Commit | Line | Data |
---|---|---|
20c8ccb1 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
86039bd3 AA |
2 | /* |
3 | * fs/userfaultfd.c | |
4 | * | |
5 | * Copyright (C) 2007 Davide Libenzi <davidel@xmailserver.org> | |
6 | * Copyright (C) 2008-2009 Red Hat, Inc. | |
7 | * Copyright (C) 2015 Red Hat, Inc. | |
8 | * | |
86039bd3 AA |
9 | * Some part derived from fs/eventfd.c (anon inode setup) and |
10 | * mm/ksm.c (mm hashing). | |
11 | */ | |
12 | ||
9cd75c3c | 13 | #include <linux/list.h> |
86039bd3 | 14 | #include <linux/hashtable.h> |
174cd4b1 | 15 | #include <linux/sched/signal.h> |
6e84f315 | 16 | #include <linux/sched/mm.h> |
86039bd3 | 17 | #include <linux/mm.h> |
6dfeaff9 | 18 | #include <linux/mmu_notifier.h> |
86039bd3 AA |
19 | #include <linux/poll.h> |
20 | #include <linux/slab.h> | |
21 | #include <linux/seq_file.h> | |
22 | #include <linux/file.h> | |
23 | #include <linux/bug.h> | |
24 | #include <linux/anon_inodes.h> | |
25 | #include <linux/syscalls.h> | |
26 | #include <linux/userfaultfd_k.h> | |
27 | #include <linux/mempolicy.h> | |
28 | #include <linux/ioctl.h> | |
29 | #include <linux/security.h> | |
cab350af | 30 | #include <linux/hugetlb.h> |
86039bd3 | 31 | |
d0d4730a | 32 | int sysctl_unprivileged_userfaultfd __read_mostly; |
cefdca0a | 33 | |
3004ec9c AA |
34 | static struct kmem_cache *userfaultfd_ctx_cachep __read_mostly; |
35 | ||
86039bd3 AA |
36 | enum userfaultfd_state { |
37 | UFFD_STATE_WAIT_API, | |
38 | UFFD_STATE_RUNNING, | |
39 | }; | |
40 | ||
3004ec9c AA |
41 | /* |
42 | * Start with fault_pending_wqh and fault_wqh so they're more likely | |
43 | * to be in the same cacheline. | |
cbcfa130 EB |
44 | * |
45 | * Locking order: | |
46 | * fd_wqh.lock | |
47 | * fault_pending_wqh.lock | |
48 | * fault_wqh.lock | |
49 | * event_wqh.lock | |
50 | * | |
51 | * To avoid deadlocks, IRQs must be disabled when taking any of the above locks, | |
52 | * since fd_wqh.lock is taken by aio_poll() while it's holding a lock that's | |
53 | * also taken in IRQ context. | |
3004ec9c | 54 | */ |
86039bd3 | 55 | struct userfaultfd_ctx { |
15b726ef AA |
56 | /* waitqueue head for the pending (i.e. not read) userfaults */ |
57 | wait_queue_head_t fault_pending_wqh; | |
58 | /* waitqueue head for the userfaults */ | |
86039bd3 AA |
59 | wait_queue_head_t fault_wqh; |
60 | /* waitqueue head for the pseudo fd to wakeup poll/read */ | |
61 | wait_queue_head_t fd_wqh; | |
9cd75c3c PE |
62 | /* waitqueue head for events */ |
63 | wait_queue_head_t event_wqh; | |
2c5b7e1b | 64 | /* a refile sequence protected by fault_pending_wqh lock */ |
2ca97ac8 | 65 | seqcount_spinlock_t refile_seq; |
3004ec9c | 66 | /* pseudo fd refcounting */ |
ca880420 | 67 | refcount_t refcount; |
86039bd3 AA |
68 | /* userfaultfd syscall flags */ |
69 | unsigned int flags; | |
9cd75c3c PE |
70 | /* features requested from the userspace */ |
71 | unsigned int features; | |
86039bd3 AA |
72 | /* state machine */ |
73 | enum userfaultfd_state state; | |
74 | /* released */ | |
75 | bool released; | |
df2cc96e MR |
76 | /* memory mappings are changing because of non-cooperative event */ |
77 | bool mmap_changing; | |
86039bd3 AA |
78 | /* mm with one ore more vmas attached to this userfaultfd_ctx */ |
79 | struct mm_struct *mm; | |
80 | }; | |
81 | ||
893e26e6 PE |
82 | struct userfaultfd_fork_ctx { |
83 | struct userfaultfd_ctx *orig; | |
84 | struct userfaultfd_ctx *new; | |
85 | struct list_head list; | |
86 | }; | |
87 | ||
897ab3e0 MR |
88 | struct userfaultfd_unmap_ctx { |
89 | struct userfaultfd_ctx *ctx; | |
90 | unsigned long start; | |
91 | unsigned long end; | |
92 | struct list_head list; | |
93 | }; | |
94 | ||
86039bd3 | 95 | struct userfaultfd_wait_queue { |
a9b85f94 | 96 | struct uffd_msg msg; |
ac6424b9 | 97 | wait_queue_entry_t wq; |
86039bd3 | 98 | struct userfaultfd_ctx *ctx; |
15a77c6f | 99 | bool waken; |
86039bd3 AA |
100 | }; |
101 | ||
102 | struct userfaultfd_wake_range { | |
103 | unsigned long start; | |
104 | unsigned long len; | |
105 | }; | |
106 | ||
ac6424b9 | 107 | static int userfaultfd_wake_function(wait_queue_entry_t *wq, unsigned mode, |
86039bd3 AA |
108 | int wake_flags, void *key) |
109 | { | |
110 | struct userfaultfd_wake_range *range = key; | |
111 | int ret; | |
112 | struct userfaultfd_wait_queue *uwq; | |
113 | unsigned long start, len; | |
114 | ||
115 | uwq = container_of(wq, struct userfaultfd_wait_queue, wq); | |
116 | ret = 0; | |
86039bd3 AA |
117 | /* len == 0 means wake all */ |
118 | start = range->start; | |
119 | len = range->len; | |
a9b85f94 AA |
120 | if (len && (start > uwq->msg.arg.pagefault.address || |
121 | start + len <= uwq->msg.arg.pagefault.address)) | |
86039bd3 | 122 | goto out; |
15a77c6f AA |
123 | WRITE_ONCE(uwq->waken, true); |
124 | /* | |
a9668cd6 PZ |
125 | * The Program-Order guarantees provided by the scheduler |
126 | * ensure uwq->waken is visible before the task is woken. | |
15a77c6f | 127 | */ |
86039bd3 | 128 | ret = wake_up_state(wq->private, mode); |
a9668cd6 | 129 | if (ret) { |
86039bd3 AA |
130 | /* |
131 | * Wake only once, autoremove behavior. | |
132 | * | |
a9668cd6 PZ |
133 | * After the effect of list_del_init is visible to the other |
134 | * CPUs, the waitqueue may disappear from under us, see the | |
135 | * !list_empty_careful() in handle_userfault(). | |
136 | * | |
137 | * try_to_wake_up() has an implicit smp_mb(), and the | |
138 | * wq->private is read before calling the extern function | |
139 | * "wake_up_state" (which in turns calls try_to_wake_up). | |
86039bd3 | 140 | */ |
2055da97 | 141 | list_del_init(&wq->entry); |
a9668cd6 | 142 | } |
86039bd3 AA |
143 | out: |
144 | return ret; | |
145 | } | |
146 | ||
147 | /** | |
148 | * userfaultfd_ctx_get - Acquires a reference to the internal userfaultfd | |
149 | * context. | |
150 | * @ctx: [in] Pointer to the userfaultfd context. | |
86039bd3 AA |
151 | */ |
152 | static void userfaultfd_ctx_get(struct userfaultfd_ctx *ctx) | |
153 | { | |
ca880420 | 154 | refcount_inc(&ctx->refcount); |
86039bd3 AA |
155 | } |
156 | ||
157 | /** | |
158 | * userfaultfd_ctx_put - Releases a reference to the internal userfaultfd | |
159 | * context. | |
160 | * @ctx: [in] Pointer to userfaultfd context. | |
161 | * | |
162 | * The userfaultfd context reference must have been previously acquired either | |
163 | * with userfaultfd_ctx_get() or userfaultfd_ctx_fdget(). | |
164 | */ | |
165 | static void userfaultfd_ctx_put(struct userfaultfd_ctx *ctx) | |
166 | { | |
ca880420 | 167 | if (refcount_dec_and_test(&ctx->refcount)) { |
86039bd3 AA |
168 | VM_BUG_ON(spin_is_locked(&ctx->fault_pending_wqh.lock)); |
169 | VM_BUG_ON(waitqueue_active(&ctx->fault_pending_wqh)); | |
170 | VM_BUG_ON(spin_is_locked(&ctx->fault_wqh.lock)); | |
171 | VM_BUG_ON(waitqueue_active(&ctx->fault_wqh)); | |
9cd75c3c PE |
172 | VM_BUG_ON(spin_is_locked(&ctx->event_wqh.lock)); |
173 | VM_BUG_ON(waitqueue_active(&ctx->event_wqh)); | |
86039bd3 AA |
174 | VM_BUG_ON(spin_is_locked(&ctx->fd_wqh.lock)); |
175 | VM_BUG_ON(waitqueue_active(&ctx->fd_wqh)); | |
d2005e3f | 176 | mmdrop(ctx->mm); |
3004ec9c | 177 | kmem_cache_free(userfaultfd_ctx_cachep, ctx); |
86039bd3 AA |
178 | } |
179 | } | |
180 | ||
a9b85f94 | 181 | static inline void msg_init(struct uffd_msg *msg) |
86039bd3 | 182 | { |
a9b85f94 AA |
183 | BUILD_BUG_ON(sizeof(struct uffd_msg) != 32); |
184 | /* | |
185 | * Must use memset to zero out the paddings or kernel data is | |
186 | * leaked to userland. | |
187 | */ | |
188 | memset(msg, 0, sizeof(struct uffd_msg)); | |
189 | } | |
190 | ||
191 | static inline struct uffd_msg userfault_msg(unsigned long address, | |
192 | unsigned int flags, | |
9d4ac934 AP |
193 | unsigned long reason, |
194 | unsigned int features) | |
a9b85f94 AA |
195 | { |
196 | struct uffd_msg msg; | |
197 | msg_init(&msg); | |
198 | msg.event = UFFD_EVENT_PAGEFAULT; | |
199 | msg.arg.pagefault.address = address; | |
7677f7fd AR |
200 | /* |
201 | * These flags indicate why the userfault occurred: | |
202 | * - UFFD_PAGEFAULT_FLAG_WP indicates a write protect fault. | |
203 | * - UFFD_PAGEFAULT_FLAG_MINOR indicates a minor fault. | |
204 | * - Neither of these flags being set indicates a MISSING fault. | |
205 | * | |
206 | * Separately, UFFD_PAGEFAULT_FLAG_WRITE indicates it was a write | |
207 | * fault. Otherwise, it was a read fault. | |
208 | */ | |
86039bd3 | 209 | if (flags & FAULT_FLAG_WRITE) |
a9b85f94 | 210 | msg.arg.pagefault.flags |= UFFD_PAGEFAULT_FLAG_WRITE; |
86039bd3 | 211 | if (reason & VM_UFFD_WP) |
a9b85f94 | 212 | msg.arg.pagefault.flags |= UFFD_PAGEFAULT_FLAG_WP; |
7677f7fd AR |
213 | if (reason & VM_UFFD_MINOR) |
214 | msg.arg.pagefault.flags |= UFFD_PAGEFAULT_FLAG_MINOR; | |
9d4ac934 | 215 | if (features & UFFD_FEATURE_THREAD_ID) |
a36985d3 | 216 | msg.arg.pagefault.feat.ptid = task_pid_vnr(current); |
a9b85f94 | 217 | return msg; |
86039bd3 AA |
218 | } |
219 | ||
369cd212 MK |
220 | #ifdef CONFIG_HUGETLB_PAGE |
221 | /* | |
222 | * Same functionality as userfaultfd_must_wait below with modifications for | |
223 | * hugepmd ranges. | |
224 | */ | |
225 | static inline bool userfaultfd_huge_must_wait(struct userfaultfd_ctx *ctx, | |
7868a208 | 226 | struct vm_area_struct *vma, |
369cd212 MK |
227 | unsigned long address, |
228 | unsigned long flags, | |
229 | unsigned long reason) | |
230 | { | |
231 | struct mm_struct *mm = ctx->mm; | |
1e2c0436 | 232 | pte_t *ptep, pte; |
369cd212 MK |
233 | bool ret = true; |
234 | ||
42fc5414 | 235 | mmap_assert_locked(mm); |
369cd212 | 236 | |
1e2c0436 JF |
237 | ptep = huge_pte_offset(mm, address, vma_mmu_pagesize(vma)); |
238 | ||
239 | if (!ptep) | |
369cd212 MK |
240 | goto out; |
241 | ||
242 | ret = false; | |
1e2c0436 | 243 | pte = huge_ptep_get(ptep); |
369cd212 MK |
244 | |
245 | /* | |
246 | * Lockless access: we're in a wait_event so it's ok if it | |
247 | * changes under us. | |
248 | */ | |
1e2c0436 | 249 | if (huge_pte_none(pte)) |
369cd212 | 250 | ret = true; |
1e2c0436 | 251 | if (!huge_pte_write(pte) && (reason & VM_UFFD_WP)) |
