1 // SPDX-License-Identifier: GPL-2.0-only 1 // SPDX-License-Identifier: GPL-2.0-only
2 /* 2 /*
3 * linux/fs/exec.c 3 * linux/fs/exec.c
4 * 4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds 5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */ 6 */
7 7
8 /* 8 /*
9 * #!-checking implemented by tytso. 9 * #!-checking implemented by tytso.
10 */ 10 */
11 /* 11 /*
12 * Demand-loading implemented 01.12.91 - no ne 12 * Demand-loading implemented 01.12.91 - no need to read anything but
13 * the header into memory. The inode of the ex 13 * the header into memory. The inode of the executable is put into
14 * "current->executable", and page faults do t 14 * "current->executable", and page faults do the actual loading. Clean.
15 * 15 *
16 * Once more I can proudly say that linux stoo 16 * Once more I can proudly say that linux stood up to being changed: it
17 * was less than 2 hours work to get demand-lo 17 * was less than 2 hours work to get demand-loading completely implemented.
18 * 18 *
19 * Demand loading changed July 1993 by Eric Yo 19 * Demand loading changed July 1993 by Eric Youngdale. Use mmap instead,
20 * current->executable is only used by the pro 20 * current->executable is only used by the procfs. This allows a dispatch
21 * table to check for several different types 21 * table to check for several different types of binary formats. We keep
22 * trying until we recognize the file or we ru 22 * trying until we recognize the file or we run out of supported binary
23 * formats. 23 * formats.
24 */ 24 */
25 25
26 #include <linux/kernel_read_file.h> 26 #include <linux/kernel_read_file.h>
27 #include <linux/slab.h> 27 #include <linux/slab.h>
28 #include <linux/file.h> 28 #include <linux/file.h>
29 #include <linux/fdtable.h> 29 #include <linux/fdtable.h>
30 #include <linux/mm.h> 30 #include <linux/mm.h>
>> 31 #include <linux/vmacache.h>
31 #include <linux/stat.h> 32 #include <linux/stat.h>
32 #include <linux/fcntl.h> 33 #include <linux/fcntl.h>
33 #include <linux/swap.h> 34 #include <linux/swap.h>
34 #include <linux/string.h> 35 #include <linux/string.h>
35 #include <linux/init.h> 36 #include <linux/init.h>
36 #include <linux/sched/mm.h> 37 #include <linux/sched/mm.h>
37 #include <linux/sched/coredump.h> 38 #include <linux/sched/coredump.h>
38 #include <linux/sched/signal.h> 39 #include <linux/sched/signal.h>
39 #include <linux/sched/numa_balancing.h> 40 #include <linux/sched/numa_balancing.h>
40 #include <linux/sched/task.h> 41 #include <linux/sched/task.h>
41 #include <linux/pagemap.h> 42 #include <linux/pagemap.h>
42 #include <linux/perf_event.h> 43 #include <linux/perf_event.h>
43 #include <linux/highmem.h> 44 #include <linux/highmem.h>
44 #include <linux/spinlock.h> 45 #include <linux/spinlock.h>
45 #include <linux/key.h> 46 #include <linux/key.h>
46 #include <linux/personality.h> 47 #include <linux/personality.h>
47 #include <linux/binfmts.h> 48 #include <linux/binfmts.h>
48 #include <linux/utsname.h> 49 #include <linux/utsname.h>
49 #include <linux/pid_namespace.h> 50 #include <linux/pid_namespace.h>
50 #include <linux/module.h> 51 #include <linux/module.h>
51 #include <linux/namei.h> 52 #include <linux/namei.h>
52 #include <linux/mount.h> 53 #include <linux/mount.h>
53 #include <linux/security.h> 54 #include <linux/security.h>
54 #include <linux/syscalls.h> 55 #include <linux/syscalls.h>
55 #include <linux/tsacct_kern.h> 56 #include <linux/tsacct_kern.h>
56 #include <linux/cn_proc.h> 57 #include <linux/cn_proc.h>
57 #include <linux/audit.h> 58 #include <linux/audit.h>
>> 59 #include <linux/tracehook.h>
58 #include <linux/kmod.h> 60 #include <linux/kmod.h>
59 #include <linux/fsnotify.h> 61 #include <linux/fsnotify.h>
60 #include <linux/fs_struct.h> 62 #include <linux/fs_struct.h>
61 #include <linux/oom.h> 63 #include <linux/oom.h>
62 #include <linux/compat.h> 64 #include <linux/compat.h>
63 #include <linux/vmalloc.h> 65 #include <linux/vmalloc.h>
64 #include <linux/io_uring.h> 66 #include <linux/io_uring.h>
65 #include <linux/syscall_user_dispatch.h> 67 #include <linux/syscall_user_dispatch.h>
66 #include <linux/coredump.h> 68 #include <linux/coredump.h>
67 #include <linux/time_namespace.h> <<
68 #include <linux/user_events.h> <<
69 #include <linux/rseq.h> <<
70 #include <linux/ksm.h> <<
71 69
72 #include <linux/uaccess.h> 70 #include <linux/uaccess.h>
73 #include <asm/mmu_context.h> 71 #include <asm/mmu_context.h>
74 #include <asm/tlb.h> 72 #include <asm/tlb.h>
75 73
76 #include <trace/events/task.h> 74 #include <trace/events/task.h>
77 #include "internal.h" 75 #include "internal.h"
78 76
79 #include <trace/events/sched.h> 77 #include <trace/events/sched.h>
80 78
81 static int bprm_creds_from_file(struct linux_b 79 static int bprm_creds_from_file(struct linux_binprm *bprm);
82 80
83 int suid_dumpable = 0; 81 int suid_dumpable = 0;
84 82
85 static LIST_HEAD(formats); 83 static LIST_HEAD(formats);
86 static DEFINE_RWLOCK(binfmt_lock); 84 static DEFINE_RWLOCK(binfmt_lock);
87 85
88 void __register_binfmt(struct linux_binfmt * f 86 void __register_binfmt(struct linux_binfmt * fmt, int insert)
89 { 87 {
90 write_lock(&binfmt_lock); 88 write_lock(&binfmt_lock);
91 insert ? list_add(&fmt->lh, &formats) 89 insert ? list_add(&fmt->lh, &formats) :
92 list_add_tail(&fmt->lh, &form 90 list_add_tail(&fmt->lh, &formats);
93 write_unlock(&binfmt_lock); 91 write_unlock(&binfmt_lock);
94 } 92 }
95 93
96 EXPORT_SYMBOL(__register_binfmt); 94 EXPORT_SYMBOL(__register_binfmt);
97 95
98 void unregister_binfmt(struct linux_binfmt * f 96 void unregister_binfmt(struct linux_binfmt * fmt)
99 { 97 {
100 write_lock(&binfmt_lock); 98 write_lock(&binfmt_lock);
101 list_del(&fmt->lh); 99 list_del(&fmt->lh);
102 write_unlock(&binfmt_lock); 100 write_unlock(&binfmt_lock);
103 } 101 }
104 102
105 EXPORT_SYMBOL(unregister_binfmt); 103 EXPORT_SYMBOL(unregister_binfmt);
106 104
107 static inline void put_binfmt(struct linux_bin 105 static inline void put_binfmt(struct linux_binfmt * fmt)
108 { 106 {
109 module_put(fmt->module); 107 module_put(fmt->module);
110 } 108 }
111 109
112 bool path_noexec(const struct path *path) 110 bool path_noexec(const struct path *path)
113 { 111 {
114 return (path->mnt->mnt_flags & MNT_NOE 112 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
115 (path->mnt->mnt_sb->s_iflags & 113 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
116 } 114 }
117 115
118 #ifdef CONFIG_USELIB 116 #ifdef CONFIG_USELIB
119 /* 117 /*
120 * Note that a shared library must be both rea 118 * Note that a shared library must be both readable and executable due to
121 * security reasons. 119 * security reasons.
122 * 120 *
123 * Also note that we take the address to load !! 121 * Also note that we take the address to load from from the file itself.
124 */ 122 */
125 SYSCALL_DEFINE1(uselib, const char __user *, l 123 SYSCALL_DEFINE1(uselib, const char __user *, library)
126 { 124 {
127 struct linux_binfmt *fmt; 125 struct linux_binfmt *fmt;
128 struct file *file; 126 struct file *file;
129 struct filename *tmp = getname(library 127 struct filename *tmp = getname(library);
130 int error = PTR_ERR(tmp); 128 int error = PTR_ERR(tmp);
131 static const struct open_flags uselib_ 129 static const struct open_flags uselib_flags = {
132 .open_flag = O_LARGEFILE | O_R !! 130 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
133 .acc_mode = MAY_READ | MAY_EXE 131 .acc_mode = MAY_READ | MAY_EXEC,
134 .intent = LOOKUP_OPEN, 132 .intent = LOOKUP_OPEN,
135 .lookup_flags = LOOKUP_FOLLOW, 133 .lookup_flags = LOOKUP_FOLLOW,
136 }; 134 };
137 135
138 if (IS_ERR(tmp)) 136 if (IS_ERR(tmp))
139 goto out; 137 goto out;
140 138
141 file = do_filp_open(AT_FDCWD, tmp, &us 139 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
142 putname(tmp); 140 putname(tmp);
143 error = PTR_ERR(file); 141 error = PTR_ERR(file);
144 if (IS_ERR(file)) 142 if (IS_ERR(file))
145 goto out; 143 goto out;
146 144
147 /* 145 /*
148 * Check do_open_execat() for an expla !! 146 * may_open() has already checked for this, so it should be
>> 147 * impossible to trip now. But we need to be extra cautious
>> 148 * and check again at the very end too.
149 */ 149 */
150 error = -EACCES; 150 error = -EACCES;
151 if (WARN_ON_ONCE(!S_ISREG(file_inode(f !! 151 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
152 path_noexec(&file->f_path)) !! 152 path_noexec(&file->f_path)))
153 goto exit; 153 goto exit;
154 154
>> 155 fsnotify_open(file);
>> 156
155 error = -ENOEXEC; 157 error = -ENOEXEC;
156 158
157 read_lock(&binfmt_lock); 159 read_lock(&binfmt_lock);
158 list_for_each_entry(fmt, &formats, lh) 160 list_for_each_entry(fmt, &formats, lh) {
159 if (!fmt->load_shlib) 161 if (!fmt->load_shlib)
160 continue; 162 continue;
161 if (!try_module_get(fmt->modul 163 if (!try_module_get(fmt->module))
162 continue; 164 continue;
163 read_unlock(&binfmt_lock); 165 read_unlock(&binfmt_lock);
164 error = fmt->load_shlib(file); 166 error = fmt->load_shlib(file);
165 read_lock(&binfmt_lock); 167 read_lock(&binfmt_lock);
166 put_binfmt(fmt); 168 put_binfmt(fmt);
167 if (error != -ENOEXEC) 169 if (error != -ENOEXEC)
168 break; 170 break;
169 } 171 }
170 read_unlock(&binfmt_lock); 172 read_unlock(&binfmt_lock);
171 exit: 173 exit:
172 fput(file); 174 fput(file);
173 out: 175 out:
174 return error; !! 176 return error;
175 } 177 }
176 #endif /* #ifdef CONFIG_USELIB */ 178 #endif /* #ifdef CONFIG_USELIB */
177 179
178 #ifdef CONFIG_MMU 180 #ifdef CONFIG_MMU
179 /* 181 /*
180 * The nascent bprm->mm is not visible until e 182 * The nascent bprm->mm is not visible until exec_mmap() but it can
181 * use a lot of memory, account these pages in 183 * use a lot of memory, account these pages in current->mm temporary
182 * for oom_badness()->get_mm_rss(). Once exec 184 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
183 * change the counter back via acct_arg_size(0 185 * change the counter back via acct_arg_size(0).
184 */ 186 */
185 static void acct_arg_size(struct linux_binprm 187 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
186 { 188 {
187 struct mm_struct *mm = current->mm; 189 struct mm_struct *mm = current->mm;
188 long diff = (long)(pages - bprm->vma_p 190 long diff = (long)(pages - bprm->vma_pages);
189 191
190 if (!mm || !diff) 192 if (!mm || !diff)
191 return; 193 return;
192 194
193 bprm->vma_pages = pages; 195 bprm->vma_pages = pages;
194 add_mm_counter(mm, MM_ANONPAGES, diff) 196 add_mm_counter(mm, MM_ANONPAGES, diff);
195 } 197 }
196 198
197 static struct page *get_arg_page(struct linux_ 199 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
198 int write) 200 int write)
199 { 201 {
200 struct page *page; 202 struct page *page;
201 struct vm_area_struct *vma = bprm->vma <<
202 struct mm_struct *mm = bprm->mm; <<
203 int ret; 203 int ret;
>> 204 unsigned int gup_flags = FOLL_FORCE;
204 205
205 /* !! 206 #ifdef CONFIG_STACK_GROWSUP
206 * Avoid relying on expanding the stac !! 207 if (write) {
207 * does not work for STACK_GROWSUP any !! 208 ret = expand_downwards(bprm->vma, pos);
208 * by hand ahead of time. !! 209 if (ret < 0)
209 */ <<
210 if (write && pos < vma->vm_start) { <<
211 mmap_write_lock(mm); <<
212 ret = expand_downwards(vma, po <<
213 if (unlikely(ret < 0)) { <<
214 mmap_write_unlock(mm); <<
215 return NULL; 210 return NULL;
216 } !! 211 }
217 mmap_write_downgrade(mm); !! 212 #endif
218 } else !! 213
219 mmap_read_lock(mm); !! 214 if (write)
>> 215 gup_flags |= FOLL_WRITE;
220 216
221 /* 217 /*
222 * We are doing an exec(). 'current' 218 * We are doing an exec(). 'current' is the process
223 * doing the exec and 'mm' is the new !! 219 * doing the exec and bprm->mm is the new process's mm.
224 */ 220 */
225 ret = get_user_pages_remote(mm, pos, 1 !! 221 mmap_read_lock(bprm->mm);
226 write ? FOLL_WRITE : 0 !! 222 ret = get_user_pages_remote(bprm->mm, pos, 1, gup_flags,
227 &page, NULL); !! 223 &page, NULL, NULL);
228 mmap_read_unlock(mm); !! 224 mmap_read_unlock(bprm->mm);
229 if (ret <= 0) 225 if (ret <= 0)
230 return NULL; 226 return NULL;
231 227
232 if (write) 228 if (write)
233 acct_arg_size(bprm, vma_pages( !! 229 acct_arg_size(bprm, vma_pages(bprm->vma));
234 230
235 return page; 231 return page;
236 } 232 }
237 233
238 static void put_arg_page(struct page *page) 234 static void put_arg_page(struct page *page)
239 { 235 {
240 put_page(page); 236 put_page(page);
241 } 237 }
242 238
243 static void free_arg_pages(struct linux_binprm 239 static void free_arg_pages(struct linux_binprm *bprm)
244 { 240 {
245 } 241 }
246 242
247 static void flush_arg_page(struct linux_binprm 243 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
248 struct page *page) 244 struct page *page)
249 { 245 {
250 flush_cache_page(bprm->vma, pos, page_ 246 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
251 } 247 }
252 248
253 static int __bprm_mm_init(struct linux_binprm 249 static int __bprm_mm_init(struct linux_binprm *bprm)
254 { 250 {
255 int err; 251 int err;
256 struct vm_area_struct *vma = NULL; 252 struct vm_area_struct *vma = NULL;
257 struct mm_struct *mm = bprm->mm; 253 struct mm_struct *mm = bprm->mm;
258 254
259 bprm->vma = vma = vm_area_alloc(mm); 255 bprm->vma = vma = vm_area_alloc(mm);
260 if (!vma) 256 if (!vma)
261 return -ENOMEM; 257 return -ENOMEM;
262 vma_set_anonymous(vma); 258 vma_set_anonymous(vma);
263 259
264 if (mmap_write_lock_killable(mm)) { 260 if (mmap_write_lock_killable(mm)) {
265 err = -EINTR; 261 err = -EINTR;
266 goto err_free; 262 goto err_free;
267 } 263 }
268 264
269 /* 265 /*
270 * Need to be called with mmap write l <<
271 * held, to avoid race with ksmd. <<
272 */ <<
273 err = ksm_execve(mm); <<
274 if (err) <<
275 goto err_ksm; <<
276 <<
277 /* <<
278 * Place the stack at the largest stac 266 * Place the stack at the largest stack address the architecture
279 * supports. Later, we'll move this to 267 * supports. Later, we'll move this to an appropriate place. We don't
280 * use STACK_TOP because that can depe 268 * use STACK_TOP because that can depend on attributes which aren't
281 * configured yet. 269 * configured yet.
