Commit | Line | Data |
---|---|---|
720e596a | 1 | // SPDX-License-Identifier: GPL-2.0+ |
2b144498 | 2 | /* |
7b2d81d4 | 3 | * User-space Probes (UProbes) |
2b144498 | 4 | * |
35aa621b | 5 | * Copyright (C) IBM Corporation, 2008-2012 |
2b144498 SD |
6 | * Authors: |
7 | * Srikar Dronamraju | |
8 | * Jim Keniston | |
90eec103 | 9 | * Copyright (C) 2011-2012 Red Hat, Inc., Peter Zijlstra |
2b144498 SD |
10 | */ |
11 | ||
12 | #include <linux/kernel.h> | |
13 | #include <linux/highmem.h> | |
14 | #include <linux/pagemap.h> /* read_mapping_page */ | |
15 | #include <linux/slab.h> | |
16 | #include <linux/sched.h> | |
6e84f315 | 17 | #include <linux/sched/mm.h> |
f7ccbae4 | 18 | #include <linux/sched/coredump.h> |
e8440c14 | 19 | #include <linux/export.h> |
2b144498 SD |
20 | #include <linux/rmap.h> /* anon_vma_prepare */ |
21 | #include <linux/mmu_notifier.h> /* set_pte_at_notify */ | |
22 | #include <linux/swap.h> /* try_to_free_swap */ | |
0326f5a9 SD |
23 | #include <linux/ptrace.h> /* user_enable_single_step */ |
24 | #include <linux/kdebug.h> /* notifier mechanism */ | |
194f8dcb | 25 | #include "../../mm/internal.h" /* munlock_vma_page */ |
32cdba1e | 26 | #include <linux/percpu-rwsem.h> |
aa59c53f | 27 | #include <linux/task_work.h> |
40814f68 | 28 | #include <linux/shmem_fs.h> |
f385cb85 | 29 | #include <linux/khugepaged.h> |
7b2d81d4 | 30 | |
2b144498 SD |
31 | #include <linux/uprobes.h> |
32 | ||
d4b3b638 SD |
33 | #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES) |
34 | #define MAX_UPROBE_XOL_SLOTS UINSNS_PER_PAGE | |
35 | ||
2b144498 | 36 | static struct rb_root uprobes_tree = RB_ROOT; |
441f1eb7 ON |
37 | /* |
38 | * allows us to skip the uprobe_mmap if there are no uprobe events active | |
39 | * at this time. Probably a fine grained per inode count is better? | |
40 | */ | |
41 | #define no_uprobe_events() RB_EMPTY_ROOT(&uprobes_tree) | |
7b2d81d4 | 42 | |
2b144498 SD |
43 | static DEFINE_SPINLOCK(uprobes_treelock); /* serialize rbtree access */ |
44 | ||
45 | #define UPROBES_HASH_SZ 13 | |
2b144498 SD |
46 | /* serialize uprobe->pending_list */ |
47 | static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ]; | |
7b2d81d4 | 48 | #define uprobes_mmap_hash(v) (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ]) |
2b144498 | 49 | |
2bf1acc2 | 50 | DEFINE_STATIC_PERCPU_RWSEM(dup_mmap_sem); |
32cdba1e | 51 | |
cb9a19fe | 52 | /* Have a copy of original instruction */ |
71434f2f | 53 | #define UPROBE_COPY_INSN 0 |
cb9a19fe | 54 | |
3ff54efd SD |
55 | struct uprobe { |
56 | struct rb_node rb_node; /* node in the rb tree */ | |
ce59b8e9 | 57 | refcount_t ref; |
e591c8d7 | 58 | struct rw_semaphore register_rwsem; |
3ff54efd SD |
59 | struct rw_semaphore consumer_rwsem; |
60 | struct list_head pending_list; | |
61 | struct uprobe_consumer *consumers; | |
62 | struct inode *inode; /* Also hold a ref to inode */ | |
63 | loff_t offset; | |
1cc33161 | 64 | loff_t ref_ctr_offset; |
71434f2f | 65 | unsigned long flags; |
ad439356 ON |
66 | |
67 | /* | |
68 | * The generic code assumes that it has two members of unknown type | |
69 | * owned by the arch-specific code: | |
70 | * | |
71 | * insn - copy_insn() saves the original instruction here for | |
72 | * arch_uprobe_analyze_insn(). | |
73 | * | |
74 | * ixol - potentially modified instruction to execute out of | |
75 | * line, copied to xol_area by xol_get_insn_slot(). | |
76 | */ | |
3ff54efd SD |
77 | struct arch_uprobe arch; |
78 | }; | |
79 | ||
1cc33161 RB |
80 | struct delayed_uprobe { |
81 | struct list_head list; | |
82 | struct uprobe *uprobe; | |
83 | struct mm_struct *mm; | |
84 | }; | |
85 | ||
86 | static DEFINE_MUTEX(delayed_uprobe_lock); | |
87 | static LIST_HEAD(delayed_uprobe_list); | |
88 | ||
c912dae6 | 89 | /* |
ad439356 ON |
90 | * Execute out of line area: anonymous executable mapping installed |
91 | * by the probed task to execute the copy of the original instruction | |
92 | * mangled by set_swbp(). | |
93 | * | |
c912dae6 ON |
94 | * On a breakpoint hit, thread contests for a slot. It frees the |
95 | * slot after singlestep. Currently a fixed number of slots are | |
96 | * allocated. | |
97 | */ | |
98 | struct xol_area { | |
704bde3c ON |
99 | wait_queue_head_t wq; /* if all slots are busy */ |
100 | atomic_t slot_count; /* number of in-use slots */ | |
101 | unsigned long *bitmap; /* 0 = free slot */ | |
c912dae6 | 102 | |
704bde3c ON |
103 | struct vm_special_mapping xol_mapping; |
104 | struct page *pages[2]; | |
c912dae6 ON |
105 | /* |
106 | * We keep the vma's vm_start rather than a pointer to the vma | |
107 | * itself. The probed process or a naughty kernel module could make | |
108 | * the vma go away, and we must handle that reasonably gracefully. | |
109 | */ | |
704bde3c | 110 | unsigned long vaddr; /* Page(s) of instruction slots */ |
c912dae6 ON |
111 | }; |
112 | ||
2b144498 SD |
113 | /* |
114 | * valid_vma: Verify if the specified vma is an executable vma | |
115 | * Relax restrictions while unregistering: vm_flags might have | |
116 | * changed after breakpoint was inserted. | |
117 | * - is_register: indicates if we are in register context. | |
118 | * - Return 1 if the specified virtual address is in an | |
119 | * executable vma. | |
120 | */ | |
121 | static bool valid_vma(struct vm_area_struct *vma, bool is_register) | |
122 | { | |
13f59c5e | 123 | vm_flags_t flags = VM_HUGETLB | VM_MAYEXEC | VM_MAYSHARE; |
2b144498 | 124 | |
e40cfce6 ON |
125 | if (is_register) |
126 | flags |= VM_WRITE; | |
2b144498 | 127 | |
e40cfce6 | 128 | return vma->vm_file && (vma->vm_flags & flags) == VM_MAYEXEC; |
2b144498 SD |
129 | } |
130 | ||
57683f72 | 131 | static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset) |
2b144498 | 132 | { |
57683f72 | 133 | return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT); |
2b144498 SD |
134 | } |
135 | ||
cb113b47 ON |
136 | static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr) |
137 | { | |
138 | return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start); | |
139 | } | |
140 | ||
2b144498 SD |
141 | /** |
142 | * __replace_page - replace page in vma by new page. | |
143 | * based on replace_page in mm/ksm.c | |
144 | * | |
145 | * @vma: vma that holds the pte pointing to page | |
c517ee74 | 146 | * @addr: address the old @page is mapped at |
fb4fb04f SL |
147 | * @old_page: the page we are replacing by new_page |
148 | * @new_page: the modified page we replace page by | |
2b144498 | 149 | * |
fb4fb04f SL |
150 | * If @new_page is NULL, only unmap @old_page. |
151 | * | |
152 | * Returns 0 on success, negative error code otherwise. | |
2b144498 | 153 | */ |
c517ee74 | 154 | static int __replace_page(struct vm_area_struct *vma, unsigned long addr, |
bdfaa2ee | 155 | struct page *old_page, struct page *new_page) |
2b144498 SD |
156 | { |
157 | struct mm_struct *mm = vma->vm_mm; | |
14fa2daa | 158 | struct page_vma_mapped_walk pvmw = { |
5a52c9df | 159 | .page = compound_head(old_page), |
14fa2daa KS |
160 | .vma = vma, |
161 | .address = addr, | |
162 | }; | |
9f92448c | 163 | int err; |
ac46d4f3 | 164 | struct mmu_notifier_range range; |
00501b53 | 165 | |
7269f999 | 166 | mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, mm, addr, |
6f4f13e8 | 167 | addr + PAGE_SIZE); |
ac46d4f3 | 168 | |
fb4fb04f | 169 | if (new_page) { |
d9eb1ea2 | 170 | err = mem_cgroup_charge(new_page, vma->vm_mm, GFP_KERNEL); |
fb4fb04f SL |
171 | if (err) |
172 | return err; | |
173 | } | |
2b144498 | 174 | |
194f8dcb | 175 | /* For try_to_free_swap() and munlock_vma_page() below */ |
bdfaa2ee | 176 | lock_page(old_page); |
9f92448c | 177 | |
ac46d4f3 | 178 | mmu_notifier_invalidate_range_start(&range); |
9f92448c | 179 | err = -EAGAIN; |
9d82c694 | 180 | if (!page_vma_mapped_walk(&pvmw)) |
9f92448c | 181 | goto unlock; |
14fa2daa | 182 | VM_BUG_ON_PAGE(addr != pvmw.address, old_page); |
2b144498 | 183 | |
fb4fb04f SL |
184 | if (new_page) { |
185 | get_page(new_page); | |
be5d0a74 | 186 | page_add_new_anon_rmap(new_page, vma, addr, false); |
b518154e | 187 | lru_cache_add_inactive_or_unevictable(new_page, vma); |
fb4fb04f SL |
188 | } else |
189 | /* no new page, just dec_mm_counter for old_page */ | |
190 | dec_mm_counter(mm, MM_ANONPAGES); | |
2b144498 | 191 | |
bdfaa2ee ON |
192 | if (!PageAnon(old_page)) { |
193 | dec_mm_counter(mm, mm_counter_file(old_page)); | |
7396fa81 SD |
194 | inc_mm_counter(mm, MM_ANONPAGES); |
195 | } | |
196 | ||
14fa2daa KS |
197 | flush_cache_page(vma, addr, pte_pfn(*pvmw.pte)); |
198 | ptep_clear_flush_notify(vma, addr, pvmw.pte); | |
fb4fb04f SL |
199 | if (new_page) |
200 | set_pte_at_notify(mm, addr, pvmw.pte, | |
201 | mk_pte(new_page, vma->vm_page_prot)); | |
2b144498 | 202 | |
bdfaa2ee ON |
203 | page_remove_rmap(old_page, false); |
204 | if (!page_mapped(old_page)) | |
205 | try_to_free_swap(old_page); | |
14fa2daa | 206 | page_vma_mapped_walk_done(&pvmw); |
2b144498 | 207 | |
194f8dcb | 208 | if (vma->vm_flags & VM_LOCKED) |
bdfaa2ee ON |
209 | munlock_vma_page(old_page); |
210 | put_page(old_page); | |
194f8dcb | 211 | |
9f92448c ON |
212 | err = 0; |
213 | unlock: | |
ac46d4f3 | 214 | mmu_notifier_invalidate_range_end(&range); |
bdfaa2ee | 215 | unlock_page(old_page); |
9f92448c | 216 | return err; |
2b144498 SD |
217 | } |
218 | ||
219 | /** | |
5cb4ac3a | 220 | * is_swbp_insn - check if instruction is breakpoint instruction. |
2b144498 | 221 | * @insn: instruction to be checked. |
5cb4ac3a | 222 | * Default implementation of is_swbp_insn |
2b144498 SD |
223 | * Returns true if @insn is a breakpoint instruction. |
224 | */ | |
5cb4ac3a | 225 | bool __weak is_swbp_insn(uprobe_opcode_t *insn) |
2b144498 | 226 | { |
5cb4ac3a | 227 | return *insn == UPROBE_SWBP_INSN; |
2b144498 SD |
228 | } |
229 | ||
0908ad6e AM |
230 | /** |
231 | * is_trap_insn - check if instruction is breakpoint instruction. | |
232 | * @insn: instruction to be checked. | |
233 | * Default implementation of is_trap_insn | |
234 | * Returns true if @insn is a breakpoint instruction. | |
235 | * | |
236 | * This function is needed for the case where an architecture has multiple | |
237 | * trap instructions (like powerpc). | |
238 | */ | |
239 | bool __weak is_trap_insn(uprobe_opcode_t *insn) | |
240 | { | |
241 | return is_swbp_insn(insn); | |
242 | } | |
243 | ||
ab0d805c | 244 | static void copy_from_page(struct page *page, unsigned long vaddr, void *dst, int len) |
cceb55aa ON |
245 | { |
246 | void *kaddr = kmap_atomic(page); | |
ab0d805c | 247 | memcpy(dst, kaddr + (vaddr & ~PAGE_MASK), len); |
cceb55aa ON |
248 | kunmap_atomic(kaddr); |
249 | } | |
250 | ||
5669ccee ON |
251 | static void copy_to_page(struct page *page, unsigned long vaddr, const void *src, int len) |
252 | { | |
253 | void *kaddr = kmap_atomic(page); | |
254 | memcpy(kaddr + (vaddr & ~PAGE_MASK), src, len); | |
255 | kunmap_atomic(kaddr); | |
256 | } | |
257 | ||
ed6f6a50 ON |
258 | static int verify_opcode(struct page *page, unsigned long vaddr, uprobe_opcode_t *new_opcode) |
259 | { | |
260 | uprobe_opcode_t old_opcode; | |
