Eric Lee / smarc-ti-linux-kernel

1

/*

1

/*

2

* linux/mm/memory.c

2

* linux/mm/memory.c

3

*

3

*

4

5

*/

5

*/

6

7

/*

7

/*

8

* demand-loading started 01.12.91 - seems it is high on the list of

8

* demand-loading started 01.12.91 - seems it is high on the list of

9

* things wanted, and it should be easy to implement. - Linus

9

* things wanted, and it should be easy to implement. - Linus

10

*/

10

*/

11

12

/*

12

/*

13

* Ok, demand-loading was easy, shared pages a little bit tricker. Shared

13

* Ok, demand-loading was easy, shared pages a little bit tricker. Shared

14

* pages started 02.12.91, seems to work. - Linus.

14

* pages started 02.12.91, seems to work. - Linus.

15

*

15

*

16

* Tested sharing by executing about 30 /bin/sh: under the old kernel it

16

* Tested sharing by executing about 30 /bin/sh: under the old kernel it

17

* would have taken more than the 6M I have free, but it worked well as

17

* would have taken more than the 6M I have free, but it worked well as

18

* far as I could see.

18

* far as I could see.

19

*

19

*

20

* Also corrected some "invalidate()"s - I wasn't doing enough of them.

20

* Also corrected some "invalidate()"s - I wasn't doing enough of them.

21

*/

21

*/

22

23

/*

23

/*

24

* Real VM (paging to/from disk) started 18.12.91. Much more work and

24

* Real VM (paging to/from disk) started 18.12.91. Much more work and

25

* thought has to go into this. Oh, well..

25

* thought has to go into this. Oh, well..

26

* 19.12.91 - works, somewhat. Sometimes I get faults, don't know why.

26

* 19.12.91 - works, somewhat. Sometimes I get faults, don't know why.

27

* Found it. Everything seems to work now.

27

* Found it. Everything seems to work now.

28

* 20.12.91 - Ok, making the swap-device changeable like the root.

28

* 20.12.91 - Ok, making the swap-device changeable like the root.

29

*/

29

*/

30

31

/*

31

/*

32

* 05.04.94 - Multi-page memory management added for v1.1.

32

* 05.04.94 - Multi-page memory management added for v1.1.

33

* Idea by Alex Bligh (alex@cconcepts.co.uk)

33

* Idea by Alex Bligh (alex@cconcepts.co.uk)

34

*

34

*

35

* 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG

35

* 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG

36

* (Gerhard.Wichert@pdb.siemens.de)

36

* (Gerhard.Wichert@pdb.siemens.de)

37

*

37

*

38

* Aug/Sep 2004 Changed to four level page tables (Andi Kleen)

38

* Aug/Sep 2004 Changed to four level page tables (Andi Kleen)

39

*/

39

*/

40

41

#include <linux/kernel_stat.h>

41

#include <linux/kernel_stat.h>

42

#include <linux/mm.h>

42

#include <linux/mm.h>

43

#include <linux/hugetlb.h>

43

#include <linux/hugetlb.h>

44

#include <linux/mman.h>

44

#include <linux/mman.h>

45

#include <linux/swap.h>

45

#include <linux/swap.h>

46

#include <linux/highmem.h>

46

#include <linux/highmem.h>

47

#include <linux/pagemap.h>

47

#include <linux/pagemap.h>

48

#include <linux/ksm.h>

48

#include <linux/ksm.h>

49

#include <linux/rmap.h>

49

#include <linux/rmap.h>

50

#include <linux/export.h>

50

#include <linux/export.h>

51

#include <linux/delayacct.h>

51

#include <linux/delayacct.h>

52

#include <linux/init.h>

52

#include <linux/init.h>

53

#include <linux/writeback.h>

53

#include <linux/writeback.h>

54

#include <linux/memcontrol.h>

54

#include <linux/memcontrol.h>

55

#include <linux/mmu_notifier.h>

55

#include <linux/mmu_notifier.h>

56

#include <linux/kallsyms.h>

56

#include <linux/kallsyms.h>

57

#include <linux/swapops.h>

57

#include <linux/swapops.h>

58

#include <linux/elf.h>

58

#include <linux/elf.h>

59

#include <linux/gfp.h>

59

#include <linux/gfp.h>

60

#include <linux/migrate.h>

60

#include <linux/migrate.h>

61

#include <linux/string.h>

61

#include <linux/string.h>

62

63

#include <asm/io.h>

63

#include <asm/io.h>

64

#include <asm/pgalloc.h>

64

#include <asm/pgalloc.h>

65

#include <asm/uaccess.h>

65

#include <asm/uaccess.h>

66

#include <asm/tlb.h>

66

#include <asm/tlb.h>

67

#include <asm/tlbflush.h>

67

#include <asm/tlbflush.h>

68

#include <asm/pgtable.h>

68

#include <asm/pgtable.h>

69

70

#include "internal.h"

70

#include "internal.h"

71

72

#ifdef LAST_NID_NOT_IN_PAGE_FLAGS

72

#ifdef LAST_NID_NOT_IN_PAGE_FLAGS

73

#warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_nid.

73

#warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_nid.

74

#endif

74

#endif

75

76

#ifndef CONFIG_NEED_MULTIPLE_NODES

76

#ifndef CONFIG_NEED_MULTIPLE_NODES

77

/* use the per-pgdat data instead for discontigmem - mbligh */

77

/* use the per-pgdat data instead for discontigmem - mbligh */

78

unsigned long max_mapnr;

78

unsigned long max_mapnr;

79

struct page *mem_map;

79

struct page *mem_map;

80

81

EXPORT_SYMBOL(max_mapnr);

81

EXPORT_SYMBOL(max_mapnr);

82

EXPORT_SYMBOL(mem_map);

82

EXPORT_SYMBOL(mem_map);

83

#endif

83

#endif

84

85

/*

85

/*

86

* A number of key systems in x86 including ioremap() rely on the assumption

86

* A number of key systems in x86 including ioremap() rely on the assumption

87

* that high_memory defines the upper bound on direct map memory, then end

87

* that high_memory defines the upper bound on direct map memory, then end

88

* of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and

88

* of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and

89

* highstart_pfn must be the same; there must be no gap between ZONE_NORMAL

89

* highstart_pfn must be the same; there must be no gap between ZONE_NORMAL

90

* and ZONE_HIGHMEM.

90

* and ZONE_HIGHMEM.

91

*/

91

*/

92

void * high_memory;

92

void * high_memory;

93

94

EXPORT_SYMBOL(high_memory);

94

EXPORT_SYMBOL(high_memory);

95

96

/*

96

/*

97

* Randomize the address space (stacks, mmaps, brk, etc.).

97

* Randomize the address space (stacks, mmaps, brk, etc.).

98

*

98

*

99

* ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,

99

* ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,

100

* as ancient (libc5 based) binaries can segfault. )

100

* as ancient (libc5 based) binaries can segfault. )

101

*/

101

*/

102

int randomize_va_space __read_mostly =

102

int randomize_va_space __read_mostly =

103

#ifdef CONFIG_COMPAT_BRK

103

#ifdef CONFIG_COMPAT_BRK

104

1;

104

1;

105

#else

105

#else

106

2;

106

2;

107

#endif

107

#endif

108

109

static int __init disable_randmaps(char *s)

109

static int __init disable_randmaps(char *s)

110

{

110

{

111

randomize_va_space = 0;

111

randomize_va_space = 0;

112

return 1;

112

return 1;

113

}

113

}

114

__setup("norandmaps", disable_randmaps);

114

__setup("norandmaps", disable_randmaps);

115

116

unsigned long zero_pfn __read_mostly;

116

unsigned long zero_pfn __read_mostly;

117

unsigned long highest_memmap_pfn __read_mostly;

117

unsigned long highest_memmap_pfn __read_mostly;

118

119

/*

119

/*

120

* CONFIG_MMU architectures set up ZERO_PAGE in their paging_init()

120

* CONFIG_MMU architectures set up ZERO_PAGE in their paging_init()

121

*/

121

*/

122

static int __init init_zero_pfn(void)

122

static int __init init_zero_pfn(void)

123

{

123

{

124

zero_pfn = page_to_pfn(ZERO_PAGE(0));

124

zero_pfn = page_to_pfn(ZERO_PAGE(0));

125

return 0;

125

return 0;

126

}

126

}

127

core_initcall(init_zero_pfn);

127

core_initcall(init_zero_pfn);

128

129

130

#if defined(SPLIT_RSS_COUNTING)

130

#if defined(SPLIT_RSS_COUNTING)

131

132

void sync_mm_rss(struct mm_struct *mm)

132

void sync_mm_rss(struct mm_struct *mm)

133

{

133

{

134

int i;

134

int i;

135

136

for (i = 0; i < NR_MM_COUNTERS; i++) {

136

for (i = 0; i < NR_MM_COUNTERS; i++) {

137

if (current->rss_stat.count[i]) {

137

if (current->rss_stat.count[i]) {

138

add_mm_counter(mm, i, current->rss_stat.count[i]);

138

add_mm_counter(mm, i, current->rss_stat.count[i]);

139

current->rss_stat.count[i] = 0;

139

current->rss_stat.count[i] = 0;

140

}

140

}

141

}

141

}

142

current->rss_stat.events = 0;

142

current->rss_stat.events = 0;

143

}

143

}

144

145

static void add_mm_counter_fast(struct mm_struct *mm, int member, int val)

145

static void add_mm_counter_fast(struct mm_struct *mm, int member, int val)

146

{

146

{

147

struct task_struct *task = current;

147

struct task_struct *task = current;

148

149

if (likely(task->mm == mm))

149

if (likely(task->mm == mm))

150

task->rss_stat.count[member] += val;

150

task->rss_stat.count[member] += val;

151

else

151

else

152

add_mm_counter(mm, member, val);

152

add_mm_counter(mm, member, val);

153

}

153

}

154

#define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1)

154

#define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1)

155

#define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1)

155

#define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1)

156

157

/* sync counter once per 64 page faults */

157

/* sync counter once per 64 page faults */

158

#define TASK_RSS_EVENTS_THRESH (64)

158

#define TASK_RSS_EVENTS_THRESH (64)

159

static void check_sync_rss_stat(struct task_struct *task)

159

static void check_sync_rss_stat(struct task_struct *task)

160

{

160

{

161

if (unlikely(task != current))

161

if (unlikely(task != current))

162

return;

162

return;

163

if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH))

163

if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH))

164

sync_mm_rss(task->mm);

164

sync_mm_rss(task->mm);

165

}

165

}

166

#else /* SPLIT_RSS_COUNTING */

166

#else /* SPLIT_RSS_COUNTING */

167

168

#define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member)

168

#define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member)

169

#define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member)

169

#define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member)

170

171

static void check_sync_rss_stat(struct task_struct *task)

171

static void check_sync_rss_stat(struct task_struct *task)

172

{

172

{

173

}

173

}

174

175

#endif /* SPLIT_RSS_COUNTING */

175

#endif /* SPLIT_RSS_COUNTING */

176

177

#ifdef HAVE_GENERIC_MMU_GATHER

177

#ifdef HAVE_GENERIC_MMU_GATHER

178

179

static int tlb_next_batch(struct mmu_gather *tlb)

179

static int tlb_next_batch(struct mmu_gather *tlb)

180

{

180

{

181

struct mmu_gather_batch *batch;

181

struct mmu_gather_batch *batch;

182

183

batch = tlb->active;

183

batch = tlb->active;

184

if (batch->next) {

184

if (batch->next) {

185

tlb->active = batch->next;

185

tlb->active = batch->next;

186

return 1;

186

return 1;

187

}

187

}

188

189

if (tlb->batch_count == MAX_GATHER_BATCH_COUNT)

189

if (tlb->batch_count == MAX_GATHER_BATCH_COUNT)

190

return 0;

190

return 0;

191

192

batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0);

192

batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0);

193

if (!batch)

193

if (!batch)

194

return 0;

194

return 0;

195

196

tlb->batch_count++;

196

tlb->batch_count++;

197

batch->next = NULL;

197

batch->next = NULL;

198

batch->nr = 0;

198

batch->nr = 0;

199

batch->max = MAX_GATHER_BATCH;

199

batch->max = MAX_GATHER_BATCH;

200

201

tlb->active->next = batch;

201

tlb->active->next = batch;

202

tlb->active = batch;

202

tlb->active = batch;

203

204

return 1;

204

return 1;

205

}

205

}

206

207

/* tlb_gather_mmu

207

/* tlb_gather_mmu

208

* Called to initialize an (on-stack) mmu_gather structure for page-table

208

* Called to initialize an (on-stack) mmu_gather structure for page-table

209

* tear-down from @mm. The @fullmm argument is used when @mm is without

209

* tear-down from @mm. The @fullmm argument is used when @mm is without

210

* users and we're going to destroy the full address space (exit/execve).

210

* users and we're going to destroy the full address space (exit/execve).

211

*/

211

*/

212

void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, unsigned long start, unsigned long end)

212

void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, unsigned long start, unsigned long end)

213

{

213

{

214

tlb->mm = mm;

214

tlb->mm = mm;

215

216

/* Is it from 0 to ~0? */

216

/* Is it from 0 to ~0? */

217

tlb->fullmm = !(start | (end+1));

217

tlb->fullmm = !(start | (end+1));

218

tlb->need_flush_all = 0;

218

tlb->need_flush_all = 0;

219

tlb->start = start;

219

tlb->start = start;

220

tlb->end = end;

220

tlb->end = end;

221

tlb->need_flush = 0;

221

tlb->need_flush = 0;

222

tlb->local.next = NULL;

222

tlb->local.next = NULL;

223

tlb->local.nr = 0;

223

tlb->local.nr = 0;

224

tlb->local.max = ARRAY_SIZE(tlb->__pages);

224

tlb->local.max = ARRAY_SIZE(tlb->__pages);

225

tlb->active = &tlb->local;

225

tlb->active = &tlb->local;

226

tlb->batch_count = 0;

226

tlb->batch_count = 0;

227

228

#ifdef CONFIG_HAVE_RCU_TABLE_FREE

228

#ifdef CONFIG_HAVE_RCU_TABLE_FREE

229

tlb->batch = NULL;

229

tlb->batch = NULL;

230

#endif

230

#endif

231

}

231

}

232

233

void tlb_flush_mmu(struct mmu_gather *tlb)

233

void tlb_flush_mmu(struct mmu_gather *tlb)

234

{

234

{

235

struct mmu_gather_batch *batch;

235

struct mmu_gather_batch *batch;

236

237

if (!tlb->need_flush)

237

if (!tlb->need_flush)

238

return;

238

return;

239

tlb->need_flush = 0;

239

tlb->need_flush = 0;

240

tlb_flush(tlb);

240

tlb_flush(tlb);

241

#ifdef CONFIG_HAVE_RCU_TABLE_FREE

241

#ifdef CONFIG_HAVE_RCU_TABLE_FREE

242

tlb_table_flush(tlb);

242

tlb_table_flush(tlb);

243

#endif

243

#endif

244

245

for (batch = &tlb->local; batch; batch = batch->next) {

245

for (batch = &tlb->local; batch; batch = batch->next) {

246

free_pages_and_swap_cache(batch->pages, batch->nr);

246

free_pages_and_swap_cache(batch->pages, batch->nr);

247

batch->nr = 0;

247

batch->nr = 0;

248

}

248

}

249

tlb->active = &tlb->local;

249

tlb->active = &tlb->local;

250

}

250

}

251

252

/* tlb_finish_mmu

252

/* tlb_finish_mmu

253

* Called at the end of the shootdown operation to free up any resources

253

* Called at the end of the shootdown operation to free up any resources

254

* that were required.

254

* that were required.

255

*/

255

*/

256

void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end)

256

void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end)

257

{

257

{

258

struct mmu_gather_batch *batch, *next;

258

struct mmu_gather_batch *batch, *next;

259

260

tlb_flush_mmu(tlb);

260

tlb_flush_mmu(tlb);

261

262

/* keep the page table cache within bounds */

262

/* keep the page table cache within bounds */

263

check_pgt_cache();

263

check_pgt_cache();

264

265

for (batch = tlb->local.next; batch; batch = next) {

265

for (batch = tlb->local.next; batch; batch = next) {

266

next = batch->next;

266

next = batch->next;

267

free_pages((unsigned long)batch, 0);

267

free_pages((unsigned long)batch, 0);

268

}

268

}

269

tlb->local.next = NULL;

269

tlb->local.next = NULL;

270

}

270

}

271

272

/* __tlb_remove_page

272

/* __tlb_remove_page

273

* Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while

273

* Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while

274

* handling the additional races in SMP caused by other CPUs caching valid

274

* handling the additional races in SMP caused by other CPUs caching valid

275

* mappings in their TLBs. Returns the number of free page slots left.

275

* mappings in their TLBs. Returns the number of free page slots left.

276

* When out of page slots we must call tlb_flush_mmu().

276

* When out of page slots we must call tlb_flush_mmu().

277

*/

277

*/

278

int __tlb_remove_page(struct mmu_gather *tlb, struct page *page)

278

int __tlb_remove_page(struct mmu_gather *tlb, struct page *page)

279

{

279

{

280

struct mmu_gather_batch *batch;

280

struct mmu_gather_batch *batch;

281

282

VM_BUG_ON(!tlb->need_flush);

282

VM_BUG_ON(!tlb->need_flush);

283

284

batch = tlb->active;

284

batch = tlb->active;

285

batch->pages[batch->nr++] = page;

285

batch->pages[batch->nr++] = page;

286

if (batch->nr == batch->max) {

286

if (batch->nr == batch->max) {

287

if (!tlb_next_batch(tlb))

287

if (!tlb_next_batch(tlb))

288

return 0;

288

return 0;

289

batch = tlb->active;

289

batch = tlb->active;

290

}

290

}

291

VM_BUG_ON(batch->nr > batch->max);

291

VM_BUG_ON(batch->nr > batch->max);

292

293

return batch->max - batch->nr;

293

return batch->max - batch->nr;

294

}

294

}

295

296

#endif /* HAVE_GENERIC_MMU_GATHER */

296

#endif /* HAVE_GENERIC_MMU_GATHER */

297

298

#ifdef CONFIG_HAVE_RCU_TABLE_FREE

298

#ifdef CONFIG_HAVE_RCU_TABLE_FREE

299

300

/*

300

/*

301

* See the comment near struct mmu_table_batch.

301

* See the comment near struct mmu_table_batch.

302

*/

302

*/

303

304

static void tlb_remove_table_smp_sync(void *arg)

304

static void tlb_remove_table_smp_sync(void *arg)

305

{

305

{

306

/* Simply deliver the interrupt */

306

/* Simply deliver the interrupt */

307

}

307

}

308

309

static void tlb_remove_table_one(void *table)

309

static void tlb_remove_table_one(void *table)

310

{

310

{

311

/*

311

/*

312

* This isn't an RCU grace period and hence the page-tables cannot be

312

* This isn't an RCU grace period and hence the page-tables cannot be

313

* assumed to be actually RCU-freed.

313

* assumed to be actually RCU-freed.

314

*

314

*

315

* It is however sufficient for software page-table walkers that rely on

315

* It is however sufficient for software page-table walkers that rely on

316

* IRQ disabling. See the comment near struct mmu_table_batch.

316

* IRQ disabling. See the comment near struct mmu_table_batch.

317

*/

317

*/

318

smp_call_function(tlb_remove_table_smp_sync, NULL, 1);

318

smp_call_function(tlb_remove_table_smp_sync, NULL, 1);

319

__tlb_remove_table(table);

319

__tlb_remove_table(table);

320

}

320

}

321

322

static void tlb_remove_table_rcu(struct rcu_head *head)

322

static void tlb_remove_table_rcu(struct rcu_head *head)

323

{

323

{

324

struct mmu_table_batch *batch;

324

struct mmu_table_batch *batch;

325

int i;

325

int i;

326

327

batch = container_of(head, struct mmu_table_batch, rcu);

327

batch = container_of(head, struct mmu_table_batch, rcu);

328

329

for (i = 0; i < batch->nr; i++)

329

for (i = 0; i < batch->nr; i++)

330

__tlb_remove_table(batch->tables[i]);

330

__tlb_remove_table(batch->tables[i]);

331

332

free_page((unsigned long)batch);

332

free_page((unsigned long)batch);

333

}

333

}

334

335

void tlb_table_flush(struct mmu_gather *tlb)

335

void tlb_table_flush(struct mmu_gather *tlb)

336

{

336

{

337

struct mmu_table_batch **batch = &tlb->batch;

337

struct mmu_table_batch **batch = &tlb->batch;

338

339

if (*batch) {

339

if (*batch) {

340

call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu);

340

call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu);

341

*batch = NULL;

341

*batch = NULL;

342

}

342

}

343

}

343

}

344

345

void tlb_remove_table(struct mmu_gather *tlb, void *table)

345

void tlb_remove_table(struct mmu_gather *tlb, void *table)

346

{

346

{

347

struct mmu_table_batch **batch = &tlb->batch;

347

struct mmu_table_batch **batch = &tlb->batch;

348

349

tlb->need_flush = 1;

349

tlb->need_flush = 1;

350

351

/*

351

/*

352

* When there's less then two users of this mm there cannot be a

352

* When there's less then two users of this mm there cannot be a

353

* concurrent page-table walk.

353

* concurrent page-table walk.

354

*/

354

*/

355

if (atomic_read(&tlb->mm->mm_users) < 2) {

355

if (atomic_read(&tlb->mm->mm_users) < 2) {

356

__tlb_remove_table(table);

356

__tlb_remove_table(table);

357

return;

357

return;

358

}

358

}

359

360

if (*batch == NULL) {

360

if (*batch == NULL) {

361

*batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN);

361

*batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN);

362

if (*batch == NULL) {

362

if (*batch == NULL) {

363

tlb_remove_table_one(table);

363

tlb_remove_table_one(table);

364

return;

364

return;

365

}

365

}

366

(*batch)->nr = 0;

366

(*batch)->nr = 0;

367

}

367

}

368

(*batch)->tables[(*batch)->nr++] = table;

368

(*batch)->tables[(*batch)->nr++] = table;

369

if ((*batch)->nr == MAX_TABLE_BATCH)

369

if ((*batch)->nr == MAX_TABLE_BATCH)

370

tlb_table_flush(tlb);

370

tlb_table_flush(tlb);

371

}

371

}

372

373

#endif /* CONFIG_HAVE_RCU_TABLE_FREE */

373

#endif /* CONFIG_HAVE_RCU_TABLE_FREE */

374

375

/*

375

/*

376

* Note: this doesn't free the actual pages themselves. That

376

* Note: this doesn't free the actual pages themselves. That

377

* has been handled earlier when unmapping all the memory regions.

377

* has been handled earlier when unmapping all the memory regions.

378

*/

378

*/

379

static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,

379

static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,

380

unsigned long addr)

380

unsigned long addr)

381

{

381

{

382

pgtable_t token = pmd_pgtable(*pmd);

382

pgtable_t token = pmd_pgtable(*pmd);

383

pmd_clear(pmd);

383

pmd_clear(pmd);

384

pte_free_tlb(tlb, token, addr);

384

pte_free_tlb(tlb, token, addr);

385

tlb->mm->nr_ptes--;

385

tlb->mm->nr_ptes--;

386

}

386

}

387

388

static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,

388

static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,

389

unsigned long addr, unsigned long end,

389

unsigned long addr, unsigned long end,

390

unsigned long floor, unsigned long ceiling)

390

unsigned long floor, unsigned long ceiling)

391

{

391

{

392

pmd_t *pmd;

392

pmd_t *pmd;

393

unsigned long next;

393

unsigned long next;

394

unsigned long start;

394

unsigned long start;

395

396

start = addr;

396

start = addr;

397

pmd = pmd_offset(pud, addr);

397

pmd = pmd_offset(pud, addr);

398

do {

398

do {

399

next = pmd_addr_end(addr, end);

399

next = pmd_addr_end(addr, end);

400

if (pmd_none_or_clear_bad(pmd))

400

if (pmd_none_or_clear_bad(pmd))

401

continue;

401

continue;

402

free_pte_range(tlb, pmd, addr);

402

free_pte_range(tlb, pmd, addr);

403

} while (pmd++, addr = next, addr != end);

403

} while (pmd++, addr = next, addr != end);

404

405

start &= PUD_MASK;

405

start &= PUD_MASK;

406

if (start < floor)

406

if (start < floor)

407

return;

407

return;

408

if (ceiling) {

408

if (ceiling) {

409

ceiling &= PUD_MASK;

409

ceiling &= PUD_MASK;

410

if (!ceiling)

410

if (!ceiling)

411

return;

411

return;

412

}

412

}

413

if (end - 1 > ceiling - 1)

413

if (end - 1 > ceiling - 1)

414

return;

414

return;

415

416

pmd = pmd_offset(pud, start);

416

pmd = pmd_offset(pud, start);

417

pud_clear(pud);

417

pud_clear(pud);

418

pmd_free_tlb(tlb, pmd, start);

418

pmd_free_tlb(tlb, pmd, start);

419

}

419

}

420

421

static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,

421

static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,

422

unsigned long addr, unsigned long end,

422

unsigned long addr, unsigned long end,

423

unsigned long floor, unsigned long ceiling)

423

unsigned long floor, unsigned long ceiling)

424

{

424

{

425

pud_t *pud;

425

pud_t *pud;

426

unsigned long next;

426

unsigned long next;

427

unsigned long start;

427

unsigned long start;

428

429

start = addr;

429

start = addr;

430

pud = pud_offset(pgd, addr);

430

pud = pud_offset(pgd, addr);

431

do {

431

do {

432

next = pud_addr_end(addr, end);

432

next = pud_addr_end(addr, end);

433

if (pud_none_or_clear_bad(pud))

433

if (pud_none_or_clear_bad(pud))

434

continue;

434

continue;

435

free_pmd_range(tlb, pud, addr, next, floor, ceiling);

435

free_pmd_range(tlb, pud, addr, next, floor, ceiling);

436

} while (pud++, addr = next, addr != end);

436

} while (pud++, addr = next, addr != end);

437

438

start &= PGDIR_MASK;

438

start &= PGDIR_MASK;

439

if (start < floor)

439

if (start < floor)

440

return;

440

return;

441

if (ceiling) {

441

if (ceiling) {

442

ceiling &= PGDIR_MASK;

442

ceiling &= PGDIR_MASK;

443

if (!ceiling)

443

if (!ceiling)

444

return;

444

return;

445

}

445

}

446

if (end - 1 > ceiling - 1)

446

if (end - 1 > ceiling - 1)

447

return;

447

return;

448

449

pud = pud_offset(pgd, start);

449

pud = pud_offset(pgd, start);

450

pgd_clear(pgd);

450

pgd_clear(pgd);

451

pud_free_tlb(tlb, pud, start);

451

pud_free_tlb(tlb, pud, start);

452

}

452

}

453

454

/*

454

/*

455

* This function frees user-level page tables of a process.

455

* This function frees user-level page tables of a process.

456

*

456

*

457

* Must be called with pagetable lock held.

457

* Must be called with pagetable lock held.

458

*/

458

*/

459

void free_pgd_range(struct mmu_gather *tlb,

459

void free_pgd_range(struct mmu_gather *tlb,

460

unsigned long addr, unsigned long end,

460

unsigned long addr, unsigned long end,

461

unsigned long floor, unsigned long ceiling)

461

unsigned long floor, unsigned long ceiling)

462

{

462

{

463

pgd_t *pgd;

463

pgd_t *pgd;

464

unsigned long next;

464

unsigned long next;

465

466

/*

466

/*

467

* The next few lines have given us lots of grief...

467

* The next few lines have given us lots of grief...

468

*

468

*

469

* Why are we testing PMD* at this top level? Because often

469

* Why are we testing PMD* at this top level? Because often

470

* there will be no work to do at all, and we'd prefer not to

470

* there will be no work to do at all, and we'd prefer not to

471

* go all the way down to the bottom just to discover that.

471

* go all the way down to the bottom just to discover that.

472

*

472

*

473

* Why all these "- 1"s? Because 0 represents both the bottom

473

* Why all these "- 1"s? Because 0 represents both the bottom

474

* of the address space and the top of it (using -1 for the

474

* of the address space and the top of it (using -1 for the

475

* top wouldn't help much: the masks would do the wrong thing).

475

* top wouldn't help much: the masks would do the wrong thing).

476

* The rule is that addr 0 and floor 0 refer to the bottom of

476

* The rule is that addr 0 and floor 0 refer to the bottom of

477

* the address space, but end 0 and ceiling 0 refer to the top

477

* the address space, but end 0 and ceiling 0 refer to the top

478

* Comparisons need to use "end - 1" and "ceiling - 1" (though

478

* Comparisons need to use "end - 1" and "ceiling - 1" (though

479

* that end 0 case should be mythical).

479

* that end 0 case should be mythical).

480

*

480

*

481

* Wherever addr is brought up or ceiling brought down, we must

481

* Wherever addr is brought up or ceiling brought down, we must

482

* be careful to reject "the opposite 0" before it confuses the

482

* be careful to reject "the opposite 0" before it confuses the

483

* subsequent tests. But what about where end is brought down

483

* subsequent tests. But what about where end is brought down

484

* by PMD_SIZE below? no, end can't go down to 0 there.

484

* by PMD_SIZE below? no, end can't go down to 0 there.

485

*

485

*

486

* Whereas we round start (addr) and ceiling down, by different

486

* Whereas we round start (addr) and ceiling down, by different

487

* masks at different levels, in order to test whether a table

487

* masks at different levels, in order to test whether a table

488

* now has no other vmas using it, so can be freed, we don't

488

* now has no other vmas using it, so can be freed, we don't

489

* bother to round floor or end up - the tests don't need that.

489

* bother to round floor or end up - the tests don't need that.

490

*/

490

*/

491

492

addr &= PMD_MASK;

492

addr &= PMD_MASK;

493

if (addr < floor) {

493

if (addr < floor) {

494

addr += PMD_SIZE;

494

addr += PMD_SIZE;

495

if (!addr)

495

if (!addr)

496

return;

496

return;

497

}

497

}

498

if (ceiling) {

498

if (ceiling) {

499

ceiling &= PMD_MASK;

499

ceiling &= PMD_MASK;

500

if (!ceiling)

500

if (!ceiling)

501

return;

501

return;

502

}

502

}

503

if (end - 1 > ceiling - 1)

503

if (end - 1 > ceiling - 1)

504

end -= PMD_SIZE;

504

end -= PMD_SIZE;

505

if (addr > end - 1)

505

if (addr > end - 1)

506

return;

506

return;

507

508

pgd = pgd_offset(tlb->mm, addr);

508

pgd = pgd_offset(tlb->mm, addr);

509

do {

509

do {

510

next = pgd_addr_end(addr, end);

510

next = pgd_addr_end(addr, end);

511

if (pgd_none_or_clear_bad(pgd))

511

if (pgd_none_or_clear_bad(pgd))

512

continue;

512

continue;

513

free_pud_range(tlb, pgd, addr, next, floor, ceiling);

513

free_pud_range(tlb, pgd, addr, next, floor, ceiling);

514

} while (pgd++, addr = next, addr != end);

514

} while (pgd++, addr = next, addr != end);

515

}

515

}

516

517

void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,

517

void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,

518

unsigned long floor, unsigned long ceiling)

518

unsigned long floor, unsigned long ceiling)

519

{

519

{

520

while (vma) {

520

while (vma) {

521

struct vm_area_struct *next = vma->vm_next;

521

struct vm_area_struct *next = vma->vm_next;

522

unsigned long addr = vma->vm_start;

522

unsigned long addr = vma->vm_start;

523

524

/*

524

/*

525

* Hide vma from rmap and truncate_pagecache before freeing

525

* Hide vma from rmap and truncate_pagecache before freeing

526

* pgtables

526

* pgtables

527

*/

527

*/

528

unlink_anon_vmas(vma);

528

unlink_anon_vmas(vma);

529

unlink_file_vma(vma);

529

unlink_file_vma(vma);

530

531

if (is_vm_hugetlb_page(vma)) {

531

if (is_vm_hugetlb_page(vma)) {

532

hugetlb_free_pgd_range(tlb, addr, vma->vm_end,

532

hugetlb_free_pgd_range(tlb, addr, vma->vm_end,

533

floor, next? next->vm_start: ceiling);

533

floor, next? next->vm_start: ceiling);

534

} else {

534

} else {

535

/*

535

/*

536

* Optimization: gather nearby vmas into one call down

536

* Optimization: gather nearby vmas into one call down

537

*/

537

*/

538

while (next && next->vm_start <= vma->vm_end + PMD_SIZE

538

while (next && next->vm_start <= vma->vm_end + PMD_SIZE

539

&& !is_vm_hugetlb_page(next)) {

539

&& !is_vm_hugetlb_page(next)) {

540

vma = next;

540

vma = next;

541

next = vma->vm_next;

541

next = vma->vm_next;

542

unlink_anon_vmas(vma);

542

unlink_anon_vmas(vma);

543

unlink_file_vma(vma);

543

unlink_file_vma(vma);

544

}

544

}

545

free_pgd_range(tlb, addr, vma->vm_end,

545

free_pgd_range(tlb, addr, vma->vm_end,

546

floor, next? next->vm_start: ceiling);

546

floor, next? next->vm_start: ceiling);

547

}

547

}

548

vma = next;

548

vma = next;

549

}

549

}

550

}

550

}

551

552

int __pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,

552

int __pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,

553

pmd_t *pmd, unsigned long address)

553

pmd_t *pmd, unsigned long address)

554

{

554

{

555

pgtable_t new = pte_alloc_one(mm, address);

555

pgtable_t new = pte_alloc_one(mm, address);

556

int wait_split_huge_page;

556

int wait_split_huge_page;

557

if (!new)

557

if (!new)

558

return -ENOMEM;

558

return -ENOMEM;

559

560

/*

560

/*

561

* Ensure all pte setup (eg. pte page lock and page clearing) are

561

* Ensure all pte setup (eg. pte page lock and page clearing) are

562

* visible before the pte is made visible to other CPUs by being

562

* visible before the pte is made visible to other CPUs by being

563

* put into page tables.

563

* put into page tables.

564

*

564

*

565

* The other side of the story is the pointer chasing in the page

565

* The other side of the story is the pointer chasing in the page

566

* table walking code (when walking the page table without locking;

566

* table walking code (when walking the page table without locking;

567

* ie. most of the time). Fortunately, these data accesses consist

567

* ie. most of the time). Fortunately, these data accesses consist

568

* of a chain of data-dependent loads, meaning most CPUs (alpha

568

* of a chain of data-dependent loads, meaning most CPUs (alpha

569

* being the notable exception) will already guarantee loads are

569

* being the notable exception) will already guarantee loads are

570

* seen in-order. See the alpha page table accessors for the

570

* seen in-order. See the alpha page table accessors for the

571

* smp_read_barrier_depends() barriers in page table walking code.

571

* smp_read_barrier_depends() barriers in page table walking code.

572

*/

572

*/

573

smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */

573

smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */

574

575

spin_lock(&mm->page_table_lock);

575

spin_lock(&mm->page_table_lock);

576

wait_split_huge_page = 0;

576

wait_split_huge_page = 0;

577

if (likely(pmd_none(*pmd))) { /* Has another populated it ? */

577

if (likely(pmd_none(*pmd))) { /* Has another populated it ? */

578

mm->nr_ptes++;

578

mm->nr_ptes++;

579

pmd_populate(mm, pmd, new);

579

pmd_populate(mm, pmd, new);

580

new = NULL;

580

new = NULL;

581

} else if (unlikely(pmd_trans_splitting(*pmd)))

581

} else if (unlikely(pmd_trans_splitting(*pmd)))

582

wait_split_huge_page = 1;

582

wait_split_huge_page = 1;

583

spin_unlock(&mm->page_table_lock);

583

spin_unlock(&mm->page_table_lock);

584

if (new)

584

if (new)

585

pte_free(mm, new);

585

pte_free(mm, new);

586

if (wait_split_huge_page)

586

if (wait_split_huge_page)

587

wait_split_huge_page(vma->anon_vma, pmd);

587

wait_split_huge_page(vma->anon_vma, pmd);

588

return 0;

588

return 0;

589

}

589

}

590

591

int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)

591

int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)

592

{

592

{

593

pte_t *new = pte_alloc_one_kernel(&init_mm, address);

593

pte_t *new = pte_alloc_one_kernel(&init_mm, address);

594

if (!new)

594

if (!new)

595

return -ENOMEM;

595

return -ENOMEM;

596

597

smp_wmb(); /* See comment in __pte_alloc */

597

smp_wmb(); /* See comment in __pte_alloc */

598

599

spin_lock(&init_mm.page_table_lock);

599

spin_lock(&init_mm.page_table_lock);

600

if (likely(pmd_none(*pmd))) { /* Has another populated it ? */

600

if (likely(pmd_none(*pmd))) { /* Has another populated it ? */

601

pmd_populate_kernel(&init_mm, pmd, new);

601

pmd_populate_kernel(&init_mm, pmd, new);

602

new = NULL;

602

new = NULL;

603

} else

603

} else

604

VM_BUG_ON(pmd_trans_splitting(*pmd));

604

VM_BUG_ON(pmd_trans_splitting(*pmd));

605

spin_unlock(&init_mm.page_table_lock);

605

spin_unlock(&init_mm.page_table_lock);

606

if (new)

606

if (new)

607

pte_free_kernel(&init_mm, new);

607

pte_free_kernel(&init_mm, new);

608

return 0;

608

return 0;

609

}

609

}

610

611

static inline void init_rss_vec(int *rss)

611

static inline void init_rss_vec(int *rss)

612

{

612

{

613

memset(rss, 0, sizeof(int) * NR_MM_COUNTERS);

613

memset(rss, 0, sizeof(int) * NR_MM_COUNTERS);

614

}

614

}

615

616

static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss)

616

static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss)

617

{

617

{

618

int i;

618

int i;

619

620

if (current->mm == mm)

620

if (current->mm == mm)

621

sync_mm_rss(mm);

621

sync_mm_rss(mm);

622

for (i = 0; i < NR_MM_COUNTERS; i++)

622

for (i = 0; i < NR_MM_COUNTERS; i++)

623

if (rss[i])

623

if (rss[i])

624

add_mm_counter(mm, i, rss[i]);

624

add_mm_counter(mm, i, rss[i]);

625

}

625

}

626

627

/*

627

/*

628

* This function is called to print an error when a bad pte

628

* This function is called to print an error when a bad pte

629

* is found. For example, we might have a PFN-mapped pte in

629

* is found. For example, we might have a PFN-mapped pte in

630

* a region that doesn't allow it.

