blob: e3e3dc2b295639306554761f8869c0bfe8bd16eb [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);
Matthew Wilcox (Oracle)99fbb6b2024-02-27 17:42:35 +0000209 folios_put(fbatch);
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800210}
211
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000212void mlock_drain_local(void)
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800213{
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000214 struct folio_batch *fbatch;
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800215
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000216 local_lock(&mlock_fbatch.lock);
217 fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
218 if (folio_batch_count(fbatch))
219 mlock_folio_batch(fbatch);
220 local_unlock(&mlock_fbatch.lock);
Sebastian Andrzej Siewioradb11e72022-04-01 11:28:33 -0700221}
222
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000223void mlock_drain_remote(int cpu)
Sebastian Andrzej Siewioradb11e72022-04-01 11:28:33 -0700224{
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000225 struct folio_batch *fbatch;
Sebastian Andrzej Siewioradb11e72022-04-01 11:28:33 -0700226
227 WARN_ON_ONCE(cpu_online(cpu));
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000228 fbatch = &per_cpu(mlock_fbatch.fbatch, cpu);
229 if (folio_batch_count(fbatch))
230 mlock_folio_batch(fbatch);
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800231}
232
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000233bool need_mlock_drain(int cpu)
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800234{
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000235 return folio_batch_count(&per_cpu(mlock_fbatch.fbatch, cpu));
Lee Schermerhorn6927c1d2009-12-14 17:59:55 -0800236}
Nick Pigginb291f002008-10-18 20:26:44 -0700237
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800238/**
Matthew Wilcox (Oracle)dcc5d332022-02-15 13:33:59 -0500239 * mlock_folio - mlock a folio already on (or temporarily off) LRU
240 * @folio: folio to be mlocked.
Hugh Dickins3d470fc2011-10-31 17:09:43 -0700241 */
Matthew Wilcox (Oracle)dcc5d332022-02-15 13:33:59 -0500242void mlock_folio(struct folio *folio)
Hugh Dickins3d470fc2011-10-31 17:09:43 -0700243{
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000244 struct folio_batch *fbatch;
Sebastian Andrzej Siewioradb11e72022-04-01 11:28:33 -0700245
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000246 local_lock(&mlock_fbatch.lock);
247 fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
Hugh Dickins3d470fc2011-10-31 17:09:43 -0700248
Matthew Wilcox (Oracle)dcc5d332022-02-15 13:33:59 -0500249 if (!folio_test_set_mlocked(folio)) {
250 int nr_pages = folio_nr_pages(folio);
Hugh Dickins3d470fc2011-10-31 17:09:43 -0700251
Matthew Wilcox (Oracle)dcc5d332022-02-15 13:33:59 -0500252 zone_stat_mod_folio(folio, NR_MLOCK, nr_pages);
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800253 __count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages);
Hugh Dickins53f79ac2009-12-14 17:58:58 -0800254 }
Nick Piggin5344b7e2008-10-18 20:26:51 -0700255
Matthew Wilcox (Oracle)dcc5d332022-02-15 13:33:59 -0500256 folio_get(folio);
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000257 if (!folio_batch_add(fbatch, mlock_lru(folio)) ||
Matthew Wilcox (Oracle)dcc5d332022-02-15 13:33:59 -0500258 folio_test_large(folio) || lru_cache_disabled())
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000259 mlock_folio_batch(fbatch);
260 local_unlock(&mlock_fbatch.lock);
Nick Piggin5344b7e2008-10-18 20:26:51 -0700261}
262
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800263/**
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000264 * mlock_new_folio - mlock a newly allocated folio not yet on LRU
265 * @folio: folio to be mlocked, either normal or a THP head.
