1 // SPDX-License-Identifier: GPL-2.0-only 1 // SPDX-License-Identifier: GPL-2.0-only
2 /* 2 /*
3 * fs/userfaultfd.c 3 * fs/userfaultfd.c
4 * 4 *
5 * Copyright (C) 2007 Davide Libenzi <davide 5 * Copyright (C) 2007 Davide Libenzi <davidel@xmailserver.org>
6 * Copyright (C) 2008-2009 Red Hat, Inc. 6 * Copyright (C) 2008-2009 Red Hat, Inc.
7 * Copyright (C) 2015 Red Hat, Inc. 7 * Copyright (C) 2015 Red Hat, Inc.
8 * 8 *
9 * Some part derived from fs/eventfd.c (anon 9 * Some part derived from fs/eventfd.c (anon inode setup) and
10 * mm/ksm.c (mm hashing). 10 * mm/ksm.c (mm hashing).
11 */ 11 */
12 12
13 #include <linux/list.h> 13 #include <linux/list.h>
14 #include <linux/hashtable.h> 14 #include <linux/hashtable.h>
15 #include <linux/sched/signal.h> 15 #include <linux/sched/signal.h>
16 #include <linux/sched/mm.h> 16 #include <linux/sched/mm.h>
17 #include <linux/mm.h> 17 #include <linux/mm.h>
18 #include <linux/mm_inline.h> 18 #include <linux/mm_inline.h>
19 #include <linux/mmu_notifier.h> 19 #include <linux/mmu_notifier.h>
20 #include <linux/poll.h> 20 #include <linux/poll.h>
21 #include <linux/slab.h> 21 #include <linux/slab.h>
22 #include <linux/seq_file.h> 22 #include <linux/seq_file.h>
23 #include <linux/file.h> 23 #include <linux/file.h>
24 #include <linux/bug.h> 24 #include <linux/bug.h>
25 #include <linux/anon_inodes.h> 25 #include <linux/anon_inodes.h>
26 #include <linux/syscalls.h> 26 #include <linux/syscalls.h>
27 #include <linux/userfaultfd_k.h> 27 #include <linux/userfaultfd_k.h>
28 #include <linux/mempolicy.h> 28 #include <linux/mempolicy.h>
29 #include <linux/ioctl.h> 29 #include <linux/ioctl.h>
30 #include <linux/security.h> 30 #include <linux/security.h>
31 #include <linux/hugetlb.h> 31 #include <linux/hugetlb.h>
32 #include <linux/swapops.h> 32 #include <linux/swapops.h>
33 #include <linux/miscdevice.h> 33 #include <linux/miscdevice.h>
34 #include <linux/uio.h> <<
35 34
36 static int sysctl_unprivileged_userfaultfd __r 35 static int sysctl_unprivileged_userfaultfd __read_mostly;
37 36
38 #ifdef CONFIG_SYSCTL 37 #ifdef CONFIG_SYSCTL
39 static struct ctl_table vm_userfaultfd_table[] 38 static struct ctl_table vm_userfaultfd_table[] = {
40 { 39 {
41 .procname = "unprivilege 40 .procname = "unprivileged_userfaultfd",
42 .data = &sysctl_unpr 41 .data = &sysctl_unprivileged_userfaultfd,
43 .maxlen = sizeof(sysct 42 .maxlen = sizeof(sysctl_unprivileged_userfaultfd),
44 .mode = 0644, 43 .mode = 0644,
45 .proc_handler = proc_dointve 44 .proc_handler = proc_dointvec_minmax,
46 .extra1 = SYSCTL_ZERO, 45 .extra1 = SYSCTL_ZERO,
47 .extra2 = SYSCTL_ONE, 46 .extra2 = SYSCTL_ONE,
48 }, 47 },
>> 48 { }
49 }; 49 };
50 #endif 50 #endif
51 51
52 static struct kmem_cache *userfaultfd_ctx_cach !! 52 static struct kmem_cache *userfaultfd_ctx_cachep __read_mostly;
>> 53
>> 54 /*
>> 55 * Start with fault_pending_wqh and fault_wqh so they're more likely
>> 56 * to be in the same cacheline.
>> 57 *
>> 58 * Locking order:
>> 59 * fd_wqh.lock
>> 60 * fault_pending_wqh.lock
>> 61 * fault_wqh.lock
>> 62 * event_wqh.lock
>> 63 *
>> 64 * To avoid deadlocks, IRQs must be disabled when taking any of the above locks,
>> 65 * since fd_wqh.lock is taken by aio_poll() while it's holding a lock that's
>> 66 * also taken in IRQ context.
>> 67 */
>> 68 struct userfaultfd_ctx {
>> 69 /* waitqueue head for the pending (i.e. not read) userfaults */
>> 70 wait_queue_head_t fault_pending_wqh;
>> 71 /* waitqueue head for the userfaults */
>> 72 wait_queue_head_t fault_wqh;
>> 73 /* waitqueue head for the pseudo fd to wakeup poll/read */
>> 74 wait_queue_head_t fd_wqh;
>> 75 /* waitqueue head for events */
>> 76 wait_queue_head_t event_wqh;
>> 77 /* a refile sequence protected by fault_pending_wqh lock */
>> 78 seqcount_spinlock_t refile_seq;
>> 79 /* pseudo fd refcounting */
>> 80 refcount_t refcount;
>> 81 /* userfaultfd syscall flags */
>> 82 unsigned int flags;
>> 83 /* features requested from the userspace */
>> 84 unsigned int features;
>> 85 /* released */
>> 86 bool released;
>> 87 /* memory mappings are changing because of non-cooperative event */
>> 88 atomic_t mmap_changing;
>> 89 /* mm with one ore more vmas attached to this userfaultfd_ctx */
>> 90 struct mm_struct *mm;
>> 91 };
53 92
54 struct userfaultfd_fork_ctx { 93 struct userfaultfd_fork_ctx {
55 struct userfaultfd_ctx *orig; 94 struct userfaultfd_ctx *orig;
56 struct userfaultfd_ctx *new; 95 struct userfaultfd_ctx *new;
57 struct list_head list; 96 struct list_head list;
58 }; 97 };
59 98
60 struct userfaultfd_unmap_ctx { 99 struct userfaultfd_unmap_ctx {
61 struct userfaultfd_ctx *ctx; 100 struct userfaultfd_ctx *ctx;
62 unsigned long start; 101 unsigned long start;
63 unsigned long end; 102 unsigned long end;
64 struct list_head list; 103 struct list_head list;
65 }; 104 };
66 105
67 struct userfaultfd_wait_queue { 106 struct userfaultfd_wait_queue {
68 struct uffd_msg msg; 107 struct uffd_msg msg;
69 wait_queue_entry_t wq; 108 wait_queue_entry_t wq;
70 struct userfaultfd_ctx *ctx; 109 struct userfaultfd_ctx *ctx;
71 bool waken; 110 bool waken;
72 }; 111 };
73 112
74 struct userfaultfd_wake_range { 113 struct userfaultfd_wake_range {
75 unsigned long start; 114 unsigned long start;
76 unsigned long len; 115 unsigned long len;
77 }; 116 };
78 117
79 /* internal indication that UFFD_API ioctl was 118 /* internal indication that UFFD_API ioctl was successfully executed */
80 #define UFFD_FEATURE_INITIALIZED 119 #define UFFD_FEATURE_INITIALIZED (1u << 31)
81 120
82 static bool userfaultfd_is_initialized(struct 121 static bool userfaultfd_is_initialized(struct userfaultfd_ctx *ctx)
83 { 122 {
84 return ctx->features & UFFD_FEATURE_IN 123 return ctx->features & UFFD_FEATURE_INITIALIZED;
85 } 124 }
86 125
87 static bool userfaultfd_wp_async_ctx(struct us <<
88 { <<
89 return ctx && (ctx->features & UFFD_FE <<
90 } <<
91 <<
92 /* 126 /*
93 * Whether WP_UNPOPULATED is enabled on the uf 127 * Whether WP_UNPOPULATED is enabled on the uffd context. It is only
94 * meaningful when userfaultfd_wp()==true on t 128 * meaningful when userfaultfd_wp()==true on the vma and when it's
95 * anonymous. 129 * anonymous.
96 */ 130 */
97 bool userfaultfd_wp_unpopulated(struct vm_area 131 bool userfaultfd_wp_unpopulated(struct vm_area_struct *vma)
98 { 132 {
99 struct userfaultfd_ctx *ctx = vma->vm_ 133 struct userfaultfd_ctx *ctx = vma->vm_userfaultfd_ctx.ctx;
100 134
101 if (!ctx) 135 if (!ctx)
102 return false; 136 return false;
103 137
104 return ctx->features & UFFD_FEATURE_WP 138 return ctx->features & UFFD_FEATURE_WP_UNPOPULATED;
105 } 139 }
106 140
107 static void userfaultfd_set_vm_flags(struct vm 141 static void userfaultfd_set_vm_flags(struct vm_area_struct *vma,
108 vm_flags_ 142 vm_flags_t flags)
109 { 143 {
110 const bool uffd_wp_changed = (vma->vm_ 144 const bool uffd_wp_changed = (vma->vm_flags ^ flags) & VM_UFFD_WP;
111 145
112 vm_flags_reset(vma, flags); 146 vm_flags_reset(vma, flags);
113 /* 147 /*
114 * For shared mappings, we want to ena 148 * For shared mappings, we want to enable writenotify while
115 * userfaultfd-wp is enabled (see vma_ 149 * userfaultfd-wp is enabled (see vma_wants_writenotify()). We'll simply
116 * recalculate vma->vm_page_prot whene 150 * recalculate vma->vm_page_prot whenever userfaultfd-wp changes.
117 */ 151 */
118 if ((vma->vm_flags & VM_SHARED) && uff 152 if ((vma->vm_flags & VM_SHARED) && uffd_wp_changed)
119 vma_set_page_prot(vma); 153 vma_set_page_prot(vma);
120 } 154 }
121 155
122 static int userfaultfd_wake_function(wait_queu 156 static int userfaultfd_wake_function(wait_queue_entry_t *wq, unsigned mode,
123 int wake_ 157 int wake_flags, void *key)
124 { 158 {
125 struct userfaultfd_wake_range *range = 159 struct userfaultfd_wake_range *range = key;
126 int ret; 160 int ret;
127 struct userfaultfd_wait_queue *uwq; 161 struct userfaultfd_wait_queue *uwq;
128 unsigned long start, len; 162 unsigned long start, len;
129 163
130 uwq = container_of(wq, struct userfaul 164 uwq = container_of(wq, struct userfaultfd_wait_queue, wq);
131 ret = 0; 165 ret = 0;
132 /* len == 0 means wake all */ 166 /* len == 0 means wake all */
133 start = range->start; 167 start = range->start;
134 len = range->len; 168 len = range->len;
135 if (len && (start > uwq->msg.arg.pagef 169 if (len && (start > uwq->msg.arg.pagefault.address ||
136 start + len <= uwq->msg.ar 170 start + len <= uwq->msg.arg.pagefault.address))
137 goto out; 171 goto out;
138 WRITE_ONCE(uwq->waken, true); 172 WRITE_ONCE(uwq->waken, true);
139 /* 173 /*
140 * The Program-Order guarantees provid 174 * The Program-Order guarantees provided by the scheduler
141 * ensure uwq->waken is visible before 175 * ensure uwq->waken is visible before the task is woken.
142 */ 176 */
143 ret = wake_up_state(wq->private, mode) 177 ret = wake_up_state(wq->private, mode);
144 if (ret) { 178 if (ret) {
145 /* 179 /*
146 * Wake only once, autoremove 180 * Wake only once, autoremove behavior.
147 * 181 *
148 * After the effect of list_de 182 * After the effect of list_del_init is visible to the other
149 * CPUs, the waitqueue may dis 183 * CPUs, the waitqueue may disappear from under us, see the
150 * !list_empty_careful() in ha 184 * !list_empty_careful() in handle_userfault().
151 * 185 *
152 * try_to_wake_up() has an imp 186 * try_to_wake_up() has an implicit smp_mb(), and the
153 * wq->private is read before 187 * wq->private is read before calling the extern function
154 * "wake_up_state" (which in t 188 * "wake_up_state" (which in turns calls try_to_wake_up).
155 */ 189 */
156 list_del_init(&wq->entry); 190 list_del_init(&wq->entry);
157 } 191 }
158 out: 192 out:
159 return ret; 193 return ret;
160 } 194 }
161 195
162 /** 196 /**
163 * userfaultfd_ctx_get - Acquires a reference 197 * userfaultfd_ctx_get - Acquires a reference to the internal userfaultfd
164 * context. 198 * context.
165 * @ctx: [in] Pointer to the userfaultfd conte 199 * @ctx: [in] Pointer to the userfaultfd context.
166 */ 200 */
167 static void userfaultfd_ctx_get(struct userfau 201 static void userfaultfd_ctx_get(struct userfaultfd_ctx *ctx)
168 { 202 {
169 refcount_inc(&ctx->refcount); 203 refcount_inc(&ctx->refcount);
170 } 204 }
171 205
172 /** 206 /**
173 * userfaultfd_ctx_put - Releases a reference 207 * userfaultfd_ctx_put - Releases a reference to the internal userfaultfd
174 * context. 208 * context.
175 * @ctx: [in] Pointer to userfaultfd context. 209 * @ctx: [in] Pointer to userfaultfd context.
176 * 210 *
177 * The userfaultfd context reference must have 211 * The userfaultfd context reference must have been previously acquired either
178 * with userfaultfd_ctx_get() or userfaultfd_c 212 * with userfaultfd_ctx_get() or userfaultfd_ctx_fdget().
179 */ 213 */
180 static void userfaultfd_ctx_put(struct userfau 214 static void userfaultfd_ctx_put(struct userfaultfd_ctx *ctx)
181 { 215 {
182 if (refcount_dec_and_test(&ctx->refcou 216 if (refcount_dec_and_test(&ctx->refcount)) {
183 VM_BUG_ON(spin_is_locked(&ctx- 217 VM_BUG_ON(spin_is_locked(&ctx->fault_pending_wqh.lock));
184 VM_BUG_ON(waitqueue_active(&ct 218 VM_BUG_ON(waitqueue_active(&ctx->fault_pending_wqh));
185 VM_BUG_ON(spin_is_locked(&ctx- 219 VM_BUG_ON(spin_is_locked(&ctx->fault_wqh.lock));
186 VM_BUG_ON(waitqueue_active(&ct 220 VM_BUG_ON(waitqueue_active(&ctx->fault_wqh));
187 VM_BUG_ON(spin_is_locked(&ctx- 221 VM_BUG_ON(spin_is_locked(&ctx->event_wqh.lock));
188 VM_BUG_ON(waitqueue_active(&ct 222 VM_BUG_ON(waitqueue_active(&ctx->event_wqh));
189 VM_BUG_ON(spin_is_locked(&ctx- 223 VM_BUG_ON(spin_is_locked(&ctx->fd_wqh.lock));
190 VM_BUG_ON(waitqueue_active(&ct 224 VM_BUG_ON(waitqueue_active(&ctx->fd_wqh));
191 mmdrop(ctx->mm); 225 mmdrop(ctx->mm);
192 kmem_cache_free(userfaultfd_ct 226 kmem_cache_free(userfaultfd_ctx_cachep, ctx);
193 } 227 }
194 } 228 }
195 229
196 static inline void msg_init(struct uffd_msg *m 230 static inline void msg_init(struct uffd_msg *msg)
197 { 231 {
198 BUILD_BUG_ON(sizeof(struct uffd_msg) ! 232 BUILD_BUG_ON(sizeof(struct uffd_msg) != 32);
199 /* 233 /*
200 * Must use memset to zero out the pad 234 * Must use memset to zero out the paddings or kernel data is
201 * leaked to userland. 235 * leaked to userland.
202 */ 236 */
203 memset(msg, 0, sizeof(struct uffd_msg) 237 memset(msg, 0, sizeof(struct uffd_msg));
204 } 238 }
205 239
206 static inline struct uffd_msg userfault_msg(un 240 static inline struct uffd_msg userfault_msg(unsigned long address,
207 un 241 unsigned long real_address,
208 un 242 unsigned int flags,
209 un 243 unsigned long reason,
210 un 244 unsigned int features)
211 { 245 {
212 struct uffd_msg msg; 246 struct uffd_msg msg;
213 247
214 msg_init(&msg); 248 msg_init(&msg);
215 msg.event = UFFD_EVENT_PAGEFAULT; 249 msg.event = UFFD_EVENT_PAGEFAULT;
216 250
217 msg.arg.pagefault.address = (features 251 msg.arg.pagefault.address = (features & UFFD_FEATURE_EXACT_ADDRESS) ?
218 real_addre 252 real_address : address;
219 253
220 /* 254 /*
221 * These flags indicate why the userfa 255 * These flags indicate why the userfault occurred:
222 * - UFFD_PAGEFAULT_FLAG_WP indicates 256 * - UFFD_PAGEFAULT_FLAG_WP indicates a write protect fault.
223 * - UFFD_PAGEFAULT_FLAG_MINOR indicat 257 * - UFFD_PAGEFAULT_FLAG_MINOR indicates a minor fault.
224 * - Neither of these flags being set 258 * - Neither of these flags being set indicates a MISSING fault.
225 * 259 *
226 * Separately, UFFD_PAGEFAULT_FLAG_WRI 260 * Separately, UFFD_PAGEFAULT_FLAG_WRITE indicates it was a write
227 * fault. Otherwise, it was a read fau 261 * fault. Otherwise, it was a read fault.
228 */ 262 */
229 if (flags & FAULT_FLAG_WRITE) 263 if (flags & FAULT_FLAG_WRITE)
230 msg.arg.pagefault.flags |= UFF 264 msg.arg.pagefault.flags |= UFFD_PAGEFAULT_FLAG_WRITE;
231 if (reason & VM_UFFD_WP) 265 if (reason & VM_UFFD_WP)
232 msg.arg.pagefault.flags |= UFF 266 msg.arg.pagefault.flags |= UFFD_PAGEFAULT_FLAG_WP;
233 if (reason & VM_UFFD_MINOR) 267 if (reason & VM_UFFD_MINOR)
234 msg.arg.pagefault.flags |= UFF 268 msg.arg.pagefault.flags |= UFFD_PAGEFAULT_FLAG_MINOR;
235 if (features & UFFD_FEATURE_THREAD_ID) 269 if (features & UFFD_FEATURE_THREAD_ID)
236 msg.arg.pagefault.feat.ptid = 270 msg.arg.pagefault.feat.ptid = task_pid_vnr(current);
237 return msg; 271 return msg;
238 } 272 }
239 273
240 #ifdef CONFIG_HUGETLB_PAGE 274 #ifdef CONFIG_HUGETLB_PAGE
241 /* 275 /*
242 * Same functionality as userfaultfd_must_wait 276 * Same functionality as userfaultfd_must_wait below with modifications for
243 * hugepmd ranges. 277 * hugepmd ranges.
244 */ 278 */
245 static inline bool userfaultfd_huge_must_wait( 279 static inline bool userfaultfd_huge_must_wait(struct userfaultfd_ctx *ctx,
246 !! 280 struct vm_area_struct *vma,
247 !! 281 unsigned long address,
>> 282 unsigned long flags,
>> 283 unsigned long reason)
248 { 284 {
249 struct vm_area_struct *vma = vmf->vma; <<
250 pte_t *ptep, pte; 285 pte_t *ptep, pte;
251 bool ret = true; 286 bool ret = true;
252 287
253 assert_fault_locked(vmf); !! 288 mmap_assert_locked(ctx->mm);
254 289
255 ptep = hugetlb_walk(vma, vmf->address, !! 290 ptep = hugetlb_walk(vma, address, vma_mmu_pagesize(vma));
256 if (!ptep) 291 if (!ptep)
257 goto out; 292 goto out;
258 293
259 ret = false; 294 ret = false;
260 pte = huge_ptep_get(vma->vm_mm, vmf->a !! 295 pte = huge_ptep_get(ptep);
261 296
262 /* 297 /*
263 * Lockless access: we're in a wait_ev 298 * Lockless access: we're in a wait_event so it's ok if it
264 * changes under us. PTE markers shou 299 * changes under us. PTE markers should be handled the same as none
265 * ptes here. 300 * ptes here.