282 */ 270 */
283 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK 271 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
284 vma->vm_end = STACK_TOP_MAX; 272 vma->vm_end = STACK_TOP_MAX;
285 vma->vm_start = vma->vm_end - PAGE_SIZ 273 vma->vm_start = vma->vm_end - PAGE_SIZE;
286 vm_flags_init(vma, VM_SOFTDIRTY | VM_S !! 274 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
287 vma->vm_page_prot = vm_get_page_prot(v 275 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
288 276
289 err = insert_vm_struct(mm, vma); 277 err = insert_vm_struct(mm, vma);
290 if (err) 278 if (err)
291 goto err; 279 goto err;
292 280
293 mm->stack_vm = mm->total_vm = 1; 281 mm->stack_vm = mm->total_vm = 1;
294 mmap_write_unlock(mm); 282 mmap_write_unlock(mm);
295 bprm->p = vma->vm_end - sizeof(void *) 283 bprm->p = vma->vm_end - sizeof(void *);
296 return 0; 284 return 0;
297 err: 285 err:
298 ksm_exit(mm); <<
299 err_ksm: <<
300 mmap_write_unlock(mm); 286 mmap_write_unlock(mm);
301 err_free: 287 err_free:
302 bprm->vma = NULL; 288 bprm->vma = NULL;
303 vm_area_free(vma); 289 vm_area_free(vma);
304 return err; 290 return err;
305 } 291 }
306 292
307 static bool valid_arg_len(struct linux_binprm 293 static bool valid_arg_len(struct linux_binprm *bprm, long len)
308 { 294 {
309 return len <= MAX_ARG_STRLEN; 295 return len <= MAX_ARG_STRLEN;
310 } 296 }
311 297
312 #else 298 #else
313 299
314 static inline void acct_arg_size(struct linux_ 300 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
315 { 301 {
316 } 302 }
317 303
318 static struct page *get_arg_page(struct linux_ 304 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
319 int write) 305 int write)
320 { 306 {
321 struct page *page; 307 struct page *page;
322 308
323 page = bprm->page[pos / PAGE_SIZE]; 309 page = bprm->page[pos / PAGE_SIZE];
324 if (!page && write) { 310 if (!page && write) {
325 page = alloc_page(GFP_HIGHUSER 311 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
326 if (!page) 312 if (!page)
327 return NULL; 313 return NULL;
328 bprm->page[pos / PAGE_SIZE] = 314 bprm->page[pos / PAGE_SIZE] = page;
329 } 315 }
330 316
331 return page; 317 return page;
332 } 318 }
333 319
334 static void put_arg_page(struct page *page) 320 static void put_arg_page(struct page *page)
335 { 321 {
336 } 322 }
337 323
338 static void free_arg_page(struct linux_binprm 324 static void free_arg_page(struct linux_binprm *bprm, int i)
339 { 325 {
340 if (bprm->page[i]) { 326 if (bprm->page[i]) {
341 __free_page(bprm->page[i]); 327 __free_page(bprm->page[i]);
342 bprm->page[i] = NULL; 328 bprm->page[i] = NULL;
343 } 329 }
344 } 330 }
345 331
346 static void free_arg_pages(struct linux_binprm 332 static void free_arg_pages(struct linux_binprm *bprm)
347 { 333 {
348 int i; 334 int i;
349 335
350 for (i = 0; i < MAX_ARG_PAGES; i++) 336 for (i = 0; i < MAX_ARG_PAGES; i++)
351 free_arg_page(bprm, i); 337 free_arg_page(bprm, i);
352 } 338 }
353 339
354 static void flush_arg_page(struct linux_binprm 340 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
355 struct page *page) 341 struct page *page)
356 { 342 {
357 } 343 }
358 344
359 static int __bprm_mm_init(struct linux_binprm 345 static int __bprm_mm_init(struct linux_binprm *bprm)
360 { 346 {
361 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - 347 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
362 return 0; 348 return 0;
363 } 349 }
364 350
365 static bool valid_arg_len(struct linux_binprm 351 static bool valid_arg_len(struct linux_binprm *bprm, long len)
366 { 352 {
367 return len <= bprm->p; 353 return len <= bprm->p;
368 } 354 }
369 355
370 #endif /* CONFIG_MMU */ 356 #endif /* CONFIG_MMU */
371 357
372 /* 358 /*
373 * Create a new mm_struct and populate it with 359 * Create a new mm_struct and populate it with a temporary stack
374 * vm_area_struct. We don't have enough conte 360 * vm_area_struct. We don't have enough context at this point to set the stack
375 * flags, permissions, and offset, so we use t 361 * flags, permissions, and offset, so we use temporary values. We'll update
376 * them later in setup_arg_pages(). 362 * them later in setup_arg_pages().
377 */ 363 */
378 static int bprm_mm_init(struct linux_binprm *b 364 static int bprm_mm_init(struct linux_binprm *bprm)
379 { 365 {
380 int err; 366 int err;
381 struct mm_struct *mm = NULL; 367 struct mm_struct *mm = NULL;
382 368
383 bprm->mm = mm = mm_alloc(); 369 bprm->mm = mm = mm_alloc();
384 err = -ENOMEM; 370 err = -ENOMEM;
385 if (!mm) 371 if (!mm)
386 goto err; 372 goto err;
387 373
388 /* Save current stack limit for all ca 374 /* Save current stack limit for all calculations made during exec. */
389 task_lock(current->group_leader); 375 task_lock(current->group_leader);
390 bprm->rlim_stack = current->signal->rl 376 bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
391 task_unlock(current->group_leader); 377 task_unlock(current->group_leader);
392 378
393 err = __bprm_mm_init(bprm); 379 err = __bprm_mm_init(bprm);
394 if (err) 380 if (err)
395 goto err; 381 goto err;
396 382
397 return 0; 383 return 0;
398 384
399 err: 385 err:
400 if (mm) { 386 if (mm) {
401 bprm->mm = NULL; 387 bprm->mm = NULL;
402 mmdrop(mm); 388 mmdrop(mm);
403 } 389 }
404 390
405 return err; 391 return err;
406 } 392 }
407 393
408 struct user_arg_ptr { 394 struct user_arg_ptr {
409 #ifdef CONFIG_COMPAT 395 #ifdef CONFIG_COMPAT
410 bool is_compat; 396 bool is_compat;
411 #endif 397 #endif
412 union { 398 union {
413 const char __user *const __use 399 const char __user *const __user *native;
414 #ifdef CONFIG_COMPAT 400 #ifdef CONFIG_COMPAT
415 const compat_uptr_t __user *co 401 const compat_uptr_t __user *compat;
416 #endif 402 #endif
417 } ptr; 403 } ptr;
418 }; 404 };
419 405
420 static const char __user *get_user_arg_ptr(str 406 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
421 { 407 {
422 const char __user *native; 408 const char __user *native;
423 409
424 #ifdef CONFIG_COMPAT 410 #ifdef CONFIG_COMPAT
425 if (unlikely(argv.is_compat)) { 411 if (unlikely(argv.is_compat)) {
426 compat_uptr_t compat; 412 compat_uptr_t compat;
427 413
428 if (get_user(compat, argv.ptr. 414 if (get_user(compat, argv.ptr.compat + nr))
429 return ERR_PTR(-EFAULT 415 return ERR_PTR(-EFAULT);
430 416
431 return compat_ptr(compat); 417 return compat_ptr(compat);
432 } 418 }
433 #endif 419 #endif
434 420
435 if (get_user(native, argv.ptr.native + 421 if (get_user(native, argv.ptr.native + nr))
436 return ERR_PTR(-EFAULT); 422 return ERR_PTR(-EFAULT);
437 423
438 return native; 424 return native;
439 } 425 }
440 426
441 /* 427 /*
442 * count() counts the number of strings in arr 428 * count() counts the number of strings in array ARGV.
443 */ 429 */
444 static int count(struct user_arg_ptr argv, int 430 static int count(struct user_arg_ptr argv, int max)
445 { 431 {
446 int i = 0; 432 int i = 0;
447 433
448 if (argv.ptr.native != NULL) { 434 if (argv.ptr.native != NULL) {
449 for (;;) { 435 for (;;) {
450 const char __user *p = 436 const char __user *p = get_user_arg_ptr(argv, i);
451 437
452 if (!p) 438 if (!p)
453 break; 439 break;
454 440
455 if (IS_ERR(p)) 441 if (IS_ERR(p))
456 return -EFAULT 442 return -EFAULT;
457 443
458 if (i >= max) 444 if (i >= max)
459 return -E2BIG; 445 return -E2BIG;
460 ++i; 446 ++i;
461 447
462 if (fatal_signal_pendi 448 if (fatal_signal_pending(current))
463 return -ERESTA 449 return -ERESTARTNOHAND;
464 cond_resched(); 450 cond_resched();
465 } 451 }
466 } 452 }
467 return i; 453 return i;
468 } 454 }
469 455
470 static int count_strings_kernel(const char *co 456 static int count_strings_kernel(const char *const *argv)
471 { 457 {
472 int i; 458 int i;
473 459
474 if (!argv) 460 if (!argv)
475 return 0; 461 return 0;
476 462
477 for (i = 0; argv[i]; ++i) { 463 for (i = 0; argv[i]; ++i) {
478 if (i >= MAX_ARG_STRINGS) 464 if (i >= MAX_ARG_STRINGS)
479 return -E2BIG; 465 return -E2BIG;
480 if (fatal_signal_pending(curre 466 if (fatal_signal_pending(current))
481 return -ERESTARTNOHAND 467 return -ERESTARTNOHAND;
482 cond_resched(); 468 cond_resched();
483 } 469 }
484 return i; 470 return i;
485 } 471 }
486 472
487 static inline int bprm_set_stack_limit(struct <<
488 unsigne <<
489 { <<
490 #ifdef CONFIG_MMU <<
491 /* Avoid a pathological bprm->p. */ <<
492 if (bprm->p < limit) <<
493 return -E2BIG; <<
494 bprm->argmin = bprm->p - limit; <<
495 #endif <<
496 return 0; <<
497 } <<
498 static inline bool bprm_hit_stack_limit(struct <<
499 { <<
500 #ifdef CONFIG_MMU <<
501 return bprm->p < bprm->argmin; <<
502 #else <<
503 return false; <<
504 #endif <<
505 } <<
506 <<
507 /* <<
508 * Calculate bprm->argmin from: <<
509 * - _STK_LIM <<
510 * - ARG_MAX <<
511 * - bprm->rlim_stack.rlim_cur <<
512 * - bprm->argc <<
513 * - bprm->envc <<
514 * - bprm->p <<
515 */ <<
516 static int bprm_stack_limits(struct linux_binp 473 static int bprm_stack_limits(struct linux_binprm *bprm)
517 { 474 {
518 unsigned long limit, ptr_size; 475 unsigned long limit, ptr_size;
519 476
520 /* 477 /*
521 * Limit to 1/4 of the max stack size 478 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
522 * (whichever is smaller) for the argv 479 * (whichever is smaller) for the argv+env strings.
523 * This ensures that: 480 * This ensures that:
524 * - the remaining binfmt code will n 481 * - the remaining binfmt code will not run out of stack space,
525 * - the program will have a reasonab 482 * - the program will have a reasonable amount of stack left
526 * to work from. 483 * to work from.
527 */ 484 */
528 limit = _STK_LIM / 4 * 3; 485 limit = _STK_LIM / 4 * 3;
529 limit = min(limit, bprm->rlim_stack.rl 486 limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
530 /* 487 /*
531 * We've historically supported up to 488 * We've historically supported up to 32 pages (ARG_MAX)
532 * of argument strings even with small 489 * of argument strings even with small stacks
533 */ 490 */
534 limit = max_t(unsigned long, limit, AR 491 limit = max_t(unsigned long, limit, ARG_MAX);
535 /* Reject totally pathological counts. <<
536 if (bprm->argc < 0 || bprm->envc < 0) <<
537 return -E2BIG; <<
538 /* 492 /*
539 * We must account for the size of all 493 * We must account for the size of all the argv and envp pointers to
540 * the argv and envp strings, since th 494 * the argv and envp strings, since they will also take up space in
541 * the stack. They aren't stored until 495 * the stack. They aren't stored until much later when we can't
542 * signal to the parent that the child 496 * signal to the parent that the child has run out of stack space.
543 * Instead, calculate it here so it's 497 * Instead, calculate it here so it's possible to fail gracefully.
544 * 498 *
545 * In the case of argc = 0, make sure 499 * In the case of argc = 0, make sure there is space for adding a
546 * empty string (which will bump argc 500 * empty string (which will bump argc to 1), to ensure confused
547 * userspace programs don't start proc 501 * userspace programs don't start processing from argv[1], thinking
548 * argc can never be 0, to keep them f 502 * argc can never be 0, to keep them from walking envp by accident.
549 * See do_execveat_common(). 503 * See do_execveat_common().
550 */ 504 */
551 if (check_add_overflow(max(bprm->argc, !! 505 ptr_size = (max(bprm->argc, 1) + bprm->envc) * sizeof(void *);
552 check_mul_overflow(ptr_size, sizeo <<
553 return -E2BIG; <<
554 if (limit <= ptr_size) 506 if (limit <= ptr_size)
555 return -E2BIG; 507 return -E2BIG;
556 limit -= ptr_size; 508 limit -= ptr_size;
557 509
558 return bprm_set_stack_limit(bprm, limi !! 510 bprm->argmin = bprm->p - limit;
>> 511 return 0;
559 } 512 }
560 513
561 /* 514 /*
562 * 'copy_strings()' copies argument/environmen 515 * 'copy_strings()' copies argument/environment strings from the old
563 * processes's memory to the new process's sta 516 * processes's memory to the new process's stack. The call to get_user_pages()
564 * ensures the destination page is created and 517 * ensures the destination page is created and not swapped out.