630

* a region that doesn't allow it.

631

*

631

*

632

* The calling function must still handle the error.

632

* The calling function must still handle the error.

633

*/

633

*/

634

static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,

634

static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,

635

pte_t pte, struct page *page)

635

pte_t pte, struct page *page)

636

{

636

{

637

pgd_t *pgd = pgd_offset(vma->vm_mm, addr);

637

pgd_t *pgd = pgd_offset(vma->vm_mm, addr);

638

pud_t *pud = pud_offset(pgd, addr);

638

pud_t *pud = pud_offset(pgd, addr);

639

pmd_t *pmd = pmd_offset(pud, addr);

639

pmd_t *pmd = pmd_offset(pud, addr);

640

struct address_space *mapping;

640

struct address_space *mapping;

641

pgoff_t index;

641

pgoff_t index;

642

static unsigned long resume;

642

static unsigned long resume;

643

static unsigned long nr_shown;

643

static unsigned long nr_shown;

644

static unsigned long nr_unshown;

644

static unsigned long nr_unshown;

645

646

/*

646

/*

647

* Allow a burst of 60 reports, then keep quiet for that minute;

647

* Allow a burst of 60 reports, then keep quiet for that minute;

648

* or allow a steady drip of one report per second.

648

* or allow a steady drip of one report per second.

649

*/

649

*/

650

if (nr_shown == 60) {

650

if (nr_shown == 60) {

651

if (time_before(jiffies, resume)) {

651

if (time_before(jiffies, resume)) {

652

nr_unshown++;

652

nr_unshown++;

653

return;

653

return;

654

}

654

}

655

if (nr_unshown) {

655

if (nr_unshown) {

656

printk(KERN_ALERT

656

printk(KERN_ALERT

657

"BUG: Bad page map: %lu messages suppressed\n",

657

"BUG: Bad page map: %lu messages suppressed\n",

658

nr_unshown);

658

nr_unshown);

659

nr_unshown = 0;

659

nr_unshown = 0;

660

}

660

}

661

nr_shown = 0;

661

nr_shown = 0;

662

}

662

}

663

if (nr_shown++ == 0)

663

if (nr_shown++ == 0)

664

resume = jiffies + 60 * HZ;

664

resume = jiffies + 60 * HZ;

665

666

mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;

666

mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;

667

index = linear_page_index(vma, addr);

667

index = linear_page_index(vma, addr);

668

669

printk(KERN_ALERT

669

printk(KERN_ALERT

670

"BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",

670

"BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",

671

current->comm,

671

current->comm,

672

(long long)pte_val(pte), (long long)pmd_val(*pmd));

672

(long long)pte_val(pte), (long long)pmd_val(*pmd));

673

if (page)

673

if (page)

674

dump_page(page);

674

dump_page(page);

675

printk(KERN_ALERT

675

printk(KERN_ALERT

676

"addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",

676

"addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",

677

(void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);

677

(void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);

678

/*

678

/*

679

* Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y

679

* Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y

680

*/

680

*/

681

if (vma->vm_ops)

681

if (vma->vm_ops)

682

printk(KERN_ALERT "vma->vm_ops->fault: %pSR\n",

682

printk(KERN_ALERT "vma->vm_ops->fault: %pSR\n",

683

vma->vm_ops->fault);

683

vma->vm_ops->fault);

684

if (vma->vm_file && vma->vm_file->f_op)

684

if (vma->vm_file && vma->vm_file->f_op)

685

printk(KERN_ALERT "vma->vm_file->f_op->mmap: %pSR\n",

685

printk(KERN_ALERT "vma->vm_file->f_op->mmap: %pSR\n",

686

vma->vm_file->f_op->mmap);

686

vma->vm_file->f_op->mmap);

687

dump_stack();

687

dump_stack();

688

add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);

688

add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);

689

}

689

}

690

691

static inline bool is_cow_mapping(vm_flags_t flags)

691

static inline bool is_cow_mapping(vm_flags_t flags)

692

{

692

{

693

return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;

693

return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;

694

}

694

}

695

696

/*

696

/*

697

* vm_normal_page -- This function gets the "struct page" associated with a pte.

697

* vm_normal_page -- This function gets the "struct page" associated with a pte.

698

*

698

*

699

* "Special" mappings do not wish to be associated with a "struct page" (either

699

* "Special" mappings do not wish to be associated with a "struct page" (either

700

* it doesn't exist, or it exists but they don't want to touch it). In this

700

* it doesn't exist, or it exists but they don't want to touch it). In this

701

* case, NULL is returned here. "Normal" mappings do have a struct page.

701

* case, NULL is returned here. "Normal" mappings do have a struct page.

702

*

702

*

703

* There are 2 broad cases. Firstly, an architecture may define a pte_special()

703

* There are 2 broad cases. Firstly, an architecture may define a pte_special()

704

* pte bit, in which case this function is trivial. Secondly, an architecture

704

* pte bit, in which case this function is trivial. Secondly, an architecture

705

* may not have a spare pte bit, which requires a more complicated scheme,

705

* may not have a spare pte bit, which requires a more complicated scheme,

706

* described below.

706

* described below.

707

*

707

*

708

* A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a

708

* A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a

709

* special mapping (even if there are underlying and valid "struct pages").

709

* special mapping (even if there are underlying and valid "struct pages").

710

* COWed pages of a VM_PFNMAP are always normal.

710

* COWed pages of a VM_PFNMAP are always normal.

711

*

711

*

712

* The way we recognize COWed pages within VM_PFNMAP mappings is through the

712

* The way we recognize COWed pages within VM_PFNMAP mappings is through the

713

* rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit

713

* rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit

714

* set, and the vm_pgoff will point to the first PFN mapped: thus every special

714

* set, and the vm_pgoff will point to the first PFN mapped: thus every special

715

* mapping will always honor the rule

715

* mapping will always honor the rule

716

*

716

*

717

* pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)

717

* pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)

718

*

718

*

719

* And for normal mappings this is false.

719

* And for normal mappings this is false.

720

*

720

*

721

* This restricts such mappings to be a linear translation from virtual address

721

* This restricts such mappings to be a linear translation from virtual address

722

* to pfn. To get around this restriction, we allow arbitrary mappings so long

722

* to pfn. To get around this restriction, we allow arbitrary mappings so long

723

* as the vma is not a COW mapping; in that case, we know that all ptes are

723

* as the vma is not a COW mapping; in that case, we know that all ptes are

724

* special (because none can have been COWed).

724

* special (because none can have been COWed).

725

*

725

*

726

*

726

*

727

* In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.

727

* In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.

728

*

728

*

729

* VM_MIXEDMAP mappings can likewise contain memory with or without "struct

729

* VM_MIXEDMAP mappings can likewise contain memory with or without "struct

730

* page" backing, however the difference is that _all_ pages with a struct

730

* page" backing, however the difference is that _all_ pages with a struct

731

* page (that is, those where pfn_valid is true) are refcounted and considered

731

* page (that is, those where pfn_valid is true) are refcounted and considered

732

* normal pages by the VM. The disadvantage is that pages are refcounted

732

* normal pages by the VM. The disadvantage is that pages are refcounted

733

* (which can be slower and simply not an option for some PFNMAP users). The

733

* (which can be slower and simply not an option for some PFNMAP users). The

734

* advantage is that we don't have to follow the strict linearity rule of

734

* advantage is that we don't have to follow the strict linearity rule of

735

* PFNMAP mappings in order to support COWable mappings.

735

* PFNMAP mappings in order to support COWable mappings.

736

*

736

*

737

*/

737

*/

738

#ifdef __HAVE_ARCH_PTE_SPECIAL

738

#ifdef __HAVE_ARCH_PTE_SPECIAL

739

# define HAVE_PTE_SPECIAL 1

739

# define HAVE_PTE_SPECIAL 1

740

#else

740

#else

741

# define HAVE_PTE_SPECIAL 0

741

# define HAVE_PTE_SPECIAL 0

742

#endif

742

#endif

743

struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,

743

struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,

744

pte_t pte)

744

pte_t pte)

745

{

745

{

746

unsigned long pfn = pte_pfn(pte);

746

unsigned long pfn = pte_pfn(pte);

747

748

if (HAVE_PTE_SPECIAL) {

748

if (HAVE_PTE_SPECIAL) {

749

if (likely(!pte_special(pte)))

749

if (likely(!pte_special(pte)))

750

goto check_pfn;

750

goto check_pfn;

751

if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))

751

if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))

752

return NULL;

752

return NULL;

753

if (!is_zero_pfn(pfn))

753

if (!is_zero_pfn(pfn))

754

print_bad_pte(vma, addr, pte, NULL);

754

print_bad_pte(vma, addr, pte, NULL);

755

return NULL;

755

return NULL;

756

}

756

}

757

758

/* !HAVE_PTE_SPECIAL case follows: */

758

/* !HAVE_PTE_SPECIAL case follows: */

759

760

if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {

760

if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {

761

if (vma->vm_flags & VM_MIXEDMAP) {

761

if (vma->vm_flags & VM_MIXEDMAP) {

762

if (!pfn_valid(pfn))

762

if (!pfn_valid(pfn))

763

return NULL;

763

return NULL;

764

goto out;

764

goto out;

765

} else {

765

} else {

766

unsigned long off;

766

unsigned long off;

767

off = (addr - vma->vm_start) >> PAGE_SHIFT;

767

off = (addr - vma->vm_start) >> PAGE_SHIFT;

768

if (pfn == vma->vm_pgoff + off)

768

if (pfn == vma->vm_pgoff + off)

769

return NULL;

769

return NULL;

770

if (!is_cow_mapping(vma->vm_flags))

770

if (!is_cow_mapping(vma->vm_flags))

771

return NULL;

771

return NULL;

772

}

772

}

773

}

773

}

774

775

if (is_zero_pfn(pfn))

775

if (is_zero_pfn(pfn))

776

return NULL;

776

return NULL;

777

check_pfn:

777

check_pfn:

778

if (unlikely(pfn > highest_memmap_pfn)) {

778

if (unlikely(pfn > highest_memmap_pfn)) {

779

print_bad_pte(vma, addr, pte, NULL);

779

print_bad_pte(vma, addr, pte, NULL);

780

return NULL;

780

return NULL;

781

}

781

}

782

783

/*

783

/*

784

* NOTE! We still have PageReserved() pages in the page tables.

784

* NOTE! We still have PageReserved() pages in the page tables.

785

* eg. VDSO mappings can cause them to exist.

785

* eg. VDSO mappings can cause them to exist.

786

*/

786

*/

787

out:

787

out:

788

return pfn_to_page(pfn);

788

return pfn_to_page(pfn);

789

}

789

}

790

791

/*

791

/*

792

* copy one vm_area from one task to the other. Assumes the page tables

792

* copy one vm_area from one task to the other. Assumes the page tables

793

* already present in the new task to be cleared in the whole range

793

* already present in the new task to be cleared in the whole range

794

* covered by this vma.

794

* covered by this vma.

795

*/

795

*/

796

797

static inline unsigned long

797

static inline unsigned long

798

copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,

798

copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,

799

pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,

799

pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,

800

unsigned long addr, int *rss)

800

unsigned long addr, int *rss)

801

{

801

{

802

unsigned long vm_flags = vma->vm_flags;

802

unsigned long vm_flags = vma->vm_flags;

803

pte_t pte = *src_pte;

803

pte_t pte = *src_pte;

804

struct page *page;

804

struct page *page;

805

806

/* pte contains position in swap or file, so copy. */

806

/* pte contains position in swap or file, so copy. */

807

if (unlikely(!pte_present(pte))) {

807

if (unlikely(!pte_present(pte))) {

808

if (!pte_file(pte)) {

808

if (!pte_file(pte)) {

809

swp_entry_t entry = pte_to_swp_entry(pte);

809

swp_entry_t entry = pte_to_swp_entry(pte);

810

811

if (swap_duplicate(entry) < 0)

811

if (swap_duplicate(entry) < 0)

812

return entry.val;

812

return entry.val;

813

814

/* make sure dst_mm is on swapoff's mmlist. */

814

/* make sure dst_mm is on swapoff's mmlist. */

815

if (unlikely(list_empty(&dst_mm->mmlist))) {

815

if (unlikely(list_empty(&dst_mm->mmlist))) {

816

spin_lock(&mmlist_lock);

816

spin_lock(&mmlist_lock);

817

if (list_empty(&dst_mm->mmlist))

817

if (list_empty(&dst_mm->mmlist))

818

list_add(&dst_mm->mmlist,

818

list_add(&dst_mm->mmlist,

819

&src_mm->mmlist);

819

&src_mm->mmlist);

820

spin_unlock(&mmlist_lock);

820

spin_unlock(&mmlist_lock);

821

}

821

}

822

if (likely(!non_swap_entry(entry)))

822

if (likely(!non_swap_entry(entry)))

823

rss[MM_SWAPENTS]++;

823

rss[MM_SWAPENTS]++;

824

else if (is_migration_entry(entry)) {

824

else if (is_migration_entry(entry)) {

825

page = migration_entry_to_page(entry);

825

page = migration_entry_to_page(entry);

826

827

if (PageAnon(page))

827

if (PageAnon(page))

828

rss[MM_ANONPAGES]++;

828

rss[MM_ANONPAGES]++;

829

else

829

else

830

rss[MM_FILEPAGES]++;

830

rss[MM_FILEPAGES]++;

831

832

if (is_write_migration_entry(entry) &&

832

if (is_write_migration_entry(entry) &&

833

is_cow_mapping(vm_flags)) {

833

is_cow_mapping(vm_flags)) {

834

/*

834

/*

835

* COW mappings require pages in both

835

* COW mappings require pages in both

836

* parent and child to be set to read.

836

* parent and child to be set to read.

837

*/

837

*/

838

make_migration_entry_read(&entry);

838

make_migration_entry_read(&entry);

839

pte = swp_entry_to_pte(entry);

839

pte = swp_entry_to_pte(entry);

840

if (pte_swp_soft_dirty(*src_pte))

840

if (pte_swp_soft_dirty(*src_pte))

841

pte = pte_swp_mksoft_dirty(pte);

841

pte = pte_swp_mksoft_dirty(pte);

842

set_pte_at(src_mm, addr, src_pte, pte);

842

set_pte_at(src_mm, addr, src_pte, pte);

843

}

843

}

844

}

844

}

845

}

845

}

846

goto out_set_pte;

846

goto out_set_pte;

847

}

847

}

848

849

/*

849

/*

850

* If it's a COW mapping, write protect it both

850

* If it's a COW mapping, write protect it both

851

* in the parent and the child

851

* in the parent and the child

852

*/

852

*/

853

if (is_cow_mapping(vm_flags)) {

853

if (is_cow_mapping(vm_flags)) {

854

ptep_set_wrprotect(src_mm, addr, src_pte);

854

ptep_set_wrprotect(src_mm, addr, src_pte);

855

pte = pte_wrprotect(pte);

855

pte = pte_wrprotect(pte);

856

}

856

}

857

858

/*

858

/*

859

* If it's a shared mapping, mark it clean in

859

* If it's a shared mapping, mark it clean in

860

* the child

860

* the child

861

*/

861

*/

862

if (vm_flags & VM_SHARED)

862

if (vm_flags & VM_SHARED)

863

pte = pte_mkclean(pte);

863

pte = pte_mkclean(pte);

864

pte = pte_mkold(pte);

864

pte = pte_mkold(pte);

865

866

page = vm_normal_page(vma, addr, pte);

866

page = vm_normal_page(vma, addr, pte);

867

if (page) {

867

if (page) {

868

get_page(page);

868

get_page(page);

869

page_dup_rmap(page);

869

page_dup_rmap(page);

870

if (PageAnon(page))

870

if (PageAnon(page))

871

rss[MM_ANONPAGES]++;

871

rss[MM_ANONPAGES]++;

872

else

872

else

873

rss[MM_FILEPAGES]++;

873

rss[MM_FILEPAGES]++;

874

}

874

}

875

876

out_set_pte:

876

out_set_pte:

877

set_pte_at(dst_mm, addr, dst_pte, pte);

877

set_pte_at(dst_mm, addr, dst_pte, pte);

878

return 0;

878

return 0;

879

}

879

}

880

881

int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,

881

static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,

882

pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,

882

pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,

883

unsigned long addr, unsigned long end)

883

unsigned long addr, unsigned long end)

884

{

884

{

885

pte_t *orig_src_pte, *orig_dst_pte;

885

pte_t *orig_src_pte, *orig_dst_pte;

886

pte_t *src_pte, *dst_pte;

886

pte_t *src_pte, *dst_pte;

887

spinlock_t *src_ptl, *dst_ptl;

887

spinlock_t *src_ptl, *dst_ptl;

888

int progress = 0;

888

int progress = 0;

889

int rss[NR_MM_COUNTERS];

889

int rss[NR_MM_COUNTERS];

890

swp_entry_t entry = (swp_entry_t){0};

890

swp_entry_t entry = (swp_entry_t){0};

891

892

again:

892

again:

893

init_rss_vec(rss);

893

init_rss_vec(rss);

894

895

dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);

895

dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);

896

if (!dst_pte)

896

if (!dst_pte)

897

return -ENOMEM;

897

return -ENOMEM;

898

src_pte = pte_offset_map(src_pmd, addr);

898

src_pte = pte_offset_map(src_pmd, addr);

899

src_ptl = pte_lockptr(src_mm, src_pmd);

899

src_ptl = pte_lockptr(src_mm, src_pmd);

900

spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);

900

spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);

901

orig_src_pte = src_pte;

901

orig_src_pte = src_pte;

902

orig_dst_pte = dst_pte;

902

orig_dst_pte = dst_pte;

903

arch_enter_lazy_mmu_mode();

903

arch_enter_lazy_mmu_mode();

904

905

do {

905

do {

906

/*

906

/*

907

* We are holding two locks at this point - either of them

907

* We are holding two locks at this point - either of them

908

* could generate latencies in another task on another CPU.

908

* could generate latencies in another task on another CPU.

909

*/

909

*/

910

if (progress >= 32) {

910

if (progress >= 32) {

911

progress = 0;

911

progress = 0;

912

if (need_resched() ||

912

if (need_resched() ||

913

spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))

913

spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))

914

break;

914

break;

915

}

915

}

916

if (pte_none(*src_pte)) {

916

if (pte_none(*src_pte)) {

917

progress++;

917

progress++;

918

continue;

918

continue;

919

}

919

}

920

entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte,

920

entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte,

921

vma, addr, rss);

921

vma, addr, rss);

922

if (entry.val)

922

if (entry.val)

923

break;

923

break;

924

progress += 8;

924

progress += 8;

925

} while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);

925

} while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);

926

927

arch_leave_lazy_mmu_mode();

927

arch_leave_lazy_mmu_mode();

928

spin_unlock(src_ptl);

928

spin_unlock(src_ptl);

929

pte_unmap(orig_src_pte);

929

pte_unmap(orig_src_pte);

930

add_mm_rss_vec(dst_mm, rss);

930

add_mm_rss_vec(dst_mm, rss);

931

pte_unmap_unlock(orig_dst_pte, dst_ptl);

931

pte_unmap_unlock(orig_dst_pte, dst_ptl);

932

cond_resched();

932

cond_resched();

933

934

if (entry.val) {

934

if (entry.val) {

935

if (add_swap_count_continuation(entry, GFP_KERNEL) < 0)

935

if (add_swap_count_continuation(entry, GFP_KERNEL) < 0)

936

return -ENOMEM;

936

return -ENOMEM;

937

progress = 0;

937

progress = 0;

938

}

938

}

939

if (addr != end)

939

if (addr != end)

940

goto again;

940

goto again;

941

return 0;

941

return 0;

942

}

942

}

943

944

static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,

944

static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,

945

pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,

945

pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,

946

unsigned long addr, unsigned long end)

946

unsigned long addr, unsigned long end)

947

{

947

{

948

pmd_t *src_pmd, *dst_pmd;

948

pmd_t *src_pmd, *dst_pmd;

949

unsigned long next;

949

unsigned long next;

950

951

dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);

951

dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);

952

if (!dst_pmd)

952

if (!dst_pmd)

953

return -ENOMEM;

953

return -ENOMEM;

954

src_pmd = pmd_offset(src_pud, addr);

954

src_pmd = pmd_offset(src_pud, addr);

955

do {

955

do {

956

next = pmd_addr_end(addr, end);

956

next = pmd_addr_end(addr, end);

957

if (pmd_trans_huge(*src_pmd)) {

957

if (pmd_trans_huge(*src_pmd)) {

958

int err;

958

int err;

959

VM_BUG_ON(next-addr != HPAGE_PMD_SIZE);

959

VM_BUG_ON(next-addr != HPAGE_PMD_SIZE);

960

err = copy_huge_pmd(dst_mm, src_mm,

960

err = copy_huge_pmd(dst_mm, src_mm,

961

dst_pmd, src_pmd, addr, vma);

961

dst_pmd, src_pmd, addr, vma);

962

if (err == -ENOMEM)

962

if (err == -ENOMEM)

963

return -ENOMEM;

963

return -ENOMEM;

964

if (!err)

964

if (!err)

965

continue;

965

continue;

966

/* fall through */

966

/* fall through */

967

}

967

}

968

if (pmd_none_or_clear_bad(src_pmd))

968

if (pmd_none_or_clear_bad(src_pmd))

969

continue;

969

continue;

970

if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,

970

if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,

971

vma, addr, next))

971

vma, addr, next))

972

return -ENOMEM;

972

return -ENOMEM;

973

} while (dst_pmd++, src_pmd++, addr = next, addr != end);

973

} while (dst_pmd++, src_pmd++, addr = next, addr != end);

974

return 0;

974

return 0;

975

}

975

}

976

977

static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,

977

static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,

978

pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,

978

pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,

979

unsigned long addr, unsigned long end)

979

unsigned long addr, unsigned long end)

980

{

980

{

981

pud_t *src_pud, *dst_pud;

981

pud_t *src_pud, *dst_pud;

982

unsigned long next;

982

unsigned long next;

983

984

dst_pud = pud_alloc(dst_mm, dst_pgd, addr);

984

dst_pud = pud_alloc(dst_mm, dst_pgd, addr);

985

if (!dst_pud)

985

if (!dst_pud)

986

return -ENOMEM;

986

return -ENOMEM;

987

src_pud = pud_offset(src_pgd, addr);

987

src_pud = pud_offset(src_pgd, addr);

988

do {

988

do {

989

next = pud_addr_end(addr, end);

989

next = pud_addr_end(addr, end);

990

if (pud_none_or_clear_bad(src_pud))

990

if (pud_none_or_clear_bad(src_pud))

991

continue;

991

continue;

992

if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,

992

if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,

993

vma, addr, next))

993

vma, addr, next))

994

return -ENOMEM;

994

return -ENOMEM;

995

} while (dst_pud++, src_pud++, addr = next, addr != end);

995

} while (dst_pud++, src_pud++, addr = next, addr != end);

996

return 0;

996

return 0;

997

}

997

}

998

999

int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,

999

int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,

1000

struct vm_area_struct *vma)

1000

struct vm_area_struct *vma)

1001

{

1001

{

1002

pgd_t *src_pgd, *dst_pgd;

1002

pgd_t *src_pgd, *dst_pgd;

1003

unsigned long next;

1003

unsigned long next;

1004

unsigned long addr = vma->vm_start;

1004

unsigned long addr = vma->vm_start;

1005

unsigned long end = vma->vm_end;

1005

unsigned long end = vma->vm_end;

1006

unsigned long mmun_start; /* For mmu_notifiers */

1006

unsigned long mmun_start; /* For mmu_notifiers */

1007

unsigned long mmun_end; /* For mmu_notifiers */

1007

unsigned long mmun_end; /* For mmu_notifiers */

1008

bool is_cow;

1008

bool is_cow;

1009

int ret;

1009

int ret;

1010

1011

/*

1011

/*

1012

* Don't copy ptes where a page fault will fill them correctly.

1012

* Don't copy ptes where a page fault will fill them correctly.

1013

* Fork becomes much lighter when there are big shared or private

1013

* Fork becomes much lighter when there are big shared or private

1014

* readonly mappings. The tradeoff is that copy_page_range is more

1014

* readonly mappings. The tradeoff is that copy_page_range is more

1015

* efficient than faulting.

1015

* efficient than faulting.

1016

*/

1016

*/

1017

if (!(vma->vm_flags & (VM_HUGETLB | VM_NONLINEAR |

1017

if (!(vma->vm_flags & (VM_HUGETLB | VM_NONLINEAR |

1018

VM_PFNMAP | VM_MIXEDMAP))) {

1018

VM_PFNMAP | VM_MIXEDMAP))) {

1019

if (!vma->anon_vma)

1019

if (!vma->anon_vma)

1020

return 0;

1020

return 0;

1021

}

1021

}

1022

1023

if (is_vm_hugetlb_page(vma))

1023

if (is_vm_hugetlb_page(vma))

1024

return copy_hugetlb_page_range(dst_mm, src_mm, vma);

1024

return copy_hugetlb_page_range(dst_mm, src_mm, vma);

1025

1026

if (unlikely(vma->vm_flags & VM_PFNMAP)) {

1026

if (unlikely(vma->vm_flags & VM_PFNMAP)) {

1027

/*

1027

/*

1028

* We do not free on error cases below as remove_vma

1028

* We do not free on error cases below as remove_vma

1029

* gets called on error from higher level routine

1029

* gets called on error from higher level routine

1030

*/

1030

*/

1031

ret = track_pfn_copy(vma);

1031

ret = track_pfn_copy(vma);

1032

if (ret)

1032

if (ret)

1033

return ret;

1033

return ret;

1034

}

1034

}

1035

1036

/*

1036

/*

1037

* We need to invalidate the secondary MMU mappings only when

1037

* We need to invalidate the secondary MMU mappings only when

1038

* there could be a permission downgrade on the ptes of the

1038

* there could be a permission downgrade on the ptes of the

1039

* parent mm. And a permission downgrade will only happen if

1039

* parent mm. And a permission downgrade will only happen if

1040

* is_cow_mapping() returns true.

1040

* is_cow_mapping() returns true.

1041

*/

1041

*/

1042

is_cow = is_cow_mapping(vma->vm_flags);

1042

is_cow = is_cow_mapping(vma->vm_flags);

1043

mmun_start = addr;

1043

mmun_start = addr;

1044

mmun_end = end;

1044

mmun_end = end;

1045

if (is_cow)

1045

if (is_cow)

1046

mmu_notifier_invalidate_range_start(src_mm, mmun_start,

1046

mmu_notifier_invalidate_range_start(src_mm, mmun_start,

1047

mmun_end);

1047

mmun_end);

1048

1049

ret = 0;

1049

ret = 0;

1050

dst_pgd = pgd_offset(dst_mm, addr);

1050

dst_pgd = pgd_offset(dst_mm, addr);

1051

src_pgd = pgd_offset(src_mm, addr);

1051

src_pgd = pgd_offset(src_mm, addr);

1052

do {

1052

do {

1053

next = pgd_addr_end(addr, end);

1053

next = pgd_addr_end(addr, end);

1054

if (pgd_none_or_clear_bad(src_pgd))

1054

if (pgd_none_or_clear_bad(src_pgd))

1055

continue;

1055

continue;

1056

if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,

1056

if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,

1057

vma, addr, next))) {

1057

vma, addr, next))) {

1058

ret = -ENOMEM;

1058

ret = -ENOMEM;

1059

break;

1059

break;

1060

}

1060

}

1061

} while (dst_pgd++, src_pgd++, addr = next, addr != end);

1061

} while (dst_pgd++, src_pgd++, addr = next, addr != end);

1062

1063

if (is_cow)

1063

if (is_cow)

1064

mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end);

1064

mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end);

1065

return ret;

1065

return ret;

1066

}

1066

}

1067

1068

static unsigned long zap_pte_range(struct mmu_gather *tlb,

1068

static unsigned long zap_pte_range(struct mmu_gather *tlb,

1069

struct vm_area_struct *vma, pmd_t *pmd,

1069

struct vm_area_struct *vma, pmd_t *pmd,

1070

unsigned long addr, unsigned long end,

1070

unsigned long addr, unsigned long end,

1071

struct zap_details *details)

1071

struct zap_details *details)

1072

{

1072

{

1073

struct mm_struct *mm = tlb->mm;

1073

struct mm_struct *mm = tlb->mm;

1074

int force_flush = 0;

1074

int force_flush = 0;

1075

int rss[NR_MM_COUNTERS];

1075

int rss[NR_MM_COUNTERS];

1076

spinlock_t *ptl;

1076

spinlock_t *ptl;

1077

pte_t *start_pte;

1077

pte_t *start_pte;

1078

pte_t *pte;

1078

pte_t *pte;

1079

1080

again:

1080

again:

1081

init_rss_vec(rss);

1081

init_rss_vec(rss);

1082

start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl);

1082

start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl);

1083

pte = start_pte;

1083

pte = start_pte;

1084

arch_enter_lazy_mmu_mode();

1084

arch_enter_lazy_mmu_mode();

1085

do {

1085

do {

1086

pte_t ptent = *pte;

1086

pte_t ptent = *pte;

1087

if (pte_none(ptent)) {

1087

if (pte_none(ptent)) {

1088

continue;

1088

continue;

1089

}

1089

}

1090

1091

if (pte_present(ptent)) {

1091

if (pte_present(ptent)) {

1092

struct page *page;

1092

struct page *page;

1093

1094

page = vm_normal_page(vma, addr, ptent);

1094

page = vm_normal_page(vma, addr, ptent);

1095

if (unlikely(details) && page) {

1095

if (unlikely(details) && page) {

1096

/*

1096

/*

1097

* unmap_shared_mapping_pages() wants to

1097

* unmap_shared_mapping_pages() wants to

1098

* invalidate cache without truncating:

1098

* invalidate cache without truncating:

1099

* unmap shared but keep private pages.

1099

* unmap shared but keep private pages.

1100

*/

1100

*/

1101

if (details->check_mapping &&

1101

if (details->check_mapping &&

1102

details->check_mapping != page->mapping)

1102

details->check_mapping != page->mapping)

1103

continue;

1103

continue;

1104

/*

1104

/*

1105

* Each page->index must be checked when

1105

* Each page->index must be checked when

1106

* invalidating or truncating nonlinear.

1106

* invalidating or truncating nonlinear.

1107

*/

1107

*/

1108

if (details->nonlinear_vma &&

1108

if (details->nonlinear_vma &&

1109

(page->index < details->first_index ||

1109

(page->index < details->first_index ||

1110

page->index > details->last_index))

1110

page->index > details->last_index))

1111

continue;

1111

continue;

1112

}

1112

}

1113

ptent = ptep_get_and_clear_full(mm, addr, pte,

1113

ptent = ptep_get_and_clear_full(mm, addr, pte,

1114

tlb->fullmm);

1114

tlb->fullmm);

1115

tlb_remove_tlb_entry(tlb, pte, addr);

1115

tlb_remove_tlb_entry(tlb, pte, addr);

1116

if (unlikely(!page))

1116

if (unlikely(!page))

1117

continue;

1117

continue;

1118

if (unlikely(details) && details->nonlinear_vma

1118

if (unlikely(details) && details->nonlinear_vma

1119

&& linear_page_index(details->nonlinear_vma,

1119

&& linear_page_index(details->nonlinear_vma,

1120

addr) != page->index) {

1120

addr) != page->index) {

1121

pte_t ptfile = pgoff_to_pte(page->index);

1121

pte_t ptfile = pgoff_to_pte(page->index);

1122

if (pte_soft_dirty(ptent))

1122

if (pte_soft_dirty(ptent))

1123

pte_file_mksoft_dirty(ptfile);

1123

pte_file_mksoft_dirty(ptfile);

1124

set_pte_at(mm, addr, pte, ptfile);

1124

set_pte_at(mm, addr, pte, ptfile);

1125

}

1125

}

1126

if (PageAnon(page))

1126

if (PageAnon(page))

1127

rss[MM_ANONPAGES]--;

1127

rss[MM_ANONPAGES]--;

1128

else {

1128

else {

1129

if (pte_dirty(ptent))

1129

if (pte_dirty(ptent))

1130

set_page_dirty(page);

1130

set_page_dirty(page);

1131

if (pte_young(ptent) &&

1131

if (pte_young(ptent) &&

1132

likely(!(vma->vm_flags & VM_SEQ_READ)))

1132

likely(!(vma->vm_flags & VM_SEQ_READ)))

1133

mark_page_accessed(page);

1133

mark_page_accessed(page);

1134

rss[MM_FILEPAGES]--;

1134

rss[MM_FILEPAGES]--;

1135

}

1135

}

1136

page_remove_rmap(page);

1136

page_remove_rmap(page);

1137

if (unlikely(page_mapcount(page) < 0))

1137

if (unlikely(page_mapcount(page) < 0))

1138

print_bad_pte(vma, addr, ptent, page);

1138

print_bad_pte(vma, addr, ptent, page);

1139

force_flush = !__tlb_remove_page(tlb, page);

1139

force_flush = !__tlb_remove_page(tlb, page);

1140

if (force_flush)

1140

if (force_flush)

1141

break;

1141

break;

1142

continue;

1142

continue;

1143

}

1143

}

1144

/*

1144

/*

1145

* If details->check_mapping, we leave swap entries;

1145

* If details->check_mapping, we leave swap entries;

1146

* if details->nonlinear_vma, we leave file entries.

1146

* if details->nonlinear_vma, we leave file entries.

1147

*/

1147

*/

1148

if (unlikely(details))

1148

if (unlikely(details))

1149

continue;

1149

continue;

1150

if (pte_file(ptent)) {

1150

if (pte_file(ptent)) {

1151

if (unlikely(!(vma->vm_flags & VM_NONLINEAR)))

1151

if (unlikely(!(vma->vm_flags & VM_NONLINEAR)))

1152

print_bad_pte(vma, addr, ptent, NULL);

1152

print_bad_pte(vma, addr, ptent, NULL);

1153

} else {

1153

} else {

1154

swp_entry_t entry = pte_to_swp_entry(ptent);

1154

swp_entry_t entry = pte_to_swp_entry(ptent);

1155

1156

if (!non_swap_entry(entry))

1156

if (!non_swap_entry(entry))

1157

rss[MM_SWAPENTS]--;

1157

rss[MM_SWAPENTS]--;

1158

else if (is_migration_entry(entry)) {

1158

else if (is_migration_entry(entry)) {

1159

struct page *page;

1159

struct page *page;

1160

1161

page = migration_entry_to_page(entry);

1161

page = migration_entry_to_page(entry);

1162

1163

if (PageAnon(page))

1163

if (PageAnon(page))

1164

rss[MM_ANONPAGES]--;

1164

rss[MM_ANONPAGES]--;

1165

else

1165

else

1166

rss[MM_FILEPAGES]--;

1166

rss[MM_FILEPAGES]--;

1167

}

1167

}

1168

if (unlikely(!free_swap_and_cache(entry)))

1168

if (unlikely(!free_swap_and_cache(entry)))

1169

print_bad_pte(vma, addr, ptent, NULL);

1169

print_bad_pte(vma, addr, ptent, NULL);

1170

}

1170

}

1171

pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);

1171

pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);

1172

} while (pte++, addr += PAGE_SIZE, addr != end);

1172

} while (pte++, addr += PAGE_SIZE, addr != end);

1173

1174

add_mm_rss_vec(mm, rss);

1174

add_mm_rss_vec(mm, rss);

1175

arch_leave_lazy_mmu_mode();

1175

arch_leave_lazy_mmu_mode();

1176

pte_unmap_unlock(start_pte, ptl);

1176

pte_unmap_unlock(start_pte, ptl);

1177

1178

/*

1178

/*

1179

* mmu_gather ran out of room to batch pages, we break out of

1179

* mmu_gather ran out of room to batch pages, we break out of

1180

* the PTE lock to avoid doing the potential expensive TLB invalidate

1180

* the PTE lock to avoid doing the potential expensive TLB invalidate

1181

* and page-free while holding it.

1181

* and page-free while holding it.

1182

*/

1182

*/

1183

if (force_flush) {

1183

if (force_flush) {

1184

unsigned long old_end;

1184

unsigned long old_end;

1185

1186

force_flush = 0;

1186

force_flush = 0;

1187

1188

/*

1188

/*

1189

* Flush the TLB just for the previous segment,

1189

* Flush the TLB just for the previous segment,

1190

* then update the range to be the remaining

1190

* then update the range to be the remaining

1191

* TLB range.

1191

* TLB range.