266 */ 301 */
267 if (huge_pte_none_mostly(pte)) 302 if (huge_pte_none_mostly(pte))
268 ret = true; 303 ret = true;
269 if (!huge_pte_write(pte) && (reason & 304 if (!huge_pte_write(pte) && (reason & VM_UFFD_WP))
270 ret = true; 305 ret = true;
271 out: 306 out:
272 return ret; 307 return ret;
273 } 308 }
274 #else 309 #else
275 static inline bool userfaultfd_huge_must_wait( 310 static inline bool userfaultfd_huge_must_wait(struct userfaultfd_ctx *ctx,
276 !! 311 struct vm_area_struct *vma,
277 !! 312 unsigned long address,
>> 313 unsigned long flags,
>> 314 unsigned long reason)
278 { 315 {
279 return false; /* should never get he 316 return false; /* should never get here */
280 } 317 }
281 #endif /* CONFIG_HUGETLB_PAGE */ 318 #endif /* CONFIG_HUGETLB_PAGE */
282 319
283 /* 320 /*
284 * Verify the pagetables are still not ok afte 321 * Verify the pagetables are still not ok after having reigstered into
285 * the fault_pending_wqh to avoid userland hav 322 * the fault_pending_wqh to avoid userland having to UFFDIO_WAKE any
286 * userfault that has already been resolved, i !! 323 * userfault that has already been resolved, if userfaultfd_read and
287 * UFFDIO_COPY|ZEROPAGE are being run simultan 324 * UFFDIO_COPY|ZEROPAGE are being run simultaneously on two different
288 * threads. 325 * threads.
289 */ 326 */
290 static inline bool userfaultfd_must_wait(struc 327 static inline bool userfaultfd_must_wait(struct userfaultfd_ctx *ctx,
291 struc !! 328 unsigned long address,
>> 329 unsigned long flags,
292 unsig 330 unsigned long reason)
293 { 331 {
294 struct mm_struct *mm = ctx->mm; 332 struct mm_struct *mm = ctx->mm;
295 unsigned long address = vmf->address; <<
296 pgd_t *pgd; 333 pgd_t *pgd;
297 p4d_t *p4d; 334 p4d_t *p4d;
298 pud_t *pud; 335 pud_t *pud;
299 pmd_t *pmd, _pmd; 336 pmd_t *pmd, _pmd;
300 pte_t *pte; 337 pte_t *pte;
301 pte_t ptent; 338 pte_t ptent;
302 bool ret = true; 339 bool ret = true;
303 340
304 assert_fault_locked(vmf); !! 341 mmap_assert_locked(mm);
305 342
306 pgd = pgd_offset(mm, address); 343 pgd = pgd_offset(mm, address);
307 if (!pgd_present(*pgd)) 344 if (!pgd_present(*pgd))
308 goto out; 345 goto out;
309 p4d = p4d_offset(pgd, address); 346 p4d = p4d_offset(pgd, address);
310 if (!p4d_present(*p4d)) 347 if (!p4d_present(*p4d))
311 goto out; 348 goto out;
312 pud = pud_offset(p4d, address); 349 pud = pud_offset(p4d, address);
313 if (!pud_present(*pud)) 350 if (!pud_present(*pud))
314 goto out; 351 goto out;
315 pmd = pmd_offset(pud, address); 352 pmd = pmd_offset(pud, address);
316 again: 353 again:
317 _pmd = pmdp_get_lockless(pmd); 354 _pmd = pmdp_get_lockless(pmd);
318 if (pmd_none(_pmd)) 355 if (pmd_none(_pmd))
319 goto out; 356 goto out;
320 357
321 ret = false; 358 ret = false;
322 if (!pmd_present(_pmd) || pmd_devmap(_ 359 if (!pmd_present(_pmd) || pmd_devmap(_pmd))
323 goto out; 360 goto out;
324 361
325 if (pmd_trans_huge(_pmd)) { 362 if (pmd_trans_huge(_pmd)) {
326 if (!pmd_write(_pmd) && (reaso 363 if (!pmd_write(_pmd) && (reason & VM_UFFD_WP))
327 ret = true; 364 ret = true;
328 goto out; 365 goto out;
329 } 366 }
330 367
331 pte = pte_offset_map(pmd, address); 368 pte = pte_offset_map(pmd, address);
332 if (!pte) { 369 if (!pte) {
333 ret = true; 370 ret = true;
334 goto again; 371 goto again;
335 } 372 }
336 /* 373 /*
337 * Lockless access: we're in a wait_ev 374 * Lockless access: we're in a wait_event so it's ok if it
338 * changes under us. PTE markers shou 375 * changes under us. PTE markers should be handled the same as none
339 * ptes here. 376 * ptes here.
340 */ 377 */
341 ptent = ptep_get(pte); 378 ptent = ptep_get(pte);
342 if (pte_none_mostly(ptent)) 379 if (pte_none_mostly(ptent))
343 ret = true; 380 ret = true;
344 if (!pte_write(ptent) && (reason & VM_ 381 if (!pte_write(ptent) && (reason & VM_UFFD_WP))
345 ret = true; 382 ret = true;
346 pte_unmap(pte); 383 pte_unmap(pte);
347 384
348 out: 385 out:
349 return ret; 386 return ret;
350 } 387 }
351 388
352 static inline unsigned int userfaultfd_get_blo 389 static inline unsigned int userfaultfd_get_blocking_state(unsigned int flags)
353 { 390 {
354 if (flags & FAULT_FLAG_INTERRUPTIBLE) 391 if (flags & FAULT_FLAG_INTERRUPTIBLE)
355 return TASK_INTERRUPTIBLE; 392 return TASK_INTERRUPTIBLE;
356 393
357 if (flags & FAULT_FLAG_KILLABLE) 394 if (flags & FAULT_FLAG_KILLABLE)
358 return TASK_KILLABLE; 395 return TASK_KILLABLE;
359 396
360 return TASK_UNINTERRUPTIBLE; 397 return TASK_UNINTERRUPTIBLE;
361 } 398 }
362 399
363 /* 400 /*
364 * The locking rules involved in returning VM_ 401 * The locking rules involved in returning VM_FAULT_RETRY depending on
365 * FAULT_FLAG_ALLOW_RETRY, FAULT_FLAG_RETRY_NO 402 * FAULT_FLAG_ALLOW_RETRY, FAULT_FLAG_RETRY_NOWAIT and
366 * FAULT_FLAG_KILLABLE are not straightforward 403 * FAULT_FLAG_KILLABLE are not straightforward. The "Caution"
367 * recommendation in __lock_page_or_retry is n 404 * recommendation in __lock_page_or_retry is not an understatement.
368 * 405 *
369 * If FAULT_FLAG_ALLOW_RETRY is set, the mmap_ 406 * If FAULT_FLAG_ALLOW_RETRY is set, the mmap_lock must be released
370 * before returning VM_FAULT_RETRY only if FAU 407 * before returning VM_FAULT_RETRY only if FAULT_FLAG_RETRY_NOWAIT is
371 * not set. 408 * not set.
372 * 409 *
373 * If FAULT_FLAG_ALLOW_RETRY is set but FAULT_ 410 * If FAULT_FLAG_ALLOW_RETRY is set but FAULT_FLAG_KILLABLE is not
374 * set, VM_FAULT_RETRY can still be returned i 411 * set, VM_FAULT_RETRY can still be returned if and only if there are
375 * fatal_signal_pending()s, and the mmap_lock 412 * fatal_signal_pending()s, and the mmap_lock must be released before
376 * returning it. 413 * returning it.
377 */ 414 */
378 vm_fault_t handle_userfault(struct vm_fault *v 415 vm_fault_t handle_userfault(struct vm_fault *vmf, unsigned long reason)
379 { 416 {
380 struct vm_area_struct *vma = vmf->vma; 417 struct vm_area_struct *vma = vmf->vma;
381 struct mm_struct *mm = vma->vm_mm; 418 struct mm_struct *mm = vma->vm_mm;
382 struct userfaultfd_ctx *ctx; 419 struct userfaultfd_ctx *ctx;
383 struct userfaultfd_wait_queue uwq; 420 struct userfaultfd_wait_queue uwq;
384 vm_fault_t ret = VM_FAULT_SIGBUS; 421 vm_fault_t ret = VM_FAULT_SIGBUS;
385 bool must_wait; 422 bool must_wait;
386 unsigned int blocking_state; 423 unsigned int blocking_state;
387 424
388 /* 425 /*
389 * We don't do userfault handling for 426 * We don't do userfault handling for the final child pid update.
390 * 427 *
391 * We also don't do userfault handling 428 * We also don't do userfault handling during
392 * coredumping. hugetlbfs has the spec 429 * coredumping. hugetlbfs has the special
393 * hugetlb_follow_page_mask() to skip !! 430 * follow_hugetlb_page() to skip missing pages in the
394 * FOLL_DUMP case, anon memory also ch 431 * FOLL_DUMP case, anon memory also checks for FOLL_DUMP with
395 * the no_page_table() helper in follo 432 * the no_page_table() helper in follow_page_mask(), but the
396 * shmem_vm_ops->fault method is invok 433 * shmem_vm_ops->fault method is invoked even during
397 * coredumping and it ends up here. !! 434 * coredumping without mmap_lock and it ends up here.
398 */ 435 */
399 if (current->flags & (PF_EXITING|PF_DU 436 if (current->flags & (PF_EXITING|PF_DUMPCORE))
400 goto out; 437 goto out;
401 438
402 assert_fault_locked(vmf); !! 439 /*
>> 440 * Coredumping runs without mmap_lock so we can only check that
>> 441 * the mmap_lock is held, if PF_DUMPCORE was not set.
>> 442 */
>> 443 mmap_assert_locked(mm);
403 444
404 ctx = vma->vm_userfaultfd_ctx.ctx; 445 ctx = vma->vm_userfaultfd_ctx.ctx;
405 if (!ctx) 446 if (!ctx)
406 goto out; 447 goto out;
407 448
408 BUG_ON(ctx->mm != mm); 449 BUG_ON(ctx->mm != mm);
409 450
410 /* Any unrecognized flag is a bug. */ 451 /* Any unrecognized flag is a bug. */
411 VM_BUG_ON(reason & ~__VM_UFFD_FLAGS); 452 VM_BUG_ON(reason & ~__VM_UFFD_FLAGS);
412 /* 0 or > 1 flags set is a bug; we exp 453 /* 0 or > 1 flags set is a bug; we expect exactly 1. */
413 VM_BUG_ON(!reason || (reason & (reason 454 VM_BUG_ON(!reason || (reason & (reason - 1)));
414 455
415 if (ctx->features & UFFD_FEATURE_SIGBU 456 if (ctx->features & UFFD_FEATURE_SIGBUS)
416 goto out; 457 goto out;
417 if (!(vmf->flags & FAULT_FLAG_USER) && 458 if (!(vmf->flags & FAULT_FLAG_USER) && (ctx->flags & UFFD_USER_MODE_ONLY))
418 goto out; 459 goto out;
419 460
420 /* 461 /*
421 * If it's already released don't get 462 * If it's already released don't get it. This avoids to loop
422 * in __get_user_pages if userfaultfd_ 463 * in __get_user_pages if userfaultfd_release waits on the
423 * caller of handle_userfault to relea 464 * caller of handle_userfault to release the mmap_lock.
424 */ 465 */
425 if (unlikely(READ_ONCE(ctx->released)) 466 if (unlikely(READ_ONCE(ctx->released))) {
426 /* 467 /*
427 * Don't return VM_FAULT_SIGBU 468 * Don't return VM_FAULT_SIGBUS in this case, so a non
428 * cooperative manager can clo 469 * cooperative manager can close the uffd after the
429 * last UFFDIO_COPY, without r 470 * last UFFDIO_COPY, without risking to trigger an
430 * involuntary SIGBUS if the p 471 * involuntary SIGBUS if the process was starting the
431 * userfaultfd while the userf 472 * userfaultfd while the userfaultfd was still armed
432 * (but after the last UFFDIO_ 473 * (but after the last UFFDIO_COPY). If the uffd
433 * wasn't already closed when 474 * wasn't already closed when the userfault reached
434 * this point, that would norm 475 * this point, that would normally be solved by
435 * userfaultfd_must_wait retur 476 * userfaultfd_must_wait returning 'false'.
436 * 477 *
437 * If we were to return VM_FAU 478 * If we were to return VM_FAULT_SIGBUS here, the non
438 * cooperative manager would b 479 * cooperative manager would be instead forced to
439 * always call UFFDIO_UNREGIST 480 * always call UFFDIO_UNREGISTER before it can safely
440 * close the uffd. 481 * close the uffd.
441 */ 482 */
442 ret = VM_FAULT_NOPAGE; 483 ret = VM_FAULT_NOPAGE;
443 goto out; 484 goto out;
444 } 485 }
445 486
446 /* 487 /*
447 * Check that we can return VM_FAULT_R 488 * Check that we can return VM_FAULT_RETRY.
448 * 489 *
449 * NOTE: it should become possible to 490 * NOTE: it should become possible to return VM_FAULT_RETRY
450 * even if FAULT_FLAG_TRIED is set wit 491 * even if FAULT_FLAG_TRIED is set without leading to gup()
451 * -EBUSY failures, if the userfaultfd 492 * -EBUSY failures, if the userfaultfd is to be extended for
452 * VM_UFFD_WP tracking and we intend t 493 * VM_UFFD_WP tracking and we intend to arm the userfault
453 * without first stopping userland acc 494 * without first stopping userland access to the memory. For
454 * VM_UFFD_MISSING userfaults this is 495 * VM_UFFD_MISSING userfaults this is enough for now.
455 */ 496 */
456 if (unlikely(!(vmf->flags & FAULT_FLAG 497 if (unlikely(!(vmf->flags & FAULT_FLAG_ALLOW_RETRY))) {
457 /* 498 /*
458 * Validate the invariant that 499 * Validate the invariant that nowait must allow retry
459 * to be sure not to return SI 500 * to be sure not to return SIGBUS erroneously on
460 * nowait invocations. 501 * nowait invocations.
461 */ 502 */
462 BUG_ON(vmf->flags & FAULT_FLAG 503 BUG_ON(vmf->flags & FAULT_FLAG_RETRY_NOWAIT);
463 #ifdef CONFIG_DEBUG_VM 504 #ifdef CONFIG_DEBUG_VM
464 if (printk_ratelimit()) { 505 if (printk_ratelimit()) {
465 printk(KERN_WARNING 506 printk(KERN_WARNING
466 "FAULT_FLAG_ALL 507 "FAULT_FLAG_ALLOW_RETRY missing %x\n",
467 vmf->flags); 508 vmf->flags);
468 dump_stack(); 509 dump_stack();
469 } 510 }
470 #endif 511 #endif
471 goto out; 512 goto out;
472 } 513 }
473 514
474 /* 515 /*
475 * Handle nowait, not much to do other 516 * Handle nowait, not much to do other than tell it to retry
476 * and wait. 517 * and wait.
477 */ 518 */
478 ret = VM_FAULT_RETRY; 519 ret = VM_FAULT_RETRY;
479 if (vmf->flags & FAULT_FLAG_RETRY_NOWA 520 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
480 goto out; 521 goto out;
481 522
482 /* take the reference before dropping 523 /* take the reference before dropping the mmap_lock */
483 userfaultfd_ctx_get(ctx); 524 userfaultfd_ctx_get(ctx);
484 525
485 init_waitqueue_func_entry(&uwq.wq, use 526 init_waitqueue_func_entry(&uwq.wq, userfaultfd_wake_function);
486 uwq.wq.private = current; 527 uwq.wq.private = current;
487 uwq.msg = userfault_msg(vmf->address, 528 uwq.msg = userfault_msg(vmf->address, vmf->real_address, vmf->flags,
488 reason, ctx->f 529 reason, ctx->features);
489 uwq.ctx = ctx; 530 uwq.ctx = ctx;
490 uwq.waken = false; 531 uwq.waken = false;
491 532
492 blocking_state = userfaultfd_get_block 533 blocking_state = userfaultfd_get_blocking_state(vmf->flags);
493 534
494 /* 535 /*
495 * Take the vma lock now, in order to 536 * Take the vma lock now, in order to safely call
496 * userfaultfd_huge_must_wait() later. 537 * userfaultfd_huge_must_wait() later. Since acquiring the
497 * (sleepable) vma lock can modify the 538 * (sleepable) vma lock can modify the current task state, that
498 * must be before explicitly calling s 539 * must be before explicitly calling set_current_state().
499 */ 540 */
500 if (is_vm_hugetlb_page(vma)) 541 if (is_vm_hugetlb_page(vma))
501 hugetlb_vma_lock_read(vma); 542 hugetlb_vma_lock_read(vma);
502 543
503 spin_lock_irq(&ctx->fault_pending_wqh. 544 spin_lock_irq(&ctx->fault_pending_wqh.lock);
504 /* 545 /*
505 * After the __add_wait_queue the uwq 546 * After the __add_wait_queue the uwq is visible to userland
506 * through poll/read(). 547 * through poll/read().
507 */ 548 */
508 __add_wait_queue(&ctx->fault_pending_w 549 __add_wait_queue(&ctx->fault_pending_wqh, &uwq.wq);
509 /* 550 /*
510 * The smp_mb() after __set_current_st 551 * The smp_mb() after __set_current_state prevents the reads
511 * following the spin_unlock to happen 552 * following the spin_unlock to happen before the list_add in
512 * __add_wait_queue. 553 * __add_wait_queue.
513 */ 554 */
514 set_current_state(blocking_state); 555 set_current_state(blocking_state);
515 spin_unlock_irq(&ctx->fault_pending_wq 556 spin_unlock_irq(&ctx->fault_pending_wqh.lock);
516 557
517 if (!is_vm_hugetlb_page(vma)) 558 if (!is_vm_hugetlb_page(vma))
518 must_wait = userfaultfd_must_w !! 559 must_wait = userfaultfd_must_wait(ctx, vmf->address, vmf->flags,
>> 560 reason);
519 else 561 else
520 must_wait = userfaultfd_huge_m !! 562 must_wait = userfaultfd_huge_must_wait(ctx, vma,
>> 563 vmf->address,
>> 564 vmf->flags, reason);
521 if (is_vm_hugetlb_page(vma)) 565 if (is_vm_hugetlb_page(vma))
522 hugetlb_vma_unlock_read(vma); 566 hugetlb_vma_unlock_read(vma);
523 release_fault_lock(vmf); !! 567 mmap_read_unlock(mm);
524 568
525 if (likely(must_wait && !READ_ONCE(ctx 569 if (likely(must_wait && !READ_ONCE(ctx->released))) {
526 wake_up_poll(&ctx->fd_wqh, EPO 570 wake_up_poll(&ctx->fd_wqh, EPOLLIN);
527 schedule(); 571 schedule();
528 } 572 }
529 573
530 __set_current_state(TASK_RUNNING); 574 __set_current_state(TASK_RUNNING);
531 575
532 /* 576 /*
533 * Here we race with the list_del; lis 577 * Here we race with the list_del; list_add in
534 * userfaultfd_ctx_read(), however bec 578 * userfaultfd_ctx_read(), however because we don't ever run
535 * list_del_init() to refile across th 579 * list_del_init() to refile across the two lists, the prev
536 * and next pointers will never point 580 * and next pointers will never point to self. list_add also
537 * would never let any of the two poin 581 * would never let any of the two pointers to point to
538 * self. So list_empty_careful won't r 582 * self. So list_empty_careful won't risk to see both pointers
539 * pointing to self at any time during 583 * pointing to self at any time during the list refile. The
540 * only case where list_del_init() is 584 * only case where list_del_init() is called is the full
541 * removal in the wake function and th 585 * removal in the wake function and there we don't re-list_add
542 * and it's fine not to block on the s 586 * and it's fine not to block on the spinlock. The uwq on this
543 * kernel stack can be released after 587 * kernel stack can be released after the list_del_init.