565 */ 518 */
566 static int copy_strings(int argc, struct user_ 519 static int copy_strings(int argc, struct user_arg_ptr argv,
567 struct linux_binprm *b 520 struct linux_binprm *bprm)
568 { 521 {
569 struct page *kmapped_page = NULL; 522 struct page *kmapped_page = NULL;
570 char *kaddr = NULL; 523 char *kaddr = NULL;
571 unsigned long kpos = 0; 524 unsigned long kpos = 0;
572 int ret; 525 int ret;
573 526
574 while (argc-- > 0) { 527 while (argc-- > 0) {
575 const char __user *str; 528 const char __user *str;
576 int len; 529 int len;
577 unsigned long pos; 530 unsigned long pos;
578 531
579 ret = -EFAULT; 532 ret = -EFAULT;
580 str = get_user_arg_ptr(argv, a 533 str = get_user_arg_ptr(argv, argc);
581 if (IS_ERR(str)) 534 if (IS_ERR(str))
582 goto out; 535 goto out;
583 536
584 len = strnlen_user(str, MAX_AR 537 len = strnlen_user(str, MAX_ARG_STRLEN);
585 if (!len) 538 if (!len)
586 goto out; 539 goto out;
587 540
588 ret = -E2BIG; 541 ret = -E2BIG;
589 if (!valid_arg_len(bprm, len)) 542 if (!valid_arg_len(bprm, len))
590 goto out; 543 goto out;
591 544
592 /* We're going to work our way !! 545 /* We're going to work our way backwords. */
593 pos = bprm->p; 546 pos = bprm->p;
594 str += len; 547 str += len;
595 bprm->p -= len; 548 bprm->p -= len;
596 if (bprm_hit_stack_limit(bprm) !! 549 #ifdef CONFIG_MMU
>> 550 if (bprm->p < bprm->argmin)
597 goto out; 551 goto out;
>> 552 #endif
598 553
599 while (len > 0) { 554 while (len > 0) {
600 int offset, bytes_to_c 555 int offset, bytes_to_copy;
601 556
602 if (fatal_signal_pendi 557 if (fatal_signal_pending(current)) {
603 ret = -ERESTAR 558 ret = -ERESTARTNOHAND;
604 goto out; 559 goto out;
605 } 560 }
606 cond_resched(); 561 cond_resched();
607 562
608 offset = pos % PAGE_SI 563 offset = pos % PAGE_SIZE;
609 if (offset == 0) 564 if (offset == 0)
610 offset = PAGE_ 565 offset = PAGE_SIZE;
611 566
612 bytes_to_copy = offset 567 bytes_to_copy = offset;
613 if (bytes_to_copy > le 568 if (bytes_to_copy > len)
614 bytes_to_copy 569 bytes_to_copy = len;
615 570
616 offset -= bytes_to_cop 571 offset -= bytes_to_copy;
617 pos -= bytes_to_copy; 572 pos -= bytes_to_copy;
618 str -= bytes_to_copy; 573 str -= bytes_to_copy;
619 len -= bytes_to_copy; 574 len -= bytes_to_copy;
620 575
621 if (!kmapped_page || k 576 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
622 struct page *p 577 struct page *page;
623 578
624 page = get_arg 579 page = get_arg_page(bprm, pos, 1);
625 if (!page) { 580 if (!page) {
626 ret = 581 ret = -E2BIG;
627 goto o 582 goto out;
628 } 583 }
629 584
630 if (kmapped_pa 585 if (kmapped_page) {
631 flush_ 586 flush_dcache_page(kmapped_page);
632 kunmap !! 587 kunmap(kmapped_page);
633 put_ar 588 put_arg_page(kmapped_page);
634 } 589 }
635 kmapped_page = 590 kmapped_page = page;
636 kaddr = kmap_l !! 591 kaddr = kmap(kmapped_page);
637 kpos = pos & P 592 kpos = pos & PAGE_MASK;
638 flush_arg_page 593 flush_arg_page(bprm, kpos, kmapped_page);
639 } 594 }
640 if (copy_from_user(kad 595 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
641 ret = -EFAULT; 596 ret = -EFAULT;
642 goto out; 597 goto out;
643 } 598 }
644 } 599 }
645 } 600 }
646 ret = 0; 601 ret = 0;
647 out: 602 out:
648 if (kmapped_page) { 603 if (kmapped_page) {
649 flush_dcache_page(kmapped_page 604 flush_dcache_page(kmapped_page);
650 kunmap_local(kaddr); !! 605 kunmap(kmapped_page);
651 put_arg_page(kmapped_page); 606 put_arg_page(kmapped_page);
652 } 607 }
653 return ret; 608 return ret;
654 } 609 }
655 610
656 /* 611 /*
657 * Copy and argument/environment string from t 612 * Copy and argument/environment string from the kernel to the processes stack.
658 */ 613 */
659 int copy_string_kernel(const char *arg, struct 614 int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
660 { 615 {
661 int len = strnlen(arg, MAX_ARG_STRLEN) 616 int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
662 unsigned long pos = bprm->p; 617 unsigned long pos = bprm->p;
663 618
664 if (len == 0) 619 if (len == 0)
665 return -EFAULT; 620 return -EFAULT;
666 if (!valid_arg_len(bprm, len)) 621 if (!valid_arg_len(bprm, len))
667 return -E2BIG; 622 return -E2BIG;
668 623
669 /* We're going to work our way backwar 624 /* We're going to work our way backwards. */
670 arg += len; 625 arg += len;
671 bprm->p -= len; 626 bprm->p -= len;
672 if (bprm_hit_stack_limit(bprm)) !! 627 if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
673 return -E2BIG; 628 return -E2BIG;
674 629
675 while (len > 0) { 630 while (len > 0) {
676 unsigned int bytes_to_copy = m 631 unsigned int bytes_to_copy = min_t(unsigned int, len,
677 min_not_zero(o 632 min_not_zero(offset_in_page(pos), PAGE_SIZE));
678 struct page *page; 633 struct page *page;
>> 634 char *kaddr;
679 635
680 pos -= bytes_to_copy; 636 pos -= bytes_to_copy;
681 arg -= bytes_to_copy; 637 arg -= bytes_to_copy;
682 len -= bytes_to_copy; 638 len -= bytes_to_copy;
683 639
684 page = get_arg_page(bprm, pos, 640 page = get_arg_page(bprm, pos, 1);
685 if (!page) 641 if (!page)
686 return -E2BIG; 642 return -E2BIG;
>> 643 kaddr = kmap_atomic(page);
687 flush_arg_page(bprm, pos & PAG 644 flush_arg_page(bprm, pos & PAGE_MASK, page);
688 memcpy_to_page(page, offset_in !! 645 memcpy(kaddr + offset_in_page(pos), arg, bytes_to_copy);
>> 646 flush_dcache_page(page);
>> 647 kunmap_atomic(kaddr);
689 put_arg_page(page); 648 put_arg_page(page);
690 } 649 }
691 650
692 return 0; 651 return 0;
693 } 652 }
694 EXPORT_SYMBOL(copy_string_kernel); 653 EXPORT_SYMBOL(copy_string_kernel);
695 654
696 static int copy_strings_kernel(int argc, const 655 static int copy_strings_kernel(int argc, const char *const *argv,
697 struct linux_bi 656 struct linux_binprm *bprm)
698 { 657 {
699 while (argc-- > 0) { 658 while (argc-- > 0) {
700 int ret = copy_string_kernel(a 659 int ret = copy_string_kernel(argv[argc], bprm);
701 if (ret < 0) 660 if (ret < 0)
702 return ret; 661 return ret;
703 if (fatal_signal_pending(curre 662 if (fatal_signal_pending(current))
704 return -ERESTARTNOHAND 663 return -ERESTARTNOHAND;
705 cond_resched(); 664 cond_resched();
706 } 665 }
707 return 0; 666 return 0;
708 } 667 }
709 668
710 #ifdef CONFIG_MMU 669 #ifdef CONFIG_MMU
711 670
712 /* 671 /*
>> 672 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
>> 673 * the binfmt code determines where the new stack should reside, we shift it to
>> 674 * its final location. The process proceeds as follows:
>> 675 *
>> 676 * 1) Use shift to calculate the new vma endpoints.
>> 677 * 2) Extend vma to cover both the old and new ranges. This ensures the
>> 678 * arguments passed to subsequent functions are consistent.
>> 679 * 3) Move vma's page tables to the new range.
>> 680 * 4) Free up any cleared pgd range.
>> 681 * 5) Shrink the vma to cover only the new range.
>> 682 */
>> 683 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
>> 684 {
>> 685 struct mm_struct *mm = vma->vm_mm;
>> 686 unsigned long old_start = vma->vm_start;
>> 687 unsigned long old_end = vma->vm_end;
>> 688 unsigned long length = old_end - old_start;
>> 689 unsigned long new_start = old_start - shift;
>> 690 unsigned long new_end = old_end - shift;
>> 691 struct mmu_gather tlb;
>> 692
>> 693 BUG_ON(new_start > new_end);
>> 694
>> 695 /*
>> 696 * ensure there are no vmas between where we want to go
>> 697 * and where we are
>> 698 */
>> 699 if (vma != find_vma(mm, new_start))
>> 700 return -EFAULT;
>> 701
>> 702 /*
>> 703 * cover the whole range: [new_start, old_end)
>> 704 */
>> 705 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
>> 706 return -ENOMEM;
>> 707
>> 708 /*
>> 709 * move the page tables downwards, on failure we rely on
>> 710 * process cleanup to remove whatever mess we made.
>> 711 */
>> 712 if (length != move_page_tables(vma, old_start,
>> 713 vma, new_start, length, false))
>> 714 return -ENOMEM;
>> 715
>> 716 lru_add_drain();
>> 717 tlb_gather_mmu(&tlb, mm);
>> 718 if (new_end > old_start) {
>> 719 /*
>> 720 * when the old and new regions overlap clear from new_end.
>> 721 */
>> 722 free_pgd_range(&tlb, new_end, old_end, new_end,
>> 723 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
>> 724 } else {
>> 725 /*
>> 726 * otherwise, clean from old_start; this is done to not touch
>> 727 * the address space in [new_end, old_start) some architectures
>> 728 * have constraints on va-space that make this illegal (IA64) -
>> 729 * for the others its just a little faster.
>> 730 */
>> 731 free_pgd_range(&tlb, old_start, old_end, new_end,
>> 732 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
>> 733 }
>> 734 tlb_finish_mmu(&tlb);
>> 735
>> 736 /*
>> 737 * Shrink the vma to just the new range. Always succeeds.
>> 738 */
>> 739 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
>> 740
>> 741 return 0;
>> 742 }
>> 743
>> 744 /*
713 * Finalizes the stack vm_area_struct. The fla 745 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
714 * the stack is optionally relocated, and some 746 * the stack is optionally relocated, and some extra space is added.
715 */ 747 */
716 int setup_arg_pages(struct linux_binprm *bprm, 748 int setup_arg_pages(struct linux_binprm *bprm,
717 unsigned long stack_top, 749 unsigned long stack_top,
718 int executable_stack) 750 int executable_stack)
719 { 751 {
720 unsigned long ret; 752 unsigned long ret;
721 unsigned long stack_shift; 753 unsigned long stack_shift;
722 struct mm_struct *mm = current->mm; 754 struct mm_struct *mm = current->mm;
723 struct vm_area_struct *vma = bprm->vma 755 struct vm_area_struct *vma = bprm->vma;
724 struct vm_area_struct *prev = NULL; 756 struct vm_area_struct *prev = NULL;
725 unsigned long vm_flags; 757 unsigned long vm_flags;
726 unsigned long stack_base; 758 unsigned long stack_base;
727 unsigned long stack_size; 759 unsigned long stack_size;
728 unsigned long stack_expand; 760 unsigned long stack_expand;
729 unsigned long rlim_stack; 761 unsigned long rlim_stack;
730 struct mmu_gather tlb; <<
731 struct vma_iterator vmi; <<
732 762
733 #ifdef CONFIG_STACK_GROWSUP 763 #ifdef CONFIG_STACK_GROWSUP
734 /* Limit stack size */ 764 /* Limit stack size */
735 stack_base = bprm->rlim_stack.rlim_max 765 stack_base = bprm->rlim_stack.rlim_max;
736 766
737 stack_base = calc_max_stack_size(stack 767 stack_base = calc_max_stack_size(stack_base);
738 768
739 /* Add space for stack randomization. 769 /* Add space for stack randomization. */
740 if (current->flags & PF_RANDOMIZE) !! 770 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
741 stack_base += (STACK_RND_MASK <<
742 771
743 /* Make sure we didn't let the argumen 772 /* Make sure we didn't let the argument array grow too large. */
744 if (vma->vm_end - vma->vm_start > stac 773 if (vma->vm_end - vma->vm_start > stack_base)
745 return -ENOMEM; 774 return -ENOMEM;
746 775
747 stack_base = PAGE_ALIGN(stack_top - st 776 stack_base = PAGE_ALIGN(stack_top - stack_base);
748 777
749 stack_shift = vma->vm_start - stack_ba 778 stack_shift = vma->vm_start - stack_base;
750 mm->arg_start = bprm->p - stack_shift; 779 mm->arg_start = bprm->p - stack_shift;
751 bprm->p = vma->vm_end - stack_shift; 780 bprm->p = vma->vm_end - stack_shift;
752 #else 781 #else
753 stack_top = arch_align_stack(stack_top 782 stack_top = arch_align_stack(stack_top);
754 stack_top = PAGE_ALIGN(stack_top); 783 stack_top = PAGE_ALIGN(stack_top);
755 784
756 if (unlikely(stack_top < mmap_min_addr 785 if (unlikely(stack_top < mmap_min_addr) ||
757 unlikely(vma->vm_end - vma->vm_sta 786 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
758 return -ENOMEM; 787 return -ENOMEM;
759 788
760 stack_shift = vma->vm_end - stack_top; 789 stack_shift = vma->vm_end - stack_top;
761 790
762 bprm->p -= stack_shift; 791 bprm->p -= stack_shift;
763 mm->arg_start = bprm->p; 792 mm->arg_start = bprm->p;
764 #endif 793 #endif
765 794
766 if (bprm->loader) 795 if (bprm->loader)
767 bprm->loader -= stack_shift; 796 bprm->loader -= stack_shift;
768 bprm->exec -= stack_shift; 797 bprm->exec -= stack_shift;
769 798
770 if (mmap_write_lock_killable(mm)) 799 if (mmap_write_lock_killable(mm))
771 return -EINTR; 800 return -EINTR;
772 801
773 vm_flags = VM_STACK_FLAGS; 802 vm_flags = VM_STACK_FLAGS;
774 803
775 /* 804 /*
776 * Adjust stack execute permissions; e 805 * Adjust stack execute permissions; explicitly enable for
777 * EXSTACK_ENABLE_X, disable for EXSTA 806 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
778 * (arch default) otherwise. 807 * (arch default) otherwise.
779 */ 808 */
780 if (unlikely(executable_stack == EXSTA 809 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
781 vm_flags |= VM_EXEC; 810 vm_flags |= VM_EXEC;
782 else if (executable_stack == EXSTACK_D 811 else if (executable_stack == EXSTACK_DISABLE_X)
783 vm_flags &= ~VM_EXEC; 812 vm_flags &= ~VM_EXEC;
784 vm_flags |= mm->def_flags; 813 vm_flags |= mm->def_flags;
785 vm_flags |= VM_STACK_INCOMPLETE_SETUP; 814 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
786 815
787 vma_iter_init(&vmi, mm, vma->vm_start) !! 816 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
788 <<
789 tlb_gather_mmu(&tlb, mm); <<
790 ret = mprotect_fixup(&vmi, &tlb, vma, <<
791 vm_flags); 817 vm_flags);
792 tlb_finish_mmu(&tlb); <<
793 <<
794 if (ret) 818 if (ret)
795 goto out_unlock; 819 goto out_unlock;
796 BUG_ON(prev != vma); 820 BUG_ON(prev != vma);
797 821
798 if (unlikely(vm_flags & VM_EXEC)) { 822 if (unlikely(vm_flags & VM_EXEC)) {
799 pr_warn_once("process '%pD4' s 823 pr_warn_once("process '%pD4' started with executable stack\n",
800 bprm->file); 824 bprm->file);
801 } 825 }
802 826
803 /* Move stack pages down in memory. */ 827 /* Move stack pages down in memory. */
804 if (stack_shift) { 828 if (stack_shift) {
805 /* !! 829 ret = shift_arg_pages(vma, stack_shift);
806 * During bprm_mm_init(), we c <<
807 * the binfmt code determines <<
808 * its final location. <<
809 */ <<
810 ret = relocate_vma_down(vma, s <<
811 if (ret) 830 if (ret)
812 goto out_unlock; 831 goto out_unlock;
813 } 832 }
814 833
815 /* mprotect_fixup is overkill to remov 834 /* mprotect_fixup is overkill to remove the temporary stack flags */
816 vm_flags_clear(vma, VM_STACK_INCOMPLET !! 835 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
817 836
818 stack_expand = 131072UL; /* randomly 3 837 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
819 stack_size = vma->vm_end - vma->vm_sta 838 stack_size = vma->vm_end - vma->vm_start;
820 /* 839 /*
821 * Align this down to a page boundary 840 * Align this down to a page boundary as expand_stack
822 * will align it up. 841 * will align it up.
823 */ 842 */
824 rlim_stack = bprm->rlim_stack.rlim_cur 843 rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
825 <<
826 stack_expand = min(rlim_stack, stack_s <<
827 <<
828 #ifdef CONFIG_STACK_GROWSUP 844 #ifdef CONFIG_STACK_GROWSUP
829 stack_base = vma->vm_start + stack_exp !! 845 if (stack_size + stack_expand > rlim_stack)
>> 846 stack_base = vma->vm_start + rlim_stack;
>> 847 else
>> 848 stack_base = vma->vm_end + stack_expand;
830 #else 849 #else
831 stack_base = vma->vm_end - stack_expan !! 850 if (stack_size + stack_expand > rlim_stack)
>> 851 stack_base = vma->vm_end - rlim_stack;
>> 852 else
>> 853 stack_base = vma->vm_start - stack_expand;
832 #endif 854 #endif
833 current->mm->start_stack = bprm->p; 855 current->mm->start_stack = bprm->p;
834 ret = expand_stack_locked(vma, stack_b !! 856 ret = expand_stack(vma, stack_base);
835 if (ret) 857 if (ret)
836 ret = -EFAULT; 858 ret = -EFAULT;
837 859
838 out_unlock: 860 out_unlock:
839 mmap_write_unlock(mm); 861 mmap_write_unlock(mm);
840 return ret; 862 return ret;
841 } 863 }
842 EXPORT_SYMBOL(setup_arg_pages); 864 EXPORT_SYMBOL(setup_arg_pages);
843 865
844 #else 866 #else
845 867
846 /* 868 /*
847 * Transfer the program arguments and environm 869 * Transfer the program arguments and environment from the holding pages
848 * onto the stack. The provided stack pointer 870 * onto the stack. The provided stack pointer is adjusted accordingly.