Lee Schermerhorn6927c1d2009-12-14 17:59:55 -0800266 */
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000267void mlock_new_folio(struct folio *folio)
Lee Schermerhorn6927c1d2009-12-14 17:59:55 -0800268{
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000269 struct folio_batch *fbatch;
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000270 int nr_pages = folio_nr_pages(folio);
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800271
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000272 local_lock(&mlock_fbatch.lock);
273 fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
274 folio_set_mlocked(folio);
275
276 zone_stat_mod_folio(folio, NR_MLOCK, nr_pages);
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800277 __count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages);
278
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000279 folio_get(folio);
280 if (!folio_batch_add(fbatch, mlock_new(folio)) ||
281 folio_test_large(folio) || lru_cache_disabled())
282 mlock_folio_batch(fbatch);
283 local_unlock(&mlock_fbatch.lock);
Nick Piggin5344b7e2008-10-18 20:26:51 -0700284}
285
Hugh Dickins2fbb0c12022-02-14 18:37:29 -0800286/**
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000287 * munlock_folio - munlock a folio
288 * @folio: folio to be munlocked, either normal or a THP head.
Rik van Rielba470de2008-10-18 20:26:50 -0700289 */
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000290void munlock_folio(struct folio *folio)
Hugh Dickins408e82b2009-09-21 17:03:23 -0700291{
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000292 struct folio_batch *fbatch;
Vlastimil Babka56afe477d2013-09-11 14:22:32 -0700293
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000294 local_lock(&mlock_fbatch.lock);
295 fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
Vlastimil Babka5b409982013-09-11 14:22:33 -0700296 /*
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000297 * folio_test_clear_mlocked(folio) must be left to __munlock_folio(),
298 * which will check whether the folio is multiply mlocked.
Vlastimil Babka5b409982013-09-11 14:22:33 -0700299 */
Lorenzo Stoakes90d07212023-01-12 12:39:29 +0000300 folio_get(folio);
301 if (!folio_batch_add(fbatch, folio) ||
302 folio_test_large(folio) || lru_cache_disabled())
303 mlock_folio_batch(fbatch);
304 local_unlock(&mlock_fbatch.lock);
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700305}
Vlastimil Babka56afe477d2013-09-11 14:22:32 -0700306
Yin Fengweidc68bad2023-09-18 15:33:18 +0800307static inline unsigned int folio_mlock_step(struct folio *folio,
308 pte_t *pte, unsigned long addr, unsigned long end)
309{
Lance Yangf7428292024-06-11 09:04:18 +0800310 const fpb_t fpb_flags = FPB_IGNORE_DIRTY | FPB_IGNORE_SOFT_DIRTY;
311 unsigned int count = (end - addr) >> PAGE_SHIFT;
Yin Fengweidc68bad2023-09-18 15:33:18 +0800312 pte_t ptent = ptep_get(pte);
313
314 if (!folio_test_large(folio))
315 return 1;
316
Lance Yangf7428292024-06-11 09:04:18 +0800317 return folio_pte_batch(folio, addr, pte, ptent, count, fpb_flags, NULL,
318 NULL, NULL);
Yin Fengweidc68bad2023-09-18 15:33:18 +0800319}
320
321static inline bool allow_mlock_munlock(struct folio *folio,
322 struct vm_area_struct *vma, unsigned long start,
323 unsigned long end, unsigned int step)
324{
325 /*
326 * For unlock, allow munlock large folio which is partially
327 * mapped to VMA. As it's possible that large folio is
328 * mlocked and VMA is split later.
329 *
330 * During memory pressure, such kind of large folio can
331 * be split. And the pages are not in VM_LOCKed VMA
332 * can be reclaimed.