544 */ 588 */
545 if (!list_empty_careful(&uwq.wq.entry) 589 if (!list_empty_careful(&uwq.wq.entry)) {
546 spin_lock_irq(&ctx->fault_pend 590 spin_lock_irq(&ctx->fault_pending_wqh.lock);
547 /* 591 /*
548 * No need of list_del_init(), 592 * No need of list_del_init(), the uwq on the stack
549 * will be freed shortly anywa 593 * will be freed shortly anyway.
550 */ 594 */
551 list_del(&uwq.wq.entry); 595 list_del(&uwq.wq.entry);
552 spin_unlock_irq(&ctx->fault_pe 596 spin_unlock_irq(&ctx->fault_pending_wqh.lock);
553 } 597 }
554 598
555 /* 599 /*
556 * ctx may go away after this if the u 600 * ctx may go away after this if the userfault pseudo fd is
557 * already released. 601 * already released.
558 */ 602 */
559 userfaultfd_ctx_put(ctx); 603 userfaultfd_ctx_put(ctx);
560 604
561 out: 605 out:
562 return ret; 606 return ret;
563 } 607 }
564 608
565 static void userfaultfd_event_wait_completion( 609 static void userfaultfd_event_wait_completion(struct userfaultfd_ctx *ctx,
566 610 struct userfaultfd_wait_queue *ewq)
567 { 611 {
568 struct userfaultfd_ctx *release_new_ct 612 struct userfaultfd_ctx *release_new_ctx;
569 613
570 if (WARN_ON_ONCE(current->flags & PF_E 614 if (WARN_ON_ONCE(current->flags & PF_EXITING))
571 goto out; 615 goto out;
572 616
573 ewq->ctx = ctx; 617 ewq->ctx = ctx;
574 init_waitqueue_entry(&ewq->wq, current 618 init_waitqueue_entry(&ewq->wq, current);
575 release_new_ctx = NULL; 619 release_new_ctx = NULL;
576 620
577 spin_lock_irq(&ctx->event_wqh.lock); 621 spin_lock_irq(&ctx->event_wqh.lock);
578 /* 622 /*
579 * After the __add_wait_queue the uwq 623 * After the __add_wait_queue the uwq is visible to userland
580 * through poll/read(). 624 * through poll/read().
581 */ 625 */
582 __add_wait_queue(&ctx->event_wqh, &ewq 626 __add_wait_queue(&ctx->event_wqh, &ewq->wq);
583 for (;;) { 627 for (;;) {
584 set_current_state(TASK_KILLABL 628 set_current_state(TASK_KILLABLE);
585 if (ewq->msg.event == 0) 629 if (ewq->msg.event == 0)
586 break; 630 break;
587 if (READ_ONCE(ctx->released) | 631 if (READ_ONCE(ctx->released) ||
588 fatal_signal_pending(curre 632 fatal_signal_pending(current)) {
589 /* 633 /*
590 * &ewq->wq may be que 634 * &ewq->wq may be queued in fork_event, but
591 * __remove_wait_queue 635 * __remove_wait_queue ignores the head
592 * parameter. It would 636 * parameter. It would be a problem if it
593 * didn't. 637 * didn't.
594 */ 638 */
595 __remove_wait_queue(&c 639 __remove_wait_queue(&ctx->event_wqh, &ewq->wq);
596 if (ewq->msg.event == 640 if (ewq->msg.event == UFFD_EVENT_FORK) {
597 struct userfau 641 struct userfaultfd_ctx *new;
598 642
599 new = (struct 643 new = (struct userfaultfd_ctx *)
600 (unsig 644 (unsigned long)
601 ewq->m 645 ewq->msg.arg.reserved.reserved1;
602 release_new_ct 646 release_new_ctx = new;
603 } 647 }
604 break; 648 break;
605 } 649 }
606 650
607 spin_unlock_irq(&ctx->event_wq 651 spin_unlock_irq(&ctx->event_wqh.lock);
608 652
609 wake_up_poll(&ctx->fd_wqh, EPO 653 wake_up_poll(&ctx->fd_wqh, EPOLLIN);
610 schedule(); 654 schedule();
611 655
612 spin_lock_irq(&ctx->event_wqh. 656 spin_lock_irq(&ctx->event_wqh.lock);
613 } 657 }
614 __set_current_state(TASK_RUNNING); 658 __set_current_state(TASK_RUNNING);
615 spin_unlock_irq(&ctx->event_wqh.lock); 659 spin_unlock_irq(&ctx->event_wqh.lock);
616 660
617 if (release_new_ctx) { 661 if (release_new_ctx) {
618 struct vm_area_struct *vma; 662 struct vm_area_struct *vma;
619 struct mm_struct *mm = release 663 struct mm_struct *mm = release_new_ctx->mm;
620 VMA_ITERATOR(vmi, mm, 0); 664 VMA_ITERATOR(vmi, mm, 0);
621 665
622 /* the various vma->vm_userfau 666 /* the various vma->vm_userfaultfd_ctx still points to it */
623 mmap_write_lock(mm); 667 mmap_write_lock(mm);
624 for_each_vma(vmi, vma) { 668 for_each_vma(vmi, vma) {
625 if (vma->vm_userfaultf 669 if (vma->vm_userfaultfd_ctx.ctx == release_new_ctx) {
626 vma_start_writ <<
627 vma->vm_userfa 670 vma->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
628 userfaultfd_se 671 userfaultfd_set_vm_flags(vma,
629 672 vma->vm_flags & ~__VM_UFFD_FLAGS);
630 } 673 }
631 } 674 }
632 mmap_write_unlock(mm); 675 mmap_write_unlock(mm);
633 676
634 userfaultfd_ctx_put(release_ne 677 userfaultfd_ctx_put(release_new_ctx);
635 } 678 }
636 679
637 /* 680 /*
638 * ctx may go away after this if the u 681 * ctx may go away after this if the userfault pseudo fd is
639 * already released. 682 * already released.
640 */ 683 */
641 out: 684 out:
642 atomic_dec(&ctx->mmap_changing); 685 atomic_dec(&ctx->mmap_changing);
643 VM_BUG_ON(atomic_read(&ctx->mmap_chang 686 VM_BUG_ON(atomic_read(&ctx->mmap_changing) < 0);
644 userfaultfd_ctx_put(ctx); 687 userfaultfd_ctx_put(ctx);
645 } 688 }
646 689
647 static void userfaultfd_event_complete(struct 690 static void userfaultfd_event_complete(struct userfaultfd_ctx *ctx,
648 struct 691 struct userfaultfd_wait_queue *ewq)
649 { 692 {
650 ewq->msg.event = 0; 693 ewq->msg.event = 0;
651 wake_up_locked(&ctx->event_wqh); 694 wake_up_locked(&ctx->event_wqh);
652 __remove_wait_queue(&ctx->event_wqh, & 695 __remove_wait_queue(&ctx->event_wqh, &ewq->wq);
653 } 696 }
654 697
655 int dup_userfaultfd(struct vm_area_struct *vma 698 int dup_userfaultfd(struct vm_area_struct *vma, struct list_head *fcs)
656 { 699 {
657 struct userfaultfd_ctx *ctx = NULL, *o 700 struct userfaultfd_ctx *ctx = NULL, *octx;
658 struct userfaultfd_fork_ctx *fctx; 701 struct userfaultfd_fork_ctx *fctx;
659 702
660 octx = vma->vm_userfaultfd_ctx.ctx; 703 octx = vma->vm_userfaultfd_ctx.ctx;
661 if (!octx) !! 704 if (!octx || !(octx->features & UFFD_FEATURE_EVENT_FORK)) {
662 return 0; <<
663 <<
664 if (!(octx->features & UFFD_FEATURE_EV <<
665 vma_start_write(vma); <<
666 vma->vm_userfaultfd_ctx = NULL 705 vma->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
667 userfaultfd_set_vm_flags(vma, 706 userfaultfd_set_vm_flags(vma, vma->vm_flags & ~__VM_UFFD_FLAGS);
668 return 0; 707 return 0;
669 } 708 }
670 709
671 list_for_each_entry(fctx, fcs, list) 710 list_for_each_entry(fctx, fcs, list)
672 if (fctx->orig == octx) { 711 if (fctx->orig == octx) {
673 ctx = fctx->new; 712 ctx = fctx->new;
674 break; 713 break;
675 } 714 }
676 715
677 if (!ctx) { 716 if (!ctx) {
678 fctx = kmalloc(sizeof(*fctx), 717 fctx = kmalloc(sizeof(*fctx), GFP_KERNEL);
679 if (!fctx) 718 if (!fctx)
680 return -ENOMEM; 719 return -ENOMEM;
681 720
682 ctx = kmem_cache_alloc(userfau 721 ctx = kmem_cache_alloc(userfaultfd_ctx_cachep, GFP_KERNEL);
683 if (!ctx) { 722 if (!ctx) {
684 kfree(fctx); 723 kfree(fctx);
685 return -ENOMEM; 724 return -ENOMEM;
686 } 725 }
687 726
688 refcount_set(&ctx->refcount, 1 727 refcount_set(&ctx->refcount, 1);
689 ctx->flags = octx->flags; 728 ctx->flags = octx->flags;
690 ctx->features = octx->features 729 ctx->features = octx->features;
691 ctx->released = false; 730 ctx->released = false;
692 init_rwsem(&ctx->map_changing_ <<
693 atomic_set(&ctx->mmap_changing 731 atomic_set(&ctx->mmap_changing, 0);
694 ctx->mm = vma->vm_mm; 732 ctx->mm = vma->vm_mm;
695 mmgrab(ctx->mm); 733 mmgrab(ctx->mm);
696 734
697 userfaultfd_ctx_get(octx); 735 userfaultfd_ctx_get(octx);
698 down_write(&octx->map_changing <<
699 atomic_inc(&octx->mmap_changin 736 atomic_inc(&octx->mmap_changing);
700 up_write(&octx->map_changing_l <<
701 fctx->orig = octx; 737 fctx->orig = octx;
702 fctx->new = ctx; 738 fctx->new = ctx;
703 list_add_tail(&fctx->list, fcs 739 list_add_tail(&fctx->list, fcs);
704 } 740 }
705 741
706 vma->vm_userfaultfd_ctx.ctx = ctx; 742 vma->vm_userfaultfd_ctx.ctx = ctx;
707 return 0; 743 return 0;
708 } 744 }
709 745
710 static void dup_fctx(struct userfaultfd_fork_c 746 static void dup_fctx(struct userfaultfd_fork_ctx *fctx)
711 { 747 {
712 struct userfaultfd_ctx *ctx = fctx->or 748 struct userfaultfd_ctx *ctx = fctx->orig;
713 struct userfaultfd_wait_queue ewq; 749 struct userfaultfd_wait_queue ewq;
714 750
715 msg_init(&ewq.msg); 751 msg_init(&ewq.msg);
716 752
717 ewq.msg.event = UFFD_EVENT_FORK; 753 ewq.msg.event = UFFD_EVENT_FORK;
718 ewq.msg.arg.reserved.reserved1 = (unsi 754 ewq.msg.arg.reserved.reserved1 = (unsigned long)fctx->new;
719 755
720 userfaultfd_event_wait_completion(ctx, 756 userfaultfd_event_wait_completion(ctx, &ewq);
721 } 757 }
722 758
723 void dup_userfaultfd_complete(struct list_head 759 void dup_userfaultfd_complete(struct list_head *fcs)
724 { 760 {
725 struct userfaultfd_fork_ctx *fctx, *n; 761 struct userfaultfd_fork_ctx *fctx, *n;
726 762
727 list_for_each_entry_safe(fctx, n, fcs, 763 list_for_each_entry_safe(fctx, n, fcs, list) {
728 dup_fctx(fctx); 764 dup_fctx(fctx);
729 list_del(&fctx->list); 765 list_del(&fctx->list);
730 kfree(fctx); 766 kfree(fctx);
731 } 767 }
732 } 768 }
733 769
734 void mremap_userfaultfd_prep(struct vm_area_st 770 void mremap_userfaultfd_prep(struct vm_area_struct *vma,
735 struct vm_userfau 771 struct vm_userfaultfd_ctx *vm_ctx)
736 { 772 {
737 struct userfaultfd_ctx *ctx; 773 struct userfaultfd_ctx *ctx;
738 774
739 ctx = vma->vm_userfaultfd_ctx.ctx; 775 ctx = vma->vm_userfaultfd_ctx.ctx;
740 776
741 if (!ctx) 777 if (!ctx)
742 return; 778 return;
743 779
744 if (ctx->features & UFFD_FEATURE_EVENT 780 if (ctx->features & UFFD_FEATURE_EVENT_REMAP) {
745 vm_ctx->ctx = ctx; 781 vm_ctx->ctx = ctx;
746 userfaultfd_ctx_get(ctx); 782 userfaultfd_ctx_get(ctx);
747 down_write(&ctx->map_changing_ <<
748 atomic_inc(&ctx->mmap_changing 783 atomic_inc(&ctx->mmap_changing);
749 up_write(&ctx->map_changing_lo <<
750 } else { 784 } else {
751 /* Drop uffd context if remap 785 /* Drop uffd context if remap feature not enabled */
752 vma_start_write(vma); <<
753 vma->vm_userfaultfd_ctx = NULL 786 vma->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
754 userfaultfd_set_vm_flags(vma, 787 userfaultfd_set_vm_flags(vma, vma->vm_flags & ~__VM_UFFD_FLAGS);
755 } 788 }
756 } 789 }
757 790
758 void mremap_userfaultfd_complete(struct vm_use 791 void mremap_userfaultfd_complete(struct vm_userfaultfd_ctx *vm_ctx,
759 unsigned long 792 unsigned long from, unsigned long to,
760 unsigned long 793 unsigned long len)
761 { 794 {
762 struct userfaultfd_ctx *ctx = vm_ctx-> 795 struct userfaultfd_ctx *ctx = vm_ctx->ctx;
763 struct userfaultfd_wait_queue ewq; 796 struct userfaultfd_wait_queue ewq;
764 797
765 if (!ctx) 798 if (!ctx)
766 return; 799 return;
767 800
768 if (to & ~PAGE_MASK) { 801 if (to & ~PAGE_MASK) {
769 userfaultfd_ctx_put(ctx); 802 userfaultfd_ctx_put(ctx);
770 return; 803 return;
771 } 804 }
772 805
773 msg_init(&ewq.msg); 806 msg_init(&ewq.msg);
774 807
775 ewq.msg.event = UFFD_EVENT_REMAP; 808 ewq.msg.event = UFFD_EVENT_REMAP;
776 ewq.msg.arg.remap.from = from; 809 ewq.msg.arg.remap.from = from;
777 ewq.msg.arg.remap.to = to; 810 ewq.msg.arg.remap.to = to;
778 ewq.msg.arg.remap.len = len; 811 ewq.msg.arg.remap.len = len;
779 812
780 userfaultfd_event_wait_completion(ctx, 813 userfaultfd_event_wait_completion(ctx, &ewq);
781 } 814 }
782 815
783 bool userfaultfd_remove(struct vm_area_struct 816 bool userfaultfd_remove(struct vm_area_struct *vma,
784 unsigned long start, u 817 unsigned long start, unsigned long end)
785 { 818 {
786 struct mm_struct *mm = vma->vm_mm; 819 struct mm_struct *mm = vma->vm_mm;
787 struct userfaultfd_ctx *ctx; 820 struct userfaultfd_ctx *ctx;
788 struct userfaultfd_wait_queue ewq; 821 struct userfaultfd_wait_queue ewq;
789 822
790 ctx = vma->vm_userfaultfd_ctx.ctx; 823 ctx = vma->vm_userfaultfd_ctx.ctx;
791 if (!ctx || !(ctx->features & UFFD_FEA 824 if (!ctx || !(ctx->features & UFFD_FEATURE_EVENT_REMOVE))
792 return true; 825 return true;
793 826
794 userfaultfd_ctx_get(ctx); 827 userfaultfd_ctx_get(ctx);
795 down_write(&ctx->map_changing_lock); <<
796 atomic_inc(&ctx->mmap_changing); 828 atomic_inc(&ctx->mmap_changing);
797 up_write(&ctx->map_changing_lock); <<
798 mmap_read_unlock(mm); 829 mmap_read_unlock(mm);
799 830
800 msg_init(&ewq.msg); 831 msg_init(&ewq.msg);
801 832
802 ewq.msg.event = UFFD_EVENT_REMOVE; 833 ewq.msg.event = UFFD_EVENT_REMOVE;
803 ewq.msg.arg.remove.start = start; 834 ewq.msg.arg.remove.start = start;
804 ewq.msg.arg.remove.end = end; 835 ewq.msg.arg.remove.end = end;
805 836
806 userfaultfd_event_wait_completion(ctx, 837 userfaultfd_event_wait_completion(ctx, &ewq);
807 838
808 return false; 839 return false;
809 } 840 }
810 841
811 static bool has_unmap_ctx(struct userfaultfd_c 842 static bool has_unmap_ctx(struct userfaultfd_ctx *ctx, struct list_head *unmaps,
812 unsigned long start, 843 unsigned long start, unsigned long end)
813 { 844 {
814 struct userfaultfd_unmap_ctx *unmap_ct 845 struct userfaultfd_unmap_ctx *unmap_ctx;
815 846
816 list_for_each_entry(unmap_ctx, unmaps, 847 list_for_each_entry(unmap_ctx, unmaps, list)
817 if (unmap_ctx->ctx == ctx && u 848 if (unmap_ctx->ctx == ctx && unmap_ctx->start == start &&
818 unmap_ctx->end == end) 849 unmap_ctx->end == end)
819 return true; 850 return true;
820 851
821 return false; 852 return false;
822 } 853 }
823 854
824 int userfaultfd_unmap_prep(struct vm_area_stru 855 int userfaultfd_unmap_prep(struct vm_area_struct *vma, unsigned long start,
825 unsigned long end, 856 unsigned long end, struct list_head *unmaps)
826 { 857 {
827 struct userfaultfd_unmap_ctx *unmap_ct 858 struct userfaultfd_unmap_ctx *unmap_ctx;
828 struct userfaultfd_ctx *ctx = vma->vm_ 859 struct userfaultfd_ctx *ctx = vma->vm_userfaultfd_ctx.ctx;
829 860
830 if (!ctx || !(ctx->features & UFFD_FEA 861 if (!ctx || !(ctx->features & UFFD_FEATURE_EVENT_UNMAP) ||
831 has_unmap_ctx(ctx, unmaps, start, 862 has_unmap_ctx(ctx, unmaps, start, end))
832 return 0; 863 return 0;
833 864
834 unmap_ctx = kzalloc(sizeof(*unmap_ctx) 865 unmap_ctx = kzalloc(sizeof(*unmap_ctx), GFP_KERNEL);
835 if (!unmap_ctx) 866 if (!unmap_ctx)
836 return -ENOMEM; 867 return -ENOMEM;
837 868
838 userfaultfd_ctx_get(ctx); 869 userfaultfd_ctx_get(ctx);
839 down_write(&ctx->map_changing_lock); <<
840 atomic_inc(&ctx->mmap_changing); 870 atomic_inc(&ctx->mmap_changing);
841 up_write(&ctx->map_changing_lock); <<
842 unmap_ctx->ctx = ctx; 871 unmap_ctx->ctx = ctx;
843 unmap_ctx->start = start; 872 unmap_ctx->start = start;
844 unmap_ctx->end = end; 873 unmap_ctx->end = end;
845 list_add_tail(&unmap_ctx->list, unmaps 874 list_add_tail(&unmap_ctx->list, unmaps);
846 875
847 return 0; 876 return 0;
848 } 877 }
849 878
850 void userfaultfd_unmap_complete(struct mm_stru 879 void userfaultfd_unmap_complete(struct mm_struct *mm, struct list_head *uf)
851 { 880 {
852 struct userfaultfd_unmap_ctx *ctx, *n; 881 struct userfaultfd_unmap_ctx *ctx, *n;
853 struct userfaultfd_wait_queue ewq; 882 struct userfaultfd_wait_queue ewq;
854 883
855 list_for_each_entry_safe(ctx, n, uf, l 884 list_for_each_entry_safe(ctx, n, uf, list) {
856 msg_init(&ewq.msg); 885 msg_init(&ewq.msg);
857 886
858 ewq.msg.event = UFFD_EVENT_UNM 887 ewq.msg.event = UFFD_EVENT_UNMAP;
859 ewq.msg.arg.remove.start = ctx 888 ewq.msg.arg.remove.start = ctx->start;
860 ewq.msg.arg.remove.end = ctx-> 889 ewq.msg.arg.remove.end = ctx->end;
861 890
862 userfaultfd_event_wait_complet 891 userfaultfd_event_wait_completion(ctx->ctx, &ewq);
863 892
864 list_del(&ctx->list); 893 list_del(&ctx->list);
865 kfree(ctx); 894 kfree(ctx);
866 } 895 }
867 } 896 }
868 897
869 static int userfaultfd_release(struct inode *i 898 static int userfaultfd_release(struct inode *inode, struct file *file)
870 { 899 {
871 struct userfaultfd_ctx *ctx = file->pr 900 struct userfaultfd_ctx *ctx = file->private_data;
872 struct mm_struct *mm = ctx->mm; 901 struct mm_struct *mm = ctx->mm;
873 struct vm_area_struct *vma, *prev; 902 struct vm_area_struct *vma, *prev;
874 /* len == 0 means wake all */ 903 /* len == 0 means wake all */
875 struct userfaultfd_wake_range range = 904 struct userfaultfd_wake_range range = { .len = 0, };
876 unsigned long new_flags; 905 unsigned long new_flags;
877 VMA_ITERATOR(vmi, mm, 0); 906 VMA_ITERATOR(vmi, mm, 0);
878 907
879 WRITE_ONCE(ctx->released, true); 908 WRITE_ONCE(ctx->released, true);
880 909
881 if (!mmget_not_zero(mm)) 910 if (!mmget_not_zero(mm))
882 goto wakeup; 911 goto wakeup;
883 912
884 /* 913 /*
885 * Flush page faults out of all CPUs. 914 * Flush page faults out of all CPUs. NOTE: all page faults
886 * must be retried without returning V 915 * must be retried without returning VM_FAULT_SIGBUS if
887 * userfaultfd_ctx_get() succeeds but 916 * userfaultfd_ctx_get() succeeds but vma->vma_userfault_ctx
888 * changes while handle_userfault rele 917 * changes while handle_userfault released the mmap_lock. So
889 * it's critical that released is set 918 * it's critical that released is set to true (above), before
890 * taking the mmap_lock for writing. 919 * taking the mmap_lock for writing.