1192

*/

1192

*/

1193

old_end = tlb->end;

1193

old_end = tlb->end;

1194

tlb->end = addr;

1194

tlb->end = addr;

1195

1196

tlb_flush_mmu(tlb);

1196

tlb_flush_mmu(tlb);

1197

1198

tlb->start = addr;

1198

tlb->start = addr;

1199

tlb->end = old_end;

1199

tlb->end = old_end;

1200

1201

if (addr != end)

1201

if (addr != end)

1202

goto again;

1202

goto again;

1203

}

1203

}

1204

1205

return addr;

1205

return addr;

1206

}

1206

}

1207

1208

static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,

1208

static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,

1209

struct vm_area_struct *vma, pud_t *pud,

1209

struct vm_area_struct *vma, pud_t *pud,

1210

unsigned long addr, unsigned long end,

1210

unsigned long addr, unsigned long end,

1211

struct zap_details *details)

1211

struct zap_details *details)

1212

{

1212

{

1213

pmd_t *pmd;

1213

pmd_t *pmd;

1214

unsigned long next;

1214

unsigned long next;

1215

1216

pmd = pmd_offset(pud, addr);

1216

pmd = pmd_offset(pud, addr);

1217

do {

1217

do {

1218

next = pmd_addr_end(addr, end);

1218

next = pmd_addr_end(addr, end);

1219

if (pmd_trans_huge(*pmd)) {

1219

if (pmd_trans_huge(*pmd)) {

1220

if (next - addr != HPAGE_PMD_SIZE) {

1220

if (next - addr != HPAGE_PMD_SIZE) {

1221

#ifdef CONFIG_DEBUG_VM

1221

#ifdef CONFIG_DEBUG_VM

1222

if (!rwsem_is_locked(&tlb->mm->mmap_sem)) {

1222

if (!rwsem_is_locked(&tlb->mm->mmap_sem)) {

1223

pr_err("%s: mmap_sem is unlocked! addr=0x%lx end=0x%lx vma->vm_start=0x%lx vma->vm_end=0x%lx\n",

1223

pr_err("%s: mmap_sem is unlocked! addr=0x%lx end=0x%lx vma->vm_start=0x%lx vma->vm_end=0x%lx\n",

1224

__func__, addr, end,

1224

__func__, addr, end,

1225

vma->vm_start,

1225

vma->vm_start,

1226

vma->vm_end);

1226

vma->vm_end);

1227

BUG();

1227

BUG();

1228

}

1228

}

1229

#endif

1229

#endif

1230

split_huge_page_pmd(vma, addr, pmd);

1230

split_huge_page_pmd(vma, addr, pmd);

1231

} else if (zap_huge_pmd(tlb, vma, pmd, addr))

1231

} else if (zap_huge_pmd(tlb, vma, pmd, addr))

1232

goto next;

1232

goto next;

1233

/* fall through */

1233

/* fall through */

1234

}

1234

}

1235

/*

1235

/*

1236

* Here there can be other concurrent MADV_DONTNEED or

1236

* Here there can be other concurrent MADV_DONTNEED or

1237

* trans huge page faults running, and if the pmd is

1237

* trans huge page faults running, and if the pmd is

1238

* none or trans huge it can change under us. This is

1238

* none or trans huge it can change under us. This is

1239

* because MADV_DONTNEED holds the mmap_sem in read

1239

* because MADV_DONTNEED holds the mmap_sem in read

1240

* mode.

1240

* mode.

1241

*/

1241

*/

1242

if (pmd_none_or_trans_huge_or_clear_bad(pmd))

1242

if (pmd_none_or_trans_huge_or_clear_bad(pmd))

1243

goto next;

1243

goto next;

1244

next = zap_pte_range(tlb, vma, pmd, addr, next, details);

1244

next = zap_pte_range(tlb, vma, pmd, addr, next, details);

1245

cond_resched();

1246

cond_resched();

1247

} while (pmd++, addr = next, addr != end);

1247

} while (pmd++, addr = next, addr != end);

1248

1249

return addr;

1249

return addr;

1250

}

1250

}

1251

1252

static inline unsigned long zap_pud_range(struct mmu_gather *tlb,

1252

static inline unsigned long zap_pud_range(struct mmu_gather *tlb,

1253

struct vm_area_struct *vma, pgd_t *pgd,

1253

struct vm_area_struct *vma, pgd_t *pgd,

1254

unsigned long addr, unsigned long end,

1254

unsigned long addr, unsigned long end,

1255

struct zap_details *details)

1255

struct zap_details *details)

1256

{

1256

{

1257

pud_t *pud;

1257

pud_t *pud;

1258

unsigned long next;

1258

unsigned long next;

1259

1260

pud = pud_offset(pgd, addr);

1260

pud = pud_offset(pgd, addr);

1261

do {

1261

do {

1262

next = pud_addr_end(addr, end);

1262

next = pud_addr_end(addr, end);

1263

if (pud_none_or_clear_bad(pud))

1263

if (pud_none_or_clear_bad(pud))

1264

continue;

1264

continue;

1265

next = zap_pmd_range(tlb, vma, pud, addr, next, details);

1265

next = zap_pmd_range(tlb, vma, pud, addr, next, details);

1266

} while (pud++, addr = next, addr != end);

1266

} while (pud++, addr = next, addr != end);

1267

1268

return addr;

1268

return addr;

1269

}

1269

}

1270

1271

static void unmap_page_range(struct mmu_gather *tlb,

1271

static void unmap_page_range(struct mmu_gather *tlb,

1272

struct vm_area_struct *vma,

1272

struct vm_area_struct *vma,

1273

unsigned long addr, unsigned long end,

1273

unsigned long addr, unsigned long end,

1274

struct zap_details *details)

1274

struct zap_details *details)

1275

{

1275

{

1276

pgd_t *pgd;

1276

pgd_t *pgd;

1277

unsigned long next;

1277

unsigned long next;

1278

1279

if (details && !details->check_mapping && !details->nonlinear_vma)

1279

if (details && !details->check_mapping && !details->nonlinear_vma)

1280

details = NULL;

1280

details = NULL;

1281

1282

BUG_ON(addr >= end);

1282

BUG_ON(addr >= end);

1283

mem_cgroup_uncharge_start();

1283

mem_cgroup_uncharge_start();

1284

tlb_start_vma(tlb, vma);

1284

tlb_start_vma(tlb, vma);

1285

pgd = pgd_offset(vma->vm_mm, addr);

1285

pgd = pgd_offset(vma->vm_mm, addr);

1286

do {

1286

do {

1287

next = pgd_addr_end(addr, end);

1287

next = pgd_addr_end(addr, end);

1288

if (pgd_none_or_clear_bad(pgd))

1288

if (pgd_none_or_clear_bad(pgd))

1289

continue;

1289

continue;

1290

next = zap_pud_range(tlb, vma, pgd, addr, next, details);

1290

next = zap_pud_range(tlb, vma, pgd, addr, next, details);

1291

} while (pgd++, addr = next, addr != end);

1291

} while (pgd++, addr = next, addr != end);

1292

tlb_end_vma(tlb, vma);

1292

tlb_end_vma(tlb, vma);

1293

mem_cgroup_uncharge_end();

1293

mem_cgroup_uncharge_end();

1294

}

1294

}

1295

1296

1297

static void unmap_single_vma(struct mmu_gather *tlb,

1297

static void unmap_single_vma(struct mmu_gather *tlb,

1298

struct vm_area_struct *vma, unsigned long start_addr,

1298

struct vm_area_struct *vma, unsigned long start_addr,

1299

unsigned long end_addr,

1299

unsigned long end_addr,

1300

struct zap_details *details)

1300

struct zap_details *details)

1301

{

1301

{

1302

unsigned long start = max(vma->vm_start, start_addr);

1302

unsigned long start = max(vma->vm_start, start_addr);

1303

unsigned long end;

1303

unsigned long end;

1304

1305

if (start >= vma->vm_end)

1305

if (start >= vma->vm_end)

1306

return;

1306

return;

1307

end = min(vma->vm_end, end_addr);

1307

end = min(vma->vm_end, end_addr);

1308

if (end <= vma->vm_start)

1308

if (end <= vma->vm_start)

1309

return;

1309

return;

1310

1311

if (vma->vm_file)

1311

if (vma->vm_file)

1312

uprobe_munmap(vma, start, end);

1312

uprobe_munmap(vma, start, end);

1313

1314

if (unlikely(vma->vm_flags & VM_PFNMAP))

1314

if (unlikely(vma->vm_flags & VM_PFNMAP))

1315

untrack_pfn(vma, 0, 0);

1315

untrack_pfn(vma, 0, 0);

1316

1317

if (start != end) {

1317

if (start != end) {

1318

if (unlikely(is_vm_hugetlb_page(vma))) {

1318

if (unlikely(is_vm_hugetlb_page(vma))) {

1319

/*

1319

/*

1320

* It is undesirable to test vma->vm_file as it

1320

* It is undesirable to test vma->vm_file as it

1321

* should be non-null for valid hugetlb area.

1321

* should be non-null for valid hugetlb area.

1322

* However, vm_file will be NULL in the error

1322

* However, vm_file will be NULL in the error

1323

* cleanup path of do_mmap_pgoff. When

1323

* cleanup path of do_mmap_pgoff. When

1324

* hugetlbfs ->mmap method fails,

1324

* hugetlbfs ->mmap method fails,

1325

* do_mmap_pgoff() nullifies vma->vm_file

1325

* do_mmap_pgoff() nullifies vma->vm_file

1326

* before calling this function to clean up.

1326

* before calling this function to clean up.

1327

* Since no pte has actually been setup, it is

1327

* Since no pte has actually been setup, it is

1328

* safe to do nothing in this case.

1328

* safe to do nothing in this case.

1329

*/

1329

*/

1330

if (vma->vm_file) {

1330

if (vma->vm_file) {

1331

mutex_lock(&vma->vm_file->f_mapping->i_mmap_mutex);

1331

mutex_lock(&vma->vm_file->f_mapping->i_mmap_mutex);

1332

__unmap_hugepage_range_final(tlb, vma, start, end, NULL);

1332

__unmap_hugepage_range_final(tlb, vma, start, end, NULL);

1333

mutex_unlock(&vma->vm_file->f_mapping->i_mmap_mutex);

1333

mutex_unlock(&vma->vm_file->f_mapping->i_mmap_mutex);

1334

}

1334

}

1335

} else

1335

} else

1336

unmap_page_range(tlb, vma, start, end, details);

1336

unmap_page_range(tlb, vma, start, end, details);

1337

}

1337

}

1338

}

1338

}

1339

1340

/**

1340

/**

1341

* unmap_vmas - unmap a range of memory covered by a list of vma's

1341

* unmap_vmas - unmap a range of memory covered by a list of vma's

1342

* @tlb: address of the caller's struct mmu_gather

1342

* @tlb: address of the caller's struct mmu_gather

1343

* @vma: the starting vma

1343

* @vma: the starting vma

1344

* @start_addr: virtual address at which to start unmapping

1344

* @start_addr: virtual address at which to start unmapping

1345

* @end_addr: virtual address at which to end unmapping

1345

* @end_addr: virtual address at which to end unmapping

1346

*

1346

*

1347

* Unmap all pages in the vma list.

1347

* Unmap all pages in the vma list.

1348

*

1348

*

1349

* Only addresses between `start' and `end' will be unmapped.

1349

* Only addresses between `start' and `end' will be unmapped.

1350

*

1350

*

1351

* The VMA list must be sorted in ascending virtual address order.

1351

* The VMA list must be sorted in ascending virtual address order.

1352

*

1352

*

1353

* unmap_vmas() assumes that the caller will flush the whole unmapped address

1353

* unmap_vmas() assumes that the caller will flush the whole unmapped address

1354

* range after unmap_vmas() returns. So the only responsibility here is to

1354

* range after unmap_vmas() returns. So the only responsibility here is to

1355

* ensure that any thus-far unmapped pages are flushed before unmap_vmas()

1355

* ensure that any thus-far unmapped pages are flushed before unmap_vmas()

1356

* drops the lock and schedules.

1356

* drops the lock and schedules.

1357

*/

1357

*/

1358

void unmap_vmas(struct mmu_gather *tlb,

1358

void unmap_vmas(struct mmu_gather *tlb,

1359

struct vm_area_struct *vma, unsigned long start_addr,

1359

struct vm_area_struct *vma, unsigned long start_addr,

1360

unsigned long end_addr)

1360

unsigned long end_addr)

1361

{

1361

{

1362

struct mm_struct *mm = vma->vm_mm;

1362

struct mm_struct *mm = vma->vm_mm;

1363

1364

mmu_notifier_invalidate_range_start(mm, start_addr, end_addr);

1364

mmu_notifier_invalidate_range_start(mm, start_addr, end_addr);

1365

for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next)

1365

for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next)

1366

unmap_single_vma(tlb, vma, start_addr, end_addr, NULL);

1366

unmap_single_vma(tlb, vma, start_addr, end_addr, NULL);

1367

mmu_notifier_invalidate_range_end(mm, start_addr, end_addr);

1367

mmu_notifier_invalidate_range_end(mm, start_addr, end_addr);

1368

}

1368

}

1369

1370

/**

1370

/**

1371

* zap_page_range - remove user pages in a given range

1371

* zap_page_range - remove user pages in a given range

1372

* @vma: vm_area_struct holding the applicable pages

1372

* @vma: vm_area_struct holding the applicable pages

1373

* @start: starting address of pages to zap

1373

* @start: starting address of pages to zap

1374

* @size: number of bytes to zap

1374

* @size: number of bytes to zap

1375

* @details: details of nonlinear truncation or shared cache invalidation

1375

* @details: details of nonlinear truncation or shared cache invalidation

1376

*

1376

*

1377

* Caller must protect the VMA list

1377

* Caller must protect the VMA list

1378

*/

1378

*/

1379

void zap_page_range(struct vm_area_struct *vma, unsigned long start,

1379

void zap_page_range(struct vm_area_struct *vma, unsigned long start,

1380

unsigned long size, struct zap_details *details)

1380

unsigned long size, struct zap_details *details)

1381

{

1381

{

1382

struct mm_struct *mm = vma->vm_mm;

1382

struct mm_struct *mm = vma->vm_mm;

1383

struct mmu_gather tlb;

1383

struct mmu_gather tlb;

1384

unsigned long end = start + size;

1384

unsigned long end = start + size;

1385

1386

lru_add_drain();

1386

lru_add_drain();

1387

tlb_gather_mmu(&tlb, mm, start, end);

1387

tlb_gather_mmu(&tlb, mm, start, end);

1388

update_hiwater_rss(mm);

1388

update_hiwater_rss(mm);

1389

mmu_notifier_invalidate_range_start(mm, start, end);

1389

mmu_notifier_invalidate_range_start(mm, start, end);

1390

for ( ; vma && vma->vm_start < end; vma = vma->vm_next)

1390

for ( ; vma && vma->vm_start < end; vma = vma->vm_next)

1391

unmap_single_vma(&tlb, vma, start, end, details);

1391

unmap_single_vma(&tlb, vma, start, end, details);

1392

mmu_notifier_invalidate_range_end(mm, start, end);

1392

mmu_notifier_invalidate_range_end(mm, start, end);

1393

tlb_finish_mmu(&tlb, start, end);

1393

tlb_finish_mmu(&tlb, start, end);

1394

}

1394

}

1395

1396

/**

1396

/**

1397

* zap_page_range_single - remove user pages in a given range

1397

* zap_page_range_single - remove user pages in a given range

1398

* @vma: vm_area_struct holding the applicable pages

1398

* @vma: vm_area_struct holding the applicable pages

1399

* @address: starting address of pages to zap

1399

* @address: starting address of pages to zap

1400

* @size: number of bytes to zap

1400

* @size: number of bytes to zap

1401

* @details: details of nonlinear truncation or shared cache invalidation

1401

* @details: details of nonlinear truncation or shared cache invalidation

1402

*

1402

*

1403

* The range must fit into one VMA.

1403

* The range must fit into one VMA.

1404

*/

1404

*/

1405

static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address,

1405

static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address,

1406

unsigned long size, struct zap_details *details)

1406

unsigned long size, struct zap_details *details)

1407

{

1407

{

1408

struct mm_struct *mm = vma->vm_mm;

1408

struct mm_struct *mm = vma->vm_mm;

1409

struct mmu_gather tlb;

1409

struct mmu_gather tlb;

1410

unsigned long end = address + size;

1410

unsigned long end = address + size;

1411

1412

lru_add_drain();

1412

lru_add_drain();

1413

tlb_gather_mmu(&tlb, mm, address, end);

1413

tlb_gather_mmu(&tlb, mm, address, end);

1414

update_hiwater_rss(mm);

1414

update_hiwater_rss(mm);

1415

mmu_notifier_invalidate_range_start(mm, address, end);

1415

mmu_notifier_invalidate_range_start(mm, address, end);

1416

unmap_single_vma(&tlb, vma, address, end, details);

1416

unmap_single_vma(&tlb, vma, address, end, details);

1417

mmu_notifier_invalidate_range_end(mm, address, end);

1417

mmu_notifier_invalidate_range_end(mm, address, end);

1418

tlb_finish_mmu(&tlb, address, end);

1418

tlb_finish_mmu(&tlb, address, end);

1419

}

1419

}

1420

1421

/**

1421

/**

1422

* zap_vma_ptes - remove ptes mapping the vma

1422

* zap_vma_ptes - remove ptes mapping the vma

1423

* @vma: vm_area_struct holding ptes to be zapped

1423

* @vma: vm_area_struct holding ptes to be zapped

1424

* @address: starting address of pages to zap

1424

* @address: starting address of pages to zap

1425

* @size: number of bytes to zap

1425

* @size: number of bytes to zap

1426

*

1426

*

1427

* This function only unmaps ptes assigned to VM_PFNMAP vmas.

1427

* This function only unmaps ptes assigned to VM_PFNMAP vmas.

1428

*

1428

*

1429

* The entire address range must be fully contained within the vma.

1429

* The entire address range must be fully contained within the vma.

1430

*

1430

*

1431

* Returns 0 if successful.

1431

* Returns 0 if successful.

1432

*/

1432

*/

1433

int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,

1433

int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,

1434

unsigned long size)

1434

unsigned long size)

1435

{

1435

{

1436

if (address < vma->vm_start || address + size > vma->vm_end ||

1436

if (address < vma->vm_start || address + size > vma->vm_end ||

1437

!(vma->vm_flags & VM_PFNMAP))

1437

!(vma->vm_flags & VM_PFNMAP))

1438

return -1;

1438

return -1;

1439

zap_page_range_single(vma, address, size, NULL);

1439

zap_page_range_single(vma, address, size, NULL);

1440

return 0;

1440

return 0;

1441

}

1441

}

1442

EXPORT_SYMBOL_GPL(zap_vma_ptes);

1442

EXPORT_SYMBOL_GPL(zap_vma_ptes);

1443

1444

/**

1444

/**

1445

* follow_page_mask - look up a page descriptor from a user-virtual address

1445

* follow_page_mask - look up a page descriptor from a user-virtual address

1446

* @vma: vm_area_struct mapping @address

1446

* @vma: vm_area_struct mapping @address

1447

* @address: virtual address to look up

1447

* @address: virtual address to look up

1448

* @flags: flags modifying lookup behaviour

1448

* @flags: flags modifying lookup behaviour

1449

* @page_mask: on output, *page_mask is set according to the size of the page

1449

* @page_mask: on output, *page_mask is set according to the size of the page

1450

*

1450

*

1451

* @flags can have FOLL_ flags set, defined in <linux/mm.h>

1451

* @flags can have FOLL_ flags set, defined in <linux/mm.h>

1452

*

1452

*

1453

* Returns the mapped (struct page *), %NULL if no mapping exists, or

1453

* Returns the mapped (struct page *), %NULL if no mapping exists, or

1454

* an error pointer if there is a mapping to something not represented

1454

* an error pointer if there is a mapping to something not represented

1455

* by a page descriptor (see also vm_normal_page()).

1455

* by a page descriptor (see also vm_normal_page()).

1456

*/

1456

*/

1457

struct page *follow_page_mask(struct vm_area_struct *vma,

1457

struct page *follow_page_mask(struct vm_area_struct *vma,

1458

unsigned long address, unsigned int flags,

1458

unsigned long address, unsigned int flags,

1459

unsigned int *page_mask)

1459

unsigned int *page_mask)

1460

{

1460

{

1461

pgd_t *pgd;

1461

pgd_t *pgd;

1462

pud_t *pud;

1462

pud_t *pud;

1463

pmd_t *pmd;

1463

pmd_t *pmd;

1464

pte_t *ptep, pte;

1464

pte_t *ptep, pte;

1465

spinlock_t *ptl;

1465

spinlock_t *ptl;

1466

struct page *page;

1466

struct page *page;

1467

struct mm_struct *mm = vma->vm_mm;

1467

struct mm_struct *mm = vma->vm_mm;

1468

1469

*page_mask = 0;

1469

*page_mask = 0;

1470

1471

page = follow_huge_addr(mm, address, flags & FOLL_WRITE);

1471

page = follow_huge_addr(mm, address, flags & FOLL_WRITE);

1472

if (!IS_ERR(page)) {

1472

if (!IS_ERR(page)) {

1473

BUG_ON(flags & FOLL_GET);

1473

BUG_ON(flags & FOLL_GET);

1474

goto out;

1474

goto out;

1475

}

1475

}

1476

1477

page = NULL;

1477

page = NULL;

1478

pgd = pgd_offset(mm, address);

1478

pgd = pgd_offset(mm, address);

1479

if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))

1479

if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))

1480

goto no_page_table;

1480

goto no_page_table;

1481

1482

pud = pud_offset(pgd, address);

1482

pud = pud_offset(pgd, address);

1483

if (pud_none(*pud))

1483

if (pud_none(*pud))

1484

goto no_page_table;

1484

goto no_page_table;

1485

if (pud_huge(*pud) && vma->vm_flags & VM_HUGETLB) {

1485

if (pud_huge(*pud) && vma->vm_flags & VM_HUGETLB) {

1486

if (flags & FOLL_GET)

1486

if (flags & FOLL_GET)

1487

goto out;

1487

goto out;

1488

page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE);

1488

page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE);

1489

goto out;

1489

goto out;

1490

}

1490

}

1491

if (unlikely(pud_bad(*pud)))

1491

if (unlikely(pud_bad(*pud)))

1492

goto no_page_table;

1492

goto no_page_table;

1493

1494

pmd = pmd_offset(pud, address);

1494

pmd = pmd_offset(pud, address);

1495

if (pmd_none(*pmd))

1495

if (pmd_none(*pmd))

1496

goto no_page_table;

1496

goto no_page_table;

1497

if (pmd_huge(*pmd) && vma->vm_flags & VM_HUGETLB) {

1497

if (pmd_huge(*pmd) && vma->vm_flags & VM_HUGETLB) {

1498

page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);

1498

page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);

1499

if (flags & FOLL_GET) {

1499

if (flags & FOLL_GET) {

1500

/*

1500

/*

1501

* Refcount on tail pages are not well-defined and

1501

* Refcount on tail pages are not well-defined and

1502

* shouldn't be taken. The caller should handle a NULL

1502

* shouldn't be taken. The caller should handle a NULL

1503

* return when trying to follow tail pages.

1503

* return when trying to follow tail pages.

1504

*/

1504

*/

1505

if (PageHead(page))

1505

if (PageHead(page))

1506

get_page(page);

1506

get_page(page);

1507

else {

1507

else {

1508

page = NULL;

1508

page = NULL;

1509

goto out;

1509

goto out;

1510

}

1510

}

1511

}

1511

}

1512

goto out;

1512

goto out;

1513

}

1513

}

1514

if ((flags & FOLL_NUMA) && pmd_numa(*pmd))

1514

if ((flags & FOLL_NUMA) && pmd_numa(*pmd))

1515

goto no_page_table;

1515

goto no_page_table;

1516

if (pmd_trans_huge(*pmd)) {

1516

if (pmd_trans_huge(*pmd)) {

1517

if (flags & FOLL_SPLIT) {

1517

if (flags & FOLL_SPLIT) {

1518

split_huge_page_pmd(vma, address, pmd);

1518

split_huge_page_pmd(vma, address, pmd);

1519

goto split_fallthrough;

1519

goto split_fallthrough;

1520

}

1520

}

1521

spin_lock(&mm->page_table_lock);

1521

spin_lock(&mm->page_table_lock);

1522

if (likely(pmd_trans_huge(*pmd))) {

1522

if (likely(pmd_trans_huge(*pmd))) {

1523

if (unlikely(pmd_trans_splitting(*pmd))) {

1523

if (unlikely(pmd_trans_splitting(*pmd))) {

1524

spin_unlock(&mm->page_table_lock);

1524

spin_unlock(&mm->page_table_lock);

1525

wait_split_huge_page(vma->anon_vma, pmd);

1525

wait_split_huge_page(vma->anon_vma, pmd);

1526

} else {

1526

} else {

1527

page = follow_trans_huge_pmd(vma, address,

1527

page = follow_trans_huge_pmd(vma, address,

1528

pmd, flags);

1528

pmd, flags);

1529

spin_unlock(&mm->page_table_lock);

1529

spin_unlock(&mm->page_table_lock);

1530

*page_mask = HPAGE_PMD_NR - 1;

1530

*page_mask = HPAGE_PMD_NR - 1;

1531

goto out;

1531

goto out;

1532

}

1532

}

1533

} else

1533

} else

1534

spin_unlock(&mm->page_table_lock);

1534

spin_unlock(&mm->page_table_lock);

1535

/* fall through */

1535

/* fall through */

1536

}

1536

}

1537

split_fallthrough:

1537

split_fallthrough:

1538

if (unlikely(pmd_bad(*pmd)))

1538

if (unlikely(pmd_bad(*pmd)))

1539

goto no_page_table;

1539

goto no_page_table;

1540

1541

ptep = pte_offset_map_lock(mm, pmd, address, &ptl);

1541

ptep = pte_offset_map_lock(mm, pmd, address, &ptl);

1542

1543

pte = *ptep;

1543

pte = *ptep;

1544

if (!pte_present(pte)) {

1544

if (!pte_present(pte)) {

1545

swp_entry_t entry;

1545

swp_entry_t entry;

1546

/*

1546

/*

1547

* KSM's break_ksm() relies upon recognizing a ksm page

1547

* KSM's break_ksm() relies upon recognizing a ksm page

1548

* even while it is being migrated, so for that case we

1548

* even while it is being migrated, so for that case we

1549

* need migration_entry_wait().

1549

* need migration_entry_wait().

1550

*/

1550

*/

1551

if (likely(!(flags & FOLL_MIGRATION)))

1551

if (likely(!(flags & FOLL_MIGRATION)))

1552

goto no_page;

1552

goto no_page;

1553

if (pte_none(pte) || pte_file(pte))

1553

if (pte_none(pte) || pte_file(pte))

1554

goto no_page;

1554

goto no_page;

1555

entry = pte_to_swp_entry(pte);

1555

entry = pte_to_swp_entry(pte);

1556

if (!is_migration_entry(entry))

1556

if (!is_migration_entry(entry))

1557

goto no_page;

1557

goto no_page;

1558

pte_unmap_unlock(ptep, ptl);

1558

pte_unmap_unlock(ptep, ptl);

1559

migration_entry_wait(mm, pmd, address);

1559

migration_entry_wait(mm, pmd, address);

1560

goto split_fallthrough;

1560

goto split_fallthrough;

1561

}

1561

}

1562

if ((flags & FOLL_NUMA) && pte_numa(pte))

1562

if ((flags & FOLL_NUMA) && pte_numa(pte))

1563

goto no_page;

1563

goto no_page;

1564

if ((flags & FOLL_WRITE) && !pte_write(pte))

1564

if ((flags & FOLL_WRITE) && !pte_write(pte))

1565

goto unlock;

1565

goto unlock;

1566

1567

page = vm_normal_page(vma, address, pte);

1567

page = vm_normal_page(vma, address, pte);

1568

if (unlikely(!page)) {

1568

if (unlikely(!page)) {

1569

if ((flags & FOLL_DUMP) ||

1569

if ((flags & FOLL_DUMP) ||

1570

!is_zero_pfn(pte_pfn(pte)))

1570

!is_zero_pfn(pte_pfn(pte)))

1571

goto bad_page;

1571

goto bad_page;

1572

page = pte_page(pte);

1572

page = pte_page(pte);

1573

}

1573

}

1574

1575

if (flags & FOLL_GET)

1575

if (flags & FOLL_GET)

1576

get_page_foll(page);

1576

get_page_foll(page);

1577

if (flags & FOLL_TOUCH) {

1577

if (flags & FOLL_TOUCH) {

1578

if ((flags & FOLL_WRITE) &&

1578

if ((flags & FOLL_WRITE) &&

1579

!pte_dirty(pte) && !PageDirty(page))

1579

!pte_dirty(pte) && !PageDirty(page))

1580

set_page_dirty(page);

1580

set_page_dirty(page);

1581

/*

1581

/*

1582

* pte_mkyoung() would be more correct here, but atomic care

1582

* pte_mkyoung() would be more correct here, but atomic care

1583

* is needed to avoid losing the dirty bit: it is easier to use

1583

* is needed to avoid losing the dirty bit: it is easier to use

1584

* mark_page_accessed().

1584

* mark_page_accessed().

1585

*/

1585

*/

1586

mark_page_accessed(page);

1586

mark_page_accessed(page);

1587

}

1587

}

1588

if ((flags & FOLL_MLOCK) && (vma->vm_flags & VM_LOCKED)) {

1588

if ((flags & FOLL_MLOCK) && (vma->vm_flags & VM_LOCKED)) {

1589

/*

1589

/*

1590

* The preliminary mapping check is mainly to avoid the

1590

* The preliminary mapping check is mainly to avoid the

1591

* pointless overhead of lock_page on the ZERO_PAGE

1591

* pointless overhead of lock_page on the ZERO_PAGE

1592

* which might bounce very badly if there is contention.

1592

* which might bounce very badly if there is contention.

1593

*

1593

*

1594

* If the page is already locked, we don't need to

1594

* If the page is already locked, we don't need to

1595

* handle it now - vmscan will handle it later if and

1595

* handle it now - vmscan will handle it later if and

1596

* when it attempts to reclaim the page.

1596

* when it attempts to reclaim the page.

1597

*/

1597

*/

1598

if (page->mapping && trylock_page(page)) {

1598

if (page->mapping && trylock_page(page)) {

1599

lru_add_drain(); /* push cached pages to LRU */

1599

lru_add_drain(); /* push cached pages to LRU */

1600

/*

1600

/*

1601

* Because we lock page here, and migration is

1601

* Because we lock page here, and migration is

1602

* blocked by the pte's page reference, and we

1602

* blocked by the pte's page reference, and we

1603

* know the page is still mapped, we don't even

1603

* know the page is still mapped, we don't even

1604

* need to check for file-cache page truncation.

1604

* need to check for file-cache page truncation.

1605

*/

1605

*/

1606

mlock_vma_page(page);

1606

mlock_vma_page(page);

1607

unlock_page(page);

1607

unlock_page(page);

1608

}

1608

}

1609

}

1609

}

1610

unlock:

1610

unlock:

1611

pte_unmap_unlock(ptep, ptl);

1611

pte_unmap_unlock(ptep, ptl);

1612

out:

1612

out:

1613

return page;

1613

return page;

1614

1615

bad_page:

1615

bad_page:

1616

pte_unmap_unlock(ptep, ptl);

1616

pte_unmap_unlock(ptep, ptl);

1617

return ERR_PTR(-EFAULT);

1617

return ERR_PTR(-EFAULT);

1618

1619

no_page:

1619

no_page:

1620

pte_unmap_unlock(ptep, ptl);

1620

pte_unmap_unlock(ptep, ptl);

1621

if (!pte_none(pte))

1621

if (!pte_none(pte))

1622

return page;

1622

return page;

1623

1624

no_page_table:

1624

no_page_table:

1625

/*

1625

/*

1626

* When core dumping an enormous anonymous area that nobody

1626

* When core dumping an enormous anonymous area that nobody

1627

* has touched so far, we don't want to allocate unnecessary pages or

1627

* has touched so far, we don't want to allocate unnecessary pages or

1628

* page tables. Return error instead of NULL to skip handle_mm_fault,

1628

* page tables. Return error instead of NULL to skip handle_mm_fault,

1629

* then get_dump_page() will return NULL to leave a hole in the dump.

1629

* then get_dump_page() will return NULL to leave a hole in the dump.

1630

* But we can only make this optimization where a hole would surely

1630

* But we can only make this optimization where a hole would surely

1631

* be zero-filled if handle_mm_fault() actually did handle it.

1631

* be zero-filled if handle_mm_fault() actually did handle it.

1632

*/

1632

*/

1633

if ((flags & FOLL_DUMP) &&

1633

if ((flags & FOLL_DUMP) &&

1634

(!vma->vm_ops || !vma->vm_ops->fault))

1634

(!vma->vm_ops || !vma->vm_ops->fault))

1635

return ERR_PTR(-EFAULT);

1635

return ERR_PTR(-EFAULT);

1636

return page;

1636

return page;

1637

}

1637

}

1638

1639

static inline int stack_guard_page(struct vm_area_struct *vma, unsigned long addr)

1639

static inline int stack_guard_page(struct vm_area_struct *vma, unsigned long addr)

1640

{

1640

{

1641

return stack_guard_page_start(vma, addr) ||

1641

return stack_guard_page_start(vma, addr) ||

1642

stack_guard_page_end(vma, addr+PAGE_SIZE);

1642

stack_guard_page_end(vma, addr+PAGE_SIZE);

1643

}

1643

}

1644

1645

/**

1645

/**

1646

* __get_user_pages() - pin user pages in memory

1646

* __get_user_pages() - pin user pages in memory

1647

* @tsk: task_struct of target task

1647

* @tsk: task_struct of target task

1648

* @mm: mm_struct of target mm

1648

* @mm: mm_struct of target mm

1649

* @start: starting user address

1649

* @start: starting user address

1650

* @nr_pages: number of pages from start to pin

1650

* @nr_pages: number of pages from start to pin

1651

* @gup_flags: flags modifying pin behaviour

1651

* @gup_flags: flags modifying pin behaviour

1652

* @pages: array that receives pointers to the pages pinned.

1652

* @pages: array that receives pointers to the pages pinned.

1653

* Should be at least nr_pages long. Or NULL, if caller

1653

* Should be at least nr_pages long. Or NULL, if caller

1654

* only intends to ensure the pages are faulted in.

1654

* only intends to ensure the pages are faulted in.

1655

* @vmas: array of pointers to vmas corresponding to each page.

1655

* @vmas: array of pointers to vmas corresponding to each page.

1656

* Or NULL if the caller does not require them.

1656

* Or NULL if the caller does not require them.

1657

* @nonblocking: whether waiting for disk IO or mmap_sem contention

1657

* @nonblocking: whether waiting for disk IO or mmap_sem contention

1658

*

1658

*

1659

* Returns number of pages pinned. This may be fewer than the number

1659

* Returns number of pages pinned. This may be fewer than the number

1660

* requested. If nr_pages is 0 or negative, returns 0. If no pages

1660

* requested. If nr_pages is 0 or negative, returns 0. If no pages

1661

* were pinned, returns -errno. Each page returned must be released

1661

* were pinned, returns -errno. Each page returned must be released

1662

* with a put_page() call when it is finished with. vmas will only

1662

* with a put_page() call when it is finished with. vmas will only

1663

* remain valid while mmap_sem is held.

1663

* remain valid while mmap_sem is held.

1664

*

1664

*

1665

* Must be called with mmap_sem held for read or write.

1665

* Must be called with mmap_sem held for read or write.

1666

*

1666

*

1667

* __get_user_pages walks a process's page tables and takes a reference to

1667

* __get_user_pages walks a process's page tables and takes a reference to

1668

* each struct page that each user address corresponds to at a given

1668

* each struct page that each user address corresponds to at a given

1669

* instant. That is, it takes the page that would be accessed if a user

1669

* instant. That is, it takes the page that would be accessed if a user

1670

* thread accesses the given user virtual address at that instant.

1670

* thread accesses the given user virtual address at that instant.

1671

*

1671

*

1672

* This does not guarantee that the page exists in the user mappings when

1672

* This does not guarantee that the page exists in the user mappings when

1673

* __get_user_pages returns, and there may even be a completely different

1673

* __get_user_pages returns, and there may even be a completely different

1674

* page there in some cases (eg. if mmapped pagecache has been invalidated

1674

* page there in some cases (eg. if mmapped pagecache has been invalidated

1675

* and subsequently re faulted). However it does guarantee that the page

1675

* and subsequently re faulted). However it does guarantee that the page

1676

* won't be freed completely. And mostly callers simply care that the page

1676

* won't be freed completely. And mostly callers simply care that the page

1677

* contains data that was valid *at some point in time*. Typically, an IO

1677

* contains data that was valid *at some point in time*. Typically, an IO

1678

* or similar operation cannot guarantee anything stronger anyway because

1678

* or similar operation cannot guarantee anything stronger anyway because

1679

* locks can't be held over the syscall boundary.

1679

* locks can't be held over the syscall boundary.

1680

*

1680

*

1681

* If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If

1681

* If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If

1682

* the page is written to, set_page_dirty (or set_page_dirty_lock, as

1682

* the page is written to, set_page_dirty (or set_page_dirty_lock, as

1683

* appropriate) must be called after the page is finished with, and

1683

* appropriate) must be called after the page is finished with, and

1684

* before put_page is called.

1684

* before put_page is called.

1685

*

1685

*

1686

* If @nonblocking != NULL, __get_user_pages will not wait for disk IO

1686

* If @nonblocking != NULL, __get_user_pages will not wait for disk IO

1687

* or mmap_sem contention, and if waiting is needed to pin all pages,

1687

* or mmap_sem contention, and if waiting is needed to pin all pages,

1688

* *@nonblocking will be set to 0.

1688

* *@nonblocking will be set to 0.

1689

*

1689

*

1690

* In most cases, get_user_pages or get_user_pages_fast should be used

1690

* In most cases, get_user_pages or get_user_pages_fast should be used

1691

* instead of __get_user_pages. __get_user_pages should be used only if

1691

* instead of __get_user_pages. __get_user_pages should be used only if

1692

* you need some special @gup_flags.

1692

* you need some special @gup_flags.

1693

*/

1693

*/

1694

long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,

1694

long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,

1695

unsigned long start, unsigned long nr_pages,

1695

unsigned long start, unsigned long nr_pages,

1696

unsigned int gup_flags, struct page **pages,

1696

unsigned int gup_flags, struct page **pages,

1697

struct vm_area_struct **vmas, int *nonblocking)

1697

struct vm_area_struct **vmas, int *nonblocking)

1698

{

1698

{

1699

long i;

1699

long i;

1700

unsigned long vm_flags;

1700

unsigned long vm_flags;

1701

unsigned int page_mask;

1701

unsigned int page_mask;

1702

1703

if (!nr_pages)

1703

if (!nr_pages)

1704

return 0;

1704

return 0;

1705

1706

VM_BUG_ON(!!pages != !!(gup_flags & FOLL_GET));

1706

VM_BUG_ON(!!pages != !!(gup_flags & FOLL_GET));

1707

1708

/*

1708

/*

1709

* Require read or write permissions.

1709

* Require read or write permissions.

1710

* If FOLL_FORCE is set, we only require the "MAY" flags.

1710

* If FOLL_FORCE is set, we only require the "MAY" flags.

1711

*/

1711

*/

1712

vm_flags = (gup_flags & FOLL_WRITE) ?

1712

vm_flags = (gup_flags & FOLL_WRITE) ?

1713

(VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);

1713

(VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);

1714

vm_flags &= (gup_flags & FOLL_FORCE) ?

1714

vm_flags &= (gup_flags & FOLL_FORCE) ?