849 */ 871 */
850 int transfer_args_to_stack(struct linux_binprm 872 int transfer_args_to_stack(struct linux_binprm *bprm,
851 unsigned long *sp_l 873 unsigned long *sp_location)
852 { 874 {
853 unsigned long index, stop, sp; 875 unsigned long index, stop, sp;
854 int ret = 0; 876 int ret = 0;
855 877
856 stop = bprm->p >> PAGE_SHIFT; 878 stop = bprm->p >> PAGE_SHIFT;
857 sp = *sp_location; 879 sp = *sp_location;
858 880
859 for (index = MAX_ARG_PAGES - 1; index 881 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
860 unsigned int offset = index == 882 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
861 char *src = kmap_local_page(bp !! 883 char *src = kmap(bprm->page[index]) + offset;
862 sp -= PAGE_SIZE - offset; 884 sp -= PAGE_SIZE - offset;
863 if (copy_to_user((void *) sp, 885 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
864 ret = -EFAULT; 886 ret = -EFAULT;
865 kunmap_local(src); !! 887 kunmap(bprm->page[index]);
866 if (ret) 888 if (ret)
867 goto out; 889 goto out;
868 } 890 }
869 891
870 bprm->exec += *sp_location - MAX_ARG_P <<
871 *sp_location = sp; 892 *sp_location = sp;
872 893
873 out: 894 out:
874 return ret; 895 return ret;
875 } 896 }
876 EXPORT_SYMBOL(transfer_args_to_stack); 897 EXPORT_SYMBOL(transfer_args_to_stack);
877 898
878 #endif /* CONFIG_MMU */ 899 #endif /* CONFIG_MMU */
879 900
880 /* <<
881 * On success, caller must call do_close_execa <<
882 * struct file to close it. <<
883 */ <<
884 static struct file *do_open_execat(int fd, str 901 static struct file *do_open_execat(int fd, struct filename *name, int flags)
885 { 902 {
886 struct file *file; 903 struct file *file;
>> 904 int err;
887 struct open_flags open_exec_flags = { 905 struct open_flags open_exec_flags = {
888 .open_flag = O_LARGEFILE | O_R 906 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
889 .acc_mode = MAY_EXEC, 907 .acc_mode = MAY_EXEC,
890 .intent = LOOKUP_OPEN, 908 .intent = LOOKUP_OPEN,
891 .lookup_flags = LOOKUP_FOLLOW, 909 .lookup_flags = LOOKUP_FOLLOW,
892 }; 910 };
893 911
894 if ((flags & ~(AT_SYMLINK_NOFOLLOW | A 912 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
895 return ERR_PTR(-EINVAL); 913 return ERR_PTR(-EINVAL);
896 if (flags & AT_SYMLINK_NOFOLLOW) 914 if (flags & AT_SYMLINK_NOFOLLOW)
897 open_exec_flags.lookup_flags & 915 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
898 if (flags & AT_EMPTY_PATH) 916 if (flags & AT_EMPTY_PATH)
899 open_exec_flags.lookup_flags | 917 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
900 918
901 file = do_filp_open(fd, name, &open_ex 919 file = do_filp_open(fd, name, &open_exec_flags);
902 if (IS_ERR(file)) 920 if (IS_ERR(file))
903 return file; !! 921 goto out;
904 922
905 /* 923 /*
906 * In the past the regular type check !! 924 * may_open() has already checked for this, so it should be
907 * 633fb6ac3980 ("exec: move S_ISREG() !! 925 * impossible to trip now. But we need to be extra cautious
908 * an invariant that all non-regular f !! 926 * and check again at the very end too.
909 */ !! 927 */
910 if (WARN_ON_ONCE(!S_ISREG(file_inode(f !! 928 err = -EACCES;
911 path_noexec(&file->f_path)) { !! 929 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
912 fput(file); !! 930 path_noexec(&file->f_path)))
913 return ERR_PTR(-EACCES); !! 931 goto exit;
914 } <<
915 932
>> 933 err = deny_write_access(file);
>> 934 if (err)
>> 935 goto exit;
>> 936
>> 937 if (name->name[0] != '\0')
>> 938 fsnotify_open(file);
>> 939
>> 940 out:
916 return file; 941 return file;
>> 942
>> 943 exit:
>> 944 fput(file);
>> 945 return ERR_PTR(err);
917 } 946 }
918 947
919 /** <<
920 * open_exec - Open a path name for execution <<
921 * <<
922 * @name: path name to open with the intent of <<
923 * <<
924 * Returns ERR_PTR on failure or allocated str <<
925 * <<
926 * As this is a wrapper for the internal do_op <<
927 * do_close_execat(). <<
928 */ <<
929 struct file *open_exec(const char *name) 948 struct file *open_exec(const char *name)
930 { 949 {
931 struct filename *filename = getname_ke 950 struct filename *filename = getname_kernel(name);
932 struct file *f = ERR_CAST(filename); 951 struct file *f = ERR_CAST(filename);
933 952
934 if (!IS_ERR(filename)) { 953 if (!IS_ERR(filename)) {
935 f = do_open_execat(AT_FDCWD, f 954 f = do_open_execat(AT_FDCWD, filename, 0);
936 putname(filename); 955 putname(filename);
937 } 956 }
938 return f; 957 return f;
939 } 958 }
940 EXPORT_SYMBOL(open_exec); 959 EXPORT_SYMBOL(open_exec);
941 960
942 #if defined(CONFIG_BINFMT_FLAT) || defined(CON !! 961 #if defined(CONFIG_HAVE_AOUT) || defined(CONFIG_BINFMT_FLAT) || \
>> 962 defined(CONFIG_BINFMT_ELF_FDPIC)
943 ssize_t read_code(struct file *file, unsigned 963 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
944 { 964 {
945 ssize_t res = vfs_read(file, (void __u 965 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
946 if (res > 0) 966 if (res > 0)
947 flush_icache_user_range(addr, 967 flush_icache_user_range(addr, addr + len);
948 return res; 968 return res;
949 } 969 }
950 EXPORT_SYMBOL(read_code); 970 EXPORT_SYMBOL(read_code);
951 #endif 971 #endif
952 972
953 /* 973 /*
954 * Maps the mm_struct mm into the current task 974 * Maps the mm_struct mm into the current task struct.
955 * On success, this function returns with exec 975 * On success, this function returns with exec_update_lock
956 * held for writing. 976 * held for writing.
957 */ 977 */
958 static int exec_mmap(struct mm_struct *mm) 978 static int exec_mmap(struct mm_struct *mm)
959 { 979 {
960 struct task_struct *tsk; 980 struct task_struct *tsk;
961 struct mm_struct *old_mm, *active_mm; 981 struct mm_struct *old_mm, *active_mm;
962 int ret; 982 int ret;
963 983
964 /* Notify parent that we're no longer 984 /* Notify parent that we're no longer interested in the old VM */
965 tsk = current; 985 tsk = current;
966 old_mm = current->mm; 986 old_mm = current->mm;
967 exec_mm_release(tsk, old_mm); 987 exec_mm_release(tsk, old_mm);
>> 988 if (old_mm)
>> 989 sync_mm_rss(old_mm);
968 990
969 ret = down_write_killable(&tsk->signal 991 ret = down_write_killable(&tsk->signal->exec_update_lock);
970 if (ret) 992 if (ret)
971 return ret; 993 return ret;
972 994
973 if (old_mm) { 995 if (old_mm) {
974 /* 996 /*
975 * If there is a pending fatal 997 * If there is a pending fatal signal perhaps a signal
976 * whose default action is to 998 * whose default action is to create a coredump get
977 * out and die instead of goin 999 * out and die instead of going through with the exec.
978 */ 1000 */
979 ret = mmap_read_lock_killable( 1001 ret = mmap_read_lock_killable(old_mm);
980 if (ret) { 1002 if (ret) {
981 up_write(&tsk->signal- 1003 up_write(&tsk->signal->exec_update_lock);
982 return ret; 1004 return ret;
983 } 1005 }
984 } 1006 }
985 1007
986 task_lock(tsk); 1008 task_lock(tsk);
987 membarrier_exec_mmap(mm); 1009 membarrier_exec_mmap(mm);
988 1010
989 local_irq_disable(); 1011 local_irq_disable();
990 active_mm = tsk->active_mm; 1012 active_mm = tsk->active_mm;
991 tsk->active_mm = mm; 1013 tsk->active_mm = mm;
992 tsk->mm = mm; 1014 tsk->mm = mm;
993 mm_init_cid(mm); <<
994 /* 1015 /*
995 * This prevents preemption while acti 1016 * This prevents preemption while active_mm is being loaded and
996 * it and mm are being updated, which 1017 * it and mm are being updated, which could cause problems for
997 * lazy tlb mm refcounting when these 1018 * lazy tlb mm refcounting when these are updated by context
998 * switches. Not all architectures can 1019 * switches. Not all architectures can handle irqs off over
999 * activate_mm yet. 1020 * activate_mm yet.
1000 */ 1021 */
1001 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS 1022 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1002 local_irq_enable(); 1023 local_irq_enable();
1003 activate_mm(active_mm, mm); 1024 activate_mm(active_mm, mm);
1004 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_ 1025 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1005 local_irq_enable(); 1026 local_irq_enable();
1006 lru_gen_add_mm(mm); !! 1027 tsk->mm->vmacache_seqnum = 0;
>> 1028 vmacache_flush(tsk);
1007 task_unlock(tsk); 1029 task_unlock(tsk);
1008 lru_gen_use_mm(mm); <<
1009 if (old_mm) { 1030 if (old_mm) {
1010 mmap_read_unlock(old_mm); 1031 mmap_read_unlock(old_mm);
1011 BUG_ON(active_mm != old_mm); 1032 BUG_ON(active_mm != old_mm);
1012 setmax_mm_hiwater_rss(&tsk->s 1033 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1013 mm_update_next_owner(old_mm); 1034 mm_update_next_owner(old_mm);
1014 mmput(old_mm); 1035 mmput(old_mm);
1015 return 0; 1036 return 0;
1016 } 1037 }
1017 mmdrop_lazy_tlb(active_mm); !! 1038 mmdrop(active_mm);
1018 return 0; 1039 return 0;
1019 } 1040 }
1020 1041
1021 static int de_thread(struct task_struct *tsk) 1042 static int de_thread(struct task_struct *tsk)
1022 { 1043 {
1023 struct signal_struct *sig = tsk->sign 1044 struct signal_struct *sig = tsk->signal;
1024 struct sighand_struct *oldsighand = t 1045 struct sighand_struct *oldsighand = tsk->sighand;
1025 spinlock_t *lock = &oldsighand->siglo 1046 spinlock_t *lock = &oldsighand->siglock;
1026 1047
1027 if (thread_group_empty(tsk)) 1048 if (thread_group_empty(tsk))
1028 goto no_thread_group; 1049 goto no_thread_group;
1029 1050
1030 /* 1051 /*
1031 * Kill all other threads in the thre 1052 * Kill all other threads in the thread group.
1032 */ 1053 */
1033 spin_lock_irq(lock); 1054 spin_lock_irq(lock);
1034 if ((sig->flags & SIGNAL_GROUP_EXIT) 1055 if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) {
1035 /* 1056 /*
1036 * Another group action in pr 1057 * Another group action in progress, just
1037 * return so that the signal 1058 * return so that the signal is processed.
1038 */ 1059 */
1039 spin_unlock_irq(lock); 1060 spin_unlock_irq(lock);
1040 return -EAGAIN; 1061 return -EAGAIN;
1041 } 1062 }
1042 1063
1043 sig->group_exec_task = tsk; 1064 sig->group_exec_task = tsk;
1044 sig->notify_count = zap_other_threads 1065 sig->notify_count = zap_other_threads(tsk);
1045 if (!thread_group_leader(tsk)) 1066 if (!thread_group_leader(tsk))
1046 sig->notify_count--; 1067 sig->notify_count--;
1047 1068
1048 while (sig->notify_count) { 1069 while (sig->notify_count) {
1049 __set_current_state(TASK_KILL 1070 __set_current_state(TASK_KILLABLE);
1050 spin_unlock_irq(lock); 1071 spin_unlock_irq(lock);
1051 schedule(); 1072 schedule();
1052 if (__fatal_signal_pending(ts 1073 if (__fatal_signal_pending(tsk))
1053 goto killed; 1074 goto killed;
1054 spin_lock_irq(lock); 1075 spin_lock_irq(lock);
1055 } 1076 }
1056 spin_unlock_irq(lock); 1077 spin_unlock_irq(lock);
1057 1078
1058 /* 1079 /*
1059 * At this point all other threads ha 1080 * At this point all other threads have exited, all we have to
1060 * do is to wait for the thread group 1081 * do is to wait for the thread group leader to become inactive,
1061 * and to assume its PID: 1082 * and to assume its PID:
1062 */ 1083 */
1063 if (!thread_group_leader(tsk)) { 1084 if (!thread_group_leader(tsk)) {
1064 struct task_struct *leader = 1085 struct task_struct *leader = tsk->group_leader;
1065 1086
1066 for (;;) { 1087 for (;;) {
1067 cgroup_threadgroup_ch 1088 cgroup_threadgroup_change_begin(tsk);
1068 write_lock_irq(&taskl 1089 write_lock_irq(&tasklist_lock);
1069 /* 1090 /*
1070 * Do this under task 1091 * Do this under tasklist_lock to ensure that
1071 * exit_notify() can' 1092 * exit_notify() can't miss ->group_exec_task
1072 */ 1093 */
1073 sig->notify_count = - 1094 sig->notify_count = -1;
1074 if (likely(leader->ex 1095 if (likely(leader->exit_state))
1075 break; 1096 break;
1076 __set_current_state(T 1097 __set_current_state(TASK_KILLABLE);
1077 write_unlock_irq(&tas 1098 write_unlock_irq(&tasklist_lock);
1078 cgroup_threadgroup_ch 1099 cgroup_threadgroup_change_end(tsk);
1079 schedule(); 1100 schedule();
1080 if (__fatal_signal_pe 1101 if (__fatal_signal_pending(tsk))
1081 goto killed; 1102 goto killed;
1082 } 1103 }
1083 1104
1084 /* 1105 /*
1085 * The only record we have of 1106 * The only record we have of the real-time age of a
1086 * process, regardless of exe 1107 * process, regardless of execs it's done, is start_time.
1087 * All the past CPU time is a 1108 * All the past CPU time is accumulated in signal_struct
1088 * from sister threads now de 1109 * from sister threads now dead. But in this non-leader
1089 * exec, nothing survives fro 1110 * exec, nothing survives from the original leader thread,
1090 * whose birth marks the true 1111 * whose birth marks the true age of this process now.
1091 * When we take on its identi 1112 * When we take on its identity by switching to its PID, we
1092 * also take its birthdate (a 1113 * also take its birthdate (always earlier than our own).
1093 */ 1114 */
1094 tsk->start_time = leader->sta 1115 tsk->start_time = leader->start_time;
1095 tsk->start_boottime = leader- 1116 tsk->start_boottime = leader->start_boottime;
1096 1117
1097 BUG_ON(!same_thread_group(lea 1118 BUG_ON(!same_thread_group(leader, tsk));
1098 /* 1119 /*
1099 * An exec() starts a new thr 1120 * An exec() starts a new thread group with the
1100 * TGID of the previous threa 1121 * TGID of the previous thread group. Rehash the
1101 * two threads with a switche 1122 * two threads with a switched PID, and release
1102 * the former thread group le 1123 * the former thread group leader:
1103 */ 1124 */
1104 1125
1105 /* Become a process group lea 1126 /* Become a process group leader with the old leader's pid.
1106 * The old leader becomes a t 1127 * The old leader becomes a thread of the this thread group.