369cd212 MK |
252 | ret = true; |
253 | out: | |
254 | return ret; | |
255 | } | |
256 | #else | |
257 | static inline bool userfaultfd_huge_must_wait(struct userfaultfd_ctx *ctx, | |
7868a208 | 258 | struct vm_area_struct *vma, |
369cd212 MK |
259 | unsigned long address, |
260 | unsigned long flags, | |
261 | unsigned long reason) | |
262 | { | |
263 | return false; /* should never get here */ | |
264 | } | |
265 | #endif /* CONFIG_HUGETLB_PAGE */ | |
266 | ||
8d2afd96 AA |
267 | /* |
268 | * Verify the pagetables are still not ok after having reigstered into | |
269 | * the fault_pending_wqh to avoid userland having to UFFDIO_WAKE any | |
270 | * userfault that has already been resolved, if userfaultfd_read and | |
271 | * UFFDIO_COPY|ZEROPAGE are being run simultaneously on two different | |
272 | * threads. | |
273 | */ | |
274 | static inline bool userfaultfd_must_wait(struct userfaultfd_ctx *ctx, | |
275 | unsigned long address, | |
276 | unsigned long flags, | |
277 | unsigned long reason) | |
278 | { | |
279 | struct mm_struct *mm = ctx->mm; | |
280 | pgd_t *pgd; | |
c2febafc | 281 | p4d_t *p4d; |
8d2afd96 AA |
282 | pud_t *pud; |
283 | pmd_t *pmd, _pmd; | |
284 | pte_t *pte; | |
285 | bool ret = true; | |
286 | ||
42fc5414 | 287 | mmap_assert_locked(mm); |
8d2afd96 AA |
288 | |
289 | pgd = pgd_offset(mm, address); | |
290 | if (!pgd_present(*pgd)) | |
291 | goto out; | |
c2febafc KS |
292 | p4d = p4d_offset(pgd, address); |
293 | if (!p4d_present(*p4d)) | |
294 | goto out; | |
295 | pud = pud_offset(p4d, address); | |
8d2afd96 AA |
296 | if (!pud_present(*pud)) |
297 | goto out; | |
298 | pmd = pmd_offset(pud, address); | |
299 | /* | |
300 | * READ_ONCE must function as a barrier with narrower scope | |
301 | * and it must be equivalent to: | |
302 | * _pmd = *pmd; barrier(); | |
303 | * | |
304 | * This is to deal with the instability (as in | |
305 | * pmd_trans_unstable) of the pmd. | |
306 | */ | |
307 | _pmd = READ_ONCE(*pmd); | |
a365ac09 | 308 | if (pmd_none(_pmd)) |
8d2afd96 AA |
309 | goto out; |
310 | ||
311 | ret = false; | |
a365ac09 HY |
312 | if (!pmd_present(_pmd)) |
313 | goto out; | |
314 | ||
63b2d417 AA |
315 | if (pmd_trans_huge(_pmd)) { |
316 | if (!pmd_write(_pmd) && (reason & VM_UFFD_WP)) | |
317 | ret = true; | |
8d2afd96 | 318 | goto out; |
63b2d417 | 319 | } |
8d2afd96 AA |
320 | |
321 | /* | |
322 | * the pmd is stable (as in !pmd_trans_unstable) so we can re-read it | |
323 | * and use the standard pte_offset_map() instead of parsing _pmd. | |
324 | */ | |
325 | pte = pte_offset_map(pmd, address); | |
326 | /* | |
327 | * Lockless access: we're in a wait_event so it's ok if it | |
328 | * changes under us. | |
329 | */ | |
330 | if (pte_none(*pte)) | |
331 | ret = true; | |
63b2d417 AA |
332 | if (!pte_write(*pte) && (reason & VM_UFFD_WP)) |
333 | ret = true; | |
8d2afd96 AA |
334 | pte_unmap(pte); |
335 | ||
336 | out: | |
337 | return ret; | |
338 | } | |
339 | ||
3e69ad08 PX |
340 | static inline long userfaultfd_get_blocking_state(unsigned int flags) |
341 | { | |
342 | if (flags & FAULT_FLAG_INTERRUPTIBLE) | |
343 | return TASK_INTERRUPTIBLE; | |
344 | ||
345 | if (flags & FAULT_FLAG_KILLABLE) | |
346 | return TASK_KILLABLE; | |
347 | ||
348 | return TASK_UNINTERRUPTIBLE; | |
349 | } | |
350 | ||
86039bd3 AA |
351 | /* |
352 | * The locking rules involved in returning VM_FAULT_RETRY depending on | |
353 | * FAULT_FLAG_ALLOW_RETRY, FAULT_FLAG_RETRY_NOWAIT and | |
354 | * FAULT_FLAG_KILLABLE are not straightforward. The "Caution" | |
355 | * recommendation in __lock_page_or_retry is not an understatement. | |
356 | * | |
c1e8d7c6 | 357 | * If FAULT_FLAG_ALLOW_RETRY is set, the mmap_lock must be released |
86039bd3 AA |
358 | * before returning VM_FAULT_RETRY only if FAULT_FLAG_RETRY_NOWAIT is |
359 | * not set. | |
360 | * | |
361 | * If FAULT_FLAG_ALLOW_RETRY is set but FAULT_FLAG_KILLABLE is not | |
362 | * set, VM_FAULT_RETRY can still be returned if and only if there are | |
c1e8d7c6 | 363 | * fatal_signal_pending()s, and the mmap_lock must be released before |
86039bd3 AA |
364 | * returning it. |
365 | */ | |
2b740303 | 366 | vm_fault_t handle_userfault(struct vm_fault *vmf, unsigned long reason) |
86039bd3 | 367 | { |
82b0f8c3 | 368 | struct mm_struct *mm = vmf->vma->vm_mm; |
86039bd3 AA |
369 | struct userfaultfd_ctx *ctx; |
370 | struct userfaultfd_wait_queue uwq; | |
2b740303 | 371 | vm_fault_t ret = VM_FAULT_SIGBUS; |
3e69ad08 | 372 | bool must_wait; |
15a77c6f | 373 | long blocking_state; |
86039bd3 | 374 | |
64c2b203 AA |
375 | /* |
376 | * We don't do userfault handling for the final child pid update. | |
377 | * | |
378 | * We also don't do userfault handling during | |
379 | * coredumping. hugetlbfs has the special | |
380 | * follow_hugetlb_page() to skip missing pages in the | |
381 | * FOLL_DUMP case, anon memory also checks for FOLL_DUMP with | |
382 | * the no_page_table() helper in follow_page_mask(), but the | |
383 | * shmem_vm_ops->fault method is invoked even during | |
c1e8d7c6 | 384 | * coredumping without mmap_lock and it ends up here. |
64c2b203 AA |
385 | */ |
386 | if (current->flags & (PF_EXITING|PF_DUMPCORE)) | |
387 | goto out; | |
388 | ||
389 | /* | |
c1e8d7c6 ML |
390 | * Coredumping runs without mmap_lock so we can only check that |
391 | * the mmap_lock is held, if PF_DUMPCORE was not set. | |
64c2b203 | 392 | */ |
42fc5414 | 393 | mmap_assert_locked(mm); |
64c2b203 | 394 | |
82b0f8c3 | 395 | ctx = vmf->vma->vm_userfaultfd_ctx.ctx; |
86039bd3 | 396 | if (!ctx) |
ba85c702 | 397 | goto out; |
86039bd3 AA |
398 | |
399 | BUG_ON(ctx->mm != mm); | |
400 | ||
7677f7fd AR |
401 | /* Any unrecognized flag is a bug. */ |
402 | VM_BUG_ON(reason & ~__VM_UFFD_FLAGS); | |
403 | /* 0 or > 1 flags set is a bug; we expect exactly 1. */ | |
404 | VM_BUG_ON(!reason || (reason & (reason - 1))); | |
86039bd3 | 405 | |
2d6d6f5a PS |
406 | if (ctx->features & UFFD_FEATURE_SIGBUS) |
407 | goto out; | |
37cd0575 LG |
408 | if ((vmf->flags & FAULT_FLAG_USER) == 0 && |
409 | ctx->flags & UFFD_USER_MODE_ONLY) { | |
410 | printk_once(KERN_WARNING "uffd: Set unprivileged_userfaultfd " | |
411 | "sysctl knob to 1 if kernel faults must be handled " | |
412 | "without obtaining CAP_SYS_PTRACE capability\n"); | |
413 | goto out; | |
414 | } | |
2d6d6f5a | 415 | |
86039bd3 AA |
416 | /* |
417 | * If it's already released don't get it. This avoids to loop | |
418 | * in __get_user_pages if userfaultfd_release waits on the | |
c1e8d7c6 | 419 | * caller of handle_userfault to release the mmap_lock. |
86039bd3 | 420 | */ |
6aa7de05 | 421 | if (unlikely(READ_ONCE(ctx->released))) { |
656710a6 AA |
422 | /* |
423 | * Don't return VM_FAULT_SIGBUS in this case, so a non | |
424 | * cooperative manager can close the uffd after the | |
425 | * last UFFDIO_COPY, without risking to trigger an | |
426 | * involuntary SIGBUS if the process was starting the | |
427 | * userfaultfd while the userfaultfd was still armed | |
428 | * (but after the last UFFDIO_COPY). If the uffd | |
429 | * wasn't already closed when the userfault reached | |
430 | * this point, that would normally be solved by | |
431 | * userfaultfd_must_wait returning 'false'. | |
432 | * | |
433 | * If we were to return VM_FAULT_SIGBUS here, the non | |
434 | * cooperative manager would be instead forced to | |
435 | * always call UFFDIO_UNREGISTER before it can safely | |
436 | * close the uffd. | |
437 | */ | |
438 | ret = VM_FAULT_NOPAGE; | |
ba85c702 | 439 | goto out; |
656710a6 | 440 | } |
86039bd3 AA |
441 | |
442 | /* | |
443 | * Check that we can return VM_FAULT_RETRY. | |
444 | * | |
445 | * NOTE: it should become possible to return VM_FAULT_RETRY | |
446 | * even if FAULT_FLAG_TRIED is set without leading to gup() | |
447 | * -EBUSY failures, if the userfaultfd is to be extended for | |
448 | * VM_UFFD_WP tracking and we intend to arm the userfault | |
449 | * without first stopping userland access to the memory. For | |
450 | * VM_UFFD_MISSING userfaults this is enough for now. | |
451 | */ | |
82b0f8c3 | 452 | if (unlikely(!(vmf->flags & FAULT_FLAG_ALLOW_RETRY))) { |
86039bd3 AA |
453 | /* |
454 | * Validate the invariant that nowait must allow retry | |
455 | * to be sure not to return SIGBUS erroneously on | |
456 | * nowait invocations. | |
457 | */ | |
82b0f8c3 | 458 | BUG_ON(vmf->flags & FAULT_FLAG_RETRY_NOWAIT); |
86039bd3 AA |
459 | #ifdef CONFIG_DEBUG_VM |
460 | if (printk_ratelimit()) { | |
461 | printk(KERN_WARNING | |
82b0f8c3 JK |
462 | "FAULT_FLAG_ALLOW_RETRY missing %x\n", |
463 | vmf->flags); | |
86039bd3 AA |
464 | dump_stack(); |
465 | } | |
466 | #endif | |
ba85c702 | 467 | goto out; |
86039bd3 AA |
468 | } |
469 | ||
470 | /* | |
471 | * Handle nowait, not much to do other than tell it to retry | |
472 | * and wait. | |
473 | */ | |
ba85c702 | 474 | ret = VM_FAULT_RETRY; |
82b0f8c3 | 475 | if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT) |
ba85c702 | 476 | goto out; |
86039bd3 | 477 | |
c1e8d7c6 | 478 | /* take the reference before dropping the mmap_lock */ |
86039bd3 AA |
479 | userfaultfd_ctx_get(ctx); |
480 | ||
86039bd3 AA |
481 | init_waitqueue_func_entry(&uwq.wq, userfaultfd_wake_function); |
482 | uwq.wq.private = current; | |
9d4ac934 AP |
483 | uwq.msg = userfault_msg(vmf->address, vmf->flags, reason, |
484 | ctx->features); | |
86039bd3 | 485 | uwq.ctx = ctx; |
15a77c6f | 486 | uwq.waken = false; |
86039bd3 | 487 | |
3e69ad08 | 488 | blocking_state = userfaultfd_get_blocking_state(vmf->flags); |
dfa37dc3 | 489 | |
cbcfa130 | 490 | spin_lock_irq(&ctx->fault_pending_wqh.lock); |
86039bd3 AA |
491 | /* |
492 | * After the __add_wait_queue the uwq is visible to userland | |
493 | * through poll/read(). | |
494 | */ | |
15b726ef AA |
495 | __add_wait_queue(&ctx->fault_pending_wqh, &uwq.wq); |
496 | /* | |
497 | * The smp_mb() after __set_current_state prevents the reads | |
498 | * following the spin_unlock to happen before the list_add in | |
499 | * __add_wait_queue. | |
500 | */ | |