1715

(VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);

1715

(VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);

1716

1717

/*

1717

/*

1718

* If FOLL_FORCE and FOLL_NUMA are both set, handle_mm_fault

1718

* If FOLL_FORCE and FOLL_NUMA are both set, handle_mm_fault

1719

* would be called on PROT_NONE ranges. We must never invoke

1719

* would be called on PROT_NONE ranges. We must never invoke

1720

* handle_mm_fault on PROT_NONE ranges or the NUMA hinting

1720

* handle_mm_fault on PROT_NONE ranges or the NUMA hinting

1721

* page faults would unprotect the PROT_NONE ranges if

1721

* page faults would unprotect the PROT_NONE ranges if

1722

* _PAGE_NUMA and _PAGE_PROTNONE are sharing the same pte/pmd

1722

* _PAGE_NUMA and _PAGE_PROTNONE are sharing the same pte/pmd

1723

* bitflag. So to avoid that, don't set FOLL_NUMA if

1723

* bitflag. So to avoid that, don't set FOLL_NUMA if

1724

* FOLL_FORCE is set.

1724

* FOLL_FORCE is set.

1725

*/

1725

*/

1726

if (!(gup_flags & FOLL_FORCE))

1726

if (!(gup_flags & FOLL_FORCE))

1727

gup_flags |= FOLL_NUMA;

1727

gup_flags |= FOLL_NUMA;

1728

1729

i = 0;

1729

i = 0;

1730

1731

do {

1731

do {

1732

struct vm_area_struct *vma;

1732

struct vm_area_struct *vma;

1733

1734

vma = find_extend_vma(mm, start);

1734

vma = find_extend_vma(mm, start);

1735

if (!vma && in_gate_area(mm, start)) {

1735

if (!vma && in_gate_area(mm, start)) {

1736

unsigned long pg = start & PAGE_MASK;

1736

unsigned long pg = start & PAGE_MASK;

1737

pgd_t *pgd;

1737

pgd_t *pgd;

1738

pud_t *pud;

1738

pud_t *pud;

1739

pmd_t *pmd;

1739

pmd_t *pmd;

1740

pte_t *pte;

1740

pte_t *pte;

1741

1742

/* user gate pages are read-only */

1742

/* user gate pages are read-only */

1743

if (gup_flags & FOLL_WRITE)

1743

if (gup_flags & FOLL_WRITE)

1744

return i ? : -EFAULT;

1744

return i ? : -EFAULT;

1745

if (pg > TASK_SIZE)

1745

if (pg > TASK_SIZE)

1746

pgd = pgd_offset_k(pg);

1746

pgd = pgd_offset_k(pg);

1747

else

1747

else

1748

pgd = pgd_offset_gate(mm, pg);

1748

pgd = pgd_offset_gate(mm, pg);

1749

BUG_ON(pgd_none(*pgd));

1749

BUG_ON(pgd_none(*pgd));

1750

pud = pud_offset(pgd, pg);

1750

pud = pud_offset(pgd, pg);

1751

BUG_ON(pud_none(*pud));

1751

BUG_ON(pud_none(*pud));

1752

pmd = pmd_offset(pud, pg);

1752

pmd = pmd_offset(pud, pg);

1753

if (pmd_none(*pmd))

1753

if (pmd_none(*pmd))

1754

return i ? : -EFAULT;

1754

return i ? : -EFAULT;

1755

VM_BUG_ON(pmd_trans_huge(*pmd));

1755

VM_BUG_ON(pmd_trans_huge(*pmd));

1756

pte = pte_offset_map(pmd, pg);

1756

pte = pte_offset_map(pmd, pg);

1757

if (pte_none(*pte)) {

1757

if (pte_none(*pte)) {

1758

pte_unmap(pte);

1758

pte_unmap(pte);

1759

return i ? : -EFAULT;

1759

return i ? : -EFAULT;

1760

}

1760

}

1761

vma = get_gate_vma(mm);

1761

vma = get_gate_vma(mm);

1762

if (pages) {

1762

if (pages) {

1763

struct page *page;

1763

struct page *page;

1764

1765

page = vm_normal_page(vma, start, *pte);

1765

page = vm_normal_page(vma, start, *pte);

1766

if (!page) {

1766

if (!page) {

1767

if (!(gup_flags & FOLL_DUMP) &&

1767

if (!(gup_flags & FOLL_DUMP) &&

1768

is_zero_pfn(pte_pfn(*pte)))

1768

is_zero_pfn(pte_pfn(*pte)))

1769

page = pte_page(*pte);

1769

page = pte_page(*pte);

1770

else {

1770

else {

1771

pte_unmap(pte);

1771

pte_unmap(pte);

1772

return i ? : -EFAULT;

1772

return i ? : -EFAULT;

1773

}

1773

}

1774

}

1774

}

1775

pages[i] = page;

1775

pages[i] = page;

1776

get_page(page);

1776

get_page(page);

1777

}

1777

}

1778

pte_unmap(pte);

1778

pte_unmap(pte);

1779

page_mask = 0;

1779

page_mask = 0;

1780

goto next_page;

1780

goto next_page;

1781

}

1781

}

1782

1783

if (!vma ||

1783

if (!vma ||

1784

(vma->vm_flags & (VM_IO | VM_PFNMAP)) ||

1784

(vma->vm_flags & (VM_IO | VM_PFNMAP)) ||

1785

!(vm_flags & vma->vm_flags))

1785

!(vm_flags & vma->vm_flags))

1786

return i ? : -EFAULT;

1786

return i ? : -EFAULT;

1787

1788

if (is_vm_hugetlb_page(vma)) {

1788

if (is_vm_hugetlb_page(vma)) {

1789

i = follow_hugetlb_page(mm, vma, pages, vmas,

1789

i = follow_hugetlb_page(mm, vma, pages, vmas,

1790

&start, &nr_pages, i, gup_flags);

1790

&start, &nr_pages, i, gup_flags);

1791

continue;

1791

continue;

1792

}

1792

}

1793

1794

do {

1794

do {

1795

struct page *page;

1795

struct page *page;

1796

unsigned int foll_flags = gup_flags;

1796

unsigned int foll_flags = gup_flags;

1797

unsigned int page_increm;

1797

unsigned int page_increm;

1798

1799

/*

1799

/*

1800

* If we have a pending SIGKILL, don't keep faulting

1800

* If we have a pending SIGKILL, don't keep faulting

1801

* pages and potentially allocating memory.

1801

* pages and potentially allocating memory.

1802

*/

1802

*/

1803

if (unlikely(fatal_signal_pending(current)))

1803

if (unlikely(fatal_signal_pending(current)))

1804

return i ? i : -ERESTARTSYS;

1804

return i ? i : -ERESTARTSYS;

1805

1806

cond_resched();

1806

cond_resched();

1807

while (!(page = follow_page_mask(vma, start,

1807

while (!(page = follow_page_mask(vma, start,

1808

foll_flags, &page_mask))) {

1808

foll_flags, &page_mask))) {

1809

int ret;

1809

int ret;

1810

unsigned int fault_flags = 0;

1810

unsigned int fault_flags = 0;

1811

1812

/* For mlock, just skip the stack guard page. */

1812

/* For mlock, just skip the stack guard page. */

1813

if (foll_flags & FOLL_MLOCK) {

1813

if (foll_flags & FOLL_MLOCK) {

1814

if (stack_guard_page(vma, start))

1814

if (stack_guard_page(vma, start))

1815

goto next_page;

1815

goto next_page;

1816

}

1816

}

1817

if (foll_flags & FOLL_WRITE)

1817

if (foll_flags & FOLL_WRITE)

1818

fault_flags |= FAULT_FLAG_WRITE;

1818

fault_flags |= FAULT_FLAG_WRITE;

1819

if (nonblocking)

1819

if (nonblocking)

1820

fault_flags |= FAULT_FLAG_ALLOW_RETRY;

1820

fault_flags |= FAULT_FLAG_ALLOW_RETRY;

1821

if (foll_flags & FOLL_NOWAIT)

1821

if (foll_flags & FOLL_NOWAIT)

1822

fault_flags |= (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT);

1822

fault_flags |= (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT);

1823

1824

ret = handle_mm_fault(mm, vma, start,

1824

ret = handle_mm_fault(mm, vma, start,

1825

fault_flags);

1825

fault_flags);

1826

1827

if (ret & VM_FAULT_ERROR) {

1827

if (ret & VM_FAULT_ERROR) {

1828

if (ret & VM_FAULT_OOM)

1828

if (ret & VM_FAULT_OOM)

1829

return i ? i : -ENOMEM;

1829

return i ? i : -ENOMEM;

1830

if (ret & (VM_FAULT_HWPOISON |

1830

if (ret & (VM_FAULT_HWPOISON |

1831

VM_FAULT_HWPOISON_LARGE)) {

1831

VM_FAULT_HWPOISON_LARGE)) {

1832

if (i)

1832

if (i)

1833

return i;

1833

return i;

1834

else if (gup_flags & FOLL_HWPOISON)

1834

else if (gup_flags & FOLL_HWPOISON)

1835

return -EHWPOISON;

1835

return -EHWPOISON;

1836

else

1836

else

1837

return -EFAULT;

1837

return -EFAULT;

1838

}

1838

}

1839

if (ret & VM_FAULT_SIGBUS)

1839

if (ret & VM_FAULT_SIGBUS)

1840

return i ? i : -EFAULT;

1840

return i ? i : -EFAULT;

1841

BUG();

1841

BUG();

1842

}

1842

}

1843

1844

if (tsk) {

1844

if (tsk) {

1845

if (ret & VM_FAULT_MAJOR)

1845

if (ret & VM_FAULT_MAJOR)

1846

tsk->maj_flt++;

1846

tsk->maj_flt++;

1847

else

1847

else

1848

tsk->min_flt++;

1848

tsk->min_flt++;

1849

}

1849

}

1850

1851

if (ret & VM_FAULT_RETRY) {

1851

if (ret & VM_FAULT_RETRY) {

1852

if (nonblocking)

1852

if (nonblocking)

1853

*nonblocking = 0;

1853

*nonblocking = 0;

1854

return i;

1854

return i;

1855

}

1855

}

1856

1857

/*

1857

/*

1858

* The VM_FAULT_WRITE bit tells us that

1858

* The VM_FAULT_WRITE bit tells us that

1859

* do_wp_page has broken COW when necessary,

1859

* do_wp_page has broken COW when necessary,

1860

* even if maybe_mkwrite decided not to set

1860

* even if maybe_mkwrite decided not to set

1861

* pte_write. We can thus safely do subsequent

1861

* pte_write. We can thus safely do subsequent

1862

* page lookups as if they were reads. But only

1862

* page lookups as if they were reads. But only

1863

* do so when looping for pte_write is futile:

1863

* do so when looping for pte_write is futile:

1864

* in some cases userspace may also be wanting

1864

* in some cases userspace may also be wanting

1865

* to write to the gotten user page, which a

1865

* to write to the gotten user page, which a

1866

* read fault here might prevent (a readonly

1866

* read fault here might prevent (a readonly

1867

* page might get reCOWed by userspace write).

1867

* page might get reCOWed by userspace write).

1868

*/

1868

*/

1869

if ((ret & VM_FAULT_WRITE) &&

1869

if ((ret & VM_FAULT_WRITE) &&

1870

!(vma->vm_flags & VM_WRITE))

1870

!(vma->vm_flags & VM_WRITE))

1871

foll_flags &= ~FOLL_WRITE;

1871

foll_flags &= ~FOLL_WRITE;

1872

1873

cond_resched();

1873

cond_resched();

1874

}

1874

}

1875

if (IS_ERR(page))

1875

if (IS_ERR(page))

1876

return i ? i : PTR_ERR(page);

1876

return i ? i : PTR_ERR(page);

1877

if (pages) {

1877

if (pages) {

1878

pages[i] = page;

1878

pages[i] = page;

1879

1880

flush_anon_page(vma, page, start);

1880

flush_anon_page(vma, page, start);

1881

flush_dcache_page(page);

1881

flush_dcache_page(page);

1882

page_mask = 0;

1882

page_mask = 0;

1883

}

1883

}

1884

next_page:

1884

next_page:

1885

if (vmas) {

1885

if (vmas) {

1886

vmas[i] = vma;

1886

vmas[i] = vma;

1887

page_mask = 0;

1887

page_mask = 0;

1888

}

1888

}

1889

page_increm = 1 + (~(start >> PAGE_SHIFT) & page_mask);

1889

page_increm = 1 + (~(start >> PAGE_SHIFT) & page_mask);

1890

if (page_increm > nr_pages)

1890

if (page_increm > nr_pages)

1891

page_increm = nr_pages;

1891

page_increm = nr_pages;

1892

i += page_increm;

1892

i += page_increm;

1893

start += page_increm * PAGE_SIZE;

1893

start += page_increm * PAGE_SIZE;

1894

nr_pages -= page_increm;

1894

nr_pages -= page_increm;

1895

} while (nr_pages && start < vma->vm_end);

1895

} while (nr_pages && start < vma->vm_end);

1896

} while (nr_pages);

1896

} while (nr_pages);

1897

return i;

1897

return i;

1898

}

1898

}

1899

EXPORT_SYMBOL(__get_user_pages);

1899

EXPORT_SYMBOL(__get_user_pages);

1900

1901

/*

1901

/*

1902

* fixup_user_fault() - manually resolve a user page fault

1902

* fixup_user_fault() - manually resolve a user page fault

1903

* @tsk: the task_struct to use for page fault accounting, or

1903

* @tsk: the task_struct to use for page fault accounting, or

1904

* NULL if faults are not to be recorded.

1904

* NULL if faults are not to be recorded.

1905

* @mm: mm_struct of target mm

1905

* @mm: mm_struct of target mm

1906

* @address: user address

1906

* @address: user address

1907

* @fault_flags:flags to pass down to handle_mm_fault()

1907

* @fault_flags:flags to pass down to handle_mm_fault()

1908

*

1908

*

1909

* This is meant to be called in the specific scenario where for locking reasons

1909

* This is meant to be called in the specific scenario where for locking reasons

1910

* we try to access user memory in atomic context (within a pagefault_disable()

1910

* we try to access user memory in atomic context (within a pagefault_disable()

1911

* section), this returns -EFAULT, and we want to resolve the user fault before

1911

* section), this returns -EFAULT, and we want to resolve the user fault before

1912

* trying again.

1912

* trying again.

1913

*

1913

*

1914

* Typically this is meant to be used by the futex code.

1914

* Typically this is meant to be used by the futex code.

1915

*

1915

*

1916

* The main difference with get_user_pages() is that this function will

1916

* The main difference with get_user_pages() is that this function will

1917

* unconditionally call handle_mm_fault() which will in turn perform all the

1917

* unconditionally call handle_mm_fault() which will in turn perform all the

1918

* necessary SW fixup of the dirty and young bits in the PTE, while

1918

* necessary SW fixup of the dirty and young bits in the PTE, while

1919

* handle_mm_fault() only guarantees to update these in the struct page.

1919

* handle_mm_fault() only guarantees to update these in the struct page.

1920

*

1920

*

1921

* This is important for some architectures where those bits also gate the

1921

* This is important for some architectures where those bits also gate the

1922

* access permission to the page because they are maintained in software. On

1922

* access permission to the page because they are maintained in software. On

1923

* such architectures, gup() will not be enough to make a subsequent access

1923

* such architectures, gup() will not be enough to make a subsequent access

1924

* succeed.

1924

* succeed.

1925

*

1925

*

1926

* This should be called with the mm_sem held for read.

1926

* This should be called with the mm_sem held for read.

1927

*/

1927

*/

1928

int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm,

1928

int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm,

1929

unsigned long address, unsigned int fault_flags)

1929

unsigned long address, unsigned int fault_flags)

1930

{

1930

{

1931

struct vm_area_struct *vma;

1931

struct vm_area_struct *vma;

1932

vm_flags_t vm_flags;

1932

vm_flags_t vm_flags;

1933

int ret;

1933

int ret;

1934

1935

vma = find_extend_vma(mm, address);

1935

vma = find_extend_vma(mm, address);

1936

if (!vma || address < vma->vm_start)

1936

if (!vma || address < vma->vm_start)

1937

return -EFAULT;

1937

return -EFAULT;

1938

1939

vm_flags = (fault_flags & FAULT_FLAG_WRITE) ? VM_WRITE : VM_READ;

1939

vm_flags = (fault_flags & FAULT_FLAG_WRITE) ? VM_WRITE : VM_READ;

1940

if (!(vm_flags & vma->vm_flags))

1940

if (!(vm_flags & vma->vm_flags))

1941

return -EFAULT;

1941

return -EFAULT;

1942

1943

ret = handle_mm_fault(mm, vma, address, fault_flags);

1943

ret = handle_mm_fault(mm, vma, address, fault_flags);

1944

if (ret & VM_FAULT_ERROR) {

1944

if (ret & VM_FAULT_ERROR) {

1945

if (ret & VM_FAULT_OOM)

1945

if (ret & VM_FAULT_OOM)

1946

return -ENOMEM;

1946

return -ENOMEM;

1947

if (ret & (VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE))

1947

if (ret & (VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE))

1948

return -EHWPOISON;

1948

return -EHWPOISON;

1949

if (ret & VM_FAULT_SIGBUS)

1949

if (ret & VM_FAULT_SIGBUS)

1950

return -EFAULT;

1950

return -EFAULT;

1951

BUG();

1951

BUG();

1952

}

1952

}

1953

if (tsk) {

1953

if (tsk) {

1954

if (ret & VM_FAULT_MAJOR)

1954

if (ret & VM_FAULT_MAJOR)

1955

tsk->maj_flt++;

1955

tsk->maj_flt++;

1956

else

1956

else

1957

tsk->min_flt++;

1957

tsk->min_flt++;

1958

}

1958

}

1959

return 0;

1959

return 0;

1960

}

1960

}

1961

1962

/*

1962

/*

1963

* get_user_pages() - pin user pages in memory

1963

* get_user_pages() - pin user pages in memory

1964

* @tsk: the task_struct to use for page fault accounting, or

1964

* @tsk: the task_struct to use for page fault accounting, or

1965

* NULL if faults are not to be recorded.

1965

* NULL if faults are not to be recorded.

1966

* @mm: mm_struct of target mm

1966

* @mm: mm_struct of target mm

1967

* @start: starting user address

1967

* @start: starting user address

1968

* @nr_pages: number of pages from start to pin

1968

* @nr_pages: number of pages from start to pin

1969

* @write: whether pages will be written to by the caller

1969

* @write: whether pages will be written to by the caller

1970

* @force: whether to force write access even if user mapping is

1970

* @force: whether to force write access even if user mapping is

1971

* readonly. This will result in the page being COWed even

1971

* readonly. This will result in the page being COWed even

1972

* in MAP_SHARED mappings. You do not want this.

1972

* in MAP_SHARED mappings. You do not want this.

1973

* @pages: array that receives pointers to the pages pinned.

1973

* @pages: array that receives pointers to the pages pinned.

1974

* Should be at least nr_pages long. Or NULL, if caller

1974

* Should be at least nr_pages long. Or NULL, if caller

1975

* only intends to ensure the pages are faulted in.

1975

* only intends to ensure the pages are faulted in.

1976

* @vmas: array of pointers to vmas corresponding to each page.

1976

* @vmas: array of pointers to vmas corresponding to each page.

1977

* Or NULL if the caller does not require them.

1977

* Or NULL if the caller does not require them.

1978

*

1978

*

1979

* Returns number of pages pinned. This may be fewer than the number

1979

* Returns number of pages pinned. This may be fewer than the number

1980

* requested. If nr_pages is 0 or negative, returns 0. If no pages

1980

* requested. If nr_pages is 0 or negative, returns 0. If no pages

1981

* were pinned, returns -errno. Each page returned must be released

1981

* were pinned, returns -errno. Each page returned must be released

1982

* with a put_page() call when it is finished with. vmas will only

1982

* with a put_page() call when it is finished with. vmas will only

1983

* remain valid while mmap_sem is held.

1983

* remain valid while mmap_sem is held.

1984

*

1984

*

1985

* Must be called with mmap_sem held for read or write.

1985

* Must be called with mmap_sem held for read or write.

1986

*

1986

*

1987

* get_user_pages walks a process's page tables and takes a reference to

1987

* get_user_pages walks a process's page tables and takes a reference to

1988

* each struct page that each user address corresponds to at a given

1988

* each struct page that each user address corresponds to at a given

1989

* instant. That is, it takes the page that would be accessed if a user

1989

* instant. That is, it takes the page that would be accessed if a user

1990

* thread accesses the given user virtual address at that instant.

1990

* thread accesses the given user virtual address at that instant.

1991

*

1991

*

1992

* This does not guarantee that the page exists in the user mappings when

1992

* This does not guarantee that the page exists in the user mappings when

1993

* get_user_pages returns, and there may even be a completely different

1993

* get_user_pages returns, and there may even be a completely different

1994

* page there in some cases (eg. if mmapped pagecache has been invalidated

1994

* page there in some cases (eg. if mmapped pagecache has been invalidated

1995

* and subsequently re faulted). However it does guarantee that the page

1995

* and subsequently re faulted). However it does guarantee that the page

1996

* won't be freed completely. And mostly callers simply care that the page

1996

* won't be freed completely. And mostly callers simply care that the page

1997

* contains data that was valid *at some point in time*. Typically, an IO

1997

* contains data that was valid *at some point in time*. Typically, an IO

1998

* or similar operation cannot guarantee anything stronger anyway because

1998

* or similar operation cannot guarantee anything stronger anyway because

1999

* locks can't be held over the syscall boundary.

1999

* locks can't be held over the syscall boundary.

2000

*

2000

*

2001

* If write=0, the page must not be written to. If the page is written to,

2001

* If write=0, the page must not be written to. If the page is written to,

2002

* set_page_dirty (or set_page_dirty_lock, as appropriate) must be called

2002

* set_page_dirty (or set_page_dirty_lock, as appropriate) must be called

2003

* after the page is finished with, and before put_page is called.

2003

* after the page is finished with, and before put_page is called.

2004

*

2004

*

2005

* get_user_pages is typically used for fewer-copy IO operations, to get a

2005

* get_user_pages is typically used for fewer-copy IO operations, to get a

2006

* handle on the memory by some means other than accesses via the user virtual

2006

* handle on the memory by some means other than accesses via the user virtual

2007

* addresses. The pages may be submitted for DMA to devices or accessed via

2007

* addresses. The pages may be submitted for DMA to devices or accessed via

2008

* their kernel linear mapping (via the kmap APIs). Care should be taken to

2008

* their kernel linear mapping (via the kmap APIs). Care should be taken to

2009

* use the correct cache flushing APIs.

2009

* use the correct cache flushing APIs.

2010

*

2010

*

2011

* See also get_user_pages_fast, for performance critical applications.

2011

* See also get_user_pages_fast, for performance critical applications.

2012

*/

2012

*/

2013

long get_user_pages(struct task_struct *tsk, struct mm_struct *mm,

2013

long get_user_pages(struct task_struct *tsk, struct mm_struct *mm,

2014

unsigned long start, unsigned long nr_pages, int write,

2014

unsigned long start, unsigned long nr_pages, int write,

2015

int force, struct page **pages, struct vm_area_struct **vmas)

2015

int force, struct page **pages, struct vm_area_struct **vmas)

2016

{

2016

{

2017

int flags = FOLL_TOUCH;

2017

int flags = FOLL_TOUCH;

2018

2019

if (pages)

2019

if (pages)

2020

flags |= FOLL_GET;

2020

flags |= FOLL_GET;

2021

if (write)

2021

if (write)

2022

flags |= FOLL_WRITE;

2022

flags |= FOLL_WRITE;

2023

if (force)

2023

if (force)

2024

flags |= FOLL_FORCE;

2024

flags |= FOLL_FORCE;

2025

2026

return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,

2026

return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,

2027

NULL);

2027

NULL);

2028

}

2028

}

2029

EXPORT_SYMBOL(get_user_pages);

2029

EXPORT_SYMBOL(get_user_pages);

2030

2031

/**

2031

/**

2032

* get_dump_page() - pin user page in memory while writing it to core dump

2032

* get_dump_page() - pin user page in memory while writing it to core dump

2033

* @addr: user address

2033

* @addr: user address

2034

*

2034

*

2035

* Returns struct page pointer of user page pinned for dump,

2035

* Returns struct page pointer of user page pinned for dump,

2036

* to be freed afterwards by page_cache_release() or put_page().

2036

* to be freed afterwards by page_cache_release() or put_page().

2037

*

2037

*

2038

* Returns NULL on any kind of failure - a hole must then be inserted into

2038

* Returns NULL on any kind of failure - a hole must then be inserted into

2039

* the corefile, to preserve alignment with its headers; and also returns

2039

* the corefile, to preserve alignment with its headers; and also returns

2040

* NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found -

2040

* NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found -

2041

* allowing a hole to be left in the corefile to save diskspace.

2041

* allowing a hole to be left in the corefile to save diskspace.

2042

*

2042

*

2043

* Called without mmap_sem, but after all other threads have been killed.

2043

* Called without mmap_sem, but after all other threads have been killed.

2044

*/

2044

*/

2045

#ifdef CONFIG_ELF_CORE

2045

#ifdef CONFIG_ELF_CORE

2046

struct page *get_dump_page(unsigned long addr)

2046

struct page *get_dump_page(unsigned long addr)

2047

{

2047

{

2048

struct vm_area_struct *vma;

2048

struct vm_area_struct *vma;

2049

struct page *page;

2049

struct page *page;

2050

2051

if (__get_user_pages(current, current->mm, addr, 1,

2051

if (__get_user_pages(current, current->mm, addr, 1,

2052

FOLL_FORCE | FOLL_DUMP | FOLL_GET, &page, &vma,

2052

FOLL_FORCE | FOLL_DUMP | FOLL_GET, &page, &vma,

2053

NULL) < 1)

2053

NULL) < 1)

2054

return NULL;

2054

return NULL;

2055

flush_cache_page(vma, addr, page_to_pfn(page));

2055

flush_cache_page(vma, addr, page_to_pfn(page));

2056

return page;

2056

return page;

2057

}

2057

}

2058

#endif /* CONFIG_ELF_CORE */

2058

#endif /* CONFIG_ELF_CORE */

2059

2060

pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr,

2060

pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr,

2061

spinlock_t **ptl)

2061

spinlock_t **ptl)

2062

{

2062

{

2063

pgd_t * pgd = pgd_offset(mm, addr);

2063

pgd_t * pgd = pgd_offset(mm, addr);

2064

pud_t * pud = pud_alloc(mm, pgd, addr);

2064

pud_t * pud = pud_alloc(mm, pgd, addr);

2065

if (pud) {

2065

if (pud) {

2066

pmd_t * pmd = pmd_alloc(mm, pud, addr);

2066

pmd_t * pmd = pmd_alloc(mm, pud, addr);

2067

if (pmd) {

2067

if (pmd) {

2068

VM_BUG_ON(pmd_trans_huge(*pmd));

2068

VM_BUG_ON(pmd_trans_huge(*pmd));

2069

return pte_alloc_map_lock(mm, pmd, addr, ptl);

2069

return pte_alloc_map_lock(mm, pmd, addr, ptl);

2070

}

2070

}

2071

}

2071

}

2072

return NULL;

2072

return NULL;

2073

}

2073

}

2074

2075

/*

2075

/*

2076

* This is the old fallback for page remapping.

2076

* This is the old fallback for page remapping.

2077

*

2077

*

2078

* For historical reasons, it only allows reserved pages. Only

2078

* For historical reasons, it only allows reserved pages. Only

2079

* old drivers should use this, and they needed to mark their

2079

* old drivers should use this, and they needed to mark their

2080

* pages reserved for the old functions anyway.

2080

* pages reserved for the old functions anyway.

2081

*/

2081

*/

2082

static int insert_page(struct vm_area_struct *vma, unsigned long addr,

2082

static int insert_page(struct vm_area_struct *vma, unsigned long addr,

2083

struct page *page, pgprot_t prot)

2083

struct page *page, pgprot_t prot)

2084

{

2084

{

2085

struct mm_struct *mm = vma->vm_mm;

2085

struct mm_struct *mm = vma->vm_mm;

2086

int retval;

2086

int retval;

2087

pte_t *pte;

2087

pte_t *pte;

2088

spinlock_t *ptl;

2088

spinlock_t *ptl;

2089

2090

retval = -EINVAL;

2090

retval = -EINVAL;

2091

if (PageAnon(page))

2091

if (PageAnon(page))

2092

goto out;

2092

goto out;

2093

retval = -ENOMEM;

2093

retval = -ENOMEM;

2094

flush_dcache_page(page);

2094

flush_dcache_page(page);

2095

pte = get_locked_pte(mm, addr, &ptl);

2095

pte = get_locked_pte(mm, addr, &ptl);

2096

if (!pte)

2096

if (!pte)

2097

goto out;

2097

goto out;

2098

retval = -EBUSY;

2098

retval = -EBUSY;

2099

if (!pte_none(*pte))

2099

if (!pte_none(*pte))

2100

goto out_unlock;

2100

goto out_unlock;

2101

2102

/* Ok, finally just insert the thing.. */

2102

/* Ok, finally just insert the thing.. */

2103

get_page(page);

2103

get_page(page);

2104

inc_mm_counter_fast(mm, MM_FILEPAGES);

2104

inc_mm_counter_fast(mm, MM_FILEPAGES);

2105

page_add_file_rmap(page);

2105

page_add_file_rmap(page);

2106

set_pte_at(mm, addr, pte, mk_pte(page, prot));

2106

set_pte_at(mm, addr, pte, mk_pte(page, prot));

2107

2108

retval = 0;

2108

retval = 0;

2109

pte_unmap_unlock(pte, ptl);

2109

pte_unmap_unlock(pte, ptl);

2110

return retval;

2110

return retval;

2111

out_unlock:

2111

out_unlock:

2112

pte_unmap_unlock(pte, ptl);

2112

pte_unmap_unlock(pte, ptl);

2113

out:

2113

out:

2114

return retval;

2114

return retval;

2115

}

2115

}

2116

2117

/**

2117

/**

2118

* vm_insert_page - insert single page into user vma

2118

* vm_insert_page - insert single page into user vma

2119

* @vma: user vma to map to

2119

* @vma: user vma to map to

2120

* @addr: target user address of this page

2120

* @addr: target user address of this page

2121

* @page: source kernel page

2121

* @page: source kernel page

2122

*

2122

*

2123

* This allows drivers to insert individual pages they've allocated

2123

* This allows drivers to insert individual pages they've allocated

2124

* into a user vma.

2124

* into a user vma.

2125

*

2125

*

2126

* The page has to be a nice clean _individual_ kernel allocation.

2126

* The page has to be a nice clean _individual_ kernel allocation.

2127

* If you allocate a compound page, you need to have marked it as

2127

* If you allocate a compound page, you need to have marked it as

2128

* such (__GFP_COMP), or manually just split the page up yourself

2128

* such (__GFP_COMP), or manually just split the page up yourself

2129

* (see split_page()).

2129

* (see split_page()).

2130

*

2130

*

2131

* NOTE! Traditionally this was done with "remap_pfn_range()" which

2131

* NOTE! Traditionally this was done with "remap_pfn_range()" which

2132

* took an arbitrary page protection parameter. This doesn't allow

2132

* took an arbitrary page protection parameter. This doesn't allow

2133

* that. Your vma protection will have to be set up correctly, which

2133

* that. Your vma protection will have to be set up correctly, which

2134

* means that if you want a shared writable mapping, you'd better

2134

* means that if you want a shared writable mapping, you'd better

2135

* ask for a shared writable mapping!

2135

* ask for a shared writable mapping!

2136

*

2136

*

2137

* The page does not need to be reserved.

2137

* The page does not need to be reserved.

2138

*

2138

*

2139

* Usually this function is called from f_op->mmap() handler

2139

* Usually this function is called from f_op->mmap() handler

2140

* under mm->mmap_sem write-lock, so it can change vma->vm_flags.

2140

* under mm->mmap_sem write-lock, so it can change vma->vm_flags.

2141

* Caller must set VM_MIXEDMAP on vma if it wants to call this

2141

* Caller must set VM_MIXEDMAP on vma if it wants to call this

2142

* function from other places, for example from page-fault handler.

2142

* function from other places, for example from page-fault handler.

2143

*/

2143

*/

2144

int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,

2144

int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,

2145

struct page *page)

2145

struct page *page)

2146

{

2146

{

2147

if (addr < vma->vm_start || addr >= vma->vm_end)

2147

if (addr < vma->vm_start || addr >= vma->vm_end)

2148

return -EFAULT;

2148

return -EFAULT;

2149

if (!page_count(page))

2149

if (!page_count(page))

2150

return -EINVAL;

2150

return -EINVAL;

2151

if (!(vma->vm_flags & VM_MIXEDMAP)) {

2151

if (!(vma->vm_flags & VM_MIXEDMAP)) {

2152

BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem));

2152

BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem));

2153

BUG_ON(vma->vm_flags & VM_PFNMAP);

2153

BUG_ON(vma->vm_flags & VM_PFNMAP);

2154

vma->vm_flags |= VM_MIXEDMAP;

2154

vma->vm_flags |= VM_MIXEDMAP;

2155

}

2155

}

2156

return insert_page(vma, addr, page, vma->vm_page_prot);

2156

return insert_page(vma, addr, page, vma->vm_page_prot);

2157

}

2157

}

2158

EXPORT_SYMBOL(vm_insert_page);

2158

EXPORT_SYMBOL(vm_insert_page);

2159

2160

static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,

2160

static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,

2161

unsigned long pfn, pgprot_t prot)

2161

unsigned long pfn, pgprot_t prot)

2162

{

2162

{

2163

struct mm_struct *mm = vma->vm_mm;

2163

struct mm_struct *mm = vma->vm_mm;

2164

int retval;

2164

int retval;

2165

pte_t *pte, entry;

2165

pte_t *pte, entry;

2166

spinlock_t *ptl;

2166

spinlock_t *ptl;

2167

2168

retval = -ENOMEM;

2168

retval = -ENOMEM;

2169

pte = get_locked_pte(mm, addr, &ptl);

2169

pte = get_locked_pte(mm, addr, &ptl);

2170

if (!pte)

2170

if (!pte)

2171

goto out;

2171

goto out;

2172

retval = -EBUSY;

2172

retval = -EBUSY;

2173

if (!pte_none(*pte))

2173

if (!pte_none(*pte))

2174

goto out_unlock;

2174

goto out_unlock;

2175

2176

/* Ok, finally just insert the thing.. */

2176

/* Ok, finally just insert the thing.. */

2177

entry = pte_mkspecial(pfn_pte(pfn, prot));

2177

entry = pte_mkspecial(pfn_pte(pfn, prot));

2178

set_pte_at(mm, addr, pte, entry);

2178

set_pte_at(mm, addr, pte, entry);

2179

update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */

2179

update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */

2180

2181

retval = 0;

2181

retval = 0;

2182

out_unlock:

2182

out_unlock:

2183

pte_unmap_unlock(pte, ptl);

2183

pte_unmap_unlock(pte, ptl);

2184

out:

2184

out:

2185

return retval;

2185

return retval;

2186

}

2186

}

2187

2188

/**

2188

/**

2189

* vm_insert_pfn - insert single pfn into user vma

2189

* vm_insert_pfn - insert single pfn into user vma

2190

* @vma: user vma to map to

2190

* @vma: user vma to map to

2191

* @addr: target user address of this page

2191

* @addr: target user address of this page

2192

* @pfn: source kernel pfn

2192

* @pfn: source kernel pfn

2193

*

2193

*

2194

* Similar to vm_insert_page, this allows drivers to insert individual pages

2194

* Similar to vm_insert_page, this allows drivers to insert individual pages

2195

* they've allocated into a user vma. Same comments apply.

2195

* they've allocated into a user vma. Same comments apply.

2196

*

2196

*

2197

* This function should only be called from a vm_ops->fault handler, and

2197

* This function should only be called from a vm_ops->fault handler, and

2198

* in that case the handler should return NULL.

2198

* in that case the handler should return NULL.

2199

*

2199

*

2200

* vma cannot be a COW mapping.

2200

* vma cannot be a COW mapping.

2201

*

2201

*

2202

* As this is called only for pages that do not currently exist, we

2202

* As this is called only for pages that do not currently exist, we

2203

* do not need to flush old virtual caches or the TLB.

2203

* do not need to flush old virtual caches or the TLB.

2204

*/

2204

*/

2205

int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,

2205

int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,

2206

unsigned long pfn)

2206

unsigned long pfn)

2207

{

2207

{

2208

int ret;

2208

int ret;

2209

pgprot_t pgprot = vma->vm_page_prot;

2209

pgprot_t pgprot = vma->vm_page_prot;

2210

/*

2210

/*

2211

* Technically, architectures with pte_special can avoid all these

2211

* Technically, architectures with pte_special can avoid all these

2212

* restrictions (same for remap_pfn_range). However we would like

2212

* restrictions (same for remap_pfn_range). However we would like

2213

* consistency in testing and feature parity among all, so we should

2213

* consistency in testing and feature parity among all, so we should

2214

* try to keep these invariants in place for everybody.

2214

* try to keep these invariants in place for everybody.

2215

*/

2215

*/

2216

BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));

2216

BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));

2217

BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==

2217

BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==

2218

(VM_PFNMAP|VM_MIXEDMAP));

2218

(VM_PFNMAP|VM_MIXEDMAP));

2219

BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));

2219

BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));

2220

BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));

2220

BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));

2221

2222

if (addr < vma->vm_start || addr >= vma->vm_end)

2222

if (addr < vma->vm_start || addr >= vma->vm_end)

2223

return -EFAULT;

2223

return -EFAULT;

2224

if (track_pfn_insert(vma, &pgprot, pfn))

2224

if (track_pfn_insert(vma, &pgprot, pfn))

2225

return -EINVAL;

2225

return -EINVAL;

2226

2227

ret = insert_pfn(vma, addr, pfn, pgprot);

2227

ret = insert_pfn(vma, addr, pfn, pgprot);

2228

2229

return ret;

2229

return ret;

2230

}

2230

}

2231

EXPORT_SYMBOL(vm_insert_pfn);

2231

EXPORT_SYMBOL(vm_insert_pfn);

2232

2233

int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,

2233

int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,

2234

unsigned long pfn)

2234

unsigned long pfn)

2235

{

2235

{

2236

BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));

2236

BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));

2237

2238

if (addr < vma->vm_start || addr >= vma->vm_end)

2238

if (addr < vma->vm_start || addr >= vma->vm_end)

2239

return -EFAULT;

2239

return -EFAULT;

2240

2241

/*

2241

/*

2242

* If we don't have pte special, then we have to use the pfn_valid()

2242

* If we don't have pte special, then we have to use the pfn_valid()

2243

* based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*

2243

* based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*

2244

* refcount the page if pfn_valid is true (hence insert_page rather

2244

* refcount the page if pfn_valid is true (hence insert_page rather

2245

* than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP

2245

* than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP

2246

* without pte special, it would there be refcounted as a normal page.