1107 */ 1128 */
1108 exchange_tids(tsk, leader); 1129 exchange_tids(tsk, leader);
1109 transfer_pid(leader, tsk, PID 1130 transfer_pid(leader, tsk, PIDTYPE_TGID);
1110 transfer_pid(leader, tsk, PID 1131 transfer_pid(leader, tsk, PIDTYPE_PGID);
1111 transfer_pid(leader, tsk, PID 1132 transfer_pid(leader, tsk, PIDTYPE_SID);
1112 1133
1113 list_replace_rcu(&leader->tas 1134 list_replace_rcu(&leader->tasks, &tsk->tasks);
1114 list_replace_init(&leader->si 1135 list_replace_init(&leader->sibling, &tsk->sibling);
1115 1136
1116 tsk->group_leader = tsk; 1137 tsk->group_leader = tsk;
1117 leader->group_leader = tsk; 1138 leader->group_leader = tsk;
1118 1139
1119 tsk->exit_signal = SIGCHLD; 1140 tsk->exit_signal = SIGCHLD;
1120 leader->exit_signal = -1; 1141 leader->exit_signal = -1;
1121 1142
1122 BUG_ON(leader->exit_state != 1143 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1123 leader->exit_state = EXIT_DEA 1144 leader->exit_state = EXIT_DEAD;
>> 1145
1124 /* 1146 /*
1125 * We are going to release_ta 1147 * We are going to release_task()->ptrace_unlink() silently,
1126 * the tracer can sleep in do 1148 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1127 * the tracer won't block aga !! 1149 * the tracer wont't block again waiting for this thread.
1128 */ 1150 */
1129 if (unlikely(leader->ptrace)) 1151 if (unlikely(leader->ptrace))
1130 __wake_up_parent(lead 1152 __wake_up_parent(leader, leader->parent);
1131 write_unlock_irq(&tasklist_lo 1153 write_unlock_irq(&tasklist_lock);
1132 cgroup_threadgroup_change_end 1154 cgroup_threadgroup_change_end(tsk);
1133 1155
1134 release_task(leader); 1156 release_task(leader);
1135 } 1157 }
1136 1158
1137 sig->group_exec_task = NULL; 1159 sig->group_exec_task = NULL;
1138 sig->notify_count = 0; 1160 sig->notify_count = 0;
1139 1161
1140 no_thread_group: 1162 no_thread_group:
1141 /* we have changed execution domain * 1163 /* we have changed execution domain */
1142 tsk->exit_signal = SIGCHLD; 1164 tsk->exit_signal = SIGCHLD;
1143 1165
1144 BUG_ON(!thread_group_leader(tsk)); 1166 BUG_ON(!thread_group_leader(tsk));
1145 return 0; 1167 return 0;
1146 1168
1147 killed: 1169 killed:
1148 /* protects against exit_notify() and 1170 /* protects against exit_notify() and __exit_signal() */
1149 read_lock(&tasklist_lock); 1171 read_lock(&tasklist_lock);
1150 sig->group_exec_task = NULL; 1172 sig->group_exec_task = NULL;
1151 sig->notify_count = 0; 1173 sig->notify_count = 0;
1152 read_unlock(&tasklist_lock); 1174 read_unlock(&tasklist_lock);
1153 return -EAGAIN; 1175 return -EAGAIN;
1154 } 1176 }
1155 1177
1156 1178
1157 /* 1179 /*
1158 * This function makes sure the current proce 1180 * This function makes sure the current process has its own signal table,
1159 * so that flush_signal_handlers can later re 1181 * so that flush_signal_handlers can later reset the handlers without
1160 * disturbing other processes. (Other proces 1182 * disturbing other processes. (Other processes might share the signal
1161 * table via the CLONE_SIGHAND option to clon 1183 * table via the CLONE_SIGHAND option to clone().)
1162 */ 1184 */
1163 static int unshare_sighand(struct task_struct 1185 static int unshare_sighand(struct task_struct *me)
1164 { 1186 {
1165 struct sighand_struct *oldsighand = m 1187 struct sighand_struct *oldsighand = me->sighand;
1166 1188
1167 if (refcount_read(&oldsighand->count) 1189 if (refcount_read(&oldsighand->count) != 1) {
1168 struct sighand_struct *newsig 1190 struct sighand_struct *newsighand;
1169 /* 1191 /*
1170 * This ->sighand is shared w 1192 * This ->sighand is shared with the CLONE_SIGHAND
1171 * but not CLONE_THREAD task, 1193 * but not CLONE_THREAD task, switch to the new one.
1172 */ 1194 */
1173 newsighand = kmem_cache_alloc 1195 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1174 if (!newsighand) 1196 if (!newsighand)
1175 return -ENOMEM; 1197 return -ENOMEM;
1176 1198
1177 refcount_set(&newsighand->cou 1199 refcount_set(&newsighand->count, 1);
>> 1200 memcpy(newsighand->action, oldsighand->action,
>> 1201 sizeof(newsighand->action));
1178 1202
1179 write_lock_irq(&tasklist_lock 1203 write_lock_irq(&tasklist_lock);
1180 spin_lock(&oldsighand->sigloc 1204 spin_lock(&oldsighand->siglock);
1181 memcpy(newsighand->action, ol <<
1182 sizeof(newsighand->act <<
1183 rcu_assign_pointer(me->sighan 1205 rcu_assign_pointer(me->sighand, newsighand);
1184 spin_unlock(&oldsighand->sigl 1206 spin_unlock(&oldsighand->siglock);
1185 write_unlock_irq(&tasklist_lo 1207 write_unlock_irq(&tasklist_lock);
1186 1208
1187 __cleanup_sighand(oldsighand) 1209 __cleanup_sighand(oldsighand);
1188 } 1210 }
1189 return 0; 1211 return 0;
1190 } 1212 }
1191 1213
1192 char *__get_task_comm(char *buf, size_t buf_s 1214 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1193 { 1215 {
1194 task_lock(tsk); 1216 task_lock(tsk);
1195 /* Always NUL terminated and zero-pad 1217 /* Always NUL terminated and zero-padded */
1196 strscpy_pad(buf, tsk->comm, buf_size) 1218 strscpy_pad(buf, tsk->comm, buf_size);
1197 task_unlock(tsk); 1219 task_unlock(tsk);
1198 return buf; 1220 return buf;
1199 } 1221 }
1200 EXPORT_SYMBOL_GPL(__get_task_comm); 1222 EXPORT_SYMBOL_GPL(__get_task_comm);
1201 1223
1202 /* 1224 /*
1203 * These functions flushes out all traces of 1225 * These functions flushes out all traces of the currently running executable
1204 * so that a new one can be started 1226 * so that a new one can be started
1205 */ 1227 */
1206 1228
1207 void __set_task_comm(struct task_struct *tsk, 1229 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1208 { 1230 {
1209 task_lock(tsk); 1231 task_lock(tsk);
1210 trace_task_rename(tsk, buf); 1232 trace_task_rename(tsk, buf);
1211 strscpy_pad(tsk->comm, buf, sizeof(ts 1233 strscpy_pad(tsk->comm, buf, sizeof(tsk->comm));
1212 task_unlock(tsk); 1234 task_unlock(tsk);
1213 perf_event_comm(tsk, exec); 1235 perf_event_comm(tsk, exec);
1214 } 1236 }
1215 1237
1216 /* 1238 /*
1217 * Calling this is the point of no return. No 1239 * Calling this is the point of no return. None of the failures will be
1218 * seen by userspace since either the process 1240 * seen by userspace since either the process is already taking a fatal
1219 * signal (via de_thread() or coredump), or w 1241 * signal (via de_thread() or coredump), or will have SEGV raised
1220 * (after exec_mmap()) by search_binary_handl 1242 * (after exec_mmap()) by search_binary_handler (see below).
1221 */ 1243 */
1222 int begin_new_exec(struct linux_binprm * bprm 1244 int begin_new_exec(struct linux_binprm * bprm)
1223 { 1245 {
1224 struct task_struct *me = current; 1246 struct task_struct *me = current;
1225 int retval; 1247 int retval;
1226 1248
1227 /* Once we are committed compute the 1249 /* Once we are committed compute the creds */
1228 retval = bprm_creds_from_file(bprm); 1250 retval = bprm_creds_from_file(bprm);
1229 if (retval) 1251 if (retval)
1230 return retval; 1252 return retval;
1231 1253
1232 /* 1254 /*
1233 * This tracepoint marks the point be <<
1234 * the current task is still unchange <<
1235 * no return). The later "sched_proce <<
1236 * the current task has successfully <<
1237 */ <<
1238 trace_sched_prepare_exec(current, bpr <<
1239 <<
1240 /* <<
1241 * Ensure all future errors are fatal 1255 * Ensure all future errors are fatal.
1242 */ 1256 */
1243 bprm->point_of_no_return = true; 1257 bprm->point_of_no_return = true;
1244 1258
1245 /* 1259 /*
1246 * Make this the only thread in the t 1260 * Make this the only thread in the thread group.
1247 */ 1261 */
1248 retval = de_thread(me); 1262 retval = de_thread(me);
1249 if (retval) 1263 if (retval)
1250 goto out; 1264 goto out;
1251 1265
1252 /* 1266 /*
1253 * Cancel any io_uring activity acros 1267 * Cancel any io_uring activity across execve
1254 */ 1268 */
1255 io_uring_task_cancel(); 1269 io_uring_task_cancel();
1256 1270
1257 /* Ensure the files table is not shar 1271 /* Ensure the files table is not shared. */
1258 retval = unshare_files(); 1272 retval = unshare_files();
1259 if (retval) 1273 if (retval)
1260 goto out; 1274 goto out;
1261 1275
1262 /* 1276 /*
1263 * Must be called _before_ exec_mmap( 1277 * Must be called _before_ exec_mmap() as bprm->mm is
1264 * not visible until then. Doing it h !! 1278 * not visibile until then. This also enables the update
1265 * we don't race against replace_mm_e !! 1279 * to be lockless.
1266 */ 1280 */
1267 retval = set_mm_exe_file(bprm->mm, bp 1281 retval = set_mm_exe_file(bprm->mm, bprm->file);
1268 if (retval) 1282 if (retval)
1269 goto out; 1283 goto out;
1270 1284
1271 /* If the binary is not readable then 1285 /* If the binary is not readable then enforce mm->dumpable=0 */
1272 would_dump(bprm, bprm->file); 1286 would_dump(bprm, bprm->file);
1273 if (bprm->have_execfd) 1287 if (bprm->have_execfd)
1274 would_dump(bprm, bprm->execut 1288 would_dump(bprm, bprm->executable);
1275 1289
1276 /* 1290 /*
1277 * Release all of the old mmap stuff 1291 * Release all of the old mmap stuff
1278 */ 1292 */
1279 acct_arg_size(bprm, 0); 1293 acct_arg_size(bprm, 0);
1280 retval = exec_mmap(bprm->mm); 1294 retval = exec_mmap(bprm->mm);
1281 if (retval) 1295 if (retval)
1282 goto out; 1296 goto out;
1283 1297
1284 bprm->mm = NULL; 1298 bprm->mm = NULL;
1285 1299
1286 retval = exec_task_namespaces(); <<
1287 if (retval) <<
1288 goto out_unlock; <<
1289 <<
1290 #ifdef CONFIG_POSIX_TIMERS 1300 #ifdef CONFIG_POSIX_TIMERS
1291 spin_lock_irq(&me->sighand->siglock); !! 1301 exit_itimers(me->signal);
1292 posix_cpu_timers_exit(me); <<
1293 spin_unlock_irq(&me->sighand->siglock <<
1294 exit_itimers(me); <<
1295 flush_itimer_signals(); 1302 flush_itimer_signals();
1296 #endif 1303 #endif
1297 1304
1298 /* 1305 /*
1299 * Make the signal table private. 1306 * Make the signal table private.
1300 */ 1307 */
1301 retval = unshare_sighand(me); 1308 retval = unshare_sighand(me);
1302 if (retval) 1309 if (retval)
1303 goto out_unlock; 1310 goto out_unlock;
1304 1311
1305 me->flags &= ~(PF_RANDOMIZE | PF_FORK !! 1312 /*
>> 1313 * Ensure that the uaccess routines can actually operate on userspace
>> 1314 * pointers:
>> 1315 */
>> 1316 force_uaccess_begin();
>> 1317
>> 1318 me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1306 PF_NO 1319 PF_NOFREEZE | PF_NO_SETAFFINITY);
1307 flush_thread(); 1320 flush_thread();
1308 me->personality &= ~bprm->per_clear; 1321 me->personality &= ~bprm->per_clear;
1309 1322
1310 clear_syscall_work_syscall_user_dispa 1323 clear_syscall_work_syscall_user_dispatch(me);
1311 1324
1312 /* 1325 /*
1313 * We have to apply CLOEXEC before we 1326 * We have to apply CLOEXEC before we change whether the process is
1314 * dumpable (in setup_new_exec) to av 1327 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1315 * trying to access the should-be-clo 1328 * trying to access the should-be-closed file descriptors of a process
1316 * undergoing exec(2). 1329 * undergoing exec(2).
1317 */ 1330 */
1318 do_close_on_exec(me->files); 1331 do_close_on_exec(me->files);
1319 1332
1320 if (bprm->secureexec) { 1333 if (bprm->secureexec) {
1321 /* Make sure parent cannot si 1334 /* Make sure parent cannot signal privileged process. */
1322 me->pdeath_signal = 0; 1335 me->pdeath_signal = 0;
1323 1336
1324 /* 1337 /*
1325 * For secureexec, reset the 1338 * For secureexec, reset the stack limit to sane default to
1326 * avoid bad behavior from th 1339 * avoid bad behavior from the prior rlimits. This has to
1327 * happen before arch_pick_mm 1340 * happen before arch_pick_mmap_layout(), which examines
1328 * RLIMIT_STACK, but after th 1341 * RLIMIT_STACK, but after the point of no return to avoid
1329 * needing to clean up the ch 1342 * needing to clean up the change on failure.
1330 */ 1343 */
1331 if (bprm->rlim_stack.rlim_cur 1344 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1332 bprm->rlim_stack.rlim 1345 bprm->rlim_stack.rlim_cur = _STK_LIM;
1333 } 1346 }
1334 1347
1335 me->sas_ss_sp = me->sas_ss_size = 0; 1348 me->sas_ss_sp = me->sas_ss_size = 0;
1336 1349
1337 /* 1350 /*
1338 * Figure out dumpability. Note that 1351 * Figure out dumpability. Note that this checking only of current
1339 * is wrong, but userspace depends on 1352 * is wrong, but userspace depends on it. This should be testing
1340 * bprm->secureexec instead. 1353 * bprm->secureexec instead.
1341 */ 1354 */
1342 if (bprm->interp_flags & BINPRM_FLAGS 1355 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1343 !(uid_eq(current_euid(), current_ 1356 !(uid_eq(current_euid(), current_uid()) &&
1344 gid_eq(current_egid(), current_ 1357 gid_eq(current_egid(), current_gid())))
1345 set_dumpable(current->mm, sui 1358 set_dumpable(current->mm, suid_dumpable);
1346 else 1359 else
1347 set_dumpable(current->mm, SUI 1360 set_dumpable(current->mm, SUID_DUMP_USER);
1348 1361
1349 perf_event_exec(); 1362 perf_event_exec();
1350 __set_task_comm(me, kbasename(bprm->f 1363 __set_task_comm(me, kbasename(bprm->filename), true);
1351 1364
1352 /* An exec changes our domain. We are 1365 /* An exec changes our domain. We are no longer part of the thread
1353 group */ 1366 group */
1354 WRITE_ONCE(me->self_exec_id, me->self 1367 WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1355 flush_signal_handlers(me, 0); 1368 flush_signal_handlers(me, 0);
1356 1369
1357 retval = set_cred_ucounts(bprm->cred) 1370 retval = set_cred_ucounts(bprm->cred);
1358 if (retval < 0) 1371 if (retval < 0)
1359 goto out_unlock; 1372 goto out_unlock;
1360 1373
1361 /* 1374 /*
1362 * install the new credentials for th 1375 * install the new credentials for this executable
1363 */ 1376 */
1364 security_bprm_committing_creds(bprm); 1377 security_bprm_committing_creds(bprm);
1365 1378
1366 commit_creds(bprm->cred); 1379 commit_creds(bprm->cred);
1367 bprm->cred = NULL; 1380 bprm->cred = NULL;
1368 1381
1369 /* 1382 /*
1370 * Disable monitoring for regular use 1383 * Disable monitoring for regular users
1371 * when executing setuid binaries. Mu 1384 * when executing setuid binaries. Must
1372 * wait until new credentials are com 1385 * wait until new credentials are committed
1373 * by commit_creds() above 1386 * by commit_creds() above
1374 */ 1387 */
1375 if (get_dumpable(me->mm) != SUID_DUMP 1388 if (get_dumpable(me->mm) != SUID_DUMP_USER)
1376 perf_event_exit_task(me); 1389 perf_event_exit_task(me);
1377 /* 1390 /*
1378 * cred_guard_mutex must be held at l 1391 * cred_guard_mutex must be held at least to this point to prevent
1379 * ptrace_attach() from altering our 1392 * ptrace_attach() from altering our determination of the task's
1380 * credentials; any time after this i 1393 * credentials; any time after this it may be unlocked.