333 */
334 if (!(vma->vm_flags & VM_LOCKED))
335 return true;
336
Hugh Dickinsb1454b42023-10-23 23:38:41 -0700337 /* folio_within_range() cannot take KSM, but any small folio is OK */
338 if (!folio_test_large(folio))
339 return true;
340
Yin Fengweidc68bad2023-09-18 15:33:18 +0800341 /* folio not in range [start, end), skip mlock */
342 if (!folio_within_range(folio, vma, start, end))
343 return false;
344
345 /* folio is not fully mapped, skip mlock */
346 if (step != folio_nr_pages(folio))
347 return false;
348
349 return true;
350}
351
Hugh Dickins34b67922022-02-14 18:31:48 -0800352static int mlock_pte_range(pmd_t *pmd, unsigned long addr,
353 unsigned long end, struct mm_walk *walk)
Hugh Dickins408e82b2009-09-21 17:03:23 -0700354
Hugh Dickins408e82b2009-09-21 17:03:23 -0700355{
Hugh Dickins34b67922022-02-14 18:31:48 -0800356 struct vm_area_struct *vma = walk->vma;
Hugh Dickins6e919712009-09-21 17:03:32 -0700357 spinlock_t *ptl;
Hugh Dickins34b67922022-02-14 18:31:48 -0800358 pte_t *start_pte, *pte;
Ryan Robertsc33c7942023-06-12 16:15:45 +0100359 pte_t ptent;
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000360 struct folio *folio;
Yin Fengweidc68bad2023-09-18 15:33:18 +0800361 unsigned int step = 1;
362 unsigned long start = addr;
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800363
Hugh Dickins34b67922022-02-14 18:31:48 -0800364 ptl = pmd_trans_huge_lock(pmd, vma);
365 if (ptl) {
366 if (!pmd_present(*pmd))
367 goto out;
368 if (is_huge_zero_pmd(*pmd))
369 goto out;
Matthew Wilcox (Oracle)e06d03d2024-03-26 20:28:23 +0000370 folio = pmd_folio(*pmd);
Hugh Dickins34b67922022-02-14 18:31:48 -0800371 if (vma->vm_flags & VM_LOCKED)
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000372 mlock_folio(folio);
Hugh Dickins34b67922022-02-14 18:31:48 -0800373 else
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000374 munlock_folio(folio);
Hugh Dickins34b67922022-02-14 18:31:48 -0800375 goto out;
Nick Pigginb291f002008-10-18 20:26:44 -0700376 }
Hugh Dickins34b67922022-02-14 18:31:48 -0800377
378 start_pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Hugh Dickins7780d042023-06-08 18:17:26 -0700379 if (!start_pte) {
380 walk->action = ACTION_AGAIN;
381 return 0;
382 }
Yin Fengweidc68bad2023-09-18 15:33:18 +0800383
Hugh Dickins34b67922022-02-14 18:31:48 -0800384 for (pte = start_pte; addr != end; pte++, addr += PAGE_SIZE) {
Ryan Robertsc33c7942023-06-12 16:15:45 +0100385 ptent = ptep_get(pte);
386 if (!pte_present(ptent))
Hugh Dickins34b67922022-02-14 18:31:48 -0800387 continue;
Ryan Robertsc33c7942023-06-12 16:15:45 +0100388 folio = vm_normal_folio(vma, addr, ptent);
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000389 if (!folio || folio_is_zone_device(folio))
Hugh Dickins34b67922022-02-14 18:31:48 -0800390 continue;
Yin Fengweidc68bad2023-09-18 15:33:18 +0800391
392 step = folio_mlock_step(folio, pte, addr, end);
393 if (!allow_mlock_munlock(folio, vma, start, end, step))
394 goto next_entry;
395
Hugh Dickins34b67922022-02-14 18:31:48 -0800396 if (vma->vm_flags & VM_LOCKED)
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000397 mlock_folio(folio);
Hugh Dickins34b67922022-02-14 18:31:48 -0800398 else
Lorenzo Stoakes96f97c42023-01-12 12:39:31 +0000399 munlock_folio(folio);
Yin Fengweidc68bad2023-09-18 15:33:18 +0800400
401next_entry:
402 pte += step - 1;
403 addr += (step - 1) << PAGE_SHIFT;
Hugh Dickins34b67922022-02-14 18:31:48 -0800404 }
405 pte_unmap(start_pte);
406out:
407 spin_unlock(ptl);
408 cond_resched();
409 return 0;
Nick Pigginb291f002008-10-18 20:26:44 -0700410}
411
412/*
Hugh Dickins34b67922022-02-14 18:31:48 -0800413 * mlock_vma_pages_range() - mlock any pages already in the range,
414 * or munlock all pages in the range.
415 * @vma - vma containing range to be mlock()ed or munlock()ed
Rik van Rielba470de2008-10-18 20:26:50 -0700416 * @start - start address in @vma of the range
Hugh Dickins34b67922022-02-14 18:31:48 -0800417 * @end - end of range in @vma
418 * @newflags - the new set of flags for @vma.
Rik van Rielba470de2008-10-18 20:26:50 -0700419 *
Hugh Dickins34b67922022-02-14 18:31:48 -0800420 * Called for mlock(), mlock2() and mlockall(), to set @vma VM_LOCKED;
421 * called for munlock() and munlockall(), to clear VM_LOCKED from @vma.