2246

* without pte special, it would there be refcounted as a normal page.

2247

*/

2247

*/

2248

if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) {

2248

if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) {

2249

struct page *page;

2249

struct page *page;

2250

2251

page = pfn_to_page(pfn);

2251

page = pfn_to_page(pfn);

2252

return insert_page(vma, addr, page, vma->vm_page_prot);

2252

return insert_page(vma, addr, page, vma->vm_page_prot);

2253

}

2253

}

2254

return insert_pfn(vma, addr, pfn, vma->vm_page_prot);

2254

return insert_pfn(vma, addr, pfn, vma->vm_page_prot);

2255

}

2255

}

2256

EXPORT_SYMBOL(vm_insert_mixed);

2256

EXPORT_SYMBOL(vm_insert_mixed);

2257

2258

/*

2258

/*

2259

* maps a range of physical memory into the requested pages. the old

2259

* maps a range of physical memory into the requested pages. the old

2260

* mappings are removed. any references to nonexistent pages results

2260

* mappings are removed. any references to nonexistent pages results

2261

* in null mappings (currently treated as "copy-on-access")

2261

* in null mappings (currently treated as "copy-on-access")

2262

*/

2262

*/

2263

static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,

2263

static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,

2264

unsigned long addr, unsigned long end,

2264

unsigned long addr, unsigned long end,

2265

unsigned long pfn, pgprot_t prot)

2265

unsigned long pfn, pgprot_t prot)

2266

{

2266

{

2267

pte_t *pte;

2267

pte_t *pte;

2268

spinlock_t *ptl;

2268

spinlock_t *ptl;

2269

2270

pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);

2270

pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);

2271

if (!pte)

2271

if (!pte)

2272

return -ENOMEM;

2272

return -ENOMEM;

2273

arch_enter_lazy_mmu_mode();

2273

arch_enter_lazy_mmu_mode();

2274

do {

2274

do {

2275

BUG_ON(!pte_none(*pte));

2275

BUG_ON(!pte_none(*pte));

2276

set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));

2276

set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));

2277

pfn++;

2277

pfn++;

2278

} while (pte++, addr += PAGE_SIZE, addr != end);

2278

} while (pte++, addr += PAGE_SIZE, addr != end);

2279

arch_leave_lazy_mmu_mode();

2279

arch_leave_lazy_mmu_mode();

2280

pte_unmap_unlock(pte - 1, ptl);

2280

pte_unmap_unlock(pte - 1, ptl);

2281

return 0;

2281

return 0;

2282

}

2282

}

2283

2284

static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,

2284

static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,

2285

unsigned long addr, unsigned long end,

2285

unsigned long addr, unsigned long end,

2286

unsigned long pfn, pgprot_t prot)

2286

unsigned long pfn, pgprot_t prot)

2287

{

2287

{

2288

pmd_t *pmd;

2288

pmd_t *pmd;

2289

unsigned long next;

2289

unsigned long next;

2290

2291

pfn -= addr >> PAGE_SHIFT;

2291

pfn -= addr >> PAGE_SHIFT;

2292

pmd = pmd_alloc(mm, pud, addr);

2292

pmd = pmd_alloc(mm, pud, addr);

2293

if (!pmd)

2293

if (!pmd)

2294

return -ENOMEM;

2294

return -ENOMEM;

2295

VM_BUG_ON(pmd_trans_huge(*pmd));

2295

VM_BUG_ON(pmd_trans_huge(*pmd));

2296

do {

2296

do {

2297

next = pmd_addr_end(addr, end);

2297

next = pmd_addr_end(addr, end);

2298

if (remap_pte_range(mm, pmd, addr, next,

2298

if (remap_pte_range(mm, pmd, addr, next,

2299

pfn + (addr >> PAGE_SHIFT), prot))

2299

pfn + (addr >> PAGE_SHIFT), prot))

2300

return -ENOMEM;

2300

return -ENOMEM;

2301

} while (pmd++, addr = next, addr != end);

2301

} while (pmd++, addr = next, addr != end);

2302

return 0;

2302

return 0;

2303

}

2303

}

2304

2305

static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,

2305

static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,

2306

unsigned long addr, unsigned long end,

2306

unsigned long addr, unsigned long end,

2307

unsigned long pfn, pgprot_t prot)

2307

unsigned long pfn, pgprot_t prot)

2308

{

2308

{

2309

pud_t *pud;

2309

pud_t *pud;

2310

unsigned long next;

2310

unsigned long next;

2311

2312

pfn -= addr >> PAGE_SHIFT;

2312

pfn -= addr >> PAGE_SHIFT;

2313

pud = pud_alloc(mm, pgd, addr);

2313

pud = pud_alloc(mm, pgd, addr);

2314

if (!pud)

2314

if (!pud)

2315

return -ENOMEM;

2315

return -ENOMEM;

2316

do {

2316

do {

2317

next = pud_addr_end(addr, end);

2317

next = pud_addr_end(addr, end);

2318

if (remap_pmd_range(mm, pud, addr, next,

2318

if (remap_pmd_range(mm, pud, addr, next,

2319

pfn + (addr >> PAGE_SHIFT), prot))

2319

pfn + (addr >> PAGE_SHIFT), prot))

2320

return -ENOMEM;

2320

return -ENOMEM;

2321

} while (pud++, addr = next, addr != end);

2321

} while (pud++, addr = next, addr != end);

2322

return 0;

2322

return 0;

2323

}

2323

}

2324

2325

/**

2325

/**

2326

* remap_pfn_range - remap kernel memory to userspace

2326

* remap_pfn_range - remap kernel memory to userspace

2327

* @vma: user vma to map to

2327

* @vma: user vma to map to

2328

* @addr: target user address to start at

2328

* @addr: target user address to start at

2329

* @pfn: physical address of kernel memory

2329

* @pfn: physical address of kernel memory

2330

* @size: size of map area

2330

* @size: size of map area

2331

* @prot: page protection flags for this mapping

2331

* @prot: page protection flags for this mapping

2332

*

2332

*

2333

* Note: this is only safe if the mm semaphore is held when called.

2333

* Note: this is only safe if the mm semaphore is held when called.

2334

*/

2334

*/

2335

int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,

2335

int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,

2336

unsigned long pfn, unsigned long size, pgprot_t prot)

2336

unsigned long pfn, unsigned long size, pgprot_t prot)

2337

{

2337

{

2338

pgd_t *pgd;

2338

pgd_t *pgd;

2339

unsigned long next;

2339

unsigned long next;

2340

unsigned long end = addr + PAGE_ALIGN(size);

2340

unsigned long end = addr + PAGE_ALIGN(size);

2341

struct mm_struct *mm = vma->vm_mm;

2341

struct mm_struct *mm = vma->vm_mm;

2342

int err;

2342

int err;

2343

2344

/*

2344

/*

2345

* Physically remapped pages are special. Tell the

2345

* Physically remapped pages are special. Tell the

2346

* rest of the world about it:

2346

* rest of the world about it:

2347

* VM_IO tells people not to look at these pages

2347

* VM_IO tells people not to look at these pages

2348

* (accesses can have side effects).

2348

* (accesses can have side effects).

2349

* VM_PFNMAP tells the core MM that the base pages are just

2349

* VM_PFNMAP tells the core MM that the base pages are just

2350

* raw PFN mappings, and do not have a "struct page" associated

2350

* raw PFN mappings, and do not have a "struct page" associated

2351

* with them.

2351

* with them.

2352

* VM_DONTEXPAND

2352

* VM_DONTEXPAND

2353

* Disable vma merging and expanding with mremap().

2353

* Disable vma merging and expanding with mremap().

2354

* VM_DONTDUMP

2354

* VM_DONTDUMP

2355

* Omit vma from core dump, even when VM_IO turned off.

2355

* Omit vma from core dump, even when VM_IO turned off.

2356

*

2356

*

2357

* There's a horrible special case to handle copy-on-write

2357

* There's a horrible special case to handle copy-on-write

2358

* behaviour that some programs depend on. We mark the "original"

2358

* behaviour that some programs depend on. We mark the "original"

2359

* un-COW'ed pages by matching them up with "vma->vm_pgoff".

2359

* un-COW'ed pages by matching them up with "vma->vm_pgoff".

2360

* See vm_normal_page() for details.

2360

* See vm_normal_page() for details.

2361

*/

2361

*/

2362

if (is_cow_mapping(vma->vm_flags)) {

2362

if (is_cow_mapping(vma->vm_flags)) {

2363

if (addr != vma->vm_start || end != vma->vm_end)

2363

if (addr != vma->vm_start || end != vma->vm_end)

2364

return -EINVAL;

2364

return -EINVAL;

2365

vma->vm_pgoff = pfn;

2365

vma->vm_pgoff = pfn;

2366

}

2366

}

2367

2368

err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size));

2368

err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size));

2369

if (err)

2369

if (err)

2370

return -EINVAL;

2370

return -EINVAL;

2371

2372

vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;

2372

vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;

2373

2374

BUG_ON(addr >= end);

2374

BUG_ON(addr >= end);

2375

pfn -= addr >> PAGE_SHIFT;

2375

pfn -= addr >> PAGE_SHIFT;

2376

pgd = pgd_offset(mm, addr);

2376

pgd = pgd_offset(mm, addr);

2377

flush_cache_range(vma, addr, end);

2377

flush_cache_range(vma, addr, end);

2378

do {

2378

do {

2379

next = pgd_addr_end(addr, end);

2379

next = pgd_addr_end(addr, end);

2380

err = remap_pud_range(mm, pgd, addr, next,

2380

err = remap_pud_range(mm, pgd, addr, next,

2381

pfn + (addr >> PAGE_SHIFT), prot);

2381

pfn + (addr >> PAGE_SHIFT), prot);

2382

if (err)

2382

if (err)

2383

break;

2383

break;

2384

} while (pgd++, addr = next, addr != end);

2384

} while (pgd++, addr = next, addr != end);

2385

2386

if (err)

2386

if (err)

2387

untrack_pfn(vma, pfn, PAGE_ALIGN(size));

2387

untrack_pfn(vma, pfn, PAGE_ALIGN(size));

2388

2389

return err;

2389

return err;

2390

}

2390

}

2391

EXPORT_SYMBOL(remap_pfn_range);

2391

EXPORT_SYMBOL(remap_pfn_range);

2392

2393

/**

2393

/**

2394

* vm_iomap_memory - remap memory to userspace

2394

* vm_iomap_memory - remap memory to userspace

2395

* @vma: user vma to map to

2395

* @vma: user vma to map to

2396

* @start: start of area

2396

* @start: start of area

2397

* @len: size of area

2397

* @len: size of area

2398

*

2398

*

2399

* This is a simplified io_remap_pfn_range() for common driver use. The

2399

* This is a simplified io_remap_pfn_range() for common driver use. The

2400

* driver just needs to give us the physical memory range to be mapped,

2400

* driver just needs to give us the physical memory range to be mapped,

2401

* we'll figure out the rest from the vma information.

2401

* we'll figure out the rest from the vma information.

2402

*

2402

*

2403

* NOTE! Some drivers might want to tweak vma->vm_page_prot first to get

2403

* NOTE! Some drivers might want to tweak vma->vm_page_prot first to get

2404

* whatever write-combining details or similar.

2404

* whatever write-combining details or similar.

2405

*/

2405

*/

2406

int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)

2406

int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)

2407

{

2407

{

2408

unsigned long vm_len, pfn, pages;

2408

unsigned long vm_len, pfn, pages;

2409

2410

/* Check that the physical memory area passed in looks valid */

2410

/* Check that the physical memory area passed in looks valid */

2411

if (start + len < start)

2411

if (start + len < start)

2412

return -EINVAL;

2412

return -EINVAL;

2413

/*

2413

/*

2414

* You *really* shouldn't map things that aren't page-aligned,

2414

* You *really* shouldn't map things that aren't page-aligned,

2415

* but we've historically allowed it because IO memory might

2415

* but we've historically allowed it because IO memory might

2416

* just have smaller alignment.

2416

* just have smaller alignment.

2417

*/

2417

*/

2418

len += start & ~PAGE_MASK;

2418

len += start & ~PAGE_MASK;

2419

pfn = start >> PAGE_SHIFT;

2419

pfn = start >> PAGE_SHIFT;

2420

pages = (len + ~PAGE_MASK) >> PAGE_SHIFT;

2420

pages = (len + ~PAGE_MASK) >> PAGE_SHIFT;

2421

if (pfn + pages < pfn)

2421

if (pfn + pages < pfn)

2422

return -EINVAL;

2422

return -EINVAL;

2423

2424

/* We start the mapping 'vm_pgoff' pages into the area */

2424

/* We start the mapping 'vm_pgoff' pages into the area */

2425

if (vma->vm_pgoff > pages)

2425

if (vma->vm_pgoff > pages)

2426

return -EINVAL;

2426

return -EINVAL;

2427

pfn += vma->vm_pgoff;

2427

pfn += vma->vm_pgoff;

2428

pages -= vma->vm_pgoff;

2428

pages -= vma->vm_pgoff;

2429

2430

/* Can we fit all of the mapping? */

2430

/* Can we fit all of the mapping? */

2431

vm_len = vma->vm_end - vma->vm_start;

2431

vm_len = vma->vm_end - vma->vm_start;

2432

if (vm_len >> PAGE_SHIFT > pages)

2432

if (vm_len >> PAGE_SHIFT > pages)

2433

return -EINVAL;

2433

return -EINVAL;

2434

2435

/* Ok, let it rip */

2435

/* Ok, let it rip */

2436

return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);

2436

return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);

2437

}

2437

}

2438

EXPORT_SYMBOL(vm_iomap_memory);

2438

EXPORT_SYMBOL(vm_iomap_memory);

2439

2440

static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,

2440

static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,

2441

unsigned long addr, unsigned long end,

2441

unsigned long addr, unsigned long end,

2442

pte_fn_t fn, void *data)

2442

pte_fn_t fn, void *data)

2443

{

2443

{

2444

pte_t *pte;

2444

pte_t *pte;

2445

int err;

2445

int err;

2446

pgtable_t token;

2446

pgtable_t token;

2447

spinlock_t *uninitialized_var(ptl);

2447

spinlock_t *uninitialized_var(ptl);

2448

2449

pte = (mm == &init_mm) ?

2449

pte = (mm == &init_mm) ?

2450

pte_alloc_kernel(pmd, addr) :

2450

pte_alloc_kernel(pmd, addr) :

2451

pte_alloc_map_lock(mm, pmd, addr, &ptl);

2451

pte_alloc_map_lock(mm, pmd, addr, &ptl);

2452

if (!pte)

2452

if (!pte)

2453

return -ENOMEM;

2453

return -ENOMEM;

2454

2455

BUG_ON(pmd_huge(*pmd));

2455

BUG_ON(pmd_huge(*pmd));

2456

2457

arch_enter_lazy_mmu_mode();

2457

arch_enter_lazy_mmu_mode();

2458

2459

token = pmd_pgtable(*pmd);

2459

token = pmd_pgtable(*pmd);

2460

2461

do {

2461

do {

2462

err = fn(pte++, token, addr, data);

2462

err = fn(pte++, token, addr, data);

2463

if (err)

2463

if (err)

2464

break;

2464

break;

2465

} while (addr += PAGE_SIZE, addr != end);

2465

} while (addr += PAGE_SIZE, addr != end);

2466

2467

arch_leave_lazy_mmu_mode();

2467

arch_leave_lazy_mmu_mode();

2468

2469

if (mm != &init_mm)

2469

if (mm != &init_mm)

2470

pte_unmap_unlock(pte-1, ptl);

2470

pte_unmap_unlock(pte-1, ptl);

2471

return err;

2471

return err;

2472

}

2472

}

2473

2474

static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,

2474

static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,

2475

unsigned long addr, unsigned long end,

2475

unsigned long addr, unsigned long end,

2476

pte_fn_t fn, void *data)

2476

pte_fn_t fn, void *data)

2477

{

2477

{

2478

pmd_t *pmd;

2478

pmd_t *pmd;

2479

unsigned long next;

2479

unsigned long next;

2480

int err;

2480

int err;

2481

2482

BUG_ON(pud_huge(*pud));

2482

BUG_ON(pud_huge(*pud));

2483

2484

pmd = pmd_alloc(mm, pud, addr);

2484

pmd = pmd_alloc(mm, pud, addr);

2485

if (!pmd)

2485

if (!pmd)

2486

return -ENOMEM;

2486

return -ENOMEM;

2487

do {

2487

do {

2488

next = pmd_addr_end(addr, end);

2488

next = pmd_addr_end(addr, end);

2489

err = apply_to_pte_range(mm, pmd, addr, next, fn, data);

2489

err = apply_to_pte_range(mm, pmd, addr, next, fn, data);

2490

if (err)

2490

if (err)

2491

break;

2491

break;

2492

} while (pmd++, addr = next, addr != end);

2492

} while (pmd++, addr = next, addr != end);

2493

return err;

2493

return err;

2494

}

2494

}

2495

2496

static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,

2496

static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,

2497

unsigned long addr, unsigned long end,

2497

unsigned long addr, unsigned long end,

2498

pte_fn_t fn, void *data)

2498

pte_fn_t fn, void *data)

2499

{

2499

{

2500

pud_t *pud;

2500

pud_t *pud;

2501

unsigned long next;

2501

unsigned long next;

2502

int err;

2502

int err;

2503

2504

pud = pud_alloc(mm, pgd, addr);

2504

pud = pud_alloc(mm, pgd, addr);

2505

if (!pud)

2505

if (!pud)

2506

return -ENOMEM;

2506

return -ENOMEM;

2507

do {

2507

do {

2508

next = pud_addr_end(addr, end);

2508

next = pud_addr_end(addr, end);

2509

err = apply_to_pmd_range(mm, pud, addr, next, fn, data);

2509

err = apply_to_pmd_range(mm, pud, addr, next, fn, data);

2510

if (err)

2510

if (err)

2511

break;

2511

break;

2512

} while (pud++, addr = next, addr != end);

2512

} while (pud++, addr = next, addr != end);

2513

return err;

2513

return err;

2514

}

2514

}

2515

2516

/*

2516

/*

2517

* Scan a region of virtual memory, filling in page tables as necessary

2517

* Scan a region of virtual memory, filling in page tables as necessary

2518

* and calling a provided function on each leaf page table.

2518

* and calling a provided function on each leaf page table.

2519

*/

2519

*/

2520

int apply_to_page_range(struct mm_struct *mm, unsigned long addr,

2520

int apply_to_page_range(struct mm_struct *mm, unsigned long addr,

2521

unsigned long size, pte_fn_t fn, void *data)

2521

unsigned long size, pte_fn_t fn, void *data)

2522

{

2522

{

2523

pgd_t *pgd;

2523

pgd_t *pgd;

2524

unsigned long next;

2524

unsigned long next;

2525

unsigned long end = addr + size;

2525

unsigned long end = addr + size;

2526

int err;

2526

int err;

2527

2528

BUG_ON(addr >= end);

2528

BUG_ON(addr >= end);

2529

pgd = pgd_offset(mm, addr);

2529

pgd = pgd_offset(mm, addr);

2530

do {

2530

do {

2531

next = pgd_addr_end(addr, end);

2531

next = pgd_addr_end(addr, end);

2532

err = apply_to_pud_range(mm, pgd, addr, next, fn, data);

2532

err = apply_to_pud_range(mm, pgd, addr, next, fn, data);

2533

if (err)

2533

if (err)

2534

break;

2534

break;

2535

} while (pgd++, addr = next, addr != end);

2535

} while (pgd++, addr = next, addr != end);

2536

2537

return err;

2537

return err;

2538

}

2538

}

2539

EXPORT_SYMBOL_GPL(apply_to_page_range);

2539

EXPORT_SYMBOL_GPL(apply_to_page_range);

2540

2541

/*

2541

/*

2542

* handle_pte_fault chooses page fault handler according to an entry

2542

* handle_pte_fault chooses page fault handler according to an entry

2543

* which was read non-atomically. Before making any commitment, on

2543

* which was read non-atomically. Before making any commitment, on

2544

* those architectures or configurations (e.g. i386 with PAE) which

2544

* those architectures or configurations (e.g. i386 with PAE) which

2545

* might give a mix of unmatched parts, do_swap_page and do_nonlinear_fault

2545

* might give a mix of unmatched parts, do_swap_page and do_nonlinear_fault

2546

* must check under lock before unmapping the pte and proceeding

2546

* must check under lock before unmapping the pte and proceeding

2547

* (but do_wp_page is only called after already making such a check;

2547

* (but do_wp_page is only called after already making such a check;

2548

* and do_anonymous_page can safely check later on).

2548

* and do_anonymous_page can safely check later on).

2549

*/

2549

*/

2550

static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,

2550

static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,

2551

pte_t *page_table, pte_t orig_pte)

2551

pte_t *page_table, pte_t orig_pte)

2552

{

2552

{

2553

int same = 1;

2553

int same = 1;

2554

#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)

2554

#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)

2555

if (sizeof(pte_t) > sizeof(unsigned long)) {

2555

if (sizeof(pte_t) > sizeof(unsigned long)) {

2556

spinlock_t *ptl = pte_lockptr(mm, pmd);

2556

spinlock_t *ptl = pte_lockptr(mm, pmd);

2557

spin_lock(ptl);

2557

spin_lock(ptl);

2558

same = pte_same(*page_table, orig_pte);

2558

same = pte_same(*page_table, orig_pte);

2559

spin_unlock(ptl);

2559

spin_unlock(ptl);

2560

}

2560

}

2561

#endif

2561

#endif

2562

pte_unmap(page_table);

2562

pte_unmap(page_table);

2563

return same;

2563

return same;

2564

}

2564

}

2565

2566

static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma)

2566

static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma)

2567

{

2567

{

2568

/*

2568

/*

2569

* If the source page was a PFN mapping, we don't have

2569

* If the source page was a PFN mapping, we don't have

2570

* a "struct page" for it. We do a best-effort copy by

2570

* a "struct page" for it. We do a best-effort copy by

2571

* just copying from the original user address. If that

2571

* just copying from the original user address. If that

2572

* fails, we just zero-fill it. Live with it.

2572

* fails, we just zero-fill it. Live with it.

2573

*/

2573

*/

2574

if (unlikely(!src)) {

2574

if (unlikely(!src)) {

2575

void *kaddr = kmap_atomic(dst);

2575

void *kaddr = kmap_atomic(dst);

2576

void __user *uaddr = (void __user *)(va & PAGE_MASK);

2576

void __user *uaddr = (void __user *)(va & PAGE_MASK);

2577

2578

/*

2578

/*

2579

* This really shouldn't fail, because the page is there

2579

* This really shouldn't fail, because the page is there

2580

* in the page tables. But it might just be unreadable,

2580

* in the page tables. But it might just be unreadable,

2581

* in which case we just give up and fill the result with

2581

* in which case we just give up and fill the result with

2582

* zeroes.

2582

* zeroes.

2583

*/

2583

*/

2584

if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))

2584

if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))

2585

clear_page(kaddr);

2585

clear_page(kaddr);

2586

kunmap_atomic(kaddr);

2586

kunmap_atomic(kaddr);

2587

flush_dcache_page(dst);

2587

flush_dcache_page(dst);

2588

} else

2588

} else

2589

copy_user_highpage(dst, src, va, vma);

2589

copy_user_highpage(dst, src, va, vma);

2590

}

2590

}

2591

2592

/*

2592

/*

2593

* This routine handles present pages, when users try to write

2593

* This routine handles present pages, when users try to write

2594

* to a shared page. It is done by copying the page to a new address

2594

* to a shared page. It is done by copying the page to a new address

2595

* and decrementing the shared-page counter for the old page.

2595

* and decrementing the shared-page counter for the old page.

2596

*

2596

*

2597

* Note that this routine assumes that the protection checks have been

2597

* Note that this routine assumes that the protection checks have been

2598

* done by the caller (the low-level page fault routine in most cases).

2598

* done by the caller (the low-level page fault routine in most cases).

2599

* Thus we can safely just mark it writable once we've done any necessary

2599

* Thus we can safely just mark it writable once we've done any necessary

2600

* COW.

2600

* COW.

2601

*

2601

*

2602

* We also mark the page dirty at this point even though the page will

2602

* We also mark the page dirty at this point even though the page will

2603

* change only once the write actually happens. This avoids a few races,

2603

* change only once the write actually happens. This avoids a few races,

2604

* and potentially makes it more efficient.

2604

* and potentially makes it more efficient.

2605

*

2605

*

2606

* We enter with non-exclusive mmap_sem (to exclude vma changes,

2606

* We enter with non-exclusive mmap_sem (to exclude vma changes,

2607

* but allow concurrent faults), with pte both mapped and locked.

2607

* but allow concurrent faults), with pte both mapped and locked.

2608

* We return with mmap_sem still held, but pte unmapped and unlocked.

2608

* We return with mmap_sem still held, but pte unmapped and unlocked.

2609

*/

2609

*/

2610

static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,

2610

static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,

2611

unsigned long address, pte_t *page_table, pmd_t *pmd,

2611

unsigned long address, pte_t *page_table, pmd_t *pmd,

2612

spinlock_t *ptl, pte_t orig_pte)

2612

spinlock_t *ptl, pte_t orig_pte)

2613

__releases(ptl)

2613

__releases(ptl)

2614

{

2614

{

2615

struct page *old_page, *new_page = NULL;

2615

struct page *old_page, *new_page = NULL;

2616

pte_t entry;

2616

pte_t entry;

2617

int ret = 0;

2617

int ret = 0;

2618

int page_mkwrite = 0;

2618

int page_mkwrite = 0;

2619

struct page *dirty_page = NULL;

2619

struct page *dirty_page = NULL;

2620

unsigned long mmun_start = 0; /* For mmu_notifiers */

2620

unsigned long mmun_start = 0; /* For mmu_notifiers */

2621

unsigned long mmun_end = 0; /* For mmu_notifiers */

2621

unsigned long mmun_end = 0; /* For mmu_notifiers */

2622

2623

old_page = vm_normal_page(vma, address, orig_pte);

2623

old_page = vm_normal_page(vma, address, orig_pte);

2624

if (!old_page) {

2624

if (!old_page) {

2625

/*

2625

/*

2626

* VM_MIXEDMAP !pfn_valid() case

2626

* VM_MIXEDMAP !pfn_valid() case

2627

*

2627

*

2628

* We should not cow pages in a shared writeable mapping.

2628

* We should not cow pages in a shared writeable mapping.

2629

* Just mark the pages writable as we can't do any dirty

2629

* Just mark the pages writable as we can't do any dirty

2630

* accounting on raw pfn maps.

2630

* accounting on raw pfn maps.

2631

*/

2631

*/

2632

if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==

2632

if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==

2633

(VM_WRITE|VM_SHARED))

2633

(VM_WRITE|VM_SHARED))

2634

goto reuse;

2634

goto reuse;

2635

goto gotten;

2635

goto gotten;

2636

}

2636

}

2637

2638

/*

2638

/*

2639

* Take out anonymous pages first, anonymous shared vmas are

2639

* Take out anonymous pages first, anonymous shared vmas are

2640

* not dirty accountable.

2640

* not dirty accountable.

2641

*/

2641

*/

2642

if (PageAnon(old_page) && !PageKsm(old_page)) {

2642

if (PageAnon(old_page) && !PageKsm(old_page)) {

2643

if (!trylock_page(old_page)) {

2643

if (!trylock_page(old_page)) {

2644

page_cache_get(old_page);

2644

page_cache_get(old_page);

2645

pte_unmap_unlock(page_table, ptl);

2645

pte_unmap_unlock(page_table, ptl);

2646

lock_page(old_page);

2646

lock_page(old_page);

2647

page_table = pte_offset_map_lock(mm, pmd, address,

2647

page_table = pte_offset_map_lock(mm, pmd, address,

2648

&ptl);

2648

&ptl);

2649

if (!pte_same(*page_table, orig_pte)) {

2649

if (!pte_same(*page_table, orig_pte)) {

2650

unlock_page(old_page);

2650

unlock_page(old_page);

2651

goto unlock;

2651

goto unlock;

2652

}

2652

}

2653

page_cache_release(old_page);

2653

page_cache_release(old_page);

2654

}

2654

}

2655

if (reuse_swap_page(old_page)) {

2655

if (reuse_swap_page(old_page)) {

2656

/*

2656

/*

2657

* The page is all ours. Move it to our anon_vma so

2657

* The page is all ours. Move it to our anon_vma so

2658

* the rmap code will not search our parent or siblings.

2658

* the rmap code will not search our parent or siblings.

2659

* Protected against the rmap code by the page lock.

2659

* Protected against the rmap code by the page lock.

2660

*/

2660

*/

2661

page_move_anon_rmap(old_page, vma, address);

2661

page_move_anon_rmap(old_page, vma, address);

2662

unlock_page(old_page);

2662

unlock_page(old_page);

2663

goto reuse;

2663

goto reuse;

2664

}

2664

}

2665

unlock_page(old_page);

2665

unlock_page(old_page);

2666

} else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==

2666

} else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==

2667

(VM_WRITE|VM_SHARED))) {

2667

(VM_WRITE|VM_SHARED))) {

2668

/*

2668

/*

2669

* Only catch write-faults on shared writable pages,

2669

* Only catch write-faults on shared writable pages,

2670

* read-only shared pages can get COWed by

2670

* read-only shared pages can get COWed by

2671

* get_user_pages(.write=1, .force=1).

2671

* get_user_pages(.write=1, .force=1).

2672

*/

2672

*/

2673

if (vma->vm_ops && vma->vm_ops->page_mkwrite) {

2673

if (vma->vm_ops && vma->vm_ops->page_mkwrite) {

2674

struct vm_fault vmf;

2674

struct vm_fault vmf;

2675

int tmp;

2675

int tmp;

2676

2677

vmf.virtual_address = (void __user *)(address &

2677

vmf.virtual_address = (void __user *)(address &

2678

PAGE_MASK);

2678

PAGE_MASK);

2679

vmf.pgoff = old_page->index;

2679

vmf.pgoff = old_page->index;

2680

vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;

2680

vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;

2681

vmf.page = old_page;

2681

vmf.page = old_page;

2682

2683

/*

2683

/*

2684

* Notify the address space that the page is about to

2684

* Notify the address space that the page is about to

2685

* become writable so that it can prohibit this or wait

2685

* become writable so that it can prohibit this or wait

2686

* for the page to get into an appropriate state.

2686

* for the page to get into an appropriate state.

2687

*

2687

*

2688

* We do this without the lock held, so that it can

2688

* We do this without the lock held, so that it can

2689

* sleep if it needs to.

2689

* sleep if it needs to.

2690

*/

2690

*/

2691

page_cache_get(old_page);

2691

page_cache_get(old_page);

2692

pte_unmap_unlock(page_table, ptl);

2692

pte_unmap_unlock(page_table, ptl);

2693

2694

tmp = vma->vm_ops->page_mkwrite(vma, &vmf);

2694

tmp = vma->vm_ops->page_mkwrite(vma, &vmf);

2695

if (unlikely(tmp &

2695

if (unlikely(tmp &

2696

(VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {

2696

(VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {

2697

ret = tmp;

2697

ret = tmp;

2698

goto unwritable_page;

2698

goto unwritable_page;

2699

}

2699

}

2700

if (unlikely(!(tmp & VM_FAULT_LOCKED))) {

2700

if (unlikely(!(tmp & VM_FAULT_LOCKED))) {

2701

lock_page(old_page);

2701

lock_page(old_page);

2702

if (!old_page->mapping) {

2702

if (!old_page->mapping) {

2703

ret = 0; /* retry the fault */

2703

ret = 0; /* retry the fault */

2704

unlock_page(old_page);

2704

unlock_page(old_page);

2705

goto unwritable_page;

2705

goto unwritable_page;

2706

}

2706

}

2707

} else

2707

} else

2708

VM_BUG_ON(!PageLocked(old_page));

2708

VM_BUG_ON(!PageLocked(old_page));

2709

2710

/*

2710

/*

2711

* Since we dropped the lock we need to revalidate

2711

* Since we dropped the lock we need to revalidate

2712

* the PTE as someone else may have changed it. If

2712

* the PTE as someone else may have changed it. If

2713

* they did, we just return, as we can count on the

2713

* they did, we just return, as we can count on the

2714

* MMU to tell us if they didn't also make it writable.

2714

* MMU to tell us if they didn't also make it writable.

2715

*/

2715

*/

2716

page_table = pte_offset_map_lock(mm, pmd, address,

2716

page_table = pte_offset_map_lock(mm, pmd, address,

2717

&ptl);

2717

&ptl);

2718

if (!pte_same(*page_table, orig_pte)) {

2718

if (!pte_same(*page_table, orig_pte)) {

2719

unlock_page(old_page);

2719

unlock_page(old_page);

2720

goto unlock;

2720

goto unlock;

2721

}

2721

}

2722

2723

page_mkwrite = 1;

2723

page_mkwrite = 1;

2724

}

2724

}

2725

dirty_page = old_page;

2725

dirty_page = old_page;

2726

get_page(dirty_page);

2726

get_page(dirty_page);

2727

2728

reuse:

2728

reuse:

2729

flush_cache_page(vma, address, pte_pfn(orig_pte));

2729

flush_cache_page(vma, address, pte_pfn(orig_pte));

2730

entry = pte_mkyoung(orig_pte);

2730

entry = pte_mkyoung(orig_pte);

2731

entry = maybe_mkwrite(pte_mkdirty(entry), vma);

2731

entry = maybe_mkwrite(pte_mkdirty(entry), vma);

2732

if (ptep_set_access_flags(vma, address, page_table, entry,1))

2732

if (ptep_set_access_flags(vma, address, page_table, entry,1))

2733

update_mmu_cache(vma, address, page_table);

2733

update_mmu_cache(vma, address, page_table);

2734

pte_unmap_unlock(page_table, ptl);

2734

pte_unmap_unlock(page_table, ptl);

2735

ret |= VM_FAULT_WRITE;

2735

ret |= VM_FAULT_WRITE;

2736

2737

if (!dirty_page)

2737

if (!dirty_page)

2738

return ret;

2738

return ret;

2739

2740

/*

2740

/*

2741

* Yes, Virginia, this is actually required to prevent a race

2741

* Yes, Virginia, this is actually required to prevent a race

2742

* with clear_page_dirty_for_io() from clearing the page dirty

2742

* with clear_page_dirty_for_io() from clearing the page dirty

2743

* bit after it clear all dirty ptes, but before a racing

2743

* bit after it clear all dirty ptes, but before a racing

2744

* do_wp_page installs a dirty pte.

2744

* do_wp_page installs a dirty pte.

2745

*

2745

*

2746

* __do_fault is protected similarly.

2746

* __do_fault is protected similarly.

2747

*/

2747

*/

2748

if (!page_mkwrite) {

2748

if (!page_mkwrite) {

2749

wait_on_page_locked(dirty_page);

2749

wait_on_page_locked(dirty_page);

2750

set_page_dirty_balance(dirty_page, page_mkwrite);

2750

set_page_dirty_balance(dirty_page, page_mkwrite);

2751

/* file_update_time outside page_lock */

2751

/* file_update_time outside page_lock */

2752

if (vma->vm_file)

2752

if (vma->vm_file)

2753

file_update_time(vma->vm_file);

2753

file_update_time(vma->vm_file);

2754

}

2754

}

2755

put_page(dirty_page);

2755

put_page(dirty_page);

2756

if (page_mkwrite) {

2756

if (page_mkwrite) {

2757

struct address_space *mapping = dirty_page->mapping;

2757

struct address_space *mapping = dirty_page->mapping;

2758

2759

set_page_dirty(dirty_page);

2759

set_page_dirty(dirty_page);

2760

unlock_page(dirty_page);

2760

unlock_page(dirty_page);

2761

page_cache_release(dirty_page);

2761

page_cache_release(dirty_page);

2762

if (mapping) {

2762

if (mapping) {

2763

/*

2763

/*

2764

* Some device drivers do not set page.mapping

2764

* Some device drivers do not set page.mapping

2765

* but still dirty their pages

2765

* but still dirty their pages

2766

*/

2766

*/

2767

balance_dirty_pages_ratelimited(mapping);

2767

balance_dirty_pages_ratelimited(mapping);

2768

}

2768

}

2769

}

2769

}

2770

2771

return ret;

2771

return ret;

2772

}

2772

}

2773

2774

/*

2774

/*

2775

* Ok, we need to copy. Oh, well..

2775

* Ok, we need to copy. Oh, well..

