blob: 617469fce96d2130cc66df9a20850aefc47e3c8d [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * linux/mm/mlock.c
4 *
5 * (C) Copyright 1995 Linus Torvalds
6 * (C) Copyright 2002 Christoph Hellwig
7 */
8
Randy.Dunlapc59ede72006-01-11 12:17:46 -08009#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070010#include <linux/mman.h>
11#include <linux/mm.h>
Ingo Molnar8703e8a2017-02-08 18:51:30 +010012#include <linux/sched/user.h>
Nick Pigginb291f002008-10-18 20:26:44 -070013#include <linux/swap.h>
14#include <linux/swapops.h>
15#include <linux/pagemap.h>
Vlastimil Babka72255222013-09-11 14:22:29 -070016#include <linux/pagevec.h>
Hugh Dickins34b67922022-02-14 18:31:48 -080017#include <linux/pagewalk.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/mempolicy.h>
19#include <linux/syscalls.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040020#include <linux/sched.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040021#include <linux/export.h>
Nick Pigginb291f002008-10-18 20:26:44 -070022#include <linux/rmap.h>
23#include <linux/mmzone.h>
24#include <linux/hugetlb.h>
Vlastimil Babka72255222013-09-11 14:22:29 -070025#include <linux/memcontrol.h>
26#include <linux/mm_inline.h>
Mike Rapoport1507f512021-07-07 18:08:03 -070027#include <linux/secretmem.h>
Nick Pigginb291f002008-10-18 20:26:44 -070028
29#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
Lorenzo Stoakes90d07212023-01-12 12:39:29 +000031struct mlock_fbatch {
Sebastian Andrzej Siewioradb11e72022-04-01 11:28:33 -070032 local_lock_t lock;
Lorenzo Stoakes90d07212023-01-12 12:39:29 +000033 struct folio_batch fbatch;
Sebastian Andrzej Siewioradb11e72022-04-01 11:28:33 -070034};
35
Lorenzo Stoakes90d07212023-01-12 12:39:29 +000036static DEFINE_PER_CPU(struct mlock_fbatch, mlock_fbatch) = {
Sebastian Andrzej Siewioradb11e72022-04-01 11:28:33 -070037 .lock = INIT_LOCAL_LOCK(lock),
38};
Hugh Dickins2fbb0c12022-02-14 18:37:29 -080039
Wang Xiaoqiang7f43add2016-01-15 16:57:22 -080040bool can_do_mlock(void)
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040041{
Jiri Slaby59e99e52010-03-05 13:41:44 -080042 if (rlimit(RLIMIT_MEMLOCK) != 0)
Wang Xiaoqiang7f43add2016-01-15 16:57:22 -080043 return true;
Jeff Vander Stoepa5a65792015-03-12 16:26:17 -070044 if (capable(CAP_IPC_LOCK))
Wang Xiaoqiang7f43add2016-01-15 16:57:22 -080045 return true;
46 return false;
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040047}
48EXPORT_SYMBOL(can_do_mlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
Nick Pigginb291f002008-10-18 20:26:44 -070050/*
Lorenzo Stoakes90d07212023-01-12 12:39:29 +000051 * Mlocked folios are marked with the PG_mlocked flag for efficient testing
Nick Pigginb291f002008-10-18 20:26:44 -070052 * in vmscan and, possibly, the fault path; and to support semi-accurate
53 * statistics.
54 *
Lorenzo Stoakes90d07212023-01-12 12:39:29 +000055 * An mlocked folio [folio_test_mlocked(folio)] is unevictable. As such, it
56 * will be ostensibly placed on the LRU "unevictable" list (actually no such
57 * list exists), rather than the [in]active lists. PG_unevictable is set to
58 * indicate the unevictable state.
Nick Pigginb291f002008-10-18 20:26:44 -070059 */
60
Lorenzo Stoakes90d07212023-01-12 12:39:29 +000061static struct lruvec *__mlock_folio(struct folio *folio, struct lruvec *lruvec)
Nick Pigginb291f002008-10-18 20:26:44 -070062{
Hugh Dickins07ca7602022-02-14 18:29:54 -080063 /* There is nothing more we can do while it's off LRU */
Lorenzo Stoakes90d07212023-01-12 12:39:29 +000064 if (!folio_test_clear_lru(folio))
Hugh Dickins2fbb0c12022-02-14 18:37:29 -080065 return lruvec;
Hugh Dickins09647302020-09-18 21:20:15 -070066
Lorenzo Stoakes90d07212023-01-12 12:39:29 +000067 lruvec = folio_lruvec_relock_irq(folio, lruvec);
Nick Pigginb291f002008-10-18 20:26:44 -070068
Lorenzo Stoakes90d07212023-01-12 12:39:29 +000069 if (unlikely(folio_evictable(folio))) {
Nick Pigginb291f002008-10-18 20:26:44 -070070 /*
Lorenzo Stoakes90d07212023-01-12 12:39:29 +000071 * This is a little surprising, but quite possible: PG_mlocked
72 * must have got cleared already by another CPU. Could this
73 * folio be unevictable? I'm not sure, but move it now if so.