15a77c6f | 501 | set_current_state(blocking_state); |
cbcfa130 | 502 | spin_unlock_irq(&ctx->fault_pending_wqh.lock); |
86039bd3 | 503 | |
369cd212 MK |
504 | if (!is_vm_hugetlb_page(vmf->vma)) |
505 | must_wait = userfaultfd_must_wait(ctx, vmf->address, vmf->flags, | |
506 | reason); | |
507 | else | |
7868a208 PA |
508 | must_wait = userfaultfd_huge_must_wait(ctx, vmf->vma, |
509 | vmf->address, | |
369cd212 | 510 | vmf->flags, reason); |
d8ed45c5 | 511 | mmap_read_unlock(mm); |
8d2afd96 | 512 | |
f9bf3522 | 513 | if (likely(must_wait && !READ_ONCE(ctx->released))) { |
a9a08845 | 514 | wake_up_poll(&ctx->fd_wqh, EPOLLIN); |
86039bd3 | 515 | schedule(); |
ba85c702 | 516 | } |
86039bd3 | 517 | |
ba85c702 | 518 | __set_current_state(TASK_RUNNING); |
15b726ef AA |
519 | |
520 | /* | |
521 | * Here we race with the list_del; list_add in | |
522 | * userfaultfd_ctx_read(), however because we don't ever run | |
523 | * list_del_init() to refile across the two lists, the prev | |
524 | * and next pointers will never point to self. list_add also | |
525 | * would never let any of the two pointers to point to | |
526 | * self. So list_empty_careful won't risk to see both pointers | |
527 | * pointing to self at any time during the list refile. The | |
528 | * only case where list_del_init() is called is the full | |
529 | * removal in the wake function and there we don't re-list_add | |
530 | * and it's fine not to block on the spinlock. The uwq on this | |
531 | * kernel stack can be released after the list_del_init. | |
532 | */ | |
2055da97 | 533 | if (!list_empty_careful(&uwq.wq.entry)) { |
cbcfa130 | 534 | spin_lock_irq(&ctx->fault_pending_wqh.lock); |
15b726ef AA |
535 | /* |
536 | * No need of list_del_init(), the uwq on the stack | |
537 | * will be freed shortly anyway. | |
538 | */ | |
2055da97 | 539 | list_del(&uwq.wq.entry); |
cbcfa130 | 540 | spin_unlock_irq(&ctx->fault_pending_wqh.lock); |
86039bd3 | 541 | } |
86039bd3 AA |
542 | |
543 | /* | |
544 | * ctx may go away after this if the userfault pseudo fd is | |
545 | * already released. | |
546 | */ | |
547 | userfaultfd_ctx_put(ctx); | |
548 | ||
ba85c702 AA |
549 | out: |
550 | return ret; | |
86039bd3 AA |
551 | } |
552 | ||
8c9e7bb7 AA |
553 | static void userfaultfd_event_wait_completion(struct userfaultfd_ctx *ctx, |
554 | struct userfaultfd_wait_queue *ewq) | |
9cd75c3c | 555 | { |
0cbb4b4f AA |
556 | struct userfaultfd_ctx *release_new_ctx; |
557 | ||
9a69a829 AA |
558 | if (WARN_ON_ONCE(current->flags & PF_EXITING)) |
559 | goto out; | |
9cd75c3c PE |
560 | |
561 | ewq->ctx = ctx; | |
562 | init_waitqueue_entry(&ewq->wq, current); | |
0cbb4b4f | 563 | release_new_ctx = NULL; |
9cd75c3c | 564 | |
cbcfa130 | 565 | spin_lock_irq(&ctx->event_wqh.lock); |
9cd75c3c PE |
566 | /* |
567 | * After the __add_wait_queue the uwq is visible to userland | |
568 | * through poll/read(). | |
569 | */ | |
570 | __add_wait_queue(&ctx->event_wqh, &ewq->wq); | |
571 | for (;;) { | |
572 | set_current_state(TASK_KILLABLE); | |
573 | if (ewq->msg.event == 0) | |
574 | break; | |
6aa7de05 | 575 | if (READ_ONCE(ctx->released) || |
9cd75c3c | 576 | fatal_signal_pending(current)) { |
384632e6 AA |
577 | /* |
578 | * &ewq->wq may be queued in fork_event, but | |
579 | * __remove_wait_queue ignores the head | |
580 | * parameter. It would be a problem if it | |
581 | * didn't. | |
582 | */ | |
9cd75c3c | 583 | __remove_wait_queue(&ctx->event_wqh, &ewq->wq); |
7eb76d45 MR |
584 | if (ewq->msg.event == UFFD_EVENT_FORK) { |
585 | struct userfaultfd_ctx *new; | |
586 | ||
587 | new = (struct userfaultfd_ctx *) | |
588 | (unsigned long) | |
589 | ewq->msg.arg.reserved.reserved1; | |
0cbb4b4f | 590 | release_new_ctx = new; |
7eb76d45 | 591 | } |
9cd75c3c PE |
592 | break; |
593 | } | |
594 | ||
cbcfa130 | 595 | spin_unlock_irq(&ctx->event_wqh.lock); |
9cd75c3c | 596 | |
a9a08845 | 597 | wake_up_poll(&ctx->fd_wqh, EPOLLIN); |
9cd75c3c PE |
598 | schedule(); |
599 | ||
cbcfa130 | 600 | spin_lock_irq(&ctx->event_wqh.lock); |
9cd75c3c PE |
601 | } |
602 | __set_current_state(TASK_RUNNING); | |
cbcfa130 | 603 | spin_unlock_irq(&ctx->event_wqh.lock); |
9cd75c3c | 604 | |
0cbb4b4f AA |
605 | if (release_new_ctx) { |
606 | struct vm_area_struct *vma; | |
607 | struct mm_struct *mm = release_new_ctx->mm; | |
608 | ||
609 | /* the various vma->vm_userfaultfd_ctx still points to it */ | |
d8ed45c5 | 610 | mmap_write_lock(mm); |
0cbb4b4f | 611 | for (vma = mm->mmap; vma; vma = vma->vm_next) |
31e810aa | 612 | if (vma->vm_userfaultfd_ctx.ctx == release_new_ctx) { |
0cbb4b4f | 613 | vma->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX; |
7677f7fd | 614 | vma->vm_flags &= ~__VM_UFFD_FLAGS; |
31e810aa | 615 | } |
d8ed45c5 | 616 | mmap_write_unlock(mm); |
0cbb4b4f AA |
617 | |
618 | userfaultfd_ctx_put(release_new_ctx); | |
619 | } | |
620 | ||
9cd75c3c PE |
621 | /* |
622 | * ctx may go away after this if the userfault pseudo fd is | |
623 | * already released. | |
624 | */ | |
9a69a829 | 625 | out: |
df2cc96e | 626 | WRITE_ONCE(ctx->mmap_changing, false); |
9cd75c3c | 627 | userfaultfd_ctx_put(ctx); |
9cd75c3c PE |
628 | } |
629 | ||
630 | static void userfaultfd_event_complete(struct userfaultfd_ctx *ctx, | |
631 | struct userfaultfd_wait_queue *ewq) | |
632 | { | |
633 | ewq->msg.event = 0; | |
634 | wake_up_locked(&ctx->event_wqh); | |
635 | __remove_wait_queue(&ctx->event_wqh, &ewq->wq); | |
636 | } | |
637 | ||
893e26e6 PE |
638 | int dup_userfaultfd(struct vm_area_struct *vma, struct list_head *fcs) |
639 | { | |
640 | struct userfaultfd_ctx *ctx = NULL, *octx; | |
641 | struct userfaultfd_fork_ctx *fctx; | |
642 | ||
643 | octx = vma->vm_userfaultfd_ctx.ctx; | |
644 | if (!octx || !(octx->features & UFFD_FEATURE_EVENT_FORK)) { | |
645 | vma->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX; | |
7677f7fd | 646 | vma->vm_flags &= ~__VM_UFFD_FLAGS; |
893e26e6 PE |
647 | return 0; |
648 | } | |
649 | ||
650 | list_for_each_entry(fctx, fcs, list) | |
651 | if (fctx->orig == octx) { | |
652 | ctx = fctx->new; | |
653 | break; | |
654 | } | |
655 | ||
656 | if (!ctx) { | |
657 | fctx = kmalloc(sizeof(*fctx), GFP_KERNEL); | |
658 | if (!fctx) | |
659 | return -ENOMEM; | |
660 | ||
661 | ctx = kmem_cache_alloc(userfaultfd_ctx_cachep, GFP_KERNEL); | |
662 | if (!ctx) { | |
663 | kfree(fctx); | |
664 | return -ENOMEM; | |
665 | } | |
666 | ||
ca880420 | 667 | refcount_set(&ctx->refcount, 1); |
893e26e6 PE |
668 | ctx->flags = octx->flags; |
669 | ctx->state = UFFD_STATE_RUNNING; | |
670 | ctx->features = octx->features; | |
671 | ctx->released = false; | |
df2cc96e | 672 | ctx->mmap_changing = false; |
893e26e6 | 673 | ctx->mm = vma->vm_mm; |
00bb31fa | 674 | mmgrab(ctx->mm); |
893e26e6 PE |
675 | |
676 | userfaultfd_ctx_get(octx); | |
df2cc96e | 677 | WRITE_ONCE(octx->mmap_changing, true); |
893e26e6 PE |
678 | fctx->orig = octx; |
679 | fctx->new = ctx; | |
680 | list_add_tail(&fctx->list, fcs); | |
681 | } | |
682 | ||
683 | vma->vm_userfaultfd_ctx.ctx = ctx; | |
684 | return 0; | |
685 | } | |
686 | ||
8c9e7bb7 | 687 | static void dup_fctx(struct userfaultfd_fork_ctx *fctx) |
893e26e6 PE |
688 | { |
689 | struct userfaultfd_ctx *ctx = fctx->orig; | |
690 | struct userfaultfd_wait_queue ewq; | |
691 | ||
692 | msg_init(&ewq.msg); | |
693 | ||
694 | ewq.msg.event = UFFD_EVENT_FORK; | |
695 | ewq.msg.arg.reserved.reserved1 = (unsigned long)fctx->new; | |
696 | ||
8c9e7bb7 | 697 | userfaultfd_event_wait_completion(ctx, &ewq); |
893e26e6 PE |
698 | } |
699 | ||
700 | void dup_userfaultfd_complete(struct list_head *fcs) | |
701 | { | |
893e26e6 PE |
702 | struct userfaultfd_fork_ctx *fctx, *n; |
703 | ||
704 | list_for_each_entry_safe(fctx, n, fcs, list) { | |
8c9e7bb7 | 705 | dup_fctx(fctx); |
893e26e6 PE |
706 | list_del(&fctx->list); |
707 | kfree(fctx); | |
708 | } | |
709 | } | |
710 | ||
72f87654 PE |
711 | void mremap_userfaultfd_prep(struct vm_area_struct *vma, |
712 | struct vm_userfaultfd_ctx *vm_ctx) | |
713 | { | |
714 | struct userfaultfd_ctx *ctx; | |
715 | ||
716 | ctx = vma->vm_userfaultfd_ctx.ctx; | |
3cfd22be PX |
717 | |
718 | if (!ctx) | |
719 | return; | |
720 | ||
721 | if (ctx->features & UFFD_FEATURE_EVENT_REMAP) { | |
72f87654 PE |
722 | vm_ctx->ctx = ctx; |
723 | userfaultfd_ctx_get(ctx); | |
df2cc96e | 724 | WRITE_ONCE(ctx->mmap_changing, true); |
3cfd22be PX |
725 | } else { |
726 | /* Drop uffd context if remap feature not enabled */ | |
727 | vma->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX; | |
7677f7fd | 728 | vma->vm_flags &= ~__VM_UFFD_FLAGS; |
72f87654 PE |
729 | } |
730 | } | |
731 | ||
90794bf1 | 732 | void mremap_userfaultfd_complete(struct vm_userfaultfd_ctx *vm_ctx, |
72f87654 PE |
733 | unsigned long from, unsigned long to, |
734 | unsigned long len) | |
735 | { | |
90794bf1 | 736 | struct userfaultfd_ctx *ctx = vm_ctx->ctx; |
72f87654 PE |
737 | struct userfaultfd_wait_queue ewq; |
738 | ||
739 | if (!ctx) | |
740 | return; | |
741 | ||
742 | if (to & ~PAGE_MASK) { | |
743 | userfaultfd_ctx_put(ctx); | |
744 | return; | |
745 | } | |
746 | ||
747 | msg_init(&ewq.msg); | |
748 | ||
749 | ewq.msg.event = UFFD_EVENT_REMAP; | |
750 | ewq.msg.arg.remap.from = from; | |
751 | ewq.msg.arg.remap.to = to; | |
752 | ewq.msg.arg.remap.len = len; | |
753 | ||
754 | userfaultfd_event_wait_completion(ctx, &ewq); | |
755 | } | |
756 | ||
70ccb92f | 757 | bool userfaultfd_remove(struct vm_area_struct *vma, |
d811914d | 758 | unsigned long start, unsigned long end) |
05ce7724 PE |
759 | { |
760 | struct mm_struct *mm = vma->vm_mm; | |
761 | struct userfaultfd_ctx *ctx; | |
762 | struct userfaultfd_wait_queue ewq; | |
763 | ||
764 | ctx = vma->vm_userfaultfd_ctx.ctx; | |
d811914d | 765 | if (!ctx || !(ctx->features & UFFD_FEATURE_EVENT_REMOVE)) |
70ccb92f | 766 | return true; |
05ce7724 PE |
767 | |
768 | userfaultfd_ctx_get(ctx); | |
df2cc96e | 769 | WRITE_ONCE(ctx->mmap_changing, true); |
d8ed45c5 | 770 | mmap_read_unlock(mm); |
05ce7724 | 771 | |
05ce7724 PE |
772 | msg_init(&ewq.msg); |
773 | ||
d811914d MR |