261 | bool is_swbp; | |
262 | ||
0908ad6e AM |
263 | /* |
264 | * Note: We only check if the old_opcode is UPROBE_SWBP_INSN here. | |
265 | * We do not check if it is any other 'trap variant' which could | |
266 | * be conditional trap instruction such as the one powerpc supports. | |
267 | * | |
268 | * The logic is that we do not care if the underlying instruction | |
269 | * is a trap variant; uprobes always wins over any other (gdb) | |
270 | * breakpoint. | |
271 | */ | |
ab0d805c | 272 | copy_from_page(page, vaddr, &old_opcode, UPROBE_SWBP_INSN_SIZE); |
ed6f6a50 ON |
273 | is_swbp = is_swbp_insn(&old_opcode); |
274 | ||
275 | if (is_swbp_insn(new_opcode)) { | |
276 | if (is_swbp) /* register: already installed? */ | |
277 | return 0; | |
278 | } else { | |
279 | if (!is_swbp) /* unregister: was it changed by us? */ | |
076a365b | 280 | return 0; |
ed6f6a50 ON |
281 | } |
282 | ||
283 | return 1; | |
284 | } | |
285 | ||
1cc33161 RB |
286 | static struct delayed_uprobe * |
287 | delayed_uprobe_check(struct uprobe *uprobe, struct mm_struct *mm) | |
288 | { | |
289 | struct delayed_uprobe *du; | |
290 | ||
291 | list_for_each_entry(du, &delayed_uprobe_list, list) | |
292 | if (du->uprobe == uprobe && du->mm == mm) | |
293 | return du; | |
294 | return NULL; | |
295 | } | |
296 | ||
297 | static int delayed_uprobe_add(struct uprobe *uprobe, struct mm_struct *mm) | |
298 | { | |
299 | struct delayed_uprobe *du; | |
300 | ||
301 | if (delayed_uprobe_check(uprobe, mm)) | |
302 | return 0; | |
303 | ||
304 | du = kzalloc(sizeof(*du), GFP_KERNEL); | |
305 | if (!du) | |
306 | return -ENOMEM; | |
307 | ||
308 | du->uprobe = uprobe; | |
309 | du->mm = mm; | |
310 | list_add(&du->list, &delayed_uprobe_list); | |
311 | return 0; | |
312 | } | |
313 | ||
314 | static void delayed_uprobe_delete(struct delayed_uprobe *du) | |
315 | { | |
316 | if (WARN_ON(!du)) | |
317 | return; | |
318 | list_del(&du->list); | |
319 | kfree(du); | |
320 | } | |
321 | ||
322 | static void delayed_uprobe_remove(struct uprobe *uprobe, struct mm_struct *mm) | |
323 | { | |
324 | struct list_head *pos, *q; | |
325 | struct delayed_uprobe *du; | |
326 | ||
327 | if (!uprobe && !mm) | |
328 | return; | |
329 | ||
330 | list_for_each_safe(pos, q, &delayed_uprobe_list) { | |
331 | du = list_entry(pos, struct delayed_uprobe, list); | |
332 | ||
333 | if (uprobe && du->uprobe != uprobe) | |
334 | continue; | |
335 | if (mm && du->mm != mm) | |
336 | continue; | |
337 | ||
338 | delayed_uprobe_delete(du); | |
339 | } | |
340 | } | |
341 | ||
342 | static bool valid_ref_ctr_vma(struct uprobe *uprobe, | |
343 | struct vm_area_struct *vma) | |
344 | { | |
345 | unsigned long vaddr = offset_to_vaddr(vma, uprobe->ref_ctr_offset); | |
346 | ||
347 | return uprobe->ref_ctr_offset && | |
348 | vma->vm_file && | |
349 | file_inode(vma->vm_file) == uprobe->inode && | |
350 | (vma->vm_flags & (VM_WRITE|VM_SHARED)) == VM_WRITE && | |
351 | vma->vm_start <= vaddr && | |
352 | vma->vm_end > vaddr; | |
353 | } | |
354 | ||
355 | static struct vm_area_struct * | |
356 | find_ref_ctr_vma(struct uprobe *uprobe, struct mm_struct *mm) | |
357 | { | |
358 | struct vm_area_struct *tmp; | |
359 | ||
360 | for (tmp = mm->mmap; tmp; tmp = tmp->vm_next) | |
361 | if (valid_ref_ctr_vma(uprobe, tmp)) | |
362 | return tmp; | |
363 | ||
364 | return NULL; | |
365 | } | |
366 | ||
367 | static int | |
368 | __update_ref_ctr(struct mm_struct *mm, unsigned long vaddr, short d) | |
369 | { | |
370 | void *kaddr; | |
371 | struct page *page; | |
372 | struct vm_area_struct *vma; | |
373 | int ret; | |
374 | short *ptr; | |
375 | ||
376 | if (!vaddr || !d) | |
377 | return -EINVAL; | |
378 | ||
379 | ret = get_user_pages_remote(NULL, mm, vaddr, 1, | |
380 | FOLL_WRITE, &page, &vma, NULL); | |
381 | if (unlikely(ret <= 0)) { | |
382 | /* | |
383 | * We are asking for 1 page. If get_user_pages_remote() fails, | |
384 | * it may return 0, in that case we have to return error. | |
385 | */ | |
386 | return ret == 0 ? -EBUSY : ret; | |
387 | } | |
388 | ||
389 | kaddr = kmap_atomic(page); | |
390 | ptr = kaddr + (vaddr & ~PAGE_MASK); | |
391 | ||
392 | if (unlikely(*ptr + d < 0)) { | |
393 | pr_warn("ref_ctr going negative. vaddr: 0x%lx, " | |
394 | "curr val: %d, delta: %d\n", vaddr, *ptr, d); | |
395 | ret = -EINVAL; | |
396 | goto out; | |
397 | } | |
398 | ||
399 | *ptr += d; | |
400 | ret = 0; | |
401 | out: | |
402 | kunmap_atomic(kaddr); | |
403 | put_page(page); | |
404 | return ret; | |
405 | } | |
406 | ||
407 | static void update_ref_ctr_warn(struct uprobe *uprobe, | |
408 | struct mm_struct *mm, short d) | |
409 | { | |
410 | pr_warn("ref_ctr %s failed for inode: 0x%lx offset: " | |
411 | "0x%llx ref_ctr_offset: 0x%llx of mm: 0x%pK\n", | |
412 | d > 0 ? "increment" : "decrement", uprobe->inode->i_ino, | |
413 | (unsigned long long) uprobe->offset, | |
414 | (unsigned long long) uprobe->ref_ctr_offset, mm); | |
415 | } | |
416 | ||
417 | static int update_ref_ctr(struct uprobe *uprobe, struct mm_struct *mm, | |
418 | short d) | |
419 | { | |
420 | struct vm_area_struct *rc_vma; | |
421 | unsigned long rc_vaddr; | |
422 | int ret = 0; | |
423 | ||
424 | rc_vma = find_ref_ctr_vma(uprobe, mm); | |
425 | ||
426 | if (rc_vma) { | |
427 | rc_vaddr = offset_to_vaddr(rc_vma, uprobe->ref_ctr_offset); | |
428 | ret = __update_ref_ctr(mm, rc_vaddr, d); | |
429 | if (ret) | |
430 | update_ref_ctr_warn(uprobe, mm, d); | |
431 | ||
432 | if (d > 0) | |
433 | return ret; | |
434 | } | |
435 | ||
436 | mutex_lock(&delayed_uprobe_lock); | |
437 | if (d > 0) | |
438 | ret = delayed_uprobe_add(uprobe, mm); | |
439 | else | |
440 | delayed_uprobe_remove(uprobe, mm); | |
441 | mutex_unlock(&delayed_uprobe_lock); | |
442 | ||
443 | return ret; | |
444 | } | |
445 | ||
2b144498 SD |
446 | /* |
447 | * NOTE: | |
448 | * Expect the breakpoint instruction to be the smallest size instruction for | |
449 | * the architecture. If an arch has variable length instruction and the | |
450 | * breakpoint instruction is not of the smallest length instruction | |
0908ad6e | 451 | * supported by that architecture then we need to modify is_trap_at_addr and |
f72d41fa ON |
452 | * uprobe_write_opcode accordingly. This would never be a problem for archs |
453 | * that have fixed length instructions. | |
29dedee0 | 454 | * |
f72d41fa | 455 | * uprobe_write_opcode - write the opcode at a given virtual address. |
2b144498 | 456 | * @mm: the probed process address space. |
2b144498 SD |
457 | * @vaddr: the virtual address to store the opcode. |
458 | * @opcode: opcode to be written at @vaddr. | |
459 | * | |
c1e8d7c6 | 460 | * Called with mm->mmap_lock held for write. |
2b144498 SD |
461 | * Return 0 (success) or a negative errno. |
462 | */ | |
6d43743e RB |
463 | int uprobe_write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm, |
464 | unsigned long vaddr, uprobe_opcode_t opcode) | |
2b144498 | 465 | { |
1cc33161 | 466 | struct uprobe *uprobe; |
2b144498 | 467 | struct page *old_page, *new_page; |
2b144498 | 468 | struct vm_area_struct *vma; |
1cc33161 | 469 | int ret, is_register, ref_ctr_updated = 0; |
f385cb85 | 470 | bool orig_page_huge = false; |
aa5de305 | 471 | unsigned int gup_flags = FOLL_FORCE; |
1cc33161 RB |
472 | |
473 | is_register = is_swbp_insn(&opcode); | |
474 | uprobe = container_of(auprobe, struct uprobe, arch); | |
f403072c | 475 | |
5323ce71 | 476 | retry: |
aa5de305 SL |
477 | if (is_register) |
478 | gup_flags |= FOLL_SPLIT_PMD; | |
2b144498 | 479 | /* Read the page with vaddr into memory */ |
aa5de305 SL |
480 | ret = get_user_pages_remote(NULL, mm, vaddr, 1, gup_flags, |
481 | &old_page, &vma, NULL); | |
2b144498 SD |
482 | if (ret <= 0) |
483 | return ret; | |
7b2d81d4 | 484 | |
ed6f6a50 ON |
485 | ret = verify_opcode(old_page, vaddr, &opcode); |
486 | if (ret <= 0) | |
487 | goto put_old; | |
488 | ||
aa5de305 SL |
489 | if (WARN(!is_register && PageCompound(old_page), |
490 | "uprobe unregister should never work on compound page\n")) { | |
491 | ret = -EINVAL; | |
492 | goto put_old; | |
493 | } | |
494 | ||
1cc33161 RB |
495 | /* We are going to replace instruction, update ref_ctr. */ |
496 | if (!ref_ctr_updated && uprobe->ref_ctr_offset) { | |
497 | ret = update_ref_ctr(uprobe, mm, is_register ? 1 : -1); | |
498 | if (ret) | |
499 | goto put_old; | |
500 | ||
501 | ref_ctr_updated = 1; | |
502 | } | |
503 | ||
fb4fb04f SL |
504 | ret = 0; |
505 | if (!is_register && !PageAnon(old_page)) | |
506 | goto put_old; | |
507 | ||
29dedee0 ON |
508 | ret = anon_vma_prepare(vma); |
509 | if (ret) | |
510 | goto put_old; | |
511 | ||
2b144498 SD |
512 | ret = -ENOMEM; |
513 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr); | |
514 | if (!new_page) | |
9f92448c | 515 | goto put_old; |
2b144498 | 516 | |
29dedee0 | 517 | __SetPageUptodate(new_page); |
3f47107c ON |
518 | copy_highpage(new_page, old_page); |
519 | copy_to_page(new_page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE); | |
2b144498 | 520 | |
fb4fb04f SL |
521 | if (!is_register) { |
522 | struct page *orig_page; | |
523 | pgoff_t index; | |
524 | ||
525 | VM_BUG_ON_PAGE(!PageAnon(old_page), old_page); | |
526 | ||
527 | index = vaddr_to_offset(vma, vaddr & PAGE_MASK) >> PAGE_SHIFT; | |
528 | orig_page = find_get_page(vma->vm_file->f_inode->i_mapping, | |
529 | index); | |
530 | ||
531 | if (orig_page) { | |
532 | if (PageUptodate(orig_page) && | |
533 | pages_identical(new_page, orig_page)) { | |
534 | /* let go new_page */ | |
535 | put_page(new_page); | |
536 | new_page = NULL; | |
f385cb85 SL |
537 | |
538 | if (PageCompound(orig_page)) | |
539 | orig_page_huge = true; | |
fb4fb04f SL |
540 | } |
541 | put_page(orig_page); | |
542 | } | |
543 | } | |
544 | ||
c517ee74 | 545 | ret = __replace_page(vma, vaddr, old_page, new_page); |
fb4fb04f SL |
546 | if (new_page) |
547 | put_page(new_page); | |
9f92448c | 548 | put_old: |
7b2d81d4 IM |
549 | put_page(old_page); |
550 | ||
5323ce71 ON |
551 | if (unlikely(ret == -EAGAIN)) |
552 | goto retry; | |
1cc33161 RB |
553 | |
554 | /* Revert back reference counter if instruction update failed. */ | |
555 | if (ret && is_register && ref_ctr_updated) | |
556 | update_ref_ctr(uprobe, mm, -1); | |
557 | ||
f385cb85 SL |
558 | /* try collapse pmd for compound page */ |
559 | if (!ret && orig_page_huge) | |
560 | collapse_pte_mapped_thp(mm, vaddr); | |
561 | ||
2b144498 SD |
562 | return ret; |
563 | } | |
564 | ||
2b144498 | 565 | /** |
5cb4ac3a | 566 | * set_swbp - store breakpoint at a given address. |
e3343e6a | 567 | * @auprobe: arch specific probepoint information. |
2b144498 | 568 | * @mm: the probed process address space. |
2b144498 SD |
569 | * @vaddr: the virtual address to insert the opcode. |
570 | * | |
571 | * For mm @mm, store the breakpoint instruction at @vaddr. | |
572 | * Return 0 (success) or a negative errno. | |
573 | */ | |
5cb4ac3a | 574 | int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 575 | { |
6d43743e | 576 | return uprobe_write_opcode(auprobe, mm, vaddr, UPROBE_SWBP_INSN); |
2b144498 SD |
577 | } |
