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