774 | ewq.msg.event = UFFD_EVENT_REMOVE; |
775 | ewq.msg.arg.remove.start = start; | |
776 | ewq.msg.arg.remove.end = end; | |
05ce7724 PE |
777 | |
778 | userfaultfd_event_wait_completion(ctx, &ewq); | |
779 | ||
70ccb92f | 780 | return false; |
05ce7724 PE |
781 | } |
782 | ||
897ab3e0 MR |
783 | static bool has_unmap_ctx(struct userfaultfd_ctx *ctx, struct list_head *unmaps, |
784 | unsigned long start, unsigned long end) | |
785 | { | |
786 | struct userfaultfd_unmap_ctx *unmap_ctx; | |
787 | ||
788 | list_for_each_entry(unmap_ctx, unmaps, list) | |
789 | if (unmap_ctx->ctx == ctx && unmap_ctx->start == start && | |
790 | unmap_ctx->end == end) | |
791 | return true; | |
792 | ||
793 | return false; | |
794 | } | |
795 | ||
796 | int userfaultfd_unmap_prep(struct vm_area_struct *vma, | |
797 | unsigned long start, unsigned long end, | |
798 | struct list_head *unmaps) | |
799 | { | |
800 | for ( ; vma && vma->vm_start < end; vma = vma->vm_next) { | |
801 | struct userfaultfd_unmap_ctx *unmap_ctx; | |
802 | struct userfaultfd_ctx *ctx = vma->vm_userfaultfd_ctx.ctx; | |
803 | ||
804 | if (!ctx || !(ctx->features & UFFD_FEATURE_EVENT_UNMAP) || | |
805 | has_unmap_ctx(ctx, unmaps, start, end)) | |
806 | continue; | |
807 | ||
808 | unmap_ctx = kzalloc(sizeof(*unmap_ctx), GFP_KERNEL); | |
809 | if (!unmap_ctx) | |
810 | return -ENOMEM; | |
811 | ||
812 | userfaultfd_ctx_get(ctx); | |
df2cc96e | 813 | WRITE_ONCE(ctx->mmap_changing, true); |
897ab3e0 MR |
814 | unmap_ctx->ctx = ctx; |
815 | unmap_ctx->start = start; | |
816 | unmap_ctx->end = end; | |
817 | list_add_tail(&unmap_ctx->list, unmaps); | |
818 | } | |
819 | ||
820 | return 0; | |
821 | } | |
822 | ||
823 | void userfaultfd_unmap_complete(struct mm_struct *mm, struct list_head *uf) | |
824 | { | |
825 | struct userfaultfd_unmap_ctx *ctx, *n; | |
826 | struct userfaultfd_wait_queue ewq; | |
827 | ||
828 | list_for_each_entry_safe(ctx, n, uf, list) { | |
829 | msg_init(&ewq.msg); | |
830 | ||
831 | ewq.msg.event = UFFD_EVENT_UNMAP; | |
832 | ewq.msg.arg.remove.start = ctx->start; | |
833 | ewq.msg.arg.remove.end = ctx->end; | |
834 | ||
835 | userfaultfd_event_wait_completion(ctx->ctx, &ewq); | |
836 | ||
837 | list_del(&ctx->list); | |
838 | kfree(ctx); | |
839 | } | |
840 | } | |
841 | ||
86039bd3 AA |
842 | static int userfaultfd_release(struct inode *inode, struct file *file) |
843 | { | |
844 | struct userfaultfd_ctx *ctx = file->private_data; | |
845 | struct mm_struct *mm = ctx->mm; | |
846 | struct vm_area_struct *vma, *prev; | |
847 | /* len == 0 means wake all */ | |
848 | struct userfaultfd_wake_range range = { .len = 0, }; | |
849 | unsigned long new_flags; | |
850 | ||
6aa7de05 | 851 | WRITE_ONCE(ctx->released, true); |
86039bd3 | 852 | |
d2005e3f ON |
853 | if (!mmget_not_zero(mm)) |
854 | goto wakeup; | |
855 | ||
86039bd3 AA |
856 | /* |
857 | * Flush page faults out of all CPUs. NOTE: all page faults | |
858 | * must be retried without returning VM_FAULT_SIGBUS if | |
859 | * userfaultfd_ctx_get() succeeds but vma->vma_userfault_ctx | |
c1e8d7c6 | 860 | * changes while handle_userfault released the mmap_lock. So |
86039bd3 | 861 | * it's critical that released is set to true (above), before |
c1e8d7c6 | 862 | * taking the mmap_lock for writing. |
86039bd3 | 863 | */ |
d8ed45c5 | 864 | mmap_write_lock(mm); |
86039bd3 AA |
865 | prev = NULL; |
866 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
867 | cond_resched(); | |
868 | BUG_ON(!!vma->vm_userfaultfd_ctx.ctx ^ | |
7677f7fd | 869 | !!(vma->vm_flags & __VM_UFFD_FLAGS)); |
86039bd3 AA |
870 | if (vma->vm_userfaultfd_ctx.ctx != ctx) { |
871 | prev = vma; | |
872 | continue; | |
873 | } | |
7677f7fd | 874 | new_flags = vma->vm_flags & ~__VM_UFFD_FLAGS; |
4d45e75a JH |
875 | prev = vma_merge(mm, prev, vma->vm_start, vma->vm_end, |
876 | new_flags, vma->anon_vma, | |
877 | vma->vm_file, vma->vm_pgoff, | |
878 | vma_policy(vma), | |
879 | NULL_VM_UFFD_CTX); | |
880 | if (prev) | |
881 | vma = prev; | |
882 | else | |
883 | prev = vma; | |
86039bd3 AA |
884 | vma->vm_flags = new_flags; |
885 | vma->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX; | |
886 | } | |
d8ed45c5 | 887 | mmap_write_unlock(mm); |
d2005e3f ON |
888 | mmput(mm); |
889 | wakeup: | |
86039bd3 | 890 | /* |
15b726ef | 891 | * After no new page faults can wait on this fault_*wqh, flush |
86039bd3 | 892 | * the last page faults that may have been already waiting on |
15b726ef | 893 | * the fault_*wqh. |
86039bd3 | 894 | */ |
cbcfa130 | 895 | spin_lock_irq(&ctx->fault_pending_wqh.lock); |
ac5be6b4 | 896 | __wake_up_locked_key(&ctx->fault_pending_wqh, TASK_NORMAL, &range); |
c430d1e8 | 897 | __wake_up(&ctx->fault_wqh, TASK_NORMAL, 1, &range); |
cbcfa130 | 898 | spin_unlock_irq(&ctx->fault_pending_wqh.lock); |
86039bd3 | 899 | |
5a18b64e MR |
900 | /* Flush pending events that may still wait on event_wqh */ |
901 | wake_up_all(&ctx->event_wqh); | |
902 | ||
a9a08845 | 903 | wake_up_poll(&ctx->fd_wqh, EPOLLHUP); |
86039bd3 AA |
904 | userfaultfd_ctx_put(ctx); |
905 | return 0; | |
906 | } | |
907 | ||
15b726ef | 908 | /* fault_pending_wqh.lock must be hold by the caller */ |
6dcc27fd PE |
909 | static inline struct userfaultfd_wait_queue *find_userfault_in( |
910 | wait_queue_head_t *wqh) | |
86039bd3 | 911 | { |
ac6424b9 | 912 | wait_queue_entry_t *wq; |
15b726ef | 913 | struct userfaultfd_wait_queue *uwq; |
86039bd3 | 914 | |
456a7378 | 915 | lockdep_assert_held(&wqh->lock); |
86039bd3 | 916 | |
15b726ef | 917 | uwq = NULL; |
6dcc27fd | 918 | if (!waitqueue_active(wqh)) |
15b726ef AA |
919 | goto out; |
920 | /* walk in reverse to provide FIFO behavior to read userfaults */ | |
2055da97 | 921 | wq = list_last_entry(&wqh->head, typeof(*wq), entry); |
15b726ef AA |
922 | uwq = container_of(wq, struct userfaultfd_wait_queue, wq); |
923 | out: | |
924 | return uwq; | |
86039bd3 | 925 | } |
6dcc27fd PE |
926 | |
927 | static inline struct userfaultfd_wait_queue *find_userfault( | |
928 | struct userfaultfd_ctx *ctx) | |
929 | { | |
930 | return find_userfault_in(&ctx->fault_pending_wqh); | |
931 | } | |
86039bd3 | 932 | |
9cd75c3c PE |
933 | static inline struct userfaultfd_wait_queue *find_userfault_evt( |
934 | struct userfaultfd_ctx *ctx) | |
935 | { | |
936 | return find_userfault_in(&ctx->event_wqh); | |
937 | } | |
938 | ||
076ccb76 | 939 | static __poll_t userfaultfd_poll(struct file *file, poll_table *wait) |
86039bd3 AA |
940 | { |
941 | struct userfaultfd_ctx *ctx = file->private_data; | |
076ccb76 | 942 | __poll_t ret; |
86039bd3 AA |
943 | |
944 | poll_wait(file, &ctx->fd_wqh, wait); | |
945 | ||
946 | switch (ctx->state) { | |
947 | case UFFD_STATE_WAIT_API: | |
a9a08845 | 948 | return EPOLLERR; |
86039bd3 | 949 | case UFFD_STATE_RUNNING: |
ba85c702 AA |
950 | /* |
951 | * poll() never guarantees that read won't block. | |
952 | * userfaults can be waken before they're read(). | |
953 | */ | |
954 | if (unlikely(!(file->f_flags & O_NONBLOCK))) | |
a9a08845 | 955 | return EPOLLERR; |
15b726ef AA |
956 | /* |
957 | * lockless access to see if there are pending faults | |
958 | * __pollwait last action is the add_wait_queue but | |
959 | * the spin_unlock would allow the waitqueue_active to | |
960 | * pass above the actual list_add inside | |
961 | * add_wait_queue critical section. So use a full | |
962 | * memory barrier to serialize the list_add write of | |
963 | * add_wait_queue() with the waitqueue_active read | |
964 | * below. | |
965 | */ | |
966 | ret = 0; | |
967 | smp_mb(); | |
968 | if (waitqueue_active(&ctx->fault_pending_wqh)) | |
a9a08845 | 969 | ret = EPOLLIN; |
9cd75c3c | 970 | else if (waitqueue_active(&ctx->event_wqh)) |
a9a08845 | 971 | ret = EPOLLIN; |
9cd75c3c | 972 | |
86039bd3 AA |
973 | return ret; |
974 | default: | |
8474901a | 975 | WARN_ON_ONCE(1); |
a9a08845 | 976 | return EPOLLERR; |
86039bd3 AA |
977 | } |
978 | } | |
979 | ||
893e26e6 PE |
980 | static const struct file_operations userfaultfd_fops; |
981 | ||
b537900f DC |
982 | static int resolve_userfault_fork(struct userfaultfd_ctx *new, |
983 | struct inode *inode, | |
893e26e6 PE |
984 | struct uffd_msg *msg) |
985 | { | |
986 | int fd; | |
893e26e6 | 987 | |
b537900f DC |
988 | fd = anon_inode_getfd_secure("[userfaultfd]", &userfaultfd_fops, new, |
989 | O_RDWR | (new->flags & UFFD_SHARED_FCNTL_FLAGS), inode); | |
893e26e6 PE |
990 | if (fd < 0) |
991 | return fd; | |
992 | ||
893e26e6 PE |
993 | msg->arg.reserved.reserved1 = 0; |
994 | msg->arg.fork.ufd = fd; | |
893e26e6 PE |
995 | return 0; |
996 | } | |
997 | ||
86039bd3 | 998 | static ssize_t userfaultfd_ctx_read(struct userfaultfd_ctx *ctx, int no_wait, |
b537900f | 999 | struct uffd_msg *msg, struct inode *inode) |
86039bd3 AA |
1000 | { |
1001 | ssize_t ret; | |
1002 | DECLARE_WAITQUEUE(wait, current); | |
15b726ef | 1003 | struct userfaultfd_wait_queue *uwq; |
893e26e6 PE |
1004 | /* |
1005 | * Handling fork event requires sleeping operations, so | |
1006 | * we drop the event_wqh lock, then do these ops, then | |
1007 | * lock it back and wake up the waiter. While the lock is | |
1008 | * dropped the ewq may go away so we keep track of it | |
1009 | * carefully. | |
1010 | */ | |
1011 | LIST_HEAD(fork_event); | |
1012 | struct userfaultfd_ctx *fork_nctx = NULL; | |
86039bd3 | 1013 | |
15b726ef | 1014 | /* always take the fd_wqh lock before the fault_pending_wqh lock */ |
ae62c16e | 1015 | spin_lock_irq(&ctx->fd_wqh.lock); |
86039bd3 AA |
1016 | __add_wait_queue(&ctx->fd_wqh, &wait); |
1017 | for (;;) { | |
1018 | set_current_state(TASK_INTERRUPTIBLE); | |
15b726ef AA |
1019 | spin_lock(&ctx->fault_pending_wqh.lock); |
1020 | uwq = find_userfault(ctx); | |
1021 | if (uwq) { | |
2c5b7e1b AA |
1022 | /* |
1023 | * Use a seqcount to repeat the lockless check | |
1024 | * in wake_userfault() to avoid missing | |
1025 | * wakeups because during the refile both | |
1026 | * waitqueue could become empty if this is the | |
1027 | * only userfault. | |
1028 | */ | |
1029 | write_seqcount_begin(&ctx->refile_seq); | |
