCE(ra->mmap_miss); !! 2170 ra->mmap_miss--;
3225 if (mmap_miss) !! 2171 if (PageReadahead(page))
3226 WRITE_ONCE(ra->mmap_miss, --m !! 2172 page_cache_async_readahead(mapping, ra, file,
3227 !! 2173 page, offset, ra->ra_pages);
3228 if (folio_test_readahead(folio)) { <<
3229 fpin = maybe_unlock_mmap_for_ <<
3230 page_cache_async_ra(&ractl, f <<
3231 } <<
3232 return fpin; <<
3233 } <<
3234 <<
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 } 2174 }
3277 2175
3278 /** 2176 /**
3279 * filemap_fault - read in file data for page 2177 * filemap_fault - read in file data for page fault handling
3280 * @vmf: struct vm_fault containing de 2178 * @vmf: struct vm_fault containing details of the fault
3281 * 2179 *
3282 * filemap_fault() is invoked via the vma ope 2180 * filemap_fault() is invoked via the vma operations vector for a
3283 * mapped memory region to read in file data 2181 * mapped memory region to read in file data during a page fault.
3284 * 2182 *
3285 * The goto's are kind of ugly, but this stre 2183 * 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 2184 * it in the page cache, and handles the special cases reasonably without
3287 * having a lot of duplicated code. 2185 * having a lot of duplicated code.
3288 * 2186 *
3289 * vma->vm_mm->mmap_lock must be held on entr !! 2187 * vma->vm_mm->mmap_sem must be held on entry.
3290 * 2188 *
3291 * If our return value has VM_FAULT_RETRY set !! 2189 * If our return value has VM_FAULT_RETRY set, it's because
3292 * may be dropped before doing I/O or by lock !! 2190 * lock_page_or_retry() returned 0.
>> 2191 * The mmap_sem has usually been released in this case.
>> 2192 * See __lock_page_or_retry() for the exception.
3293 * 2193 *
3294 * If our return value does not have VM_FAULT !! 2194 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
3295 * has not been released. 2195 * has not been released.
3296 * 2196 *
3297 * We never return with VM_FAULT_RETRY and a 2197 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3298 * <<
3299 * Return: bitwise-OR of %VM_FAULT_ codes. <<
3300 */ 2198 */
3301 vm_fault_t filemap_fault(struct vm_fault *vmf !! 2199 int filemap_fault(struct vm_fault *vmf)
3302 { 2200 {
3303 int error; 2201 int error;
3304 struct file *file = vmf->vma->vm_file 2202 struct file *file = vmf->vma->vm_file;
3305 struct file *fpin = NULL; <<
3306 struct address_space *mapping = file- 2203 struct address_space *mapping = file->f_mapping;
>> 2204 struct file_ra_state *ra = &file->f_ra;
3307 struct inode *inode = mapping->host; 2205 struct inode *inode = mapping->host;
3308 pgoff_t max_idx, index = vmf->pgoff; !! 2206 pgoff_t offset = vmf->pgoff;
3309 struct folio *folio; !! 2207 pgoff_t max_off;
3310 vm_fault_t ret = 0; !! 2208 struct page *page;
3311 bool mapping_locked = false; !! 2209 int ret = 0;
3312 2210
3313 max_idx = DIV_ROUND_UP(i_size_read(in !! 2211 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3314 if (unlikely(index >= max_idx)) !! 2212 if (unlikely(offset >= max_off))
3315 return VM_FAULT_SIGBUS; 2213 return VM_FAULT_SIGBUS;
3316 2214
3317 trace_mm_filemap_fault(mapping, index <<
3318 <<
3319 /* 2215 /*
3320 * Do we have something in the page c 2216 * Do we have something in the page cache already?
3321 */ 2217 */
3322 folio = filemap_get_folio(mapping, in !! 2218 page = find_get_page(mapping, offset);
3323 if (likely(!IS_ERR(folio))) { !! 2219 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
3324 /* 2220 /*
3325 * We found the page, so try !! 2221 * We found the page, so try async readahead before
3326 * the lock. !! 2222 * waiting for the lock.
3327 */ 2223 */
3328 if (!(vmf->flags & FAULT_FLAG !! 2224 do_async_mmap_readahead(vmf->vma, ra, file, page, offset);
3329 fpin = do_async_mmap_ !! 2225 } 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 2226 /* No page in the page cache at all */
>> 2227 do_sync_mmap_readahead(vmf->vma, ra, file, offset);
3340 count_vm_event(PGMAJFAULT); 2228 count_vm_event(PGMAJFAULT);
3341 count_memcg_event_mm(vmf->vma !! 2229 mem_cgroup_count_vm_event(vmf->vma->vm_mm, PGMAJFAULT);
3342 ret = VM_FAULT_MAJOR; 2230 ret = VM_FAULT_MAJOR;
3343 fpin = do_sync_mmap_readahead <<
3344 retry_find: 2231 retry_find:
3345 /* !! 2232 page = find_get_page(mapping, offset);
3346 * See comment in filemap_cre !! 2233 if (!page)
3347 * invalidate_lock !! 2234 goto no_cached_page;
3348 */ <<
3349 if (!mapping_locked) { <<
3350 filemap_invalidate_lo <<
3351 mapping_locked = true <<
3352 } <<
3353 folio = __filemap_get_folio(m <<
3354 FGP <<
3355 vmf <<
3356 if (IS_ERR(folio)) { <<
3357 if (fpin) <<
3358 goto out_retr <<
3359 filemap_invalidate_un <<
3360 return VM_FAULT_OOM; <<
3361 } <<
3362 } 2235 }
3363 2236
3364 if (!lock_folio_maybe_drop_mmap(vmf, !! 2237 if (!lock_page_or_retry(page, vmf->vma->vm_mm, vmf->flags)) {
3365 goto out_retry; !! 2238 put_page(page);
>> 2239 return ret | VM_FAULT_RETRY;
>> 2240 }
3366 2241
3367 /* Did it get truncated? */ 2242 /* Did it get truncated? */
3368 if (unlikely(folio->mapping != mappin !! 2243 if (unlikely(page->mapping != mapping)) {
3369 folio_unlock(folio); !! 2244 unlock_page(page);
3370 folio_put(folio); !! 2245 put_page(page);
3371 goto retry_find; 2246 goto retry_find;
3372 } 2247 }
3373 VM_BUG_ON_FOLIO(!folio_contains(folio !! 2248 VM_BUG_ON_PAGE(page->index != offset, page);
3374 2249
3375 /* 2250 /*
3376 * We have a locked folio in the page !! 2251 * 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 !! 2252 * that it's up-to-date. If not, it is going to be due to an error.
3378 * or because readahead was otherwise <<
3379 */ 2253 */
3380 if (unlikely(!folio_test_uptodate(fol !! 2254 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; 2255 goto page_not_uptodate;
3399 } <<
3400 <<
3401 /* <<
3402 * We've made it this far and we had <<
3403 * time to return to the upper layer <<
3404 * redo the fault. <<
3405 */ <<
3406 if (fpin) { <<
3407 folio_unlock(folio); <<
3408 goto out_retry; <<
3409 } <<
3410 if (mapping_locked) <<
3411 filemap_invalidate_unlock_sha <<
3412 2256
3413 /* 2257 /*
3414 * Found the page and have a referenc 2258 * Found the page and have a reference on it.