Nick Pigginb291f002008-10-18 20:26:44 -070074 */
Lorenzo Stoakes90d07212023-01-12 12:39:29 +000075 if (folio_test_unevictable(folio)) {
76 lruvec_del_folio(lruvec, folio);
77 folio_clear_unevictable(folio);
78 lruvec_add_folio(lruvec, folio);
79
Hugh Dickins2fbb0c12022-02-14 18:37:29 -080080 __count_vm_events(UNEVICTABLE_PGRESCUED,
Lorenzo Stoakes90d07212023-01-12 12:39:29 +000081 folio_nr_pages(folio));
Hugh Dickins2fbb0c12022-02-14 18:37:29 -080082 }
83 goto out;
Nick Pigginb291f002008-10-18 20:26:44 -070084 }
Nick Pigginb291f002008-10-18 20:26:44 -070085
Lorenzo Stoakes90d07212023-01-12 12:39:29 +000086 if (folio_test_unevictable(folio)) {
87 if (folio_test_mlocked(folio))
88 folio->mlock_count++;
Hugh Dickins07ca7602022-02-14 18:29:54 -080089 goto out;
Nick Piggin5344b7e2008-10-18 20:26:51 -070090 }
Hugh Dickins07ca7602022-02-14 18:29:54 -080091
Lorenzo Stoakes90d07212023-01-12 12:39:29 +000092 lruvec_del_folio(lruvec, folio);
93 folio_clear_active(folio);
94 folio_set_unevictable(folio);
95 folio->mlock_count = !!folio_test_mlocked(folio);
96 lruvec_add_folio(lruvec, folio);
97 __count_vm_events(UNEVICTABLE_PGCULLED, folio_nr_pages(folio));
Hugh Dickins07ca7602022-02-14 18:29:54 -080098out:
Lorenzo Stoakes90d07212023-01-12 12:39:29 +000099 folio_set_lru(folio);
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800100 return lruvec;
Nick Pigginb291f002008-10-18 20:26:44 -0700101}
102
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000103static struct lruvec *__mlock_new_folio(struct folio *folio, struct lruvec *lruvec)
Vlastimil Babka72255222013-09-11 14:22:29 -0700104{
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000105 VM_BUG_ON_FOLIO(folio_test_lru(folio), folio);
Vlastimil Babka72255222013-09-11 14:22:29 -0700106
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000107 lruvec = folio_lruvec_relock_irq(folio, lruvec);
Vlastimil Babka72255222013-09-11 14:22:29 -0700108
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800109 /* As above, this is a little surprising, but possible */
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000110 if (unlikely(folio_evictable(folio)))
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800111 goto out;
112
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000113 folio_set_unevictable(folio);
114 folio->mlock_count = !!folio_test_mlocked(folio);
115 __count_vm_events(UNEVICTABLE_PGCULLED, folio_nr_pages(folio));
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800116out:
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000117 lruvec_add_folio(lruvec, folio);
118 folio_set_lru(folio);
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800119 return lruvec;
Vlastimil Babka72255222013-09-11 14:22:29 -0700120}
121
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000122static struct lruvec *__munlock_folio(struct folio *folio, struct lruvec *lruvec)
Vlastimil Babka72255222013-09-11 14:22:29 -0700123{
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000124 int nr_pages = folio_nr_pages(folio);
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800125 bool isolated = false;
Hugh Dickins09647302020-09-18 21:20:15 -0700126
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000127 if (!folio_test_clear_lru(folio))
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800128 goto munlock;
Kirill A. Shutemove90309c2016-01-15 16:54:33 -0800129
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800130 isolated = true;
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000131 lruvec = folio_lruvec_relock_irq(folio, lruvec);
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800132
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000133 if (folio_test_unevictable(folio)) {
Hugh Dickins07ca7602022-02-14 18:29:54 -0800134 /* Then mlock_count is maintained, but might undercount */
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000135 if (folio->mlock_count)
136 folio->mlock_count--;
137 if (folio->mlock_count)
Hugh Dickins07ca7602022-02-14 18:29:54 -0800138 goto out;
Vlastimil Babka72255222013-09-11 14:22:29 -0700139 }
Hugh Dickins07ca7602022-02-14 18:29:54 -0800140 /* else assume that was the last mlock: reclaim will fix it if not */
141
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800142munlock:
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000143 if (folio_test_clear_mlocked(folio)) {
144 __zone_stat_mod_folio(folio, NR_MLOCK, -nr_pages);
145 if (isolated || !folio_test_unevictable(folio))
Hugh Dickins07ca7602022-02-14 18:29:54 -0800146 __count_vm_events(UNEVICTABLE_PGMUNLOCKED, nr_pages);
147 else
148 __count_vm_events(UNEVICTABLE_PGSTRANDED, nr_pages);
149 }
150
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000151 /* folio_evictable() has to be checked *after* clearing Mlocked */
152 if (isolated && folio_test_unevictable(folio) && folio_evictable(folio)) {
153 lruvec_del_folio(lruvec, folio);
154 folio_clear_unevictable(folio);
155 lruvec_add_folio(lruvec, folio);
Hugh Dickins07ca7602022-02-14 18:29:54 -0800156 __count_vm_events(UNEVICTABLE_PGRESCUED, nr_pages);
157 }
158out:
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800159 if (isolated)
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000160 folio_set_lru(folio);
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800161 return lruvec;
162}
163
164/*
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000165 * Flags held in the low bits of a struct folio pointer on the mlock_fbatch.
