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/stat.h> 31 #include <linux/stat.h>
32 #include <linux/fcntl.h> 32 #include <linux/fcntl.h>
33 #include <linux/swap.h> 33 #include <linux/swap.h>
34 #include <linux/string.h> 34 #include <linux/string.h>
35 #include <linux/init.h> 35 #include <linux/init.h>
36 #include <linux/sched/mm.h> 36 #include <linux/sched/mm.h>
37 #include <linux/sched/coredump.h> 37 #include <linux/sched/coredump.h>
38 #include <linux/sched/signal.h> 38 #include <linux/sched/signal.h>
39 #include <linux/sched/numa_balancing.h> 39 #include <linux/sched/numa_balancing.h>
40 #include <linux/sched/task.h> 40 #include <linux/sched/task.h>
41 #include <linux/pagemap.h> 41 #include <linux/pagemap.h>
42 #include <linux/perf_event.h> 42 #include <linux/perf_event.h>
43 #include <linux/highmem.h> 43 #include <linux/highmem.h>
44 #include <linux/spinlock.h> 44 #include <linux/spinlock.h>
45 #include <linux/key.h> 45 #include <linux/key.h>
46 #include <linux/personality.h> 46 #include <linux/personality.h>
47 #include <linux/binfmts.h> 47 #include <linux/binfmts.h>
48 #include <linux/utsname.h> 48 #include <linux/utsname.h>
49 #include <linux/pid_namespace.h> 49 #include <linux/pid_namespace.h>
50 #include <linux/module.h> 50 #include <linux/module.h>
51 #include <linux/namei.h> 51 #include <linux/namei.h>
52 #include <linux/mount.h> 52 #include <linux/mount.h>
53 #include <linux/security.h> 53 #include <linux/security.h>
54 #include <linux/syscalls.h> 54 #include <linux/syscalls.h>
55 #include <linux/tsacct_kern.h> 55 #include <linux/tsacct_kern.h>
56 #include <linux/cn_proc.h> 56 #include <linux/cn_proc.h>
57 #include <linux/audit.h> 57 #include <linux/audit.h>
58 #include <linux/kmod.h> 58 #include <linux/kmod.h>
59 #include <linux/fsnotify.h> 59 #include <linux/fsnotify.h>
60 #include <linux/fs_struct.h> 60 #include <linux/fs_struct.h>
61 #include <linux/oom.h> 61 #include <linux/oom.h>
62 #include <linux/compat.h> 62 #include <linux/compat.h>
63 #include <linux/vmalloc.h> 63 #include <linux/vmalloc.h>
64 #include <linux/io_uring.h> 64 #include <linux/io_uring.h>
65 #include <linux/syscall_user_dispatch.h> 65 #include <linux/syscall_user_dispatch.h>
66 #include <linux/coredump.h> 66 #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 67
72 #include <linux/uaccess.h> 68 #include <linux/uaccess.h>
73 #include <asm/mmu_context.h> 69 #include <asm/mmu_context.h>
74 #include <asm/tlb.h> 70 #include <asm/tlb.h>
75 71
76 #include <trace/events/task.h> 72 #include <trace/events/task.h>
77 #include "internal.h" 73 #include "internal.h"
78 74
79 #include <trace/events/sched.h> 75 #include <trace/events/sched.h>
80 76
81 static int bprm_creds_from_file(struct linux_b 77 static int bprm_creds_from_file(struct linux_binprm *bprm);
82 78
83 int suid_dumpable = 0; 79 int suid_dumpable = 0;
84 80
85 static LIST_HEAD(formats); 81 static LIST_HEAD(formats);
86 static DEFINE_RWLOCK(binfmt_lock); 82 static DEFINE_RWLOCK(binfmt_lock);
87 83
88 void __register_binfmt(struct linux_binfmt * f 84 void __register_binfmt(struct linux_binfmt * fmt, int insert)
89 { 85 {
90 write_lock(&binfmt_lock); 86 write_lock(&binfmt_lock);
91 insert ? list_add(&fmt->lh, &formats) 87 insert ? list_add(&fmt->lh, &formats) :
92 list_add_tail(&fmt->lh, &form 88 list_add_tail(&fmt->lh, &formats);
93 write_unlock(&binfmt_lock); 89 write_unlock(&binfmt_lock);
94 } 90 }
95 91
96 EXPORT_SYMBOL(__register_binfmt); 92 EXPORT_SYMBOL(__register_binfmt);
97 93
98 void unregister_binfmt(struct linux_binfmt * f 94 void unregister_binfmt(struct linux_binfmt * fmt)
99 { 95 {
100 write_lock(&binfmt_lock); 96 write_lock(&binfmt_lock);
101 list_del(&fmt->lh); 97 list_del(&fmt->lh);
102 write_unlock(&binfmt_lock); 98 write_unlock(&binfmt_lock);
103 } 99 }
104 100
105 EXPORT_SYMBOL(unregister_binfmt); 101 EXPORT_SYMBOL(unregister_binfmt);
106 102
107 static inline void put_binfmt(struct linux_bin 103 static inline void put_binfmt(struct linux_binfmt * fmt)
108 { 104 {
109 module_put(fmt->module); 105 module_put(fmt->module);
110 } 106 }
111 107
112 bool path_noexec(const struct path *path) 108 bool path_noexec(const struct path *path)
113 { 109 {
114 return (path->mnt->mnt_flags & MNT_NOE 110 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
115 (path->mnt->mnt_sb->s_iflags & 111 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
116 } 112 }
117 113
118 #ifdef CONFIG_USELIB 114 #ifdef CONFIG_USELIB
119 /* 115 /*
120 * Note that a shared library must be both rea 116 * Note that a shared library must be both readable and executable due to
121 * security reasons. 117 * security reasons.