3415 * We must recheck i_size under page 2259 * We must recheck i_size under page lock.
3416 */ 2260 */
3417 max_idx = DIV_ROUND_UP(i_size_read(in !! 2261 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3418 if (unlikely(index >= max_idx)) { !! 2262 if (unlikely(offset >= max_off)) {
3419 folio_unlock(folio); !! 2263 unlock_page(page);
3420 folio_put(folio); !! 2264 put_page(page);
3421 return VM_FAULT_SIGBUS; 2265 return VM_FAULT_SIGBUS;
3422 } 2266 }
3423 2267
3424 vmf->page = folio_file_page(folio, in !! 2268 vmf->page = page;
3425 return ret | VM_FAULT_LOCKED; 2269 return ret | VM_FAULT_LOCKED;
3426 2270
>> 2271 no_cached_page:
>> 2272 /*
>> 2273 * We're only likely to ever get here if MADV_RANDOM is in
>> 2274 * effect.
>> 2275 */
>> 2276 error = page_cache_read(file, offset, vmf->gfp_mask);
>> 2277
>> 2278 /*
>> 2279 * The page we want has now been added to the page cache.
>> 2280 * In the unlikely event that someone removed it in the
>> 2281 * meantime, we'll just come back here and read it again.
>> 2282 */
>> 2283 if (error >= 0)
>> 2284 goto retry_find;
>> 2285
>> 2286 /*
>> 2287 * An error return from page_cache_read can result if the
>> 2288 * system is low on memory, or a problem occurs while trying
>> 2289 * to schedule I/O.
>> 2290 */
>> 2291 if (error == -ENOMEM)
>> 2292 return VM_FAULT_OOM;
>> 2293 return VM_FAULT_SIGBUS;
>> 2294
3427 page_not_uptodate: 2295 page_not_uptodate:
3428 /* 2296 /*
3429 * Umm, take care of errors if the pa 2297 * Umm, take care of errors if the page isn't up-to-date.
3430 * Try to re-read it _once_. We do th 2298 * Try to re-read it _once_. We do this synchronously,
3431 * because there really aren't any pe 2299 * because there really aren't any performance issues here
3432 * and we need to check for errors. 2300 * and we need to check for errors.
3433 */ 2301 */
3434 fpin = maybe_unlock_mmap_for_io(vmf, !! 2302 ClearPageError(page);
3435 error = filemap_read_folio(file, mapp !! 2303 error = mapping->a_ops->readpage(file, page);
3436 if (fpin) !! 2304 if (!error) {
3437 goto out_retry; !! 2305 wait_on_page_locked(page);
3438 folio_put(folio); !! 2306 if (!PageUptodate(page))
>> 2307 error = -EIO;
>> 2308 }
>> 2309 put_page(page);
3439 2310
3440 if (!error || error == AOP_TRUNCATED_ 2311 if (!error || error == AOP_TRUNCATED_PAGE)
3441 goto retry_find; 2312 goto retry_find;
3442 filemap_invalidate_unlock_shared(mapp <<
3443 2313
>> 2314 /* Things didn't work out. Return zero to tell the mm layer so. */
>> 2315 shrink_readahead_size_eio(file, ra);
3444 return VM_FAULT_SIGBUS; 2316 return VM_FAULT_SIGBUS;
3445 <<
3446 out_retry: <<
3447 /* <<
3448 * We dropped the mmap_lock, we need <<
3449 * re-find the vma and come back and <<
3450 * page. <<
3451 */ <<
3452 if (!IS_ERR(folio)) <<
3453 folio_put(folio); <<
3454 if (mapping_locked) <<
3455 filemap_invalidate_unlock_sha <<
3456 if (fpin) <<
3457 fput(fpin); <<
3458 return ret | VM_FAULT_RETRY; <<
3459 } 2317 }
3460 EXPORT_SYMBOL(filemap_fault); 2318 EXPORT_SYMBOL(filemap_fault);
3461 2319
3462 static bool filemap_map_pmd(struct vm_fault * !! 2320 void filemap_map_pages(struct vm_fault *vmf,
3463 pgoff_t start) !! 2321 pgoff_t start_pgoff, pgoff_t end_pgoff)
3464 { 2322 {
3465 struct mm_struct *mm = vmf->vma->vm_m !! 2323 struct radix_tree_iter iter;
3466 !! 2324 void **slot;
3467 /* Huge page is mapped? No need to pr !! 2325 struct file *file = vmf->vma->vm_file;
3468 if (pmd_trans_huge(*vmf->pmd)) { !! 2326 struct address_space *mapping = file->f_mapping;
3469 folio_unlock(folio); !! 2327 pgoff_t last_pgoff = start_pgoff;
3470 folio_put(folio); !! 2328 unsigned long max_idx;
3471 return true; !! 2329 struct page *head, *page;
3472 } <<
3473 2330
3474 if (pmd_none(*vmf->pmd) && folio_test !! 2331 rcu_read_lock();
3475 struct page *page = folio_fil !! 2332 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter,
3476 vm_fault_t ret = do_set_pmd(v !! 2333 start_pgoff) {
3477 if (!ret) { !! 2334 if (iter.index > end_pgoff)
3478 /* The page is mapped !! 2335 break;
3479 folio_unlock(folio); !! 2336 repeat:
3480 return true; !! 2337 page = radix_tree_deref_slot(slot);
>> 2338 if (unlikely(!page))
>> 2339 goto next;
>> 2340 if (radix_tree_exception(page)) {
>> 2341 if (radix_tree_deref_retry(page)) {
>> 2342 slot = radix_tree_iter_retry(&iter);
>> 2343 continue;
>> 2344 }
>> 2345 goto next;
3481 } 2346 }
3482 } <<
3483 2347
3484 if (pmd_none(*vmf->pmd) && vmf->preal !! 2348 head = compound_head(page);
3485 pmd_install(mm, vmf->pmd, &vm !! 2349 if (!page_cache_get_speculative(head))
>> 2350 goto repeat;
3486 2351
3487 return false; !! 2352 /* The page was split under us? */
3488 } !! 2353 if (compound_head(page) != head) {
>> 2354 put_page(head);
>> 2355 goto repeat;
>> 2356 }
3489 2357
3490 static struct folio *next_uptodate_folio(stru !! 2358 /* Has the page moved? */
3491 struct address_space *mapping !! 2359 if (unlikely(page != *slot)) {
3492 { !! 2360 put_page(head);
3493 struct folio *folio = xas_next_entry( !! 2361 goto repeat;
3494 unsigned long max_idx; !! 2362 }
3495 2363
3496 do { !! 2364 if (!PageUptodate(page) ||
3497 if (!folio) !! 2365 PageReadahead(page) ||
3498 return NULL; !! 2366 PageHWPoison(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; <<
3507 /* Has the page moved or been <<
3508 if (unlikely(folio != xas_rel <<
3509 goto skip; <<
3510 if (!folio_test_uptodate(foli <<
3511 goto skip; 2367 goto skip;
3512 if (!folio_trylock(folio)) !! 2368 if (!trylock_page(page))