578 | ||
579 | /** | |
580 | * set_orig_insn - Restore the original instruction. | |
581 | * @mm: the probed process address space. | |
e3343e6a | 582 | * @auprobe: arch specific probepoint information. |
2b144498 | 583 | * @vaddr: the virtual address to insert the opcode. |
2b144498 SD |
584 | * |
585 | * For mm @mm, restore the original opcode (opcode) at @vaddr. | |
586 | * Return 0 (success) or a negative errno. | |
587 | */ | |
7b2d81d4 | 588 | int __weak |
ded86e7c | 589 | set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 590 | { |
6d43743e RB |
591 | return uprobe_write_opcode(auprobe, mm, vaddr, |
592 | *(uprobe_opcode_t *)&auprobe->insn); | |
2b144498 SD |
593 | } |
594 | ||
f231722a ON |
595 | static struct uprobe *get_uprobe(struct uprobe *uprobe) |
596 | { | |
ce59b8e9 | 597 | refcount_inc(&uprobe->ref); |
f231722a ON |
598 | return uprobe; |
599 | } | |
600 | ||
601 | static void put_uprobe(struct uprobe *uprobe) | |
602 | { | |
ce59b8e9 | 603 | if (refcount_dec_and_test(&uprobe->ref)) { |
1cc33161 RB |
604 | /* |
605 | * If application munmap(exec_vma) before uprobe_unregister() | |
606 | * gets called, we don't get a chance to remove uprobe from | |
607 | * delayed_uprobe_list from remove_breakpoint(). Do it here. | |
608 | */ | |
1aed58e6 | 609 | mutex_lock(&delayed_uprobe_lock); |
1cc33161 | 610 | delayed_uprobe_remove(uprobe, NULL); |
1aed58e6 | 611 | mutex_unlock(&delayed_uprobe_lock); |
f231722a | 612 | kfree(uprobe); |
1cc33161 | 613 | } |
f231722a ON |
614 | } |
615 | ||
2b144498 SD |
616 | static int match_uprobe(struct uprobe *l, struct uprobe *r) |
617 | { | |
618 | if (l->inode < r->inode) | |
619 | return -1; | |
7b2d81d4 | 620 | |
2b144498 SD |
621 | if (l->inode > r->inode) |
622 | return 1; | |
2b144498 | 623 | |
7b2d81d4 IM |
624 | if (l->offset < r->offset) |
625 | return -1; | |
626 | ||
627 | if (l->offset > r->offset) | |
628 | return 1; | |
2b144498 SD |
629 | |
630 | return 0; | |
631 | } | |
632 | ||
633 | static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset) | |
634 | { | |
635 | struct uprobe u = { .inode = inode, .offset = offset }; | |
636 | struct rb_node *n = uprobes_tree.rb_node; | |
637 | struct uprobe *uprobe; | |
638 | int match; | |
639 | ||
640 | while (n) { | |
641 | uprobe = rb_entry(n, struct uprobe, rb_node); | |
642 | match = match_uprobe(&u, uprobe); | |
f231722a ON |
643 | if (!match) |
644 | return get_uprobe(uprobe); | |
7b2d81d4 | 645 | |
2b144498 SD |
646 | if (match < 0) |
647 | n = n->rb_left; | |
648 | else | |
649 | n = n->rb_right; | |
650 | } | |
651 | return NULL; | |
652 | } | |
653 | ||
654 | /* | |
655 | * Find a uprobe corresponding to a given inode:offset | |
656 | * Acquires uprobes_treelock | |
657 | */ | |
658 | static struct uprobe *find_uprobe(struct inode *inode, loff_t offset) | |
659 | { | |
660 | struct uprobe *uprobe; | |
2b144498 | 661 | |
6f47caa0 | 662 | spin_lock(&uprobes_treelock); |
2b144498 | 663 | uprobe = __find_uprobe(inode, offset); |
6f47caa0 | 664 | spin_unlock(&uprobes_treelock); |
7b2d81d4 | 665 | |
2b144498 SD |
666 | return uprobe; |
667 | } | |
668 | ||
669 | static struct uprobe *__insert_uprobe(struct uprobe *uprobe) | |
670 | { | |
671 | struct rb_node **p = &uprobes_tree.rb_node; | |
672 | struct rb_node *parent = NULL; | |
673 | struct uprobe *u; | |
674 | int match; | |
675 | ||
676 | while (*p) { | |
677 | parent = *p; | |
678 | u = rb_entry(parent, struct uprobe, rb_node); | |
679 | match = match_uprobe(uprobe, u); | |
f231722a ON |
680 | if (!match) |
681 | return get_uprobe(u); | |
2b144498 SD |
682 | |
683 | if (match < 0) | |
684 | p = &parent->rb_left; | |
685 | else | |
686 | p = &parent->rb_right; | |
687 | ||
688 | } | |
7b2d81d4 | 689 | |
2b144498 SD |
690 | u = NULL; |
691 | rb_link_node(&uprobe->rb_node, parent, p); | |
692 | rb_insert_color(&uprobe->rb_node, &uprobes_tree); | |
693 | /* get access + creation ref */ | |
ce59b8e9 | 694 | refcount_set(&uprobe->ref, 2); |
7b2d81d4 | 695 | |
2b144498 SD |
696 | return u; |
697 | } | |
698 | ||
699 | /* | |
7b2d81d4 | 700 | * Acquire uprobes_treelock. |
2b144498 SD |
701 | * Matching uprobe already exists in rbtree; |
702 | * increment (access refcount) and return the matching uprobe. | |
703 | * | |
704 | * No matching uprobe; insert the uprobe in rb_tree; | |
705 | * get a double refcount (access + creation) and return NULL. | |
706 | */ | |
707 | static struct uprobe *insert_uprobe(struct uprobe *uprobe) | |
708 | { | |
2b144498 SD |
709 | struct uprobe *u; |
710 | ||
6f47caa0 | 711 | spin_lock(&uprobes_treelock); |
2b144498 | 712 | u = __insert_uprobe(uprobe); |
6f47caa0 | 713 | spin_unlock(&uprobes_treelock); |
7b2d81d4 | 714 | |
2b144498 SD |
715 | return u; |
716 | } | |
717 | ||
22bad382 RB |
718 | static void |
719 | ref_ctr_mismatch_warn(struct uprobe *cur_uprobe, struct uprobe *uprobe) | |
720 | { | |
721 | pr_warn("ref_ctr_offset mismatch. inode: 0x%lx offset: 0x%llx " | |
722 | "ref_ctr_offset(old): 0x%llx ref_ctr_offset(new): 0x%llx\n", | |
723 | uprobe->inode->i_ino, (unsigned long long) uprobe->offset, | |
724 | (unsigned long long) cur_uprobe->ref_ctr_offset, | |
725 | (unsigned long long) uprobe->ref_ctr_offset); | |
726 | } | |
727 | ||
1cc33161 RB |
728 | static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset, |
729 | loff_t ref_ctr_offset) | |
2b144498 SD |
730 | { |
731 | struct uprobe *uprobe, *cur_uprobe; | |
732 | ||
733 | uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL); | |
734 | if (!uprobe) | |
735 | return NULL; | |
736 | ||
61f94203 | 737 | uprobe->inode = inode; |
2b144498 | 738 | uprobe->offset = offset; |
1cc33161 | 739 | uprobe->ref_ctr_offset = ref_ctr_offset; |
e591c8d7 | 740 | init_rwsem(&uprobe->register_rwsem); |
2b144498 | 741 | init_rwsem(&uprobe->consumer_rwsem); |
2b144498 SD |
742 | |
743 | /* add to uprobes_tree, sorted on inode:offset */ | |
744 | cur_uprobe = insert_uprobe(uprobe); | |
2b144498 SD |
745 | /* a uprobe exists for this inode:offset combination */ |
746 | if (cur_uprobe) { | |
22bad382 RB |
747 | if (cur_uprobe->ref_ctr_offset != uprobe->ref_ctr_offset) { |
748 | ref_ctr_mismatch_warn(cur_uprobe, uprobe); | |
749 | put_uprobe(cur_uprobe); | |
750 | kfree(uprobe); | |
751 | return ERR_PTR(-EINVAL); | |
752 | } | |
2b144498 SD |
753 | kfree(uprobe); |
754 | uprobe = cur_uprobe; | |
7b2d81d4 IM |
755 | } |
756 | ||
2b144498 SD |
757 | return uprobe; |
758 | } | |
759 | ||
9a98e03c | 760 | static void consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
761 | { |
762 | down_write(&uprobe->consumer_rwsem); | |
e3343e6a SD |
763 | uc->next = uprobe->consumers; |
764 | uprobe->consumers = uc; | |
2b144498 | 765 | up_write(&uprobe->consumer_rwsem); |
2b144498 SD |
766 | } |
767 | ||
768 | /* | |
e3343e6a SD |
769 | * For uprobe @uprobe, delete the consumer @uc. |
770 | * Return true if the @uc is deleted successfully | |
2b144498 SD |
771 | * or return false. |
772 | */ | |
e3343e6a | 773 | static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc) |
2b144498 SD |
774 | { |
775 | struct uprobe_consumer **con; | |
776 | bool ret = false; | |
777 | ||
778 | down_write(&uprobe->consumer_rwsem); | |
779 | for (con = &uprobe->consumers; *con; con = &(*con)->next) { | |
e3343e6a SD |
780 | if (*con == uc) { |
781 | *con = uc->next; | |
2b144498 SD |
782 | ret = true; |
783 | break; | |
784 | } | |
785 | } | |
786 | up_write(&uprobe->consumer_rwsem); | |
7b2d81d4 | 787 | |
2b144498 SD |
788 | return ret; |
789 | } | |
790 | ||
2ded0980 ON |
791 | static int __copy_insn(struct address_space *mapping, struct file *filp, |
792 | void *insn, int nbytes, loff_t offset) | |
2b144498 | 793 | { |
2b144498 | 794 | struct page *page; |
2b144498 | 795 | /* |
40814f68 ON |
796 | * Ensure that the page that has the original instruction is populated |
797 | * and in page-cache. If ->readpage == NULL it must be shmem_mapping(), | |
798 | * see uprobe_register(). | |
2b144498 | 799 | */ |
40814f68 | 800 | if (mapping->a_ops->readpage) |
09cbfeaf | 801 | page = read_mapping_page(mapping, offset >> PAGE_SHIFT, filp); |
40814f68 | 802 | else |
09cbfeaf | 803 | page = shmem_read_mapping_page(mapping, offset >> PAGE_SHIFT); |
2b144498 SD |
804 | if (IS_ERR(page)) |
805 | return PTR_ERR(page); | |
806 | ||
2edb7b55 | 807 | copy_from_page(page, offset, insn, nbytes); |
09cbfeaf | 808 | put_page(page); |
7b2d81d4 | 809 | |
2b144498 SD |
810 | return 0; |
811 | } | |
812 | ||
d436615e | 813 | static int copy_insn(struct uprobe *uprobe, struct file *filp) |
2b144498 | 814 | { |
2ded0980 ON |
815 | struct address_space *mapping = uprobe->inode->i_mapping; |
816 | loff_t offs = uprobe->offset; | |
803200e2 ON |
817 | void *insn = &uprobe->arch.insn; |
818 | int size = sizeof(uprobe->arch.insn); | |
2ded0980 ON |
819 | int len, err = -EIO; |
820 | ||
821 | /* Copy only available bytes, -EIO if nothing was read */ | |
822 | do { | |
823 | if (offs >= i_size_read(uprobe->inode)) | |
824 | break; | |
825 | ||
826 | len = min_t(int, size, PAGE_SIZE - (offs & ~PAGE_MASK)); | |
827 | err = __copy_insn(mapping, filp, insn, len, offs); | |
fc36f595 | 828 | if (err) |
2ded0980 ON |
829 | break; |
830 | ||
831 | insn += len; | |
832 | offs += len; | |
833 | size -= len; | |
834 | } while (size); | |
835 | ||
836 | return err; | |
2b144498 SD |
837 | } |
838 | ||
cb9a19fe ON |
839 | static int prepare_uprobe(struct uprobe *uprobe, struct file *file, |
840 | struct mm_struct *mm, unsigned long vaddr) | |
841 | { | |
842 | int ret = 0; | |
843 | ||
71434f2f | 844 | if (test_bit(UPROBE_COPY_INSN, &uprobe->flags)) |
cb9a19fe ON |
845 | return ret; |
846 | ||
d4d3ccc6 ON |
847 | /* TODO: move this into _register, until then we abuse this sem. */ |
848 | down_write(&uprobe->consumer_rwsem); | |
71434f2f | 849 | if (test_bit(UPROBE_COPY_INSN, &uprobe->flags)) |
4710f05f ON |
850 | goto out; |
851 | ||
cb9a19fe ON |
852 | ret = copy_insn(uprobe, file); |
853 | if (ret) | |
854 | goto out; | |
855 | ||
856 | ret = -ENOTSUPP; | |
803200e2 | 857 | if (is_trap_insn((uprobe_opcode_t *)&uprobe->arch.insn)) |
cb9a19fe ON |
858 | goto out; |
859 | ||
860 | ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr); | |
861 | if (ret) | |
862 | goto out; | |
863 | ||
09d3f015 | 864 | smp_wmb(); /* pairs with the smp_rmb() in handle_swbp() */ |
71434f2f | 865 | set_bit(UPROBE_COPY_INSN, &uprobe->flags); |
cb9a19fe ON |
866 | |
867 | out: | |
d4d3ccc6 | 868 | up_write(&uprobe->consumer_rwsem); |
4710f05f | 869 | |
cb9a19fe ON |
870 | return ret; |
871 | } | |
872 | ||
8a7f2fa0 ON |
873 | static inline bool consumer_filter(struct uprobe_consumer *uc, |
874 | enum uprobe_filter_ctx ctx, struct mm_struct *mm) | |
806a98bd | 875 | { |
8a7f2fa0 | 876 | return !uc->filter || uc->filter(uc, ctx, mm); |
806a98bd ON |
877 | } |
878 | ||
8a7f2fa0 ON |
879 | static bool filter_chain(struct uprobe *uprobe, |
880 | enum uprobe_filter_ctx ctx, struct mm_struct *mm) | |
63633cbf | 881 | { |
1ff6fee5 ON |
882 | struct uprobe_consumer *uc; |
883 | bool ret = false; | |
884 | ||