Nick Pigginb291f002008-10-18 20:26:44 -0700422 */
Hugh Dickins34b67922022-02-14 18:31:48 -0800423static void mlock_vma_pages_range(struct vm_area_struct *vma,
424 unsigned long start, unsigned long end, vm_flags_t newflags)
Nick Pigginb291f002008-10-18 20:26:44 -0700425{
Hugh Dickins34b67922022-02-14 18:31:48 -0800426 static const struct mm_walk_ops mlock_walk_ops = {
427 .pmd_entry = mlock_pte_range,
Suren Baghdasaryan49b06382023-08-04 08:27:19 -0700428 .walk_lock = PGWALK_WRLOCK_VERIFY,
Hugh Dickins34b67922022-02-14 18:31:48 -0800429 };
Hugh Dickins408e82b2009-09-21 17:03:23 -0700430
Hugh Dickins34b67922022-02-14 18:31:48 -0800431 /*
432 * There is a slight chance that concurrent page migration,
433 * or page reclaim finding a page of this now-VM_LOCKED vma,
Matthew Wilcox (Oracle)7efecff2023-01-16 19:28:25 +0000434 * will call mlock_vma_folio() and raise page's mlock_count:
Hugh Dickins34b67922022-02-14 18:31:48 -0800435 * double counting, leaving the page unevictable indefinitely.
Matthew Wilcox (Oracle)7efecff2023-01-16 19:28:25 +0000436 * Communicate this danger to mlock_vma_folio() with VM_IO,
Hugh Dickins34b67922022-02-14 18:31:48 -0800437 * which is a VM_SPECIAL flag not allowed on VM_LOCKED vmas.
438 * mmap_lock is held in write mode here, so this weird
439 * combination should not be visible to other mmap_lock users;
440 * but WRITE_ONCE so rmap walkers must see VM_IO if VM_LOCKED.
441 */
442 if (newflags & VM_LOCKED)
443 newflags |= VM_IO;
Suren Baghdasaryan60081bf2023-08-04 08:27:22 -0700444 vma_start_write(vma);
Suren Baghdasaryan601c3c22023-01-31 16:01:16 -0800445 vm_flags_reset_once(vma, newflags);
Michel Lespinasseff6a6da2013-02-27 17:02:44 -0800446
Hugh Dickins34b67922022-02-14 18:31:48 -0800447 lru_add_drain();
448 walk_page_range(vma->vm_mm, start, end, &mlock_walk_ops, NULL);
449 lru_add_drain();
Vlastimil Babka7a8010c2013-09-11 14:22:35 -0700450
Hugh Dickins34b67922022-02-14 18:31:48 -0800451 if (newflags & VM_IO) {
452 newflags &= ~VM_IO;
Suren Baghdasaryan601c3c22023-01-31 16:01:16 -0800453 vm_flags_reset_once(vma, newflags);
Hugh Dickins408e82b2009-09-21 17:03:23 -0700454 }
Nick Pigginb291f002008-10-18 20:26:44 -0700455}
456
457/*
458 * mlock_fixup - handle mlock[all]/munlock[all] requests.
459 *
460 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
461 * munlock is a no-op. However, for some special vmas, we go ahead and
Michel Lespinassecea10a12013-02-22 16:32:44 -0800462 * populate the ptes.
Nick Pigginb291f002008-10-18 20:26:44 -0700463 *
464 * For vmas that pass the filters, merge/split as appropriate.