3513 goto skip; 2369 goto skip;
3514 if (folio->mapping != mapping !! 2370
3515 goto unlock; !! 2371 if (page->mapping != mapping || !PageUptodate(page))
3516 if (!folio_test_uptodate(foli <<
3517 goto unlock; 2372 goto unlock;
>> 2373
3518 max_idx = DIV_ROUND_UP(i_size 2374 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3519 if (xas->xa_index >= max_idx) !! 2375 if (page->index >= max_idx)
3520 goto unlock; 2376 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 2377
3549 /* !! 2378 if (file->f_ra.mmap_miss > 0)
3550 * If there are too many foli !! 2379 file->f_ra.mmap_miss--;
3551 * in a file, they will proba <<
3552 * In such situation, read-ah <<
3553 * Don't decrease mmap_miss i <<
3554 * we can stop read-ahead. <<
3555 */ <<
3556 if (!folio_test_workingset(fo <<
3557 (*mmap_miss)++; <<
3558 2380
3559 /* !! 2381 vmf->address += (iter.index - last_pgoff) << PAGE_SHIFT;
3560 * NOTE: If there're PTE mark !! 2382 if (vmf->pte)
3561 * handled in the specific fa !! 2383 vmf->pte += iter.index - last_pgoff;
3562 * fault-around logic. !! 2384 last_pgoff = iter.index;
3563 */ !! 2385 if (alloc_set_pte(vmf, NULL, page))
3564 if (!pte_none(ptep_get(&vmf-> !! 2386 goto unlock;
3565 goto skip; !! 2387 unlock_page(page);
3566 !! 2388 goto next;
3567 count++; !! 2389 unlock:
3568 continue; !! 2390 unlock_page(page);
3569 skip: 2391 skip:
3570 if (count) { !! 2392 put_page(page);
3571 set_pte_range(vmf, fo !! 2393 next:
3572 *rss += count; !! 2394 /* Huge page is mapped? No need to proceed. */
3573 folio_ref_add(folio, !! 2395 if (pmd_trans_huge(*vmf->pmd))
3574 if (in_range(vmf->add !! 2396 break;
3575 ret = VM_FAUL !! 2397 if (iter.index == end_pgoff)
3576 } !! 2398 break;
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 <<
3654 addr = vma->vm_start + ((start_pgoff <<
3655 vmf->pte = pte_offset_map_lock(vma->v <<
3656 if (!vmf->pte) { <<
3657 folio_unlock(folio); <<
3658 folio_put(folio); <<
3659 goto out; <<
3660 } 2399 }
3661 <<
3662 file_end = DIV_ROUND_UP(i_size_read(m <<
3663 if (end_pgoff > file_end) <<
3664 end_pgoff = file_end; <<
3665 <<
3666 folio_type = mm_counter_file(folio); <<
3667 do { <<
3668 unsigned long end; <<
3669 <<
3670 addr += (xas.xa_index - last_ <<
3671 vmf->pte += xas.xa_index - la <<
3672 last_pgoff = xas.xa_index; <<
3673 end = folio_next_index(folio) <<
3674 nr_pages = min(end, end_pgoff <<
3675 <<
3676 if (!folio_test_large(folio)) <<
3677 ret |= filemap_map_or <<
3678 folio <<
3679 else <<
3680 ret |= filemap_map_fo <<
3681 xas.x <<
3682 nr_pa <<
3683 <<
3684 folio_unlock(folio); <<
3685 folio_put(folio); <<
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(); 2400 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 } 2401 }
3701 EXPORT_SYMBOL(filemap_map_pages); 2402 EXPORT_SYMBOL(filemap_map_pages);
3702 2403
3703 vm_fault_t filemap_page_mkwrite(struct vm_fau !! 2404 int filemap_page_mkwrite(struct vm_fault *vmf)
3704 { 2405 {
3705 struct address_space *mapping = vmf-> !! 2406 struct page *page = vmf->page;
3706 struct folio *folio = page_folio(vmf- !! 2407 struct inode *inode = file_inode(vmf->vma->vm_file);
3707 vm_fault_t ret = VM_FAULT_LOCKED; !! 2408 int ret = VM_FAULT_LOCKED;
3708 2409
3709 sb_start_pagefault(mapping->host->i_s !! 2410 sb_start_pagefault(inode->i_sb);
3710 file_update_time(vmf->vma->vm_file); 2411 file_update_time(vmf->vma->vm_file);
3711 folio_lock(folio); !! 2412 lock_page(page);
3712 if (folio->mapping != mapping) { !! 2413 if (page->mapping != inode->i_mapping) {
3713 folio_unlock(folio); !! 2414 unlock_page(page);
3714 ret = VM_FAULT_NOPAGE; 2415 ret = VM_FAULT_NOPAGE;
3715 goto out; 2416 goto out;
3716 } 2417 }
3717 /* 2418 /*
3718 * We mark the folio dirty already he !! 2419 * We mark the page dirty already here so that when freeze is in
3719 * progress, we are guaranteed that w 2420 * progress, we are guaranteed that writeback during freezing will
3720 * see the dirty folio and writeprote !! 2421 * see the dirty page and writeprotect it again.
3721 */ 2422 */
3722 folio_mark_dirty(folio); !! 2423 set_page_dirty(page);
3723 folio_wait_stable(folio); !! 2424 wait_for_stable_page(page);
3724 out: 2425 out:
3725 sb_end_pagefault(mapping->host->i_sb) !! 2426 sb_end_pagefault(inode->i_sb);
3726 return ret; 2427 return ret;
3727 } 2428 }
>> 2429 EXPORT_SYMBOL(filemap_page_mkwrite);
3728 2430
3729 const struct vm_operations_struct generic_fil 2431 const struct vm_operations_struct generic_file_vm_ops = {
3730 .fault = filemap_fault, 2432 .fault = filemap_fault,
3731 .map_pages = filemap_map_pages, 2433 .map_pages = filemap_map_pages,
3732 .page_mkwrite = filemap_page_mkwrit 2434 .page_mkwrite = filemap_page_mkwrite,
3733 }; 2435 };
3734 2436
3735 /* This is used for a general mmap of a disk 2437 /* This is used for a general mmap of a disk file */
3736 2438
3737 int generic_file_mmap(struct file *file, stru !! 2439 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
3738 { 2440 {
3739 struct address_space *mapping = file- 2441 struct address_space *mapping = file->f_mapping;
3740 2442
3741 if (!mapping->a_ops->read_folio) !! 2443 if (!mapping->a_ops->readpage)
3742 return -ENOEXEC; 2444 return -ENOEXEC;
3743 file_accessed(file); 2445 file_accessed(file);
3744 vma->vm_ops = &generic_file_vm_ops; 2446 vma->vm_ops = &generic_file_vm_ops;
3745 return 0; 2447 return 0;
3746 } 2448 }
3747 2449
3748 /* 2450 /*
3749 * This is for filesystems which do not imple 2451 * This is for filesystems which do not implement ->writepage.