891 */ 920 */
892 mmap_write_lock(mm); 921 mmap_write_lock(mm);
893 prev = NULL; 922 prev = NULL;
894 for_each_vma(vmi, vma) { 923 for_each_vma(vmi, vma) {
895 cond_resched(); 924 cond_resched();
896 BUG_ON(!!vma->vm_userfaultfd_c 925 BUG_ON(!!vma->vm_userfaultfd_ctx.ctx ^
897 !!(vma->vm_flags & __VM 926 !!(vma->vm_flags & __VM_UFFD_FLAGS));
898 if (vma->vm_userfaultfd_ctx.ct 927 if (vma->vm_userfaultfd_ctx.ctx != ctx) {
899 prev = vma; 928 prev = vma;
900 continue; 929 continue;
901 } 930 }
902 /* Reset ptes for the whole vm <<
903 if (userfaultfd_wp(vma)) <<
904 uffd_wp_range(vma, vma <<
905 vma->vm_ <<
906 new_flags = vma->vm_flags & ~_ 931 new_flags = vma->vm_flags & ~__VM_UFFD_FLAGS;
907 vma = vma_modify_flags_uffd(&v !! 932 prev = vma_merge(&vmi, mm, prev, vma->vm_start, vma->vm_end,
908 vm !! 933 new_flags, vma->anon_vma,
909 NU !! 934 vma->vm_file, vma->vm_pgoff,
>> 935 vma_policy(vma),
>> 936 NULL_VM_UFFD_CTX, anon_vma_name(vma));
>> 937 if (prev) {
>> 938 vma = prev;
>> 939 } else {
>> 940 prev = vma;
>> 941 }
910 942
911 vma_start_write(vma); <<
912 userfaultfd_set_vm_flags(vma, 943 userfaultfd_set_vm_flags(vma, new_flags);
913 vma->vm_userfaultfd_ctx = NULL 944 vma->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
914 <<
915 prev = vma; <<
916 } 945 }
917 mmap_write_unlock(mm); 946 mmap_write_unlock(mm);
918 mmput(mm); 947 mmput(mm);
919 wakeup: 948 wakeup:
920 /* 949 /*
921 * After no new page faults can wait o 950 * After no new page faults can wait on this fault_*wqh, flush
922 * the last page faults that may have 951 * the last page faults that may have been already waiting on
923 * the fault_*wqh. 952 * the fault_*wqh.
924 */ 953 */
925 spin_lock_irq(&ctx->fault_pending_wqh. 954 spin_lock_irq(&ctx->fault_pending_wqh.lock);
926 __wake_up_locked_key(&ctx->fault_pendi 955 __wake_up_locked_key(&ctx->fault_pending_wqh, TASK_NORMAL, &range);
927 __wake_up(&ctx->fault_wqh, TASK_NORMAL 956 __wake_up(&ctx->fault_wqh, TASK_NORMAL, 1, &range);
928 spin_unlock_irq(&ctx->fault_pending_wq 957 spin_unlock_irq(&ctx->fault_pending_wqh.lock);
929 958
930 /* Flush pending events that may still 959 /* Flush pending events that may still wait on event_wqh */
931 wake_up_all(&ctx->event_wqh); 960 wake_up_all(&ctx->event_wqh);
932 961
933 wake_up_poll(&ctx->fd_wqh, EPOLLHUP); 962 wake_up_poll(&ctx->fd_wqh, EPOLLHUP);
934 userfaultfd_ctx_put(ctx); 963 userfaultfd_ctx_put(ctx);
935 return 0; 964 return 0;
936 } 965 }
937 966
938 /* fault_pending_wqh.lock must be hold by the 967 /* fault_pending_wqh.lock must be hold by the caller */
939 static inline struct userfaultfd_wait_queue *f 968 static inline struct userfaultfd_wait_queue *find_userfault_in(
940 wait_queue_head_t *wqh) 969 wait_queue_head_t *wqh)
941 { 970 {
942 wait_queue_entry_t *wq; 971 wait_queue_entry_t *wq;
943 struct userfaultfd_wait_queue *uwq; 972 struct userfaultfd_wait_queue *uwq;
944 973
945 lockdep_assert_held(&wqh->lock); 974 lockdep_assert_held(&wqh->lock);
946 975
947 uwq = NULL; 976 uwq = NULL;
948 if (!waitqueue_active(wqh)) 977 if (!waitqueue_active(wqh))
949 goto out; 978 goto out;
950 /* walk in reverse to provide FIFO beh 979 /* walk in reverse to provide FIFO behavior to read userfaults */
951 wq = list_last_entry(&wqh->head, typeo 980 wq = list_last_entry(&wqh->head, typeof(*wq), entry);
952 uwq = container_of(wq, struct userfaul 981 uwq = container_of(wq, struct userfaultfd_wait_queue, wq);
953 out: 982 out:
954 return uwq; 983 return uwq;
955 } 984 }
956 985
957 static inline struct userfaultfd_wait_queue *f 986 static inline struct userfaultfd_wait_queue *find_userfault(
958 struct userfaultfd_ctx *ctx) 987 struct userfaultfd_ctx *ctx)
959 { 988 {
960 return find_userfault_in(&ctx->fault_p 989 return find_userfault_in(&ctx->fault_pending_wqh);
961 } 990 }
962 991
963 static inline struct userfaultfd_wait_queue *f 992 static inline struct userfaultfd_wait_queue *find_userfault_evt(
964 struct userfaultfd_ctx *ctx) 993 struct userfaultfd_ctx *ctx)
965 { 994 {
966 return find_userfault_in(&ctx->event_w 995 return find_userfault_in(&ctx->event_wqh);
967 } 996 }
968 997
969 static __poll_t userfaultfd_poll(struct file * 998 static __poll_t userfaultfd_poll(struct file *file, poll_table *wait)
970 { 999 {
971 struct userfaultfd_ctx *ctx = file->pr 1000 struct userfaultfd_ctx *ctx = file->private_data;
972 __poll_t ret; 1001 __poll_t ret;
973 1002
974 poll_wait(file, &ctx->fd_wqh, wait); 1003 poll_wait(file, &ctx->fd_wqh, wait);
975 1004
976 if (!userfaultfd_is_initialized(ctx)) 1005 if (!userfaultfd_is_initialized(ctx))
977 return EPOLLERR; 1006 return EPOLLERR;
978 1007
979 /* 1008 /*
980 * poll() never guarantees that read w 1009 * poll() never guarantees that read won't block.
981 * userfaults can be waken before they 1010 * userfaults can be waken before they're read().
982 */ 1011 */
983 if (unlikely(!(file->f_flags & O_NONBL 1012 if (unlikely(!(file->f_flags & O_NONBLOCK)))
984 return EPOLLERR; 1013 return EPOLLERR;
985 /* 1014 /*
986 * lockless access to see if there are 1015 * lockless access to see if there are pending faults
987 * __pollwait last action is the add_w 1016 * __pollwait last action is the add_wait_queue but
988 * the spin_unlock would allow the wai 1017 * the spin_unlock would allow the waitqueue_active to
989 * pass above the actual list_add insi 1018 * pass above the actual list_add inside
990 * add_wait_queue critical section. So 1019 * add_wait_queue critical section. So use a full
991 * memory barrier to serialize the lis 1020 * memory barrier to serialize the list_add write of
992 * add_wait_queue() with the waitqueue 1021 * add_wait_queue() with the waitqueue_active read
993 * below. 1022 * below.
994 */ 1023 */
995 ret = 0; 1024 ret = 0;
996 smp_mb(); 1025 smp_mb();
997 if (waitqueue_active(&ctx->fault_pendi 1026 if (waitqueue_active(&ctx->fault_pending_wqh))
998 ret = EPOLLIN; 1027 ret = EPOLLIN;
999 else if (waitqueue_active(&ctx->event_ 1028 else if (waitqueue_active(&ctx->event_wqh))
1000 ret = EPOLLIN; 1029 ret = EPOLLIN;
1001 1030
1002 return ret; 1031 return ret;
1003 } 1032 }
1004 1033
1005 static const struct file_operations userfault 1034 static const struct file_operations userfaultfd_fops;
1006 1035
1007 static int resolve_userfault_fork(struct user 1036 static int resolve_userfault_fork(struct userfaultfd_ctx *new,
1008 struct inod 1037 struct inode *inode,
1009 struct uffd 1038 struct uffd_msg *msg)
1010 { 1039 {
1011 int fd; 1040 int fd;
1012 1041
1013 fd = anon_inode_create_getfd("[userfa !! 1042 fd = anon_inode_getfd_secure("[userfaultfd]", &userfaultfd_fops, new,
1014 O_RDONLY | (new->flag 1043 O_RDONLY | (new->flags & UFFD_SHARED_FCNTL_FLAGS), inode);
1015 if (fd < 0) 1044 if (fd < 0)
1016 return fd; 1045 return fd;
1017 1046
1018 msg->arg.reserved.reserved1 = 0; 1047 msg->arg.reserved.reserved1 = 0;
1019 msg->arg.fork.ufd = fd; 1048 msg->arg.fork.ufd = fd;
1020 return 0; 1049 return 0;
1021 } 1050 }
1022 1051
1023 static ssize_t userfaultfd_ctx_read(struct us 1052 static ssize_t userfaultfd_ctx_read(struct userfaultfd_ctx *ctx, int no_wait,
1024 struct uf 1053 struct uffd_msg *msg, struct inode *inode)
1025 { 1054 {
1026 ssize_t ret; 1055 ssize_t ret;
1027 DECLARE_WAITQUEUE(wait, current); 1056 DECLARE_WAITQUEUE(wait, current);
1028 struct userfaultfd_wait_queue *uwq; 1057 struct userfaultfd_wait_queue *uwq;
1029 /* 1058 /*
1030 * Handling fork event requires sleep 1059 * Handling fork event requires sleeping operations, so
1031 * we drop the event_wqh lock, then d 1060 * we drop the event_wqh lock, then do these ops, then
1032 * lock it back and wake up the waite 1061 * lock it back and wake up the waiter. While the lock is
1033 * dropped the ewq may go away so we 1062 * dropped the ewq may go away so we keep track of it
1034 * carefully. 1063 * carefully.
1035 */ 1064 */
1036 LIST_HEAD(fork_event); 1065 LIST_HEAD(fork_event);
1037 struct userfaultfd_ctx *fork_nctx = N 1066 struct userfaultfd_ctx *fork_nctx = NULL;
1038 1067
1039 /* always take the fd_wqh lock before 1068 /* always take the fd_wqh lock before the fault_pending_wqh lock */
1040 spin_lock_irq(&ctx->fd_wqh.lock); 1069 spin_lock_irq(&ctx->fd_wqh.lock);
1041 __add_wait_queue(&ctx->fd_wqh, &wait) 1070 __add_wait_queue(&ctx->fd_wqh, &wait);
1042 for (;;) { 1071 for (;;) {
1043 set_current_state(TASK_INTERR 1072 set_current_state(TASK_INTERRUPTIBLE);
1044 spin_lock(&ctx->fault_pending 1073 spin_lock(&ctx->fault_pending_wqh.lock);
1045 uwq = find_userfault(ctx); 1074 uwq = find_userfault(ctx);
1046 if (uwq) { 1075 if (uwq) {
1047 /* 1076 /*
1048 * Use a seqcount to 1077 * Use a seqcount to repeat the lockless check
1049 * in wake_userfault( 1078 * in wake_userfault() to avoid missing
1050 * wakeups because du 1079 * wakeups because during the refile both
1051 * waitqueue could be 1080 * waitqueue could become empty if this is the
1052 * only userfault. 1081 * only userfault.
1053 */ 1082 */
1054 write_seqcount_begin( 1083 write_seqcount_begin(&ctx->refile_seq);
1055 1084
1056 /* 1085 /*
1057 * The fault_pending_ 1086 * The fault_pending_wqh.lock prevents the uwq
1058 * to disappear from 1087 * to disappear from under us.
1059 * 1088 *
1060 * Refile this userfa 1089 * Refile this userfault from
1061 * fault_pending_wqh 1090 * fault_pending_wqh to fault_wqh, it's not
1062 * pending anymore af 1091 * pending anymore after we read it.
1063 * 1092 *
1064 * Use list_del() by 1093 * Use list_del() by hand (as
1065 * userfaultfd_wake_f 1094 * userfaultfd_wake_function also uses
1066 * list_del_init() by 1095 * list_del_init() by hand) to be sure nobody
1067 * changes __remove_w 1096 * changes __remove_wait_queue() to use
1068 * list_del_init() in 1097 * list_del_init() in turn breaking the
1069 * !list_empty_carefu 1098 * !list_empty_careful() check in
1070 * handle_userfault() 1099 * handle_userfault(). The uwq->wq.head list
1071 * must never be empt 1100 * must never be empty at any time during the
1072 * refile, or the wai 1101 * refile, or the waitqueue could disappear
1073 * from under us. The 1102 * from under us. The "wait_queue_head_t"
1074 * parameter of __rem 1103 * parameter of __remove_wait_queue() is unused
1075 * anyway. 1104 * anyway.
1076 */ 1105 */
1077 list_del(&uwq->wq.ent 1106 list_del(&uwq->wq.entry);
1078 add_wait_queue(&ctx-> 1107 add_wait_queue(&ctx->fault_wqh, &uwq->wq);
1079 1108
1080 write_seqcount_end(&c 1109 write_seqcount_end(&ctx->refile_seq);
1081 1110
1082 /* careful to always 1111 /* careful to always initialize msg if ret == 0 */
1083 *msg = uwq->msg; 1112 *msg = uwq->msg;
1084 spin_unlock(&ctx->fau 1113 spin_unlock(&ctx->fault_pending_wqh.lock);
1085 ret = 0; 1114 ret = 0;
1086 break; 1115 break;
1087 } 1116 }
1088 spin_unlock(&ctx->fault_pendi 1117 spin_unlock(&ctx->fault_pending_wqh.lock);
1089 1118
1090 spin_lock(&ctx->event_wqh.loc 1119 spin_lock(&ctx->event_wqh.lock);
1091 uwq = find_userfault_evt(ctx) 1120 uwq = find_userfault_evt(ctx);
1092 if (uwq) { 1121 if (uwq) {
1093 *msg = uwq->msg; 1122 *msg = uwq->msg;
1094 1123
1095 if (uwq->msg.event == 1124 if (uwq->msg.event == UFFD_EVENT_FORK) {
1096 fork_nctx = ( 1125 fork_nctx = (struct userfaultfd_ctx *)
1097 (unsi 1126 (unsigned long)
1098 uwq-> 1127 uwq->msg.arg.reserved.reserved1;
1099 list_move(&uw 1128 list_move(&uwq->wq.entry, &fork_event);
1100 /* 1129 /*
1101 * fork_nctx 1130 * fork_nctx can be freed as soon as
1102 * we drop th 1131 * we drop the lock, unless we take a
1103 * reference 1132 * reference on it.
1104 */ 1133 */
1105 userfaultfd_c 1134 userfaultfd_ctx_get(fork_nctx);
1106 spin_unlock(& 1135 spin_unlock(&ctx->event_wqh.lock);
1107 ret = 0; 1136 ret = 0;
1108 break; 1137 break;
1109 } 1138 }
1110 1139
1111 userfaultfd_event_com 1140 userfaultfd_event_complete(ctx, uwq);
1112 spin_unlock(&ctx->eve 1141 spin_unlock(&ctx->event_wqh.lock);
1113 ret = 0; 1142 ret = 0;
1114 break; 1143 break;
1115 } 1144 }
1116 spin_unlock(&ctx->event_wqh.l 1145 spin_unlock(&ctx->event_wqh.lock);
1117 1146
1118 if (signal_pending(current)) 1147 if (signal_pending(current)) {
1119 ret = -ERESTARTSYS; 1148 ret = -ERESTARTSYS;
1120 break; 1149 break;
1121 } 1150 }
1122 if (no_wait) { 1151 if (no_wait) {
1123 ret = -EAGAIN; 1152 ret = -EAGAIN;
1124 break; 1153 break;
1125 } 1154 }
1126 spin_unlock_irq(&ctx->fd_wqh. 1155 spin_unlock_irq(&ctx->fd_wqh.lock);
1127 schedule(); 1156 schedule();
1128 spin_lock_irq(&ctx->fd_wqh.lo 1157 spin_lock_irq(&ctx->fd_wqh.lock);
1129 } 1158 }
1130 __remove_wait_queue(&ctx->fd_wqh, &wa 1159 __remove_wait_queue(&ctx->fd_wqh, &wait);
1131 __set_current_state(TASK_RUNNING); 1160 __set_current_state(TASK_RUNNING);
1132 spin_unlock_irq(&ctx->fd_wqh.lock); 1161 spin_unlock_irq(&ctx->fd_wqh.lock);
1133 1162
1134 if (!ret && msg->event == UFFD_EVENT_ 1163 if (!ret && msg->event == UFFD_EVENT_FORK) {
1135 ret = resolve_userfault_fork( 1164 ret = resolve_userfault_fork(fork_nctx, inode, msg);
1136 spin_lock_irq(&ctx->event_wqh 1165 spin_lock_irq(&ctx->event_wqh.lock);
1137 if (!list_empty(&fork_event)) 1166 if (!list_empty(&fork_event)) {
1138 /* 1167 /*
1139 * The fork thread di 1168 * The fork thread didn't abort, so we can
1140 * drop the temporary 1169 * drop the temporary refcount.