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800166 */
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000167#define LRU_FOLIO 0x1
168#define NEW_FOLIO 0x2
169static inline struct folio *mlock_lru(struct folio *folio)
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800170{
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000171 return (struct folio *)((unsigned long)folio + LRU_FOLIO);
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800172}
173
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000174static inline struct folio *mlock_new(struct folio *folio)
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800175{
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000176 return (struct folio *)((unsigned long)folio + NEW_FOLIO);
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800177}
178
179/*
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000180 * mlock_folio_batch() is derived from folio_batch_move_lru(): perhaps that can
181 * make use of such folio pointer flags in future, but for now just keep it for
182 * mlock. We could use three separate folio batches instead, but one feels
183 * better (munlocking a full folio batch does not need to drain mlocking folio
184 * batches first).
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800185 */
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000186static void mlock_folio_batch(struct folio_batch *fbatch)
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800187{
188 struct lruvec *lruvec = NULL;
189 unsigned long mlock;
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000190 struct folio *folio;
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800191 int i;
192
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000193 for (i = 0; i < folio_batch_count(fbatch); i++) {
194 folio = fbatch->folios[i];
195 mlock = (unsigned long)folio & (LRU_FOLIO | NEW_FOLIO);
196 folio = (struct folio *)((unsigned long)folio - mlock);
197 fbatch->folios[i] = folio;
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800198
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000199 if (mlock & LRU_FOLIO)
200 lruvec = __mlock_folio(folio, lruvec);
201 else if (mlock & NEW_FOLIO)
202 lruvec = __mlock_new_folio(folio, lruvec);
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800203 else
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000204 lruvec = __munlock_folio(folio, lruvec);
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800205 }
206
207 if (lruvec)
208 unlock_page_lruvec_irq(lruvec);
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000209 release_pages(fbatch->folios, fbatch->nr);
210 folio_batch_reinit(fbatch);
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800211}
212
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000213void mlock_drain_local(void)
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800214{
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000215 struct folio_batch *fbatch;
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800216
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000217 local_lock(&mlock_fbatch.lock);
218 fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
219 if (folio_batch_count(fbatch))
220 mlock_folio_batch(fbatch);
221 local_unlock(&mlock_fbatch.lock);
Sebastian Andrzej Siewioradb11e72022-04-01 11:28:33 -0700222}
223
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000224void mlock_drain_remote(int cpu)
Sebastian Andrzej Siewioradb11e72022-04-01 11:28:33 -0700225{
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000226 struct folio_batch *fbatch;
Sebastian Andrzej Siewioradb11e72022-04-01 11:28:33 -0700227
228 WARN_ON_ONCE(cpu_online(cpu));
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000229 fbatch = &per_cpu(mlock_fbatch.fbatch, cpu);
230 if (folio_batch_count(fbatch))
231 mlock_folio_batch(fbatch);
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800232}
233
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000234bool need_mlock_drain(int cpu)
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800235{
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000236 return folio_batch_count(&per_cpu(mlock_fbatch.fbatch, cpu));
Lee Schermerhorn6927c1d2009-12-14 17:59:55 -0800237}
Nick Pigginb291f002008-10-18 20:26:44 -0700238
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800239/**
Matthew Wilcox (Oracle)dcc5d332022-02-15 13:33:59 -0500240 * mlock_folio - mlock a folio already on (or temporarily off) LRU
241 * @folio: folio to be mlocked.
Hugh Dickins3d470fc2011-10-31 17:09:43 -0700242 */
Matthew Wilcox (Oracle)dcc5d332022-02-15 13:33:59 -0500243void mlock_folio(struct folio *folio)
Hugh Dickins3d470fc2011-10-31 17:09:43 -0700244{
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000245 struct folio_batch *fbatch;
Sebastian Andrzej Siewioradb11e72022-04-01 11:28:33 -0700246
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000247 local_lock(&mlock_fbatch.lock);
248 fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
Hugh Dickins3d470fc2011-10-31 17:09:43 -0700249
Matthew Wilcox (Oracle)dcc5d332022-02-15 13:33:59 -0500250 if (!folio_test_set_mlocked(folio)) {
251 int nr_pages = folio_nr_pages(folio);
Hugh Dickins3d470fc2011-10-31 17:09:43 -0700252
Matthew Wilcox (Oracle)dcc5d332022-02-15 13:33:59 -0500253 zone_stat_mod_folio(folio, NR_MLOCK, nr_pages);
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800254 __count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages);
Hugh Dickins53f79ac2009-12-14 17:58:58 -0800255 }
Nick Piggin5344b7e2008-10-18 20:26:51 -0700256
Matthew Wilcox (Oracle)dcc5d332022-02-15 13:33:59 -0500257 folio_get(folio);
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000258 if (!folio_batch_add(fbatch, mlock_lru(folio)) ||
Matthew Wilcox (Oracle)dcc5d332022-02-15 13:33:59 -0500259 folio_test_large(folio) || lru_cache_disabled())
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000260 mlock_folio_batch(fbatch);
261 local_unlock(&mlock_fbatch.lock);
Nick Piggin5344b7e2008-10-18 20:26:51 -0700262}
263
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800264/**
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000265 * mlock_new_folio - mlock a newly allocated folio not yet on LRU
266 * @folio: folio to be mlocked, either normal or a THP head.