122 * 118 *
123 * Also note that we take the address to load 119 * Also note that we take the address to load from the file itself.
124 */ 120 */
125 SYSCALL_DEFINE1(uselib, const char __user *, l 121 SYSCALL_DEFINE1(uselib, const char __user *, library)
126 { 122 {
127 struct linux_binfmt *fmt; 123 struct linux_binfmt *fmt;
128 struct file *file; 124 struct file *file;
129 struct filename *tmp = getname(library 125 struct filename *tmp = getname(library);
130 int error = PTR_ERR(tmp); 126 int error = PTR_ERR(tmp);
131 static const struct open_flags uselib_ 127 static const struct open_flags uselib_flags = {
132 .open_flag = O_LARGEFILE | O_R !! 128 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
133 .acc_mode = MAY_READ | MAY_EXE 129 .acc_mode = MAY_READ | MAY_EXEC,
134 .intent = LOOKUP_OPEN, 130 .intent = LOOKUP_OPEN,
135 .lookup_flags = LOOKUP_FOLLOW, 131 .lookup_flags = LOOKUP_FOLLOW,
136 }; 132 };
137 133
138 if (IS_ERR(tmp)) 134 if (IS_ERR(tmp))
139 goto out; 135 goto out;
140 136
141 file = do_filp_open(AT_FDCWD, tmp, &us 137 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
142 putname(tmp); 138 putname(tmp);
143 error = PTR_ERR(file); 139 error = PTR_ERR(file);
144 if (IS_ERR(file)) 140 if (IS_ERR(file))
145 goto out; 141 goto out;
146 142
147 /* 143 /*
148 * Check do_open_execat() for an expla 144 * Check do_open_execat() for an explanation.
149 */ 145 */
150 error = -EACCES; 146 error = -EACCES;
151 if (WARN_ON_ONCE(!S_ISREG(file_inode(f 147 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode)) ||
152 path_noexec(&file->f_path)) 148 path_noexec(&file->f_path))
153 goto exit; 149 goto exit;
154 150
>> 151 fsnotify_open(file);
>> 152
155 error = -ENOEXEC; 153 error = -ENOEXEC;
156 154
157 read_lock(&binfmt_lock); 155 read_lock(&binfmt_lock);
158 list_for_each_entry(fmt, &formats, lh) 156 list_for_each_entry(fmt, &formats, lh) {
159 if (!fmt->load_shlib) 157 if (!fmt->load_shlib)
160 continue; 158 continue;
161 if (!try_module_get(fmt->modul 159 if (!try_module_get(fmt->module))
162 continue; 160 continue;
163 read_unlock(&binfmt_lock); 161 read_unlock(&binfmt_lock);
164 error = fmt->load_shlib(file); 162 error = fmt->load_shlib(file);
165 read_lock(&binfmt_lock); 163 read_lock(&binfmt_lock);
166 put_binfmt(fmt); 164 put_binfmt(fmt);
167 if (error != -ENOEXEC) 165 if (error != -ENOEXEC)
168 break; 166 break;
169 } 167 }
170 read_unlock(&binfmt_lock); 168 read_unlock(&binfmt_lock);
171 exit: 169 exit:
172 fput(file); 170 fput(file);
173 out: 171 out:
174 return error; !! 172 return error;
175 } 173 }
176 #endif /* #ifdef CONFIG_USELIB */ 174 #endif /* #ifdef CONFIG_USELIB */
177 175
178 #ifdef CONFIG_MMU 176 #ifdef CONFIG_MMU
179 /* 177 /*
180 * The nascent bprm->mm is not visible until e 178 * The nascent bprm->mm is not visible until exec_mmap() but it can
181 * use a lot of memory, account these pages in 179 * use a lot of memory, account these pages in current->mm temporary
182 * for oom_badness()->get_mm_rss(). Once exec 180 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
183 * change the counter back via acct_arg_size(0 181 * change the counter back via acct_arg_size(0).