1381 */ 1394 */
1382 security_bprm_committed_creds(bprm); 1395 security_bprm_committed_creds(bprm);
1383 1396
1384 /* Pass the opened binary to the inte 1397 /* Pass the opened binary to the interpreter. */
1385 if (bprm->have_execfd) { 1398 if (bprm->have_execfd) {
1386 retval = get_unused_fd_flags( 1399 retval = get_unused_fd_flags(0);
1387 if (retval < 0) 1400 if (retval < 0)
1388 goto out_unlock; 1401 goto out_unlock;
1389 fd_install(retval, bprm->exec 1402 fd_install(retval, bprm->executable);
1390 bprm->executable = NULL; 1403 bprm->executable = NULL;
1391 bprm->execfd = retval; 1404 bprm->execfd = retval;
1392 } 1405 }
1393 return 0; 1406 return 0;
1394 1407
1395 out_unlock: 1408 out_unlock:
1396 up_write(&me->signal->exec_update_loc 1409 up_write(&me->signal->exec_update_lock);
1397 if (!bprm->cred) <<
1398 mutex_unlock(&me->signal->cre <<
1399 <<
1400 out: 1410 out:
1401 return retval; 1411 return retval;
1402 } 1412 }
1403 EXPORT_SYMBOL(begin_new_exec); 1413 EXPORT_SYMBOL(begin_new_exec);
1404 1414
1405 void would_dump(struct linux_binprm *bprm, st 1415 void would_dump(struct linux_binprm *bprm, struct file *file)
1406 { 1416 {
1407 struct inode *inode = file_inode(file 1417 struct inode *inode = file_inode(file);
1408 struct mnt_idmap *idmap = file_mnt_id !! 1418 struct user_namespace *mnt_userns = file_mnt_user_ns(file);
1409 if (inode_permission(idmap, inode, MA !! 1419 if (inode_permission(mnt_userns, inode, MAY_READ) < 0) {
1410 struct user_namespace *old, * 1420 struct user_namespace *old, *user_ns;
1411 bprm->interp_flags |= BINPRM_ 1421 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1412 1422
1413 /* Ensure mm->user_ns contain 1423 /* Ensure mm->user_ns contains the executable */
1414 user_ns = old = bprm->mm->use 1424 user_ns = old = bprm->mm->user_ns;
1415 while ((user_ns != &init_user 1425 while ((user_ns != &init_user_ns) &&
1416 !privileged_wrt_inode_ !! 1426 !privileged_wrt_inode_uidgid(user_ns, mnt_userns, inode))
1417 user_ns = user_ns->pa 1427 user_ns = user_ns->parent;
1418 1428
1419 if (old != user_ns) { 1429 if (old != user_ns) {
1420 bprm->mm->user_ns = g 1430 bprm->mm->user_ns = get_user_ns(user_ns);
1421 put_user_ns(old); 1431 put_user_ns(old);
1422 } 1432 }
1423 } 1433 }
1424 } 1434 }
1425 EXPORT_SYMBOL(would_dump); 1435 EXPORT_SYMBOL(would_dump);
1426 1436
1427 void setup_new_exec(struct linux_binprm * bpr 1437 void setup_new_exec(struct linux_binprm * bprm)
1428 { 1438 {
1429 /* Setup things that can depend upon 1439 /* Setup things that can depend upon the personality */
1430 struct task_struct *me = current; 1440 struct task_struct *me = current;
1431 1441
1432 arch_pick_mmap_layout(me->mm, &bprm-> 1442 arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1433 1443
1434 arch_setup_new_exec(); 1444 arch_setup_new_exec();
1435 1445
1436 /* Set the new mm task size. We have 1446 /* Set the new mm task size. We have to do that late because it may
1437 * depend on TIF_32BIT which is only 1447 * depend on TIF_32BIT which is only updated in flush_thread() on
1438 * some architectures like powerpc 1448 * some architectures like powerpc
1439 */ 1449 */
1440 me->mm->task_size = TASK_SIZE; 1450 me->mm->task_size = TASK_SIZE;
1441 up_write(&me->signal->exec_update_loc 1451 up_write(&me->signal->exec_update_lock);
1442 mutex_unlock(&me->signal->cred_guard_ 1452 mutex_unlock(&me->signal->cred_guard_mutex);
1443 } 1453 }
1444 EXPORT_SYMBOL(setup_new_exec); 1454 EXPORT_SYMBOL(setup_new_exec);
1445 1455
1446 /* Runs immediately before start_thread() tak 1456 /* Runs immediately before start_thread() takes over. */
1447 void finalize_exec(struct linux_binprm *bprm) 1457 void finalize_exec(struct linux_binprm *bprm)
1448 { 1458 {
1449 /* Store any stack rlimit changes bef 1459 /* Store any stack rlimit changes before starting thread. */
1450 task_lock(current->group_leader); 1460 task_lock(current->group_leader);
1451 current->signal->rlim[RLIMIT_STACK] = 1461 current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1452 task_unlock(current->group_leader); 1462 task_unlock(current->group_leader);
1453 } 1463 }
1454 EXPORT_SYMBOL(finalize_exec); 1464 EXPORT_SYMBOL(finalize_exec);
1455 1465
1456 /* 1466 /*
1457 * Prepare credentials and lock ->cred_guard_ 1467 * Prepare credentials and lock ->cred_guard_mutex.
1458 * setup_new_exec() commits the new creds and 1468 * setup_new_exec() commits the new creds and drops the lock.
1459 * Or, if exec fails before, free_bprm() shou 1469 * Or, if exec fails before, free_bprm() should release ->cred
1460 * and unlock. 1470 * and unlock.
1461 */ 1471 */
1462 static int prepare_bprm_creds(struct linux_bi 1472 static int prepare_bprm_creds(struct linux_binprm *bprm)
1463 { 1473 {
1464 if (mutex_lock_interruptible(¤t 1474 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1465 return -ERESTARTNOINTR; 1475 return -ERESTARTNOINTR;
1466 1476
1467 bprm->cred = prepare_exec_creds(); 1477 bprm->cred = prepare_exec_creds();
1468 if (likely(bprm->cred)) 1478 if (likely(bprm->cred))
1469 return 0; 1479 return 0;
1470 1480
1471 mutex_unlock(¤t->signal->cred_g 1481 mutex_unlock(¤t->signal->cred_guard_mutex);
1472 return -ENOMEM; 1482 return -ENOMEM;
1473 } 1483 }
1474 1484
1475 /* Matches do_open_execat() */ <<
1476 static void do_close_execat(struct file *file <<
1477 { <<
1478 if (file) <<
1479 fput(file); <<
1480 } <<
1481 <<
1482 static void free_bprm(struct linux_binprm *bp 1485 static void free_bprm(struct linux_binprm *bprm)
1483 { 1486 {
1484 if (bprm->mm) { 1487 if (bprm->mm) {
1485 acct_arg_size(bprm, 0); 1488 acct_arg_size(bprm, 0);
1486 mmput(bprm->mm); 1489 mmput(bprm->mm);
1487 } 1490 }
1488 free_arg_pages(bprm); 1491 free_arg_pages(bprm);
1489 if (bprm->cred) { 1492 if (bprm->cred) {
1490 mutex_unlock(¤t->signal 1493 mutex_unlock(¤t->signal->cred_guard_mutex);
1491 abort_creds(bprm->cred); 1494 abort_creds(bprm->cred);
1492 } 1495 }
1493 do_close_execat(bprm->file); !! 1496 if (bprm->file) {
>> 1497 allow_write_access(bprm->file);
>> 1498 fput(bprm->file);
>> 1499 }
1494 if (bprm->executable) 1500 if (bprm->executable)
1495 fput(bprm->executable); 1501 fput(bprm->executable);
1496 /* If a binfmt changed the interp, fr 1502 /* If a binfmt changed the interp, free it. */
1497 if (bprm->interp != bprm->filename) 1503 if (bprm->interp != bprm->filename)
1498 kfree(bprm->interp); 1504 kfree(bprm->interp);
1499 kfree(bprm->fdpath); 1505 kfree(bprm->fdpath);
1500 kfree(bprm); 1506 kfree(bprm);
1501 } 1507 }
1502 1508
1503 static struct linux_binprm *alloc_bprm(int fd !! 1509 static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1504 { 1510 {
1505 struct linux_binprm *bprm; !! 1511 struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1506 struct file *file; <<
1507 int retval = -ENOMEM; 1512 int retval = -ENOMEM;
1508 !! 1513 if (!bprm)
1509 file = do_open_execat(fd, filename, f !! 1514 goto out;
1510 if (IS_ERR(file)) <<
1511 return ERR_CAST(file); <<
1512 <<
1513 bprm = kzalloc(sizeof(*bprm), GFP_KER <<
1514 if (!bprm) { <<
1515 do_close_execat(file); <<
1516 return ERR_PTR(-ENOMEM); <<
1517 } <<
1518 <<
1519 bprm->file = file; <<
1520 1515
1521 if (fd == AT_FDCWD || filename->name[ 1516 if (fd == AT_FDCWD || filename->name[0] == '/') {
1522 bprm->filename = filename->na 1517 bprm->filename = filename->name;
1523 } else { 1518 } else {
1524 if (filename->name[0] == '\0' 1519 if (filename->name[0] == '\0')
1525 bprm->fdpath = kaspri 1520 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1526 else 1521 else
1527 bprm->fdpath = kaspri 1522 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1528 1523 fd, filename->name);
1529 if (!bprm->fdpath) 1524 if (!bprm->fdpath)
1530 goto out_free; 1525 goto out_free;
1531 1526
1532 /* <<
1533 * Record that a name derived <<
1534 * inaccessible after exec. <<
1535 * choose to fail when the ex <<
1536 * interpreter and an open fi <<
1537 * the interpreter. This mak <<
1538 * than having the interprete <<
1539 * when it finds the executab <<
1540 */ <<
1541 if (get_close_on_exec(fd)) <<
1542 bprm->interp_flags |= <<
1543 <<
1544 bprm->filename = bprm->fdpath 1527 bprm->filename = bprm->fdpath;
1545 } 1528 }
1546 bprm->interp = bprm->filename; 1529 bprm->interp = bprm->filename;
1547 1530
1548 retval = bprm_mm_init(bprm); 1531 retval = bprm_mm_init(bprm);
1549 if (!retval) !! 1532 if (retval)
1550 return bprm; !! 1533 goto out_free;
>> 1534 return bprm;
1551 1535
1552 out_free: 1536 out_free:
1553 free_bprm(bprm); 1537 free_bprm(bprm);
>> 1538 out:
1554 return ERR_PTR(retval); 1539 return ERR_PTR(retval);
1555 } 1540 }
1556 1541
1557 int bprm_change_interp(const char *interp, st 1542 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1558 { 1543 {
1559 /* If a binfmt changed the interp, fr 1544 /* If a binfmt changed the interp, free it first. */
1560 if (bprm->interp != bprm->filename) 1545 if (bprm->interp != bprm->filename)
1561 kfree(bprm->interp); 1546 kfree(bprm->interp);
1562 bprm->interp = kstrdup(interp, GFP_KE 1547 bprm->interp = kstrdup(interp, GFP_KERNEL);
1563 if (!bprm->interp) 1548 if (!bprm->interp)
1564 return -ENOMEM; 1549 return -ENOMEM;
1565 return 0; 1550 return 0;
1566 } 1551 }
1567 EXPORT_SYMBOL(bprm_change_interp); 1552 EXPORT_SYMBOL(bprm_change_interp);
1568 1553
1569 /* 1554 /*
1570 * determine how safe it is to execute the pr 1555 * determine how safe it is to execute the proposed program
1571 * - the caller must hold ->cred_guard_mutex 1556 * - the caller must hold ->cred_guard_mutex to protect against
1572 * PTRACE_ATTACH or seccomp thread-sync 1557 * PTRACE_ATTACH or seccomp thread-sync
1573 */ 1558 */
1574 static void check_unsafe_exec(struct linux_bi 1559 static void check_unsafe_exec(struct linux_binprm *bprm)
1575 { 1560 {
1576 struct task_struct *p = current, *t; 1561 struct task_struct *p = current, *t;
1577 unsigned n_fs; 1562 unsigned n_fs;
1578 1563
1579 if (p->ptrace) 1564 if (p->ptrace)
1580 bprm->unsafe |= LSM_UNSAFE_PT 1565 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1581 1566
1582 /* 1567 /*
1583 * This isn't strictly necessary, but 1568 * This isn't strictly necessary, but it makes it harder for LSMs to
1584 * mess up. 1569 * mess up.
1585 */ 1570 */
1586 if (task_no_new_privs(current)) 1571 if (task_no_new_privs(current))
1587 bprm->unsafe |= LSM_UNSAFE_NO 1572 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1588 1573
1589 /* !! 1574 t = p;
1590 * If another task is sharing our fs, <<
1591 * suid exec because the differently <<
1592 * will be able to manipulate the cur <<
1593 * It would be nice to force an unsha <<
1594 */ <<
1595 n_fs = 1; 1575 n_fs = 1;
1596 spin_lock(&p->fs->lock); 1576 spin_lock(&p->fs->lock);
1597 rcu_read_lock(); 1577 rcu_read_lock();
1598 for_other_threads(p, t) { !! 1578 while_each_thread(p, t) {
1599 if (t->fs == p->fs) 1579 if (t->fs == p->fs)
1600 n_fs++; 1580 n_fs++;
1601 } 1581 }
1602 rcu_read_unlock(); 1582 rcu_read_unlock();
1603 1583
1604 /* "users" and "in_exec" locked for c <<
1605 if (p->fs->users > n_fs) 1584 if (p->fs->users > n_fs)
1606 bprm->unsafe |= LSM_UNSAFE_SH 1585 bprm->unsafe |= LSM_UNSAFE_SHARE;
1607 else 1586 else
1608 p->fs->in_exec = 1; 1587 p->fs->in_exec = 1;
1609 spin_unlock(&p->fs->lock); 1588 spin_unlock(&p->fs->lock);
1610 } 1589 }
1611 1590
1612 static void bprm_fill_uid(struct linux_binprm 1591 static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1613 { 1592 {
1614 /* Handle suid and sgid on files */ 1593 /* Handle suid and sgid on files */
1615 struct mnt_idmap *idmap; !! 1594 struct user_namespace *mnt_userns;
1616 struct inode *inode = file_inode(file !! 1595 struct inode *inode;
1617 unsigned int mode; 1596 unsigned int mode;
1618 vfsuid_t vfsuid; !! 1597 kuid_t uid;
1619 vfsgid_t vfsgid; !! 1598 kgid_t gid;
1620 int err; <<
1621 1599
1622 if (!mnt_may_suid(file->f_path.mnt)) 1600 if (!mnt_may_suid(file->f_path.mnt))
1623 return; 1601 return;
1624 1602
1625 if (task_no_new_privs(current)) 1603 if (task_no_new_privs(current))
1626 return; 1604 return;
1627 1605
>> 1606 inode = file->f_path.dentry->d_inode;
1628 mode = READ_ONCE(inode->i_mode); 1607 mode = READ_ONCE(inode->i_mode);
1629 if (!(mode & (S_ISUID|S_ISGID))) 1608 if (!(mode & (S_ISUID|S_ISGID)))
1630 return; 1609 return;
1631 1610
1632 idmap = file_mnt_idmap(file); !! 1611 mnt_userns = file_mnt_user_ns(file);
1633 1612
1634 /* Be careful if suid/sgid is set */ 1613 /* Be careful if suid/sgid is set */
1635 inode_lock(inode); 1614 inode_lock(inode);
1636 1615
1637 /* Atomically reload and check mode/u !! 1616 /* reload atomically mode/uid/gid now that lock held */
1638 mode = inode->i_mode; 1617 mode = inode->i_mode;
1639 vfsuid = i_uid_into_vfsuid(idmap, ino !! 1618 uid = i_uid_into_mnt(mnt_userns, inode);
1640 vfsgid = i_gid_into_vfsgid(idmap, ino !! 1619 gid = i_gid_into_mnt(mnt_userns, inode);
1641 err = inode_permission(idmap, inode, <<
1642 inode_unlock(inode); 1620 inode_unlock(inode);
1643 1621
1644 /* Did the exec bit vanish out from u <<
1645 if (err) <<
1646 return; <<
1647 <<
1648 /* We ignore suid/sgid if there are n 1622 /* We ignore suid/sgid if there are no mappings for them in the ns */
1649 if (!vfsuid_has_mapping(bprm->cred->u !! 1623 if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1650 !vfsgid_has_mapping(bprm->cred->u !! 1624 !kgid_has_mapping(bprm->cred->user_ns, gid))
1651 return; 1625 return;
1652 1626
1653 if (mode & S_ISUID) { 1627 if (mode & S_ISUID) {
1654 bprm->per_clear |= PER_CLEAR_ 1628 bprm->per_clear |= PER_CLEAR_ON_SETID;
1655 bprm->cred->euid = vfsuid_int !! 1629 bprm->cred->euid = uid;
1656 } 1630 }
1657 1631
1658 if ((mode & (S_ISGID | S_IXGRP)) == ( 1632 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1659 bprm->per_clear |= PER_CLEAR_ 1633 bprm->per_clear |= PER_CLEAR_ON_SETID;
1660 bprm->cred->egid = vfsgid_int !! 1634 bprm->cred->egid = gid;
1661 } 1635 }
1662 } 1636 }
1663 1637
1664 /* 1638 /*
1665 * Compute brpm->cred based upon the final bi 1639 * Compute brpm->cred based upon the final binary.