1141 */ 1170 */
1142 userfaultfd_ctx_put(f 1171 userfaultfd_ctx_put(fork_nctx);
1143 1172
1144 uwq = list_first_entr 1173 uwq = list_first_entry(&fork_event,
1145 1174 typeof(*uwq),
1146 1175 wq.entry);
1147 /* 1176 /*
1148 * If fork_event list 1177 * If fork_event list wasn't empty and in turn
1149 * the event wasn't a 1178 * the event wasn't already released by fork
1150 * (the event is allo 1179 * (the event is allocated on fork kernel
1151 * stack), put the ev 1180 * stack), put the event back to its place in
1152 * the event_wq. fork 1181 * the event_wq. fork_event head will be freed
1153 * as soon as we retu 1182 * as soon as we return so the event cannot
1154 * stay queued there 1183 * stay queued there no matter the current
1155 * "ret" value. 1184 * "ret" value.
1156 */ 1185 */
1157 list_del(&uwq->wq.ent 1186 list_del(&uwq->wq.entry);
1158 __add_wait_queue(&ctx 1187 __add_wait_queue(&ctx->event_wqh, &uwq->wq);
1159 1188
1160 /* 1189 /*
1161 * Leave the event in 1190 * Leave the event in the waitqueue and report
1162 * error to userland 1191 * error to userland if we failed to resolve
1163 * the userfault fork 1192 * the userfault fork.
1164 */ 1193 */
1165 if (likely(!ret)) 1194 if (likely(!ret))
1166 userfaultfd_e 1195 userfaultfd_event_complete(ctx, uwq);
1167 } else { 1196 } else {
1168 /* 1197 /*
1169 * Here the fork thre 1198 * Here the fork thread aborted and the
1170 * refcount from the 1199 * refcount from the fork thread on fork_nctx
1171 * has already been r 1200 * has already been released. We still hold
1172 * the reference we t 1201 * the reference we took before releasing the
1173 * lock above. If res 1202 * lock above. If resolve_userfault_fork
1174 * failed we've to dr 1203 * failed we've to drop it because the
1175 * fork_nctx has to b 1204 * fork_nctx has to be freed in such case. If
1176 * it succeeded we'll 1205 * it succeeded we'll hold it because the new
1177 * uffd references it 1206 * uffd references it.
1178 */ 1207 */
1179 if (ret) 1208 if (ret)
1180 userfaultfd_c 1209 userfaultfd_ctx_put(fork_nctx);
1181 } 1210 }
1182 spin_unlock_irq(&ctx->event_w 1211 spin_unlock_irq(&ctx->event_wqh.lock);
1183 } 1212 }
1184 1213
1185 return ret; 1214 return ret;
1186 } 1215 }
1187 1216
1188 static ssize_t userfaultfd_read_iter(struct k !! 1217 static ssize_t userfaultfd_read(struct file *file, char __user *buf,
>> 1218 size_t count, loff_t *ppos)
1189 { 1219 {
1190 struct file *file = iocb->ki_filp; <<
1191 struct userfaultfd_ctx *ctx = file->p 1220 struct userfaultfd_ctx *ctx = file->private_data;
1192 ssize_t _ret, ret = 0; 1221 ssize_t _ret, ret = 0;
1193 struct uffd_msg msg; 1222 struct uffd_msg msg;
>> 1223 int no_wait = file->f_flags & O_NONBLOCK;
1194 struct inode *inode = file_inode(file 1224 struct inode *inode = file_inode(file);
1195 bool no_wait; <<
1196 1225
1197 if (!userfaultfd_is_initialized(ctx)) 1226 if (!userfaultfd_is_initialized(ctx))
1198 return -EINVAL; 1227 return -EINVAL;
1199 1228
1200 no_wait = file->f_flags & O_NONBLOCK <<
1201 for (;;) { 1229 for (;;) {
1202 if (iov_iter_count(to) < size !! 1230 if (count < sizeof(msg))
1203 return ret ? ret : -E 1231 return ret ? ret : -EINVAL;
1204 _ret = userfaultfd_ctx_read(c 1232 _ret = userfaultfd_ctx_read(ctx, no_wait, &msg, inode);
1205 if (_ret < 0) 1233 if (_ret < 0)
1206 return ret ? ret : _r 1234 return ret ? ret : _ret;
1207 _ret = !copy_to_iter_full(&ms !! 1235 if (copy_to_user((__u64 __user *) buf, &msg, sizeof(msg)))
1208 if (_ret) <<
1209 return ret ? ret : -E 1236 return ret ? ret : -EFAULT;
1210 ret += sizeof(msg); 1237 ret += sizeof(msg);
>> 1238 buf += sizeof(msg);
>> 1239 count -= sizeof(msg);
1211 /* 1240 /*
1212 * Allow to read more than on 1241 * Allow to read more than one fault at time but only
1213 * block if waiting for the v 1242 * block if waiting for the very first one.
1214 */ 1243 */
1215 no_wait = true; !! 1244 no_wait = O_NONBLOCK;
1216 } 1245 }
1217 } 1246 }
1218 1247
1219 static void __wake_userfault(struct userfault 1248 static void __wake_userfault(struct userfaultfd_ctx *ctx,
1220 struct userfault 1249 struct userfaultfd_wake_range *range)
1221 { 1250 {
1222 spin_lock_irq(&ctx->fault_pending_wqh 1251 spin_lock_irq(&ctx->fault_pending_wqh.lock);
1223 /* wake all in the range and autoremo 1252 /* wake all in the range and autoremove */
1224 if (waitqueue_active(&ctx->fault_pend 1253 if (waitqueue_active(&ctx->fault_pending_wqh))
1225 __wake_up_locked_key(&ctx->fa 1254 __wake_up_locked_key(&ctx->fault_pending_wqh, TASK_NORMAL,
1226 range); 1255 range);
1227 if (waitqueue_active(&ctx->fault_wqh) 1256 if (waitqueue_active(&ctx->fault_wqh))
1228 __wake_up(&ctx->fault_wqh, TA 1257 __wake_up(&ctx->fault_wqh, TASK_NORMAL, 1, range);
1229 spin_unlock_irq(&ctx->fault_pending_w 1258 spin_unlock_irq(&ctx->fault_pending_wqh.lock);
1230 } 1259 }
1231 1260
1232 static __always_inline void wake_userfault(st 1261 static __always_inline void wake_userfault(struct userfaultfd_ctx *ctx,
1233 st 1262 struct userfaultfd_wake_range *range)
1234 { 1263 {
1235 unsigned seq; 1264 unsigned seq;
1236 bool need_wakeup; 1265 bool need_wakeup;
1237 1266
1238 /* 1267 /*
1239 * To be sure waitqueue_active() is n 1268 * To be sure waitqueue_active() is not reordered by the CPU
1240 * before the pagetable update, use a 1269 * before the pagetable update, use an explicit SMP memory
1241 * barrier here. PT lock release or m 1270 * barrier here. PT lock release or mmap_read_unlock(mm) still
1242 * have release semantics that can al 1271 * have release semantics that can allow the
1243 * waitqueue_active() to be reordered 1272 * waitqueue_active() to be reordered before the pte update.
1244 */ 1273 */
1245 smp_mb(); 1274 smp_mb();
1246 1275
1247 /* 1276 /*
1248 * Use waitqueue_active because it's 1277 * Use waitqueue_active because it's very frequent to
1249 * change the address space atomicall 1278 * change the address space atomically even if there are no
1250 * userfaults yet. So we take the spi 1279 * userfaults yet. So we take the spinlock only when we're
1251 * sure we've userfaults to wake. 1280 * sure we've userfaults to wake.
1252 */ 1281 */
1253 do { 1282 do {
1254 seq = read_seqcount_begin(&ct 1283 seq = read_seqcount_begin(&ctx->refile_seq);
1255 need_wakeup = waitqueue_activ 1284 need_wakeup = waitqueue_active(&ctx->fault_pending_wqh) ||
1256 waitqueue_active(&ctx 1285 waitqueue_active(&ctx->fault_wqh);
1257 cond_resched(); 1286 cond_resched();
1258 } while (read_seqcount_retry(&ctx->re 1287 } while (read_seqcount_retry(&ctx->refile_seq, seq));
1259 if (need_wakeup) 1288 if (need_wakeup)
1260 __wake_userfault(ctx, range); 1289 __wake_userfault(ctx, range);
1261 } 1290 }
1262 1291
1263 static __always_inline int validate_unaligned !! 1292 static __always_inline int validate_range(struct mm_struct *mm,
1264 struct mm_struct *mm, __u64 start, __ !! 1293 __u64 start, __u64 len)
1265 { 1294 {
1266 __u64 task_size = mm->task_size; 1295 __u64 task_size = mm->task_size;
1267 1296
>> 1297 if (start & ~PAGE_MASK)
>> 1298 return -EINVAL;
1268 if (len & ~PAGE_MASK) 1299 if (len & ~PAGE_MASK)
1269 return -EINVAL; 1300 return -EINVAL;
1270 if (!len) 1301 if (!len)
1271 return -EINVAL; 1302 return -EINVAL;
1272 if (start < mmap_min_addr) 1303 if (start < mmap_min_addr)
1273 return -EINVAL; 1304 return -EINVAL;
1274 if (start >= task_size) 1305 if (start >= task_size)
1275 return -EINVAL; 1306 return -EINVAL;
1276 if (len > task_size - start) 1307 if (len > task_size - start)
1277 return -EINVAL; 1308 return -EINVAL;
1278 if (start + len <= start) <<
1279 return -EINVAL; <<
1280 return 0; 1309 return 0;
1281 } 1310 }
1282 1311
1283 static __always_inline int validate_range(str <<
1284 __u <<
1285 { <<
1286 if (start & ~PAGE_MASK) <<
1287 return -EINVAL; <<
1288 <<
1289 return validate_unaligned_range(mm, s <<
1290 } <<
1291 <<
1292 static int userfaultfd_register(struct userfa 1312 static int userfaultfd_register(struct userfaultfd_ctx *ctx,
1293 unsigned long 1313 unsigned long arg)
1294 { 1314 {
1295 struct mm_struct *mm = ctx->mm; 1315 struct mm_struct *mm = ctx->mm;
1296 struct vm_area_struct *vma, *prev, *c 1316 struct vm_area_struct *vma, *prev, *cur;
1297 int ret; 1317 int ret;
1298 struct uffdio_register uffdio_registe 1318 struct uffdio_register uffdio_register;
1299 struct uffdio_register __user *user_u 1319 struct uffdio_register __user *user_uffdio_register;
1300 unsigned long vm_flags, new_flags; 1320 unsigned long vm_flags, new_flags;
1301 bool found; 1321 bool found;
1302 bool basic_ioctls; 1322 bool basic_ioctls;
1303 unsigned long start, end, vma_end; 1323 unsigned long start, end, vma_end;
1304 struct vma_iterator vmi; 1324 struct vma_iterator vmi;
1305 bool wp_async = userfaultfd_wp_async_ !! 1325 pgoff_t pgoff;
1306 1326
1307 user_uffdio_register = (struct uffdio 1327 user_uffdio_register = (struct uffdio_register __user *) arg;
1308 1328
1309 ret = -EFAULT; 1329 ret = -EFAULT;
1310 if (copy_from_user(&uffdio_register, 1330 if (copy_from_user(&uffdio_register, user_uffdio_register,
1311 sizeof(uffdio_regi 1331 sizeof(uffdio_register)-sizeof(__u64)))
1312 goto out; 1332 goto out;
1313 1333
1314 ret = -EINVAL; 1334 ret = -EINVAL;
1315 if (!uffdio_register.mode) 1335 if (!uffdio_register.mode)
1316 goto out; 1336 goto out;
1317 if (uffdio_register.mode & ~UFFD_API_ 1337 if (uffdio_register.mode & ~UFFD_API_REGISTER_MODES)
1318 goto out; 1338 goto out;
1319 vm_flags = 0; 1339 vm_flags = 0;
1320 if (uffdio_register.mode & UFFDIO_REG 1340 if (uffdio_register.mode & UFFDIO_REGISTER_MODE_MISSING)
1321 vm_flags |= VM_UFFD_MISSING; 1341 vm_flags |= VM_UFFD_MISSING;
1322 if (uffdio_register.mode & UFFDIO_REG 1342 if (uffdio_register.mode & UFFDIO_REGISTER_MODE_WP) {
1323 #ifndef CONFIG_HAVE_ARCH_USERFAULTFD_WP 1343 #ifndef CONFIG_HAVE_ARCH_USERFAULTFD_WP
1324 goto out; 1344 goto out;
1325 #endif 1345 #endif
1326 vm_flags |= VM_UFFD_WP; 1346 vm_flags |= VM_UFFD_WP;
1327 } 1347 }
1328 if (uffdio_register.mode & UFFDIO_REG 1348 if (uffdio_register.mode & UFFDIO_REGISTER_MODE_MINOR) {
1329 #ifndef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR 1349 #ifndef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
1330 goto out; 1350 goto out;
1331 #endif 1351 #endif
1332 vm_flags |= VM_UFFD_MINOR; 1352 vm_flags |= VM_UFFD_MINOR;
1333 } 1353 }
1334 1354
1335 ret = validate_range(mm, uffdio_regis 1355 ret = validate_range(mm, uffdio_register.range.start,
1336 uffdio_register. 1356 uffdio_register.range.len);
1337 if (ret) 1357 if (ret)
1338 goto out; 1358 goto out;
1339 1359
1340 start = uffdio_register.range.start; 1360 start = uffdio_register.range.start;
1341 end = start + uffdio_register.range.l 1361 end = start + uffdio_register.range.len;
1342 1362
1343 ret = -ENOMEM; 1363 ret = -ENOMEM;
1344 if (!mmget_not_zero(mm)) 1364 if (!mmget_not_zero(mm))
1345 goto out; 1365 goto out;
1346 1366
1347 ret = -EINVAL; 1367 ret = -EINVAL;
1348 mmap_write_lock(mm); 1368 mmap_write_lock(mm);
1349 vma_iter_init(&vmi, mm, start); 1369 vma_iter_init(&vmi, mm, start);
1350 vma = vma_find(&vmi, end); 1370 vma = vma_find(&vmi, end);
1351 if (!vma) 1371 if (!vma)
1352 goto out_unlock; 1372 goto out_unlock;
1353 1373
1354 /* 1374 /*
1355 * If the first vma contains huge pag 1375 * If the first vma contains huge pages, make sure start address
1356 * is aligned to huge page size. 1376 * is aligned to huge page size.
1357 */ 1377 */
1358 if (is_vm_hugetlb_page(vma)) { 1378 if (is_vm_hugetlb_page(vma)) {
1359 unsigned long vma_hpagesize = 1379 unsigned long vma_hpagesize = vma_kernel_pagesize(vma);
1360 1380
1361 if (start & (vma_hpagesize - 1381 if (start & (vma_hpagesize - 1))
1362 goto out_unlock; 1382 goto out_unlock;
1363 } 1383 }
1364 1384
1365 /* 1385 /*
1366 * Search for not compatible vmas. 1386 * Search for not compatible vmas.
1367 */ 1387 */
1368 found = false; 1388 found = false;
1369 basic_ioctls = false; 1389 basic_ioctls = false;
1370 cur = vma; 1390 cur = vma;
1371 do { 1391 do {
1372 cond_resched(); 1392 cond_resched();
1373 1393
1374 BUG_ON(!!cur->vm_userfaultfd_ 1394 BUG_ON(!!cur->vm_userfaultfd_ctx.ctx ^
1375 !!(cur->vm_flags & __V 1395 !!(cur->vm_flags & __VM_UFFD_FLAGS));
1376 1396
1377 /* check not compatible vmas 1397 /* check not compatible vmas */
1378 ret = -EINVAL; 1398 ret = -EINVAL;
1379 if (!vma_can_userfault(cur, v !! 1399 if (!vma_can_userfault(cur, vm_flags))
1380 goto out_unlock; 1400 goto out_unlock;
1381 1401
1382 /* 1402 /*
1383 * UFFDIO_COPY will fill file 1403 * UFFDIO_COPY will fill file holes even without
1384 * PROT_WRITE. This check enf 1404 * PROT_WRITE. This check enforces that if this is a
1385 * MAP_SHARED, the process ha 1405 * MAP_SHARED, the process has write permission to the backing
1386 * file. If VM_MAYWRITE is se 1406 * file. If VM_MAYWRITE is set it also enforces that on a
1387 * MAP_SHARED vma: there is n 1407 * MAP_SHARED vma: there is no F_WRITE_SEAL and no further
1388 * F_WRITE_SEAL can be taken 1408 * F_WRITE_SEAL can be taken until the vma is destroyed.
1389 */ 1409 */
1390 ret = -EPERM; 1410 ret = -EPERM;
1391 if (unlikely(!(cur->vm_flags 1411 if (unlikely(!(cur->vm_flags & VM_MAYWRITE)))
1392 goto out_unlock; 1412 goto out_unlock;
1393 1413
1394 /* 1414 /*
1395 * If this vma contains endin 1415 * If this vma contains ending address, and huge pages
1396 * check alignment. 1416 * check alignment.
1397 */ 1417 */
1398 if (is_vm_hugetlb_page(cur) & 1418 if (is_vm_hugetlb_page(cur) && end <= cur->vm_end &&
1399 end > cur->vm_start) { 1419 end > cur->vm_start) {
1400 unsigned long vma_hpa 1420 unsigned long vma_hpagesize = vma_kernel_pagesize(cur);
1401 1421
1402 ret = -EINVAL; 1422 ret = -EINVAL;
1403 1423
1404 if (end & (vma_hpages 1424 if (end & (vma_hpagesize - 1))
1405 goto out_unlo 1425 goto out_unlock;
1406 } 1426 }
1407 if ((vm_flags & VM_UFFD_WP) & 1427 if ((vm_flags & VM_UFFD_WP) && !(cur->vm_flags & VM_MAYWRITE))
1408 goto out_unlock; 1428 goto out_unlock;
1409 1429
1410 /* 1430 /*
1411 * Check that this vma isn't 1431 * Check that this vma isn't already owned by a
1412 * different userfaultfd. We 1432 * different userfaultfd. We can't allow more than one
1413 * userfaultfd to own a singl 1433 * userfaultfd to own a single vma simultaneously or we
1414 * wouldn't know which one to 1434 * wouldn't know which one to deliver the userfaults to.