3750 */ 2452 */
3751 int generic_file_readonly_mmap(struct file *f 2453 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3752 { 2454 {
3753 if (vma_is_shared_maywrite(vma)) !! 2455 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3754 return -EINVAL; 2456 return -EINVAL;
3755 return generic_file_mmap(file, vma); 2457 return generic_file_mmap(file, vma);
3756 } 2458 }
3757 #else 2459 #else
3758 vm_fault_t filemap_page_mkwrite(struct vm_fau !! 2460 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
3759 { <<
3760 return VM_FAULT_SIGBUS; <<
3761 } <<
3762 int generic_file_mmap(struct file *file, stru <<
3763 { 2461 {
3764 return -ENOSYS; 2462 return -ENOSYS;
3765 } 2463 }
3766 int generic_file_readonly_mmap(struct file *f !! 2464 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
3767 { 2465 {
3768 return -ENOSYS; 2466 return -ENOSYS;
3769 } 2467 }
3770 #endif /* CONFIG_MMU */ 2468 #endif /* CONFIG_MMU */
3771 2469
3772 EXPORT_SYMBOL(filemap_page_mkwrite); <<
3773 EXPORT_SYMBOL(generic_file_mmap); 2470 EXPORT_SYMBOL(generic_file_mmap);
3774 EXPORT_SYMBOL(generic_file_readonly_mmap); 2471 EXPORT_SYMBOL(generic_file_readonly_mmap);
3775 2472
3776 static struct folio *do_read_cache_folio(stru !! 2473 static struct page *wait_on_page_read(struct page *page)
3777 pgoff_t index, filler_t fille <<
3778 { 2474 {
3779 struct folio *folio; !! 2475 if (!IS_ERR(page)) {
3780 int err; !! 2476 wait_on_page_locked(page);
>> 2477 if (!PageUptodate(page)) {
>> 2478 put_page(page);
>> 2479 page = ERR_PTR(-EIO);
>> 2480 }
>> 2481 }
>> 2482 return page;
>> 2483 }
3781 2484
3782 if (!filler) !! 2485 static struct page *do_read_cache_page(struct address_space *mapping,
3783 filler = mapping->a_ops->read !! 2486 pgoff_t index,
>> 2487 int (*filler)(void *, struct page *),
>> 2488 void *data,
>> 2489 gfp_t gfp)
>> 2490 {
>> 2491 struct page *page;
>> 2492 int err;
3784 repeat: 2493 repeat:
3785 folio = filemap_get_folio(mapping, in !! 2494 page = find_get_page(mapping, index);
3786 if (IS_ERR(folio)) { !! 2495 if (!page) {
3787 folio = filemap_alloc_folio(g !! 2496 page = __page_cache_alloc(gfp | __GFP_COLD);
3788 m !! 2497 if (!page)
3789 if (!folio) <<
3790 return ERR_PTR(-ENOME 2498 return ERR_PTR(-ENOMEM);
3791 index = mapping_align_index(m !! 2499 err = add_to_page_cache_lru(page, mapping, index, gfp);
3792 err = filemap_add_folio(mappi <<
3793 if (unlikely(err)) { 2500 if (unlikely(err)) {
3794 folio_put(folio); !! 2501 put_page(page);
3795 if (err == -EEXIST) 2502 if (err == -EEXIST)
3796 goto repeat; 2503 goto repeat;
3797 /* Presumably ENOMEM !! 2504 /* Presumably ENOMEM for radix tree node */
3798 return ERR_PTR(err); 2505 return ERR_PTR(err);
3799 } 2506 }
3800 2507
3801 goto filler; !! 2508 filler:
>> 2509 err = filler(data, page);
>> 2510 if (err < 0) {
>> 2511 put_page(page);
>> 2512 return ERR_PTR(err);
>> 2513 }
>> 2514
>> 2515 page = wait_on_page_read(page);
>> 2516 if (IS_ERR(page))
>> 2517 return page;
>> 2518 goto out;
3802 } 2519 }
3803 if (folio_test_uptodate(folio)) !! 2520 if (PageUptodate(page))
3804 goto out; 2521 goto out;
3805 2522
3806 if (!folio_trylock(folio)) { !! 2523 /*
3807 folio_put_wait_locked(folio, !! 2524 * Page is not up to date and may be locked due one of the following
3808 goto repeat; !! 2525 * case a: Page is being filled and the page lock is held
3809 } !! 2526 * case b: Read/write error clearing the page uptodate status
>> 2527 * case c: Truncation in progress (page locked)
>> 2528 * case d: Reclaim in progress
>> 2529 *
>> 2530 * Case a, the page will be up to date when the page is unlocked.
>> 2531 * There is no need to serialise on the page lock here as the page
>> 2532 * is pinned so the lock gives no additional protection. Even if the
>> 2533 * the page is truncated, the data is still valid if PageUptodate as
>> 2534 * it's a race vs truncate race.
>> 2535 * Case b, the page will not be up to date
>> 2536 * Case c, the page may be truncated but in itself, the data may still
>> 2537 * be valid after IO completes as it's a read vs truncate race. The
>> 2538 * operation must restart if the page is not uptodate on unlock but
>> 2539 * otherwise serialising on page lock to stabilise the mapping gives
>> 2540 * no additional guarantees to the caller as the page lock is
>> 2541 * released before return.
>> 2542 * Case d, similar to truncation. If reclaim holds the page lock, it
>> 2543 * will be a race with remove_mapping that determines if the mapping
>> 2544 * is valid on unlock but otherwise the data is valid and there is
>> 2545 * no need to serialise with page lock.
>> 2546 *
>> 2547 * As the page lock gives no additional guarantee, we optimistically
>> 2548 * wait on the page to be unlocked and check if it's up to date and
>> 2549 * use the page if it is. Otherwise, the page lock is required to
>> 2550 * distinguish between the different cases. The motivation is that we
>> 2551 * avoid spurious serialisations and wakeups when multiple processes
>> 2552 * wait on the same page for IO to complete.
>> 2553 */
>> 2554 wait_on_page_locked(page);
>> 2555 if (PageUptodate(page))
>> 2556 goto out;
>> 2557
>> 2558 /* Distinguish between all the cases under the safety of the lock */
>> 2559 lock_page(page);
3810 2560
3811 /* Folio was truncated from mapping * !! 2561 /* Case c or d, restart the operation */
3812 if (!folio->mapping) { !! 2562 if (!page->mapping) {
3813 folio_unlock(folio); !! 2563 unlock_page(page);
3814 folio_put(folio); !! 2564 put_page(page);
3815 goto repeat; 2565 goto repeat;
3816 } 2566 }
3817 2567
3818 /* Someone else locked and filled the 2568 /* Someone else locked and filled the page in a very small window */
3819 if (folio_test_uptodate(folio)) { !! 2569 if (PageUptodate(page)) {
3820 folio_unlock(folio); !! 2570 unlock_page(page);
3821 goto out; 2571 goto out;
3822 } 2572 }
3823 !! 2573 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 2574
3833 out: 2575 out:
3834 folio_mark_accessed(folio); !! 2576 mark_page_accessed(page);
3835 return folio; !! 2577 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 } 2578 }
3860 EXPORT_SYMBOL(read_cache_folio); <<
3861 2579
3862 /** 2580 /**
3863 * mapping_read_folio_gfp - Read into page ca !! 2581 * read_cache_page - read into page cache, fill it if needed
3864 * @mapping: The address_space for the fol !! 2582 * @mapping: the page's address_space
3865 * @index: The index that the allocated !! 2583 * @index: the page index
3866 * @gfp: The page allocator flags to u !! 2584 * @filler: function to perform the read
3867 * !! 2585 * @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 * <<
3871 * The most likely error from this function i <<
3872 * possible and so is EINTR. If ->read_folio <<
3873 * that will be returned to the caller. <<
3874 * 2586 *
3875 * The function expects mapping->invalidate_l !! 2587 * Read into the page cache. If a page already exists, and PageUptodate() is
>> 2588 * not set, try to fill the page and wait for it to become unlocked.