465 */
Liam R. Howlett37598f52023-01-20 11:26:19 -0500466static int mlock_fixup(struct vma_iterator *vmi, struct vm_area_struct *vma,
467 struct vm_area_struct **prev, unsigned long start,
468 unsigned long end, vm_flags_t newflags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469{
Nick Pigginb291f002008-10-18 20:26:44 -0700470 struct mm_struct *mm = vma->vm_mm;
Nick Pigginb291f002008-10-18 20:26:44 -0700471 int nr_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 int ret = 0;
Hugh Dickins34b67922022-02-14 18:31:48 -0800473 vm_flags_t oldflags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474
Hugh Dickins34b67922022-02-14 18:31:48 -0800475 if (newflags == oldflags || (oldflags & VM_SPECIAL) ||
Dave Jiange1fb4a02018-08-17 15:43:40 -0700476 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm) ||
Jason A. Donenfeld9651fce2022-12-08 17:55:04 +0100477 vma_is_dax(vma) || vma_is_secretmem(vma) || (oldflags & VM_DROPPABLE))
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800478 /* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */
479 goto out;
Nick Pigginb291f002008-10-18 20:26:44 -0700480
Lorenzo Stoakes94d7d922023-10-11 18:04:28 +0100481 vma = vma_modify_flags(vmi, *prev, vma, start, end, newflags);
482 if (IS_ERR(vma)) {
483 ret = PTR_ERR(vma);
484 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 }
486
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 /*
Nick Pigginb291f002008-10-18 20:26:44 -0700488 * Keep track of amount of locked VM.
489 */
490 nr_pages = (end - start) >> PAGE_SHIFT;
Hugh Dickins34b67922022-02-14 18:31:48 -0800491 if (!(newflags & VM_LOCKED))
Nick Pigginb291f002008-10-18 20:26:44 -0700492 nr_pages = -nr_pages;
Hugh Dickins34b67922022-02-14 18:31:48 -0800493 else if (oldflags & VM_LOCKED)
Simon Guob155b4f2016-10-07 16:59:40 -0700494 nr_pages = 0;
Nick Pigginb291f002008-10-18 20:26:44 -0700495 mm->locked_vm += nr_pages;
496
497 /*
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -0700498 * vm_flags is protected by the mmap_lock held in write mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 * It's okay if try_to_unmap_one unmaps a page just after we
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -0700500 * set VM_LOCKED, populate_vma_page_range will bring it back.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 */
Hugh Dickins34b67922022-02-14 18:31:48 -0800502 if ((newflags & VM_LOCKED) && (oldflags & VM_LOCKED)) {
503 /* No work to do, and mlocking twice would be wrong */
Suren Baghdasaryan60081bf2023-08-04 08:27:22 -0700504 vma_start_write(vma);
Suren Baghdasaryan1c712222023-01-26 11:37:49 -0800505 vm_flags_reset(vma, newflags);
Hugh Dickins34b67922022-02-14 18:31:48 -0800506 } else {
507 mlock_vma_pages_range(vma, start, end, newflags);
508 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509out:
Nick Pigginb291f002008-10-18 20:26:44 -0700510 *prev = vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 return ret;
512}
513
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800514static int apply_vma_lock_flags(unsigned long start, size_t len,
515 vm_flags_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516{
517 unsigned long nstart, end, tmp;
Zhiyuan Dai68d68ff2021-05-04 18:40:12 -0700518 struct vm_area_struct *vma, *prev;
Liam R. Howlett37598f52023-01-20 11:26:19 -0500519 VMA_ITERATOR(vmi, current->mm, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
Alexander Kuleshov8fd9e482015-11-05 18:46:49 -0800521 VM_BUG_ON(offset_in_page(start));
Michel Lespinassefed067d2011-01-13 15:46:10 -0800522 VM_BUG_ON(len != PAGE_ALIGN(len));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 end = start + len;
524 if (end < start)
525 return -EINVAL;
526 if (end == start)
527 return 0;
Liam R. Howlett37598f52023-01-20 11:26:19 -0500528 vma = vma_iter_load(&vmi);
Matthew Wilcox (Oracle)33108b02022-09-06 19:49:02 +0000529 if (!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 return -ENOMEM;
531
Liam R. Howlett37598f52023-01-20 11:26:19 -0500532 prev = vma_prev(&vmi);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 if (start > vma->vm_start)
534 prev = vma;
535
Liam R. Howlett37598f52023-01-20 11:26:19 -0500536 nstart = start;
537 tmp = vma->vm_start;
538 for_each_vma_range(vmi, vma, end) {
Liam R. Howlett2658f94d62023-07-11 13:50:20 -0400539 int error;
Liam R. Howlett37598f52023-01-20 11:26:19 -0500540 vm_flags_t newflags;
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800541
Liam R. Howlett37598f52023-01-20 11:26:19 -0500542 if (vma->vm_start != tmp)
543 return -ENOMEM;
544
Suren Baghdasaryane430a952023-01-26 11:37:48 -0800545 newflags = vma->vm_flags & ~VM_LOCKED_MASK;
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800546 newflags |= flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 tmp = vma->vm_end;
549 if (tmp > end)
550 tmp = end;
Liam R. Howlett37598f52023-01-20 11:26:19 -0500551 error = mlock_fixup(&vmi, vma, &prev, nstart, tmp, newflags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 if (error)
Liam R. Howlett2658f94d62023-07-11 13:50:20 -0400553 return error;
554 tmp = vma_iter_end(&vmi);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 nstart = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 }
Liam R. Howlett37598f52023-01-20 11:26:19 -0500557
Liam R. Howlett2658f94d62023-07-11 13:50:20 -0400558 if (tmp < end)
Liam R. Howlett37598f52023-01-20 11:26:19 -0500559 return -ENOMEM;
560
Liam R. Howlett2658f94d62023-07-11 13:50:20 -0400561 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562}
563
Simon Guo0cf2f6f2016-10-07 16:59:36 -0700564/*
565 * Go through vma areas and sum size of mlocked
566 * vma pages, as return value.