1415 */ 1435 */
1416 ret = -EBUSY; 1436 ret = -EBUSY;
1417 if (cur->vm_userfaultfd_ctx.c 1437 if (cur->vm_userfaultfd_ctx.ctx &&
1418 cur->vm_userfaultfd_ctx.c 1438 cur->vm_userfaultfd_ctx.ctx != ctx)
1419 goto out_unlock; 1439 goto out_unlock;
1420 1440
1421 /* 1441 /*
1422 * Note vmas containing huge 1442 * Note vmas containing huge pages
1423 */ 1443 */
1424 if (is_vm_hugetlb_page(cur)) 1444 if (is_vm_hugetlb_page(cur))
1425 basic_ioctls = true; 1445 basic_ioctls = true;
1426 1446
1427 found = true; 1447 found = true;
1428 } for_each_vma_range(vmi, cur, end); 1448 } for_each_vma_range(vmi, cur, end);
1429 BUG_ON(!found); 1449 BUG_ON(!found);
1430 1450
1431 vma_iter_set(&vmi, start); 1451 vma_iter_set(&vmi, start);
1432 prev = vma_prev(&vmi); 1452 prev = vma_prev(&vmi);
1433 if (vma->vm_start < start) 1453 if (vma->vm_start < start)
1434 prev = vma; 1454 prev = vma;
1435 1455
1436 ret = 0; 1456 ret = 0;
1437 for_each_vma_range(vmi, vma, end) { 1457 for_each_vma_range(vmi, vma, end) {
1438 cond_resched(); 1458 cond_resched();
1439 1459
1440 BUG_ON(!vma_can_userfault(vma !! 1460 BUG_ON(!vma_can_userfault(vma, vm_flags));
1441 BUG_ON(vma->vm_userfaultfd_ct 1461 BUG_ON(vma->vm_userfaultfd_ctx.ctx &&
1442 vma->vm_userfaultfd_ct 1462 vma->vm_userfaultfd_ctx.ctx != ctx);
1443 WARN_ON(!(vma->vm_flags & VM_ 1463 WARN_ON(!(vma->vm_flags & VM_MAYWRITE));
1444 1464
1445 /* 1465 /*
1446 * Nothing to do: this vma is 1466 * Nothing to do: this vma is already registered into this
1447 * userfaultfd and with the r 1467 * userfaultfd and with the right tracking mode too.
1448 */ 1468 */
1449 if (vma->vm_userfaultfd_ctx.c 1469 if (vma->vm_userfaultfd_ctx.ctx == ctx &&
1450 (vma->vm_flags & vm_flags 1470 (vma->vm_flags & vm_flags) == vm_flags)
1451 goto skip; 1471 goto skip;
1452 1472
1453 if (vma->vm_start > start) 1473 if (vma->vm_start > start)
1454 start = vma->vm_start 1474 start = vma->vm_start;
1455 vma_end = min(end, vma->vm_en 1475 vma_end = min(end, vma->vm_end);
1456 1476
1457 new_flags = (vma->vm_flags & 1477 new_flags = (vma->vm_flags & ~__VM_UFFD_FLAGS) | vm_flags;
1458 vma = vma_modify_flags_uffd(& !! 1478 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
1459 n !! 1479 prev = vma_merge(&vmi, mm, prev, start, vma_end, new_flags,
1460 ( !! 1480 vma->anon_vma, vma->vm_file, pgoff,
1461 if (IS_ERR(vma)) { !! 1481 vma_policy(vma),
1462 ret = PTR_ERR(vma); !! 1482 ((struct vm_userfaultfd_ctx){ ctx }),
1463 break; !! 1483 anon_vma_name(vma));
>> 1484 if (prev) {
>> 1485 /* vma_merge() invalidated the mas */
>> 1486 vma = prev;
>> 1487 goto next;
1464 } 1488 }
1465 !! 1489 if (vma->vm_start < start) {
>> 1490 ret = split_vma(&vmi, vma, start, 1);
>> 1491 if (ret)
>> 1492 break;
>> 1493 }
>> 1494 if (vma->vm_end > end) {
>> 1495 ret = split_vma(&vmi, vma, end, 0);
>> 1496 if (ret)
>> 1497 break;
>> 1498 }
>> 1499 next:
1466 /* 1500 /*
1467 * In the vma_merge() success 1501 * In the vma_merge() successful mprotect-like case 8:
1468 * the next vma was merged in 1502 * the next vma was merged into the current one and
1469 * the current one has not be 1503 * the current one has not been updated yet.
1470 */ 1504 */
1471 vma_start_write(vma); <<
1472 userfaultfd_set_vm_flags(vma, 1505 userfaultfd_set_vm_flags(vma, new_flags);
1473 vma->vm_userfaultfd_ctx.ctx = 1506 vma->vm_userfaultfd_ctx.ctx = ctx;
1474 1507
1475 if (is_vm_hugetlb_page(vma) & 1508 if (is_vm_hugetlb_page(vma) && uffd_disable_huge_pmd_share(vma))
1476 hugetlb_unshare_all_p 1509 hugetlb_unshare_all_pmds(vma);
1477 1510
1478 skip: 1511 skip:
1479 prev = vma; 1512 prev = vma;
1480 start = vma->vm_end; 1513 start = vma->vm_end;
1481 } 1514 }
1482 1515
1483 out_unlock: 1516 out_unlock:
1484 mmap_write_unlock(mm); 1517 mmap_write_unlock(mm);
1485 mmput(mm); 1518 mmput(mm);
1486 if (!ret) { 1519 if (!ret) {
1487 __u64 ioctls_out; 1520 __u64 ioctls_out;
1488 1521
1489 ioctls_out = basic_ioctls ? U 1522 ioctls_out = basic_ioctls ? UFFD_API_RANGE_IOCTLS_BASIC :
1490 UFFD_API_RANGE_IOCTLS; 1523 UFFD_API_RANGE_IOCTLS;
1491 1524
1492 /* 1525 /*
1493 * Declare the WP ioctl only 1526 * Declare the WP ioctl only if the WP mode is
1494 * specified and all checks p 1527 * specified and all checks passed with the range
1495 */ 1528 */
1496 if (!(uffdio_register.mode & 1529 if (!(uffdio_register.mode & UFFDIO_REGISTER_MODE_WP))
1497 ioctls_out &= ~((__u6 1530 ioctls_out &= ~((__u64)1 << _UFFDIO_WRITEPROTECT);
1498 1531
1499 /* CONTINUE ioctl is only sup 1532 /* CONTINUE ioctl is only supported for MINOR ranges. */
1500 if (!(uffdio_register.mode & 1533 if (!(uffdio_register.mode & UFFDIO_REGISTER_MODE_MINOR))
1501 ioctls_out &= ~((__u6 1534 ioctls_out &= ~((__u64)1 << _UFFDIO_CONTINUE);
1502 1535
1503 /* 1536 /*
1504 * Now that we scanned all vm 1537 * Now that we scanned all vmas we can already tell
1505 * userland which ioctls meth 1538 * userland which ioctls methods are guaranteed to
1506 * succeed on this range. 1539 * succeed on this range.
1507 */ 1540 */
1508 if (put_user(ioctls_out, &use 1541 if (put_user(ioctls_out, &user_uffdio_register->ioctls))
1509 ret = -EFAULT; 1542 ret = -EFAULT;
1510 } 1543 }
1511 out: 1544 out:
1512 return ret; 1545 return ret;
1513 } 1546 }
1514 1547
1515 static int userfaultfd_unregister(struct user 1548 static int userfaultfd_unregister(struct userfaultfd_ctx *ctx,
1516 unsigned lo 1549 unsigned long arg)
1517 { 1550 {
1518 struct mm_struct *mm = ctx->mm; 1551 struct mm_struct *mm = ctx->mm;
1519 struct vm_area_struct *vma, *prev, *c 1552 struct vm_area_struct *vma, *prev, *cur;
1520 int ret; 1553 int ret;
1521 struct uffdio_range uffdio_unregister 1554 struct uffdio_range uffdio_unregister;
1522 unsigned long new_flags; 1555 unsigned long new_flags;
1523 bool found; 1556 bool found;
1524 unsigned long start, end, vma_end; 1557 unsigned long start, end, vma_end;
1525 const void __user *buf = (void __user 1558 const void __user *buf = (void __user *)arg;
1526 struct vma_iterator vmi; 1559 struct vma_iterator vmi;
1527 bool wp_async = userfaultfd_wp_async_ !! 1560 pgoff_t pgoff;
1528 1561
1529 ret = -EFAULT; 1562 ret = -EFAULT;
1530 if (copy_from_user(&uffdio_unregister 1563 if (copy_from_user(&uffdio_unregister, buf, sizeof(uffdio_unregister)))
1531 goto out; 1564 goto out;
1532 1565
1533 ret = validate_range(mm, uffdio_unreg 1566 ret = validate_range(mm, uffdio_unregister.start,
1534 uffdio_unregiste 1567 uffdio_unregister.len);
1535 if (ret) 1568 if (ret)
1536 goto out; 1569 goto out;
1537 1570
1538 start = uffdio_unregister.start; 1571 start = uffdio_unregister.start;
1539 end = start + uffdio_unregister.len; 1572 end = start + uffdio_unregister.len;
1540 1573
1541 ret = -ENOMEM; 1574 ret = -ENOMEM;
1542 if (!mmget_not_zero(mm)) 1575 if (!mmget_not_zero(mm))
1543 goto out; 1576 goto out;
1544 1577
1545 mmap_write_lock(mm); 1578 mmap_write_lock(mm);
1546 ret = -EINVAL; 1579 ret = -EINVAL;
1547 vma_iter_init(&vmi, mm, start); 1580 vma_iter_init(&vmi, mm, start);
1548 vma = vma_find(&vmi, end); 1581 vma = vma_find(&vmi, end);
1549 if (!vma) 1582 if (!vma)
1550 goto out_unlock; 1583 goto out_unlock;
1551 1584
1552 /* 1585 /*
1553 * If the first vma contains huge pag 1586 * If the first vma contains huge pages, make sure start address
1554 * is aligned to huge page size. 1587 * is aligned to huge page size.
1555 */ 1588 */
1556 if (is_vm_hugetlb_page(vma)) { 1589 if (is_vm_hugetlb_page(vma)) {
1557 unsigned long vma_hpagesize = 1590 unsigned long vma_hpagesize = vma_kernel_pagesize(vma);
1558 1591
1559 if (start & (vma_hpagesize - 1592 if (start & (vma_hpagesize - 1))
1560 goto out_unlock; 1593 goto out_unlock;
1561 } 1594 }
1562 1595
1563 /* 1596 /*
1564 * Search for not compatible vmas. 1597 * Search for not compatible vmas.
1565 */ 1598 */
1566 found = false; 1599 found = false;
1567 cur = vma; 1600 cur = vma;
1568 do { 1601 do {
1569 cond_resched(); 1602 cond_resched();
1570 1603
1571 BUG_ON(!!cur->vm_userfaultfd_ 1604 BUG_ON(!!cur->vm_userfaultfd_ctx.ctx ^
1572 !!(cur->vm_flags & __V 1605 !!(cur->vm_flags & __VM_UFFD_FLAGS));
1573 1606
1574 /* 1607 /*
1575 * Check not compatible vmas, 1608 * Check not compatible vmas, not strictly required
1576 * here as not compatible vma 1609 * here as not compatible vmas cannot have an
1577 * userfaultfd_ctx registered 1610 * userfaultfd_ctx registered on them, but this
1578 * provides for more strict b 1611 * provides for more strict behavior to notice
1579 * unregistration errors. 1612 * unregistration errors.
1580 */ 1613 */
1581 if (!vma_can_userfault(cur, c !! 1614 if (!vma_can_userfault(cur, cur->vm_flags))
1582 goto out_unlock; 1615 goto out_unlock;
1583 1616
1584 found = true; 1617 found = true;
1585 } for_each_vma_range(vmi, cur, end); 1618 } for_each_vma_range(vmi, cur, end);
1586 BUG_ON(!found); 1619 BUG_ON(!found);
1587 1620
1588 vma_iter_set(&vmi, start); 1621 vma_iter_set(&vmi, start);
1589 prev = vma_prev(&vmi); 1622 prev = vma_prev(&vmi);
1590 if (vma->vm_start < start) 1623 if (vma->vm_start < start)
1591 prev = vma; 1624 prev = vma;
1592 1625
1593 ret = 0; 1626 ret = 0;
1594 for_each_vma_range(vmi, vma, end) { 1627 for_each_vma_range(vmi, vma, end) {
1595 cond_resched(); 1628 cond_resched();
1596 1629
1597 BUG_ON(!vma_can_userfault(vma !! 1630 BUG_ON(!vma_can_userfault(vma, vma->vm_flags));
1598 1631
1599 /* 1632 /*
1600 * Nothing to do: this vma is 1633 * Nothing to do: this vma is already registered into this
1601 * userfaultfd and with the r 1634 * userfaultfd and with the right tracking mode too.
1602 */ 1635 */
1603 if (!vma->vm_userfaultfd_ctx. 1636 if (!vma->vm_userfaultfd_ctx.ctx)
1604 goto skip; 1637 goto skip;
1605 1638
1606 WARN_ON(!(vma->vm_flags & VM_ 1639 WARN_ON(!(vma->vm_flags & VM_MAYWRITE));
1607 1640
1608 if (vma->vm_start > start) 1641 if (vma->vm_start > start)
1609 start = vma->vm_start 1642 start = vma->vm_start;
1610 vma_end = min(end, vma->vm_en 1643 vma_end = min(end, vma->vm_end);
1611 1644
1612 if (userfaultfd_missing(vma)) 1645 if (userfaultfd_missing(vma)) {
1613 /* 1646 /*
1614 * Wake any concurren 1647 * Wake any concurrent pending userfault while
1615 * we unregister, so 1648 * we unregister, so they will not hang
1616 * permanently and it 1649 * permanently and it avoids userland to call
1617 * UFFDIO_WAKE explic 1650 * UFFDIO_WAKE explicitly.
1618 */ 1651 */
1619 struct userfaultfd_wa 1652 struct userfaultfd_wake_range range;
1620 range.start = start; 1653 range.start = start;
1621 range.len = vma_end - 1654 range.len = vma_end - start;
1622 wake_userfault(vma->v 1655 wake_userfault(vma->vm_userfaultfd_ctx.ctx, &range);
1623 } 1656 }
1624 1657
1625 /* Reset ptes for the whole v 1658 /* Reset ptes for the whole vma range if wr-protected */
1626 if (userfaultfd_wp(vma)) 1659 if (userfaultfd_wp(vma))
1627 uffd_wp_range(vma, st 1660 uffd_wp_range(vma, start, vma_end - start, false);
1628 1661
1629 new_flags = vma->vm_flags & ~ 1662 new_flags = vma->vm_flags & ~__VM_UFFD_FLAGS;
1630 vma = vma_modify_flags_uffd(& !! 1663 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
1631 n !! 1664 prev = vma_merge(&vmi, mm, prev, start, vma_end, new_flags,
1632 if (IS_ERR(vma)) { !! 1665 vma->anon_vma, vma->vm_file, pgoff,
1633 ret = PTR_ERR(vma); !! 1666 vma_policy(vma),
1634 break; !! 1667 NULL_VM_UFFD_CTX, anon_vma_name(vma));
>> 1668 if (prev) {
>> 1669 vma = prev;
>> 1670 goto next;
1635 } 1671 }
1636 !! 1672 if (vma->vm_start < start) {
>> 1673 ret = split_vma(&vmi, vma, start, 1);
>> 1674 if (ret)
>> 1675 break;
>> 1676 }
>> 1677 if (vma->vm_end > end) {
>> 1678 ret = split_vma(&vmi, vma, end, 0);
>> 1679 if (ret)
>> 1680 break;
>> 1681 }
>> 1682 next:
1637 /* 1683 /*
1638 * In the vma_merge() success 1684 * In the vma_merge() successful mprotect-like case 8:
1639 * the next vma was merged in 1685 * the next vma was merged into the current one and
1640 * the current one has not be 1686 * the current one has not been updated yet.
1641 */ 1687 */
1642 vma_start_write(vma); <<
1643 userfaultfd_set_vm_flags(vma, 1688 userfaultfd_set_vm_flags(vma, new_flags);
1644 vma->vm_userfaultfd_ctx = NUL 1689 vma->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
1645 1690
1646 skip: 1691 skip:
1647 prev = vma; 1692 prev = vma;
1648 start = vma->vm_end; 1693 start = vma->vm_end;
1649 } 1694 }
1650 1695
1651 out_unlock: 1696 out_unlock:
1652 mmap_write_unlock(mm); 1697 mmap_write_unlock(mm);
1653 mmput(mm); 1698 mmput(mm);
1654 out: 1699 out:
1655 return ret; 1700 return ret;
1656 } 1701 }
1657 1702
1658 /* 1703 /*
1659 * userfaultfd_wake may be used in combinatio 1704 * userfaultfd_wake may be used in combination with the
1660 * UFFDIO_*_MODE_DONTWAKE to wakeup userfault 1705 * UFFDIO_*_MODE_DONTWAKE to wakeup userfaults in batches.
1661 */ 1706 */
1662 static int userfaultfd_wake(struct userfaultf 1707 static int userfaultfd_wake(struct userfaultfd_ctx *ctx,
1663 unsigned long arg 1708 unsigned long arg)
1664 { 1709 {
1665 int ret; 1710 int ret;
1666 struct uffdio_range uffdio_wake; 1711 struct uffdio_range uffdio_wake;
1667 struct userfaultfd_wake_range range; 1712 struct userfaultfd_wake_range range;
1668 const void __user *buf = (void __user 1713 const void __user *buf = (void __user *)arg;
1669 1714
1670 ret = -EFAULT; 1715 ret = -EFAULT;
1671 if (copy_from_user(&uffdio_wake, buf, 1716 if (copy_from_user(&uffdio_wake, buf, sizeof(uffdio_wake)))
1672 goto out; 1717 goto out;
1673 1718
1674 ret = validate_range(ctx->mm, uffdio_ 1719 ret = validate_range(ctx->mm, uffdio_wake.start, uffdio_wake.len);
1675 if (ret) 1720 if (ret)
1676 goto out; 1721 goto out;
1677 1722
1678 range.start = uffdio_wake.start; 1723 range.start = uffdio_wake.start;
1679 range.len = uffdio_wake.len; 1724 range.len = uffdio_wake.len;
1680 1725
1681 /* 1726 /*
1682 * len == 0 means wake all and we don 1727 * len == 0 means wake all and we don't want to wake all here,
1683 * so check it again to be sure. 1728 * so check it again to be sure.
1684 */ 1729 */
1685 VM_BUG_ON(!range.len); 1730 VM_BUG_ON(!range.len);
1686 1731
1687 wake_userfault(ctx, &range); 1732 wake_userfault(ctx, &range);
1688 ret = 0; 1733 ret = 0;
1689 1734
1690 out: 1735 out:
1691 return ret; 1736 return ret;
1692 } 1737 }
1693 1738
1694 static int userfaultfd_copy(struct userfaultf 1739 static int userfaultfd_copy(struct userfaultfd_ctx *ctx,
1695 unsigned long arg 1740 unsigned long arg)
1696 { 1741 {
1697 __s64 ret; 1742 __s64 ret;
1698 struct uffdio_copy uffdio_copy; 1743 struct uffdio_copy uffdio_copy;
1699 struct uffdio_copy __user *user_uffdi 1744 struct uffdio_copy __user *user_uffdio_copy;
1700 struct userfaultfd_wake_range range; 1745 struct userfaultfd_wake_range range;
1701 uffd_flags_t flags = 0; 1746 uffd_flags_t flags = 0;
1702 1747
1703 user_uffdio_copy = (struct uffdio_cop 1748 user_uffdio_copy = (struct uffdio_copy __user *) arg;
1704 1749
1705 ret = -EAGAIN; 1750 ret = -EAGAIN;
1706 if (atomic_read(&ctx->mmap_changing)) 1751 if (atomic_read(&ctx->mmap_changing))
1707 goto out; 1752 goto out;
1708 1753
1709 ret = -EFAULT; 1754 ret = -EFAULT;
1710 if (copy_from_user(&uffdio_copy, user 1755 if (copy_from_user(&uffdio_copy, user_uffdio_copy,
1711 /* don't copy "cop 1756 /* don't copy "copy" last field */
1712 sizeof(uffdio_copy 1757 sizeof(uffdio_copy)-sizeof(__s64)))
1713 goto out; 1758 goto out;
1714 1759
1715 ret = validate_unaligned_range(ctx->m <<
1716 uffdio <<
1717 if (ret) <<
1718 goto out; <<
1719 ret = validate_range(ctx->mm, uffdio_ 1760 ret = validate_range(ctx->mm, uffdio_copy.dst, uffdio_copy.len);
1720 if (ret) 1761 if (ret)
1721 goto out; 1762 goto out;
1722 !! 1763 /*
>> 1764 * double check for wraparound just in case. copy_from_user()
>> 1765 * will later check uffdio_copy.src + uffdio_copy.len to fit
>> 1766 * in the userland range.