1666 */ 1640 */
1667 static int bprm_creds_from_file(struct linux_ 1641 static int bprm_creds_from_file(struct linux_binprm *bprm)
1668 { 1642 {
1669 /* Compute creds based on which file? 1643 /* Compute creds based on which file? */
1670 struct file *file = bprm->execfd_cred 1644 struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1671 1645
1672 bprm_fill_uid(bprm, file); 1646 bprm_fill_uid(bprm, file);
1673 return security_bprm_creds_from_file( 1647 return security_bprm_creds_from_file(bprm, file);
1674 } 1648 }
1675 1649
1676 /* 1650 /*
1677 * Fill the binprm structure from the inode. 1651 * Fill the binprm structure from the inode.
1678 * Read the first BINPRM_BUF_SIZE bytes 1652 * Read the first BINPRM_BUF_SIZE bytes
1679 * 1653 *
1680 * This may be called multiple times for bina 1654 * This may be called multiple times for binary chains (scripts for example).
1681 */ 1655 */
1682 static int prepare_binprm(struct linux_binprm 1656 static int prepare_binprm(struct linux_binprm *bprm)
1683 { 1657 {
1684 loff_t pos = 0; 1658 loff_t pos = 0;
1685 1659
1686 memset(bprm->buf, 0, BINPRM_BUF_SIZE) 1660 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1687 return kernel_read(bprm->file, bprm-> 1661 return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1688 } 1662 }
1689 1663
1690 /* 1664 /*
1691 * Arguments are '\0' separated strings found 1665 * Arguments are '\0' separated strings found at the location bprm->p
1692 * points to; chop off the first by relocatin 1666 * points to; chop off the first by relocating brpm->p to right after
1693 * the first '\0' encountered. 1667 * the first '\0' encountered.
1694 */ 1668 */
1695 int remove_arg_zero(struct linux_binprm *bprm 1669 int remove_arg_zero(struct linux_binprm *bprm)
1696 { 1670 {
>> 1671 int ret = 0;
1697 unsigned long offset; 1672 unsigned long offset;
1698 char *kaddr; 1673 char *kaddr;
1699 struct page *page; 1674 struct page *page;
1700 1675
1701 if (!bprm->argc) 1676 if (!bprm->argc)
1702 return 0; 1677 return 0;
1703 1678
1704 do { 1679 do {
1705 offset = bprm->p & ~PAGE_MASK 1680 offset = bprm->p & ~PAGE_MASK;
1706 page = get_arg_page(bprm, bpr 1681 page = get_arg_page(bprm, bprm->p, 0);
1707 if (!page) !! 1682 if (!page) {
1708 return -EFAULT; !! 1683 ret = -EFAULT;
1709 kaddr = kmap_local_page(page) !! 1684 goto out;
>> 1685 }
>> 1686 kaddr = kmap_atomic(page);
1710 1687
1711 for (; offset < PAGE_SIZE && 1688 for (; offset < PAGE_SIZE && kaddr[offset];
1712 offset++, bpr 1689 offset++, bprm->p++)
1713 ; 1690 ;
1714 1691
1715 kunmap_local(kaddr); !! 1692 kunmap_atomic(kaddr);
1716 put_arg_page(page); 1693 put_arg_page(page);
1717 } while (offset == PAGE_SIZE); 1694 } while (offset == PAGE_SIZE);
1718 1695
1719 bprm->p++; 1696 bprm->p++;
1720 bprm->argc--; 1697 bprm->argc--;
>> 1698 ret = 0;
1721 1699
1722 return 0; !! 1700 out:
>> 1701 return ret;
1723 } 1702 }
1724 EXPORT_SYMBOL(remove_arg_zero); 1703 EXPORT_SYMBOL(remove_arg_zero);
1725 1704
1726 #define printable(c) (((c)=='\t') || ((c)=='\ 1705 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1727 /* 1706 /*
1728 * cycle the list of binary formats handler, 1707 * cycle the list of binary formats handler, until one recognizes the image
1729 */ 1708 */
1730 static int search_binary_handler(struct linux 1709 static int search_binary_handler(struct linux_binprm *bprm)
1731 { 1710 {
1732 bool need_retry = IS_ENABLED(CONFIG_M 1711 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1733 struct linux_binfmt *fmt; 1712 struct linux_binfmt *fmt;
1734 int retval; 1713 int retval;
1735 1714
1736 retval = prepare_binprm(bprm); 1715 retval = prepare_binprm(bprm);
1737 if (retval < 0) 1716 if (retval < 0)
1738 return retval; 1717 return retval;
1739 1718
1740 retval = security_bprm_check(bprm); 1719 retval = security_bprm_check(bprm);
1741 if (retval) 1720 if (retval)
1742 return retval; 1721 return retval;
1743 1722
1744 retval = -ENOENT; 1723 retval = -ENOENT;
1745 retry: 1724 retry:
1746 read_lock(&binfmt_lock); 1725 read_lock(&binfmt_lock);
1747 list_for_each_entry(fmt, &formats, lh 1726 list_for_each_entry(fmt, &formats, lh) {
1748 if (!try_module_get(fmt->modu 1727 if (!try_module_get(fmt->module))
1749 continue; 1728 continue;
1750 read_unlock(&binfmt_lock); 1729 read_unlock(&binfmt_lock);
1751 1730
1752 retval = fmt->load_binary(bpr 1731 retval = fmt->load_binary(bprm);
1753 1732
1754 read_lock(&binfmt_lock); 1733 read_lock(&binfmt_lock);
1755 put_binfmt(fmt); 1734 put_binfmt(fmt);
1756 if (bprm->point_of_no_return 1735 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1757 read_unlock(&binfmt_l 1736 read_unlock(&binfmt_lock);
1758 return retval; 1737 return retval;
1759 } 1738 }
1760 } 1739 }
1761 read_unlock(&binfmt_lock); 1740 read_unlock(&binfmt_lock);
1762 1741
1763 if (need_retry) { 1742 if (need_retry) {
1764 if (printable(bprm->buf[0]) & 1743 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1765 printable(bprm->buf[2]) & 1744 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1766 return retval; 1745 return retval;
1767 if (request_module("binfmt-%0 1746 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1768 return retval; 1747 return retval;
1769 need_retry = false; 1748 need_retry = false;
1770 goto retry; 1749 goto retry;
1771 } 1750 }
1772 1751
1773 return retval; 1752 return retval;
1774 } 1753 }
1775 1754
1776 /* binfmt handlers will call back into begin_ <<
1777 static int exec_binprm(struct linux_binprm *b 1755 static int exec_binprm(struct linux_binprm *bprm)
1778 { 1756 {
1779 pid_t old_pid, old_vpid; 1757 pid_t old_pid, old_vpid;
1780 int ret, depth; 1758 int ret, depth;
1781 1759
1782 /* Need to fetch pid before load_bina 1760 /* Need to fetch pid before load_binary changes it */
1783 old_pid = current->pid; 1761 old_pid = current->pid;
1784 rcu_read_lock(); 1762 rcu_read_lock();
1785 old_vpid = task_pid_nr_ns(current, ta 1763 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1786 rcu_read_unlock(); 1764 rcu_read_unlock();
1787 1765
1788 /* This allows 4 levels of binfmt rew 1766 /* This allows 4 levels of binfmt rewrites before failing hard. */
1789 for (depth = 0;; depth++) { 1767 for (depth = 0;; depth++) {
1790 struct file *exec; 1768 struct file *exec;
1791 if (depth > 5) 1769 if (depth > 5)
1792 return -ELOOP; 1770 return -ELOOP;
1793 1771
1794 ret = search_binary_handler(b 1772 ret = search_binary_handler(bprm);
1795 if (ret < 0) 1773 if (ret < 0)
1796 return ret; 1774 return ret;
1797 if (!bprm->interpreter) 1775 if (!bprm->interpreter)
1798 break; 1776 break;
1799 1777
1800 exec = bprm->file; 1778 exec = bprm->file;
1801 bprm->file = bprm->interprete 1779 bprm->file = bprm->interpreter;
1802 bprm->interpreter = NULL; 1780 bprm->interpreter = NULL;
1803 1781
>> 1782 allow_write_access(exec);
1804 if (unlikely(bprm->have_execf 1783 if (unlikely(bprm->have_execfd)) {
1805 if (bprm->executable) 1784 if (bprm->executable) {
1806 fput(exec); 1785 fput(exec);
1807 return -ENOEX 1786 return -ENOEXEC;
1808 } 1787 }
1809 bprm->executable = ex 1788 bprm->executable = exec;
1810 } else 1789 } else
1811 fput(exec); 1790 fput(exec);
1812 } 1791 }
1813 1792
1814 audit_bprm(bprm); 1793 audit_bprm(bprm);
1815 trace_sched_process_exec(current, old 1794 trace_sched_process_exec(current, old_pid, bprm);
1816 ptrace_event(PTRACE_EVENT_EXEC, old_v 1795 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1817 proc_exec_connector(current); 1796 proc_exec_connector(current);
1818 return 0; 1797 return 0;
1819 } 1798 }
1820 1799
1821 static int bprm_execve(struct linux_binprm *b !! 1800 /*
>> 1801 * sys_execve() executes a new program.
>> 1802 */
>> 1803 static int bprm_execve(struct linux_binprm *bprm,
>> 1804 int fd, struct filename *filename, int flags)
1822 { 1805 {
>> 1806 struct file *file;
1823 int retval; 1807 int retval;
1824 1808
1825 retval = prepare_bprm_creds(bprm); 1809 retval = prepare_bprm_creds(bprm);
1826 if (retval) 1810 if (retval)
1827 return retval; 1811 return retval;
1828 1812
1829 /* <<
1830 * Check for unsafe execution states <<
1831 * will call back into begin_new_exec <<
1832 * where setuid-ness is evaluated. <<
1833 */ <<
1834 check_unsafe_exec(bprm); 1813 check_unsafe_exec(bprm);
1835 current->in_execve = 1; 1814 current->in_execve = 1;
1836 sched_mm_cid_before_execve(current); !! 1815
>> 1816 file = do_open_execat(fd, filename, flags);
>> 1817 retval = PTR_ERR(file);
>> 1818 if (IS_ERR(file))
>> 1819 goto out_unmark;
1837 1820
1838 sched_exec(); 1821 sched_exec();
1839 1822
>> 1823 bprm->file = file;
>> 1824 /*
>> 1825 * Record that a name derived from an O_CLOEXEC fd will be
>> 1826 * inaccessible after exec. This allows the code in exec to
>> 1827 * choose to fail when the executable is not mmaped into the
>> 1828 * interpreter and an open file descriptor is not passed to
>> 1829 * the interpreter. This makes for a better user experience
>> 1830 * than having the interpreter start and then immediately fail
>> 1831 * when it finds the executable is inaccessible.
>> 1832 */
>> 1833 if (bprm->fdpath && get_close_on_exec(fd))
>> 1834 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
>> 1835
1840 /* Set the unchanging part of bprm->c 1836 /* Set the unchanging part of bprm->cred */
1841 retval = security_bprm_creds_for_exec 1837 retval = security_bprm_creds_for_exec(bprm);
1842 if (retval) 1838 if (retval)
1843 goto out; 1839 goto out;
1844 1840
1845 retval = ccs_exec_binprm(bprm); 1841 retval = ccs_exec_binprm(bprm);
1846 if (retval < 0) 1842 if (retval < 0)
1847 goto out; 1843 goto out;
1848 1844
1849 sched_mm_cid_after_execve(current); <<
1850 /* execve succeeded */ 1845 /* execve succeeded */
1851 current->fs->in_exec = 0; 1846 current->fs->in_exec = 0;
1852 current->in_execve = 0; 1847 current->in_execve = 0;
1853 rseq_execve(current); 1848 rseq_execve(current);
1854 user_events_execve(current); <<
1855 acct_update_integrals(current); 1849 acct_update_integrals(current);
1856 task_numa_free(current, false); 1850 task_numa_free(current, false);
1857 return retval; 1851 return retval;
1858 1852
1859 out: 1853 out:
1860 /* 1854 /*
1861 * If past the point of no return ens 1855 * If past the point of no return ensure the code never
1862 * returns to the userspace process. 1856 * returns to the userspace process. Use an existing fatal
1863 * signal if present otherwise termin 1857 * signal if present otherwise terminate the process with
1864 * SIGSEGV. 1858 * SIGSEGV.
1865 */ 1859 */
1866 if (bprm->point_of_no_return && !fata 1860 if (bprm->point_of_no_return && !fatal_signal_pending(current))
1867 force_fatal_sig(SIGSEGV); 1861 force_fatal_sig(SIGSEGV);
1868 1862
1869 sched_mm_cid_after_execve(current); !! 1863 out_unmark:
1870 current->fs->in_exec = 0; 1864 current->fs->in_exec = 0;
1871 current->in_execve = 0; 1865 current->in_execve = 0;
1872 1866
1873 return retval; 1867 return retval;
1874 } 1868 }
1875 1869
1876 static int do_execveat_common(int fd, struct 1870 static int do_execveat_common(int fd, struct filename *filename,
1877 struct user_arg 1871 struct user_arg_ptr argv,
1878 struct user_arg 1872 struct user_arg_ptr envp,
1879 int flags) 1873 int flags)
1880 { 1874 {
1881 struct linux_binprm *bprm; 1875 struct linux_binprm *bprm;
1882 int retval; 1876 int retval;
1883 1877
1884 if (IS_ERR(filename)) 1878 if (IS_ERR(filename))
1885 return PTR_ERR(filename); 1879 return PTR_ERR(filename);
1886 1880
1887 /* 1881 /*
1888 * We move the actual failure in case 1882 * We move the actual failure in case of RLIMIT_NPROC excess from
1889 * set*uid() to execve() because too 1883 * set*uid() to execve() because too many poorly written programs
1890 * don't check setuid() return code. 1884 * don't check setuid() return code. Here we additionally recheck
1891 * whether NPROC limit is still excee 1885 * whether NPROC limit is still exceeded.