1030 | ||
86039bd3 | 1031 | /* |
15b726ef AA |
1032 | * The fault_pending_wqh.lock prevents the uwq |
1033 | * to disappear from under us. | |
1034 | * | |
1035 | * Refile this userfault from | |
1036 | * fault_pending_wqh to fault_wqh, it's not | |
1037 | * pending anymore after we read it. | |
1038 | * | |
1039 | * Use list_del() by hand (as | |
1040 | * userfaultfd_wake_function also uses | |
1041 | * list_del_init() by hand) to be sure nobody | |
1042 | * changes __remove_wait_queue() to use | |
1043 | * list_del_init() in turn breaking the | |
1044 | * !list_empty_careful() check in | |
2055da97 | 1045 | * handle_userfault(). The uwq->wq.head list |
15b726ef AA |
1046 | * must never be empty at any time during the |
1047 | * refile, or the waitqueue could disappear | |
1048 | * from under us. The "wait_queue_head_t" | |
1049 | * parameter of __remove_wait_queue() is unused | |
1050 | * anyway. | |
86039bd3 | 1051 | */ |
2055da97 | 1052 | list_del(&uwq->wq.entry); |
c430d1e8 | 1053 | add_wait_queue(&ctx->fault_wqh, &uwq->wq); |
15b726ef | 1054 | |
2c5b7e1b AA |
1055 | write_seqcount_end(&ctx->refile_seq); |
1056 | ||
a9b85f94 AA |
1057 | /* careful to always initialize msg if ret == 0 */ |
1058 | *msg = uwq->msg; | |
15b726ef | 1059 | spin_unlock(&ctx->fault_pending_wqh.lock); |
86039bd3 AA |
1060 | ret = 0; |
1061 | break; | |
1062 | } | |
15b726ef | 1063 | spin_unlock(&ctx->fault_pending_wqh.lock); |
9cd75c3c PE |
1064 | |
1065 | spin_lock(&ctx->event_wqh.lock); | |
1066 | uwq = find_userfault_evt(ctx); | |
1067 | if (uwq) { | |
1068 | *msg = uwq->msg; | |
1069 | ||
893e26e6 PE |
1070 | if (uwq->msg.event == UFFD_EVENT_FORK) { |
1071 | fork_nctx = (struct userfaultfd_ctx *) | |
1072 | (unsigned long) | |
1073 | uwq->msg.arg.reserved.reserved1; | |
2055da97 | 1074 | list_move(&uwq->wq.entry, &fork_event); |
384632e6 AA |
1075 | /* |
1076 | * fork_nctx can be freed as soon as | |
1077 | * we drop the lock, unless we take a | |
1078 | * reference on it. | |
1079 | */ | |
1080 | userfaultfd_ctx_get(fork_nctx); | |
893e26e6 PE |
1081 | spin_unlock(&ctx->event_wqh.lock); |
1082 | ret = 0; | |
1083 | break; | |
1084 | } | |
1085 | ||
9cd75c3c PE |
1086 | userfaultfd_event_complete(ctx, uwq); |
1087 | spin_unlock(&ctx->event_wqh.lock); | |
1088 | ret = 0; | |
1089 | break; | |
1090 | } | |
1091 | spin_unlock(&ctx->event_wqh.lock); | |
1092 | ||
86039bd3 AA |
1093 | if (signal_pending(current)) { |
1094 | ret = -ERESTARTSYS; | |
1095 | break; | |
1096 | } | |
1097 | if (no_wait) { | |
1098 | ret = -EAGAIN; | |
1099 | break; | |
1100 | } | |
ae62c16e | 1101 | spin_unlock_irq(&ctx->fd_wqh.lock); |
86039bd3 | 1102 | schedule(); |
ae62c16e | 1103 | spin_lock_irq(&ctx->fd_wqh.lock); |
86039bd3 AA |
1104 | } |
1105 | __remove_wait_queue(&ctx->fd_wqh, &wait); | |
1106 | __set_current_state(TASK_RUNNING); | |
ae62c16e | 1107 | spin_unlock_irq(&ctx->fd_wqh.lock); |
86039bd3 | 1108 | |
893e26e6 | 1109 | if (!ret && msg->event == UFFD_EVENT_FORK) { |
b537900f | 1110 | ret = resolve_userfault_fork(fork_nctx, inode, msg); |
cbcfa130 | 1111 | spin_lock_irq(&ctx->event_wqh.lock); |
384632e6 AA |
1112 | if (!list_empty(&fork_event)) { |
1113 | /* | |
1114 | * The fork thread didn't abort, so we can | |
1115 | * drop the temporary refcount. | |
1116 | */ | |
1117 | userfaultfd_ctx_put(fork_nctx); | |
1118 | ||
1119 | uwq = list_first_entry(&fork_event, | |
1120 | typeof(*uwq), | |
1121 | wq.entry); | |
1122 | /* | |
1123 | * If fork_event list wasn't empty and in turn | |
1124 | * the event wasn't already released by fork | |
1125 | * (the event is allocated on fork kernel | |
1126 | * stack), put the event back to its place in | |
1127 | * the event_wq. fork_event head will be freed | |
1128 | * as soon as we return so the event cannot | |
1129 | * stay queued there no matter the current | |
1130 | * "ret" value. | |
1131 | */ | |
1132 | list_del(&uwq->wq.entry); | |
1133 | __add_wait_queue(&ctx->event_wqh, &uwq->wq); | |
893e26e6 | 1134 | |
384632e6 AA |
1135 | /* |
1136 | * Leave the event in the waitqueue and report | |
1137 | * error to userland if we failed to resolve | |
1138 | * the userfault fork. | |
1139 | */ | |
1140 | if (likely(!ret)) | |
893e26e6 | 1141 | userfaultfd_event_complete(ctx, uwq); |
384632e6 AA |
1142 | } else { |
1143 | /* | |
1144 | * Here the fork thread aborted and the | |
1145 | * refcount from the fork thread on fork_nctx | |
1146 | * has already been released. We still hold | |
1147 | * the reference we took before releasing the | |
1148 | * lock above. If resolve_userfault_fork | |
1149 | * failed we've to drop it because the | |
1150 | * fork_nctx has to be freed in such case. If | |
1151 | * it succeeded we'll hold it because the new | |
1152 | * uffd references it. | |
1153 | */ | |
1154 | if (ret) | |
1155 | userfaultfd_ctx_put(fork_nctx); | |
893e26e6 | 1156 | } |
cbcfa130 | 1157 | spin_unlock_irq(&ctx->event_wqh.lock); |
893e26e6 PE |
1158 | } |
1159 | ||
86039bd3 AA |
1160 | return ret; |
1161 | } | |
1162 | ||
1163 | static ssize_t userfaultfd_read(struct file *file, char __user *buf, | |
1164 | size_t count, loff_t *ppos) | |
1165 | { | |
1166 | struct userfaultfd_ctx *ctx = file->private_data; | |
1167 | ssize_t _ret, ret = 0; | |
a9b85f94 | 1168 | struct uffd_msg msg; |
86039bd3 | 1169 | int no_wait = file->f_flags & O_NONBLOCK; |
b537900f | 1170 | struct inode *inode = file_inode(file); |
86039bd3 AA |
1171 | |
1172 | if (ctx->state == UFFD_STATE_WAIT_API) | |
1173 | return -EINVAL; | |
86039bd3 AA |
1174 | |
1175 | for (;;) { | |
a9b85f94 | 1176 | if (count < sizeof(msg)) |
86039bd3 | 1177 | return ret ? ret : -EINVAL; |
b537900f | 1178 | _ret = userfaultfd_ctx_read(ctx, no_wait, &msg, inode); |
86039bd3 AA |
1179 | if (_ret < 0) |
1180 | return ret ? ret : _ret; | |
a9b85f94 | 1181 | if (copy_to_user((__u64 __user *) buf, &msg, sizeof(msg))) |
86039bd3 | 1182 | return ret ? ret : -EFAULT; |
a9b85f94 AA |
1183 | ret += sizeof(msg); |
1184 | buf += sizeof(msg); | |
1185 | count -= sizeof(msg); | |
86039bd3 AA |
1186 | /* |
1187 | * Allow to read more than one fault at time but only | |
1188 | * block if waiting for the very first one. | |
1189 | */ | |
1190 | no_wait = O_NONBLOCK; | |
1191 | } | |
1192 | } | |
1193 | ||
1194 | static void __wake_userfault(struct userfaultfd_ctx *ctx, | |
1195 | struct userfaultfd_wake_range *range) | |
1196 | { | |
cbcfa130 | 1197 | spin_lock_irq(&ctx->fault_pending_wqh.lock); |
86039bd3 | 1198 | /* wake all in the range and autoremove */ |
15b726ef | 1199 | if (waitqueue_active(&ctx->fault_pending_wqh)) |
ac5be6b4 | 1200 | __wake_up_locked_key(&ctx->fault_pending_wqh, TASK_NORMAL, |
15b726ef AA |
1201 | range); |
1202 | if (waitqueue_active(&ctx->fault_wqh)) | |
c430d1e8 | 1203 | __wake_up(&ctx->fault_wqh, TASK_NORMAL, 1, range); |
cbcfa130 | 1204 | spin_unlock_irq(&ctx->fault_pending_wqh.lock); |
86039bd3 AA |
1205 | } |
1206 | ||
1207 | static __always_inline void wake_userfault(struct userfaultfd_ctx *ctx, | |
1208 | struct userfaultfd_wake_range *range) | |
1209 | { | |
2c5b7e1b AA |
1210 | unsigned seq; |
1211 | bool need_wakeup; | |
1212 | ||
86039bd3 AA |
1213 | /* |
1214 | * To be sure waitqueue_active() is not reordered by the CPU | |
1215 | * before the pagetable update, use an explicit SMP memory | |
3e4e28c5 | 1216 | * barrier here. PT lock release or mmap_read_unlock(mm) still |
86039bd3 AA |
1217 | * have release semantics that can allow the |
1218 | * waitqueue_active() to be reordered before the pte update. | |
1219 | */ | |
1220 | smp_mb(); | |
1221 | ||
1222 | /* | |
1223 | * Use waitqueue_active because it's very frequent to | |
1224 | * change the address space atomically even if there are no | |
1225 | * userfaults yet. So we take the spinlock only when we're | |
1226 | * sure we've userfaults to wake. | |
1227 | */ | |
2c5b7e1b AA |
1228 | do { |
1229 | seq = read_seqcount_begin(&ctx->refile_seq); | |
1230 | need_wakeup = waitqueue_active(&ctx->fault_pending_wqh) || | |
1231 | waitqueue_active(&ctx->fault_wqh); | |
1232 | cond_resched(); | |
1233 | } while (read_seqcount_retry(&ctx->refile_seq, seq)); | |
1234 | if (need_wakeup) | |
86039bd3 AA |
1235 | __wake_userfault(ctx, range); |
1236 | } | |
1237 | ||
1238 | static __always_inline int validate_range(struct mm_struct *mm, | |
7d032574 | 1239 | __u64 *start, __u64 len) |
86039bd3 AA |
1240 | { |
1241 | __u64 task_size = mm->task_size; | |
1242 | ||
7d032574 AK |
1243 | *start = untagged_addr(*start); |
1244 | ||
1245 | if (*start & ~PAGE_MASK) | |
86039bd3 AA |
1246 | return -EINVAL; |
1247 | if (len & ~PAGE_MASK) | |
1248 | return -EINVAL; | |
1249 | if (!len) | |
1250 | return -EINVAL; | |
7d032574 | 1251 | if (*start < mmap_min_addr) |
86039bd3 | 1252 | return -EINVAL; |
7d032574 | 1253 | if (*start >= task_size) |
86039bd3 | 1254 | return -EINVAL; |
7d032574 | 1255 | if (len > task_size - *start) |
86039bd3 AA |
1256 | return -EINVAL; |
1257 | return 0; | |
1258 | } | |
1259 | ||
63b2d417 AA |
1260 | static inline bool vma_can_userfault(struct vm_area_struct *vma, |
1261 | unsigned long vm_flags) | |
ba6907db | 1262 | { |
63b2d417 | 1263 | /* FIXME: add WP support to hugetlbfs and shmem */ |
7677f7fd AR |
1264 | if (vm_flags & VM_UFFD_WP) { |
1265 | if (is_vm_hugetlb_page(vma) || vma_is_shmem(vma)) | |
1266 | return false; | |
1267 | } | |
1268 | ||
1269 | if (vm_flags & VM_UFFD_MINOR) { | |
1270 | /* FIXME: Add minor fault interception for shmem. */ | |
1271 | if (!is_vm_hugetlb_page(vma)) | |
1272 | return false; | |
1273 | } | |
1274 | ||
1275 | return vma_is_anonymous(vma) || is_vm_hugetlb_page(vma) || | |
1276 | vma_is_shmem(vma); | |
ba6907db MR |
1277 | } |
1278 | ||
86039bd3 AA |
1279 | static int userfaultfd_register(struct userfaultfd_ctx *ctx, |
1280 | unsigned long arg) | |
1281 | { | |
1282 | struct mm_struct *mm = ctx->mm; | |
1283 | struct vm_area_struct *vma, *prev, *cur; | |
1284 | int ret; | |
1285 | struct uffdio_register uffdio_register; | |
1286 | struct uffdio_register __user *user_uffdio_register; | |