>> 1767 */
1723 ret = -EINVAL; 1768 ret = -EINVAL;
>> 1769 if (uffdio_copy.src + uffdio_copy.len <= uffdio_copy.src)
>> 1770 goto out;
1724 if (uffdio_copy.mode & ~(UFFDIO_COPY_ 1771 if (uffdio_copy.mode & ~(UFFDIO_COPY_MODE_DONTWAKE|UFFDIO_COPY_MODE_WP))
1725 goto out; 1772 goto out;
1726 if (uffdio_copy.mode & UFFDIO_COPY_MO 1773 if (uffdio_copy.mode & UFFDIO_COPY_MODE_WP)
1727 flags |= MFILL_ATOMIC_WP; 1774 flags |= MFILL_ATOMIC_WP;
1728 if (mmget_not_zero(ctx->mm)) { 1775 if (mmget_not_zero(ctx->mm)) {
1729 ret = mfill_atomic_copy(ctx, !! 1776 ret = mfill_atomic_copy(ctx->mm, uffdio_copy.dst, uffdio_copy.src,
1730 uffdi !! 1777 uffdio_copy.len, &ctx->mmap_changing,
>> 1778 flags);
1731 mmput(ctx->mm); 1779 mmput(ctx->mm);
1732 } else { 1780 } else {
1733 return -ESRCH; 1781 return -ESRCH;
1734 } 1782 }
1735 if (unlikely(put_user(ret, &user_uffd 1783 if (unlikely(put_user(ret, &user_uffdio_copy->copy)))
1736 return -EFAULT; 1784 return -EFAULT;
1737 if (ret < 0) 1785 if (ret < 0)
1738 goto out; 1786 goto out;
1739 BUG_ON(!ret); 1787 BUG_ON(!ret);
1740 /* len == 0 would wake all */ 1788 /* len == 0 would wake all */
1741 range.len = ret; 1789 range.len = ret;
1742 if (!(uffdio_copy.mode & UFFDIO_COPY_ 1790 if (!(uffdio_copy.mode & UFFDIO_COPY_MODE_DONTWAKE)) {
1743 range.start = uffdio_copy.dst 1791 range.start = uffdio_copy.dst;
1744 wake_userfault(ctx, &range); 1792 wake_userfault(ctx, &range);
1745 } 1793 }
1746 ret = range.len == uffdio_copy.len ? 1794 ret = range.len == uffdio_copy.len ? 0 : -EAGAIN;
1747 out: 1795 out:
1748 return ret; 1796 return ret;
1749 } 1797 }
1750 1798
1751 static int userfaultfd_zeropage(struct userfa 1799 static int userfaultfd_zeropage(struct userfaultfd_ctx *ctx,
1752 unsigned long 1800 unsigned long arg)
1753 { 1801 {
1754 __s64 ret; 1802 __s64 ret;
1755 struct uffdio_zeropage uffdio_zeropag 1803 struct uffdio_zeropage uffdio_zeropage;
1756 struct uffdio_zeropage __user *user_u 1804 struct uffdio_zeropage __user *user_uffdio_zeropage;
1757 struct userfaultfd_wake_range range; 1805 struct userfaultfd_wake_range range;
1758 1806
1759 user_uffdio_zeropage = (struct uffdio 1807 user_uffdio_zeropage = (struct uffdio_zeropage __user *) arg;
1760 1808
1761 ret = -EAGAIN; 1809 ret = -EAGAIN;
1762 if (atomic_read(&ctx->mmap_changing)) 1810 if (atomic_read(&ctx->mmap_changing))
1763 goto out; 1811 goto out;
1764 1812
1765 ret = -EFAULT; 1813 ret = -EFAULT;
1766 if (copy_from_user(&uffdio_zeropage, 1814 if (copy_from_user(&uffdio_zeropage, user_uffdio_zeropage,
1767 /* don't copy "zer 1815 /* don't copy "zeropage" last field */
1768 sizeof(uffdio_zero 1816 sizeof(uffdio_zeropage)-sizeof(__s64)))
1769 goto out; 1817 goto out;
1770 1818
1771 ret = validate_range(ctx->mm, uffdio_ 1819 ret = validate_range(ctx->mm, uffdio_zeropage.range.start,
1772 uffdio_zeropage. 1820 uffdio_zeropage.range.len);
1773 if (ret) 1821 if (ret)
1774 goto out; 1822 goto out;
1775 ret = -EINVAL; 1823 ret = -EINVAL;
1776 if (uffdio_zeropage.mode & ~UFFDIO_ZE 1824 if (uffdio_zeropage.mode & ~UFFDIO_ZEROPAGE_MODE_DONTWAKE)
1777 goto out; 1825 goto out;
1778 1826
1779 if (mmget_not_zero(ctx->mm)) { 1827 if (mmget_not_zero(ctx->mm)) {
1780 ret = mfill_atomic_zeropage(c !! 1828 ret = mfill_atomic_zeropage(ctx->mm, uffdio_zeropage.range.start,
1781 uf !! 1829 uffdio_zeropage.range.len,
>> 1830 &ctx->mmap_changing);
1782 mmput(ctx->mm); 1831 mmput(ctx->mm);
1783 } else { 1832 } else {
1784 return -ESRCH; 1833 return -ESRCH;
1785 } 1834 }
1786 if (unlikely(put_user(ret, &user_uffd 1835 if (unlikely(put_user(ret, &user_uffdio_zeropage->zeropage)))
1787 return -EFAULT; 1836 return -EFAULT;
1788 if (ret < 0) 1837 if (ret < 0)
1789 goto out; 1838 goto out;
1790 /* len == 0 would wake all */ 1839 /* len == 0 would wake all */
1791 BUG_ON(!ret); 1840 BUG_ON(!ret);
1792 range.len = ret; 1841 range.len = ret;
1793 if (!(uffdio_zeropage.mode & UFFDIO_Z 1842 if (!(uffdio_zeropage.mode & UFFDIO_ZEROPAGE_MODE_DONTWAKE)) {
1794 range.start = uffdio_zeropage 1843 range.start = uffdio_zeropage.range.start;
1795 wake_userfault(ctx, &range); 1844 wake_userfault(ctx, &range);
1796 } 1845 }
1797 ret = range.len == uffdio_zeropage.ra 1846 ret = range.len == uffdio_zeropage.range.len ? 0 : -EAGAIN;
1798 out: 1847 out:
1799 return ret; 1848 return ret;
1800 } 1849 }
1801 1850
1802 static int userfaultfd_writeprotect(struct us 1851 static int userfaultfd_writeprotect(struct userfaultfd_ctx *ctx,
1803 unsigned 1852 unsigned long arg)
1804 { 1853 {
1805 int ret; 1854 int ret;
1806 struct uffdio_writeprotect uffdio_wp; 1855 struct uffdio_writeprotect uffdio_wp;
1807 struct uffdio_writeprotect __user *us 1856 struct uffdio_writeprotect __user *user_uffdio_wp;
1808 struct userfaultfd_wake_range range; 1857 struct userfaultfd_wake_range range;
1809 bool mode_wp, mode_dontwake; 1858 bool mode_wp, mode_dontwake;
1810 1859
1811 if (atomic_read(&ctx->mmap_changing)) 1860 if (atomic_read(&ctx->mmap_changing))
1812 return -EAGAIN; 1861 return -EAGAIN;
1813 1862
1814 user_uffdio_wp = (struct uffdio_write 1863 user_uffdio_wp = (struct uffdio_writeprotect __user *) arg;
1815 1864
1816 if (copy_from_user(&uffdio_wp, user_u 1865 if (copy_from_user(&uffdio_wp, user_uffdio_wp,
1817 sizeof(struct uffd 1866 sizeof(struct uffdio_writeprotect)))
1818 return -EFAULT; 1867 return -EFAULT;
1819 1868
1820 ret = validate_range(ctx->mm, uffdio_ 1869 ret = validate_range(ctx->mm, uffdio_wp.range.start,
1821 uffdio_wp.range. 1870 uffdio_wp.range.len);
1822 if (ret) 1871 if (ret)
1823 return ret; 1872 return ret;
1824 1873
1825 if (uffdio_wp.mode & ~(UFFDIO_WRITEPR 1874 if (uffdio_wp.mode & ~(UFFDIO_WRITEPROTECT_MODE_DONTWAKE |
1826 UFFDIO_WRITEPR 1875 UFFDIO_WRITEPROTECT_MODE_WP))
1827 return -EINVAL; 1876 return -EINVAL;
1828 1877
1829 mode_wp = uffdio_wp.mode & UFFDIO_WRI 1878 mode_wp = uffdio_wp.mode & UFFDIO_WRITEPROTECT_MODE_WP;
1830 mode_dontwake = uffdio_wp.mode & UFFD 1879 mode_dontwake = uffdio_wp.mode & UFFDIO_WRITEPROTECT_MODE_DONTWAKE;
1831 1880
1832 if (mode_wp && mode_dontwake) 1881 if (mode_wp && mode_dontwake)
1833 return -EINVAL; 1882 return -EINVAL;
1834 1883
1835 if (mmget_not_zero(ctx->mm)) { 1884 if (mmget_not_zero(ctx->mm)) {
1836 ret = mwriteprotect_range(ctx !! 1885 ret = mwriteprotect_range(ctx->mm, uffdio_wp.range.start,
1837 uff !! 1886 uffdio_wp.range.len, mode_wp,
>> 1887 &ctx->mmap_changing);
1838 mmput(ctx->mm); 1888 mmput(ctx->mm);
1839 } else { 1889 } else {
1840 return -ESRCH; 1890 return -ESRCH;
1841 } 1891 }
1842 1892
1843 if (ret) 1893 if (ret)
1844 return ret; 1894 return ret;
1845 1895
1846 if (!mode_wp && !mode_dontwake) { 1896 if (!mode_wp && !mode_dontwake) {
1847 range.start = uffdio_wp.range 1897 range.start = uffdio_wp.range.start;
1848 range.len = uffdio_wp.range.l 1898 range.len = uffdio_wp.range.len;
1849 wake_userfault(ctx, &range); 1899 wake_userfault(ctx, &range);
1850 } 1900 }
1851 return ret; 1901 return ret;
1852 } 1902 }
1853 1903
1854 static int userfaultfd_continue(struct userfa 1904 static int userfaultfd_continue(struct userfaultfd_ctx *ctx, unsigned long arg)
1855 { 1905 {
1856 __s64 ret; 1906 __s64 ret;
1857 struct uffdio_continue uffdio_continu 1907 struct uffdio_continue uffdio_continue;
1858 struct uffdio_continue __user *user_u 1908 struct uffdio_continue __user *user_uffdio_continue;
1859 struct userfaultfd_wake_range range; 1909 struct userfaultfd_wake_range range;
1860 uffd_flags_t flags = 0; 1910 uffd_flags_t flags = 0;
1861 1911
1862 user_uffdio_continue = (struct uffdio 1912 user_uffdio_continue = (struct uffdio_continue __user *)arg;
1863 1913
1864 ret = -EAGAIN; 1914 ret = -EAGAIN;
1865 if (atomic_read(&ctx->mmap_changing)) 1915 if (atomic_read(&ctx->mmap_changing))
1866 goto out; 1916 goto out;
1867 1917
1868 ret = -EFAULT; 1918 ret = -EFAULT;
1869 if (copy_from_user(&uffdio_continue, 1919 if (copy_from_user(&uffdio_continue, user_uffdio_continue,
1870 /* don't copy the 1920 /* don't copy the output fields */
1871 sizeof(uffdio_cont 1921 sizeof(uffdio_continue) - (sizeof(__s64))))
1872 goto out; 1922 goto out;
1873 1923
1874 ret = validate_range(ctx->mm, uffdio_ 1924 ret = validate_range(ctx->mm, uffdio_continue.range.start,
1875 uffdio_continue. 1925 uffdio_continue.range.len);
1876 if (ret) 1926 if (ret)
1877 goto out; 1927 goto out;
1878 1928
1879 ret = -EINVAL; 1929 ret = -EINVAL;
>> 1930 /* double check for wraparound just in case. */
>> 1931 if (uffdio_continue.range.start + uffdio_continue.range.len <=
>> 1932 uffdio_continue.range.start) {
>> 1933 goto out;
>> 1934 }
1880 if (uffdio_continue.mode & ~(UFFDIO_C 1935 if (uffdio_continue.mode & ~(UFFDIO_CONTINUE_MODE_DONTWAKE |
1881 UFFDIO_C 1936 UFFDIO_CONTINUE_MODE_WP))
1882 goto out; 1937 goto out;
1883 if (uffdio_continue.mode & UFFDIO_CON 1938 if (uffdio_continue.mode & UFFDIO_CONTINUE_MODE_WP)
1884 flags |= MFILL_ATOMIC_WP; 1939 flags |= MFILL_ATOMIC_WP;
1885 1940
1886 if (mmget_not_zero(ctx->mm)) { 1941 if (mmget_not_zero(ctx->mm)) {
1887 ret = mfill_atomic_continue(c !! 1942 ret = mfill_atomic_continue(ctx->mm, uffdio_continue.range.start,
1888 u !! 1943 uffdio_continue.range.len,
>> 1944 &ctx->mmap_changing, flags);
1889 mmput(ctx->mm); 1945 mmput(ctx->mm);
1890 } else { 1946 } else {
1891 return -ESRCH; 1947 return -ESRCH;
1892 } 1948 }
1893 1949
1894 if (unlikely(put_user(ret, &user_uffd 1950 if (unlikely(put_user(ret, &user_uffdio_continue->mapped)))
1895 return -EFAULT; 1951 return -EFAULT;
1896 if (ret < 0) 1952 if (ret < 0)
1897 goto out; 1953 goto out;
1898 1954
1899 /* len == 0 would wake all */ 1955 /* len == 0 would wake all */
1900 BUG_ON(!ret); 1956 BUG_ON(!ret);
1901 range.len = ret; 1957 range.len = ret;
1902 if (!(uffdio_continue.mode & UFFDIO_C 1958 if (!(uffdio_continue.mode & UFFDIO_CONTINUE_MODE_DONTWAKE)) {
1903 range.start = uffdio_continue 1959 range.start = uffdio_continue.range.start;
1904 wake_userfault(ctx, &range); 1960 wake_userfault(ctx, &range);
1905 } 1961 }
1906 ret = range.len == uffdio_continue.ra 1962 ret = range.len == uffdio_continue.range.len ? 0 : -EAGAIN;
1907 1963
1908 out: 1964 out:
1909 return ret; 1965 return ret;
1910 } 1966 }
1911 1967
1912 static inline int userfaultfd_poison(struct u <<
1913 { <<
1914 __s64 ret; <<
1915 struct uffdio_poison uffdio_poison; <<
1916 struct uffdio_poison __user *user_uff <<
1917 struct userfaultfd_wake_range range; <<
1918 <<
1919 user_uffdio_poison = (struct uffdio_p <<
1920 <<
1921 ret = -EAGAIN; <<
1922 if (atomic_read(&ctx->mmap_changing)) <<
1923 goto out; <<
1924 <<
1925 ret = -EFAULT; <<
1926 if (copy_from_user(&uffdio_poison, us <<
1927 /* don't copy the <<
1928 sizeof(uffdio_pois <<
1929 goto out; <<
1930 <<
1931 ret = validate_range(ctx->mm, uffdio_ <<
1932 uffdio_poison.ra <<
1933 if (ret) <<
1934 goto out; <<
1935 <<
1936 ret = -EINVAL; <<
1937 if (uffdio_poison.mode & ~UFFDIO_POIS <<
1938 goto out; <<
1939 <<
1940 if (mmget_not_zero(ctx->mm)) { <<
1941 ret = mfill_atomic_poison(ctx <<
1942 uff <<
1943 mmput(ctx->mm); <<
1944 } else { <<
1945 return -ESRCH; <<
1946 } <<
1947 <<
1948 if (unlikely(put_user(ret, &user_uffd <<
1949 return -EFAULT; <<
1950 if (ret < 0) <<
1951 goto out; <<
1952 <<
1953 /* len == 0 would wake all */ <<
1954 BUG_ON(!ret); <<
1955 range.len = ret; <<
1956 if (!(uffdio_poison.mode & UFFDIO_POI <<
1957 range.start = uffdio_poison.r <<
1958 wake_userfault(ctx, &range); <<
1959 } <<
1960 ret = range.len == uffdio_poison.rang <<
1961 <<
1962 out: <<
1963 return ret; <<
1964 } <<
1965 <<
1966 bool userfaultfd_wp_async(struct vm_area_stru <<
1967 { <<
1968 return userfaultfd_wp_async_ctx(vma-> <<
1969 } <<
1970 <<
1971 static inline unsigned int uffd_ctx_features( 1968 static inline unsigned int uffd_ctx_features(__u64 user_features)
1972 { 1969 {
1973 /* 1970 /*
1974 * For the current set of features th 1971 * For the current set of features the bits just coincide. Set
1975 * UFFD_FEATURE_INITIALIZED to mark t 1972 * UFFD_FEATURE_INITIALIZED to mark the features as enabled.
1976 */ 1973 */
1977 return (unsigned int)user_features | 1974 return (unsigned int)user_features | UFFD_FEATURE_INITIALIZED;
1978 } 1975 }
1979 1976
1980 static int userfaultfd_move(struct userfaultf <<
1981 unsigned long arg <<
1982 { <<
1983 __s64 ret; <<
1984 struct uffdio_move uffdio_move; <<
1985 struct uffdio_move __user *user_uffdi <<
1986 struct userfaultfd_wake_range range; <<
1987 struct mm_struct *mm = ctx->mm; <<
1988 <<
1989 user_uffdio_move = (struct uffdio_mov <<
1990 <<
1991 if (atomic_read(&ctx->mmap_changing)) <<
1992 return -EAGAIN; <<
1993 <<
1994 if (copy_from_user(&uffdio_move, user <<
1995 /* don't copy "mov <<
1996 sizeof(uffdio_move <<
1997 return -EFAULT; <<
1998 <<
1999 /* Do not allow cross-mm moves. */ <<
2000 if (mm != current->mm) <<
2001 return -EINVAL; <<
2002 <<
2003 ret = validate_range(mm, uffdio_move. <<
2004 if (ret) <<
2005 return ret; <<
2006 <<
2007 ret = validate_range(mm, uffdio_move. <<
2008 if (ret) <<
2009 return ret; <<
2010 <<
2011 if (uffdio_move.mode & ~(UFFDIO_MOVE_ <<
2012 UFFDIO_MOVE <<
2013 return -EINVAL; <<
2014 <<
2015 if (mmget_not_zero(mm)) { <<
2016 ret = move_pages(ctx, uffdio_ <<
2017 uffdio_move. <<
2018 mmput(mm); <<
2019 } else { <<
2020 return -ESRCH; <<
2021 } <<
2022 <<
2023 if (unlikely(put_user(ret, &user_uffd <<
2024 return -EFAULT; <<
2025 if (ret < 0) <<
2026 goto out; <<
2027 <<
2028 /* len == 0 would wake all */ <<
2029 VM_WARN_ON(!ret); <<
2030 range.len = ret; <<
2031 if (!(uffdio_move.mode & UFFDIO_MOVE_ <<
2032 range.start = uffdio_move.dst <<
2033 wake_userfault(ctx, &range); <<
2034 } <<
2035 ret = range.len == uffdio_move.len ? <<
2036 <<
2037 out: <<
2038 return ret; <<
2039 } <<
2040 <<
2041 /* 1977 /*
2042 * userland asks for a certain API version an 1978 * userland asks for a certain API version and we return which bits
2043 * and ioctl commands are implemented in this 1979 * and ioctl commands are implemented in this kernel for such API
2044 * version or -EINVAL if unknown. 1980 * version or -EINVAL if unknown.