3876 * 2589 *
3877 * Return: Uptodate folio on success, ERR_PTR !! 2590 * If the page does not get brought uptodate, return -EIO.
3878 */ 2591 */
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 2592 struct page *read_cache_page(struct address_space *mapping,
3898 pgoff_t index, filler !! 2593 pgoff_t index,
>> 2594 int (*filler)(void *, struct page *),
>> 2595 void *data)
3899 { 2596 {
3900 return do_read_cache_page(mapping, in !! 2597 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
3901 mapping_gfp_mask(mapp <<
3902 } 2598 }
3903 EXPORT_SYMBOL(read_cache_page); 2599 EXPORT_SYMBOL(read_cache_page);
3904 2600
3905 /** 2601 /**
3906 * read_cache_page_gfp - read into page cache 2602 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3907 * @mapping: the page's address_space 2603 * @mapping: the page's address_space
3908 * @index: the page index 2604 * @index: the page index
3909 * @gfp: the page allocator flags to u 2605 * @gfp: the page allocator flags to use if allocating
3910 * 2606 *
3911 * This is the same as "read_mapping_page(map 2607 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3912 * any new page allocations done using the sp 2608 * any new page allocations done using the specified allocation flags.
3913 * 2609 *
3914 * If the page does not get brought uptodate, 2610 * If the page does not get brought uptodate, return -EIO.
3915 * <<
3916 * The function expects mapping->invalidate_l <<
3917 * <<
3918 * Return: up to date page on success, ERR_PT <<
3919 */ 2611 */
3920 struct page *read_cache_page_gfp(struct addre 2612 struct page *read_cache_page_gfp(struct address_space *mapping,
3921 pgoff_t index 2613 pgoff_t index,
3922 gfp_t gfp) 2614 gfp_t gfp)
3923 { 2615 {
3924 return do_read_cache_page(mapping, in !! 2616 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
>> 2617
>> 2618 return do_read_cache_page(mapping, index, filler, NULL, gfp);
3925 } 2619 }
3926 EXPORT_SYMBOL(read_cache_page_gfp); 2620 EXPORT_SYMBOL(read_cache_page_gfp);
3927 2621
3928 /* 2622 /*
3929 * Warn about a page cache invalidation failu !! 2623 * Performs necessary checks before doing a write
>> 2624 *
>> 2625 * Can adjust writing position or amount of bytes to write.
>> 2626 * Returns appropriate error code that caller should return or
>> 2627 * zero in case that write should be allowed.
3930 */ 2628 */
3931 static void dio_warn_stale_pagecache(struct f !! 2629 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
3932 { 2630 {
3933 static DEFINE_RATELIMIT_STATE(_rs, 86 !! 2631 struct file *file = iocb->ki_filp;
3934 char pathname[128]; !! 2632 struct inode *inode = file->f_mapping->host;
3935 char *path; !! 2633 unsigned long limit = rlimit(RLIMIT_FSIZE);
>> 2634 loff_t pos;
>> 2635
>> 2636 if (!iov_iter_count(from))
>> 2637 return 0;
>> 2638
>> 2639 /* FIXME: this is for backwards compatibility with 2.4 */
>> 2640 if (iocb->ki_flags & IOCB_APPEND)
>> 2641 iocb->ki_pos = i_size_read(inode);
3936 2642
3937 errseq_set(&filp->f_mapping->wb_err, !! 2643 pos = iocb->ki_pos;
3938 if (__ratelimit(&_rs)) { !! 2644
3939 path = file_path(filp, pathna !! 2645 if (limit != RLIM_INFINITY) {
3940 if (IS_ERR(path)) !! 2646 if (iocb->ki_pos >= limit) {
3941 path = "(unknown)"; !! 2647 send_sig(SIGXFSZ, current, 0);
3942 pr_crit("Page cache invalidat !! 2648 return -EFBIG;
3943 pr_crit("File: %s PID: %d Com !! 2649 }
3944 current->comm); !! 2650 iov_iter_truncate(from, limit - (unsigned long)pos);
3945 } 2651 }
>> 2652
>> 2653 /*
>> 2654 * LFS rule
>> 2655 */
>> 2656 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
>> 2657 !(file->f_flags & O_LARGEFILE))) {
>> 2658 if (pos >= MAX_NON_LFS)
>> 2659 return -EFBIG;
>> 2660 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
>> 2661 }
>> 2662
>> 2663 /*
>> 2664 * Are we about to exceed the fs block limit ?
>> 2665 *
>> 2666 * If we have written data it becomes a short write. If we have
>> 2667 * exceeded without writing data we send a signal and return EFBIG.
>> 2668 * Linus frestrict idea will clean these up nicely..
>> 2669 */
>> 2670 if (unlikely(pos >= inode->i_sb->s_maxbytes))
>> 2671 return -EFBIG;
>> 2672
>> 2673 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
>> 2674 return iov_iter_count(from);
>> 2675 }
>> 2676 EXPORT_SYMBOL(generic_write_checks);
>> 2677
>> 2678 int pagecache_write_begin(struct file *file, struct address_space *mapping,
>> 2679 loff_t pos, unsigned len, unsigned flags,
>> 2680 struct page **pagep, void **fsdata)
>> 2681 {
>> 2682 const struct address_space_operations *aops = mapping->a_ops;
>> 2683
>> 2684 return aops->write_begin(file, mapping, pos, len, flags,
>> 2685 pagep, fsdata);
3946 } 2686 }
>> 2687 EXPORT_SYMBOL(pagecache_write_begin);
3947 2688
3948 void kiocb_invalidate_post_direct_write(struc !! 2689 int pagecache_write_end(struct file *file, struct address_space *mapping,
>> 2690 loff_t pos, unsigned len, unsigned copied,
>> 2691 struct page *page, void *fsdata)
3949 { 2692 {
3950 struct address_space *mapping = iocb- !! 2693 const struct address_space_operations *aops = mapping->a_ops;
3951 2694
3952 if (mapping->nrpages && !! 2695 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3953 invalidate_inode_pages2_range(map <<
3954 iocb->ki_pos >> PAGE_ <<
3955 (iocb->ki_pos + count <<
3956 dio_warn_stale_pagecache(iocb <<
3957 } 2696 }
>> 2697 EXPORT_SYMBOL(pagecache_write_end);
3958 2698
3959 ssize_t 2699 ssize_t
3960 generic_file_direct_write(struct kiocb *iocb, 2700 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3961 { 2701 {
3962 struct address_space *mapping = iocb- !! 2702 struct file *file = iocb->ki_filp;
3963 size_t write_len = iov_iter_count(fro !! 2703 struct address_space *mapping = file->f_mapping;
3964 ssize_t written; !! 2704 struct inode *inode = mapping->host;
>> 2705 loff_t pos = iocb->ki_pos;
>> 2706 ssize_t written;
>> 2707 size_t write_len;
>> 2708 pgoff_t end;
>> 2709
>> 2710 write_len = iov_iter_count(from);
>> 2711 end = (pos + write_len - 1) >> PAGE_SHIFT;
>> 2712
>> 2713 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
>> 2714 if (written)
>> 2715 goto out;
3965 2716
3966 /* 2717 /*
>> 2718 * After a write we want buffered reads to be sure to go to disk to get
>> 2719 * the new data. We invalidate clean cached page from the region we're
>> 2720 * about to write. We do this *before* the write so that we can return
>> 2721 * without clobbering -EIOCBQUEUED from ->direct_IO().