885 | down_read(&uprobe->consumer_rwsem); | |
886 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
8a7f2fa0 | 887 | ret = consumer_filter(uc, ctx, mm); |
1ff6fee5 ON |
888 | if (ret) |
889 | break; | |
890 | } | |
891 | up_read(&uprobe->consumer_rwsem); | |
892 | ||
893 | return ret; | |
63633cbf ON |
894 | } |
895 | ||
e3343e6a SD |
896 | static int |
897 | install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, | |
816c03fb | 898 | struct vm_area_struct *vma, unsigned long vaddr) |
2b144498 | 899 | { |
f8ac4ec9 | 900 | bool first_uprobe; |
2b144498 SD |
901 | int ret; |
902 | ||
cb9a19fe ON |
903 | ret = prepare_uprobe(uprobe, vma->vm_file, mm, vaddr); |
904 | if (ret) | |
905 | return ret; | |
682968e0 | 906 | |
f8ac4ec9 ON |
907 | /* |
908 | * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(), | |
909 | * the task can hit this breakpoint right after __replace_page(). | |
910 | */ | |
911 | first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags); | |
912 | if (first_uprobe) | |
913 | set_bit(MMF_HAS_UPROBES, &mm->flags); | |
914 | ||
816c03fb | 915 | ret = set_swbp(&uprobe->arch, mm, vaddr); |
9f68f672 ON |
916 | if (!ret) |
917 | clear_bit(MMF_RECALC_UPROBES, &mm->flags); | |
918 | else if (first_uprobe) | |
f8ac4ec9 | 919 | clear_bit(MMF_HAS_UPROBES, &mm->flags); |
2b144498 SD |
920 | |
921 | return ret; | |
922 | } | |
923 | ||
076a365b | 924 | static int |
816c03fb | 925 | remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr) |
2b144498 | 926 | { |
9f68f672 | 927 | set_bit(MMF_RECALC_UPROBES, &mm->flags); |
076a365b | 928 | return set_orig_insn(&uprobe->arch, mm, vaddr); |
2b144498 SD |
929 | } |
930 | ||
06b7bcd8 ON |
931 | static inline bool uprobe_is_active(struct uprobe *uprobe) |
932 | { | |
933 | return !RB_EMPTY_NODE(&uprobe->rb_node); | |
934 | } | |
0326f5a9 | 935 | /* |
778b032d ON |
936 | * There could be threads that have already hit the breakpoint. They |
937 | * will recheck the current insn and restart if find_uprobe() fails. | |
938 | * See find_active_uprobe(). | |
0326f5a9 | 939 | */ |
2b144498 SD |
940 | static void delete_uprobe(struct uprobe *uprobe) |
941 | { | |
06b7bcd8 ON |
942 | if (WARN_ON(!uprobe_is_active(uprobe))) |
943 | return; | |
944 | ||
6f47caa0 | 945 | spin_lock(&uprobes_treelock); |
2b144498 | 946 | rb_erase(&uprobe->rb_node, &uprobes_tree); |
6f47caa0 | 947 | spin_unlock(&uprobes_treelock); |
06b7bcd8 | 948 | RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */ |
2b144498 | 949 | put_uprobe(uprobe); |
2b144498 SD |
950 | } |
951 | ||
26872090 ON |
952 | struct map_info { |
953 | struct map_info *next; | |
954 | struct mm_struct *mm; | |
816c03fb | 955 | unsigned long vaddr; |
26872090 ON |
956 | }; |
957 | ||
958 | static inline struct map_info *free_map_info(struct map_info *info) | |
2b144498 | 959 | { |
26872090 ON |
960 | struct map_info *next = info->next; |
961 | kfree(info); | |
962 | return next; | |
963 | } | |
964 | ||
965 | static struct map_info * | |
966 | build_map_info(struct address_space *mapping, loff_t offset, bool is_register) | |
967 | { | |
968 | unsigned long pgoff = offset >> PAGE_SHIFT; | |
2b144498 | 969 | struct vm_area_struct *vma; |
26872090 ON |
970 | struct map_info *curr = NULL; |
971 | struct map_info *prev = NULL; | |
972 | struct map_info *info; | |
973 | int more = 0; | |
2b144498 | 974 | |
26872090 | 975 | again: |
4a23717a | 976 | i_mmap_lock_read(mapping); |
6b2dbba8 | 977 | vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) { |
2b144498 SD |
978 | if (!valid_vma(vma, is_register)) |
979 | continue; | |
980 | ||
7a5bfb66 ON |
981 | if (!prev && !more) { |
982 | /* | |
c8c06efa | 983 | * Needs GFP_NOWAIT to avoid i_mmap_rwsem recursion through |
7a5bfb66 ON |
984 | * reclaim. This is optimistic, no harm done if it fails. |
985 | */ | |
986 | prev = kmalloc(sizeof(struct map_info), | |
987 | GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN); | |
988 | if (prev) | |
989 | prev->next = NULL; | |
990 | } | |
26872090 ON |
991 | if (!prev) { |
992 | more++; | |
993 | continue; | |
2b144498 | 994 | } |
2b144498 | 995 | |
388f7934 | 996 | if (!mmget_not_zero(vma->vm_mm)) |
26872090 | 997 | continue; |
7b2d81d4 | 998 | |
26872090 ON |
999 | info = prev; |
1000 | prev = prev->next; | |
1001 | info->next = curr; | |
1002 | curr = info; | |
2b144498 | 1003 | |
26872090 | 1004 | info->mm = vma->vm_mm; |
57683f72 | 1005 | info->vaddr = offset_to_vaddr(vma, offset); |
26872090 | 1006 | } |
4a23717a | 1007 | i_mmap_unlock_read(mapping); |
2b144498 | 1008 | |
26872090 ON |
1009 | if (!more) |
1010 | goto out; | |
1011 | ||
1012 | prev = curr; | |
1013 | while (curr) { | |
1014 | mmput(curr->mm); | |
1015 | curr = curr->next; | |
1016 | } | |
7b2d81d4 | 1017 | |
26872090 ON |
1018 | do { |
1019 | info = kmalloc(sizeof(struct map_info), GFP_KERNEL); | |
1020 | if (!info) { | |
1021 | curr = ERR_PTR(-ENOMEM); | |
1022 | goto out; | |
1023 | } | |
1024 | info->next = prev; | |
1025 | prev = info; | |
1026 | } while (--more); | |
1027 | ||
1028 | goto again; | |
1029 | out: | |
1030 | while (prev) | |
1031 | prev = free_map_info(prev); | |
1032 | return curr; | |
2b144498 SD |
1033 | } |
1034 | ||
bdf8647c ON |
1035 | static int |
1036 | register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new) | |
2b144498 | 1037 | { |
bdf8647c | 1038 | bool is_register = !!new; |
26872090 ON |
1039 | struct map_info *info; |
1040 | int err = 0; | |
2b144498 | 1041 | |
32cdba1e | 1042 | percpu_down_write(&dup_mmap_sem); |
26872090 ON |
1043 | info = build_map_info(uprobe->inode->i_mapping, |
1044 | uprobe->offset, is_register); | |
32cdba1e ON |
1045 | if (IS_ERR(info)) { |
1046 | err = PTR_ERR(info); | |
1047 | goto out; | |
1048 | } | |
7b2d81d4 | 1049 | |
26872090 ON |
1050 | while (info) { |
1051 | struct mm_struct *mm = info->mm; | |
1052 | struct vm_area_struct *vma; | |
7b2d81d4 | 1053 | |
076a365b | 1054 | if (err && is_register) |
26872090 | 1055 | goto free; |
7b2d81d4 | 1056 | |
d8ed45c5 | 1057 | mmap_write_lock(mm); |
f4d6dfe5 ON |
1058 | vma = find_vma(mm, info->vaddr); |
1059 | if (!vma || !valid_vma(vma, is_register) || | |
f281769e | 1060 | file_inode(vma->vm_file) != uprobe->inode) |
26872090 ON |
1061 | goto unlock; |
1062 | ||
f4d6dfe5 ON |
1063 | if (vma->vm_start > info->vaddr || |
1064 | vaddr_to_offset(vma, info->vaddr) != uprobe->offset) | |
26872090 | 1065 | goto unlock; |
2b144498 | 1066 | |
806a98bd ON |
1067 | if (is_register) { |
1068 | /* consult only the "caller", new consumer. */ | |
bdf8647c | 1069 | if (consumer_filter(new, |
8a7f2fa0 | 1070 | UPROBE_FILTER_REGISTER, mm)) |
806a98bd ON |
1071 | err = install_breakpoint(uprobe, mm, vma, info->vaddr); |
1072 | } else if (test_bit(MMF_HAS_UPROBES, &mm->flags)) { | |
8a7f2fa0 ON |
1073 | if (!filter_chain(uprobe, |
1074 | UPROBE_FILTER_UNREGISTER, mm)) | |
806a98bd ON |
1075 | err |= remove_breakpoint(uprobe, mm, info->vaddr); |
1076 | } | |
78f74116 | 1077 | |
26872090 | 1078 | unlock: |
d8ed45c5 | 1079 | mmap_write_unlock(mm); |
26872090 ON |
1080 | free: |
1081 | mmput(mm); | |
1082 | info = free_map_info(info); | |
2b144498 | 1083 | } |
32cdba1e ON |
1084 | out: |
1085 | percpu_up_write(&dup_mmap_sem); | |
26872090 | 1086 | return err; |
2b144498 SD |
1087 | } |
1088 | ||
38e967ae RB |
1089 | static void |
1090 | __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc) | |
2b144498 | 1091 | { |
04aab9b2 ON |
1092 | int err; |
1093 | ||
06d07139 | 1094 | if (WARN_ON(!consumer_del(uprobe, uc))) |
04aab9b2 | 1095 | return; |
2b144498 | 1096 | |
bdf8647c | 1097 | err = register_for_each_vma(uprobe, NULL); |
bb929284 ON |
1098 | /* TODO : cant unregister? schedule a worker thread */ |
1099 | if (!uprobe->consumers && !err) | |
1100 | delete_uprobe(uprobe); | |
2b144498 SD |
1101 | } |
1102 | ||
1103 | /* | |
7140ad38 | 1104 | * uprobe_unregister - unregister an already registered probe. |
38e967ae RB |
1105 | * @inode: the file in which the probe has to be removed. |
1106 | * @offset: offset from the start of the file. | |
1107 | * @uc: identify which probe if multiple probes are colocated. | |
1108 | */ | |
1109 | void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc) | |
1110 | { | |
1111 | struct uprobe *uprobe; | |
1112 | ||
1113 | uprobe = find_uprobe(inode, offset); | |
1114 | if (WARN_ON(!uprobe)) | |
1115 | return; | |
1116 | ||
1117 | down_write(&uprobe->register_rwsem); | |
1118 | __uprobe_unregister(uprobe, uc); | |
1119 | up_write(&uprobe->register_rwsem); | |
1120 | put_uprobe(uprobe); | |
1121 | } | |
1122 | EXPORT_SYMBOL_GPL(uprobe_unregister); | |
1123 | ||
1124 | /* | |
1125 | * __uprobe_register - register a probe | |
2b144498 SD |
1126 | * @inode: the file in which the probe has to be placed. |
1127 | * @offset: offset from the start of the file. | |
e3343e6a | 1128 | * @uc: information on howto handle the probe.. |
2b144498 | 1129 | * |
38e967ae | 1130 | * Apart from the access refcount, __uprobe_register() takes a creation |
2b144498 SD |
1131 | * refcount (thro alloc_uprobe) if and only if this @uprobe is getting |
1132 | * inserted into the rbtree (i.e first consumer for a @inode:@offset | |
7b2d81d4 | 1133 | * tuple). Creation refcount stops uprobe_unregister from freeing the |
2b144498 | 1134 | * @uprobe even before the register operation is complete. Creation |
e3343e6a | 1135 | * refcount is released when the last @uc for the @uprobe |
38e967ae | 1136 | * unregisters. Caller of __uprobe_register() is required to keep @inode |
61f94203 | 1137 | * (and the containing mount) referenced. |
2b144498 SD |
1138 | * |
1139 | * Return errno if it cannot successully install probes | |
1140 | * else return 0 (success) | |
1141 | */ | |
38e967ae | 1142 | static int __uprobe_register(struct inode *inode, loff_t offset, |
1cc33161 | 1143 | loff_t ref_ctr_offset, struct uprobe_consumer *uc) |
2b144498 SD |
1144 | { |
1145 | struct uprobe *uprobe; | |
7b2d81d4 | 1146 | int ret; |
2b144498 | 1147 | |
ea024870 AA |
1148 | /* Uprobe must have at least one set consumer */ |
1149 | if (!uc->handler && !uc->ret_handler) | |
1150 | return -EINVAL; | |
1151 | ||
40814f68 ON |
1152 | /* copy_insn() uses read_mapping_page() or shmem_read_mapping_page() */ |
1153 | if (!inode->i_mapping->a_ops->readpage && !shmem_mapping(inode->i_mapping)) | |
41ccba02 | 1154 | return -EIO; |
f0744af7 | 1155 | /* Racy, just to catch the obvious mistakes */ |
2b144498 | 1156 | if (offset > i_size_read(inode)) |
7b2d81d4 | 1157 | return -EINVAL; |
2b144498 | 1158 | |
013b2deb ON |
1159 | /* |
1160 | * This ensures that copy_from_page(), copy_to_page() and | |
1161 | * __update_ref_ctr() can't cross page boundary. | |
1162 | */ | |
1163 | if (!IS_ALIGNED(offset, UPROBE_SWBP_INSN_SIZE)) | |
1164 | return -EINVAL; | |
1165 | if (!IS_ALIGNED(ref_ctr_offset, sizeof(short))) | |
1166 | return -EINVAL; | |
1167 | ||
66d06dff | 1168 | retry: |
1cc33161 | 1169 | uprobe = alloc_uprobe(inode, offset, ref_ctr_offset); |
66d06dff ON |
1170 | if (!uprobe) |
1171 | return -ENOMEM; | |
22bad382 RB |
1172 | if (IS_ERR(uprobe)) |