1287 | unsigned long vm_flags, new_flags; | |
1288 | bool found; | |
ce53e8e6 | 1289 | bool basic_ioctls; |
86039bd3 AA |
1290 | unsigned long start, end, vma_end; |
1291 | ||
1292 | user_uffdio_register = (struct uffdio_register __user *) arg; | |
1293 | ||
1294 | ret = -EFAULT; | |
1295 | if (copy_from_user(&uffdio_register, user_uffdio_register, | |
1296 | sizeof(uffdio_register)-sizeof(__u64))) | |
1297 | goto out; | |
1298 | ||
1299 | ret = -EINVAL; | |
1300 | if (!uffdio_register.mode) | |
1301 | goto out; | |
7677f7fd | 1302 | if (uffdio_register.mode & ~UFFD_API_REGISTER_MODES) |
86039bd3 AA |
1303 | goto out; |
1304 | vm_flags = 0; | |
1305 | if (uffdio_register.mode & UFFDIO_REGISTER_MODE_MISSING) | |
1306 | vm_flags |= VM_UFFD_MISSING; | |
63b2d417 | 1307 | if (uffdio_register.mode & UFFDIO_REGISTER_MODE_WP) |
86039bd3 | 1308 | vm_flags |= VM_UFFD_WP; |
7677f7fd AR |
1309 | if (uffdio_register.mode & UFFDIO_REGISTER_MODE_MINOR) { |
1310 | #ifndef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR | |
1311 | goto out; | |
1312 | #endif | |
1313 | vm_flags |= VM_UFFD_MINOR; | |
1314 | } | |
86039bd3 | 1315 | |
7d032574 | 1316 | ret = validate_range(mm, &uffdio_register.range.start, |
86039bd3 AA |
1317 | uffdio_register.range.len); |
1318 | if (ret) | |
1319 | goto out; | |
1320 | ||
1321 | start = uffdio_register.range.start; | |
1322 | end = start + uffdio_register.range.len; | |
1323 | ||
d2005e3f ON |
1324 | ret = -ENOMEM; |
1325 | if (!mmget_not_zero(mm)) | |
1326 | goto out; | |
1327 | ||
d8ed45c5 | 1328 | mmap_write_lock(mm); |
86039bd3 | 1329 | vma = find_vma_prev(mm, start, &prev); |
86039bd3 AA |
1330 | if (!vma) |
1331 | goto out_unlock; | |
1332 | ||
1333 | /* check that there's at least one vma in the range */ | |
1334 | ret = -EINVAL; | |
1335 | if (vma->vm_start >= end) | |
1336 | goto out_unlock; | |
1337 | ||
cab350af MK |
1338 | /* |
1339 | * If the first vma contains huge pages, make sure start address | |
1340 | * is aligned to huge page size. | |
1341 | */ | |
1342 | if (is_vm_hugetlb_page(vma)) { | |
1343 | unsigned long vma_hpagesize = vma_kernel_pagesize(vma); | |
1344 | ||
1345 | if (start & (vma_hpagesize - 1)) | |
1346 | goto out_unlock; | |
1347 | } | |
1348 | ||
86039bd3 AA |
1349 | /* |
1350 | * Search for not compatible vmas. | |
86039bd3 AA |
1351 | */ |
1352 | found = false; | |
ce53e8e6 | 1353 | basic_ioctls = false; |
86039bd3 AA |
1354 | for (cur = vma; cur && cur->vm_start < end; cur = cur->vm_next) { |
1355 | cond_resched(); | |
1356 | ||
1357 | BUG_ON(!!cur->vm_userfaultfd_ctx.ctx ^ | |
7677f7fd | 1358 | !!(cur->vm_flags & __VM_UFFD_FLAGS)); |
86039bd3 AA |
1359 | |
1360 | /* check not compatible vmas */ | |
1361 | ret = -EINVAL; | |
63b2d417 | 1362 | if (!vma_can_userfault(cur, vm_flags)) |
86039bd3 | 1363 | goto out_unlock; |
29ec9066 AA |
1364 | |
1365 | /* | |
1366 | * UFFDIO_COPY will fill file holes even without | |
1367 | * PROT_WRITE. This check enforces that if this is a | |
1368 | * MAP_SHARED, the process has write permission to the backing | |
1369 | * file. If VM_MAYWRITE is set it also enforces that on a | |
1370 | * MAP_SHARED vma: there is no F_WRITE_SEAL and no further | |
1371 | * F_WRITE_SEAL can be taken until the vma is destroyed. | |
1372 | */ | |
1373 | ret = -EPERM; | |
1374 | if (unlikely(!(cur->vm_flags & VM_MAYWRITE))) | |
1375 | goto out_unlock; | |
1376 | ||
cab350af MK |
1377 | /* |
1378 | * If this vma contains ending address, and huge pages | |
1379 | * check alignment. | |
1380 | */ | |
1381 | if (is_vm_hugetlb_page(cur) && end <= cur->vm_end && | |
1382 | end > cur->vm_start) { | |
1383 | unsigned long vma_hpagesize = vma_kernel_pagesize(cur); | |
1384 | ||
1385 | ret = -EINVAL; | |
1386 | ||
1387 | if (end & (vma_hpagesize - 1)) | |
1388 | goto out_unlock; | |
1389 | } | |
63b2d417 AA |
1390 | if ((vm_flags & VM_UFFD_WP) && !(cur->vm_flags & VM_MAYWRITE)) |
1391 | goto out_unlock; | |
86039bd3 AA |
1392 | |
1393 | /* | |
1394 | * Check that this vma isn't already owned by a | |
1395 | * different userfaultfd. We can't allow more than one | |
1396 | * userfaultfd to own a single vma simultaneously or we | |
1397 | * wouldn't know which one to deliver the userfaults to. | |
1398 | */ | |
1399 | ret = -EBUSY; | |
1400 | if (cur->vm_userfaultfd_ctx.ctx && | |
1401 | cur->vm_userfaultfd_ctx.ctx != ctx) | |
1402 | goto out_unlock; | |
1403 | ||
cab350af MK |
1404 | /* |
1405 | * Note vmas containing huge pages | |
1406 | */ | |
ce53e8e6 MR |
1407 | if (is_vm_hugetlb_page(cur)) |
1408 | basic_ioctls = true; | |
cab350af | 1409 | |
86039bd3 AA |
1410 | found = true; |
1411 | } | |
1412 | BUG_ON(!found); | |
1413 | ||
1414 | if (vma->vm_start < start) | |
1415 | prev = vma; | |
1416 | ||
1417 | ret = 0; | |
1418 | do { | |
1419 | cond_resched(); | |
1420 | ||
63b2d417 | 1421 | BUG_ON(!vma_can_userfault(vma, vm_flags)); |
86039bd3 AA |
1422 | BUG_ON(vma->vm_userfaultfd_ctx.ctx && |
1423 | vma->vm_userfaultfd_ctx.ctx != ctx); | |
29ec9066 | 1424 | WARN_ON(!(vma->vm_flags & VM_MAYWRITE)); |
86039bd3 AA |
1425 | |
1426 | /* | |
1427 | * Nothing to do: this vma is already registered into this | |
1428 | * userfaultfd and with the right tracking mode too. | |
1429 | */ | |
1430 | if (vma->vm_userfaultfd_ctx.ctx == ctx && | |
1431 | (vma->vm_flags & vm_flags) == vm_flags) | |
1432 | goto skip; | |
1433 | ||
1434 | if (vma->vm_start > start) | |
1435 | start = vma->vm_start; | |
1436 | vma_end = min(end, vma->vm_end); | |
1437 | ||
7677f7fd | 1438 | new_flags = (vma->vm_flags & ~__VM_UFFD_FLAGS) | vm_flags; |
86039bd3 AA |
1439 | prev = vma_merge(mm, prev, start, vma_end, new_flags, |
1440 | vma->anon_vma, vma->vm_file, vma->vm_pgoff, | |
1441 | vma_policy(vma), | |
1442 | ((struct vm_userfaultfd_ctx){ ctx })); | |
1443 | if (prev) { | |
1444 | vma = prev; | |
1445 | goto next; | |
1446 | } | |
1447 | if (vma->vm_start < start) { | |
1448 | ret = split_vma(mm, vma, start, 1); | |
1449 | if (ret) | |
1450 | break; | |
1451 | } | |
1452 | if (vma->vm_end > end) { | |
1453 | ret = split_vma(mm, vma, end, 0); | |
1454 | if (ret) | |
1455 | break; | |
1456 | } | |
1457 | next: | |
1458 | /* | |
1459 | * In the vma_merge() successful mprotect-like case 8: | |
1460 | * the next vma was merged into the current one and | |
1461 | * the current one has not been updated yet. | |
1462 | */ | |
1463 | vma->vm_flags = new_flags; | |
1464 | vma->vm_userfaultfd_ctx.ctx = ctx; | |
1465 | ||
6dfeaff9 PX |
1466 | if (is_vm_hugetlb_page(vma) && uffd_disable_huge_pmd_share(vma)) |
1467 | hugetlb_unshare_all_pmds(vma); | |
1468 | ||
86039bd3 AA |
1469 | skip: |
1470 | prev = vma; | |
1471 | start = vma->vm_end; | |
1472 | vma = vma->vm_next; | |
1473 | } while (vma && vma->vm_start < end); | |
1474 | out_unlock: | |
d8ed45c5 | 1475 | mmap_write_unlock(mm); |
d2005e3f | 1476 | mmput(mm); |
86039bd3 | 1477 | if (!ret) { |
14819305 PX |
1478 | __u64 ioctls_out; |
1479 | ||
1480 | ioctls_out = basic_ioctls ? UFFD_API_RANGE_IOCTLS_BASIC : | |
1481 | UFFD_API_RANGE_IOCTLS; | |
1482 | ||
1483 | /* | |
1484 | * Declare the WP ioctl only if the WP mode is | |
1485 | * specified and all checks passed with the range | |
1486 | */ | |
1487 | if (!(uffdio_register.mode & UFFDIO_REGISTER_MODE_WP)) | |
1488 | ioctls_out &= ~((__u64)1 << _UFFDIO_WRITEPROTECT); | |
1489 | ||
f6191471 AR |
1490 | /* CONTINUE ioctl is only supported for MINOR ranges. */ |
1491 | if (!(uffdio_register.mode & UFFDIO_REGISTER_MODE_MINOR)) | |
1492 | ioctls_out &= ~((__u64)1 << _UFFDIO_CONTINUE); | |
1493 | ||
86039bd3 AA |
1494 | /* |
1495 | * Now that we scanned all vmas we can already tell | |
1496 | * userland which ioctls methods are guaranteed to | |
1497 | * succeed on this range. | |
1498 | */ | |
14819305 | 1499 | if (put_user(ioctls_out, &user_uffdio_register->ioctls)) |
86039bd3 AA |
1500 | ret = -EFAULT; |
1501 | } | |
1502 | out: | |
1503 | return ret; | |
1504 | } | |
1505 | ||
1506 | static int userfaultfd_unregister(struct userfaultfd_ctx *ctx, | |
1507 | unsigned long arg) | |
1508 | { | |
1509 | struct mm_struct *mm = ctx->mm; | |
1510 | struct vm_area_struct *vma, *prev, *cur; | |
1511 | int ret; | |
1512 | struct uffdio_range uffdio_unregister; | |
1513 | unsigned long new_flags; | |
1514 | bool found; | |
1515 | unsigned long start, end, vma_end; | |
1516 | const void __user *buf = (void __user *)arg; | |
1517 | ||
1518 | ret = -EFAULT; | |
1519 | if (copy_from_user(&uffdio_unregister, buf, sizeof(uffdio_unregister))) | |
1520 | goto out; | |
1521 | ||
7d032574 | 1522 | ret = validate_range(mm, &uffdio_unregister.start, |
86039bd3 AA |
1523 | uffdio_unregister.len); |
1524 | if (ret) | |
1525 | goto out; | |
1526 | ||
1527 | start = uffdio_unregister.start; | |
1528 | end = start + uffdio_unregister.len; | |
1529 | ||
d2005e3f ON |
1530 | ret = -ENOMEM; |
1531 | if (!mmget_not_zero(mm)) | |
1532 | goto out; | |
1533 | ||
d8ed45c5 | 1534 | mmap_write_lock(mm); |
86039bd3 | 1535 | vma = find_vma_prev(mm, start, &prev); |
86039bd3 AA |
1536 | if (!vma) |
1537 | goto out_unlock; | |
1538 | ||
1539 | /* check that there's at least one vma in the range */ | |
1540 | ret = -EINVAL; | |
1541 | if (vma->vm_start >= end) | |
1542 | goto out_unlock; | |
1543 | ||
cab350af MK |
1544 | /* |
1545 | * If the first vma contains huge pages, make sure start address | |
1546 | * is aligned to huge page size. | |
1547 | */ | |
1548 | if (is_vm_hugetlb_page(vma)) { | |
1549 | unsigned long vma_hpagesize = vma_kernel_pagesize(vma); | |
1550 | ||
1551 | if (start & (vma_hpagesize - 1)) | |
1552 | goto out_unlock; | |
1553 | } | |
1554 | ||
86039bd3 AA |
1555 | /* |
1556 | * Search for not compatible vmas. | |
86039bd3 AA |
1557 | */ |
1558 | found = false; | |
1559 | ret = -EINVAL; | |
1560 | for (cur = vma; cur && cur->vm_start < end; cur = cur->vm_next) { | |
1561 | cond_resched(); | |
1562 | ||
1563 | BUG_ON(!!cur->vm_userfaultfd_ctx.ctx ^ | |
7677f7fd | 1564 | !!(cur->vm_flags & __VM_UFFD_FLAGS)); |
86039bd3 AA |
1565 | |
1566 | /* | |