Lee Schermerhorn6927c1d2009-12-14 17:59:55 -0800267 */
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000268void mlock_new_folio(struct folio *folio)
Lee Schermerhorn6927c1d2009-12-14 17:59:55 -0800269{
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000270 struct folio_batch *fbatch;
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000271 int nr_pages = folio_nr_pages(folio);
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800272
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000273 local_lock(&mlock_fbatch.lock);
274 fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
275 folio_set_mlocked(folio);
276
277 zone_stat_mod_folio(folio, NR_MLOCK, nr_pages);
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800278 __count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages);
279
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000280 folio_get(folio);
281 if (!folio_batch_add(fbatch, mlock_new(folio)) ||
282 folio_test_large(folio) || lru_cache_disabled())
283 mlock_folio_batch(fbatch);
284 local_unlock(&mlock_fbatch.lock);
Nick Piggin5344b7e2008-10-18 20:26:51 -0700285}
286
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800287/**
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000288 * munlock_folio - munlock a folio
289 * @folio: folio to be munlocked, either normal or a THP head.
Rik van Rielba470de2008-10-18 20:26:50 -0700290 */
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000291void munlock_folio(struct folio *folio)
Hugh Dickins408e82b2009-09-21 17:03:23 -0700292{
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000293 struct folio_batch *fbatch;
Vlastimil Babka56afe477d2013-09-11 14:22:32 -0700294
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000295 local_lock(&mlock_fbatch.lock);
296 fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
Vlastimil Babka5b409982013-09-11 14:22:33 -0700297 /*
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000298 * folio_test_clear_mlocked(folio) must be left to __munlock_folio(),
299 * which will check whether the folio is multiply mlocked.
Vlastimil Babka5b409982013-09-11 14:22:33 -0700300 */
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000301 folio_get(folio);
302 if (!folio_batch_add(fbatch, folio) ||
303 folio_test_large(folio) || lru_cache_disabled())
304 mlock_folio_batch(fbatch);
305 local_unlock(&mlock_fbatch.lock);
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700306}
Vlastimil Babka56afe477d2013-09-11 14:22:32 -0700307
Hugh Dickins34b67922022-02-14 18:31:48 -0800308static int mlock_pte_range(pmd_t *pmd, unsigned long addr,
309 unsigned long end, struct mm_walk *walk)
Hugh Dickins408e82b2009-09-21 17:03:23 -0700310
Hugh Dickins408e82b2009-09-21 17:03:23 -0700311{
Hugh Dickins34b67922022-02-14 18:31:48 -0800312 struct vm_area_struct *vma = walk->vma;
Hugh Dickins6e919712009-09-21 17:03:32 -0700313 spinlock_t *ptl;
Hugh Dickins34b67922022-02-14 18:31:48 -0800314 pte_t *start_pte, *pte;
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000315 struct folio *folio;
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800316
Hugh Dickins34b67922022-02-14 18:31:48 -0800317 ptl = pmd_trans_huge_lock(pmd, vma);
318 if (ptl) {
319 if (!pmd_present(*pmd))
320 goto out;
321 if (is_huge_zero_pmd(*pmd))
322 goto out;
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000323 folio = page_folio(pmd_page(*pmd));
Hugh Dickins34b67922022-02-14 18:31:48 -0800324 if (vma->vm_flags & VM_LOCKED)
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000325 mlock_folio(folio);
Hugh Dickins34b67922022-02-14 18:31:48 -0800326 else
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000327 munlock_folio(folio);
Hugh Dickins34b67922022-02-14 18:31:48 -0800328 goto out;
Nick Pigginb291f002008-10-18 20:26:44 -0700329 }
Hugh Dickins34b67922022-02-14 18:31:48 -0800330
331 start_pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
332 for (pte = start_pte; addr != end; pte++, addr += PAGE_SIZE) {
333 if (!pte_present(*pte))
334 continue;
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000335 folio = vm_normal_folio(vma, addr, *pte);
336 if (!folio || folio_is_zone_device(folio))
Hugh Dickins34b67922022-02-14 18:31:48 -0800337 continue;
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000338 if (folio_test_large(folio))
Hugh Dickins34b67922022-02-14 18:31:48 -0800339 continue;
340 if (vma->vm_flags & VM_LOCKED)
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000341 mlock_folio(folio);
Hugh Dickins34b67922022-02-14 18:31:48 -0800342 else
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000343 munlock_folio(folio);
Hugh Dickins34b67922022-02-14 18:31:48 -0800344 }
345 pte_unmap(start_pte);
346out:
347 spin_unlock(ptl);
348 cond_resched();
349 return 0;
Nick Pigginb291f002008-10-18 20:26:44 -0700350}
351
352/*
Hugh Dickins34b67922022-02-14 18:31:48 -0800353 * mlock_vma_pages_range() - mlock any pages already in the range,
354 * or munlock all pages in the range.