567 * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT)
568 * is also counted.
569 * Return value: previously mlocked page counts
570 */
swkhack0874bb42019-06-13 15:56:08 -0700571static unsigned long count_mm_mlocked_page_nr(struct mm_struct *mm,
Simon Guo0cf2f6f2016-10-07 16:59:36 -0700572 unsigned long start, size_t len)
573{
574 struct vm_area_struct *vma;
swkhack0874bb42019-06-13 15:56:08 -0700575 unsigned long count = 0;
Matthew Wilcox (Oracle)33108b02022-09-06 19:49:02 +0000576 unsigned long end;
577 VMA_ITERATOR(vmi, mm, start);
Simon Guo0cf2f6f2016-10-07 16:59:36 -0700578
Matthew Wilcox (Oracle)33108b02022-09-06 19:49:02 +0000579 /* Don't overflow past ULONG_MAX */
580 if (unlikely(ULONG_MAX - len < start))
581 end = ULONG_MAX;
582 else
583 end = start + len;
Liam Howlett66071892022-06-15 17:40:58 +0000584
Matthew Wilcox (Oracle)33108b02022-09-06 19:49:02 +0000585 for_each_vma_range(vmi, vma, end) {
Simon Guo0cf2f6f2016-10-07 16:59:36 -0700586 if (vma->vm_flags & VM_LOCKED) {
587 if (start > vma->vm_start)
588 count -= (start - vma->vm_start);
Matthew Wilcox (Oracle)33108b02022-09-06 19:49:02 +0000589 if (end < vma->vm_end) {
590 count += end - vma->vm_start;
Simon Guo0cf2f6f2016-10-07 16:59:36 -0700591 break;
592 }
593 count += vma->vm_end - vma->vm_start;
594 }
595 }
596
597 return count >> PAGE_SHIFT;
598}
599
Hugh Dickinsebcbc6e2022-02-14 18:20:24 -0800600/*
601 * convert get_user_pages() return value to posix mlock() error
602 */
603static int __mlock_posix_error_return(long retval)
604{
605 if (retval == -EFAULT)
606 retval = -ENOMEM;
607 else if (retval == -ENOMEM)
608 retval = -EAGAIN;
609 return retval;
610}
611
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700612static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613{
614 unsigned long locked;
615 unsigned long lock_limit;
616 int error = -ENOMEM;
617
Andrey Konovalov057d33892019-09-25 16:48:30 -0700618 start = untagged_addr(start);
619
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 if (!can_do_mlock())
621 return -EPERM;
622
Alexander Kuleshov8fd9e482015-11-05 18:46:49 -0800623 len = PAGE_ALIGN(len + (offset_in_page(start)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 start &= PAGE_MASK;
625
Jiri Slaby59e99e52010-03-05 13:41:44 -0800626 lock_limit = rlimit(RLIMIT_MEMLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 lock_limit >>= PAGE_SHIFT;
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800628 locked = len >> PAGE_SHIFT;
629
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700630 if (mmap_write_lock_killable(current->mm))
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700631 return -EINTR;
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800632
633 locked += current->mm->locked_vm;
Simon Guo0cf2f6f2016-10-07 16:59:36 -0700634 if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) {
635 /*
636 * It is possible that the regions requested intersect with
637 * previously mlocked areas, that part area in "mm->locked_vm"
638 * should not be counted to new mlock increment count. So check
639 * and adjust locked count if necessary.