1567 | * Check not compatible vmas, not strictly required | |
1568 | * here as not compatible vmas cannot have an | |
1569 | * userfaultfd_ctx registered on them, but this | |
1570 | * provides for more strict behavior to notice | |
1571 | * unregistration errors. | |
1572 | */ | |
63b2d417 | 1573 | if (!vma_can_userfault(cur, cur->vm_flags)) |
86039bd3 AA |
1574 | goto out_unlock; |
1575 | ||
1576 | found = true; | |
1577 | } | |
1578 | BUG_ON(!found); | |
1579 | ||
1580 | if (vma->vm_start < start) | |
1581 | prev = vma; | |
1582 | ||
1583 | ret = 0; | |
1584 | do { | |
1585 | cond_resched(); | |
1586 | ||
63b2d417 | 1587 | BUG_ON(!vma_can_userfault(vma, vma->vm_flags)); |
86039bd3 AA |
1588 | |
1589 | /* | |
1590 | * Nothing to do: this vma is already registered into this | |
1591 | * userfaultfd and with the right tracking mode too. | |
1592 | */ | |
1593 | if (!vma->vm_userfaultfd_ctx.ctx) | |
1594 | goto skip; | |
1595 | ||
01e881f5 AA |
1596 | WARN_ON(!(vma->vm_flags & VM_MAYWRITE)); |
1597 | ||
86039bd3 AA |
1598 | if (vma->vm_start > start) |
1599 | start = vma->vm_start; | |
1600 | vma_end = min(end, vma->vm_end); | |
1601 | ||
09fa5296 AA |
1602 | if (userfaultfd_missing(vma)) { |
1603 | /* | |
1604 | * Wake any concurrent pending userfault while | |
1605 | * we unregister, so they will not hang | |
1606 | * permanently and it avoids userland to call | |
1607 | * UFFDIO_WAKE explicitly. | |
1608 | */ | |
1609 | struct userfaultfd_wake_range range; | |
1610 | range.start = start; | |
1611 | range.len = vma_end - start; | |
1612 | wake_userfault(vma->vm_userfaultfd_ctx.ctx, &range); | |
1613 | } | |
1614 | ||
7677f7fd | 1615 | new_flags = vma->vm_flags & ~__VM_UFFD_FLAGS; |
86039bd3 AA |
1616 | prev = vma_merge(mm, prev, start, vma_end, new_flags, |
1617 | vma->anon_vma, vma->vm_file, vma->vm_pgoff, | |
1618 | vma_policy(vma), | |
1619 | NULL_VM_UFFD_CTX); | |
1620 | if (prev) { | |
1621 | vma = prev; | |
1622 | goto next; | |
1623 | } | |
1624 | if (vma->vm_start < start) { | |
1625 | ret = split_vma(mm, vma, start, 1); | |
1626 | if (ret) | |
1627 | break; | |
1628 | } | |
1629 | if (vma->vm_end > end) { | |
1630 | ret = split_vma(mm, vma, end, 0); | |
1631 | if (ret) | |
1632 | break; | |
1633 | } | |
1634 | next: | |
1635 | /* | |
1636 | * In the vma_merge() successful mprotect-like case 8: | |
1637 | * the next vma was merged into the current one and | |
1638 | * the current one has not been updated yet. | |
1639 | */ | |
1640 | vma->vm_flags = new_flags; | |
1641 | vma->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX; | |
1642 | ||
1643 | skip: | |
1644 | prev = vma; | |
1645 | start = vma->vm_end; | |
1646 | vma = vma->vm_next; | |
1647 | } while (vma && vma->vm_start < end); | |
1648 | out_unlock: | |
d8ed45c5 | 1649 | mmap_write_unlock(mm); |
d2005e3f | 1650 | mmput(mm); |
86039bd3 AA |
1651 | out: |
1652 | return ret; | |
1653 | } | |
1654 | ||
1655 | /* | |
ba85c702 AA |
1656 | * userfaultfd_wake may be used in combination with the |
1657 | * UFFDIO_*_MODE_DONTWAKE to wakeup userfaults in batches. | |
86039bd3 AA |
1658 | */ |
1659 | static int userfaultfd_wake(struct userfaultfd_ctx *ctx, | |
1660 | unsigned long arg) | |
1661 | { | |
1662 | int ret; | |
1663 | struct uffdio_range uffdio_wake; | |
1664 | struct userfaultfd_wake_range range; | |
1665 | const void __user *buf = (void __user *)arg; | |
1666 | ||
1667 | ret = -EFAULT; | |
1668 | if (copy_from_user(&uffdio_wake, buf, sizeof(uffdio_wake))) | |
1669 | goto out; | |
1670 | ||
7d032574 | 1671 | ret = validate_range(ctx->mm, &uffdio_wake.start, uffdio_wake.len); |
86039bd3 AA |
1672 | if (ret) |
1673 | goto out; | |
1674 | ||
1675 | range.start = uffdio_wake.start; | |
1676 | range.len = uffdio_wake.len; | |
1677 | ||
1678 | /* | |
1679 | * len == 0 means wake all and we don't want to wake all here, | |
1680 | * so check it again to be sure. | |
1681 | */ | |
1682 | VM_BUG_ON(!range.len); | |
1683 | ||
1684 | wake_userfault(ctx, &range); | |
1685 | ret = 0; | |
1686 | ||
1687 | out: | |
1688 | return ret; | |
1689 | } | |
1690 | ||
ad465cae AA |
1691 | static int userfaultfd_copy(struct userfaultfd_ctx *ctx, |
1692 | unsigned long arg) | |
1693 | { | |
1694 | __s64 ret; | |
1695 | struct uffdio_copy uffdio_copy; | |
1696 | struct uffdio_copy __user *user_uffdio_copy; | |
1697 | struct userfaultfd_wake_range range; | |
1698 | ||
1699 | user_uffdio_copy = (struct uffdio_copy __user *) arg; | |
1700 | ||
df2cc96e MR |
1701 | ret = -EAGAIN; |
1702 | if (READ_ONCE(ctx->mmap_changing)) | |
1703 | goto out; | |
1704 | ||
ad465cae AA |
1705 | ret = -EFAULT; |
1706 | if (copy_from_user(&uffdio_copy, user_uffdio_copy, | |
1707 | /* don't copy "copy" last field */ | |
1708 | sizeof(uffdio_copy)-sizeof(__s64))) | |
1709 | goto out; | |
1710 | ||
7d032574 | 1711 | ret = validate_range(ctx->mm, &uffdio_copy.dst, uffdio_copy.len); |
ad465cae AA |
1712 | if (ret) |
1713 | goto out; | |
1714 | /* | |
1715 | * double check for wraparound just in case. copy_from_user() | |
1716 | * will later check uffdio_copy.src + uffdio_copy.len to fit | |
1717 | * in the userland range. | |
1718 | */ | |
1719 | ret = -EINVAL; | |
1720 | if (uffdio_copy.src + uffdio_copy.len <= uffdio_copy.src) | |
1721 | goto out; | |
72981e0e | 1722 | if (uffdio_copy.mode & ~(UFFDIO_COPY_MODE_DONTWAKE|UFFDIO_COPY_MODE_WP)) |
ad465cae | 1723 | goto out; |
d2005e3f ON |
1724 | if (mmget_not_zero(ctx->mm)) { |
1725 | ret = mcopy_atomic(ctx->mm, uffdio_copy.dst, uffdio_copy.src, | |
72981e0e AA |
1726 | uffdio_copy.len, &ctx->mmap_changing, |
1727 | uffdio_copy.mode); | |
d2005e3f | 1728 | mmput(ctx->mm); |
96333187 | 1729 | } else { |
e86b298b | 1730 | return -ESRCH; |
d2005e3f | 1731 | } |
ad465cae AA |
1732 | if (unlikely(put_user(ret, &user_uffdio_copy->copy))) |
1733 | return -EFAULT; | |
1734 | if (ret < 0) | |
1735 | goto out; | |
1736 | BUG_ON(!ret); | |
1737 | /* len == 0 would wake all */ | |
1738 | range.len = ret; | |
1739 | if (!(uffdio_copy.mode & UFFDIO_COPY_MODE_DONTWAKE)) { | |
1740 | range.start = uffdio_copy.dst; | |
1741 | wake_userfault(ctx, &range); | |
1742 | } | |
1743 | ret = range.len == uffdio_copy.len ? 0 : -EAGAIN; | |
1744 | out: | |
1745 | return ret; | |
1746 | } | |
1747 | ||
1748 | static int userfaultfd_zeropage(struct userfaultfd_ctx *ctx, | |
1749 | unsigned long arg) | |
1750 | { | |
1751 | __s64 ret; | |
1752 | struct uffdio_zeropage uffdio_zeropage; | |
1753 | struct uffdio_zeropage __user *user_uffdio_zeropage; | |
1754 | struct userfaultfd_wake_range range; | |
1755 | ||
1756 | user_uffdio_zeropage = (struct uffdio_zeropage __user *) arg; | |
1757 | ||
df2cc96e MR |
1758 | ret = -EAGAIN; |
1759 | if (READ_ONCE(ctx->mmap_changing)) | |
1760 | goto out; | |
1761 | ||
ad465cae AA |
1762 | ret = -EFAULT; |
1763 | if (copy_from_user(&uffdio_zeropage, user_uffdio_zeropage, | |
1764 | /* don't copy "zeropage" last field */ | |
1765 | sizeof(uffdio_zeropage)-sizeof(__s64))) | |
1766 | goto out; | |
1767 | ||
7d032574 | 1768 | ret = validate_range(ctx->mm, &uffdio_zeropage.range.start, |
ad465cae AA |
1769 | uffdio_zeropage.range.len); |
1770 | if (ret) | |
1771 | goto out; | |
1772 | ret = -EINVAL; | |
1773 | if (uffdio_zeropage.mode & ~UFFDIO_ZEROPAGE_MODE_DONTWAKE) | |
1774 | goto out; | |
1775 | ||
d2005e3f ON |
1776 | if (mmget_not_zero(ctx->mm)) { |
1777 | ret = mfill_zeropage(ctx->mm, uffdio_zeropage.range.start, | |
df2cc96e MR |
1778 | uffdio_zeropage.range.len, |
1779 | &ctx->mmap_changing); | |
d2005e3f | 1780 | mmput(ctx->mm); |
9d95aa4b | 1781 | } else { |
e86b298b | 1782 | return -ESRCH; |
d2005e3f | 1783 | } |
ad465cae AA |
1784 | if (unlikely(put_user(ret, &user_uffdio_zeropage->zeropage))) |
1785 | return -EFAULT; | |
1786 | if (ret < 0) | |
1787 | goto out; | |
1788 | /* len == 0 would wake all */ | |
1789 | BUG_ON(!ret); | |
1790 | range.len = ret; | |
1791 | if (!(uffdio_zeropage.mode & UFFDIO_ZEROPAGE_MODE_DONTWAKE)) { | |
1792 | range.start = uffdio_zeropage.range.start; | |
1793 | wake_userfault(ctx, &range); | |
1794 | } | |
1795 | ret = range.len == uffdio_zeropage.range.len ? 0 : -EAGAIN; | |
1796 | out: | |
1797 | return ret; | |
1798 | } | |
1799 | ||
63b2d417 AA |
1800 | static int userfaultfd_writeprotect(struct userfaultfd_ctx *ctx, |
1801 | unsigned long arg) | |
1802 | { | |
1803 | int ret; | |
1804 | struct uffdio_writeprotect uffdio_wp; | |
1805 | struct uffdio_writeprotect __user *user_uffdio_wp; | |
1806 | struct userfaultfd_wake_range range; | |
23080e27 | 1807 | bool mode_wp, mode_dontwake; |
63b2d417 AA |
1808 | |
1809 | if (READ_ONCE(ctx->mmap_changing)) | |
1810 | return -EAGAIN; | |
1811 | ||
1812 | user_uffdio_wp = (struct uffdio_writeprotect __user *) arg; | |
1813 | ||
1814 | if (copy_from_user(&uffdio_wp, user_uffdio_wp, | |
1815 | sizeof(struct uffdio_writeprotect))) | |
1816 | return -EFAULT; | |
1817 | ||
1818 | ret = validate_range(ctx->mm, &uffdio_wp.range.start, | |
1819 | uffdio_wp.range.len); | |
1820 | if (ret) | |
1821 | return ret; | |
1822 | ||
1823 | if (uffdio_wp.mode & ~(UFFDIO_WRITEPROTECT_MODE_DONTWAKE | | |
1824 | UFFDIO_WRITEPROTECT_MODE_WP)) | |
1825 | return -EINVAL; | |
23080e27 PX |
1826 | |
1827 | mode_wp = uffdio_wp.mode & UFFDIO_WRITEPROTECT_MODE_WP; | |
1828 | mode_dontwake = uffdio_wp.mode & UFFDIO_WRITEPROTECT_MODE_DONTWAKE; | |
1829 | ||
1830 | if (mode_wp && mode_dontwake) | |
63b2d417 AA |
1831 | return -EINVAL; |
1832 | ||
1833 | ret = mwriteprotect_range(ctx->mm, uffdio_wp.range.start, | |
23080e27 | 1834 | uffdio_wp.range.len, mode_wp, |
63b2d417 AA |
1835 | &ctx->mmap_changing); |
1836 | if (ret) | |
1837 | return ret; | |
1838 | ||
23080e27 | 1839 | if (!mode_wp && !mode_dontwake) { |
63b2d417 AA |
1840 | range.start = uffdio_wp.range.start; |
1841 | range.len = uffdio_wp.range.len; | |
1842 | wake_userfault(ctx, &range); | |
1843 | } | |
1844 | return ret; | |
1845 | } | |
1846 | ||
f6191471 AR |
1847 | static int userfaultfd_continue(struct userfaultfd_ctx *ctx, unsigned long arg) |
1848 | { | |
1849 | __s64 ret; | |