355 * @vma - vma containing range to be mlock()ed or munlock()ed
Rik van Rielba470de2008-10-18 20:26:50 -0700356 * @start - start address in @vma of the range
Hugh Dickins34b67922022-02-14 18:31:48 -0800357 * @end - end of range in @vma
358 * @newflags - the new set of flags for @vma.
Rik van Rielba470de2008-10-18 20:26:50 -0700359 *
Hugh Dickins34b67922022-02-14 18:31:48 -0800360 * Called for mlock(), mlock2() and mlockall(), to set @vma VM_LOCKED;
361 * called for munlock() and munlockall(), to clear VM_LOCKED from @vma.
Nick Pigginb291f002008-10-18 20:26:44 -0700362 */
Hugh Dickins34b67922022-02-14 18:31:48 -0800363static void mlock_vma_pages_range(struct vm_area_struct *vma,
364 unsigned long start, unsigned long end, vm_flags_t newflags)
Nick Pigginb291f002008-10-18 20:26:44 -0700365{
Hugh Dickins34b67922022-02-14 18:31:48 -0800366 static const struct mm_walk_ops mlock_walk_ops = {
367 .pmd_entry = mlock_pte_range,
368 };
Hugh Dickins408e82b2009-09-21 17:03:23 -0700369
Hugh Dickins34b67922022-02-14 18:31:48 -0800370 /*
371 * There is a slight chance that concurrent page migration,
372 * or page reclaim finding a page of this now-VM_LOCKED vma,
Matthew Wilcox (Oracle)7efecff2023-01-16 19:28:25 +0000373 * will call mlock_vma_folio() and raise page's mlock_count:
Hugh Dickins34b67922022-02-14 18:31:48 -0800374 * double counting, leaving the page unevictable indefinitely.
Matthew Wilcox (Oracle)7efecff2023-01-16 19:28:25 +0000375 * Communicate this danger to mlock_vma_folio() with VM_IO,
Hugh Dickins34b67922022-02-14 18:31:48 -0800376 * which is a VM_SPECIAL flag not allowed on VM_LOCKED vmas.
377 * mmap_lock is held in write mode here, so this weird
378 * combination should not be visible to other mmap_lock users;
379 * but WRITE_ONCE so rmap walkers must see VM_IO if VM_LOCKED.
380 */
381 if (newflags & VM_LOCKED)
382 newflags |= VM_IO;
Suren Baghdasaryan601c3c22023-01-31 16:01:16 -0800383 vm_flags_reset_once(vma, newflags);
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800384
Hugh Dickins34b67922022-02-14 18:31:48 -0800385 lru_add_drain();
386 walk_page_range(vma->vm_mm, start, end, &mlock_walk_ops, NULL);
387 lru_add_drain();
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700388
Hugh Dickins34b67922022-02-14 18:31:48 -0800389 if (newflags & VM_IO) {
390 newflags &= ~VM_IO;
Suren Baghdasaryan601c3c22023-01-31 16:01:16 -0800391 vm_flags_reset_once(vma, newflags);
Hugh Dickins408e82b2009-09-21 17:03:23 -0700392 }
Nick Pigginb291f002008-10-18 20:26:44 -0700393}
394
395/*
396 * mlock_fixup - handle mlock[all]/munlock[all] requests.
397 *
398 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
399 * munlock is a no-op. However, for some special vmas, we go ahead and
Michel Lespinassecea10a12013-02-22 16:32:44 -0800400 * populate the ptes.
Nick Pigginb291f002008-10-18 20:26:44 -0700401 *
402 * For vmas that pass the filters, merge/split as appropriate.
403 */
Liam R. Howlett37598f52023-01-20 11:26:19 -0500404static int mlock_fixup(struct vma_iterator *vmi, struct vm_area_struct *vma,
405 struct vm_area_struct **prev, unsigned long start,
406 unsigned long end, vm_flags_t newflags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407{
Nick Pigginb291f002008-10-18 20:26:44 -0700408 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 pgoff_t pgoff;
Nick Pigginb291f002008-10-18 20:26:44 -0700410 int nr_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 int ret = 0;
Hugh Dickins34b67922022-02-14 18:31:48 -0800412 vm_flags_t oldflags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413
Hugh Dickins34b67922022-02-14 18:31:48 -0800414 if (newflags == oldflags || (oldflags & VM_SPECIAL) ||
Dave Jiange1fb4a02018-08-17 15:43:40 -0700415 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm) ||
Mike Rapoport1507f512021-07-07 18:08:03 -0700416 vma_is_dax(vma) || vma_is_secretmem(vma))
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800417 /* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */
418 goto out;
Nick Pigginb291f002008-10-18 20:26:44 -0700419
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
Liam R. Howlett9760ebf2023-01-20 11:26:30 -0500421 *prev = vma_merge(vmi, mm, *prev, start, end, newflags,
Liam R. Howlett37598f52023-01-20 11:26:19 -0500422 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
423 vma->vm_userfaultfd_ctx, anon_vma_name(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 if (*prev) {
425 vma = *prev;
426 goto success;
427 }
428
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 if (start != vma->vm_start) {
Liam R. Howlett9760ebf2023-01-20 11:26:30 -0500430 ret = split_vma(vmi, vma, start, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 if (ret)
432 goto out;
433 }
434
435 if (end != vma->vm_end) {
Liam R. Howlett9760ebf2023-01-20 11:26:30 -0500436 ret = split_vma(vmi, vma, end, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 if (ret)
438 goto out;
439 }
440
441success:
442 /*
Nick Pigginb291f002008-10-18 20:26:44 -0700443 * Keep track of amount of locked VM.