>> 2722 */
>> 2723 written = invalidate_inode_pages2_range(mapping,
>> 2724 pos >> PAGE_SHIFT, end);
>> 2725 /*
3967 * If a page can not be invalidated, 2726 * If a page can not be invalidated, return 0 to fall back
3968 * to buffered write. 2727 * to buffered write.
3969 */ 2728 */
3970 written = kiocb_invalidate_pages(iocb <<
3971 if (written) { 2729 if (written) {
3972 if (written == -EBUSY) 2730 if (written == -EBUSY)
3973 return 0; 2731 return 0;
3974 return written; !! 2732 goto out;
3975 } 2733 }
3976 2734
3977 written = mapping->a_ops->direct_IO(i 2735 written = mapping->a_ops->direct_IO(iocb, from);
3978 2736
3979 /* 2737 /*
3980 * Finally, try again to invalidate c 2738 * Finally, try again to invalidate clean pages which might have been
3981 * cached by non-direct readahead, or 2739 * cached by non-direct readahead, or faulted in by get_user_pages()
3982 * if the source of the write was an 2740 * if the source of the write was an mmap'ed region of the file
3983 * we're writing. Either one is a pr 2741 * we're writing. Either one is a pretty crazy thing to do,
3984 * so we don't support it 100%. If t 2742 * so we don't support it 100%. If this invalidation
3985 * fails, tough, the write still work 2743 * fails, tough, the write still worked...
3986 * <<
3987 * Most of the time we do not need th <<
3988 * the invalidation for us. However t <<
3989 * do not end up with dio_complete() <<
3990 * them by removing it completely. <<
3991 * <<
3992 * Noticeable example is a blkdev_dir <<
3993 * <<
3994 * Skip invalidation for async writes <<
3995 */ 2744 */
3996 if (written > 0) { !! 2745 invalidate_inode_pages2_range(mapping,
3997 struct inode *inode = mapping !! 2746 pos >> PAGE_SHIFT, end);
3998 loff_t pos = iocb->ki_pos; <<
3999 2747
4000 kiocb_invalidate_post_direct_ !! 2748 if (written > 0) {
4001 pos += written; 2749 pos += written;
4002 write_len -= written; 2750 write_len -= written;
4003 if (pos > i_size_read(inode) 2751 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
4004 i_size_write(inode, p 2752 i_size_write(inode, pos);
4005 mark_inode_dirty(inod 2753 mark_inode_dirty(inode);
4006 } 2754 }
4007 iocb->ki_pos = pos; 2755 iocb->ki_pos = pos;
4008 } 2756 }
4009 if (written != -EIOCBQUEUED) !! 2757 iov_iter_revert(from, write_len - iov_iter_count(from));
4010 iov_iter_revert(from, write_l !! 2758 out:
4011 return written; 2759 return written;
4012 } 2760 }
4013 EXPORT_SYMBOL(generic_file_direct_write); 2761 EXPORT_SYMBOL(generic_file_direct_write);
4014 2762
4015 ssize_t generic_perform_write(struct kiocb *i !! 2763 /*
>> 2764 * Find or create a page at the given pagecache position. Return the locked
>> 2765 * page. This function is specifically for buffered writes.
>> 2766 */
>> 2767 struct page *grab_cache_page_write_begin(struct address_space *mapping,
>> 2768 pgoff_t index, unsigned flags)
>> 2769 {
>> 2770 struct page *page;
>> 2771 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
>> 2772
>> 2773 if (flags & AOP_FLAG_NOFS)
>> 2774 fgp_flags |= FGP_NOFS;
>> 2775
>> 2776 page = pagecache_get_page(mapping, index, fgp_flags,
>> 2777 mapping_gfp_mask(mapping));
>> 2778 if (page)
>> 2779 wait_for_stable_page(page);
>> 2780
>> 2781 return page;
>> 2782 }
>> 2783 EXPORT_SYMBOL(grab_cache_page_write_begin);
>> 2784
>> 2785 ssize_t generic_perform_write(struct file *file,
>> 2786 struct iov_iter *i, loff_t pos)
4016 { 2787 {
4017 struct file *file = iocb->ki_filp; <<
4018 loff_t pos = iocb->ki_pos; <<
4019 struct address_space *mapping = file- 2788 struct address_space *mapping = file->f_mapping;
4020 const struct address_space_operations 2789 const struct address_space_operations *a_ops = mapping->a_ops;
4021 size_t chunk = mapping_max_folio_size <<
4022 long status = 0; 2790 long status = 0;
4023 ssize_t written = 0; 2791 ssize_t written = 0;
>> 2792 unsigned int flags = 0;
4024 2793
4025 do { 2794 do {
4026 struct folio *folio; !! 2795 struct page *page;
4027 size_t offset; /* Of !! 2796 unsigned long offset; /* Offset into pagecache page */
4028 size_t bytes; /* By !! 2797 unsigned long bytes; /* Bytes to write to page */
4029 size_t copied; /* By 2798 size_t copied; /* Bytes copied from user */
4030 void *fsdata = NULL; !! 2799 void *fsdata;
4031 2800
4032 bytes = iov_iter_count(i); !! 2801 offset = (pos & (PAGE_SIZE - 1));
4033 retry: !! 2802 bytes = min_t(unsigned long, PAGE_SIZE - offset,
4034 offset = pos & (chunk - 1); !! 2803 iov_iter_count(i));
4035 bytes = min(chunk - offset, b <<
4036 balance_dirty_pages_ratelimit <<
4037 2804
>> 2805 again:
4038 /* 2806 /*
4039 * Bring in the user page tha 2807 * Bring in the user page that we will copy from _first_.
4040 * Otherwise there's a nasty 2808 * Otherwise there's a nasty deadlock on copying from the
4041 * same page as we're writing 2809 * same page as we're writing to, without it being marked
4042 * up-to-date. 2810 * up-to-date.
>> 2811 *
>> 2812 * Not only is this an optimisation, but it is also required
>> 2813 * to check that the address is actually valid, when atomic
>> 2814 * usercopies are used, below.