1850 | struct uffdio_continue uffdio_continue; | |
1851 | struct uffdio_continue __user *user_uffdio_continue; | |
1852 | struct userfaultfd_wake_range range; | |
1853 | ||
1854 | user_uffdio_continue = (struct uffdio_continue __user *)arg; | |
1855 | ||
1856 | ret = -EAGAIN; | |
1857 | if (READ_ONCE(ctx->mmap_changing)) | |
1858 | goto out; | |
1859 | ||
1860 | ret = -EFAULT; | |
1861 | if (copy_from_user(&uffdio_continue, user_uffdio_continue, | |
1862 | /* don't copy the output fields */ | |
1863 | sizeof(uffdio_continue) - (sizeof(__s64)))) | |
1864 | goto out; | |
1865 | ||
1866 | ret = validate_range(ctx->mm, &uffdio_continue.range.start, | |
1867 | uffdio_continue.range.len); | |
1868 | if (ret) | |
1869 | goto out; | |
1870 | ||
1871 | ret = -EINVAL; | |
1872 | /* double check for wraparound just in case. */ | |
1873 | if (uffdio_continue.range.start + uffdio_continue.range.len <= | |
1874 | uffdio_continue.range.start) { | |
1875 | goto out; | |
1876 | } | |
1877 | if (uffdio_continue.mode & ~UFFDIO_CONTINUE_MODE_DONTWAKE) | |
1878 | goto out; | |
1879 | ||
1880 | if (mmget_not_zero(ctx->mm)) { | |
1881 | ret = mcopy_continue(ctx->mm, uffdio_continue.range.start, | |
1882 | uffdio_continue.range.len, | |
1883 | &ctx->mmap_changing); | |
1884 | mmput(ctx->mm); | |
1885 | } else { | |
1886 | return -ESRCH; | |
1887 | } | |
1888 | ||
1889 | if (unlikely(put_user(ret, &user_uffdio_continue->mapped))) | |
1890 | return -EFAULT; | |
1891 | if (ret < 0) | |
1892 | goto out; | |
1893 | ||
1894 | /* len == 0 would wake all */ | |
1895 | BUG_ON(!ret); | |
1896 | range.len = ret; | |
1897 | if (!(uffdio_continue.mode & UFFDIO_CONTINUE_MODE_DONTWAKE)) { | |
1898 | range.start = uffdio_continue.range.start; | |
1899 | wake_userfault(ctx, &range); | |
1900 | } | |
1901 | ret = range.len == uffdio_continue.range.len ? 0 : -EAGAIN; | |
1902 | ||
1903 | out: | |
1904 | return ret; | |
1905 | } | |
1906 | ||
9cd75c3c PE |
1907 | static inline unsigned int uffd_ctx_features(__u64 user_features) |
1908 | { | |
1909 | /* | |
1910 | * For the current set of features the bits just coincide | |
1911 | */ | |
1912 | return (unsigned int)user_features; | |
1913 | } | |
1914 | ||
86039bd3 AA |
1915 | /* |
1916 | * userland asks for a certain API version and we return which bits | |
1917 | * and ioctl commands are implemented in this kernel for such API | |
1918 | * version or -EINVAL if unknown. | |
1919 | */ | |
1920 | static int userfaultfd_api(struct userfaultfd_ctx *ctx, | |
1921 | unsigned long arg) | |
1922 | { | |
1923 | struct uffdio_api uffdio_api; | |
1924 | void __user *buf = (void __user *)arg; | |
1925 | int ret; | |
65603144 | 1926 | __u64 features; |
86039bd3 AA |
1927 | |
1928 | ret = -EINVAL; | |
1929 | if (ctx->state != UFFD_STATE_WAIT_API) | |
1930 | goto out; | |
1931 | ret = -EFAULT; | |
a9b85f94 | 1932 | if (copy_from_user(&uffdio_api, buf, sizeof(uffdio_api))) |
86039bd3 | 1933 | goto out; |
65603144 | 1934 | features = uffdio_api.features; |
3c1c24d9 MR |
1935 | ret = -EINVAL; |
1936 | if (uffdio_api.api != UFFD_API || (features & ~UFFD_API_FEATURES)) | |
1937 | goto err_out; | |
1938 | ret = -EPERM; | |
1939 | if ((features & UFFD_FEATURE_EVENT_FORK) && !capable(CAP_SYS_PTRACE)) | |
1940 | goto err_out; | |
65603144 AA |
1941 | /* report all available features and ioctls to userland */ |
1942 | uffdio_api.features = UFFD_API_FEATURES; | |
7677f7fd AR |
1943 | #ifndef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR |
1944 | uffdio_api.features &= ~UFFD_FEATURE_MINOR_HUGETLBFS; | |
1945 | #endif | |
86039bd3 AA |
1946 | uffdio_api.ioctls = UFFD_API_IOCTLS; |
1947 | ret = -EFAULT; | |
1948 | if (copy_to_user(buf, &uffdio_api, sizeof(uffdio_api))) | |
1949 | goto out; | |
1950 | ctx->state = UFFD_STATE_RUNNING; | |
65603144 AA |
1951 | /* only enable the requested features for this uffd context */ |
1952 | ctx->features = uffd_ctx_features(features); | |
86039bd3 AA |
1953 | ret = 0; |
1954 | out: | |
1955 | return ret; | |
3c1c24d9 MR |
1956 | err_out: |
1957 | memset(&uffdio_api, 0, sizeof(uffdio_api)); | |
1958 | if (copy_to_user(buf, &uffdio_api, sizeof(uffdio_api))) | |
1959 | ret = -EFAULT; | |
1960 | goto out; | |
86039bd3 AA |
1961 | } |
1962 | ||
1963 | static long userfaultfd_ioctl(struct file *file, unsigned cmd, | |
1964 | unsigned long arg) | |
1965 | { | |
1966 | int ret = -EINVAL; | |
1967 | struct userfaultfd_ctx *ctx = file->private_data; | |
1968 | ||
e6485a47 AA |
1969 | if (cmd != UFFDIO_API && ctx->state == UFFD_STATE_WAIT_API) |
1970 | return -EINVAL; | |
1971 | ||
86039bd3 AA |
1972 | switch(cmd) { |
1973 | case UFFDIO_API: | |
1974 | ret = userfaultfd_api(ctx, arg); | |
1975 | break; | |
1976 | case UFFDIO_REGISTER: | |
1977 | ret = userfaultfd_register(ctx, arg); | |
1978 | break; | |
1979 | case UFFDIO_UNREGISTER: | |
1980 | ret = userfaultfd_unregister(ctx, arg); | |
1981 | break; | |
1982 | case UFFDIO_WAKE: | |
1983 | ret = userfaultfd_wake(ctx, arg); | |
1984 | break; | |
ad465cae AA |
1985 | case UFFDIO_COPY: |
1986 | ret = userfaultfd_copy(ctx, arg); | |
1987 | break; | |
1988 | case UFFDIO_ZEROPAGE: | |
1989 | ret = userfaultfd_zeropage(ctx, arg); | |
1990 | break; | |
63b2d417 AA |
1991 | case UFFDIO_WRITEPROTECT: |
1992 | ret = userfaultfd_writeprotect(ctx, arg); | |
1993 | break; | |
f6191471 AR |
1994 | case UFFDIO_CONTINUE: |
1995 | ret = userfaultfd_continue(ctx, arg); | |
1996 | break; | |
86039bd3 AA |
1997 | } |
1998 | return ret; | |
1999 | } | |
2000 | ||
2001 | #ifdef CONFIG_PROC_FS | |
2002 | static void userfaultfd_show_fdinfo(struct seq_file *m, struct file *f) | |
2003 | { | |
2004 | struct userfaultfd_ctx *ctx = f->private_data; | |
ac6424b9 | 2005 | wait_queue_entry_t *wq; |
86039bd3 AA |
2006 | unsigned long pending = 0, total = 0; |
2007 | ||
cbcfa130 | 2008 | spin_lock_irq(&ctx->fault_pending_wqh.lock); |
2055da97 | 2009 | list_for_each_entry(wq, &ctx->fault_pending_wqh.head, entry) { |
15b726ef AA |
2010 | pending++; |
2011 | total++; | |
2012 | } | |
2055da97 | 2013 | list_for_each_entry(wq, &ctx->fault_wqh.head, entry) { |
86039bd3 AA |
2014 | total++; |
2015 | } | |
cbcfa130 | 2016 | spin_unlock_irq(&ctx->fault_pending_wqh.lock); |
86039bd3 AA |
2017 | |
2018 | /* | |
2019 | * If more protocols will be added, there will be all shown | |
2020 | * separated by a space. Like this: | |
2021 | * protocols: aa:... bb:... | |
2022 | */ | |
2023 | seq_printf(m, "pending:\t%lu\ntotal:\t%lu\nAPI:\t%Lx:%x:%Lx\n", | |
045098e9 | 2024 | pending, total, UFFD_API, ctx->features, |
86039bd3 AA |
2025 | UFFD_API_IOCTLS|UFFD_API_RANGE_IOCTLS); |
2026 | } | |
2027 | #endif | |
2028 | ||
2029 | static const struct file_operations userfaultfd_fops = { | |
2030 | #ifdef CONFIG_PROC_FS | |
2031 | .show_fdinfo = userfaultfd_show_fdinfo, | |
2032 | #endif | |
2033 | .release = userfaultfd_release, | |
2034 | .poll = userfaultfd_poll, | |
2035 | .read = userfaultfd_read, | |
2036 | .unlocked_ioctl = userfaultfd_ioctl, | |
1832f2d8 | 2037 | .compat_ioctl = compat_ptr_ioctl, |
86039bd3 AA |
2038 | .llseek = noop_llseek, |
2039 | }; | |
2040 | ||
3004ec9c AA |
2041 | static void init_once_userfaultfd_ctx(void *mem) |
2042 | { | |
2043 | struct userfaultfd_ctx *ctx = (struct userfaultfd_ctx *) mem; | |
2044 | ||
2045 | init_waitqueue_head(&ctx->fault_pending_wqh); | |
2046 | init_waitqueue_head(&ctx->fault_wqh); | |
9cd75c3c | 2047 | init_waitqueue_head(&ctx->event_wqh); |
3004ec9c | 2048 | init_waitqueue_head(&ctx->fd_wqh); |
2ca97ac8 | 2049 | seqcount_spinlock_init(&ctx->refile_seq, &ctx->fault_pending_wqh.lock); |
3004ec9c AA |
2050 | } |
2051 | ||
284cd241 | 2052 | SYSCALL_DEFINE1(userfaultfd, int, flags) |
86039bd3 | 2053 | { |
86039bd3 | 2054 | struct userfaultfd_ctx *ctx; |
284cd241 | 2055 | int fd; |
86039bd3 | 2056 | |
d0d4730a LG |
2057 | if (!sysctl_unprivileged_userfaultfd && |
2058 | (flags & UFFD_USER_MODE_ONLY) == 0 && | |
2059 | !capable(CAP_SYS_PTRACE)) { | |
2060 | printk_once(KERN_WARNING "uffd: Set unprivileged_userfaultfd " | |
2061 | "sysctl knob to 1 if kernel faults must be handled " | |
2062 | "without obtaining CAP_SYS_PTRACE capability\n"); | |
cefdca0a | 2063 | return -EPERM; |
d0d4730a | 2064 | } |
cefdca0a | 2065 | |
86039bd3 AA |
2066 | BUG_ON(!current->mm); |
2067 | ||
2068 | /* Check the UFFD_* constants for consistency. */ | |
37cd0575 | 2069 | BUILD_BUG_ON(UFFD_USER_MODE_ONLY & UFFD_SHARED_FCNTL_FLAGS); |
86039bd3 AA |
2070 | BUILD_BUG_ON(UFFD_CLOEXEC != O_CLOEXEC); |
2071 | BUILD_BUG_ON(UFFD_NONBLOCK != O_NONBLOCK); | |
2072 | ||
37cd0575 | 2073 | if (flags & ~(UFFD_SHARED_FCNTL_FLAGS | UFFD_USER_MODE_ONLY)) |
284cd241 | 2074 | return -EINVAL; |
86039bd3 | 2075 | |
3004ec9c | 2076 | ctx = kmem_cache_alloc(userfaultfd_ctx_cachep, GFP_KERNEL); |
86039bd3 | 2077 | if (!ctx) |
284cd241 | 2078 | return -ENOMEM; |
86039bd3 | 2079 | |
ca880420 | 2080 | refcount_set(&ctx->refcount, 1); |
86039bd3 | 2081 | ctx->flags = flags; |
9cd75c3c | 2082 | ctx->features = 0; |
86039bd3 AA |
2083 | ctx->state = UFFD_STATE_WAIT_API; |
2084 | ctx->released = false; | |
df2cc96e | 2085 | ctx->mmap_changing = false; |
86039bd3 AA |
2086 | ctx->mm = current->mm; |
2087 | /* prevent the mm struct to be freed */ | |
f1f10076 | 2088 | mmgrab(ctx->mm); |
86039bd3 | 2089 | |
b537900f DC |
2090 | fd = anon_inode_getfd_secure("[userfaultfd]", &userfaultfd_fops, ctx, |
2091 | O_RDWR | (flags & UFFD_SHARED_FCNTL_FLAGS), NULL); | |
284cd241 | 2092 | if (fd < 0) { |
d2005e3f | 2093 | mmdrop(ctx->mm); |
3004ec9c | 2094 | kmem_cache_free(userfaultfd_ctx_cachep, ctx); |
c03e946f | 2095 | } |
86039bd3 | 2096 | return fd; |
86039bd3 | 2097 | } |
3004ec9c AA |
2098 | |
2099 | static int __init userfaultfd_init(void) | |
2100 | { | |
2101 | userfaultfd_ctx_cachep = kmem_cache_create("userfaultfd_ctx_cache", | |
2102 | sizeof(struct userfaultfd_ctx), | |
2103 | 0, | |
2104 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, | |
2105 | init_once_userfaultfd_ctx); | |
2106 | return 0; | |
2107 | } | |
2108 | __initcall(userfaultfd_init); |