1173 | return PTR_ERR(uprobe); | |
1174 | ||
66d06dff ON |
1175 | /* |
1176 | * We can race with uprobe_unregister()->delete_uprobe(). | |
1177 | * Check uprobe_is_active() and retry if it is false. | |
1178 | */ | |
1179 | down_write(&uprobe->register_rwsem); | |
1180 | ret = -EAGAIN; | |
1181 | if (likely(uprobe_is_active(uprobe))) { | |
38e967ae RB |
1182 | consumer_add(uprobe, uc); |
1183 | ret = register_for_each_vma(uprobe, uc); | |
9a98e03c | 1184 | if (ret) |
04aab9b2 | 1185 | __uprobe_unregister(uprobe, uc); |
2b144498 | 1186 | } |
66d06dff ON |
1187 | up_write(&uprobe->register_rwsem); |
1188 | put_uprobe(uprobe); | |
2b144498 | 1189 | |
66d06dff ON |
1190 | if (unlikely(ret == -EAGAIN)) |
1191 | goto retry; | |
2b144498 SD |
1192 | return ret; |
1193 | } | |
38e967ae RB |
1194 | |
1195 | int uprobe_register(struct inode *inode, loff_t offset, | |
1196 | struct uprobe_consumer *uc) | |
1197 | { | |
1cc33161 | 1198 | return __uprobe_register(inode, offset, 0, uc); |
38e967ae | 1199 | } |
e8440c14 | 1200 | EXPORT_SYMBOL_GPL(uprobe_register); |
2b144498 | 1201 | |
1cc33161 RB |
1202 | int uprobe_register_refctr(struct inode *inode, loff_t offset, |
1203 | loff_t ref_ctr_offset, struct uprobe_consumer *uc) | |
1204 | { | |
1205 | return __uprobe_register(inode, offset, ref_ctr_offset, uc); | |
1206 | } | |
1207 | EXPORT_SYMBOL_GPL(uprobe_register_refctr); | |
1208 | ||
bdf8647c | 1209 | /* |
788faab7 | 1210 | * uprobe_apply - unregister an already registered probe. |
bdf8647c ON |
1211 | * @inode: the file in which the probe has to be removed. |
1212 | * @offset: offset from the start of the file. | |
1213 | * @uc: consumer which wants to add more or remove some breakpoints | |
1214 | * @add: add or remove the breakpoints | |
1215 | */ | |
1216 | int uprobe_apply(struct inode *inode, loff_t offset, | |
1217 | struct uprobe_consumer *uc, bool add) | |
1218 | { | |
1219 | struct uprobe *uprobe; | |
1220 | struct uprobe_consumer *con; | |
1221 | int ret = -ENOENT; | |
1222 | ||
1223 | uprobe = find_uprobe(inode, offset); | |
06d07139 | 1224 | if (WARN_ON(!uprobe)) |
bdf8647c ON |
1225 | return ret; |
1226 | ||
1227 | down_write(&uprobe->register_rwsem); | |
1228 | for (con = uprobe->consumers; con && con != uc ; con = con->next) | |
1229 | ; | |
1230 | if (con) | |
1231 | ret = register_for_each_vma(uprobe, add ? uc : NULL); | |
1232 | up_write(&uprobe->register_rwsem); | |
1233 | put_uprobe(uprobe); | |
1234 | ||
1235 | return ret; | |
1236 | } | |
1237 | ||
da1816b1 ON |
1238 | static int unapply_uprobe(struct uprobe *uprobe, struct mm_struct *mm) |
1239 | { | |
1240 | struct vm_area_struct *vma; | |
1241 | int err = 0; | |
1242 | ||
d8ed45c5 | 1243 | mmap_read_lock(mm); |
da1816b1 ON |
1244 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
1245 | unsigned long vaddr; | |
1246 | loff_t offset; | |
1247 | ||
1248 | if (!valid_vma(vma, false) || | |
f281769e | 1249 | file_inode(vma->vm_file) != uprobe->inode) |
da1816b1 ON |
1250 | continue; |
1251 | ||
1252 | offset = (loff_t)vma->vm_pgoff << PAGE_SHIFT; | |
1253 | if (uprobe->offset < offset || | |
1254 | uprobe->offset >= offset + vma->vm_end - vma->vm_start) | |
1255 | continue; | |
1256 | ||
1257 | vaddr = offset_to_vaddr(vma, uprobe->offset); | |
1258 | err |= remove_breakpoint(uprobe, mm, vaddr); | |
1259 | } | |
d8ed45c5 | 1260 | mmap_read_unlock(mm); |
da1816b1 ON |
1261 | |
1262 | return err; | |
1263 | } | |
1264 | ||
891c3970 ON |
1265 | static struct rb_node * |
1266 | find_node_in_range(struct inode *inode, loff_t min, loff_t max) | |
2b144498 | 1267 | { |
2b144498 | 1268 | struct rb_node *n = uprobes_tree.rb_node; |
2b144498 SD |
1269 | |
1270 | while (n) { | |
891c3970 | 1271 | struct uprobe *u = rb_entry(n, struct uprobe, rb_node); |
2b144498 | 1272 | |
891c3970 | 1273 | if (inode < u->inode) { |
2b144498 | 1274 | n = n->rb_left; |
891c3970 | 1275 | } else if (inode > u->inode) { |
2b144498 | 1276 | n = n->rb_right; |
891c3970 ON |
1277 | } else { |
1278 | if (max < u->offset) | |
1279 | n = n->rb_left; | |
1280 | else if (min > u->offset) | |
1281 | n = n->rb_right; | |
1282 | else | |
1283 | break; | |
1284 | } | |
2b144498 | 1285 | } |
7b2d81d4 | 1286 | |
891c3970 | 1287 | return n; |
2b144498 SD |
1288 | } |
1289 | ||
1290 | /* | |
891c3970 | 1291 | * For a given range in vma, build a list of probes that need to be inserted. |
2b144498 | 1292 | */ |
891c3970 ON |
1293 | static void build_probe_list(struct inode *inode, |
1294 | struct vm_area_struct *vma, | |
1295 | unsigned long start, unsigned long end, | |
1296 | struct list_head *head) | |
2b144498 | 1297 | { |
891c3970 | 1298 | loff_t min, max; |
891c3970 ON |
1299 | struct rb_node *n, *t; |
1300 | struct uprobe *u; | |
7b2d81d4 | 1301 | |
891c3970 | 1302 | INIT_LIST_HEAD(head); |
cb113b47 | 1303 | min = vaddr_to_offset(vma, start); |
891c3970 | 1304 | max = min + (end - start) - 1; |
2b144498 | 1305 | |
6f47caa0 | 1306 | spin_lock(&uprobes_treelock); |
891c3970 ON |
1307 | n = find_node_in_range(inode, min, max); |
1308 | if (n) { | |
1309 | for (t = n; t; t = rb_prev(t)) { | |
1310 | u = rb_entry(t, struct uprobe, rb_node); | |
1311 | if (u->inode != inode || u->offset < min) | |
1312 | break; | |
1313 | list_add(&u->pending_list, head); | |
f231722a | 1314 | get_uprobe(u); |
891c3970 ON |
1315 | } |
1316 | for (t = n; (t = rb_next(t)); ) { | |
1317 | u = rb_entry(t, struct uprobe, rb_node); | |
1318 | if (u->inode != inode || u->offset > max) | |
1319 | break; | |
1320 | list_add(&u->pending_list, head); | |
f231722a | 1321 | get_uprobe(u); |
891c3970 | 1322 | } |
2b144498 | 1323 | } |
6f47caa0 | 1324 | spin_unlock(&uprobes_treelock); |
2b144498 SD |
1325 | } |
1326 | ||
1cc33161 RB |
1327 | /* @vma contains reference counter, not the probed instruction. */ |
1328 | static int delayed_ref_ctr_inc(struct vm_area_struct *vma) | |
1329 | { | |
1330 | struct list_head *pos, *q; | |
1331 | struct delayed_uprobe *du; | |
1332 | unsigned long vaddr; | |
1333 | int ret = 0, err = 0; | |
1334 | ||
1335 | mutex_lock(&delayed_uprobe_lock); | |
1336 | list_for_each_safe(pos, q, &delayed_uprobe_list) { | |
1337 | du = list_entry(pos, struct delayed_uprobe, list); | |
1338 | ||
1339 | if (du->mm != vma->vm_mm || | |
1340 | !valid_ref_ctr_vma(du->uprobe, vma)) | |
1341 | continue; | |
1342 | ||
1343 | vaddr = offset_to_vaddr(vma, du->uprobe->ref_ctr_offset); | |
1344 | ret = __update_ref_ctr(vma->vm_mm, vaddr, 1); | |
1345 | if (ret) { | |
1346 | update_ref_ctr_warn(du->uprobe, vma->vm_mm, 1); | |
1347 | if (!err) | |
1348 | err = ret; | |
1349 | } | |
1350 | delayed_uprobe_delete(du); | |
1351 | } | |
1352 | mutex_unlock(&delayed_uprobe_lock); | |
1353 | return err; | |
1354 | } | |
1355 | ||
2b144498 | 1356 | /* |
c1e8d7c6 | 1357 | * Called from mmap_region/vma_adjust with mm->mmap_lock acquired. |
2b144498 | 1358 | * |
5e5be71a ON |
1359 | * Currently we ignore all errors and always return 0, the callers |
1360 | * can't handle the failure anyway. | |
2b144498 | 1361 | */ |
7b2d81d4 | 1362 | int uprobe_mmap(struct vm_area_struct *vma) |
2b144498 SD |
1363 | { |
1364 | struct list_head tmp_list; | |
665605a2 | 1365 | struct uprobe *uprobe, *u; |
2b144498 | 1366 | struct inode *inode; |
2b144498 | 1367 | |
1cc33161 RB |
1368 | if (no_uprobe_events()) |
1369 | return 0; | |
1370 | ||
1371 | if (vma->vm_file && | |
1372 | (vma->vm_flags & (VM_WRITE|VM_SHARED)) == VM_WRITE && | |
1373 | test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags)) | |
1374 | delayed_ref_ctr_inc(vma); | |
1375 | ||
1376 | if (!valid_vma(vma, true)) | |
7b2d81d4 | 1377 | return 0; |
2b144498 | 1378 | |
f281769e | 1379 | inode = file_inode(vma->vm_file); |
2b144498 | 1380 | if (!inode) |
7b2d81d4 | 1381 | return 0; |
2b144498 | 1382 | |
2b144498 | 1383 | mutex_lock(uprobes_mmap_hash(inode)); |
891c3970 | 1384 | build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list); |
806a98bd ON |
1385 | /* |
1386 | * We can race with uprobe_unregister(), this uprobe can be already | |
1387 | * removed. But in this case filter_chain() must return false, all | |
1388 | * consumers have gone away. | |
1389 | */ | |
665605a2 | 1390 | list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) { |
806a98bd | 1391 | if (!fatal_signal_pending(current) && |
8a7f2fa0 | 1392 | filter_chain(uprobe, UPROBE_FILTER_MMAP, vma->vm_mm)) { |
57683f72 | 1393 | unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset); |
5e5be71a | 1394 | install_breakpoint(uprobe, vma->vm_mm, vma, vaddr); |
2b144498 SD |
1395 | } |
1396 | put_uprobe(uprobe); | |
1397 | } | |
2b144498 SD |
1398 | mutex_unlock(uprobes_mmap_hash(inode)); |
1399 | ||
5e5be71a | 1400 | return 0; |
2b144498 SD |
1401 | } |
1402 | ||
9f68f672 ON |
1403 | static bool |
1404 | vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end) | |
1405 | { | |
1406 | loff_t min, max; | |
1407 | struct inode *inode; | |
1408 | struct rb_node *n; | |
1409 | ||
f281769e | 1410 | inode = file_inode(vma->vm_file); |
9f68f672 ON |
1411 | |
1412 | min = vaddr_to_offset(vma, start); | |
1413 | max = min + (end - start) - 1; | |
1414 | ||
1415 | spin_lock(&uprobes_treelock); | |
1416 | n = find_node_in_range(inode, min, max); | |
1417 | spin_unlock(&uprobes_treelock); | |
1418 | ||
1419 | return !!n; | |
1420 | } | |
1421 | ||
682968e0 SD |
1422 | /* |
1423 | * Called in context of a munmap of a vma. | |
1424 | */ | |
cbc91f71 | 1425 | void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end) |
682968e0 | 1426 | { |
441f1eb7 | 1427 | if (no_uprobe_events() || !valid_vma(vma, false)) |
682968e0 SD |
1428 | return; |
1429 | ||
2fd611a9 ON |
1430 | if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */ |
1431 | return; | |
1432 | ||
9f68f672 ON |
1433 | if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) || |
1434 | test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags)) | |
f8ac4ec9 ON |
1435 | return; |
1436 | ||
9f68f672 ON |
1437 | if (vma_has_uprobes(vma, start, end)) |
1438 | set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags); | |
682968e0 SD |
1439 | } |
1440 | ||
d4b3b638 | 1441 | /* Slot allocation for XOL */ |
6441ec8b | 1442 | static int xol_add_vma(struct mm_struct *mm, struct xol_area *area) |
d4b3b638 | 1443 | { |
704bde3c ON |
1444 | struct vm_area_struct *vma; |
1445 | int ret; | |
d4b3b638 | 1446 | |
d8ed45c5 | 1447 | if (mmap_write_lock_killable(mm)) |
598fdc1d MH |
1448 | return -EINTR; |
1449 | ||
704bde3c ON |
1450 | if (mm->uprobes_state.xol_area) { |
1451 | ret = -EALREADY; | |
d4b3b638 | 1452 | goto fail; |
704bde3c | 1453 | } |
d4b3b638 | 1454 | |
af0d95af ON |
1455 | if (!area->vaddr) { |
1456 | /* Try to map as high as possible, this is only a hint. */ | |
1457 | area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, | |
1458 | PAGE_SIZE, 0, 0); | |
ff68dac6 | 1459 | if (IS_ERR_VALUE(area->vaddr)) { |
af0d95af ON |
1460 | ret = area->vaddr; |
1461 | goto fail; | |
1462 | } | |
d4b3b638 SD |
1463 | } |
1464 | ||
704bde3c ON |
1465 | vma = _install_special_mapping(mm, area->vaddr, PAGE_SIZE, |