2776

*/

2776

*/

2777

page_cache_get(old_page);

2777

page_cache_get(old_page);

2778

gotten:

2778

gotten:

2779

pte_unmap_unlock(page_table, ptl);

2779

pte_unmap_unlock(page_table, ptl);

2780

2781

if (unlikely(anon_vma_prepare(vma)))

2781

if (unlikely(anon_vma_prepare(vma)))

2782

goto oom;

2782

goto oom;

2783

2784

if (is_zero_pfn(pte_pfn(orig_pte))) {

2784

if (is_zero_pfn(pte_pfn(orig_pte))) {

2785

new_page = alloc_zeroed_user_highpage_movable(vma, address);

2785

new_page = alloc_zeroed_user_highpage_movable(vma, address);

2786

if (!new_page)

2786

if (!new_page)

2787

goto oom;

2787

goto oom;

2788

} else {

2788

} else {

2789

new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);

2789

new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);

2790

if (!new_page)

2790

if (!new_page)

2791

goto oom;

2791

goto oom;

2792

cow_user_page(new_page, old_page, address, vma);

2792

cow_user_page(new_page, old_page, address, vma);

2793

}

2793

}

2794

__SetPageUptodate(new_page);

2794

__SetPageUptodate(new_page);

2795

2796

if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL))

2796

if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL))

2797

goto oom_free_new;

2797

goto oom_free_new;

2798

2799

mmun_start = address & PAGE_MASK;

2799

mmun_start = address & PAGE_MASK;

2800

mmun_end = mmun_start + PAGE_SIZE;

2800

mmun_end = mmun_start + PAGE_SIZE;

2801

mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);

2801

mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);

2802

2803

/*

2803

/*

2804

* Re-check the pte - we dropped the lock

2804

* Re-check the pte - we dropped the lock

2805

*/

2805

*/

2806

page_table = pte_offset_map_lock(mm, pmd, address, &ptl);

2806

page_table = pte_offset_map_lock(mm, pmd, address, &ptl);

2807

if (likely(pte_same(*page_table, orig_pte))) {

2807

if (likely(pte_same(*page_table, orig_pte))) {

2808

if (old_page) {

2808

if (old_page) {

2809

if (!PageAnon(old_page)) {

2809

if (!PageAnon(old_page)) {

2810

dec_mm_counter_fast(mm, MM_FILEPAGES);

2810

dec_mm_counter_fast(mm, MM_FILEPAGES);

2811

inc_mm_counter_fast(mm, MM_ANONPAGES);

2811

inc_mm_counter_fast(mm, MM_ANONPAGES);

2812

}

2812

}

2813

} else

2813

} else

2814

inc_mm_counter_fast(mm, MM_ANONPAGES);

2814

inc_mm_counter_fast(mm, MM_ANONPAGES);

2815

flush_cache_page(vma, address, pte_pfn(orig_pte));

2815

flush_cache_page(vma, address, pte_pfn(orig_pte));

2816

entry = mk_pte(new_page, vma->vm_page_prot);

2816

entry = mk_pte(new_page, vma->vm_page_prot);

2817

entry = maybe_mkwrite(pte_mkdirty(entry), vma);

2817

entry = maybe_mkwrite(pte_mkdirty(entry), vma);

2818

/*

2818

/*

2819

* Clear the pte entry and flush it first, before updating the

2819

* Clear the pte entry and flush it first, before updating the

2820

* pte with the new entry. This will avoid a race condition

2820

* pte with the new entry. This will avoid a race condition

2821

* seen in the presence of one thread doing SMC and another

2821

* seen in the presence of one thread doing SMC and another

2822

* thread doing COW.

2822

* thread doing COW.

2823

*/

2823

*/

2824

ptep_clear_flush(vma, address, page_table);

2824

ptep_clear_flush(vma, address, page_table);

2825

page_add_new_anon_rmap(new_page, vma, address);

2825

page_add_new_anon_rmap(new_page, vma, address);

2826

/*

2826

/*

2827

* We call the notify macro here because, when using secondary

2827

* We call the notify macro here because, when using secondary

2828

* mmu page tables (such as kvm shadow page tables), we want the

2828

* mmu page tables (such as kvm shadow page tables), we want the

2829

* new page to be mapped directly into the secondary page table.

2829

* new page to be mapped directly into the secondary page table.

2830

*/

2830

*/

2831

set_pte_at_notify(mm, address, page_table, entry);

2831

set_pte_at_notify(mm, address, page_table, entry);

2832

update_mmu_cache(vma, address, page_table);

2832

update_mmu_cache(vma, address, page_table);

2833

if (old_page) {

2833

if (old_page) {

2834

/*

2834

/*

2835

* Only after switching the pte to the new page may

2835

* Only after switching the pte to the new page may

2836

* we remove the mapcount here. Otherwise another

2836

* we remove the mapcount here. Otherwise another

2837

* process may come and find the rmap count decremented

2837

* process may come and find the rmap count decremented

2838

* before the pte is switched to the new page, and

2838

* before the pte is switched to the new page, and

2839

* "reuse" the old page writing into it while our pte

2839

* "reuse" the old page writing into it while our pte

2840

* here still points into it and can be read by other

2840

* here still points into it and can be read by other

2841

* threads.

2841

* threads.

2842

*

2842

*

2843

* The critical issue is to order this

2843

* The critical issue is to order this

2844

* page_remove_rmap with the ptp_clear_flush above.

2844

* page_remove_rmap with the ptp_clear_flush above.

2845

* Those stores are ordered by (if nothing else,)

2845

* Those stores are ordered by (if nothing else,)

2846

* the barrier present in the atomic_add_negative

2846

* the barrier present in the atomic_add_negative

2847

* in page_remove_rmap.

2847

* in page_remove_rmap.

2848

*

2848

*

2849

* Then the TLB flush in ptep_clear_flush ensures that

2849

* Then the TLB flush in ptep_clear_flush ensures that

2850

* no process can access the old page before the

2850

* no process can access the old page before the

2851

* decremented mapcount is visible. And the old page

2851

* decremented mapcount is visible. And the old page

2852

* cannot be reused until after the decremented

2852

* cannot be reused until after the decremented

2853

* mapcount is visible. So transitively, TLBs to

2853

* mapcount is visible. So transitively, TLBs to

2854

* old page will be flushed before it can be reused.

2854

* old page will be flushed before it can be reused.

2855

*/

2855

*/

2856

page_remove_rmap(old_page);

2856

page_remove_rmap(old_page);

2857

}

2857

}

2858

2859

/* Free the old page.. */

2859

/* Free the old page.. */

2860

new_page = old_page;

2860

new_page = old_page;

2861

ret |= VM_FAULT_WRITE;

2861

ret |= VM_FAULT_WRITE;

2862

} else

2862

} else

2863

mem_cgroup_uncharge_page(new_page);

2863

mem_cgroup_uncharge_page(new_page);

2864

2865

if (new_page)

2865

if (new_page)

2866

page_cache_release(new_page);

2866

page_cache_release(new_page);

2867

unlock:

2867

unlock:

2868

pte_unmap_unlock(page_table, ptl);

2868

pte_unmap_unlock(page_table, ptl);

2869

if (mmun_end > mmun_start)

2869

if (mmun_end > mmun_start)

2870

mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);

2870

mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);

2871

if (old_page) {

2871

if (old_page) {

2872

/*

2872

/*

2873

* Don't let another task, with possibly unlocked vma,

2873

* Don't let another task, with possibly unlocked vma,

2874

* keep the mlocked page.

2874

* keep the mlocked page.

2875

*/

2875

*/

2876

if ((ret & VM_FAULT_WRITE) && (vma->vm_flags & VM_LOCKED)) {

2876

if ((ret & VM_FAULT_WRITE) && (vma->vm_flags & VM_LOCKED)) {

2877

lock_page(old_page); /* LRU manipulation */

2877

lock_page(old_page); /* LRU manipulation */

2878

munlock_vma_page(old_page);

2878

munlock_vma_page(old_page);

2879

unlock_page(old_page);

2879

unlock_page(old_page);

2880

}

2880

}

2881

page_cache_release(old_page);

2881

page_cache_release(old_page);

2882

}

2882

}

2883

return ret;

2883

return ret;

2884

oom_free_new:

2884

oom_free_new:

2885

page_cache_release(new_page);

2885

page_cache_release(new_page);

2886

oom:

2886

oom:

2887

if (old_page)

2887

if (old_page)

2888

page_cache_release(old_page);

2888

page_cache_release(old_page);

2889

return VM_FAULT_OOM;

2889

return VM_FAULT_OOM;

2890

2891

unwritable_page:

2891

unwritable_page:

2892

page_cache_release(old_page);

2892

page_cache_release(old_page);

2893

return ret;

2893

return ret;

2894

}

2894

}

2895

2896

static void unmap_mapping_range_vma(struct vm_area_struct *vma,

2896

static void unmap_mapping_range_vma(struct vm_area_struct *vma,

2897

unsigned long start_addr, unsigned long end_addr,

2897

unsigned long start_addr, unsigned long end_addr,

2898

struct zap_details *details)

2898

struct zap_details *details)

2899

{

2899

{

2900

zap_page_range_single(vma, start_addr, end_addr - start_addr, details);

2900

zap_page_range_single(vma, start_addr, end_addr - start_addr, details);

2901

}

2901

}

2902

2903

static inline void unmap_mapping_range_tree(struct rb_root *root,

2903

static inline void unmap_mapping_range_tree(struct rb_root *root,

2904

struct zap_details *details)

2904

struct zap_details *details)

2905

{

2905

{

2906

struct vm_area_struct *vma;

2906

struct vm_area_struct *vma;

2907

pgoff_t vba, vea, zba, zea;

2907

pgoff_t vba, vea, zba, zea;

2908

2909

vma_interval_tree_foreach(vma, root,

2909

vma_interval_tree_foreach(vma, root,

2910

details->first_index, details->last_index) {

2910

details->first_index, details->last_index) {

2911

2912

vba = vma->vm_pgoff;

2912

vba = vma->vm_pgoff;

2913

vea = vba + vma_pages(vma) - 1;

2913

vea = vba + vma_pages(vma) - 1;

2914

/* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */

2914

/* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */

2915

zba = details->first_index;

2915

zba = details->first_index;

2916

if (zba < vba)

2916

if (zba < vba)

2917

zba = vba;

2917

zba = vba;

2918

zea = details->last_index;

2918

zea = details->last_index;

2919

if (zea > vea)

2919

if (zea > vea)

2920

zea = vea;

2920

zea = vea;

2921

2922

unmap_mapping_range_vma(vma,

2922

unmap_mapping_range_vma(vma,

2923

((zba - vba) << PAGE_SHIFT) + vma->vm_start,

2923

((zba - vba) << PAGE_SHIFT) + vma->vm_start,

2924

((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,

2924

((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,

2925

details);

2925

details);

2926

}

2926

}

2927

}

2927

}

2928

2929

static inline void unmap_mapping_range_list(struct list_head *head,

2929

static inline void unmap_mapping_range_list(struct list_head *head,

2930

struct zap_details *details)

2930

struct zap_details *details)

2931

{

2931

{

2932

struct vm_area_struct *vma;

2932

struct vm_area_struct *vma;

2933

2934

/*

2934

/*

2935

* In nonlinear VMAs there is no correspondence between virtual address

2935

* In nonlinear VMAs there is no correspondence between virtual address

2936

* offset and file offset. So we must perform an exhaustive search

2936

* offset and file offset. So we must perform an exhaustive search

2937

* across *all* the pages in each nonlinear VMA, not just the pages

2937

* across *all* the pages in each nonlinear VMA, not just the pages

2938

* whose virtual address lies outside the file truncation point.

2938

* whose virtual address lies outside the file truncation point.

2939

*/

2939

*/

2940

list_for_each_entry(vma, head, shared.nonlinear) {

2940

list_for_each_entry(vma, head, shared.nonlinear) {

2941

details->nonlinear_vma = vma;

2941

details->nonlinear_vma = vma;

2942

unmap_mapping_range_vma(vma, vma->vm_start, vma->vm_end, details);

2942

unmap_mapping_range_vma(vma, vma->vm_start, vma->vm_end, details);

2943

}

2943

}

2944

}

2944

}

2945

2946

/**

2946

/**

2947

* unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file.

2947

* unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file.

2948

* @mapping: the address space containing mmaps to be unmapped.

2948

* @mapping: the address space containing mmaps to be unmapped.

2949

* @holebegin: byte in first page to unmap, relative to the start of

2949

* @holebegin: byte in first page to unmap, relative to the start of

2950

* the underlying file. This will be rounded down to a PAGE_SIZE

2950

* the underlying file. This will be rounded down to a PAGE_SIZE

2951

* boundary. Note that this is different from truncate_pagecache(), which

2951

* boundary. Note that this is different from truncate_pagecache(), which

2952

* must keep the partial page. In contrast, we must get rid of

2952

* must keep the partial page. In contrast, we must get rid of

2953

* partial pages.

2953

* partial pages.

2954

* @holelen: size of prospective hole in bytes. This will be rounded

2954

* @holelen: size of prospective hole in bytes. This will be rounded

2955

* up to a PAGE_SIZE boundary. A holelen of zero truncates to the

2955

* up to a PAGE_SIZE boundary. A holelen of zero truncates to the

2956

* end of the file.

2956

* end of the file.

2957

* @even_cows: 1 when truncating a file, unmap even private COWed pages;

2957

* @even_cows: 1 when truncating a file, unmap even private COWed pages;

2958

* but 0 when invalidating pagecache, don't throw away private data.

2958

* but 0 when invalidating pagecache, don't throw away private data.

2959

*/

2959

*/

2960

void unmap_mapping_range(struct address_space *mapping,

2960

void unmap_mapping_range(struct address_space *mapping,

2961

loff_t const holebegin, loff_t const holelen, int even_cows)

2961

loff_t const holebegin, loff_t const holelen, int even_cows)

2962

{

2962

{

2963

struct zap_details details;

2963

struct zap_details details;

2964

pgoff_t hba = holebegin >> PAGE_SHIFT;

2964

pgoff_t hba = holebegin >> PAGE_SHIFT;

2965

pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;

2965

pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;

2966

2967

/* Check for overflow. */

2967

/* Check for overflow. */

2968

if (sizeof(holelen) > sizeof(hlen)) {

2968

if (sizeof(holelen) > sizeof(hlen)) {

2969

long long holeend =

2969

long long holeend =

2970

(holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;

2970

(holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;

2971

if (holeend & ~(long long)ULONG_MAX)

2971

if (holeend & ~(long long)ULONG_MAX)

2972

hlen = ULONG_MAX - hba + 1;

2972

hlen = ULONG_MAX - hba + 1;

2973

}

2973

}

2974

2975

details.check_mapping = even_cows? NULL: mapping;

2975

details.check_mapping = even_cows? NULL: mapping;

2976

details.nonlinear_vma = NULL;

2976

details.nonlinear_vma = NULL;

2977

details.first_index = hba;

2977

details.first_index = hba;

2978

details.last_index = hba + hlen - 1;

2978

details.last_index = hba + hlen - 1;

2979

if (details.last_index < details.first_index)

2979

if (details.last_index < details.first_index)

2980

details.last_index = ULONG_MAX;

2980

details.last_index = ULONG_MAX;

2981

2982

2983

mutex_lock(&mapping->i_mmap_mutex);

2983

mutex_lock(&mapping->i_mmap_mutex);

2984

if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap)))

2984

if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap)))

2985

unmap_mapping_range_tree(&mapping->i_mmap, &details);

2985

unmap_mapping_range_tree(&mapping->i_mmap, &details);

2986

if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))

2986

if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))

2987

unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);

2987

unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);

2988

mutex_unlock(&mapping->i_mmap_mutex);

2988

mutex_unlock(&mapping->i_mmap_mutex);

2989

}

2989

}

2990

EXPORT_SYMBOL(unmap_mapping_range);

2990

EXPORT_SYMBOL(unmap_mapping_range);

2991

2992

/*

2992

/*

2993

* We enter with non-exclusive mmap_sem (to exclude vma changes,

2993

* We enter with non-exclusive mmap_sem (to exclude vma changes,

2994

* but allow concurrent faults), and pte mapped but not yet locked.

2994

* but allow concurrent faults), and pte mapped but not yet locked.

2995

* We return with mmap_sem still held, but pte unmapped and unlocked.

2995

* We return with mmap_sem still held, but pte unmapped and unlocked.

2996

*/

2996

*/

2997

static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,

2997

static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,

2998

unsigned long address, pte_t *page_table, pmd_t *pmd,

2998

unsigned long address, pte_t *page_table, pmd_t *pmd,

2999

unsigned int flags, pte_t orig_pte)

2999

unsigned int flags, pte_t orig_pte)

3000

{

3000

{

3001

spinlock_t *ptl;

3001

spinlock_t *ptl;

3002

struct page *page, *swapcache;

3002

struct page *page, *swapcache;

3003

swp_entry_t entry;

3003

swp_entry_t entry;

3004

pte_t pte;

3004

pte_t pte;

3005

int locked;

3005

int locked;

3006

struct mem_cgroup *ptr;

3006

struct mem_cgroup *ptr;

3007

int exclusive = 0;

3007

int exclusive = 0;

3008

int ret = 0;

3008

int ret = 0;

3009

3010

if (!pte_unmap_same(mm, pmd, page_table, orig_pte))

3010

if (!pte_unmap_same(mm, pmd, page_table, orig_pte))

3011

goto out;

3011

goto out;

3012

3013

entry = pte_to_swp_entry(orig_pte);

3013

entry = pte_to_swp_entry(orig_pte);

3014

if (unlikely(non_swap_entry(entry))) {

3014

if (unlikely(non_swap_entry(entry))) {

3015

if (is_migration_entry(entry)) {

3015

if (is_migration_entry(entry)) {

3016

migration_entry_wait(mm, pmd, address);

3016

migration_entry_wait(mm, pmd, address);

3017

} else if (is_hwpoison_entry(entry)) {

3017

} else if (is_hwpoison_entry(entry)) {

3018

ret = VM_FAULT_HWPOISON;

3018

ret = VM_FAULT_HWPOISON;

3019

} else {

3019

} else {

3020

print_bad_pte(vma, address, orig_pte, NULL);

3020

print_bad_pte(vma, address, orig_pte, NULL);

3021

ret = VM_FAULT_SIGBUS;

3021

ret = VM_FAULT_SIGBUS;

3022

}

3022

}

3023

goto out;

3023

goto out;

3024

}

3024

}

3025

delayacct_set_flag(DELAYACCT_PF_SWAPIN);

3025

delayacct_set_flag(DELAYACCT_PF_SWAPIN);

3026

page = lookup_swap_cache(entry);

3026

page = lookup_swap_cache(entry);

3027

if (!page) {

3027

if (!page) {

3028

page = swapin_readahead(entry,

3028

page = swapin_readahead(entry,

3029

GFP_HIGHUSER_MOVABLE, vma, address);

3029

GFP_HIGHUSER_MOVABLE, vma, address);

3030

if (!page) {

3030

if (!page) {

3031

/*

3031

/*

3032

* Back out if somebody else faulted in this pte

3032

* Back out if somebody else faulted in this pte

3033

* while we released the pte lock.

3033

* while we released the pte lock.

3034

*/

3034

*/

3035

page_table = pte_offset_map_lock(mm, pmd, address, &ptl);

3035

page_table = pte_offset_map_lock(mm, pmd, address, &ptl);

3036

if (likely(pte_same(*page_table, orig_pte)))

3036

if (likely(pte_same(*page_table, orig_pte)))

3037

ret = VM_FAULT_OOM;

3037

ret = VM_FAULT_OOM;

3038

delayacct_clear_flag(DELAYACCT_PF_SWAPIN);

3038

delayacct_clear_flag(DELAYACCT_PF_SWAPIN);

3039

goto unlock;

3039

goto unlock;

3040

}

3040

}

3041

3042

/* Had to read the page from swap area: Major fault */

3042

/* Had to read the page from swap area: Major fault */

3043

ret = VM_FAULT_MAJOR;

3043

ret = VM_FAULT_MAJOR;

3044

count_vm_event(PGMAJFAULT);

3044

count_vm_event(PGMAJFAULT);

3045

mem_cgroup_count_vm_event(mm, PGMAJFAULT);

3045

mem_cgroup_count_vm_event(mm, PGMAJFAULT);

3046

} else if (PageHWPoison(page)) {

3046

} else if (PageHWPoison(page)) {

3047

/*

3047

/*

3048

* hwpoisoned dirty swapcache pages are kept for killing

3048

* hwpoisoned dirty swapcache pages are kept for killing

3049

* owner processes (which may be unknown at hwpoison time)

3049

* owner processes (which may be unknown at hwpoison time)

3050

*/

3050

*/

3051

ret = VM_FAULT_HWPOISON;

3051

ret = VM_FAULT_HWPOISON;

3052

delayacct_clear_flag(DELAYACCT_PF_SWAPIN);

3052

delayacct_clear_flag(DELAYACCT_PF_SWAPIN);

3053

swapcache = page;

3053

swapcache = page;

3054

goto out_release;

3054

goto out_release;

3055

}

3055

}

3056

3057

swapcache = page;

3057

swapcache = page;

3058

locked = lock_page_or_retry(page, mm, flags);

3058

locked = lock_page_or_retry(page, mm, flags);

3059

3060

delayacct_clear_flag(DELAYACCT_PF_SWAPIN);

3060

delayacct_clear_flag(DELAYACCT_PF_SWAPIN);

3061

if (!locked) {

3061

if (!locked) {

3062

ret |= VM_FAULT_RETRY;

3062

ret |= VM_FAULT_RETRY;

3063

goto out_release;

3063

goto out_release;

3064

}

3064

}

3065

3066

/*

3066

/*

3067

* Make sure try_to_free_swap or reuse_swap_page or swapoff did not

3067

* Make sure try_to_free_swap or reuse_swap_page or swapoff did not

3068

* release the swapcache from under us. The page pin, and pte_same

3068

* release the swapcache from under us. The page pin, and pte_same

3069

* test below, are not enough to exclude that. Even if it is still

3069

* test below, are not enough to exclude that. Even if it is still

3070

* swapcache, we need to check that the page's swap has not changed.

3070

* swapcache, we need to check that the page's swap has not changed.

3071

*/

3071

*/

3072

if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val))

3072

if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val))

3073

goto out_page;

3073

goto out_page;

3074

3075

page = ksm_might_need_to_copy(page, vma, address);

3075

page = ksm_might_need_to_copy(page, vma, address);

3076

if (unlikely(!page)) {

3076

if (unlikely(!page)) {

3077

ret = VM_FAULT_OOM;

3077

ret = VM_FAULT_OOM;

3078

page = swapcache;

3078

page = swapcache;

3079

goto out_page;

3079

goto out_page;

3080

}

3080

}

3081

3082

if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) {

3082

if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) {

3083

ret = VM_FAULT_OOM;

3083

ret = VM_FAULT_OOM;

3084

goto out_page;

3084

goto out_page;

3085

}

3085

}

3086

3087

/*

3087

/*

3088

* Back out if somebody else already faulted in this pte.

3088

* Back out if somebody else already faulted in this pte.

3089

*/

3089

*/

3090

page_table = pte_offset_map_lock(mm, pmd, address, &ptl);

3090

page_table = pte_offset_map_lock(mm, pmd, address, &ptl);

3091

if (unlikely(!pte_same(*page_table, orig_pte)))

3091

if (unlikely(!pte_same(*page_table, orig_pte)))

3092

goto out_nomap;

3092

goto out_nomap;

3093

3094

if (unlikely(!PageUptodate(page))) {

3094

if (unlikely(!PageUptodate(page))) {

3095

ret = VM_FAULT_SIGBUS;

3095

ret = VM_FAULT_SIGBUS;

3096

goto out_nomap;

3096

goto out_nomap;

3097

}

3097

}

3098

3099

/*

3099

/*

3100

* The page isn't present yet, go ahead with the fault.

3100

* The page isn't present yet, go ahead with the fault.

3101

*

3101

*

3102

* Be careful about the sequence of operations here.

3102

* Be careful about the sequence of operations here.

3103

* To get its accounting right, reuse_swap_page() must be called

3103

* To get its accounting right, reuse_swap_page() must be called

3104

* while the page is counted on swap but not yet in mapcount i.e.

3104

* while the page is counted on swap but not yet in mapcount i.e.

3105

* before page_add_anon_rmap() and swap_free(); try_to_free_swap()

3105

* before page_add_anon_rmap() and swap_free(); try_to_free_swap()

3106

* must be called after the swap_free(), or it will never succeed.

3106

* must be called after the swap_free(), or it will never succeed.

3107

* Because delete_from_swap_page() may be called by reuse_swap_page(),

3107

* Because delete_from_swap_page() may be called by reuse_swap_page(),

3108

* mem_cgroup_commit_charge_swapin() may not be able to find swp_entry

3108

* mem_cgroup_commit_charge_swapin() may not be able to find swp_entry

3109

* in page->private. In this case, a record in swap_cgroup is silently

3109

* in page->private. In this case, a record in swap_cgroup is silently

3110

* discarded at swap_free().

3110

* discarded at swap_free().

3111

*/

3111

*/

3112

3113

inc_mm_counter_fast(mm, MM_ANONPAGES);

3113

inc_mm_counter_fast(mm, MM_ANONPAGES);

3114

dec_mm_counter_fast(mm, MM_SWAPENTS);

3114

dec_mm_counter_fast(mm, MM_SWAPENTS);

3115

pte = mk_pte(page, vma->vm_page_prot);

3115

pte = mk_pte(page, vma->vm_page_prot);

3116

if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) {

3116

if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) {

3117

pte = maybe_mkwrite(pte_mkdirty(pte), vma);

3117

pte = maybe_mkwrite(pte_mkdirty(pte), vma);

3118

flags &= ~FAULT_FLAG_WRITE;

3118

flags &= ~FAULT_FLAG_WRITE;

3119

ret |= VM_FAULT_WRITE;

3119

ret |= VM_FAULT_WRITE;

3120

exclusive = 1;

3120

exclusive = 1;

3121

}

3121

}

3122

flush_icache_page(vma, page);

3122

flush_icache_page(vma, page);

3123

if (pte_swp_soft_dirty(orig_pte))

3123

if (pte_swp_soft_dirty(orig_pte))

3124

pte = pte_mksoft_dirty(pte);

3124

pte = pte_mksoft_dirty(pte);

3125

set_pte_at(mm, address, page_table, pte);

3125

set_pte_at(mm, address, page_table, pte);

3126

if (page == swapcache)

3126

if (page == swapcache)

3127

do_page_add_anon_rmap(page, vma, address, exclusive);

3127

do_page_add_anon_rmap(page, vma, address, exclusive);

3128

else /* ksm created a completely new copy */

3128

else /* ksm created a completely new copy */

3129

page_add_new_anon_rmap(page, vma, address);

3129

page_add_new_anon_rmap(page, vma, address);

3130

/* It's better to call commit-charge after rmap is established */

3130

/* It's better to call commit-charge after rmap is established */

3131

mem_cgroup_commit_charge_swapin(page, ptr);

3131

mem_cgroup_commit_charge_swapin(page, ptr);

3132

3133

swap_free(entry);

3133

swap_free(entry);

3134

if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page))

3134

if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page))

3135

try_to_free_swap(page);

3135

try_to_free_swap(page);

3136

unlock_page(page);

3136

unlock_page(page);

3137

if (page != swapcache) {

3137

if (page != swapcache) {

3138

/*

3138

/*

3139

* Hold the lock to avoid the swap entry to be reused

3139

* Hold the lock to avoid the swap entry to be reused

3140

* until we take the PT lock for the pte_same() check

3140

* until we take the PT lock for the pte_same() check

3141

* (to avoid false positives from pte_same). For

3141

* (to avoid false positives from pte_same). For

3142

* further safety release the lock after the swap_free

3142

* further safety release the lock after the swap_free

3143

* so that the swap count won't change under a

3143

* so that the swap count won't change under a

3144

* parallel locked swapcache.

3144

* parallel locked swapcache.

3145

*/

3145

*/

3146

unlock_page(swapcache);

3146

unlock_page(swapcache);

3147

page_cache_release(swapcache);

3147

page_cache_release(swapcache);

3148

}

3148

}

3149

3150

if (flags & FAULT_FLAG_WRITE) {

3150

if (flags & FAULT_FLAG_WRITE) {

3151

ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);

3151

ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);

3152

if (ret & VM_FAULT_ERROR)

3152

if (ret & VM_FAULT_ERROR)

3153

ret &= VM_FAULT_ERROR;

3153

ret &= VM_FAULT_ERROR;

3154

goto out;

3154

goto out;

3155

}

3155

}

3156

3157

/* No need to invalidate - it was non-present before */

3157

/* No need to invalidate - it was non-present before */

3158

update_mmu_cache(vma, address, page_table);

3158

update_mmu_cache(vma, address, page_table);

3159

unlock:

3159

unlock:

3160

pte_unmap_unlock(page_table, ptl);

3160

pte_unmap_unlock(page_table, ptl);

3161

out:

3161

out:

3162

return ret;

3162

return ret;

3163

out_nomap:

3163

out_nomap:

3164

mem_cgroup_cancel_charge_swapin(ptr);

3164

mem_cgroup_cancel_charge_swapin(ptr);

3165

pte_unmap_unlock(page_table, ptl);

3165

pte_unmap_unlock(page_table, ptl);

3166

out_page:

3166

out_page:

3167

unlock_page(page);

3167

unlock_page(page);

3168

out_release:

3168

out_release:

3169

page_cache_release(page);

3169

page_cache_release(page);

3170

if (page != swapcache) {

3170

if (page != swapcache) {

3171

unlock_page(swapcache);

3171

unlock_page(swapcache);

3172

page_cache_release(swapcache);

3172

page_cache_release(swapcache);

3173

}

3173

}

3174

return ret;

3174

return ret;

3175

}

3175

}

3176

3177

/*

3177

/*

3178

* This is like a special single-page "expand_{down|up}wards()",

3178

* This is like a special single-page "expand_{down|up}wards()",

3179

* except we must first make sure that 'address{-|+}PAGE_SIZE'

3179

* except we must first make sure that 'address{-|+}PAGE_SIZE'

3180

* doesn't hit another vma.

3180

* doesn't hit another vma.

3181

*/

3181

*/

3182

static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address)

3182

static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address)

3183

{

3183

{

3184

address &= PAGE_MASK;

3184

address &= PAGE_MASK;

3185

if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) {

3185

if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) {

3186

struct vm_area_struct *prev = vma->vm_prev;

3186

struct vm_area_struct *prev = vma->vm_prev;

3187

3188

/*

3188

/*

3189

* Is there a mapping abutting this one below?

3189

* Is there a mapping abutting this one below?

3190

*

3190

*

3191

* That's only ok if it's the same stack mapping

3191

* That's only ok if it's the same stack mapping

3192

* that has gotten split..

3192

* that has gotten split..

3193

*/

3193

*/

3194

if (prev && prev->vm_end == address)

3194

if (prev && prev->vm_end == address)

3195

return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM;

3195

return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM;

3196

3197

expand_downwards(vma, address - PAGE_SIZE);

3197

expand_downwards(vma, address - PAGE_SIZE);

3198

}

3198

}

3199

if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) {

3199

if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) {

3200

struct vm_area_struct *next = vma->vm_next;

3200

struct vm_area_struct *next = vma->vm_next;

3201

3202

/* As VM_GROWSDOWN but s/below/above/ */

3202

/* As VM_GROWSDOWN but s/below/above/ */

3203

if (next && next->vm_start == address + PAGE_SIZE)

3203

if (next && next->vm_start == address + PAGE_SIZE)

3204

return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM;

3204

return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM;

3205

3206

expand_upwards(vma, address + PAGE_SIZE);

3206

expand_upwards(vma, address + PAGE_SIZE);

3207

}

3207

}

3208

return 0;

3208

return 0;

3209

}

3209

}

3210

3211

/*

3211

/*

3212

* We enter with non-exclusive mmap_sem (to exclude vma changes,

3212

* We enter with non-exclusive mmap_sem (to exclude vma changes,

3213

* but allow concurrent faults), and pte mapped but not yet locked.

3213

* but allow concurrent faults), and pte mapped but not yet locked.

3214

* We return with mmap_sem still held, but pte unmapped and unlocked.

3214

* We return with mmap_sem still held, but pte unmapped and unlocked.

3215

*/

3215

*/

3216

static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,

3216

static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,

3217

unsigned long address, pte_t *page_table, pmd_t *pmd,

3217

unsigned long address, pte_t *page_table, pmd_t *pmd,

3218

unsigned int flags)

3218

unsigned int flags)

3219

{

3219

{

3220

struct page *page;

3220

struct page *page;

3221

spinlock_t *ptl;

3221

spinlock_t *ptl;

3222

pte_t entry;

3222

pte_t entry;

3223

3224

pte_unmap(page_table);

3224

pte_unmap(page_table);

3225

3226

/* Check if we need to add a guard page to the stack */

3226

/* Check if we need to add a guard page to the stack */

3227

if (check_stack_guard_page(vma, address) < 0)

3227

if (check_stack_guard_page(vma, address) < 0)

3228

return VM_FAULT_SIGBUS;

3228

return VM_FAULT_SIGBUS;

3229

3230

/* Use the zero-page for reads */

3230

/* Use the zero-page for reads */

3231

if (!(flags & FAULT_FLAG_WRITE)) {

3231

if (!(flags & FAULT_FLAG_WRITE)) {

3232

entry = pte_mkspecial(pfn_pte(my_zero_pfn(address),

3232

entry = pte_mkspecial(pfn_pte(my_zero_pfn(address),

3233

vma->vm_page_prot));

3233

vma->vm_page_prot));

3234

page_table = pte_offset_map_lock(mm, pmd, address, &ptl);

3234

page_table = pte_offset_map_lock(mm, pmd, address, &ptl);

3235

if (!pte_none(*page_table))

3235

if (!pte_none(*page_table))

3236

goto unlock;

3236

goto unlock;

3237

goto setpte;

3237

goto setpte;

3238

}

3238

}

3239

3240

/* Allocate our own private page. */

3240

/* Allocate our own private page. */

3241

if (unlikely(anon_vma_prepare(vma)))

3241

if (unlikely(anon_vma_prepare(vma)))

3242

goto oom;

3242

goto oom;

3243

page = alloc_zeroed_user_highpage_movable(vma, address);

3243

page = alloc_zeroed_user_highpage_movable(vma, address);

3244

if (!page)

3244

if (!page)

3245

goto oom;

3245

goto oom;

3246

/*

3246

/*

3247

* The memory barrier inside __SetPageUptodate makes sure that

3247

* The memory barrier inside __SetPageUptodate makes sure that

3248

* preceeding stores to the page contents become visible before

3248

* preceeding stores to the page contents become visible before

3249

* the set_pte_at() write.

3249

* the set_pte_at() write.

3250

*/

3250

*/

3251

__SetPageUptodate(page);

3251

__SetPageUptodate(page);

3252

3253

if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL))

3253

if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL))

3254

goto oom_free_page;

3254

goto oom_free_page;

3255

3256

entry = mk_pte(page, vma->vm_page_prot);

3256

entry = mk_pte(page, vma->vm_page_prot);

3257

if (vma->vm_flags & VM_WRITE)

3257

if (vma->vm_flags & VM_WRITE)

3258

entry = pte_mkwrite(pte_mkdirty(entry));

3258

entry = pte_mkwrite(pte_mkdirty(entry));

3259

3260

page_table = pte_offset_map_lock(mm, pmd, address, &ptl);

3260

page_table = pte_offset_map_lock(mm, pmd, address, &ptl);

3261

if (!pte_none(*page_table))

3261

if (!pte_none(*page_table))

3262

goto release;

3262

goto release;

3263

3264

inc_mm_counter_fast(mm, MM_ANONPAGES);

3264

inc_mm_counter_fast(mm, MM_ANONPAGES);

3265

page_add_new_anon_rmap(page, vma, address);

3265

page_add_new_anon_rmap(page, vma, address);

3266

setpte:

3266

setpte:

3267

set_pte_at(mm, address, page_table, entry);

3267

set_pte_at(mm, address, page_table, entry);

3268

3269

/* No need to invalidate - it was non-present before */

3269

/* No need to invalidate - it was non-present before */

3270

update_mmu_cache(vma, address, page_table);

3270

update_mmu_cache(vma, address, page_table);

3271

unlock:

3271

unlock:

3272

pte_unmap_unlock(page_table, ptl);

3272

pte_unmap_unlock(page_table, ptl);

3273

return 0;

3273

return 0;

3274

release:

3274

release:

3275

mem_cgroup_uncharge_page(page);

3275

mem_cgroup_uncharge_page(page);

3276

page_cache_release(page);

3276

page_cache_release(page);

3277

goto unlock;

3277

goto unlock;

3278

oom_free_page:

3278

oom_free_page:

3279

page_cache_release(page);

3279

page_cache_release(page);

3280

oom:

3280

oom:

3281

return VM_FAULT_OOM;

3281

return VM_FAULT_OOM;

3282

}

3282

}

3283

3284

/*

3284

/*

3285

* __do_fault() tries to create a new page mapping. It aggressively

3285

* __do_fault() tries to create a new page mapping. It aggressively

3286

* tries to share with existing pages, but makes a separate copy if

3286

* tries to share with existing pages, but makes a separate copy if

3287

* the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid

3287

* the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid

3288

* the next page fault.

3288

* the next page fault.

3289

*

3289

*

3290

* As this is called only for pages that do not currently exist, we

3290

* As this is called only for pages that do not currently exist, we

3291

* do not need to flush old virtual caches or the TLB.

3291

* do not need to flush old virtual caches or the TLB.

3292

*

3292

*

3293

* We enter with non-exclusive mmap_sem (to exclude vma changes,

3293

* We enter with non-exclusive mmap_sem (to exclude vma changes,

3294

* but allow concurrent faults), and pte neither mapped nor locked.

3294

* but allow concurrent faults), and pte neither mapped nor locked.

3295

* We return with mmap_sem still held, but pte unmapped and unlocked.

3295

* We return with mmap_sem still held, but pte unmapped and unlocked.

3296

*/

3296

*/

3297

static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,

3297

static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,

3298

unsigned long address, pmd_t *pmd,

3298

unsigned long address, pmd_t *pmd,

3299

pgoff_t pgoff, unsigned int flags, pte_t orig_pte)

3299

pgoff_t pgoff, unsigned int flags, pte_t orig_pte)

3300

{

3300

{

3301

pte_t *page_table;

3301

pte_t *page_table;

3302

spinlock_t *ptl;

3302

spinlock_t *ptl;

3303

struct page *page;

3303

struct page *page;

3304

struct page *cow_page;

3304

struct page *cow_page;

3305

pte_t entry;

3305

pte_t entry;

3306

int anon = 0;

3306

int anon = 0;

3307

struct page *dirty_page = NULL;

3307

struct page *dirty_page = NULL;

3308

struct vm_fault vmf;

3308

struct vm_fault vmf;

3309

int ret;

3309

int ret;

3310

int page_mkwrite = 0;

3310

int page_mkwrite = 0;

3311

3312

/*

3312

/*

3313

* If we do COW later, allocate page befor taking lock_page()

3313

* If we do COW later, allocate page befor taking lock_page()

3314

* on the file cache page. This will reduce lock holding time.

3314

* on the file cache page. This will reduce lock holding time.