184 */ 182 */
185 static void acct_arg_size(struct linux_binprm 183 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
186 { 184 {
187 struct mm_struct *mm = current->mm; 185 struct mm_struct *mm = current->mm;
188 long diff = (long)(pages - bprm->vma_p 186 long diff = (long)(pages - bprm->vma_pages);
189 187
190 if (!mm || !diff) 188 if (!mm || !diff)
191 return; 189 return;
192 190
193 bprm->vma_pages = pages; 191 bprm->vma_pages = pages;
194 add_mm_counter(mm, MM_ANONPAGES, diff) 192 add_mm_counter(mm, MM_ANONPAGES, diff);
195 } 193 }
196 194
197 static struct page *get_arg_page(struct linux_ 195 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
198 int write) 196 int write)
199 { 197 {
200 struct page *page; 198 struct page *page;
201 struct vm_area_struct *vma = bprm->vma 199 struct vm_area_struct *vma = bprm->vma;
202 struct mm_struct *mm = bprm->mm; 200 struct mm_struct *mm = bprm->mm;
203 int ret; 201 int ret;
204 202
205 /* 203 /*
206 * Avoid relying on expanding the stac 204 * Avoid relying on expanding the stack down in GUP (which
207 * does not work for STACK_GROWSUP any 205 * does not work for STACK_GROWSUP anyway), and just do it
208 * by hand ahead of time. 206 * by hand ahead of time.
209 */ 207 */
210 if (write && pos < vma->vm_start) { 208 if (write && pos < vma->vm_start) {
211 mmap_write_lock(mm); 209 mmap_write_lock(mm);
212 ret = expand_downwards(vma, po 210 ret = expand_downwards(vma, pos);
213 if (unlikely(ret < 0)) { 211 if (unlikely(ret < 0)) {
214 mmap_write_unlock(mm); 212 mmap_write_unlock(mm);
215 return NULL; 213 return NULL;
216 } 214 }
217 mmap_write_downgrade(mm); 215 mmap_write_downgrade(mm);
218 } else 216 } else
219 mmap_read_lock(mm); 217 mmap_read_lock(mm);
220 218
221 /* 219 /*
222 * We are doing an exec(). 'current' 220 * We are doing an exec(). 'current' is the process
223 * doing the exec and 'mm' is the new 221 * doing the exec and 'mm' is the new process's mm.
224 */ 222 */
225 ret = get_user_pages_remote(mm, pos, 1 223 ret = get_user_pages_remote(mm, pos, 1,
226 write ? FOLL_WRITE : 0 224 write ? FOLL_WRITE : 0,
227 &page, NULL); !! 225 &page, NULL, NULL);
228 mmap_read_unlock(mm); 226 mmap_read_unlock(mm);
229 if (ret <= 0) 227 if (ret <= 0)
230 return NULL; 228 return NULL;
231 229
232 if (write) 230 if (write)
233 acct_arg_size(bprm, vma_pages( 231 acct_arg_size(bprm, vma_pages(vma));
234 232
235 return page; 233 return page;
236 } 234 }
237 235
238 static void put_arg_page(struct page *page) 236 static void put_arg_page(struct page *page)
239 { 237 {
240 put_page(page); 238 put_page(page);
241 } 239 }
242 240
243 static void free_arg_pages(struct linux_binprm 241 static void free_arg_pages(struct linux_binprm *bprm)
244 { 242 {
245 } 243 }
246 244
247 static void flush_arg_page(struct linux_binprm 245 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
248 struct page *page) 246 struct page *page)
249 { 247 {
250 flush_cache_page(bprm->vma, pos, page_ 248 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
251 } 249 }
252 250
253 static int __bprm_mm_init(struct linux_binprm 251 static int __bprm_mm_init(struct linux_binprm *bprm)
254 { 252 {
255 int err; 253 int err;
256 struct vm_area_struct *vma = NULL; 254 struct vm_area_struct *vma = NULL;
257 struct mm_struct *mm = bprm->mm; 255 struct mm_struct *mm = bprm->mm;
258 256
259 bprm->vma = vma = vm_area_alloc(mm); 257 bprm->vma = vma = vm_area_alloc(mm);
260 if (!vma) 258 if (!vma)
261 return -ENOMEM; 259 return -ENOMEM;
262 vma_set_anonymous(vma); 260 vma_set_anonymous(vma);
263 261
264 if (mmap_write_lock_killable(mm)) { 262 if (mmap_write_lock_killable(mm)) {
265 err = -EINTR; 263 err = -EINTR;
266 goto err_free; 264 goto err_free;
267 } 265 }
268 266
269 /* 267 /*
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 268 * Place the stack at the largest stack address the architecture
279 * supports. Later, we'll move this to 269 * supports. Later, we'll move this to an appropriate place. We don't
280 * use STACK_TOP because that can depe 270 * use STACK_TOP because that can depend on attributes which aren't
281 * configured yet. 271 * configured yet.