1466 | VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, | |
1467 | &area->xol_mapping); | |
1468 | if (IS_ERR(vma)) { | |
1469 | ret = PTR_ERR(vma); | |
d4b3b638 | 1470 | goto fail; |
704bde3c | 1471 | } |
d4b3b638 | 1472 | |
704bde3c | 1473 | ret = 0; |
5c6338b4 PM |
1474 | /* pairs with get_xol_area() */ |
1475 | smp_store_release(&mm->uprobes_state.xol_area, area); /* ^^^ */ | |
c8a82538 | 1476 | fail: |
d8ed45c5 | 1477 | mmap_write_unlock(mm); |
d4b3b638 SD |
1478 | |
1479 | return ret; | |
1480 | } | |
1481 | ||
af0d95af | 1482 | static struct xol_area *__create_xol_area(unsigned long vaddr) |
d4b3b638 | 1483 | { |
9b545df8 | 1484 | struct mm_struct *mm = current->mm; |
e78aebfd | 1485 | uprobe_opcode_t insn = UPROBE_SWBP_INSN; |
6441ec8b | 1486 | struct xol_area *area; |
9b545df8 | 1487 | |
af0d95af | 1488 | area = kmalloc(sizeof(*area), GFP_KERNEL); |
d4b3b638 | 1489 | if (unlikely(!area)) |
c8a82538 | 1490 | goto out; |
d4b3b638 | 1491 | |
6396bb22 KC |
1492 | area->bitmap = kcalloc(BITS_TO_LONGS(UINSNS_PER_PAGE), sizeof(long), |
1493 | GFP_KERNEL); | |
d4b3b638 | 1494 | if (!area->bitmap) |
c8a82538 ON |
1495 | goto free_area; |
1496 | ||
704bde3c | 1497 | area->xol_mapping.name = "[uprobes]"; |
869ae761 | 1498 | area->xol_mapping.fault = NULL; |
704bde3c | 1499 | area->xol_mapping.pages = area->pages; |
f58bea2f ON |
1500 | area->pages[0] = alloc_page(GFP_HIGHUSER); |
1501 | if (!area->pages[0]) | |
c8a82538 | 1502 | goto free_bitmap; |
f58bea2f | 1503 | area->pages[1] = NULL; |
d4b3b638 | 1504 | |
af0d95af | 1505 | area->vaddr = vaddr; |
6441ec8b ON |
1506 | init_waitqueue_head(&area->wq); |
1507 | /* Reserve the 1st slot for get_trampoline_vaddr() */ | |
e78aebfd | 1508 | set_bit(0, area->bitmap); |
e78aebfd | 1509 | atomic_set(&area->slot_count, 1); |
297e765e | 1510 | arch_uprobe_copy_ixol(area->pages[0], 0, &insn, UPROBE_SWBP_INSN_SIZE); |
e78aebfd | 1511 | |
6441ec8b | 1512 | if (!xol_add_vma(mm, area)) |
d4b3b638 SD |
1513 | return area; |
1514 | ||
f58bea2f | 1515 | __free_page(area->pages[0]); |
c8a82538 | 1516 | free_bitmap: |
d4b3b638 | 1517 | kfree(area->bitmap); |
c8a82538 | 1518 | free_area: |
d4b3b638 | 1519 | kfree(area); |
c8a82538 | 1520 | out: |
6441ec8b ON |
1521 | return NULL; |
1522 | } | |
1523 | ||
1524 | /* | |
1525 | * get_xol_area - Allocate process's xol_area if necessary. | |
1526 | * This area will be used for storing instructions for execution out of line. | |
1527 | * | |
1528 | * Returns the allocated area or NULL. | |
1529 | */ | |
1530 | static struct xol_area *get_xol_area(void) | |
1531 | { | |
1532 | struct mm_struct *mm = current->mm; | |
1533 | struct xol_area *area; | |
1534 | ||
1535 | if (!mm->uprobes_state.xol_area) | |
af0d95af | 1536 | __create_xol_area(0); |
6441ec8b | 1537 | |
5c6338b4 PM |
1538 | /* Pairs with xol_add_vma() smp_store_release() */ |
1539 | area = READ_ONCE(mm->uprobes_state.xol_area); /* ^^^ */ | |
9b545df8 | 1540 | return area; |
d4b3b638 SD |
1541 | } |
1542 | ||
1543 | /* | |
1544 | * uprobe_clear_state - Free the area allocated for slots. | |
1545 | */ | |
1546 | void uprobe_clear_state(struct mm_struct *mm) | |
1547 | { | |
1548 | struct xol_area *area = mm->uprobes_state.xol_area; | |
1549 | ||
1cc33161 RB |
1550 | mutex_lock(&delayed_uprobe_lock); |
1551 | delayed_uprobe_remove(NULL, mm); | |
1552 | mutex_unlock(&delayed_uprobe_lock); | |
1553 | ||
d4b3b638 SD |
1554 | if (!area) |
1555 | return; | |
1556 | ||
f58bea2f | 1557 | put_page(area->pages[0]); |
d4b3b638 SD |
1558 | kfree(area->bitmap); |
1559 | kfree(area); | |
1560 | } | |
1561 | ||
32cdba1e ON |
1562 | void uprobe_start_dup_mmap(void) |
1563 | { | |
1564 | percpu_down_read(&dup_mmap_sem); | |
1565 | } | |
1566 | ||
1567 | void uprobe_end_dup_mmap(void) | |
1568 | { | |
1569 | percpu_up_read(&dup_mmap_sem); | |
1570 | } | |
1571 | ||
f8ac4ec9 ON |
1572 | void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm) |
1573 | { | |
9f68f672 | 1574 | if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) { |
f8ac4ec9 | 1575 | set_bit(MMF_HAS_UPROBES, &newmm->flags); |
9f68f672 ON |
1576 | /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */ |
1577 | set_bit(MMF_RECALC_UPROBES, &newmm->flags); | |
1578 | } | |
f8ac4ec9 ON |
1579 | } |
1580 | ||
d4b3b638 SD |
1581 | /* |
1582 | * - search for a free slot. | |
1583 | */ | |
1584 | static unsigned long xol_take_insn_slot(struct xol_area *area) | |
1585 | { | |
1586 | unsigned long slot_addr; | |
1587 | int slot_nr; | |
1588 | ||
1589 | do { | |
1590 | slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE); | |
1591 | if (slot_nr < UINSNS_PER_PAGE) { | |
1592 | if (!test_and_set_bit(slot_nr, area->bitmap)) | |
1593 | break; | |
1594 | ||
1595 | slot_nr = UINSNS_PER_PAGE; | |
1596 | continue; | |
1597 | } | |
1598 | wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE)); | |
1599 | } while (slot_nr >= UINSNS_PER_PAGE); | |
1600 | ||
1601 | slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES); | |
1602 | atomic_inc(&area->slot_count); | |
1603 | ||
1604 | return slot_addr; | |
1605 | } | |
1606 | ||
1607 | /* | |
a6cb3f6d | 1608 | * xol_get_insn_slot - allocate a slot for xol. |
d4b3b638 SD |
1609 | * Returns the allocated slot address or 0. |
1610 | */ | |
a6cb3f6d | 1611 | static unsigned long xol_get_insn_slot(struct uprobe *uprobe) |
d4b3b638 SD |
1612 | { |
1613 | struct xol_area *area; | |
a6cb3f6d | 1614 | unsigned long xol_vaddr; |
d4b3b638 | 1615 | |
9b545df8 ON |
1616 | area = get_xol_area(); |
1617 | if (!area) | |
1618 | return 0; | |
d4b3b638 | 1619 | |
a6cb3f6d ON |
1620 | xol_vaddr = xol_take_insn_slot(area); |
1621 | if (unlikely(!xol_vaddr)) | |
d4b3b638 SD |
1622 | return 0; |
1623 | ||
f58bea2f | 1624 | arch_uprobe_copy_ixol(area->pages[0], xol_vaddr, |
72e6ae28 | 1625 | &uprobe->arch.ixol, sizeof(uprobe->arch.ixol)); |
d4b3b638 | 1626 | |
a6cb3f6d | 1627 | return xol_vaddr; |
d4b3b638 SD |
1628 | } |
1629 | ||
1630 | /* | |
1631 | * xol_free_insn_slot - If slot was earlier allocated by | |
1632 | * @xol_get_insn_slot(), make the slot available for | |
1633 | * subsequent requests. | |
1634 | */ | |
1635 | static void xol_free_insn_slot(struct task_struct *tsk) | |
1636 | { | |
1637 | struct xol_area *area; | |
1638 | unsigned long vma_end; | |
1639 | unsigned long slot_addr; | |
1640 | ||
1641 | if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask) | |
1642 | return; | |
1643 | ||
1644 | slot_addr = tsk->utask->xol_vaddr; | |
af4355e9 | 1645 | if (unlikely(!slot_addr)) |
d4b3b638 SD |
1646 | return; |
1647 | ||
1648 | area = tsk->mm->uprobes_state.xol_area; | |
1649 | vma_end = area->vaddr + PAGE_SIZE; | |
1650 | if (area->vaddr <= slot_addr && slot_addr < vma_end) { | |
1651 | unsigned long offset; | |
1652 | int slot_nr; | |
1653 | ||
1654 | offset = slot_addr - area->vaddr; | |
1655 | slot_nr = offset / UPROBE_XOL_SLOT_BYTES; | |
1656 | if (slot_nr >= UINSNS_PER_PAGE) | |
1657 | return; | |
1658 | ||
1659 | clear_bit(slot_nr, area->bitmap); | |
1660 | atomic_dec(&area->slot_count); | |
2a742ced | 1661 | smp_mb__after_atomic(); /* pairs with prepare_to_wait() */ |
d4b3b638 SD |
1662 | if (waitqueue_active(&area->wq)) |
1663 | wake_up(&area->wq); | |
1664 | ||
1665 | tsk->utask->xol_vaddr = 0; | |
1666 | } | |
1667 | } | |
1668 | ||
72e6ae28 VK |
1669 | void __weak arch_uprobe_copy_ixol(struct page *page, unsigned long vaddr, |
1670 | void *src, unsigned long len) | |
1671 | { | |
1672 | /* Initialize the slot */ | |
1673 | copy_to_page(page, vaddr, src, len); | |
1674 | ||
1675 | /* | |
885f7f8e | 1676 | * We probably need flush_icache_user_page() but it needs vma. |
72e6ae28 VK |
1677 | * This should work on most of architectures by default. If |
1678 | * architecture needs to do something different it can define | |
1679 | * its own version of the function. | |
1680 | */ | |
1681 | flush_dcache_page(page); | |
1682 | } | |
1683 | ||
0326f5a9 SD |
1684 | /** |
1685 | * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs | |
1686 | * @regs: Reflects the saved state of the task after it has hit a breakpoint | |
1687 | * instruction. | |
1688 | * Return the address of the breakpoint instruction. | |
1689 | */ | |
1690 | unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs) | |
1691 | { | |
1692 | return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE; | |
1693 | } | |
1694 | ||
b02ef20a ON |
1695 | unsigned long uprobe_get_trap_addr(struct pt_regs *regs) |
1696 | { | |
1697 | struct uprobe_task *utask = current->utask; | |
1698 | ||
1699 | if (unlikely(utask && utask->active_uprobe)) | |
1700 | return utask->vaddr; | |
1701 | ||
1702 | return instruction_pointer(regs); | |
1703 | } | |
1704 | ||
2bb5e840 ON |
1705 | static struct return_instance *free_ret_instance(struct return_instance *ri) |
1706 | { | |
1707 | struct return_instance *next = ri->next; | |
1708 | put_uprobe(ri->uprobe); | |
1709 | kfree(ri); | |
1710 | return next; | |
1711 | } | |
1712 | ||
0326f5a9 SD |
1713 | /* |
1714 | * Called with no locks held. | |
788faab7 | 1715 | * Called in context of an exiting or an exec-ing thread. |
0326f5a9 SD |
1716 | */ |
1717 | void uprobe_free_utask(struct task_struct *t) | |
1718 | { | |
1719 | struct uprobe_task *utask = t->utask; | |
2bb5e840 | 1720 | struct return_instance *ri; |
0326f5a9 | 1721 | |
0326f5a9 SD |
1722 | if (!utask) |
1723 | return; | |
1724 | ||
1725 | if (utask->active_uprobe) | |
1726 | put_uprobe(utask->active_uprobe); | |
1727 | ||
0dfd0eb8 | 1728 | ri = utask->return_instances; |
2bb5e840 ON |
1729 | while (ri) |
1730 | ri = free_ret_instance(ri); | |
0dfd0eb8 | 1731 | |
d4b3b638 | 1732 | xol_free_insn_slot(t); |
0326f5a9 SD |
1733 | kfree(utask); |
1734 | t->utask = NULL; | |
1735 | } | |
1736 | ||
0326f5a9 | 1737 | /* |
5a2df662 ON |
1738 | * Allocate a uprobe_task object for the task if if necessary. |
1739 | * Called when the thread hits a breakpoint. | |
0326f5a9 SD |
1740 | * |
1741 | * Returns: | |
1742 | * - pointer to new uprobe_task on success | |
1743 | * - NULL otherwise | |
1744 | */ | |
5a2df662 | 1745 | static struct uprobe_task *get_utask(void) |
0326f5a9 | 1746 | { |
5a2df662 ON |
1747 | if (!current->utask) |
1748 | current->utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL); | |
1749 | return current->utask; | |
0326f5a9 SD |
1750 | } |
1751 | ||
248d3a7b ON |
1752 | static int dup_utask(struct task_struct *t, struct uprobe_task *o_utask) |
1753 | { | |
1754 | struct uprobe_task *n_utask; | |
1755 | struct return_instance **p, *o, *n; | |
1756 | ||
1757 | n_utask = kzalloc(sizeof(struct uprobe_task), GFP_KERNEL); | |
1758 | if (!n_utask) | |
1759 | return -ENOMEM; | |
1760 | t->utask = n_utask; | |
1761 | ||
1762 | p = &n_utask->return_instances; | |
1763 | for (o = o_utask->return_instances; o; o = o->next) { | |
1764 | n = kmalloc(sizeof(struct return_instance), GFP_KERNEL); | |
1765 | if (!n) | |
1766 | return -ENOMEM; | |
1767 | ||
1768 | *n = *o; | |
f231722a | 1769 | get_uprobe(n->uprobe); |
248d3a7b ON |
1770 | n->next = NULL; |
1771 | ||
1772 | *p = n; | |