444 */
445 nr_pages = (end - start) >> PAGE_SHIFT;
Hugh Dickins34b67922022-02-14 18:31:48 -0800446 if (!(newflags & VM_LOCKED))
Nick Pigginb291f002008-10-18 20:26:44 -0700447 nr_pages = -nr_pages;
Hugh Dickins34b67922022-02-14 18:31:48 -0800448 else if (oldflags & VM_LOCKED)
Simon Guob155b4f2016-10-07 16:59:40 -0700449 nr_pages = 0;
Nick Pigginb291f002008-10-18 20:26:44 -0700450 mm->locked_vm += nr_pages;
451
452 /*
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -0700453 * vm_flags is protected by the mmap_lock held in write mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 * It's okay if try_to_unmap_one unmaps a page just after we
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -0700455 * set VM_LOCKED, populate_vma_page_range will bring it back.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
Hugh Dickins34b67922022-02-14 18:31:48 -0800458 if ((newflags & VM_LOCKED) && (oldflags & VM_LOCKED)) {
459 /* No work to do, and mlocking twice would be wrong */
Suren Baghdasaryan1c712222023-01-26 11:37:49 -0800460 vm_flags_reset(vma, newflags);
Hugh Dickins34b67922022-02-14 18:31:48 -0800461 } else {
462 mlock_vma_pages_range(vma, start, end, newflags);
463 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464out:
Nick Pigginb291f002008-10-18 20:26:44 -0700465 *prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 return ret;
467}
468
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800469static int apply_vma_lock_flags(unsigned long start, size_t len,
470 vm_flags_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471{
472 unsigned long nstart, end, tmp;
Zhiyuan Dai68d68ff2021-05-04 18:40:12 -0700473 struct vm_area_struct *vma, *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 int error;
Liam R. Howlett37598f52023-01-20 11:26:19 -0500475 VMA_ITERATOR(vmi, current->mm, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476
Alexander Kuleshov8fd9e482015-11-05 18:46:49 -0800477 VM_BUG_ON(offset_in_page(start));
Michel Lespinassefed067d2011-01-13 15:46:10 -0800478 VM_BUG_ON(len != PAGE_ALIGN(len));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 end = start + len;
480 if (end < start)
481 return -EINVAL;
482 if (end == start)
483 return 0;
Liam R. Howlett37598f52023-01-20 11:26:19 -0500484 vma = vma_iter_load(&vmi);
Matthew Wilcox (Oracle)33108b02022-09-06 19:49:02 +0000485 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 return -ENOMEM;
487
Liam R. Howlett37598f52023-01-20 11:26:19 -0500488 prev = vma_prev(&vmi);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 if (start > vma->vm_start)
490 prev = vma;
491
Liam R. Howlett37598f52023-01-20 11:26:19 -0500492 nstart = start;
493 tmp = vma->vm_start;
494 for_each_vma_range(vmi, vma, end) {
495 vm_flags_t newflags;
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800496
Liam R. Howlett37598f52023-01-20 11:26:19 -0500497 if (vma->vm_start != tmp)
498 return -ENOMEM;
499
Suren Baghdasaryane430a952023-01-26 11:37:48 -0800500 newflags = vma->vm_flags & ~VM_LOCKED_MASK;
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800501 newflags |= flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 tmp = vma->vm_end;
504 if (tmp > end)
505 tmp = end;
Liam R. Howlett37598f52023-01-20 11:26:19 -0500506 error = mlock_fixup(&vmi, vma, &prev, nstart, tmp, newflags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 if (error)
508 break;
509 nstart = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 }
Liam R. Howlett37598f52023-01-20 11:26:19 -0500511
512 if (vma_iter_end(&vmi) < end)
513 return -ENOMEM;
514
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 return error;
516}
517
Simon Guo0cf2f6f2016-10-07 16:59:36 -0700518/*
519 * Go through vma areas and sum size of mlocked
520 * vma pages, as return value.
521 * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT)
522 * is also counted.