2045 */ 1981 */
2046 static int userfaultfd_api(struct userfaultfd 1982 static int userfaultfd_api(struct userfaultfd_ctx *ctx,
2047 unsigned long arg) 1983 unsigned long arg)
2048 { 1984 {
2049 struct uffdio_api uffdio_api; 1985 struct uffdio_api uffdio_api;
2050 void __user *buf = (void __user *)arg 1986 void __user *buf = (void __user *)arg;
2051 unsigned int ctx_features; 1987 unsigned int ctx_features;
2052 int ret; 1988 int ret;
2053 __u64 features; 1989 __u64 features;
2054 1990
2055 ret = -EFAULT; 1991 ret = -EFAULT;
2056 if (copy_from_user(&uffdio_api, buf, 1992 if (copy_from_user(&uffdio_api, buf, sizeof(uffdio_api)))
2057 goto out; 1993 goto out;
2058 features = uffdio_api.features; 1994 features = uffdio_api.features;
2059 ret = -EINVAL; 1995 ret = -EINVAL;
2060 if (uffdio_api.api != UFFD_API) !! 1996 if (uffdio_api.api != UFFD_API || (features & ~UFFD_API_FEATURES))
2061 goto err_out; 1997 goto err_out;
2062 ret = -EPERM; 1998 ret = -EPERM;
2063 if ((features & UFFD_FEATURE_EVENT_FO 1999 if ((features & UFFD_FEATURE_EVENT_FORK) && !capable(CAP_SYS_PTRACE))
2064 goto err_out; 2000 goto err_out;
2065 <<
2066 /* WP_ASYNC relies on WP_UNPOPULATED, <<
2067 if (features & UFFD_FEATURE_WP_ASYNC) <<
2068 features |= UFFD_FEATURE_WP_U <<
2069 <<
2070 /* report all available features and 2001 /* report all available features and ioctls to userland */
2071 uffdio_api.features = UFFD_API_FEATUR 2002 uffdio_api.features = UFFD_API_FEATURES;
2072 #ifndef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR 2003 #ifndef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
2073 uffdio_api.features &= 2004 uffdio_api.features &=
2074 ~(UFFD_FEATURE_MINOR_HUGETLBF 2005 ~(UFFD_FEATURE_MINOR_HUGETLBFS | UFFD_FEATURE_MINOR_SHMEM);
2075 #endif 2006 #endif
2076 #ifndef CONFIG_HAVE_ARCH_USERFAULTFD_WP 2007 #ifndef CONFIG_HAVE_ARCH_USERFAULTFD_WP
2077 uffdio_api.features &= ~UFFD_FEATURE_ 2008 uffdio_api.features &= ~UFFD_FEATURE_PAGEFAULT_FLAG_WP;
2078 #endif 2009 #endif
2079 #ifndef CONFIG_PTE_MARKER_UFFD_WP 2010 #ifndef CONFIG_PTE_MARKER_UFFD_WP
2080 uffdio_api.features &= ~UFFD_FEATURE_ 2011 uffdio_api.features &= ~UFFD_FEATURE_WP_HUGETLBFS_SHMEM;
2081 uffdio_api.features &= ~UFFD_FEATURE_ 2012 uffdio_api.features &= ~UFFD_FEATURE_WP_UNPOPULATED;
2082 uffdio_api.features &= ~UFFD_FEATURE_ <<
2083 #endif 2013 #endif
2084 <<
2085 ret = -EINVAL; <<
2086 if (features & ~uffdio_api.features) <<
2087 goto err_out; <<
2088 <<
2089 uffdio_api.ioctls = UFFD_API_IOCTLS; 2014 uffdio_api.ioctls = UFFD_API_IOCTLS;
2090 ret = -EFAULT; 2015 ret = -EFAULT;
2091 if (copy_to_user(buf, &uffdio_api, si 2016 if (copy_to_user(buf, &uffdio_api, sizeof(uffdio_api)))
2092 goto out; 2017 goto out;
2093 2018
2094 /* only enable the requested features 2019 /* only enable the requested features for this uffd context */
2095 ctx_features = uffd_ctx_features(feat 2020 ctx_features = uffd_ctx_features(features);
2096 ret = -EINVAL; 2021 ret = -EINVAL;
2097 if (cmpxchg(&ctx->features, 0, ctx_fe 2022 if (cmpxchg(&ctx->features, 0, ctx_features) != 0)
2098 goto err_out; 2023 goto err_out;
2099 2024
2100 ret = 0; 2025 ret = 0;
2101 out: 2026 out:
2102 return ret; 2027 return ret;
2103 err_out: 2028 err_out:
2104 memset(&uffdio_api, 0, sizeof(uffdio_ 2029 memset(&uffdio_api, 0, sizeof(uffdio_api));
2105 if (copy_to_user(buf, &uffdio_api, si 2030 if (copy_to_user(buf, &uffdio_api, sizeof(uffdio_api)))
2106 ret = -EFAULT; 2031 ret = -EFAULT;
2107 goto out; 2032 goto out;
2108 } 2033 }
2109 2034
2110 static long userfaultfd_ioctl(struct file *fi 2035 static long userfaultfd_ioctl(struct file *file, unsigned cmd,
2111 unsigned long a 2036 unsigned long arg)
2112 { 2037 {
2113 int ret = -EINVAL; 2038 int ret = -EINVAL;
2114 struct userfaultfd_ctx *ctx = file->p 2039 struct userfaultfd_ctx *ctx = file->private_data;
2115 2040
2116 if (cmd != UFFDIO_API && !userfaultfd 2041 if (cmd != UFFDIO_API && !userfaultfd_is_initialized(ctx))
2117 return -EINVAL; 2042 return -EINVAL;
2118 2043
2119 switch(cmd) { 2044 switch(cmd) {
2120 case UFFDIO_API: 2045 case UFFDIO_API:
2121 ret = userfaultfd_api(ctx, ar 2046 ret = userfaultfd_api(ctx, arg);
2122 break; 2047 break;
2123 case UFFDIO_REGISTER: 2048 case UFFDIO_REGISTER:
2124 ret = userfaultfd_register(ct 2049 ret = userfaultfd_register(ctx, arg);
2125 break; 2050 break;
2126 case UFFDIO_UNREGISTER: 2051 case UFFDIO_UNREGISTER:
2127 ret = userfaultfd_unregister( 2052 ret = userfaultfd_unregister(ctx, arg);
2128 break; 2053 break;
2129 case UFFDIO_WAKE: 2054 case UFFDIO_WAKE:
2130 ret = userfaultfd_wake(ctx, a 2055 ret = userfaultfd_wake(ctx, arg);
2131 break; 2056 break;
2132 case UFFDIO_COPY: 2057 case UFFDIO_COPY:
2133 ret = userfaultfd_copy(ctx, a 2058 ret = userfaultfd_copy(ctx, arg);
2134 break; 2059 break;
2135 case UFFDIO_ZEROPAGE: 2060 case UFFDIO_ZEROPAGE:
2136 ret = userfaultfd_zeropage(ct 2061 ret = userfaultfd_zeropage(ctx, arg);
2137 break; 2062 break;
2138 case UFFDIO_MOVE: <<
2139 ret = userfaultfd_move(ctx, a <<
2140 break; <<
2141 case UFFDIO_WRITEPROTECT: 2063 case UFFDIO_WRITEPROTECT:
2142 ret = userfaultfd_writeprotec 2064 ret = userfaultfd_writeprotect(ctx, arg);
2143 break; 2065 break;
2144 case UFFDIO_CONTINUE: 2066 case UFFDIO_CONTINUE:
2145 ret = userfaultfd_continue(ct 2067 ret = userfaultfd_continue(ctx, arg);
2146 break; 2068 break;
2147 case UFFDIO_POISON: <<
2148 ret = userfaultfd_poison(ctx, <<
2149 break; <<
2150 } 2069 }
2151 return ret; 2070 return ret;
2152 } 2071 }
2153 2072
2154 #ifdef CONFIG_PROC_FS 2073 #ifdef CONFIG_PROC_FS
2155 static void userfaultfd_show_fdinfo(struct se 2074 static void userfaultfd_show_fdinfo(struct seq_file *m, struct file *f)
2156 { 2075 {
2157 struct userfaultfd_ctx *ctx = f->priv 2076 struct userfaultfd_ctx *ctx = f->private_data;
2158 wait_queue_entry_t *wq; 2077 wait_queue_entry_t *wq;
2159 unsigned long pending = 0, total = 0; 2078 unsigned long pending = 0, total = 0;
2160 2079
2161 spin_lock_irq(&ctx->fault_pending_wqh 2080 spin_lock_irq(&ctx->fault_pending_wqh.lock);
2162 list_for_each_entry(wq, &ctx->fault_p 2081 list_for_each_entry(wq, &ctx->fault_pending_wqh.head, entry) {
2163 pending++; 2082 pending++;
2164 total++; 2083 total++;
2165 } 2084 }
2166 list_for_each_entry(wq, &ctx->fault_w 2085 list_for_each_entry(wq, &ctx->fault_wqh.head, entry) {
2167 total++; 2086 total++;
2168 } 2087 }
2169 spin_unlock_irq(&ctx->fault_pending_w 2088 spin_unlock_irq(&ctx->fault_pending_wqh.lock);
2170 2089
2171 /* 2090 /*
2172 * If more protocols will be added, t 2091 * If more protocols will be added, there will be all shown
2173 * separated by a space. Like this: 2092 * separated by a space. Like this:
2174 * protocols: aa:... bb:... 2093 * protocols: aa:... bb:...
2175 */ 2094 */
2176 seq_printf(m, "pending:\t%lu\ntotal:\ 2095 seq_printf(m, "pending:\t%lu\ntotal:\t%lu\nAPI:\t%Lx:%x:%Lx\n",
2177 pending, total, UFFD_API, 2096 pending, total, UFFD_API, ctx->features,
2178 UFFD_API_IOCTLS|UFFD_API_R 2097 UFFD_API_IOCTLS|UFFD_API_RANGE_IOCTLS);
2179 } 2098 }
2180 #endif 2099 #endif
2181 2100
2182 static const struct file_operations userfault 2101 static const struct file_operations userfaultfd_fops = {
2183 #ifdef CONFIG_PROC_FS 2102 #ifdef CONFIG_PROC_FS
2184 .show_fdinfo = userfaultfd_show_fd 2103 .show_fdinfo = userfaultfd_show_fdinfo,
2185 #endif 2104 #endif
2186 .release = userfaultfd_release 2105 .release = userfaultfd_release,
2187 .poll = userfaultfd_poll, 2106 .poll = userfaultfd_poll,
2188 .read_iter = userfaultfd_read_it !! 2107 .read = userfaultfd_read,
2189 .unlocked_ioctl = userfaultfd_ioctl, 2108 .unlocked_ioctl = userfaultfd_ioctl,
2190 .compat_ioctl = compat_ptr_ioctl, 2109 .compat_ioctl = compat_ptr_ioctl,
2191 .llseek = noop_llseek, 2110 .llseek = noop_llseek,
2192 }; 2111 };
2193 2112
2194 static void init_once_userfaultfd_ctx(void *m 2113 static void init_once_userfaultfd_ctx(void *mem)
2195 { 2114 {
2196 struct userfaultfd_ctx *ctx = (struct 2115 struct userfaultfd_ctx *ctx = (struct userfaultfd_ctx *) mem;
2197 2116
2198 init_waitqueue_head(&ctx->fault_pendi 2117 init_waitqueue_head(&ctx->fault_pending_wqh);
2199 init_waitqueue_head(&ctx->fault_wqh); 2118 init_waitqueue_head(&ctx->fault_wqh);
2200 init_waitqueue_head(&ctx->event_wqh); 2119 init_waitqueue_head(&ctx->event_wqh);
2201 init_waitqueue_head(&ctx->fd_wqh); 2120 init_waitqueue_head(&ctx->fd_wqh);
2202 seqcount_spinlock_init(&ctx->refile_s 2121 seqcount_spinlock_init(&ctx->refile_seq, &ctx->fault_pending_wqh.lock);
2203 } 2122 }
2204 2123
2205 static int new_userfaultfd(int flags) 2124 static int new_userfaultfd(int flags)
2206 { 2125 {
2207 struct userfaultfd_ctx *ctx; 2126 struct userfaultfd_ctx *ctx;
2208 struct file *file; <<
2209 int fd; 2127 int fd;
2210 2128
2211 BUG_ON(!current->mm); 2129 BUG_ON(!current->mm);
2212 2130
2213 /* Check the UFFD_* constants for con 2131 /* Check the UFFD_* constants for consistency. */
2214 BUILD_BUG_ON(UFFD_USER_MODE_ONLY & UF 2132 BUILD_BUG_ON(UFFD_USER_MODE_ONLY & UFFD_SHARED_FCNTL_FLAGS);
2215 BUILD_BUG_ON(UFFD_CLOEXEC != O_CLOEXE 2133 BUILD_BUG_ON(UFFD_CLOEXEC != O_CLOEXEC);
2216 BUILD_BUG_ON(UFFD_NONBLOCK != O_NONBL 2134 BUILD_BUG_ON(UFFD_NONBLOCK != O_NONBLOCK);
2217 2135
2218 if (flags & ~(UFFD_SHARED_FCNTL_FLAGS 2136 if (flags & ~(UFFD_SHARED_FCNTL_FLAGS | UFFD_USER_MODE_ONLY))
2219 return -EINVAL; 2137 return -EINVAL;
2220 2138
2221 ctx = kmem_cache_alloc(userfaultfd_ct 2139 ctx = kmem_cache_alloc(userfaultfd_ctx_cachep, GFP_KERNEL);
2222 if (!ctx) 2140 if (!ctx)
2223 return -ENOMEM; 2141 return -ENOMEM;
2224 2142
2225 refcount_set(&ctx->refcount, 1); 2143 refcount_set(&ctx->refcount, 1);
2226 ctx->flags = flags; 2144 ctx->flags = flags;
2227 ctx->features = 0; 2145 ctx->features = 0;
2228 ctx->released = false; 2146 ctx->released = false;
2229 init_rwsem(&ctx->map_changing_lock); <<
2230 atomic_set(&ctx->mmap_changing, 0); 2147 atomic_set(&ctx->mmap_changing, 0);
2231 ctx->mm = current->mm; 2148 ctx->mm = current->mm;
>> 2149 /* prevent the mm struct to be freed */
>> 2150 mmgrab(ctx->mm);
2232 2151
2233 fd = get_unused_fd_flags(flags & UFFD !! 2152 fd = anon_inode_getfd_secure("[userfaultfd]", &userfaultfd_fops, ctx,
2234 if (fd < 0) <<
2235 goto err_out; <<
2236 <<
2237 /* Create a new inode so that the LSM <<
2238 file = anon_inode_create_getfile("[us <<
2239 O_RDONLY | (flags & U 2153 O_RDONLY | (flags & UFFD_SHARED_FCNTL_FLAGS), NULL);
2240 if (IS_ERR(file)) { !! 2154 if (fd < 0) {
2241 put_unused_fd(fd); !! 2155 mmdrop(ctx->mm);
2242 fd = PTR_ERR(file); !! 2156 kmem_cache_free(userfaultfd_ctx_cachep, ctx);
2243 goto err_out; <<
2244 } 2157 }
2245 /* prevent the mm struct to be freed <<
2246 mmgrab(ctx->mm); <<
2247 file->f_mode |= FMODE_NOWAIT; <<
2248 fd_install(fd, file); <<
2249 return fd; <<
2250 err_out: <<
2251 kmem_cache_free(userfaultfd_ctx_cache <<
2252 return fd; 2158 return fd;
2253 } 2159 }
2254 2160
2255 static inline bool userfaultfd_syscall_allowe 2161 static inline bool userfaultfd_syscall_allowed(int flags)
2256 { 2162 {
2257 /* Userspace-only page faults are alw 2163 /* Userspace-only page faults are always allowed */
2258 if (flags & UFFD_USER_MODE_ONLY) 2164 if (flags & UFFD_USER_MODE_ONLY)
2259 return true; 2165 return true;
2260 2166
2261 /* 2167 /*
2262 * The user is requesting a userfault 2168 * The user is requesting a userfaultfd which can handle kernel faults.
2263 * Privileged users are always allowe 2169 * Privileged users are always allowed to do this.
2264 */ 2170 */
2265 if (capable(CAP_SYS_PTRACE)) 2171 if (capable(CAP_SYS_PTRACE))
2266 return true; 2172 return true;
2267 2173
2268 /* Otherwise, access to kernel fault 2174 /* Otherwise, access to kernel fault handling is sysctl controlled. */
2269 return sysctl_unprivileged_userfaultf 2175 return sysctl_unprivileged_userfaultfd;
2270 } 2176 }
2271 2177
2272 SYSCALL_DEFINE1(userfaultfd, int, flags) 2178 SYSCALL_DEFINE1(userfaultfd, int, flags)
2273 { 2179 {
2274 if (!userfaultfd_syscall_allowed(flag 2180 if (!userfaultfd_syscall_allowed(flags))
2275 return -EPERM; 2181 return -EPERM;
2276 2182
2277 return new_userfaultfd(flags); 2183 return new_userfaultfd(flags);
2278 } 2184 }
2279 2185
2280 static long userfaultfd_dev_ioctl(struct file 2186 static long userfaultfd_dev_ioctl(struct file *file, unsigned int cmd, unsigned long flags)
2281 { 2187 {
2282 if (cmd != USERFAULTFD_IOC_NEW) 2188 if (cmd != USERFAULTFD_IOC_NEW)
2283 return -EINVAL; 2189 return -EINVAL;
2284 2190
2285 return new_userfaultfd(flags); 2191 return new_userfaultfd(flags);
2286 } 2192 }
2287 2193
2288 static const struct file_operations userfault 2194 static const struct file_operations userfaultfd_dev_fops = {
2289 .unlocked_ioctl = userfaultfd_dev_ioc 2195 .unlocked_ioctl = userfaultfd_dev_ioctl,
2290 .compat_ioctl = userfaultfd_dev_ioctl 2196 .compat_ioctl = userfaultfd_dev_ioctl,
2291 .owner = THIS_MODULE, 2197 .owner = THIS_MODULE,
2292 .llseek = noop_llseek, 2198 .llseek = noop_llseek,
2293 }; 2199 };
2294 2200
2295 static struct miscdevice userfaultfd_misc = { 2201 static struct miscdevice userfaultfd_misc = {
2296 .minor = MISC_DYNAMIC_MINOR, 2202 .minor = MISC_DYNAMIC_MINOR,
2297 .name = "userfaultfd", 2203 .name = "userfaultfd",
2298 .fops = &userfaultfd_dev_fops 2204 .fops = &userfaultfd_dev_fops
2299 }; 2205 };
2300 2206
2301 static int __init userfaultfd_init(void) 2207 static int __init userfaultfd_init(void)
2302 { 2208 {
2303 int ret; 2209 int ret;
2304 2210
2305 ret = misc_register(&userfaultfd_misc 2211 ret = misc_register(&userfaultfd_misc);
2306 if (ret) 2212 if (ret)
2307 return ret; 2213 return ret;
2308 2214
2309 userfaultfd_ctx_cachep = kmem_cache_c 2215 userfaultfd_ctx_cachep = kmem_cache_create("userfaultfd_ctx_cache",
2310 2216 sizeof(struct userfaultfd_ctx),
2311 2217 0,
2312 2218 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2313 2219 init_once_userfaultfd_ctx);
2314 #ifdef CONFIG_SYSCTL 2220 #ifdef CONFIG_SYSCTL
2315 register_sysctl_init("vm", vm_userfau 2221 register_sysctl_init("vm", vm_userfaultfd_table);
2316 #endif 2222 #endif
2317 return 0; 2223 return 0;
2318 } 2224 }
2319 __initcall(userfaultfd_init); 2225 __initcall(userfaultfd_init);
2320 2226
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.