3315

*/

3315

*/

3316

if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) {

3316

if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) {

3317

3318

if (unlikely(anon_vma_prepare(vma)))

3318

if (unlikely(anon_vma_prepare(vma)))

3319

return VM_FAULT_OOM;

3319

return VM_FAULT_OOM;

3320

3321

cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);

3321

cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);

3322

if (!cow_page)

3322

if (!cow_page)

3323

return VM_FAULT_OOM;

3323

return VM_FAULT_OOM;

3324

3325

if (mem_cgroup_newpage_charge(cow_page, mm, GFP_KERNEL)) {

3325

if (mem_cgroup_newpage_charge(cow_page, mm, GFP_KERNEL)) {

3326

page_cache_release(cow_page);

3326

page_cache_release(cow_page);

3327

return VM_FAULT_OOM;

3327

return VM_FAULT_OOM;

3328

}

3328

}

3329

} else

3329

} else

3330

cow_page = NULL;

3330

cow_page = NULL;

3331

3332

vmf.virtual_address = (void __user *)(address & PAGE_MASK);

3332

vmf.virtual_address = (void __user *)(address & PAGE_MASK);

3333

vmf.pgoff = pgoff;

3333

vmf.pgoff = pgoff;

3334

vmf.flags = flags;

3334

vmf.flags = flags;

3335

vmf.page = NULL;

3335

vmf.page = NULL;

3336

3337

ret = vma->vm_ops->fault(vma, &vmf);

3337

ret = vma->vm_ops->fault(vma, &vmf);

3338

if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE |

3338

if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE |

3339

VM_FAULT_RETRY)))

3339

VM_FAULT_RETRY)))

3340

goto uncharge_out;

3340

goto uncharge_out;

3341

3342

if (unlikely(PageHWPoison(vmf.page))) {

3342

if (unlikely(PageHWPoison(vmf.page))) {

3343

if (ret & VM_FAULT_LOCKED)

3343

if (ret & VM_FAULT_LOCKED)

3344

unlock_page(vmf.page);

3344

unlock_page(vmf.page);

3345

ret = VM_FAULT_HWPOISON;

3345

ret = VM_FAULT_HWPOISON;

3346

goto uncharge_out;

3346

goto uncharge_out;

3347

}

3347

}

3348

3349

/*

3349

/*

3350

* For consistency in subsequent calls, make the faulted page always

3350

* For consistency in subsequent calls, make the faulted page always

3351

* locked.

3351

* locked.

3352

*/

3352

*/

3353

if (unlikely(!(ret & VM_FAULT_LOCKED)))

3353

if (unlikely(!(ret & VM_FAULT_LOCKED)))

3354

lock_page(vmf.page);

3354

lock_page(vmf.page);

3355

else

3355

else

3356

VM_BUG_ON(!PageLocked(vmf.page));

3356

VM_BUG_ON(!PageLocked(vmf.page));

3357

3358

/*

3358

/*

3359

* Should we do an early C-O-W break?

3359

* Should we do an early C-O-W break?

3360

*/

3360

*/

3361

page = vmf.page;

3361

page = vmf.page;

3362

if (flags & FAULT_FLAG_WRITE) {

3362

if (flags & FAULT_FLAG_WRITE) {

3363

if (!(vma->vm_flags & VM_SHARED)) {

3363

if (!(vma->vm_flags & VM_SHARED)) {

3364

page = cow_page;

3364

page = cow_page;

3365

anon = 1;

3365

anon = 1;

3366

copy_user_highpage(page, vmf.page, address, vma);

3366

copy_user_highpage(page, vmf.page, address, vma);

3367

__SetPageUptodate(page);

3367

__SetPageUptodate(page);

3368

} else {

3368

} else {

3369

/*

3369

/*

3370

* If the page will be shareable, see if the backing

3370

* If the page will be shareable, see if the backing

3371

* address space wants to know that the page is about

3371

* address space wants to know that the page is about

3372

* to become writable

3372

* to become writable

3373

*/

3373

*/

3374

if (vma->vm_ops->page_mkwrite) {

3374

if (vma->vm_ops->page_mkwrite) {

3375

int tmp;

3375

int tmp;

3376

3377

unlock_page(page);

3377

unlock_page(page);

3378

vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;

3378

vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;

3379

tmp = vma->vm_ops->page_mkwrite(vma, &vmf);

3379

tmp = vma->vm_ops->page_mkwrite(vma, &vmf);

3380

if (unlikely(tmp &

3380

if (unlikely(tmp &

3381

(VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {

3381

(VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {

3382

ret = tmp;

3382

ret = tmp;

3383

goto unwritable_page;

3383

goto unwritable_page;

3384

}

3384

}

3385

if (unlikely(!(tmp & VM_FAULT_LOCKED))) {

3385

if (unlikely(!(tmp & VM_FAULT_LOCKED))) {

3386

lock_page(page);

3386

lock_page(page);

3387

if (!page->mapping) {

3387

if (!page->mapping) {

3388

ret = 0; /* retry the fault */

3388

ret = 0; /* retry the fault */

3389

unlock_page(page);

3389

unlock_page(page);

3390

goto unwritable_page;

3390

goto unwritable_page;

3391

}

3391

}

3392

} else

3392

} else

3393

VM_BUG_ON(!PageLocked(page));

3393

VM_BUG_ON(!PageLocked(page));

3394

page_mkwrite = 1;

3394

page_mkwrite = 1;

3395

}

3395

}

3396

}

3396

}

3397

3398

}

3398

}

3399

3400

page_table = pte_offset_map_lock(mm, pmd, address, &ptl);

3400

page_table = pte_offset_map_lock(mm, pmd, address, &ptl);

3401

3402

/*

3402

/*

3403

* This silly early PAGE_DIRTY setting removes a race

3403

* This silly early PAGE_DIRTY setting removes a race

3404

* due to the bad i386 page protection. But it's valid

3404

* due to the bad i386 page protection. But it's valid

3405

* for other architectures too.

3405

* for other architectures too.

3406

*

3406

*

3407

* Note that if FAULT_FLAG_WRITE is set, we either now have

3407

* Note that if FAULT_FLAG_WRITE is set, we either now have

3408

* an exclusive copy of the page, or this is a shared mapping,

3408

* an exclusive copy of the page, or this is a shared mapping,

3409

* so we can make it writable and dirty to avoid having to

3409

* so we can make it writable and dirty to avoid having to

3410

* handle that later.

3410

* handle that later.

3411

*/

3411

*/

3412

/* Only go through if we didn't race with anybody else... */

3412

/* Only go through if we didn't race with anybody else... */

3413

if (likely(pte_same(*page_table, orig_pte))) {

3413

if (likely(pte_same(*page_table, orig_pte))) {

3414

flush_icache_page(vma, page);

3414

flush_icache_page(vma, page);

3415

entry = mk_pte(page, vma->vm_page_prot);

3415

entry = mk_pte(page, vma->vm_page_prot);

3416

if (flags & FAULT_FLAG_WRITE)

3416

if (flags & FAULT_FLAG_WRITE)

3417

entry = maybe_mkwrite(pte_mkdirty(entry), vma);

3417

entry = maybe_mkwrite(pte_mkdirty(entry), vma);

3418

else if (pte_file(orig_pte) && pte_file_soft_dirty(orig_pte))

3418

else if (pte_file(orig_pte) && pte_file_soft_dirty(orig_pte))

3419

pte_mksoft_dirty(entry);

3419

pte_mksoft_dirty(entry);

3420

if (anon) {

3420

if (anon) {

3421

inc_mm_counter_fast(mm, MM_ANONPAGES);

3421

inc_mm_counter_fast(mm, MM_ANONPAGES);

3422

page_add_new_anon_rmap(page, vma, address);

3422

page_add_new_anon_rmap(page, vma, address);

3423

} else {

3423

} else {

3424

inc_mm_counter_fast(mm, MM_FILEPAGES);

3424

inc_mm_counter_fast(mm, MM_FILEPAGES);

3425

page_add_file_rmap(page);

3425

page_add_file_rmap(page);

3426

if (flags & FAULT_FLAG_WRITE) {

3426

if (flags & FAULT_FLAG_WRITE) {

3427

dirty_page = page;

3427

dirty_page = page;

3428

get_page(dirty_page);

3428

get_page(dirty_page);

3429

}

3429

}

3430

}

3430

}

3431

set_pte_at(mm, address, page_table, entry);

3431

set_pte_at(mm, address, page_table, entry);

3432

3433

/* no need to invalidate: a not-present page won't be cached */

3433

/* no need to invalidate: a not-present page won't be cached */

3434

update_mmu_cache(vma, address, page_table);

3434

update_mmu_cache(vma, address, page_table);

3435

} else {

3435

} else {

3436

if (cow_page)

3436

if (cow_page)

3437

mem_cgroup_uncharge_page(cow_page);

3437

mem_cgroup_uncharge_page(cow_page);

3438

if (anon)

3438

if (anon)

3439

page_cache_release(page);

3439

page_cache_release(page);

3440

else

3440

else

3441

anon = 1; /* no anon but release faulted_page */

3441

anon = 1; /* no anon but release faulted_page */

3442

}

3442

}

3443

3444

pte_unmap_unlock(page_table, ptl);

3444

pte_unmap_unlock(page_table, ptl);

3445

3446

if (dirty_page) {

3446

if (dirty_page) {

3447

struct address_space *mapping = page->mapping;

3447

struct address_space *mapping = page->mapping;

3448

int dirtied = 0;

3448

int dirtied = 0;

3449

3450

if (set_page_dirty(dirty_page))

3450

if (set_page_dirty(dirty_page))

3451

dirtied = 1;

3451

dirtied = 1;

3452

unlock_page(dirty_page);

3452

unlock_page(dirty_page);

3453

put_page(dirty_page);

3453

put_page(dirty_page);

3454

if ((dirtied || page_mkwrite) && mapping) {

3454

if ((dirtied || page_mkwrite) && mapping) {

3455

/*

3455

/*

3456

* Some device drivers do not set page.mapping but still

3456

* Some device drivers do not set page.mapping but still

3457

* dirty their pages

3457

* dirty their pages

3458

*/

3458

*/

3459

balance_dirty_pages_ratelimited(mapping);

3459

balance_dirty_pages_ratelimited(mapping);

3460

}

3460

}

3461

3462

/* file_update_time outside page_lock */

3462

/* file_update_time outside page_lock */

3463

if (vma->vm_file && !page_mkwrite)

3463

if (vma->vm_file && !page_mkwrite)

3464

file_update_time(vma->vm_file);

3464

file_update_time(vma->vm_file);

3465

} else {

3465

} else {

3466

unlock_page(vmf.page);

3466

unlock_page(vmf.page);

3467

if (anon)

3467

if (anon)

3468

page_cache_release(vmf.page);

3468

page_cache_release(vmf.page);

3469

}

3469

}

3470

3471

return ret;

3471

return ret;

3472

3473

unwritable_page:

3473

unwritable_page:

3474

page_cache_release(page);

3474

page_cache_release(page);

3475

return ret;

3475

return ret;

3476

uncharge_out:

3476

uncharge_out:

3477

/* fs's fault handler get error */

3477

/* fs's fault handler get error */

3478

if (cow_page) {

3478

if (cow_page) {

3479

mem_cgroup_uncharge_page(cow_page);

3479

mem_cgroup_uncharge_page(cow_page);

3480

page_cache_release(cow_page);

3480

page_cache_release(cow_page);

3481

}

3481

}

3482

return ret;

3482

return ret;

3483

}

3483

}

3484

3485

static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,

3485

static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,

3486

unsigned long address, pte_t *page_table, pmd_t *pmd,

3486

unsigned long address, pte_t *page_table, pmd_t *pmd,

3487

unsigned int flags, pte_t orig_pte)

3487

unsigned int flags, pte_t orig_pte)

3488

{

3488

{

3489

pgoff_t pgoff = (((address & PAGE_MASK)

3489

pgoff_t pgoff = (((address & PAGE_MASK)

3490

- vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;

3490

- vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;

3491

3492

pte_unmap(page_table);

3492

pte_unmap(page_table);

3493

return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);

3493

return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);

3494

}

3494

}

3495

3496

/*

3496

/*

3497

* Fault of a previously existing named mapping. Repopulate the pte

3497

* Fault of a previously existing named mapping. Repopulate the pte

3498

* from the encoded file_pte if possible. This enables swappable

3498

* from the encoded file_pte if possible. This enables swappable

3499

* nonlinear vmas.

3499

* nonlinear vmas.

3500

*

3500

*

3501

* We enter with non-exclusive mmap_sem (to exclude vma changes,

3501

* We enter with non-exclusive mmap_sem (to exclude vma changes,

3502

* but allow concurrent faults), and pte mapped but not yet locked.

3502

* but allow concurrent faults), and pte mapped but not yet locked.

3503

* We return with mmap_sem still held, but pte unmapped and unlocked.

3503

* We return with mmap_sem still held, but pte unmapped and unlocked.

3504

*/

3504

*/

3505

static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,

3505

static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,

3506

unsigned long address, pte_t *page_table, pmd_t *pmd,

3506

unsigned long address, pte_t *page_table, pmd_t *pmd,

3507

unsigned int flags, pte_t orig_pte)

3507

unsigned int flags, pte_t orig_pte)

3508

{

3508

{

3509

pgoff_t pgoff;

3509

pgoff_t pgoff;

3510

3511

flags |= FAULT_FLAG_NONLINEAR;

3511

flags |= FAULT_FLAG_NONLINEAR;

3512

3513

if (!pte_unmap_same(mm, pmd, page_table, orig_pte))

3513

if (!pte_unmap_same(mm, pmd, page_table, orig_pte))

3514

return 0;

3514

return 0;

3515

3516

if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {

3516

if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {

3517

/*

3517

/*

3518

* Page table corrupted: show pte and kill process.

3518

* Page table corrupted: show pte and kill process.

3519

*/

3519

*/

3520

print_bad_pte(vma, address, orig_pte, NULL);

3520

print_bad_pte(vma, address, orig_pte, NULL);

3521

return VM_FAULT_SIGBUS;

3521

return VM_FAULT_SIGBUS;

3522

}

3522

}

3523

3524

pgoff = pte_to_pgoff(orig_pte);

3524

pgoff = pte_to_pgoff(orig_pte);

3525

return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);

3525

return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);

3526

}

3526

}

3527

3528

int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,

3528

int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,

3529

unsigned long addr, int page_nid)

3529

unsigned long addr, int page_nid)

3530

{

3530

{

3531

get_page(page);

3531

get_page(page);

3532

3533

count_vm_numa_event(NUMA_HINT_FAULTS);

3533

count_vm_numa_event(NUMA_HINT_FAULTS);

3534

if (page_nid == numa_node_id())

3534

if (page_nid == numa_node_id())

3535

count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL);

3535

count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL);

3536

3537

return mpol_misplaced(page, vma, addr);

3537

return mpol_misplaced(page, vma, addr);

3538

}

3538

}

3539

3540

int do_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,

3540

int do_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,

3541

unsigned long addr, pte_t pte, pte_t *ptep, pmd_t *pmd)

3541

unsigned long addr, pte_t pte, pte_t *ptep, pmd_t *pmd)

3542

{

3542

{

3543

struct page *page = NULL;

3543

struct page *page = NULL;

3544

spinlock_t *ptl;

3544

spinlock_t *ptl;

3545

int page_nid = -1;

3545

int page_nid = -1;

3546

int target_nid;

3546

int target_nid;

3547

bool migrated = false;

3547

bool migrated = false;

3548

3549

/*

3549

/*

3550

* The "pte" at this point cannot be used safely without

3550

* The "pte" at this point cannot be used safely without

3551

* validation through pte_unmap_same(). It's of NUMA type but

3551

* validation through pte_unmap_same(). It's of NUMA type but

3552

* the pfn may be screwed if the read is non atomic.

3552

* the pfn may be screwed if the read is non atomic.

3553

*

3553

*

3554

* ptep_modify_prot_start is not called as this is clearing

3554

* ptep_modify_prot_start is not called as this is clearing

3555

* the _PAGE_NUMA bit and it is not really expected that there

3555

* the _PAGE_NUMA bit and it is not really expected that there

3556

* would be concurrent hardware modifications to the PTE.

3556

* would be concurrent hardware modifications to the PTE.

3557

*/

3557

*/

3558

ptl = pte_lockptr(mm, pmd);

3558

ptl = pte_lockptr(mm, pmd);

3559

spin_lock(ptl);

3559

spin_lock(ptl);

3560

if (unlikely(!pte_same(*ptep, pte))) {

3560

if (unlikely(!pte_same(*ptep, pte))) {

3561

pte_unmap_unlock(ptep, ptl);

3561

pte_unmap_unlock(ptep, ptl);

3562

goto out;

3562

goto out;

3563

}

3563

}

3564

3565

pte = pte_mknonnuma(pte);

3565

pte = pte_mknonnuma(pte);

3566

set_pte_at(mm, addr, ptep, pte);

3566

set_pte_at(mm, addr, ptep, pte);

3567

update_mmu_cache(vma, addr, ptep);

3567

update_mmu_cache(vma, addr, ptep);

3568

3569

page = vm_normal_page(vma, addr, pte);

3569

page = vm_normal_page(vma, addr, pte);

3570

if (!page) {

3570

if (!page) {

3571

pte_unmap_unlock(ptep, ptl);

3571

pte_unmap_unlock(ptep, ptl);

3572

return 0;

3572

return 0;

3573

}

3573

}

3574

3575

page_nid = page_to_nid(page);

3575

page_nid = page_to_nid(page);

3576

target_nid = numa_migrate_prep(page, vma, addr, page_nid);

3576

target_nid = numa_migrate_prep(page, vma, addr, page_nid);

3577

pte_unmap_unlock(ptep, ptl);

3577

pte_unmap_unlock(ptep, ptl);

3578

if (target_nid == -1) {

3578

if (target_nid == -1) {

3579

put_page(page);

3579

put_page(page);

3580

goto out;

3580

goto out;

3581

}

3581

}

3582

3583

/* Migrate to the requested node */

3583

/* Migrate to the requested node */

3584

migrated = migrate_misplaced_page(page, target_nid);

3584

migrated = migrate_misplaced_page(page, target_nid);

3585

if (migrated)

3585

if (migrated)

3586

page_nid = target_nid;

3586

page_nid = target_nid;

3587

3588

out:

3588

out:

3589

if (page_nid != -1)

3589

if (page_nid != -1)

3590

task_numa_fault(page_nid, 1, migrated);

3590

task_numa_fault(page_nid, 1, migrated);

3591

return 0;

3591

return 0;

3592

}

3592

}

3593

3594

/* NUMA hinting page fault entry point for regular pmds */

3594

/* NUMA hinting page fault entry point for regular pmds */

3595

#ifdef CONFIG_NUMA_BALANCING

3595

#ifdef CONFIG_NUMA_BALANCING

3596

static int do_pmd_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,

3596

static int do_pmd_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,

3597

unsigned long addr, pmd_t *pmdp)

3597

unsigned long addr, pmd_t *pmdp)

3598

{

3598

{

3599

pmd_t pmd;

3599

pmd_t pmd;

3600

pte_t *pte, *orig_pte;

3600

pte_t *pte, *orig_pte;

3601

unsigned long _addr = addr & PMD_MASK;

3601

unsigned long _addr = addr & PMD_MASK;

3602

unsigned long offset;

3602

unsigned long offset;

3603

spinlock_t *ptl;

3603

spinlock_t *ptl;

3604

bool numa = false;

3604

bool numa = false;

3605

3606

spin_lock(&mm->page_table_lock);

3606

spin_lock(&mm->page_table_lock);

3607

pmd = *pmdp;

3607

pmd = *pmdp;

3608

if (pmd_numa(pmd)) {

3608

if (pmd_numa(pmd)) {

3609

set_pmd_at(mm, _addr, pmdp, pmd_mknonnuma(pmd));

3609

set_pmd_at(mm, _addr, pmdp, pmd_mknonnuma(pmd));

3610

numa = true;

3610

numa = true;

3611

}

3611

}

3612

spin_unlock(&mm->page_table_lock);

3612

spin_unlock(&mm->page_table_lock);

3613

3614

if (!numa)

3614

if (!numa)

3615

return 0;

3615

return 0;

3616

3617

/* we're in a page fault so some vma must be in the range */

3617

/* we're in a page fault so some vma must be in the range */

3618

BUG_ON(!vma);

3618

BUG_ON(!vma);

3619

BUG_ON(vma->vm_start >= _addr + PMD_SIZE);

3619

BUG_ON(vma->vm_start >= _addr + PMD_SIZE);

3620

offset = max(_addr, vma->vm_start) & ~PMD_MASK;

3620

offset = max(_addr, vma->vm_start) & ~PMD_MASK;

3621

VM_BUG_ON(offset >= PMD_SIZE);

3621

VM_BUG_ON(offset >= PMD_SIZE);

3622

orig_pte = pte = pte_offset_map_lock(mm, pmdp, _addr, &ptl);

3622

orig_pte = pte = pte_offset_map_lock(mm, pmdp, _addr, &ptl);

3623

pte += offset >> PAGE_SHIFT;

3623

pte += offset >> PAGE_SHIFT;

3624

for (addr = _addr + offset; addr < _addr + PMD_SIZE; pte++, addr += PAGE_SIZE) {

3624

for (addr = _addr + offset; addr < _addr + PMD_SIZE; pte++, addr += PAGE_SIZE) {

3625

pte_t pteval = *pte;

3625

pte_t pteval = *pte;

3626

struct page *page;

3626

struct page *page;

3627

int page_nid = -1;

3627

int page_nid = -1;

3628

int target_nid;

3628

int target_nid;

3629

bool migrated = false;

3629

bool migrated = false;

3630

3631

if (!pte_present(pteval))

3631

if (!pte_present(pteval))

3632

continue;

3632

continue;

3633

if (!pte_numa(pteval))

3633

if (!pte_numa(pteval))

3634

continue;

3634

continue;

3635

if (addr >= vma->vm_end) {

3635

if (addr >= vma->vm_end) {

3636

vma = find_vma(mm, addr);

3636

vma = find_vma(mm, addr);

3637

/* there's a pte present so there must be a vma */

3637

/* there's a pte present so there must be a vma */

3638

BUG_ON(!vma);

3638

BUG_ON(!vma);

3639

BUG_ON(addr < vma->vm_start);

3639

BUG_ON(addr < vma->vm_start);

3640

}

3640

}

3641

if (pte_numa(pteval)) {

3641

if (pte_numa(pteval)) {

3642

pteval = pte_mknonnuma(pteval);

3642

pteval = pte_mknonnuma(pteval);

3643

set_pte_at(mm, addr, pte, pteval);

3643

set_pte_at(mm, addr, pte, pteval);

3644

}

3644

}

3645

page = vm_normal_page(vma, addr, pteval);

3645

page = vm_normal_page(vma, addr, pteval);

3646

if (unlikely(!page))

3646

if (unlikely(!page))

3647

continue;

3647

continue;

3648

/* only check non-shared pages */

3648

/* only check non-shared pages */

3649

if (unlikely(page_mapcount(page) != 1))

3649

if (unlikely(page_mapcount(page) != 1))

3650

continue;

3650

continue;

3651

3652

page_nid = page_to_nid(page);

3652

page_nid = page_to_nid(page);

3653

target_nid = numa_migrate_prep(page, vma, addr, page_nid);

3653

target_nid = numa_migrate_prep(page, vma, addr, page_nid);

3654

pte_unmap_unlock(pte, ptl);

3654

pte_unmap_unlock(pte, ptl);

3655

if (target_nid != -1) {

3655

if (target_nid != -1) {

3656

migrated = migrate_misplaced_page(page, target_nid);

3656

migrated = migrate_misplaced_page(page, target_nid);

3657

if (migrated)

3657

if (migrated)

3658

page_nid = target_nid;

3658

page_nid = target_nid;

3659

} else {

3659

} else {

3660

put_page(page);

3660

put_page(page);

3661

}

3661

}

3662

3663

if (page_nid != -1)

3663

if (page_nid != -1)

3664

task_numa_fault(page_nid, 1, migrated);

3664

task_numa_fault(page_nid, 1, migrated);

3665

3666

pte = pte_offset_map_lock(mm, pmdp, addr, &ptl);

3666

pte = pte_offset_map_lock(mm, pmdp, addr, &ptl);

3667

}

3667

}

3668

pte_unmap_unlock(orig_pte, ptl);

3668

pte_unmap_unlock(orig_pte, ptl);

3669

3670

return 0;

3670

return 0;

3671

}

3671

}

3672

#else

3672

#else

3673

static int do_pmd_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,

3673

static int do_pmd_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,

3674

unsigned long addr, pmd_t *pmdp)

3674

unsigned long addr, pmd_t *pmdp)

3675

{

3675

{

3676

BUG();

3676

BUG();

3677

return 0;

3677

return 0;

3678

}

3678

}

3679

#endif /* CONFIG_NUMA_BALANCING */

3679

#endif /* CONFIG_NUMA_BALANCING */

3680

3681

/*

3681

/*

3682

* These routines also need to handle stuff like marking pages dirty

3682

* These routines also need to handle stuff like marking pages dirty

3683

* and/or accessed for architectures that don't do it in hardware (most

3683

* and/or accessed for architectures that don't do it in hardware (most

3684

* RISC architectures). The early dirtying is also good on the i386.

3684

* RISC architectures). The early dirtying is also good on the i386.

3685

*

3685

*

3686

* There is also a hook called "update_mmu_cache()" that architectures

3686

* There is also a hook called "update_mmu_cache()" that architectures

3687

* with external mmu caches can use to update those (ie the Sparc or

3687

* with external mmu caches can use to update those (ie the Sparc or

3688

* PowerPC hashed page tables that act as extended TLBs).

3688

* PowerPC hashed page tables that act as extended TLBs).

3689

*

3689

*

3690

* We enter with non-exclusive mmap_sem (to exclude vma changes,

3690

* We enter with non-exclusive mmap_sem (to exclude vma changes,

3691

* but allow concurrent faults), and pte mapped but not yet locked.

3691

* but allow concurrent faults), and pte mapped but not yet locked.

3692

* We return with mmap_sem still held, but pte unmapped and unlocked.

3692

* We return with mmap_sem still held, but pte unmapped and unlocked.

3693

*/

3693

*/

3694

static int handle_pte_fault(struct mm_struct *mm,

3694

static int handle_pte_fault(struct mm_struct *mm,

3695

struct vm_area_struct *vma, unsigned long address,

3695

struct vm_area_struct *vma, unsigned long address,

3696

pte_t *pte, pmd_t *pmd, unsigned int flags)

3696

pte_t *pte, pmd_t *pmd, unsigned int flags)

3697

{

3697

{

3698

pte_t entry;

3698

pte_t entry;

3699

spinlock_t *ptl;

3699

spinlock_t *ptl;

3700

3701

entry = ACCESS_ONCE(*pte);

3701

entry = ACCESS_ONCE(*pte);

3702

if (!pte_present(entry)) {

3702

if (!pte_present(entry)) {

3703

if (pte_none(entry)) {

3703

if (pte_none(entry)) {

3704

if (vma->vm_ops) {

3704

if (vma->vm_ops) {

3705

if (likely(vma->vm_ops->fault))

3705

if (likely(vma->vm_ops->fault))

3706

return do_linear_fault(mm, vma, address,

3706

return do_linear_fault(mm, vma, address,

3707

pte, pmd, flags, entry);

3707

pte, pmd, flags, entry);

3708

}

3708

}

3709

return do_anonymous_page(mm, vma, address,

3709

return do_anonymous_page(mm, vma, address,

3710

pte, pmd, flags);

3710

pte, pmd, flags);

3711

}

3711

}

3712

if (pte_file(entry))

3712

if (pte_file(entry))

3713

return do_nonlinear_fault(mm, vma, address,

3713

return do_nonlinear_fault(mm, vma, address,

3714

pte, pmd, flags, entry);

3714

pte, pmd, flags, entry);

3715

return do_swap_page(mm, vma, address,

3715

return do_swap_page(mm, vma, address,

3716

pte, pmd, flags, entry);

3716

pte, pmd, flags, entry);

3717

}

3717

}

3718

3719

if (pte_numa(entry))

3719

if (pte_numa(entry))

3720

return do_numa_page(mm, vma, address, entry, pte, pmd);

3720

return do_numa_page(mm, vma, address, entry, pte, pmd);

3721

3722

ptl = pte_lockptr(mm, pmd);

3722

ptl = pte_lockptr(mm, pmd);

3723

spin_lock(ptl);

3723

spin_lock(ptl);

3724

if (unlikely(!pte_same(*pte, entry)))

3724

if (unlikely(!pte_same(*pte, entry)))

3725

goto unlock;

3725

goto unlock;

3726

if (flags & FAULT_FLAG_WRITE) {

3726

if (flags & FAULT_FLAG_WRITE) {

3727

if (!pte_write(entry))

3727

if (!pte_write(entry))

3728

return do_wp_page(mm, vma, address,

3728

return do_wp_page(mm, vma, address,

3729

pte, pmd, ptl, entry);

3729

pte, pmd, ptl, entry);

3730

entry = pte_mkdirty(entry);

3730

entry = pte_mkdirty(entry);

3731

}

3731

}

3732

entry = pte_mkyoung(entry);

3732

entry = pte_mkyoung(entry);

3733

if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) {

3733

if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) {

3734

update_mmu_cache(vma, address, pte);

3734

update_mmu_cache(vma, address, pte);

3735

} else {

3735

} else {

3736

/*

3736

/*

3737

* This is needed only for protection faults but the arch code

3737

* This is needed only for protection faults but the arch code

3738

* is not yet telling us if this is a protection fault or not.

3738

* is not yet telling us if this is a protection fault or not.

3739

* This still avoids useless tlb flushes for .text page faults

3739

* This still avoids useless tlb flushes for .text page faults

3740

* with threads.

3740

* with threads.

3741

*/

3741

*/

3742

if (flags & FAULT_FLAG_WRITE)

3742

if (flags & FAULT_FLAG_WRITE)

3743

flush_tlb_fix_spurious_fault(vma, address);

3743

flush_tlb_fix_spurious_fault(vma, address);

3744

}

3744

}

3745

unlock:

3745

unlock:

3746

pte_unmap_unlock(pte, ptl);

3746

pte_unmap_unlock(pte, ptl);

3747

return 0;

3747

return 0;

3748

}

3748

}

3749

3750

/*

3750

/*

3751

* By the time we get here, we already hold the mm semaphore

3751

* By the time we get here, we already hold the mm semaphore

3752

*/

3752

*/

3753

static int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,

3753

static int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,

3754

unsigned long address, unsigned int flags)

3754

unsigned long address, unsigned int flags)

3755

{

3755

{

3756

pgd_t *pgd;

3756

pgd_t *pgd;

3757

pud_t *pud;

3757

pud_t *pud;

3758

pmd_t *pmd;

3758

pmd_t *pmd;

3759

pte_t *pte;

3759

pte_t *pte;

3760

3761

if (unlikely(is_vm_hugetlb_page(vma)))

3761

if (unlikely(is_vm_hugetlb_page(vma)))

3762

return hugetlb_fault(mm, vma, address, flags);

3762

return hugetlb_fault(mm, vma, address, flags);

3763

3764

pgd = pgd_offset(mm, address);

3764

pgd = pgd_offset(mm, address);

3765

pud = pud_alloc(mm, pgd, address);

3765

pud = pud_alloc(mm, pgd, address);

3766

if (!pud)

3766

if (!pud)

3767

return VM_FAULT_OOM;

3767

return VM_FAULT_OOM;

3768

pmd = pmd_alloc(mm, pud, address);

3768

pmd = pmd_alloc(mm, pud, address);

3769

if (!pmd)

3769

if (!pmd)

3770

return VM_FAULT_OOM;

3770

return VM_FAULT_OOM;

3771

if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) {

3771

if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) {

3772

int ret = VM_FAULT_FALLBACK;

3772

int ret = VM_FAULT_FALLBACK;

3773

if (!vma->vm_ops)

3773

if (!vma->vm_ops)

3774

ret = do_huge_pmd_anonymous_page(mm, vma, address,

3774

ret = do_huge_pmd_anonymous_page(mm, vma, address,

3775

pmd, flags);

3775

pmd, flags);

3776

if (!(ret & VM_FAULT_FALLBACK))

3776

if (!(ret & VM_FAULT_FALLBACK))

3777

return ret;

3777

return ret;

3778

} else {

3778

} else {

3779

pmd_t orig_pmd = *pmd;

3779

pmd_t orig_pmd = *pmd;

3780

int ret;

3780

int ret;

3781

3782

barrier();

3782

barrier();

3783

if (pmd_trans_huge(orig_pmd)) {

3783

if (pmd_trans_huge(orig_pmd)) {

3784

unsigned int dirty = flags & FAULT_FLAG_WRITE;

3784

unsigned int dirty = flags & FAULT_FLAG_WRITE;

3785

3786

/*

3786

/*

3787

* If the pmd is splitting, return and retry the

3787

* If the pmd is splitting, return and retry the

3788

* the fault. Alternative: wait until the split

3788

* the fault. Alternative: wait until the split

3789

* is done, and goto retry.

3789

* is done, and goto retry.

3790

*/

3790

*/

3791

if (pmd_trans_splitting(orig_pmd))

3791

if (pmd_trans_splitting(orig_pmd))

3792

return 0;

3792

return 0;

3793

3794

if (pmd_numa(orig_pmd))

3794

if (pmd_numa(orig_pmd))

3795

return do_huge_pmd_numa_page(mm, vma, address,

3795

return do_huge_pmd_numa_page(mm, vma, address,

3796

orig_pmd, pmd);

3796

orig_pmd, pmd);

3797

3798

if (dirty && !pmd_write(orig_pmd)) {

3798

if (dirty && !pmd_write(orig_pmd)) {

3799

ret = do_huge_pmd_wp_page(mm, vma, address, pmd,

3799

ret = do_huge_pmd_wp_page(mm, vma, address, pmd,

3800

orig_pmd);

3800

orig_pmd);

3801

if (!(ret & VM_FAULT_FALLBACK))

3801

if (!(ret & VM_FAULT_FALLBACK))

3802

return ret;

3802

return ret;

3803

} else {

3803

} else {

3804

huge_pmd_set_accessed(mm, vma, address, pmd,

3804

huge_pmd_set_accessed(mm, vma, address, pmd,

3805

orig_pmd, dirty);

3805

orig_pmd, dirty);

3806

return 0;

3806

return 0;

3807

}

3807

}

3808

}

3808

}

3809

}

3809

}

3810

3811

if (pmd_numa(*pmd))

3811

if (pmd_numa(*pmd))

3812

return do_pmd_numa_page(mm, vma, address, pmd);

3812

return do_pmd_numa_page(mm, vma, address, pmd);

3813

3814

/*

3814

/*

3815

* Use __pte_alloc instead of pte_alloc_map, because we can't

3815

* Use __pte_alloc instead of pte_alloc_map, because we can't

3816

* run pte_offset_map on the pmd, if an huge pmd could

3816

* run pte_offset_map on the pmd, if an huge pmd could

3817

* materialize from under us from a different thread.

3817

* materialize from under us from a different thread.

3818

*/

3818

*/

3819

if (unlikely(pmd_none(*pmd)) &&

3819

if (unlikely(pmd_none(*pmd)) &&

3820

unlikely(__pte_alloc(mm, vma, pmd, address)))

3820

unlikely(__pte_alloc(mm, vma, pmd, address)))

3821

return VM_FAULT_OOM;

3821

return VM_FAULT_OOM;

3822

/* if an huge pmd materialized from under us just retry later */

3822

/* if an huge pmd materialized from under us just retry later */

3823

if (unlikely(pmd_trans_huge(*pmd)))

3823

if (unlikely(pmd_trans_huge(*pmd)))

3824

return 0;

3824

return 0;

3825

/*

3825

/*

3826

* A regular pmd is established and it can't morph into a huge pmd

3826

* A regular pmd is established and it can't morph into a huge pmd

3827

* from under us anymore at this point because we hold the mmap_sem

3827

* from under us anymore at this point because we hold the mmap_sem

3828

* read mode and khugepaged takes it in write mode. So now it's

3828

* read mode and khugepaged takes it in write mode. So now it's

3829

* safe to run pte_offset_map().

3829

* safe to run pte_offset_map().

3830

*/

3830

*/

3831

pte = pte_offset_map(pmd, address);

3831

pte = pte_offset_map(pmd, address);

3832

3833

return handle_pte_fault(mm, vma, address, pte, pmd, flags);

3833

return handle_pte_fault(mm, vma, address, pte, pmd, flags);

3834

}

3834

}

3835

3836

int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,

3836

int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,

3837

unsigned long address, unsigned int flags)

3837

unsigned long address, unsigned int flags)

3838

{

3838

{

3839

int ret;

3839

int ret;

3840

3841

__set_current_state(TASK_RUNNING);

3841

__set_current_state(TASK_RUNNING);

3842

3843

count_vm_event(PGFAULT);

3843

count_vm_event(PGFAULT);

3844

mem_cgroup_count_vm_event(mm, PGFAULT);

3844

mem_cgroup_count_vm_event(mm, PGFAULT);

3845

3846

/* do counter updates before entering really critical section. */

3846

/* do counter updates before entering really critical section. */

3847

check_sync_rss_stat(current);

3847

check_sync_rss_stat(current);

3848

3849

/*

3849

/*

3850

* Enable the memcg OOM handling for faults triggered in user

3850

* Enable the memcg OOM handling for faults triggered in user

3851

* space. Kernel faults are handled more gracefully.

3851

* space. Kernel faults are handled more gracefully.

3852

*/

3852

*/

3853

if (flags & FAULT_FLAG_USER)

3853

if (flags & FAULT_FLAG_USER)

3854

mem_cgroup_oom_enable();

3854

mem_cgroup_oom_enable();

3855

3856

ret = __handle_mm_fault(mm, vma, address, flags);

3856

ret = __handle_mm_fault(mm, vma, address, flags);

3857

3858

if (flags & FAULT_FLAG_USER) {

3858

if (flags & FAULT_FLAG_USER) {

3859

mem_cgroup_oom_disable();

3859

mem_cgroup_oom_disable();

3860

/*

3860

/*

3861

* The task may have entered a memcg OOM situation but

3861

* The task may have entered a memcg OOM situation but

3862

* if the allocation error was handled gracefully (no

3862

* if the allocation error was handled gracefully (no

3863

* VM_FAULT_OOM), there is no need to kill anything.