4043 */ 2815 */
4044 if (unlikely(fault_in_iov_ite !! 2816 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
4045 status = -EFAULT; 2817 status = -EFAULT;
4046 break; 2818 break;
4047 } 2819 }
4048 2820
4049 if (fatal_signal_pending(curr 2821 if (fatal_signal_pending(current)) {
4050 status = -EINTR; 2822 status = -EINTR;
4051 break; 2823 break;
4052 } 2824 }
4053 2825
4054 status = a_ops->write_begin(f !! 2826 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
4055 !! 2827 &page, &fsdata);
4056 if (unlikely(status < 0)) 2828 if (unlikely(status < 0))
4057 break; 2829 break;
4058 2830
4059 offset = offset_in_folio(foli <<
4060 if (bytes > folio_size(folio) <<
4061 bytes = folio_size(fo <<
4062 <<
4063 if (mapping_writably_mapped(m 2831 if (mapping_writably_mapped(mapping))
4064 flush_dcache_folio(fo !! 2832 flush_dcache_page(page);
4065 2833
4066 copied = copy_folio_from_iter !! 2834 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
4067 flush_dcache_folio(folio); !! 2835 flush_dcache_page(page);
4068 2836
4069 status = a_ops->write_end(fil 2837 status = a_ops->write_end(file, mapping, pos, bytes, copied,
4070 !! 2838 page, fsdata);
4071 if (unlikely(status != copied !! 2839 if (unlikely(status < 0))
4072 iov_iter_revert(i, co !! 2840 break;
4073 if (unlikely(status < !! 2841 copied = status;
4074 break; !! 2842
4075 } <<
4076 cond_resched(); 2843 cond_resched();
4077 2844
4078 if (unlikely(status == 0)) { !! 2845 iov_iter_advance(i, copied);
>> 2846 if (unlikely(copied == 0)) {
4079 /* 2847 /*
4080 * A short copy made !! 2848 * If we were unable to copy any data at all, we must
4081 * thing entirely. M !! 2849 * fall back to a single segment length write.
4082 * halfway through, m !! 2850 *
4083 * might be severe me !! 2851 * If we didn't fallback here, we could livelock
>> 2852 * because not all segments in the iov can be copied at
>> 2853 * once without a pagefault.
4084 */ 2854 */
4085 if (chunk > PAGE_SIZE !! 2855 bytes = min_t(unsigned long, PAGE_SIZE - offset,
4086 chunk /= 2; !! 2856 iov_iter_single_seg_count(i));
4087 if (copied) { !! 2857 goto again;
4088 bytes = copie <<
4089 goto retry; <<
4090 } <<
4091 } else { <<
4092 pos += status; <<
4093 written += status; <<
4094 } 2858 }
>> 2859 pos += copied;
>> 2860 written += copied;
>> 2861
>> 2862 balance_dirty_pages_ratelimited(mapping);
4095 } while (iov_iter_count(i)); 2863 } while (iov_iter_count(i));
4096 2864
4097 if (!written) !! 2865 return written ? written : status;
4098 return status; <<
4099 iocb->ki_pos += written; <<
4100 return written; <<
4101 } 2866 }
4102 EXPORT_SYMBOL(generic_perform_write); 2867 EXPORT_SYMBOL(generic_perform_write);
4103 2868
4104 /** 2869 /**
4105 * __generic_file_write_iter - write data to 2870 * __generic_file_write_iter - write data to a file
4106 * @iocb: IO state structure (file, off 2871 * @iocb: IO state structure (file, offset, etc.)
4107 * @from: iov_iter with data to write 2872 * @from: iov_iter with data to write
4108 * 2873 *
4109 * This function does all the work needed for 2874 * This function does all the work needed for actually writing data to a
4110 * file. It does all basic checks, removes SU 2875 * file. It does all basic checks, removes SUID from the file, updates
4111 * modification times and calls proper subrou 2876 * modification times and calls proper subroutines depending on whether we
4112 * do direct IO or a standard buffered write. 2877 * do direct IO or a standard buffered write.
4113 * 2878 *
4114 * It expects i_rwsem to be grabbed unless we !! 2879 * 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. 2880 * object which does not need locking at all.
4116 * 2881 *
4117 * This function does *not* take care of sync 2882 * 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 2883 * A caller has to handle it. This is mainly due to the fact that we want to
4119 * avoid syncing under i_rwsem. !! 2884 * avoid syncing under i_mutex.
4120 * <<
4121 * Return: <<
4122 * * number of bytes written, even for trunca <<
4123 * * negative error code if no data has been <<
4124 */ 2885 */
4125 ssize_t __generic_file_write_iter(struct kioc 2886 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4126 { 2887 {
4127 struct file *file = iocb->ki_filp; 2888 struct file *file = iocb->ki_filp;
4128 struct address_space *mapping = file- !! 2889 struct address_space * mapping = file->f_mapping;
4129 struct inode *inode = mapping->host; !! 2890 struct inode *inode = mapping->host;
4130 ssize_t ret; !! 2891 ssize_t written = 0;
4131 !! 2892 ssize_t err;
4132 ret = file_remove_privs(file); !! 2893 ssize_t status;
4133 if (ret) !! 2894
4134 return ret; !! 2895 /* We can write back this queue in page reclaim */
>> 2896 current->backing_dev_info = inode_to_bdi(inode);
>> 2897 err = file_remove_privs(file);
>> 2898 if (err)
>> 2899 goto out;
4135 2900
4136 ret = file_update_time(file); !! 2901 err = file_update_time(file);
4137 if (ret) !! 2902 if (err)
4138 return ret; !! 2903 goto out;
4139 2904
4140 if (iocb->ki_flags & IOCB_DIRECT) { 2905 if (iocb->ki_flags & IOCB_DIRECT) {
4141 ret = generic_file_direct_wri !! 2906 loff_t pos, endbyte;
>> 2907
>> 2908 written = generic_file_direct_write(iocb, from);
4142 /* 2909 /*
4143 * If the write stopped short 2910 * If the write stopped short of completing, fall back to
4144 * buffered writes. Some fil 2911 * buffered writes. Some filesystems do this for writes to
4145 * holes, for example. For D 2912 * holes, for example. For DAX files, a buffered write will
4146 * not succeed (even if it di 2913 * not succeed (even if it did, DAX does not handle dirty
4147 * page-cache pages correctly 2914 * page-cache pages correctly).
4148 */ 2915 */
4149 if (ret < 0 || !iov_iter_coun !! 2916 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
4150 return ret; !! 2917 goto out;
4151 return direct_write_fallback( <<
4152 generic_perfo <<
4153 } <<
4154 2918
4155 return generic_perform_write(iocb, fr !! 2919 status = generic_perform_write(file, from, pos = iocb->ki_pos);
>> 2920 /*
>> 2921 * If generic_perform_write() returned a synchronous error
>> 2922 * then we want to return the number of bytes which were
>> 2923 * direct-written, or the error code if that was zero. Note
>> 2924 * that this differs from normal direct-io semantics, which
>> 2925 * will return -EFOO even if some bytes were written.
>> 2926 */
>> 2927 if (unlikely(status < 0)) {
>> 2928 err = status;
>> 2929 goto out;
>> 2930 }
>> 2931 /*
>> 2932 * We need to ensure that the page cache pages are written to
>> 2933 * disk and invalidated to preserve the expected O_DIRECT
>> 2934 * semantics.