523 * Return value: previously mlocked page counts
524 */
swkhack0874bb42019-06-13 15:56:08 -0700525static unsigned long count_mm_mlocked_page_nr(struct mm_struct *mm,
Simon Guo0cf2f6f2016-10-07 16:59:36 -0700526 unsigned long start, size_t len)
527{
528 struct vm_area_struct *vma;
swkhack0874bb42019-06-13 15:56:08 -0700529 unsigned long count = 0;
Matthew Wilcox (Oracle)33108b02022-09-06 19:49:02 +0000530 unsigned long end;
531 VMA_ITERATOR(vmi, mm, start);
Simon Guo0cf2f6f2016-10-07 16:59:36 -0700532
Matthew Wilcox (Oracle)33108b02022-09-06 19:49:02 +0000533 /* Don't overflow past ULONG_MAX */
534 if (unlikely(ULONG_MAX - len < start))
535 end = ULONG_MAX;
536 else
537 end = start + len;
Liam Howlett66071892022-06-15 17:40:58 +0000538
Matthew Wilcox (Oracle)33108b02022-09-06 19:49:02 +0000539 for_each_vma_range(vmi, vma, end) {
Simon Guo0cf2f6f2016-10-07 16:59:36 -0700540 if (vma->vm_flags & VM_LOCKED) {
541 if (start > vma->vm_start)
542 count -= (start - vma->vm_start);
Matthew Wilcox (Oracle)33108b02022-09-06 19:49:02 +0000543 if (end < vma->vm_end) {
544 count += end - vma->vm_start;
Simon Guo0cf2f6f2016-10-07 16:59:36 -0700545 break;
546 }
547 count += vma->vm_end - vma->vm_start;
548 }
549 }
550
551 return count >> PAGE_SHIFT;
552}
553
Hugh Dickinsebcbc6e2022-02-14 18:20:24 -0800554/*
555 * convert get_user_pages() return value to posix mlock() error
556 */
557static int __mlock_posix_error_return(long retval)
558{
559 if (retval == -EFAULT)
560 retval = -ENOMEM;
561 else if (retval == -ENOMEM)
562 retval = -EAGAIN;
563 return retval;
564}
565
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700566static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567{
568 unsigned long locked;
569 unsigned long lock_limit;
570 int error = -ENOMEM;
571
Andrey Konovalov057d33892019-09-25 16:48:30 -0700572 start = untagged_addr(start);
573
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 if (!can_do_mlock())
575 return -EPERM;
576
Alexander Kuleshov8fd9e482015-11-05 18:46:49 -0800577 len = PAGE_ALIGN(len + (offset_in_page(start)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 start &= PAGE_MASK;
579
Jiri Slaby59e99e52010-03-05 13:41:44 -0800580 lock_limit = rlimit(RLIMIT_MEMLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 lock_limit >>= PAGE_SHIFT;
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800582 locked = len >> PAGE_SHIFT;
583
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700584 if (mmap_write_lock_killable(current->mm))
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700585 return -EINTR;
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800586
587 locked += current->mm->locked_vm;
Simon Guo0cf2f6f2016-10-07 16:59:36 -0700588 if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) {
589 /*
590 * It is possible that the regions requested intersect with
591 * previously mlocked areas, that part area in "mm->locked_vm"
592 * should not be counted to new mlock increment count. So check
593 * and adjust locked count if necessary.
594 */
595 locked -= count_mm_mlocked_page_nr(current->mm,
596 start, len);
597 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
599 /* check against resource limits */
600 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800601 error = apply_vma_lock_flags(start, len, flags);
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800602
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700603 mmap_write_unlock(current->mm);
Kirill A. Shutemovc5612592015-04-14 15:44:42 -0700604 if (error)
605 return error;
606
607 error = __mm_populate(start, len, 0);
608 if (error)
609 return __mlock_posix_error_return(error);
610 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611}
612
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800613SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
614{
615 return do_mlock(start, len, VM_LOCKED);
616}
617
Eric B Munsona8ca5d02015-11-05 18:51:33 -0800618SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags)
619{
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800620 vm_flags_t vm_flags = VM_LOCKED;
621
622 if (flags & ~MLOCK_ONFAULT)
Eric B Munsona8ca5d02015-11-05 18:51:33 -0800623 return -EINVAL;
624
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800625 if (flags & MLOCK_ONFAULT)
626 vm_flags |= VM_LOCKONFAULT;
627
628 return do_mlock(start, len, vm_flags);
Eric B Munsona8ca5d02015-11-05 18:51:33 -0800629}
630
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100631SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632{
633 int ret;
634
Andrey Konovalov057d33892019-09-25 16:48:30 -0700635 start = untagged_addr(start);
636
Alexander Kuleshov8fd9e482015-11-05 18:46:49 -0800637 len = PAGE_ALIGN(len + (offset_in_page(start)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 start &= PAGE_MASK;
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800639
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700640 if (mmap_write_lock_killable(current->mm))
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700641 return -EINTR;
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800642 ret = apply_vma_lock_flags(start, len, 0);
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700643 mmap_write_unlock(current->mm);
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800644
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 return ret;
646}
647
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800648/*
649 * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall)
650 * and translate into the appropriate modifications to mm->def_flags and/or the
651 * flags for all current VMAs.
652 *
653 * There are a couple of subtleties with this. If mlockall() is called multiple
654 * times with different flags, the values do not necessarily stack. If mlockall
655 * is called once including the MCL_FUTURE flag and then a second time without
656 * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags.