1892 */ 1886 */
1893 if ((current->flags & PF_NPROC_EXCEED 1887 if ((current->flags & PF_NPROC_EXCEEDED) &&
1894 is_rlimit_overlimit(current_ucoun !! 1888 is_ucounts_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1895 retval = -EAGAIN; 1889 retval = -EAGAIN;
1896 goto out_ret; 1890 goto out_ret;
1897 } 1891 }
1898 1892
1899 /* We're below the limit (still or ag 1893 /* We're below the limit (still or again), so we don't want to make
1900 * further execve() calls fail. */ 1894 * further execve() calls fail. */
1901 current->flags &= ~PF_NPROC_EXCEEDED; 1895 current->flags &= ~PF_NPROC_EXCEEDED;
1902 1896
1903 bprm = alloc_bprm(fd, filename, flags !! 1897 bprm = alloc_bprm(fd, filename);
1904 if (IS_ERR(bprm)) { 1898 if (IS_ERR(bprm)) {
1905 retval = PTR_ERR(bprm); 1899 retval = PTR_ERR(bprm);
1906 goto out_ret; 1900 goto out_ret;
1907 } 1901 }
1908 1902
1909 retval = count(argv, MAX_ARG_STRINGS) 1903 retval = count(argv, MAX_ARG_STRINGS);
1910 if (retval == 0) 1904 if (retval == 0)
1911 pr_warn_once("process '%s' la 1905 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1912 current->comm, b 1906 current->comm, bprm->filename);
1913 if (retval < 0) 1907 if (retval < 0)
1914 goto out_free; 1908 goto out_free;
1915 bprm->argc = retval; 1909 bprm->argc = retval;
1916 1910
1917 retval = count(envp, MAX_ARG_STRINGS) 1911 retval = count(envp, MAX_ARG_STRINGS);
1918 if (retval < 0) 1912 if (retval < 0)
1919 goto out_free; 1913 goto out_free;
1920 bprm->envc = retval; 1914 bprm->envc = retval;
1921 1915
1922 retval = bprm_stack_limits(bprm); 1916 retval = bprm_stack_limits(bprm);
1923 if (retval < 0) 1917 if (retval < 0)
1924 goto out_free; 1918 goto out_free;
1925 1919
1926 retval = copy_string_kernel(bprm->fil 1920 retval = copy_string_kernel(bprm->filename, bprm);
1927 if (retval < 0) 1921 if (retval < 0)
1928 goto out_free; 1922 goto out_free;
1929 bprm->exec = bprm->p; 1923 bprm->exec = bprm->p;
1930 1924
1931 retval = copy_strings(bprm->envc, env 1925 retval = copy_strings(bprm->envc, envp, bprm);
1932 if (retval < 0) 1926 if (retval < 0)
1933 goto out_free; 1927 goto out_free;
1934 1928
1935 retval = copy_strings(bprm->argc, arg 1929 retval = copy_strings(bprm->argc, argv, bprm);
1936 if (retval < 0) 1930 if (retval < 0)
1937 goto out_free; 1931 goto out_free;
1938 1932
1939 /* 1933 /*
1940 * When argv is empty, add an empty s 1934 * When argv is empty, add an empty string ("") as argv[0] to
1941 * ensure confused userspace programs 1935 * ensure confused userspace programs that start processing
1942 * from argv[1] won't end up walking 1936 * from argv[1] won't end up walking envp. See also
1943 * bprm_stack_limits(). 1937 * bprm_stack_limits().
1944 */ 1938 */
1945 if (bprm->argc == 0) { 1939 if (bprm->argc == 0) {
1946 retval = copy_string_kernel(" 1940 retval = copy_string_kernel("", bprm);
1947 if (retval < 0) 1941 if (retval < 0)
1948 goto out_free; 1942 goto out_free;
1949 bprm->argc = 1; 1943 bprm->argc = 1;
1950 } 1944 }
1951 1945
1952 retval = bprm_execve(bprm); !! 1946 retval = bprm_execve(bprm, fd, filename, flags);
1953 out_free: 1947 out_free:
1954 free_bprm(bprm); 1948 free_bprm(bprm);
1955 1949
1956 out_ret: 1950 out_ret:
1957 putname(filename); 1951 putname(filename);
1958 return retval; 1952 return retval;
1959 } 1953 }
1960 1954
1961 int kernel_execve(const char *kernel_filename 1955 int kernel_execve(const char *kernel_filename,
1962 const char *const *argv, co 1956 const char *const *argv, const char *const *envp)
1963 { 1957 {
1964 struct filename *filename; 1958 struct filename *filename;
1965 struct linux_binprm *bprm; 1959 struct linux_binprm *bprm;
1966 int fd = AT_FDCWD; 1960 int fd = AT_FDCWD;
1967 int retval; 1961 int retval;
1968 1962
1969 /* It is non-sense for kernel threads !! 1963 if (WARN_ON_ONCE((current->flags & PF_KTHREAD) &&
1970 if (WARN_ON_ONCE(current->flags & PF_ !! 1964 (current->worker_private)))
1971 return -EINVAL; 1965 return -EINVAL;
1972 1966
1973 filename = getname_kernel(kernel_file 1967 filename = getname_kernel(kernel_filename);
1974 if (IS_ERR(filename)) 1968 if (IS_ERR(filename))
1975 return PTR_ERR(filename); 1969 return PTR_ERR(filename);
1976 1970
1977 bprm = alloc_bprm(fd, filename, 0); !! 1971 bprm = alloc_bprm(fd, filename);
1978 if (IS_ERR(bprm)) { 1972 if (IS_ERR(bprm)) {
1979 retval = PTR_ERR(bprm); 1973 retval = PTR_ERR(bprm);
1980 goto out_ret; 1974 goto out_ret;
1981 } 1975 }
1982 1976
1983 retval = count_strings_kernel(argv); 1977 retval = count_strings_kernel(argv);
1984 if (WARN_ON_ONCE(retval == 0)) 1978 if (WARN_ON_ONCE(retval == 0))
1985 retval = -EINVAL; 1979 retval = -EINVAL;
1986 if (retval < 0) 1980 if (retval < 0)
1987 goto out_free; 1981 goto out_free;
1988 bprm->argc = retval; 1982 bprm->argc = retval;
1989 1983
1990 retval = count_strings_kernel(envp); 1984 retval = count_strings_kernel(envp);
1991 if (retval < 0) 1985 if (retval < 0)
1992 goto out_free; 1986 goto out_free;
1993 bprm->envc = retval; 1987 bprm->envc = retval;
1994 1988
1995 retval = bprm_stack_limits(bprm); 1989 retval = bprm_stack_limits(bprm);
1996 if (retval < 0) 1990 if (retval < 0)
1997 goto out_free; 1991 goto out_free;
1998 1992
1999 retval = copy_string_kernel(bprm->fil 1993 retval = copy_string_kernel(bprm->filename, bprm);
2000 if (retval < 0) 1994 if (retval < 0)
2001 goto out_free; 1995 goto out_free;
2002 bprm->exec = bprm->p; 1996 bprm->exec = bprm->p;
2003 1997
2004 retval = copy_strings_kernel(bprm->en 1998 retval = copy_strings_kernel(bprm->envc, envp, bprm);
2005 if (retval < 0) 1999 if (retval < 0)
2006 goto out_free; 2000 goto out_free;
2007 2001
2008 retval = copy_strings_kernel(bprm->ar 2002 retval = copy_strings_kernel(bprm->argc, argv, bprm);
2009 if (retval < 0) 2003 if (retval < 0)
2010 goto out_free; 2004 goto out_free;
2011 2005
2012 retval = bprm_execve(bprm); !! 2006 retval = bprm_execve(bprm, fd, filename, 0);
2013 out_free: 2007 out_free:
2014 free_bprm(bprm); 2008 free_bprm(bprm);
2015 out_ret: 2009 out_ret:
2016 putname(filename); 2010 putname(filename);
2017 return retval; 2011 return retval;
2018 } 2012 }
2019 2013
2020 static int do_execve(struct filename *filenam 2014 static int do_execve(struct filename *filename,
2021 const char __user *const __user *__ar 2015 const char __user *const __user *__argv,
2022 const char __user *const __user *__en 2016 const char __user *const __user *__envp)
2023 { 2017 {
2024 struct user_arg_ptr argv = { .ptr.nat 2018 struct user_arg_ptr argv = { .ptr.native = __argv };
2025 struct user_arg_ptr envp = { .ptr.nat 2019 struct user_arg_ptr envp = { .ptr.native = __envp };
2026 return do_execveat_common(AT_FDCWD, f 2020 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2027 } 2021 }
2028 2022
2029 static int do_execveat(int fd, struct filenam 2023 static int do_execveat(int fd, struct filename *filename,
2030 const char __user *const __us 2024 const char __user *const __user *__argv,
2031 const char __user *const __us 2025 const char __user *const __user *__envp,
2032 int flags) 2026 int flags)
2033 { 2027 {
2034 struct user_arg_ptr argv = { .ptr.nat 2028 struct user_arg_ptr argv = { .ptr.native = __argv };
2035 struct user_arg_ptr envp = { .ptr.nat 2029 struct user_arg_ptr envp = { .ptr.native = __envp };
2036 2030
2037 return do_execveat_common(fd, filenam 2031 return do_execveat_common(fd, filename, argv, envp, flags);
2038 } 2032 }
2039 2033
2040 #ifdef CONFIG_COMPAT 2034 #ifdef CONFIG_COMPAT
2041 static int compat_do_execve(struct filename * 2035 static int compat_do_execve(struct filename *filename,
2042 const compat_uptr_t __user *__argv, 2036 const compat_uptr_t __user *__argv,
2043 const compat_uptr_t __user *__envp) 2037 const compat_uptr_t __user *__envp)
2044 { 2038 {
2045 struct user_arg_ptr argv = { 2039 struct user_arg_ptr argv = {
2046 .is_compat = true, 2040 .is_compat = true,
2047 .ptr.compat = __argv, 2041 .ptr.compat = __argv,
2048 }; 2042 };
2049 struct user_arg_ptr envp = { 2043 struct user_arg_ptr envp = {
2050 .is_compat = true, 2044 .is_compat = true,
2051 .ptr.compat = __envp, 2045 .ptr.compat = __envp,
2052 }; 2046 };
2053 return do_execveat_common(AT_FDCWD, f 2047 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2054 } 2048 }
2055 2049
2056 static int compat_do_execveat(int fd, struct 2050 static int compat_do_execveat(int fd, struct filename *filename,
2057 const compat_up 2051 const compat_uptr_t __user *__argv,
2058 const compat_up 2052 const compat_uptr_t __user *__envp,
2059 int flags) 2053 int flags)
2060 { 2054 {
2061 struct user_arg_ptr argv = { 2055 struct user_arg_ptr argv = {
2062 .is_compat = true, 2056 .is_compat = true,
2063 .ptr.compat = __argv, 2057 .ptr.compat = __argv,
2064 }; 2058 };
2065 struct user_arg_ptr envp = { 2059 struct user_arg_ptr envp = {
2066 .is_compat = true, 2060 .is_compat = true,
2067 .ptr.compat = __envp, 2061 .ptr.compat = __envp,
2068 }; 2062 };
2069 return do_execveat_common(fd, filenam 2063 return do_execveat_common(fd, filename, argv, envp, flags);
2070 } 2064 }
2071 #endif 2065 #endif
2072 2066
2073 void set_binfmt(struct linux_binfmt *new) 2067 void set_binfmt(struct linux_binfmt *new)
2074 { 2068 {
2075 struct mm_struct *mm = current->mm; 2069 struct mm_struct *mm = current->mm;
2076 2070
2077 if (mm->binfmt) 2071 if (mm->binfmt)
2078 module_put(mm->binfmt->module 2072 module_put(mm->binfmt->module);
2079 2073
2080 mm->binfmt = new; 2074 mm->binfmt = new;
2081 if (new) 2075 if (new)
2082 __module_get(new->module); 2076 __module_get(new->module);
2083 } 2077 }
2084 EXPORT_SYMBOL(set_binfmt); 2078 EXPORT_SYMBOL(set_binfmt);
2085 2079
2086 /* 2080 /*
2087 * set_dumpable stores three-value SUID_DUMP_ 2081 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2088 */ 2082 */
2089 void set_dumpable(struct mm_struct *mm, int v 2083 void set_dumpable(struct mm_struct *mm, int value)
2090 { 2084 {
2091 if (WARN_ON((unsigned)value > SUID_DU 2085 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2092 return; 2086 return;
2093 2087
2094 set_mask_bits(&mm->flags, MMF_DUMPABL 2088 set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2095 } 2089 }
2096 2090
2097 SYSCALL_DEFINE3(execve, 2091 SYSCALL_DEFINE3(execve,
2098 const char __user *, filename 2092 const char __user *, filename,
2099 const char __user *const __us 2093 const char __user *const __user *, argv,
2100 const char __user *const __us 2094 const char __user *const __user *, envp)
2101 { 2095 {
2102 return do_execve(getname(filename), a 2096 return do_execve(getname(filename), argv, envp);
2103 } 2097 }
2104 2098
2105 SYSCALL_DEFINE5(execveat, 2099 SYSCALL_DEFINE5(execveat,
2106 int, fd, const char __user *, 2100 int, fd, const char __user *, filename,
2107 const char __user *const __us 2101 const char __user *const __user *, argv,
2108 const char __user *const __us 2102 const char __user *const __user *, envp,
2109 int, flags) 2103 int, flags)
2110 { 2104 {
2111 return do_execveat(fd, 2105 return do_execveat(fd,
2112 getname_uflags(fil 2106 getname_uflags(filename, flags),
2113 argv, envp, flags) 2107 argv, envp, flags);
2114 } 2108 }
2115 2109
2116 #ifdef CONFIG_COMPAT 2110 #ifdef CONFIG_COMPAT
2117 COMPAT_SYSCALL_DEFINE3(execve, const char __u 2111 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2118 const compat_uptr_t __user *, argv, 2112 const compat_uptr_t __user *, argv,
2119 const compat_uptr_t __user *, envp) 2113 const compat_uptr_t __user *, envp)
2120 { 2114 {
2121 return compat_do_execve(getname(filen 2115 return compat_do_execve(getname(filename), argv, envp);
2122 } 2116 }
2123 2117
2124 COMPAT_SYSCALL_DEFINE5(execveat, int, fd, 2118 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2125 const char __user *, f 2119 const char __user *, filename,
2126 const compat_uptr_t __ 2120 const compat_uptr_t __user *, argv,
2127 const compat_uptr_t __ 2121 const compat_uptr_t __user *, envp,
2128 int, flags) 2122 int, flags)
2129 { 2123 {
2130 return compat_do_execveat(fd, 2124 return compat_do_execveat(fd,
2131 getname_ufl 2125 getname_uflags(filename, flags),
2132 argv, envp, 2126 argv, envp, flags);
2133 } 2127 }
2134 #endif 2128 #endif
2135 2129
2136 #ifdef CONFIG_SYSCTL 2130 #ifdef CONFIG_SYSCTL
2137 2131
2138 static int proc_dointvec_minmax_coredump(cons !! 2132 static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
2139 void *buffer, size_t *lenp, l 2133 void *buffer, size_t *lenp, loff_t *ppos)
2140 { 2134 {
2141 int error = proc_dointvec_minmax(tabl 2135 int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2142 2136
2143 if (!error) 2137 if (!error)
2144 validate_coredump_safety(); 2138 validate_coredump_safety();
2145 return error; 2139 return error;
2146 } 2140 }
2147 2141
2148 static struct ctl_table fs_exec_sysctls[] = { 2142 static struct ctl_table fs_exec_sysctls[] = {
2149 { 2143 {
2150 .procname = "suid_dumpa 2144 .procname = "suid_dumpable",
2151 .data = &suid_dumpa 2145 .data = &suid_dumpable,
2152 .maxlen = sizeof(int) 2146 .maxlen = sizeof(int),
2153 .mode = 0644, 2147 .mode = 0644,
2154 .proc_handler = proc_dointv 2148 .proc_handler = proc_dointvec_minmax_coredump,
2155 .extra1 = SYSCTL_ZERO 2149 .extra1 = SYSCTL_ZERO,
2156 .extra2 = SYSCTL_TWO, 2150 .extra2 = SYSCTL_TWO,
2157 }, 2151 },
>> 2152 { }
2158 }; 2153 };
2159 2154
2160 static int __init init_fs_exec_sysctls(void) 2155 static int __init init_fs_exec_sysctls(void)
2161 { 2156 {
2162 register_sysctl_init("fs", fs_exec_sy 2157 register_sysctl_init("fs", fs_exec_sysctls);
2163 return 0; 2158 return 0;
2164 } 2159 }
2165 2160
2166 fs_initcall(init_fs_exec_sysctls); 2161 fs_initcall(init_fs_exec_sysctls);
2167 #endif /* CONFIG_SYSCTL */ 2162 #endif /* CONFIG_SYSCTL */
2168 <<
2169 #ifdef CONFIG_EXEC_KUNIT_TEST <<
2170 #include "tests/exec_kunit.c" <<
2171 #endif <<
2172 2163
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.