>> 2935 */
>> 2936 endbyte = pos + status - 1;
>> 2937 err = filemap_write_and_wait_range(mapping, pos, endbyte);
>> 2938 if (err == 0) {
>> 2939 iocb->ki_pos = endbyte + 1;
>> 2940 written += status;
>> 2941 invalidate_mapping_pages(mapping,
>> 2942 pos >> PAGE_SHIFT,
>> 2943 endbyte >> PAGE_SHIFT);
>> 2944 } else {
>> 2945 /*
>> 2946 * We don't know how much we wrote, so just return
>> 2947 * the number of bytes which were direct-written
>> 2948 */
>> 2949 }
>> 2950 } else {
>> 2951 written = generic_perform_write(file, from, iocb->ki_pos);
>> 2952 if (likely(written > 0))
>> 2953 iocb->ki_pos += written;
>> 2954 }
>> 2955 out:
>> 2956 current->backing_dev_info = NULL;
>> 2957 return written ? written : err;
4156 } 2958 }
4157 EXPORT_SYMBOL(__generic_file_write_iter); 2959 EXPORT_SYMBOL(__generic_file_write_iter);
4158 2960
4159 /** 2961 /**
4160 * generic_file_write_iter - write data to a 2962 * generic_file_write_iter - write data to a file
4161 * @iocb: IO state structure 2963 * @iocb: IO state structure
4162 * @from: iov_iter with data to write 2964 * @from: iov_iter with data to write
4163 * 2965 *
4164 * This is a wrapper around __generic_file_wr 2966 * This is a wrapper around __generic_file_write_iter() to be used by most
4165 * filesystems. It takes care of syncing the 2967 * filesystems. It takes care of syncing the file in case of O_SYNC file
4166 * and acquires i_rwsem as needed. !! 2968 * and acquires i_mutex as needed.
4167 * Return: <<
4168 * * negative error code if no data has been <<
4169 * vfs_fsync_range() failed for a synchrono <<
4170 * * number of bytes written, even for trunca <<
4171 */ 2969 */
4172 ssize_t generic_file_write_iter(struct kiocb 2970 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4173 { 2971 {
4174 struct file *file = iocb->ki_filp; 2972 struct file *file = iocb->ki_filp;
4175 struct inode *inode = file->f_mapping 2973 struct inode *inode = file->f_mapping->host;
4176 ssize_t ret; 2974 ssize_t ret;
4177 2975
4178 inode_lock(inode); 2976 inode_lock(inode);
4179 ret = generic_write_checks(iocb, from 2977 ret = generic_write_checks(iocb, from);
4180 if (ret > 0) 2978 if (ret > 0)
4181 ret = __generic_file_write_it 2979 ret = __generic_file_write_iter(iocb, from);
4182 inode_unlock(inode); 2980 inode_unlock(inode);
4183 2981
4184 if (ret > 0) 2982 if (ret > 0)
4185 ret = generic_write_sync(iocb 2983 ret = generic_write_sync(iocb, ret);
4186 return ret; 2984 return ret;
4187 } 2985 }
4188 EXPORT_SYMBOL(generic_file_write_iter); 2986 EXPORT_SYMBOL(generic_file_write_iter);
4189 2987
4190 /** 2988 /**
4191 * filemap_release_folio() - Release fs-speci !! 2989 * 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 * 2990 *
4195 * The address_space is trying to release any !! 2991 * @page: the page which the kernel is trying to free
4196 * (presumably at folio->private). !! 2992 * @gfp_mask: memory allocation flags (and I/O mode)
4197 * 2993 *
4198 * This will also be called if the private_2 !! 2994 * The address_space is to try to release any data against the page
4199 * indicating that the folio has other metada !! 2995 * (presumably at page->private). If the release was successful, return '1'.
>> 2996 * Otherwise return zero.
4200 * 2997 *
4201 * The @gfp argument specifies whether I/O ma !! 2998 * 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 !! 2999 * page is known to the local caching routines.
4203 * (__GFP_RECLAIM & __GFP_FS). !! 3000 *
>> 3001 * The @gfp_mask argument specifies whether I/O may be performed to release
>> 3002 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
4204 * 3003 *
4205 * Return: %true if the release was successfu <<
4206 */ <<
4207 bool filemap_release_folio(struct folio *foli <<
4208 { <<
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 */ 3004 */
4290 static void filemap_cachestat(struct address_ !! 3005 int try_to_release_page(struct page *page, gfp_t gfp_mask)
4291 pgoff_t first_index, pgoff_t <<
4292 { 3006 {
4293 XA_STATE(xas, &mapping->i_pages, firs !! 3007 struct address_space * const mapping = page->mapping;
4294 struct folio *folio; <<
4295 3008
4296 /* Flush stats (and potentially sleep !! 3009 BUG_ON(!PageLocked(page));
4297 mem_cgroup_flush_stats_ratelimited(NU !! 3010 if (PageWriteback(page))
4298 !! 3011 return 0;
4299 rcu_read_lock(); <<
4300 xas_for_each(&xas, folio, last_index) <<
4301 int order; <<
4302 unsigned long nr_pages; <<
4303 pgoff_t folio_first_index, fo <<
4304 <<
4305 /* <<
4306 * Don't deref the folio. It <<
4307 * get freed (and reused) und <<
4308 * <<
4309 * We *could* pin it, but tha <<
4310 * what should be a fast and <<
4311 * <<
4312 * Instead, derive all inform <<
4313 * the rcu-protected xarray. <<
4314 */ <<
4315 <<
4316 if (xas_retry(&xas, folio)) <<
4317 continue; <<
4318 <<
4319 order = xas_get_order(&xas); <<
4320 nr_pages = 1 << order; <<
4321 folio_first_index = round_dow <<
4322 folio_last_index = folio_firs <<
4323 <<
4324 /* Folios might straddle the <<
4325 if (folio_first_index < first <<
4326 nr_pages -= first_ind <<
4327 <<
4328 if (folio_last_index > last_i <<
4329 nr_pages -= folio_las <<
4330 <<
4331 if (xa_is_value(folio)) { <<
4332 /* page is evicted */ <<
4333 void *shadow = (void <<
4334 bool workingset; /* n <<
4335 <<
4336 cs->nr_evicted += nr_ <<
4337 <<
4338 #ifdef CONFIG_SWAP /* implies CONFIG_MMU */ <<
4339 if (shmem_mapping(map <<
4340 /* shmem file <<
4341 swp_entry_t s <<
4342 <<
4343 /* swapin err <<
4344 if (non_swap_ <<
4345 goto <<
4346 <<
4347 /* <<
4348 * Getting a <<
4349 * inode mean <<
4350 * shmem_unus <<
4351 * ensures sw <<
4352 * freeing th <<
4353 * we can rac <<
4354 * invalidati <<
4355 * a shadow i <<
4356 */ <<
4357 shadow = get_ <<
4358 if (!shadow) <<
4359 goto <<
4360 } <<
4361 #endif <<
4362 if (workingset_test_r <<
4363 cs->nr_recent <<
4364 <<
4365 goto resched; <<
4366 } <<
4367 <<
4368 /* page is in cache */ <<
4369 cs->nr_cache += nr_pages; <<
4370 <<
4371 if (xas_get_mark(&xas, PAGECA <<
4372 cs->nr_dirty += nr_pa <<
4373 <<
4374 if (xas_get_mark(&xas, PAGECA <<
4375 cs->nr_writeback += n <<
4376 3012
4377 resched: !! 3013 if (mapping && mapping->a_ops->releasepage)
4378 if (need_resched()) { !! 3014 return mapping->a_ops->releasepage(page, gfp_mask);
4379 xas_pause(&xas); !! 3015 return try_to_free_buffers(page);
4380 cond_resched_rcu(); <<
4381 } <<
4382 } <<
4383 rcu_read_unlock(); <<
4384 } 3016 }
4385 3017
4386 /* !! 3018 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 3019
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.