3863

* VM_FAULT_OOM), there is no need to kill anything.

3864

* Just clean up the OOM state peacefully.

3864

* Just clean up the OOM state peacefully.

3865

*/

3865

*/

3866

if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM))

3866

if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM))

3867

mem_cgroup_oom_synchronize(false);

3867

mem_cgroup_oom_synchronize(false);

3868

}

3868

}

3869

3870

return ret;

3870

return ret;

3871

}

3871

}

3872

3873

#ifndef __PAGETABLE_PUD_FOLDED

3873

#ifndef __PAGETABLE_PUD_FOLDED

3874

/*

3874

/*

3875

* Allocate page upper directory.

3875

* Allocate page upper directory.

3876

* We've already handled the fast-path in-line.

3876

* We've already handled the fast-path in-line.

3877

*/

3877

*/

3878

int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)

3878

int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)

3879

{

3879

{

3880

pud_t *new = pud_alloc_one(mm, address);

3880

pud_t *new = pud_alloc_one(mm, address);

3881

if (!new)

3881

if (!new)

3882

return -ENOMEM;

3882

return -ENOMEM;

3883

3884

smp_wmb(); /* See comment in __pte_alloc */

3884

smp_wmb(); /* See comment in __pte_alloc */

3885

3886

spin_lock(&mm->page_table_lock);

3886

spin_lock(&mm->page_table_lock);

3887

if (pgd_present(*pgd)) /* Another has populated it */

3887

if (pgd_present(*pgd)) /* Another has populated it */

3888

pud_free(mm, new);

3888

pud_free(mm, new);

3889

else

3889

else

3890

pgd_populate(mm, pgd, new);

3890

pgd_populate(mm, pgd, new);

3891

spin_unlock(&mm->page_table_lock);

3891

spin_unlock(&mm->page_table_lock);

3892

return 0;

3892

return 0;

3893

}

3893

}

3894

#endif /* __PAGETABLE_PUD_FOLDED */

3894

#endif /* __PAGETABLE_PUD_FOLDED */

3895

3896

#ifndef __PAGETABLE_PMD_FOLDED

3896

#ifndef __PAGETABLE_PMD_FOLDED

3897

/*

3897

/*

3898

* Allocate page middle directory.

3898

* Allocate page middle directory.

3899

* We've already handled the fast-path in-line.

3899

* We've already handled the fast-path in-line.

3900

*/

3900

*/

3901

int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)

3901

int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)

3902

{

3902

{

3903

pmd_t *new = pmd_alloc_one(mm, address);

3903

pmd_t *new = pmd_alloc_one(mm, address);

3904

if (!new)

3904

if (!new)

3905

return -ENOMEM;

3905

return -ENOMEM;

3906

3907

smp_wmb(); /* See comment in __pte_alloc */

3907

smp_wmb(); /* See comment in __pte_alloc */

3908

3909

spin_lock(&mm->page_table_lock);

3909

spin_lock(&mm->page_table_lock);

3910

#ifndef __ARCH_HAS_4LEVEL_HACK

3910

#ifndef __ARCH_HAS_4LEVEL_HACK

3911

if (pud_present(*pud)) /* Another has populated it */

3911

if (pud_present(*pud)) /* Another has populated it */

3912

pmd_free(mm, new);

3912

pmd_free(mm, new);

3913

else

3913

else

3914

pud_populate(mm, pud, new);

3914

pud_populate(mm, pud, new);

3915

#else

3915

#else

3916

if (pgd_present(*pud)) /* Another has populated it */

3916

if (pgd_present(*pud)) /* Another has populated it */

3917

pmd_free(mm, new);

3917

pmd_free(mm, new);

3918

else

3918

else

3919

pgd_populate(mm, pud, new);

3919

pgd_populate(mm, pud, new);

3920

#endif /* __ARCH_HAS_4LEVEL_HACK */

3920

#endif /* __ARCH_HAS_4LEVEL_HACK */

3921

spin_unlock(&mm->page_table_lock);

3921

spin_unlock(&mm->page_table_lock);

3922

return 0;

3922

return 0;

3923

}

3923

}

3924

#endif /* __PAGETABLE_PMD_FOLDED */

3924

#endif /* __PAGETABLE_PMD_FOLDED */

3925

3926

#if !defined(__HAVE_ARCH_GATE_AREA)

3926

#if !defined(__HAVE_ARCH_GATE_AREA)

3927

3928

#if defined(AT_SYSINFO_EHDR)

3928

#if defined(AT_SYSINFO_EHDR)

3929

static struct vm_area_struct gate_vma;

3929

static struct vm_area_struct gate_vma;

3930

3931

static int __init gate_vma_init(void)

3931

static int __init gate_vma_init(void)

3932

{

3932

{

3933

gate_vma.vm_mm = NULL;

3933

gate_vma.vm_mm = NULL;

3934

gate_vma.vm_start = FIXADDR_USER_START;

3934

gate_vma.vm_start = FIXADDR_USER_START;

3935

gate_vma.vm_end = FIXADDR_USER_END;

3935

gate_vma.vm_end = FIXADDR_USER_END;

3936

gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;

3936

gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;

3937

gate_vma.vm_page_prot = __P101;

3937

gate_vma.vm_page_prot = __P101;

3938

3939

return 0;

3939

return 0;

3940

}

3940

}

3941

__initcall(gate_vma_init);

3941

__initcall(gate_vma_init);

3942

#endif

3942

#endif

3943

3944

struct vm_area_struct *get_gate_vma(struct mm_struct *mm)

3944

struct vm_area_struct *get_gate_vma(struct mm_struct *mm)

3945

{

3945

{

3946

#ifdef AT_SYSINFO_EHDR

3946

#ifdef AT_SYSINFO_EHDR

3947

return &gate_vma;

3947

return &gate_vma;

3948

#else

3948

#else

3949

return NULL;

3949

return NULL;

3950

#endif

3950

#endif

3951

}

3951

}

3952

3953

int in_gate_area_no_mm(unsigned long addr)

3953

int in_gate_area_no_mm(unsigned long addr)

3954

{

3954

{

3955

#ifdef AT_SYSINFO_EHDR

3955

#ifdef AT_SYSINFO_EHDR

3956

if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))

3956

if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))

3957

return 1;

3957

return 1;

3958

#endif

3958

#endif

3959

return 0;

3959

return 0;

3960

}

3960

}

3961

3962

#endif /* __HAVE_ARCH_GATE_AREA */

3962

#endif /* __HAVE_ARCH_GATE_AREA */

3963

3964

static int __follow_pte(struct mm_struct *mm, unsigned long address,

3964

static int __follow_pte(struct mm_struct *mm, unsigned long address,

3965

pte_t **ptepp, spinlock_t **ptlp)

3965

pte_t **ptepp, spinlock_t **ptlp)

3966

{

3966

{

3967

pgd_t *pgd;

3967

pgd_t *pgd;

3968

pud_t *pud;

3968

pud_t *pud;

3969

pmd_t *pmd;

3969

pmd_t *pmd;

3970

pte_t *ptep;

3970

pte_t *ptep;

3971

3972

pgd = pgd_offset(mm, address);

3972

pgd = pgd_offset(mm, address);

3973

if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))

3973

if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))

3974

goto out;

3974

goto out;

3975

3976

pud = pud_offset(pgd, address);

3976

pud = pud_offset(pgd, address);

3977

if (pud_none(*pud) || unlikely(pud_bad(*pud)))

3977

if (pud_none(*pud) || unlikely(pud_bad(*pud)))

3978

goto out;

3978

goto out;

3979

3980

pmd = pmd_offset(pud, address);

3980

pmd = pmd_offset(pud, address);

3981

VM_BUG_ON(pmd_trans_huge(*pmd));

3981

VM_BUG_ON(pmd_trans_huge(*pmd));

3982

if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))

3982

if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))

3983

goto out;

3983

goto out;

3984

3985

/* We cannot handle huge page PFN maps. Luckily they don't exist. */

3985

/* We cannot handle huge page PFN maps. Luckily they don't exist. */

3986

if (pmd_huge(*pmd))

3986

if (pmd_huge(*pmd))

3987

goto out;

3987

goto out;

3988

3989

ptep = pte_offset_map_lock(mm, pmd, address, ptlp);

3989

ptep = pte_offset_map_lock(mm, pmd, address, ptlp);

3990

if (!ptep)

3990

if (!ptep)

3991

goto out;

3991

goto out;

3992

if (!pte_present(*ptep))

3992

if (!pte_present(*ptep))

3993

goto unlock;

3993

goto unlock;

3994

*ptepp = ptep;

3994

*ptepp = ptep;

3995

return 0;

3995

return 0;

3996

unlock:

3996

unlock:

3997

pte_unmap_unlock(ptep, *ptlp);

3997

pte_unmap_unlock(ptep, *ptlp);

3998

out:

3998

out:

3999

return -EINVAL;

3999

return -EINVAL;

4000

}

4000

}

4001

4002

static inline int follow_pte(struct mm_struct *mm, unsigned long address,

4002

static inline int follow_pte(struct mm_struct *mm, unsigned long address,

4003

pte_t **ptepp, spinlock_t **ptlp)

4003

pte_t **ptepp, spinlock_t **ptlp)

4004

{

4004

{

4005

int res;

4005

int res;

4006

4007

/* (void) is needed to make gcc happy */

4007

/* (void) is needed to make gcc happy */

4008

(void) __cond_lock(*ptlp,

4008

(void) __cond_lock(*ptlp,

4009

!(res = __follow_pte(mm, address, ptepp, ptlp)));

4009

!(res = __follow_pte(mm, address, ptepp, ptlp)));

4010

return res;

4010

return res;

4011

}

4011

}

4012

4013

/**

4013

/**

4014

* follow_pfn - look up PFN at a user virtual address

4014

* follow_pfn - look up PFN at a user virtual address

4015

* @vma: memory mapping

4015

* @vma: memory mapping

4016

* @address: user virtual address

4016

* @address: user virtual address

4017

* @pfn: location to store found PFN

4017

* @pfn: location to store found PFN

4018

*

4018

*

4019

* Only IO mappings and raw PFN mappings are allowed.

4019

* Only IO mappings and raw PFN mappings are allowed.

4020

*

4020

*

4021

* Returns zero and the pfn at @pfn on success, -ve otherwise.

4021

* Returns zero and the pfn at @pfn on success, -ve otherwise.

4022

*/

4022

*/

4023

int follow_pfn(struct vm_area_struct *vma, unsigned long address,

4023

int follow_pfn(struct vm_area_struct *vma, unsigned long address,

4024

unsigned long *pfn)

4024

unsigned long *pfn)

4025

{

4025

{

4026

int ret = -EINVAL;

4026

int ret = -EINVAL;

4027

spinlock_t *ptl;

4027

spinlock_t *ptl;

4028

pte_t *ptep;

4028

pte_t *ptep;

4029

4030

if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))

4030

if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))

4031

return ret;

4031

return ret;

4032

4033

ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);

4033

ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);

4034

if (ret)

4034

if (ret)

4035

return ret;

4035

return ret;

4036

*pfn = pte_pfn(*ptep);

4036

*pfn = pte_pfn(*ptep);

4037

pte_unmap_unlock(ptep, ptl);

4037

pte_unmap_unlock(ptep, ptl);

4038

return 0;

4038

return 0;

4039

}

4039

}

4040

EXPORT_SYMBOL(follow_pfn);

4040

EXPORT_SYMBOL(follow_pfn);

4041

4042

#ifdef CONFIG_HAVE_IOREMAP_PROT

4042

#ifdef CONFIG_HAVE_IOREMAP_PROT

4043

int follow_phys(struct vm_area_struct *vma,

4043

int follow_phys(struct vm_area_struct *vma,

4044

unsigned long address, unsigned int flags,

4044

unsigned long address, unsigned int flags,

4045

unsigned long *prot, resource_size_t *phys)

4045

unsigned long *prot, resource_size_t *phys)

4046

{

4046

{

4047

int ret = -EINVAL;

4047

int ret = -EINVAL;

4048

pte_t *ptep, pte;

4048

pte_t *ptep, pte;

4049

spinlock_t *ptl;

4049

spinlock_t *ptl;

4050

4051

if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))

4051

if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))

4052

goto out;

4052

goto out;

4053

4054

if (follow_pte(vma->vm_mm, address, &ptep, &ptl))

4054

if (follow_pte(vma->vm_mm, address, &ptep, &ptl))

4055

goto out;

4055

goto out;

4056

pte = *ptep;

4056

pte = *ptep;

4057

4058

if ((flags & FOLL_WRITE) && !pte_write(pte))

4058

if ((flags & FOLL_WRITE) && !pte_write(pte))

4059

goto unlock;

4059

goto unlock;

4060

4061

*prot = pgprot_val(pte_pgprot(pte));

4061

*prot = pgprot_val(pte_pgprot(pte));

4062

*phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;

4062

*phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;

4063

4064

ret = 0;

4064

ret = 0;

4065

unlock:

4065

unlock:

4066

pte_unmap_unlock(ptep, ptl);

4066

pte_unmap_unlock(ptep, ptl);

4067

out:

4067

out:

4068

return ret;

4068

return ret;

4069

}

4069

}

4070

4071

int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,

4071

int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,

4072

void *buf, int len, int write)

4072

void *buf, int len, int write)

4073

{

4073

{

4074

resource_size_t phys_addr;

4074

resource_size_t phys_addr;

4075

unsigned long prot = 0;

4075

unsigned long prot = 0;

4076

void __iomem *maddr;

4076

void __iomem *maddr;

4077

int offset = addr & (PAGE_SIZE-1);

4077

int offset = addr & (PAGE_SIZE-1);

4078

4079

if (follow_phys(vma, addr, write, &prot, &phys_addr))

4079

if (follow_phys(vma, addr, write, &prot, &phys_addr))

4080

return -EINVAL;

4080

return -EINVAL;

4081

4082

maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot);

4082

maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot);

4083

if (write)

4083

if (write)

4084

memcpy_toio(maddr + offset, buf, len);

4084

memcpy_toio(maddr + offset, buf, len);

4085

else

4085

else

4086

memcpy_fromio(buf, maddr + offset, len);

4086

memcpy_fromio(buf, maddr + offset, len);

4087

iounmap(maddr);

4087

iounmap(maddr);

4088

4089

return len;

4089

return len;

4090

}

4090

}

4091

EXPORT_SYMBOL_GPL(generic_access_phys);

4091

EXPORT_SYMBOL_GPL(generic_access_phys);

4092

#endif

4092

#endif

4093

4094

/*

4094

/*

4095

* Access another process' address space as given in mm. If non-NULL, use the

4095

* Access another process' address space as given in mm. If non-NULL, use the

4096

* given task for page fault accounting.

4096

* given task for page fault accounting.

4097

*/

4097

*/

4098

static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,

4098

static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,

4099

unsigned long addr, void *buf, int len, int write)

4099

unsigned long addr, void *buf, int len, int write)

4100

{

4100

{

4101

struct vm_area_struct *vma;

4101

struct vm_area_struct *vma;

4102

void *old_buf = buf;

4102

void *old_buf = buf;

4103

4104

down_read(&mm->mmap_sem);

4104

down_read(&mm->mmap_sem);

4105

/* ignore errors, just check how much was successfully transferred */

4105

/* ignore errors, just check how much was successfully transferred */

4106

while (len) {

4106

while (len) {

4107

int bytes, ret, offset;

4107

int bytes, ret, offset;

4108

void *maddr;

4108

void *maddr;

4109

struct page *page = NULL;

4109

struct page *page = NULL;

4110

4111

ret = get_user_pages(tsk, mm, addr, 1,

4111

ret = get_user_pages(tsk, mm, addr, 1,

4112

write, 1, &page, &vma);

4112

write, 1, &page, &vma);

4113

if (ret <= 0) {

4113

if (ret <= 0) {

4114

/*

4114

/*

4115

* Check if this is a VM_IO | VM_PFNMAP VMA, which

4115

* Check if this is a VM_IO | VM_PFNMAP VMA, which

4116

* we can access using slightly different code.

4116

* we can access using slightly different code.

4117

*/

4117

*/

4118

#ifdef CONFIG_HAVE_IOREMAP_PROT

4118

#ifdef CONFIG_HAVE_IOREMAP_PROT

4119

vma = find_vma(mm, addr);

4119

vma = find_vma(mm, addr);

4120

if (!vma || vma->vm_start > addr)

4120

if (!vma || vma->vm_start > addr)

4121

break;

4121

break;

4122

if (vma->vm_ops && vma->vm_ops->access)

4122

if (vma->vm_ops && vma->vm_ops->access)

4123

ret = vma->vm_ops->access(vma, addr, buf,

4123

ret = vma->vm_ops->access(vma, addr, buf,

4124

len, write);

4124

len, write);

4125

if (ret <= 0)

4125

if (ret <= 0)

4126

#endif

4126

#endif

4127

break;

4127

break;

4128

bytes = ret;

4128

bytes = ret;

4129

} else {

4129

} else {

4130

bytes = len;

4130

bytes = len;

4131

offset = addr & (PAGE_SIZE-1);

4131

offset = addr & (PAGE_SIZE-1);

4132

if (bytes > PAGE_SIZE-offset)

4132

if (bytes > PAGE_SIZE-offset)

4133

bytes = PAGE_SIZE-offset;

4133

bytes = PAGE_SIZE-offset;

4134

4135

maddr = kmap(page);

4135

maddr = kmap(page);

4136

if (write) {

4136

if (write) {

4137

copy_to_user_page(vma, page, addr,

4137

copy_to_user_page(vma, page, addr,

4138

maddr + offset, buf, bytes);

4138

maddr + offset, buf, bytes);

4139

set_page_dirty_lock(page);

4139

set_page_dirty_lock(page);

4140

} else {

4140

} else {

4141

copy_from_user_page(vma, page, addr,

4141

copy_from_user_page(vma, page, addr,

4142

buf, maddr + offset, bytes);

4142

buf, maddr + offset, bytes);

4143

}

4143

}

4144

kunmap(page);

4144

kunmap(page);

4145

page_cache_release(page);

4145

page_cache_release(page);

4146

}

4146

}

4147

len -= bytes;

4147

len -= bytes;

4148

buf += bytes;

4148

buf += bytes;

4149

addr += bytes;

4149

addr += bytes;

4150

}

4150

}

4151

up_read(&mm->mmap_sem);

4151

up_read(&mm->mmap_sem);

4152

4153

return buf - old_buf;

4153

return buf - old_buf;

4154

}

4154

}

4155

4156

/**

4156

/**

4157

* access_remote_vm - access another process' address space

4157

* access_remote_vm - access another process' address space

4158

* @mm: the mm_struct of the target address space

4158

* @mm: the mm_struct of the target address space

4159

* @addr: start address to access

4159

* @addr: start address to access

4160

* @buf: source or destination buffer

4160

* @buf: source or destination buffer

4161

* @len: number of bytes to transfer

4161

* @len: number of bytes to transfer

4162

* @write: whether the access is a write

4162

* @write: whether the access is a write

4163

*

4163

*

4164

* The caller must hold a reference on @mm.

4164

* The caller must hold a reference on @mm.

4165

*/

4165

*/

4166

int access_remote_vm(struct mm_struct *mm, unsigned long addr,

4166

int access_remote_vm(struct mm_struct *mm, unsigned long addr,

4167

void *buf, int len, int write)

4167

void *buf, int len, int write)

4168

{

4168

{

4169

return __access_remote_vm(NULL, mm, addr, buf, len, write);

4169

return __access_remote_vm(NULL, mm, addr, buf, len, write);

4170

}

4170

}

4171

4172

/*

4172

/*

4173

* Access another process' address space.

4173

* Access another process' address space.

4174

* Source/target buffer must be kernel space,

4174

* Source/target buffer must be kernel space,

4175

* Do not walk the page table directly, use get_user_pages

4175

* Do not walk the page table directly, use get_user_pages

4176

*/

4176

*/

4177

int access_process_vm(struct task_struct *tsk, unsigned long addr,

4177

int access_process_vm(struct task_struct *tsk, unsigned long addr,

4178

void *buf, int len, int write)

4178

void *buf, int len, int write)

4179

{

4179

{

4180

struct mm_struct *mm;

4180

struct mm_struct *mm;

4181

int ret;

4181

int ret;

4182

4183

mm = get_task_mm(tsk);

4183

mm = get_task_mm(tsk);

4184

if (!mm)

4184

if (!mm)

4185

return 0;

4185

return 0;

4186

4187

ret = __access_remote_vm(tsk, mm, addr, buf, len, write);

4187

ret = __access_remote_vm(tsk, mm, addr, buf, len, write);

4188

mmput(mm);

4188

mmput(mm);

4189

4190

return ret;

4190

return ret;

4191

}

4191

}

4192

4193

/*

4193

/*

4194

* Print the name of a VMA.

4194

* Print the name of a VMA.

4195

*/

4195

*/

4196

void print_vma_addr(char *prefix, unsigned long ip)

4196

void print_vma_addr(char *prefix, unsigned long ip)

4197

{

4197

{

4198

struct mm_struct *mm = current->mm;

4198

struct mm_struct *mm = current->mm;

4199

struct vm_area_struct *vma;

4199

struct vm_area_struct *vma;

4200

4201

/*

4201

/*

4202

* Do not print if we are in atomic

4202

* Do not print if we are in atomic

4203

* contexts (in exception stacks, etc.):

4203

* contexts (in exception stacks, etc.):

4204

*/

4204

*/

4205

if (preempt_count())

4205

if (preempt_count())

4206

return;

4206

return;

4207

4208

down_read(&mm->mmap_sem);

4208

down_read(&mm->mmap_sem);

4209

vma = find_vma(mm, ip);

4209

vma = find_vma(mm, ip);

4210

if (vma && vma->vm_file) {

4210

if (vma && vma->vm_file) {

4211

struct file *f = vma->vm_file;

4211

struct file *f = vma->vm_file;

4212

char *buf = (char *)__get_free_page(GFP_KERNEL);

4212

char *buf = (char *)__get_free_page(GFP_KERNEL);

4213

if (buf) {

4213

if (buf) {

4214

char *p;

4214

char *p;

4215

4216

p = d_path(&f->f_path, buf, PAGE_SIZE);

4216

p = d_path(&f->f_path, buf, PAGE_SIZE);

4217

if (IS_ERR(p))

4217

if (IS_ERR(p))

4218

p = "?";

4218

p = "?";

4219

printk("%s%s[%lx+%lx]", prefix, kbasename(p),

4219

printk("%s%s[%lx+%lx]", prefix, kbasename(p),

4220

vma->vm_start,

4220

vma->vm_start,

4221

vma->vm_end - vma->vm_start);

4221

vma->vm_end - vma->vm_start);

4222

free_page((unsigned long)buf);

4222

free_page((unsigned long)buf);

4223

}

4223

}

4224

}

4224

}

4225

up_read(&mm->mmap_sem);

4225

up_read(&mm->mmap_sem);

4226

}

4226

}

4227

4228

#if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP)

4228

#if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP)

4229

void might_fault(void)

4229

void might_fault(void)

4230

{

4230

{

4231

/*

4231

/*

4232

* Some code (nfs/sunrpc) uses socket ops on kernel memory while

4232

* Some code (nfs/sunrpc) uses socket ops on kernel memory while

4233

* holding the mmap_sem, this is safe because kernel memory doesn't

4233

* holding the mmap_sem, this is safe because kernel memory doesn't

4234

* get paged out, therefore we'll never actually fault, and the

4234

* get paged out, therefore we'll never actually fault, and the

4235

* below annotations will generate false positives.

4235

* below annotations will generate false positives.

4236

*/

4236

*/

4237

if (segment_eq(get_fs(), KERNEL_DS))

4237

if (segment_eq(get_fs(), KERNEL_DS))

4238

return;

4238

return;

4239

4240

/*

4240

/*

4241

* it would be nicer only to annotate paths which are not under

4241

* it would be nicer only to annotate paths which are not under

4242

* pagefault_disable, however that requires a larger audit and

4242

* pagefault_disable, however that requires a larger audit and

4243

* providing helpers like get_user_atomic.

4243

* providing helpers like get_user_atomic.

4244

*/

4244

*/

4245

if (in_atomic())

4245

if (in_atomic())

4246

return;

4246

return;

4247

4248

__might_sleep(__FILE__, __LINE__, 0);

4248

__might_sleep(__FILE__, __LINE__, 0);

4249

4250

if (current->mm)

4250

if (current->mm)

4251

might_lock_read(&current->mm->mmap_sem);

4251

might_lock_read(&current->mm->mmap_sem);

4252

}

4252

}

4253

EXPORT_SYMBOL(might_fault);

4253

EXPORT_SYMBOL(might_fault);

4254

#endif

4254

#endif

4255

4256

#if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)

4256

#if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)

4257

static void clear_gigantic_page(struct page *page,

4257

static void clear_gigantic_page(struct page *page,

4258

unsigned long addr,

4258

unsigned long addr,

4259

unsigned int pages_per_huge_page)

4259

unsigned int pages_per_huge_page)

4260

{

4260

{

4261

int i;

4261

int i;

4262

struct page *p = page;

4262

struct page *p = page;

4263

4264

might_sleep();

4264

might_sleep();

4265

for (i = 0; i < pages_per_huge_page;

4265

for (i = 0; i < pages_per_huge_page;

4266

i++, p = mem_map_next(p, page, i)) {

4266

i++, p = mem_map_next(p, page, i)) {

4267

cond_resched();

4267

cond_resched();

4268

clear_user_highpage(p, addr + i * PAGE_SIZE);

4268

clear_user_highpage(p, addr + i * PAGE_SIZE);

4269

}

4269

}

4270

}

4270

}

4271

void clear_huge_page(struct page *page,

4271

void clear_huge_page(struct page *page,

4272

unsigned long addr, unsigned int pages_per_huge_page)

4272

unsigned long addr, unsigned int pages_per_huge_page)

4273

{

4273

{

4274

int i;

4274

int i;

4275

4276

if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {

4276

if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {

4277

clear_gigantic_page(page, addr, pages_per_huge_page);

4277

clear_gigantic_page(page, addr, pages_per_huge_page);

4278

return;

4278

return;

4279

}

4279

}

4280

4281

might_sleep();

4281

might_sleep();

4282

for (i = 0; i < pages_per_huge_page; i++) {

4282

for (i = 0; i < pages_per_huge_page; i++) {

4283

cond_resched();

4283

cond_resched();

4284

clear_user_highpage(page + i, addr + i * PAGE_SIZE);

4284

clear_user_highpage(page + i, addr + i * PAGE_SIZE);

4285

}

4285

}

4286

}

4286

}

4287

4288

static void copy_user_gigantic_page(struct page *dst, struct page *src,

4288

static void copy_user_gigantic_page(struct page *dst, struct page *src,

4289

unsigned long addr,

4289

unsigned long addr,

4290

struct vm_area_struct *vma,

4290

struct vm_area_struct *vma,

4291

unsigned int pages_per_huge_page)

4291

unsigned int pages_per_huge_page)

4292

{

4292

{

4293

int i;

4293

int i;

4294

struct page *dst_base = dst;

4294

struct page *dst_base = dst;

4295

struct page *src_base = src;

4295

struct page *src_base = src;

4296

4297

for (i = 0; i < pages_per_huge_page; ) {

4297

for (i = 0; i < pages_per_huge_page; ) {

4298

cond_resched();

4298

cond_resched();

4299

copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma);

4299

copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma);

4300

4301

i++;

4301

i++;

4302

dst = mem_map_next(dst, dst_base, i);

4302

dst = mem_map_next(dst, dst_base, i);

4303

src = mem_map_next(src, src_base, i);

4303

src = mem_map_next(src, src_base, i);

4304

}

4304

}

4305

}

4305

}

4306

4307

void copy_user_huge_page(struct page *dst, struct page *src,

4307

void copy_user_huge_page(struct page *dst, struct page *src,

4308

unsigned long addr, struct vm_area_struct *vma,

4308

unsigned long addr, struct vm_area_struct *vma,

4309

unsigned int pages_per_huge_page)

4309

unsigned int pages_per_huge_page)

4310

{

4310

{

4311

int i;

4311

int i;

4312

4313

if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {

4313

if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {

4314

copy_user_gigantic_page(dst, src, addr, vma,

4314

copy_user_gigantic_page(dst, src, addr, vma,

4315

pages_per_huge_page);

4315

pages_per_huge_page);

4316

return;

4316

return;

4317

}

4317

}

4318

4319

might_sleep();

4319

might_sleep();

4320

for (i = 0; i < pages_per_huge_page; i++) {

4320

for (i = 0; i < pages_per_huge_page; i++) {

4321

cond_resched();

4321

cond_resched();

4322

copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma);

4322

copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma);

4323

}

4323

}

4324

}

4324

}

4325

#endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */

4325

#endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */

4326

GITLAB

mm: make copy_pte_range static again

 #ifndef _LINUX_HUGE_MM_H
 #define _LINUX_HUGE_MM_H
 extern int do_huge_pmd_anonymous_page(struct mm_struct *mm,
 				      struct vm_area_struct *vma,
 				      unsigned long address, pmd_t *pmd,
 				      unsigned int flags);
 extern int copy_huge_pmd(struct mm_struct *dst_mm, struct mm_struct *src_mm,
 			 pmd_t *dst_pmd, pmd_t *src_pmd, unsigned long addr,
 			 struct vm_area_struct *vma);
 extern void huge_pmd_set_accessed(struct mm_struct *mm,
 				  struct vm_area_struct *vma,
 				  unsigned long address, pmd_t *pmd,
 				  pmd_t orig_pmd, int dirty);
 extern int do_huge_pmd_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
 			       unsigned long address, pmd_t *pmd,
 			       pmd_t orig_pmd);
 extern struct page *follow_trans_huge_pmd(struct vm_area_struct *vma,
 					  unsigned long addr,
 					  pmd_t *pmd,
 					  unsigned int flags);
 extern int zap_huge_pmd(struct mmu_gather *tlb,
 			struct vm_area_struct *vma,
 			pmd_t *pmd, unsigned long addr);
 extern int mincore_huge_pmd(struct vm_area_struct *vma, pmd_t *pmd,
 			unsigned long addr, unsigned long end,
 			unsigned char *vec);
 extern int move_huge_pmd(struct vm_area_struct *vma,
 			 struct vm_area_struct *new_vma,
 			 unsigned long old_addr,
 			 unsigned long new_addr, unsigned long old_end,
 			 pmd_t *old_pmd, pmd_t *new_pmd);
 extern int change_huge_pmd(struct vm_area_struct *vma, pmd_t *pmd,
 			unsigned long addr, pgprot_t newprot,
 			int prot_numa);
 enum transparent_hugepage_flag {
 	TRANSPARENT_HUGEPAGE_FLAG,
 	TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG,
 	TRANSPARENT_HUGEPAGE_DEFRAG_FLAG,
 	TRANSPARENT_HUGEPAGE_DEFRAG_REQ_MADV_FLAG,
 	TRANSPARENT_HUGEPAGE_DEFRAG_KHUGEPAGED_FLAG,
 	TRANSPARENT_HUGEPAGE_USE_ZERO_PAGE_FLAG,
 #ifdef CONFIG_DEBUG_VM
 	TRANSPARENT_HUGEPAGE_DEBUG_COW_FLAG,
 #endif
 };
 enum page_check_address_pmd_flag {
 	PAGE_CHECK_ADDRESS_PMD_FLAG,
 	PAGE_CHECK_ADDRESS_PMD_NOTSPLITTING_FLAG,
 	PAGE_CHECK_ADDRESS_PMD_SPLITTING_FLAG,
 };
 extern pmd_t *page_check_address_pmd(struct page *page,
 				     struct mm_struct *mm,
 				     unsigned long address,
 				     enum page_check_address_pmd_flag flag);
 #define HPAGE_PMD_ORDER (HPAGE_PMD_SHIFT-PAGE_SHIFT)
 #define HPAGE_PMD_NR (1<<HPAGE_PMD_ORDER)
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 #define HPAGE_PMD_SHIFT PMD_SHIFT
 #define HPAGE_PMD_SIZE	((1UL) << HPAGE_PMD_SHIFT)
 #define HPAGE_PMD_MASK	(~(HPAGE_PMD_SIZE - 1))
 extern bool is_vma_temporary_stack(struct vm_area_struct *vma);
 #define transparent_hugepage_enabled(__vma)				\
 	((transparent_hugepage_flags &					\
 	  (1<<TRANSPARENT_HUGEPAGE_FLAG) ||				\
 	  (transparent_hugepage_flags &					\
 	   (1<<TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG) &&			\
 	   ((__vma)->vm_flags & VM_HUGEPAGE))) &&			\
 	 !((__vma)->vm_flags & VM_NOHUGEPAGE) &&			\
 	 !is_vma_temporary_stack(__vma))
 #define transparent_hugepage_defrag(__vma)				\
 	((transparent_hugepage_flags &					\
 	  (1<<TRANSPARENT_HUGEPAGE_DEFRAG_FLAG)) ||			\
 	 (transparent_hugepage_flags &					\
 	  (1<<TRANSPARENT_HUGEPAGE_DEFRAG_REQ_MADV_FLAG) &&		\
 	  (__vma)->vm_flags & VM_HUGEPAGE))
 #define transparent_hugepage_use_zero_page()				\
 	(transparent_hugepage_flags &					\
 	 (1<<TRANSPARENT_HUGEPAGE_USE_ZERO_PAGE_FLAG))
 #ifdef CONFIG_DEBUG_VM
 #define transparent_hugepage_debug_cow()				\
 	(transparent_hugepage_flags &					\
 	 (1<<TRANSPARENT_HUGEPAGE_DEBUG_COW_FLAG))
 #else /* CONFIG_DEBUG_VM */
 #define transparent_hugepage_debug_cow() 0
 #endif /* CONFIG_DEBUG_VM */
 extern unsigned long transparent_hugepage_flags;
-extern int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
-			  pmd_t *dst_pmd, pmd_t *src_pmd,
-			  struct vm_area_struct *vma,
-			  unsigned long addr, unsigned long end);
 extern int split_huge_page_to_list(struct page *page, struct list_head *list);
 static inline int split_huge_page(struct page *page)
 {
 	return split_huge_page_to_list(page, NULL);
 }
 extern void __split_huge_page_pmd(struct vm_area_struct *vma,
 		unsigned long address, pmd_t *pmd);
 #define split_huge_page_pmd(__vma, __address, __pmd)			\
 	do {								\
 		pmd_t *____pmd = (__pmd);				\
 		if (unlikely(pmd_trans_huge(*____pmd)))			\
 			__split_huge_page_pmd(__vma, __address,		\
 					____pmd);			\
 	}  while (0)
 #define wait_split_huge_page(__anon_vma, __pmd)				\
 	do {								\
 		pmd_t *____pmd = (__pmd);				\
 		anon_vma_lock_write(__anon_vma);			\
 		anon_vma_unlock_write(__anon_vma);			\
 		BUG_ON(pmd_trans_splitting(*____pmd) ||			\
 		       pmd_trans_huge(*____pmd));			\
 	} while (0)
 extern void split_huge_page_pmd_mm(struct mm_struct *mm, unsigned long address,
 		pmd_t *pmd);
 #if HPAGE_PMD_ORDER >= MAX_ORDER
 #error "hugepages can't be allocated by the buddy allocator"
 #endif
 extern int hugepage_madvise(struct vm_area_struct *vma,
 			    unsigned long *vm_flags, int advice);
 extern void __vma_adjust_trans_huge(struct vm_area_struct *vma,
 				    unsigned long start,
 				    unsigned long end,
 				    long adjust_next);
 extern int __pmd_trans_huge_lock(pmd_t *pmd,
 				 struct vm_area_struct *vma);
 /* mmap_sem must be held on entry */
 static inline int pmd_trans_huge_lock(pmd_t *pmd,
 				      struct vm_area_struct *vma)
 {
 	VM_BUG_ON(!rwsem_is_locked(&vma->vm_mm->mmap_sem));
 	if (pmd_trans_huge(*pmd))
 		return __pmd_trans_huge_lock(pmd, vma);
 	else
 		return 0;
 }
 static inline void vma_adjust_trans_huge(struct vm_area_struct *vma,
 					 unsigned long start,
 					 unsigned long end,
 					 long adjust_next)
 {
 	if (!vma->anon_vma || vma->vm_ops)
 		return;
 	__vma_adjust_trans_huge(vma, start, end, adjust_next);
 }
 static inline int hpage_nr_pages(struct page *page)
 {
 	if (unlikely(PageTransHuge(page)))
 		return HPAGE_PMD_NR;
 	return 1;
 }
 extern int do_huge_pmd_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,
 				unsigned long addr, pmd_t pmd, pmd_t *pmdp);
 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
 #define HPAGE_PMD_SHIFT ({ BUILD_BUG(); 0; })
 #define HPAGE_PMD_MASK ({ BUILD_BUG(); 0; })
 #define HPAGE_PMD_SIZE ({ BUILD_BUG(); 0; })
 #define hpage_nr_pages(x) 1
 #define transparent_hugepage_enabled(__vma) 0
 #define transparent_hugepage_flags 0UL
 static inline int
 split_huge_page_to_list(struct page *page, struct list_head *list)
 {
 	return 0;
 }
 static inline int split_huge_page(struct page *page)
 {
 	return 0;
 }
 #define split_huge_page_pmd(__vma, __address, __pmd)	\
 	do { } while (0)
 #define wait_split_huge_page(__anon_vma, __pmd)	\
 	do { } while (0)
 #define split_huge_page_pmd_mm(__mm, __address, __pmd)	\
 	do { } while (0)
 static inline int hugepage_madvise(struct vm_area_struct *vma,
 				   unsigned long *vm_flags, int advice)
 {
 	BUG();
 	return 0;
 }
 static inline void vma_adjust_trans_huge(struct vm_area_struct *vma,
 					 unsigned long start,
 					 unsigned long end,
 					 long adjust_next)
 {
 }
 static inline int pmd_trans_huge_lock(pmd_t *pmd,
 				      struct vm_area_struct *vma)
 {
 	return 0;
 }
 static inline int do_huge_pmd_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,
 					unsigned long addr, pmd_t pmd, pmd_t *pmdp)
 {
 	return 0;
 }
 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 #endif /* _LINUX_HUGE_MM_H */