1773 | p = &n->next; | |
1774 | n_utask->depth++; | |
1775 | } | |
1776 | ||
1777 | return 0; | |
1778 | } | |
1779 | ||
1780 | static void uprobe_warn(struct task_struct *t, const char *msg) | |
1781 | { | |
1782 | pr_warn("uprobe: %s:%d failed to %s\n", | |
1783 | current->comm, current->pid, msg); | |
1784 | } | |
1785 | ||
aa59c53f ON |
1786 | static void dup_xol_work(struct callback_head *work) |
1787 | { | |
aa59c53f ON |
1788 | if (current->flags & PF_EXITING) |
1789 | return; | |
1790 | ||
598fdc1d MH |
1791 | if (!__create_xol_area(current->utask->dup_xol_addr) && |
1792 | !fatal_signal_pending(current)) | |
aa59c53f ON |
1793 | uprobe_warn(current, "dup xol area"); |
1794 | } | |
1795 | ||
b68e0749 ON |
1796 | /* |
1797 | * Called in context of a new clone/fork from copy_process. | |
1798 | */ | |
3ab67966 | 1799 | void uprobe_copy_process(struct task_struct *t, unsigned long flags) |
b68e0749 | 1800 | { |
248d3a7b ON |
1801 | struct uprobe_task *utask = current->utask; |
1802 | struct mm_struct *mm = current->mm; | |
aa59c53f | 1803 | struct xol_area *area; |
248d3a7b | 1804 | |
b68e0749 | 1805 | t->utask = NULL; |
248d3a7b | 1806 | |
3ab67966 ON |
1807 | if (!utask || !utask->return_instances) |
1808 | return; | |
1809 | ||
1810 | if (mm == t->mm && !(flags & CLONE_VFORK)) | |
248d3a7b ON |
1811 | return; |
1812 | ||
1813 | if (dup_utask(t, utask)) | |
1814 | return uprobe_warn(t, "dup ret instances"); | |
aa59c53f ON |
1815 | |
1816 | /* The task can fork() after dup_xol_work() fails */ | |
1817 | area = mm->uprobes_state.xol_area; | |
1818 | if (!area) | |
1819 | return uprobe_warn(t, "dup xol area"); | |
1820 | ||
3ab67966 ON |
1821 | if (mm == t->mm) |
1822 | return; | |
1823 | ||
32473431 ON |
1824 | t->utask->dup_xol_addr = area->vaddr; |
1825 | init_task_work(&t->utask->dup_xol_work, dup_xol_work); | |
1826 | task_work_add(t, &t->utask->dup_xol_work, true); | |
b68e0749 ON |
1827 | } |
1828 | ||
e78aebfd AA |
1829 | /* |
1830 | * Current area->vaddr notion assume the trampoline address is always | |
1831 | * equal area->vaddr. | |
1832 | * | |
1833 | * Returns -1 in case the xol_area is not allocated. | |
1834 | */ | |
1835 | static unsigned long get_trampoline_vaddr(void) | |
1836 | { | |
1837 | struct xol_area *area; | |
1838 | unsigned long trampoline_vaddr = -1; | |
1839 | ||
5c6338b4 PM |
1840 | /* Pairs with xol_add_vma() smp_store_release() */ |
1841 | area = READ_ONCE(current->mm->uprobes_state.xol_area); /* ^^^ */ | |
e78aebfd AA |
1842 | if (area) |
1843 | trampoline_vaddr = area->vaddr; | |
1844 | ||
1845 | return trampoline_vaddr; | |
1846 | } | |
1847 | ||
db087ef6 ON |
1848 | static void cleanup_return_instances(struct uprobe_task *utask, bool chained, |
1849 | struct pt_regs *regs) | |
a5b7e1a8 ON |
1850 | { |
1851 | struct return_instance *ri = utask->return_instances; | |
db087ef6 | 1852 | enum rp_check ctx = chained ? RP_CHECK_CHAIN_CALL : RP_CHECK_CALL; |
86dcb702 ON |
1853 | |
1854 | while (ri && !arch_uretprobe_is_alive(ri, ctx, regs)) { | |
a5b7e1a8 ON |
1855 | ri = free_ret_instance(ri); |
1856 | utask->depth--; | |
1857 | } | |
1858 | utask->return_instances = ri; | |
1859 | } | |
1860 | ||
0dfd0eb8 AA |
1861 | static void prepare_uretprobe(struct uprobe *uprobe, struct pt_regs *regs) |
1862 | { | |
1863 | struct return_instance *ri; | |
1864 | struct uprobe_task *utask; | |
1865 | unsigned long orig_ret_vaddr, trampoline_vaddr; | |
db087ef6 | 1866 | bool chained; |
0dfd0eb8 AA |
1867 | |
1868 | if (!get_xol_area()) | |
1869 | return; | |
1870 | ||
1871 | utask = get_utask(); | |
1872 | if (!utask) | |
1873 | return; | |
1874 | ||
ded49c55 AA |
1875 | if (utask->depth >= MAX_URETPROBE_DEPTH) { |
1876 | printk_ratelimited(KERN_INFO "uprobe: omit uretprobe due to" | |
1877 | " nestedness limit pid/tgid=%d/%d\n", | |
1878 | current->pid, current->tgid); | |
1879 | return; | |
1880 | } | |
1881 | ||
6c58d0e4 | 1882 | ri = kmalloc(sizeof(struct return_instance), GFP_KERNEL); |
0dfd0eb8 | 1883 | if (!ri) |
6c58d0e4 | 1884 | return; |
0dfd0eb8 AA |
1885 | |
1886 | trampoline_vaddr = get_trampoline_vaddr(); | |
1887 | orig_ret_vaddr = arch_uretprobe_hijack_return_addr(trampoline_vaddr, regs); | |
1888 | if (orig_ret_vaddr == -1) | |
1889 | goto fail; | |
1890 | ||
a5b7e1a8 | 1891 | /* drop the entries invalidated by longjmp() */ |
db087ef6 ON |
1892 | chained = (orig_ret_vaddr == trampoline_vaddr); |
1893 | cleanup_return_instances(utask, chained, regs); | |
a5b7e1a8 | 1894 | |
0dfd0eb8 AA |
1895 | /* |
1896 | * We don't want to keep trampoline address in stack, rather keep the | |
1897 | * original return address of first caller thru all the consequent | |
1898 | * instances. This also makes breakpoint unwrapping easier. | |
1899 | */ | |
db087ef6 | 1900 | if (chained) { |
0dfd0eb8 AA |
1901 | if (!utask->return_instances) { |
1902 | /* | |
1903 | * This situation is not possible. Likely we have an | |
1904 | * attack from user-space. | |
1905 | */ | |
6c58d0e4 | 1906 | uprobe_warn(current, "handle tail call"); |
0dfd0eb8 AA |
1907 | goto fail; |
1908 | } | |
0dfd0eb8 AA |
1909 | orig_ret_vaddr = utask->return_instances->orig_ret_vaddr; |
1910 | } | |
1911 | ||
f231722a | 1912 | ri->uprobe = get_uprobe(uprobe); |
0dfd0eb8 | 1913 | ri->func = instruction_pointer(regs); |
7b868e48 | 1914 | ri->stack = user_stack_pointer(regs); |
0dfd0eb8 AA |
1915 | ri->orig_ret_vaddr = orig_ret_vaddr; |
1916 | ri->chained = chained; | |
1917 | ||
ded49c55 | 1918 | utask->depth++; |
0dfd0eb8 AA |
1919 | ri->next = utask->return_instances; |
1920 | utask->return_instances = ri; | |
1921 | ||
1922 | return; | |
0dfd0eb8 AA |
1923 | fail: |
1924 | kfree(ri); | |
1925 | } | |
1926 | ||
0326f5a9 SD |
1927 | /* Prepare to single-step probed instruction out of line. */ |
1928 | static int | |
a6cb3f6d | 1929 | pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long bp_vaddr) |
0326f5a9 | 1930 | { |
a6cb3f6d ON |
1931 | struct uprobe_task *utask; |
1932 | unsigned long xol_vaddr; | |
aba51024 | 1933 | int err; |
a6cb3f6d | 1934 | |
608e7427 ON |
1935 | utask = get_utask(); |
1936 | if (!utask) | |
1937 | return -ENOMEM; | |
a6cb3f6d ON |
1938 | |
1939 | xol_vaddr = xol_get_insn_slot(uprobe); | |
1940 | if (!xol_vaddr) | |
1941 | return -ENOMEM; | |
1942 | ||
1943 | utask->xol_vaddr = xol_vaddr; | |
1944 | utask->vaddr = bp_vaddr; | |
d4b3b638 | 1945 | |
aba51024 ON |
1946 | err = arch_uprobe_pre_xol(&uprobe->arch, regs); |
1947 | if (unlikely(err)) { | |
1948 | xol_free_insn_slot(current); | |
1949 | return err; | |
1950 | } | |
1951 | ||
608e7427 ON |
1952 | utask->active_uprobe = uprobe; |
1953 | utask->state = UTASK_SSTEP; | |
aba51024 | 1954 | return 0; |
0326f5a9 SD |
1955 | } |
1956 | ||
1957 | /* | |
1958 | * If we are singlestepping, then ensure this thread is not connected to | |
1959 | * non-fatal signals until completion of singlestep. When xol insn itself | |
1960 | * triggers the signal, restart the original insn even if the task is | |
1961 | * already SIGKILL'ed (since coredump should report the correct ip). This | |
1962 | * is even more important if the task has a handler for SIGSEGV/etc, The | |
1963 | * _same_ instruction should be repeated again after return from the signal | |
1964 | * handler, and SSTEP can never finish in this case. | |
1965 | */ | |
1966 | bool uprobe_deny_signal(void) | |
1967 | { | |
1968 | struct task_struct *t = current; | |
1969 | struct uprobe_task *utask = t->utask; | |
1970 | ||
1971 | if (likely(!utask || !utask->active_uprobe)) | |
1972 | return false; | |
1973 | ||
1974 | WARN_ON_ONCE(utask->state != UTASK_SSTEP); | |
1975 | ||
1976 | if (signal_pending(t)) { | |
1977 | spin_lock_irq(&t->sighand->siglock); | |
1978 | clear_tsk_thread_flag(t, TIF_SIGPENDING); | |
1979 | spin_unlock_irq(&t->sighand->siglock); | |
1980 | ||
1981 | if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) { | |
1982 | utask->state = UTASK_SSTEP_TRAPPED; | |
1983 | set_tsk_thread_flag(t, TIF_UPROBE); | |
0326f5a9 SD |
1984 | } |
1985 | } | |
1986 | ||
1987 | return true; | |
1988 | } | |
1989 | ||
499a4f3e ON |
1990 | static void mmf_recalc_uprobes(struct mm_struct *mm) |
1991 | { | |
1992 | struct vm_area_struct *vma; | |
1993 | ||
1994 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
1995 | if (!valid_vma(vma, false)) | |
1996 | continue; | |
1997 | /* | |
1998 | * This is not strictly accurate, we can race with | |
1999 | * uprobe_unregister() and see the already removed | |
2000 | * uprobe if delete_uprobe() was not yet called. | |
63633cbf | 2001 | * Or this uprobe can be filtered out. |
499a4f3e ON |
2002 | */ |
2003 | if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end)) | |
2004 | return; | |
2005 | } | |
2006 | ||
2007 | clear_bit(MMF_HAS_UPROBES, &mm->flags); | |
2008 | } | |
2009 | ||
0908ad6e | 2010 | static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr) |
ec75fba9 ON |
2011 | { |
2012 | struct page *page; | |
2013 | uprobe_opcode_t opcode; | |
2014 | int result; | |
2015 | ||
013b2deb ON |
2016 | if (WARN_ON_ONCE(!IS_ALIGNED(vaddr, UPROBE_SWBP_INSN_SIZE))) |
2017 | return -EINVAL; | |
2018 | ||
ec75fba9 | 2019 | pagefault_disable(); |
bd28b145 | 2020 | result = __get_user(opcode, (uprobe_opcode_t __user *)vaddr); |
ec75fba9 ON |
2021 | pagefault_enable(); |
2022 | ||
2023 | if (likely(result == 0)) | |
2024 | goto out; | |
2025 | ||
1e987790 DH |
2026 | /* |
2027 | * The NULL 'tsk' here ensures that any faults that occur here | |
2028 | * will not be accounted to the task. 'mm' *is* current->mm, | |
2029 | * but we treat this as a 'remote' access since it is | |
2030 | * essentially a kernel access to the memory. | |
2031 | */ | |
9beae1ea | 2032 | result = get_user_pages_remote(NULL, mm, vaddr, 1, FOLL_FORCE, &page, |
5b56d49f | 2033 | NULL, NULL); |
ec75fba9 ON |
2034 | if (result < 0) |
2035 | return result; | |
2036 | ||
ab0d805c | 2037 | copy_from_page(page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE); |
ec75fba9 ON |
2038 | put_page(page); |
2039 | out: | |
0908ad6e AM |
2040 | /* This needs to return true for any variant of the trap insn */ |
2041 | return is_trap_insn(&opcode); | |
ec75fba9 ON |
2042 | } |
2043 | ||
d790d346 | 2044 | static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp) |
0326f5a9 | 2045 | { |
3a9ea052 ON |
2046 | struct mm_struct *mm = current->mm; |
2047 | struct uprobe *uprobe = NULL; | |
0326f5a9 | 2048 | struct vm_area_struct *vma; |
0326f5a9 | 2049 | |
d8ed45c5 | 2050 | mmap_read_lock(mm); |
0326f5a9 | 2051 | vma = find_vma(mm, bp_vaddr); |
3a9ea052 ON |
2052 | if (vma && vma->vm_start <= bp_vaddr) { |
2053 | if (valid_vma(vma, false)) { | |
f281769e | 2054 | struct inode *inode = file_inode(vma->vm_file); |
cb113b47 | 2055 | loff_t offset = vaddr_to_offset(vma, bp_vaddr); |
0326f5a9 | 2056 | |
3a9ea052 ON |
2057 | uprobe = find_uprobe(inode, offset); |
2058 | } | |
d790d346 ON |
2059 | |
2060 | if (!uprobe) | |
0908ad6e | 2061 | *is_swbp = is_trap_at_addr(mm, bp_vaddr); |
d790d346 ON |
2062 | } else { |
2063 | *is_swbp = -EFAULT; | |