640 */
641 locked -= count_mm_mlocked_page_nr(current->mm,
642 start, len);
643 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644
645 /* check against resource limits */
646 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800647 error = apply_vma_lock_flags(start, len, flags);
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800648
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700649 mmap_write_unlock(current->mm);
Kirill A. Shutemovc5612592015-04-14 15:44:42 -0700650 if (error)
651 return error;
652
653 error = __mm_populate(start, len, 0);
654 if (error)
655 return __mlock_posix_error_return(error);
656 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657}
658
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800659SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
660{
661 return do_mlock(start, len, VM_LOCKED);
662}
663
Eric B Munsona8ca5d02015-11-05 18:51:33 -0800664SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags)
665{
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800666 vm_flags_t vm_flags = VM_LOCKED;
667
668 if (flags & ~MLOCK_ONFAULT)
Eric B Munsona8ca5d02015-11-05 18:51:33 -0800669 return -EINVAL;
670
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800671 if (flags & MLOCK_ONFAULT)
672 vm_flags |= VM_LOCKONFAULT;
673
674 return do_mlock(start, len, vm_flags);
Eric B Munsona8ca5d02015-11-05 18:51:33 -0800675}
676
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100677SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678{
679 int ret;
680
Andrey Konovalov057d33892019-09-25 16:48:30 -0700681 start = untagged_addr(start);
682
Alexander Kuleshov8fd9e482015-11-05 18:46:49 -0800683 len = PAGE_ALIGN(len + (offset_in_page(start)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 start &= PAGE_MASK;
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800685
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700686 if (mmap_write_lock_killable(current->mm))
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700687 return -EINTR;
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800688 ret = apply_vma_lock_flags(start, len, 0);
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700689 mmap_write_unlock(current->mm);
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800690
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 return ret;
692}
693
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800694/*
695 * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall)
696 * and translate into the appropriate modifications to mm->def_flags and/or the
697 * flags for all current VMAs.
698 *
699 * There are a couple of subtleties with this. If mlockall() is called multiple
700 * times with different flags, the values do not necessarily stack. If mlockall
701 * is called once including the MCL_FUTURE flag and then a second time without
702 * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags.
703 */
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800704static int apply_mlockall_flags(int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705{
Liam R. Howlett37598f52023-01-20 11:26:19 -0500706 VMA_ITERATOR(vmi, current->mm, 0);
Zhiyuan Dai68d68ff2021-05-04 18:40:12 -0700707 struct vm_area_struct *vma, *prev = NULL;
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800708 vm_flags_t to_add = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709
Suren Baghdasaryane430a952023-01-26 11:37:48 -0800710 current->mm->def_flags &= ~VM_LOCKED_MASK;
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800711 if (flags & MCL_FUTURE) {
Michel Lespinasse09a9f1d2013-03-28 16:26:23 -0700712 current->mm->def_flags |= VM_LOCKED;
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800713
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800714 if (flags & MCL_ONFAULT)
715 current->mm->def_flags |= VM_LOCKONFAULT;
716
717 if (!(flags & MCL_CURRENT))
718 goto out;
719 }
720
721 if (flags & MCL_CURRENT) {
722 to_add |= VM_LOCKED;
723 if (flags & MCL_ONFAULT)
724 to_add |= VM_LOCKONFAULT;
725 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726
Liam R. Howlett37598f52023-01-20 11:26:19 -0500727 for_each_vma(vmi, vma) {
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900728 vm_flags_t newflags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729
Suren Baghdasaryane430a952023-01-26 11:37:48 -0800730 newflags = vma->vm_flags & ~VM_LOCKED_MASK;
Eric B Munsonb0f205c2015-11-05 18:51:39 -0800731 newflags |= to_add;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
733 /* Ignore errors */