657 */
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800658static int apply_mlockall_flags(int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659{
Liam R. Howlett37598f52023-01-20 11:26:19 -0500660 VMA_ITERATOR(vmi, current->mm, 0);
Zhiyuan Dai68d68ff2021-05-04 18:40:12 -0700661 struct vm_area_struct *vma, *prev = NULL;
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800662 vm_flags_t to_add = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
Suren Baghdasaryane430a952023-01-26 11:37:48 -0800664 current->mm->def_flags &= ~VM_LOCKED_MASK;
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800665 if (flags & MCL_FUTURE) {
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700666 current->mm->def_flags |= VM_LOCKED;
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800667
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800668 if (flags & MCL_ONFAULT)
669 current->mm->def_flags |= VM_LOCKONFAULT;
670
671 if (!(flags & MCL_CURRENT))
672 goto out;
673 }
674
675 if (flags & MCL_CURRENT) {
676 to_add |= VM_LOCKED;
677 if (flags & MCL_ONFAULT)
678 to_add |= VM_LOCKONFAULT;
679 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680
Liam R. Howlett37598f52023-01-20 11:26:19 -0500681 for_each_vma(vmi, vma) {
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900682 vm_flags_t newflags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683
Suren Baghdasaryane430a952023-01-26 11:37:48 -0800684 newflags = vma->vm_flags & ~VM_LOCKED_MASK;
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800685 newflags |= to_add;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686
687 /* Ignore errors */
Liam R. Howlett37598f52023-01-20 11:26:19 -0500688 mlock_fixup(&vmi, vma, &prev, vma->vm_start, vma->vm_end,
689 newflags);
Paul E. McKenney50d4fb72017-10-24 08:22:18 -0700690 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 }
692out:
693 return 0;
694}
695
Heiko Carstens3480b252009-01-14 14:14:16 +0100696SYSCALL_DEFINE1(mlockall, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697{
698 unsigned long lock_limit;
Alexey Klimov86d2adcc2015-11-05 18:46:00 -0800699 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
Potyra, Stefandedca632019-06-13 15:55:55 -0700701 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)) ||
702 flags == MCL_ONFAULT)
Alexey Klimov86d2adcc2015-11-05 18:46:00 -0800703 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 if (!can_do_mlock())
Alexey Klimov86d2adcc2015-11-05 18:46:00 -0800706 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707
Jiri Slaby59e99e52010-03-05 13:41:44 -0800708 lock_limit = rlimit(RLIMIT_MEMLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 lock_limit >>= PAGE_SHIFT;
710
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700711 if (mmap_write_lock_killable(current->mm))
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700712 return -EINTR;
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800713
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700714 ret = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
716 capable(CAP_IPC_LOCK))
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800717 ret = apply_mlockall_flags(flags);
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700718 mmap_write_unlock(current->mm);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -0800719 if (!ret && (flags & MCL_CURRENT))
720 mm_populate(0, TASK_SIZE);
Alexey Klimov86d2adcc2015-11-05 18:46:00 -0800721
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 return ret;
723}
724
Heiko Carstens3480b252009-01-14 14:14:16 +0100725SYSCALL_DEFINE0(munlockall)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726{
727 int ret;
728
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700729 if (mmap_write_lock_killable(current->mm))
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700730 return -EINTR;
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800731 ret = apply_mlockall_flags(0);
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700732 mmap_write_unlock(current->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 return ret;
734}
735
736/*
737 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
738 * shm segments) get accounted against the user_struct instead.
739 */
740static DEFINE_SPINLOCK(shmlock_user_lock);
741
Alexey Gladkovd7c9e992021-04-22 14:27:14 +0200742int user_shm_lock(size_t size, struct ucounts *ucounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743{
744 unsigned long lock_limit, locked;
Alexey Gladkovd7c9e992021-04-22 14:27:14 +0200745 long memlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 int allowed = 0;
747
748 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jiri Slaby59e99e52010-03-05 13:41:44 -0800749 lock_limit = rlimit(RLIMIT_MEMLOCK);
Miaohe Line97824f2022-03-22 16:09:18 +0800750 if (lock_limit != RLIM_INFINITY)
751 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 spin_lock(&shmlock_user_lock);
Alexey Gladkovd7c9e992021-04-22 14:27:14 +0200753 memlock = inc_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
754
Miaohe Line97824f2022-03-22 16:09:18 +0800755 if ((memlock == LONG_MAX || memlock > lock_limit) && !capable(CAP_IPC_LOCK)) {
Alexey Gladkovd7c9e992021-04-22 14:27:14 +0200756 dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 goto out;
Alexey Gladkovd7c9e992021-04-22 14:27:14 +0200758 }
759 if (!get_ucounts(ucounts)) {
760 dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
Miaohe Lin5c2a9562022-03-22 14:44:56 -0700761 allowed = 0;
Alexey Gladkovd7c9e992021-04-22 14:27:14 +0200762 goto out;
763 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 allowed = 1;
765out:
766 spin_unlock(&shmlock_user_lock);
767 return allowed;
768}
769
Alexey Gladkovd7c9e992021-04-22 14:27:14 +0200770void user_shm_unlock(size_t size, struct ucounts *ucounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771{
772 spin_lock(&shmlock_user_lock);
Alexey Gladkovd7c9e992021-04-22 14:27:14 +0200773 dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, (size + PAGE_SIZE - 1) >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 spin_unlock(&shmlock_user_lock);
Alexey Gladkovd7c9e992021-04-22 14:27:14 +0200775 put_ucounts(ucounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776}