0326f5a9 | 2064 | } |
499a4f3e ON |
2065 | |
2066 | if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags)) | |
2067 | mmf_recalc_uprobes(mm); | |
d8ed45c5 | 2068 | mmap_read_unlock(mm); |
0326f5a9 | 2069 | |
3a9ea052 ON |
2070 | return uprobe; |
2071 | } | |
2072 | ||
da1816b1 ON |
2073 | static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs) |
2074 | { | |
2075 | struct uprobe_consumer *uc; | |
2076 | int remove = UPROBE_HANDLER_REMOVE; | |
0dfd0eb8 | 2077 | bool need_prep = false; /* prepare return uprobe, when needed */ |
da1816b1 ON |
2078 | |
2079 | down_read(&uprobe->register_rwsem); | |
2080 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
ea024870 | 2081 | int rc = 0; |
da1816b1 | 2082 | |
ea024870 AA |
2083 | if (uc->handler) { |
2084 | rc = uc->handler(uc, regs); | |
2085 | WARN(rc & ~UPROBE_HANDLER_MASK, | |
d75f773c | 2086 | "bad rc=0x%x from %ps()\n", rc, uc->handler); |
ea024870 | 2087 | } |
0dfd0eb8 AA |
2088 | |
2089 | if (uc->ret_handler) | |
2090 | need_prep = true; | |
2091 | ||
da1816b1 ON |
2092 | remove &= rc; |
2093 | } | |
2094 | ||
0dfd0eb8 AA |
2095 | if (need_prep && !remove) |
2096 | prepare_uretprobe(uprobe, regs); /* put bp at return */ | |
2097 | ||
da1816b1 ON |
2098 | if (remove && uprobe->consumers) { |
2099 | WARN_ON(!uprobe_is_active(uprobe)); | |
2100 | unapply_uprobe(uprobe, current->mm); | |
2101 | } | |
2102 | up_read(&uprobe->register_rwsem); | |
2103 | } | |
2104 | ||
fec8898d AA |
2105 | static void |
2106 | handle_uretprobe_chain(struct return_instance *ri, struct pt_regs *regs) | |
2107 | { | |
2108 | struct uprobe *uprobe = ri->uprobe; | |
2109 | struct uprobe_consumer *uc; | |
2110 | ||
2111 | down_read(&uprobe->register_rwsem); | |
2112 | for (uc = uprobe->consumers; uc; uc = uc->next) { | |
2113 | if (uc->ret_handler) | |
2114 | uc->ret_handler(uc, ri->func, regs); | |
2115 | } | |
2116 | up_read(&uprobe->register_rwsem); | |
2117 | } | |
2118 | ||
a83cfeb9 ON |
2119 | static struct return_instance *find_next_ret_chain(struct return_instance *ri) |
2120 | { | |
2121 | bool chained; | |
2122 | ||
2123 | do { | |
2124 | chained = ri->chained; | |
2125 | ri = ri->next; /* can't be NULL if chained */ | |
2126 | } while (chained); | |
2127 | ||
2128 | return ri; | |
2129 | } | |
2130 | ||
0b5256c7 | 2131 | static void handle_trampoline(struct pt_regs *regs) |
fec8898d AA |
2132 | { |
2133 | struct uprobe_task *utask; | |
a83cfeb9 | 2134 | struct return_instance *ri, *next; |
5eeb50de | 2135 | bool valid; |
fec8898d AA |
2136 | |
2137 | utask = current->utask; | |
2138 | if (!utask) | |
0b5256c7 | 2139 | goto sigill; |
fec8898d AA |
2140 | |
2141 | ri = utask->return_instances; | |
2142 | if (!ri) | |
0b5256c7 | 2143 | goto sigill; |
fec8898d | 2144 | |
a83cfeb9 | 2145 | do { |
5eeb50de ON |
2146 | /* |
2147 | * We should throw out the frames invalidated by longjmp(). | |
2148 | * If this chain is valid, then the next one should be alive | |
2149 | * or NULL; the latter case means that nobody but ri->func | |
2150 | * could hit this trampoline on return. TODO: sigaltstack(). | |
2151 | */ | |
2152 | next = find_next_ret_chain(ri); | |
86dcb702 | 2153 | valid = !next || arch_uretprobe_is_alive(next, RP_CHECK_RET, regs); |
5eeb50de ON |
2154 | |
2155 | instruction_pointer_set(regs, ri->orig_ret_vaddr); | |
2156 | do { | |
2157 | if (valid) | |
2158 | handle_uretprobe_chain(ri, regs); | |
2159 | ri = free_ret_instance(ri); | |
2160 | utask->depth--; | |
2161 | } while (ri != next); | |
2162 | } while (!valid); | |
fec8898d AA |
2163 | |
2164 | utask->return_instances = ri; | |
0b5256c7 ON |
2165 | return; |
2166 | ||
2167 | sigill: | |
2168 | uprobe_warn(current, "handle uretprobe, sending SIGILL."); | |
3cf5d076 | 2169 | force_sig(SIGILL); |
fec8898d | 2170 | |
fec8898d AA |
2171 | } |
2172 | ||
6fe50a28 DL |
2173 | bool __weak arch_uprobe_ignore(struct arch_uprobe *aup, struct pt_regs *regs) |
2174 | { | |
2175 | return false; | |
2176 | } | |
2177 | ||
86dcb702 ON |
2178 | bool __weak arch_uretprobe_is_alive(struct return_instance *ret, enum rp_check ctx, |
2179 | struct pt_regs *regs) | |
97da8976 ON |
2180 | { |
2181 | return true; | |
2182 | } | |
2183 | ||
3a9ea052 ON |
2184 | /* |
2185 | * Run handler and ask thread to singlestep. | |
2186 | * Ensure all non-fatal signals cannot interrupt thread while it singlesteps. | |
2187 | */ | |
2188 | static void handle_swbp(struct pt_regs *regs) | |
2189 | { | |
3a9ea052 ON |
2190 | struct uprobe *uprobe; |
2191 | unsigned long bp_vaddr; | |
3f649ab7 | 2192 | int is_swbp; |
3a9ea052 ON |
2193 | |
2194 | bp_vaddr = uprobe_get_swbp_addr(regs); | |
0b5256c7 ON |
2195 | if (bp_vaddr == get_trampoline_vaddr()) |
2196 | return handle_trampoline(regs); | |
fec8898d AA |
2197 | |
2198 | uprobe = find_active_uprobe(bp_vaddr, &is_swbp); | |
0326f5a9 | 2199 | if (!uprobe) { |
56bb4cf6 ON |
2200 | if (is_swbp > 0) { |
2201 | /* No matching uprobe; signal SIGTRAP. */ | |
fe5ed7ab | 2202 | force_sig(SIGTRAP); |
56bb4cf6 ON |
2203 | } else { |
2204 | /* | |
2205 | * Either we raced with uprobe_unregister() or we can't | |
2206 | * access this memory. The latter is only possible if | |
2207 | * another thread plays with our ->mm. In both cases | |
2208 | * we can simply restart. If this vma was unmapped we | |
2209 | * can pretend this insn was not executed yet and get | |
2210 | * the (correct) SIGSEGV after restart. | |
2211 | */ | |
2212 | instruction_pointer_set(regs, bp_vaddr); | |
2213 | } | |
0326f5a9 SD |
2214 | return; |
2215 | } | |
74e59dfc ON |
2216 | |
2217 | /* change it in advance for ->handler() and restart */ | |
2218 | instruction_pointer_set(regs, bp_vaddr); | |
2219 | ||
142b18dd ON |
2220 | /* |
2221 | * TODO: move copy_insn/etc into _register and remove this hack. | |
2222 | * After we hit the bp, _unregister + _register can install the | |
2223 | * new and not-yet-analyzed uprobe at the same address, restart. | |
2224 | */ | |
71434f2f | 2225 | if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags))) |
74e59dfc | 2226 | goto out; |
0326f5a9 | 2227 | |
09d3f015 AP |
2228 | /* |
2229 | * Pairs with the smp_wmb() in prepare_uprobe(). | |
2230 | * | |
2231 | * Guarantees that if we see the UPROBE_COPY_INSN bit set, then | |
2232 | * we must also see the stores to &uprobe->arch performed by the | |
2233 | * prepare_uprobe() call. | |
2234 | */ | |
2235 | smp_rmb(); | |
2236 | ||
72fd293a ON |
2237 | /* Tracing handlers use ->utask to communicate with fetch methods */ |
2238 | if (!get_utask()) | |
2239 | goto out; | |
2240 | ||
6fe50a28 DL |
2241 | if (arch_uprobe_ignore(&uprobe->arch, regs)) |
2242 | goto out; | |
2243 | ||
0326f5a9 | 2244 | handler_chain(uprobe, regs); |
6fe50a28 | 2245 | |
8a6b1732 | 2246 | if (arch_uprobe_skip_sstep(&uprobe->arch, regs)) |
0578a970 | 2247 | goto out; |
0326f5a9 | 2248 | |
608e7427 | 2249 | if (!pre_ssout(uprobe, regs, bp_vaddr)) |
0326f5a9 | 2250 | return; |
0326f5a9 | 2251 | |
8a6b1732 | 2252 | /* arch_uprobe_skip_sstep() succeeded, or restart if can't singlestep */ |
0578a970 | 2253 | out: |
8bd87445 | 2254 | put_uprobe(uprobe); |
0326f5a9 SD |
2255 | } |
2256 | ||
2257 | /* | |
2258 | * Perform required fix-ups and disable singlestep. | |
2259 | * Allow pending signals to take effect. | |
2260 | */ | |
2261 | static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs) | |
2262 | { | |
2263 | struct uprobe *uprobe; | |
014940ba | 2264 | int err = 0; |
0326f5a9 SD |
2265 | |
2266 | uprobe = utask->active_uprobe; | |
2267 | if (utask->state == UTASK_SSTEP_ACK) | |
014940ba | 2268 | err = arch_uprobe_post_xol(&uprobe->arch, regs); |
0326f5a9 SD |
2269 | else if (utask->state == UTASK_SSTEP_TRAPPED) |
2270 | arch_uprobe_abort_xol(&uprobe->arch, regs); | |
2271 | else | |
2272 | WARN_ON_ONCE(1); | |
2273 | ||
2274 | put_uprobe(uprobe); | |
2275 | utask->active_uprobe = NULL; | |
2276 | utask->state = UTASK_RUNNING; | |
d4b3b638 | 2277 | xol_free_insn_slot(current); |
0326f5a9 SD |
2278 | |
2279 | spin_lock_irq(¤t->sighand->siglock); | |
2280 | recalc_sigpending(); /* see uprobe_deny_signal() */ | |
2281 | spin_unlock_irq(¤t->sighand->siglock); | |
014940ba ON |
2282 | |
2283 | if (unlikely(err)) { | |
2284 | uprobe_warn(current, "execute the probed insn, sending SIGILL."); | |
3cf5d076 | 2285 | force_sig(SIGILL); |
014940ba | 2286 | } |
0326f5a9 SD |
2287 | } |
2288 | ||
2289 | /* | |
1b08e907 ON |
2290 | * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag and |
2291 | * allows the thread to return from interrupt. After that handle_swbp() | |
2292 | * sets utask->active_uprobe. | |
0326f5a9 | 2293 | * |
1b08e907 ON |
2294 | * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag |
2295 | * and allows the thread to return from interrupt. | |
0326f5a9 SD |
2296 | * |
2297 | * While returning to userspace, thread notices the TIF_UPROBE flag and calls | |
2298 | * uprobe_notify_resume(). | |
2299 | */ | |
2300 | void uprobe_notify_resume(struct pt_regs *regs) | |
2301 | { | |
2302 | struct uprobe_task *utask; | |
2303 | ||
db023ea5 ON |
2304 | clear_thread_flag(TIF_UPROBE); |
2305 | ||
0326f5a9 | 2306 | utask = current->utask; |
1b08e907 | 2307 | if (utask && utask->active_uprobe) |
0326f5a9 | 2308 | handle_singlestep(utask, regs); |
1b08e907 ON |
2309 | else |
2310 | handle_swbp(regs); | |
0326f5a9 SD |
2311 | } |
2312 | ||
2313 | /* | |
2314 | * uprobe_pre_sstep_notifier gets called from interrupt context as part of | |
2315 | * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit. | |
2316 | */ | |
2317 | int uprobe_pre_sstep_notifier(struct pt_regs *regs) | |
2318 | { | |
0dfd0eb8 AA |
2319 | if (!current->mm) |
2320 | return 0; | |
2321 | ||
2322 | if (!test_bit(MMF_HAS_UPROBES, ¤t->mm->flags) && | |
2323 | (!current->utask || !current->utask->return_instances)) | |
0326f5a9 SD |
2324 | return 0; |
2325 | ||
0326f5a9 | 2326 | set_thread_flag(TIF_UPROBE); |
0326f5a9 SD |
2327 | return 1; |
2328 | } | |
2329 | ||
2330 | /* | |
2331 | * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier | |
2332 | * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep. | |
2333 | */ | |
2334 | int uprobe_post_sstep_notifier(struct pt_regs *regs) | |
2335 | { | |
2336 | struct uprobe_task *utask = current->utask; | |
2337 | ||
2338 | if (!current->mm || !utask || !utask->active_uprobe) | |
2339 | /* task is currently not uprobed */ | |
2340 | return 0; | |
2341 | ||
2342 | utask->state = UTASK_SSTEP_ACK; | |
2343 | set_thread_flag(TIF_UPROBE); | |
2344 | return 1; | |
2345 | } | |
2346 | ||
2347 | static struct notifier_block uprobe_exception_nb = { | |
2348 | .notifier_call = arch_uprobe_exception_notify, | |
2349 | .priority = INT_MAX-1, /* notified after kprobes, kgdb */ | |
2350 | }; | |
2351 | ||
aad42dd4 | 2352 | void __init uprobes_init(void) |
2b144498 SD |
2353 | { |
2354 | int i; | |
2355 | ||
66d06dff | 2356 | for (i = 0; i < UPROBES_HASH_SZ; i++) |
2b144498 | 2357 | mutex_init(&uprobes_mmap_mutex[i]); |
0326f5a9 | 2358 | |
aad42dd4 | 2359 | BUG_ON(register_die_notifier(&uprobe_exception_nb)); |
2b144498 | 2360 | } |