Liam R. Howlett37598f52023-01-20 11:26:19 -0500734 mlock_fixup(&vmi, vma, &prev, vma->vm_start, vma->vm_end,
735 newflags);
Paul E. McKenney50d4fb72017-10-24 08:22:18 -0700736 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 }
738out:
739 return 0;
740}
741
Heiko Carstens3480b252009-01-14 14:14:16 +0100742SYSCALL_DEFINE1(mlockall, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743{
744 unsigned long lock_limit;
Alexey Klimov86d2adcc2015-11-05 18:46:00 -0800745 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746
Potyra, Stefandedca632019-06-13 15:55:55 -0700747 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)) ||
748 flags == MCL_ONFAULT)
Alexey Klimov86d2adcc2015-11-05 18:46:00 -0800749 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 if (!can_do_mlock())
Alexey Klimov86d2adcc2015-11-05 18:46:00 -0800752 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753
Jiri Slaby59e99e52010-03-05 13:41:44 -0800754 lock_limit = rlimit(RLIMIT_MEMLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 lock_limit >>= PAGE_SHIFT;
756
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700757 if (mmap_write_lock_killable(current->mm))
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700758 return -EINTR;
Davidlohr Bueso1f1cd702014-01-21 15:49:16 -0800759
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700760 ret = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
762 capable(CAP_IPC_LOCK))
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800763 ret = apply_mlockall_flags(flags);
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700764 mmap_write_unlock(current->mm);
Michel Lespinassebebeb3d2013-02-22 16:32:37 -0800765 if (!ret && (flags & MCL_CURRENT))
766 mm_populate(0, TASK_SIZE);
Alexey Klimov86d2adcc2015-11-05 18:46:00 -0800767
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 return ret;
769}
770
Heiko Carstens3480b252009-01-14 14:14:16 +0100771SYSCALL_DEFINE0(munlockall)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772{
773 int ret;
774
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700775 if (mmap_write_lock_killable(current->mm))
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700776 return -EINTR;
Eric B Munson1aab92ec2015-11-05 18:51:29 -0800777 ret = apply_mlockall_flags(0);
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700778 mmap_write_unlock(current->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 return ret;
780}
781
782/*
783 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
784 * shm segments) get accounted against the user_struct instead.
785 */
786static DEFINE_SPINLOCK(shmlock_user_lock);
787
Alexey Gladkovd7c9e992021-04-22 14:27:14 +0200788int user_shm_lock(size_t size, struct ucounts *ucounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789{
790 unsigned long lock_limit, locked;
Alexey Gladkovd7c9e992021-04-22 14:27:14 +0200791 long memlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 int allowed = 0;
793
794 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jiri Slaby59e99e52010-03-05 13:41:44 -0800795 lock_limit = rlimit(RLIMIT_MEMLOCK);
Miaohe Line97824f2022-03-22 16:09:18 +0800796 if (lock_limit != RLIM_INFINITY)
797 lock_limit >>= PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 spin_lock(&shmlock_user_lock);
Alexey Gladkovd7c9e992021-04-22 14:27:14 +0200799 memlock = inc_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
800
Miaohe Line97824f2022-03-22 16:09:18 +0800801 if ((memlock == LONG_MAX || memlock > lock_limit) && !capable(CAP_IPC_LOCK)) {
Alexey Gladkovd7c9e992021-04-22 14:27:14 +0200802 dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 goto out;
Alexey Gladkovd7c9e992021-04-22 14:27:14 +0200804 }
805 if (!get_ucounts(ucounts)) {
806 dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
Miaohe Lin5c2a9562022-03-22 14:44:56 -0700807 allowed = 0;
Alexey Gladkovd7c9e992021-04-22 14:27:14 +0200808 goto out;
809 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 allowed = 1;
811out:
812 spin_unlock(&shmlock_user_lock);
813 return allowed;
814}
815
Alexey Gladkovd7c9e992021-04-22 14:27:14 +0200816void user_shm_unlock(size_t size, struct ucounts *ucounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817{
818 spin_lock(&shmlock_user_lock);
Alexey Gladkovd7c9e992021-04-22 14:27:14 +0200819 dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, (size + PAGE_SIZE - 1) >> PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 spin_unlock(&shmlock_user_lock);
Alexey Gladkovd7c9e992021-04-22 14:27:14 +0200821 put_ucounts(ucounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822}