282 */ 272 */
283 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK 273 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
284 vma->vm_end = STACK_TOP_MAX; 274 vma->vm_end = STACK_TOP_MAX;
285 vma->vm_start = vma->vm_end - PAGE_SIZ 275 vma->vm_start = vma->vm_end - PAGE_SIZE;
286 vm_flags_init(vma, VM_SOFTDIRTY | VM_S !! 276 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
287 vma->vm_page_prot = vm_get_page_prot(v 277 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
288 278
289 err = insert_vm_struct(mm, vma); 279 err = insert_vm_struct(mm, vma);
290 if (err) 280 if (err)
291 goto err; 281 goto err;
292 282
293 mm->stack_vm = mm->total_vm = 1; 283 mm->stack_vm = mm->total_vm = 1;
294 mmap_write_unlock(mm); 284 mmap_write_unlock(mm);
295 bprm->p = vma->vm_end - sizeof(void *) 285 bprm->p = vma->vm_end - sizeof(void *);
296 return 0; 286 return 0;
297 err: 287 err:
298 ksm_exit(mm); <<
299 err_ksm: <<
300 mmap_write_unlock(mm); 288 mmap_write_unlock(mm);
301 err_free: 289 err_free:
302 bprm->vma = NULL; 290 bprm->vma = NULL;
303 vm_area_free(vma); 291 vm_area_free(vma);
304 return err; 292 return err;
305 } 293 }
306 294
307 static bool valid_arg_len(struct linux_binprm 295 static bool valid_arg_len(struct linux_binprm *bprm, long len)
308 { 296 {
309 return len <= MAX_ARG_STRLEN; 297 return len <= MAX_ARG_STRLEN;
310 } 298 }
311 299
312 #else 300 #else
313 301
314 static inline void acct_arg_size(struct linux_ 302 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
315 { 303 {
316 } 304 }
317 305
318 static struct page *get_arg_page(struct linux_ 306 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
319 int write) 307 int write)
320 { 308 {
321 struct page *page; 309 struct page *page;
322 310
323 page = bprm->page[pos / PAGE_SIZE]; 311 page = bprm->page[pos / PAGE_SIZE];
324 if (!page && write) { 312 if (!page && write) {
325 page = alloc_page(GFP_HIGHUSER 313 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
326 if (!page) 314 if (!page)
327 return NULL; 315 return NULL;
328 bprm->page[pos / PAGE_SIZE] = 316 bprm->page[pos / PAGE_SIZE] = page;
329 } 317 }
330 318
331 return page; 319 return page;
332 } 320 }
333 321
334 static void put_arg_page(struct page *page) 322 static void put_arg_page(struct page *page)
335 { 323 {
336 } 324 }
337 325
338 static void free_arg_page(struct linux_binprm 326 static void free_arg_page(struct linux_binprm *bprm, int i)
339 { 327 {
340 if (bprm->page[i]) { 328 if (bprm->page[i]) {
341 __free_page(bprm->page[i]); 329 __free_page(bprm->page[i]);
342 bprm->page[i] = NULL; 330 bprm->page[i] = NULL;
343 } 331 }
344 } 332 }
345 333
346 static void free_arg_pages(struct linux_binprm 334 static void free_arg_pages(struct linux_binprm *bprm)
347 { 335 {
348 int i; 336 int i;
349 337
350 for (i = 0; i < MAX_ARG_PAGES; i++) 338 for (i = 0; i < MAX_ARG_PAGES; i++)
351 free_arg_page(bprm, i); 339 free_arg_page(bprm, i);
352 } 340 }
353 341
354 static void flush_arg_page(struct linux_binprm 342 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
355 struct page *page) 343 struct page *page)
356 { 344 {
357 } 345 }
358 346
359 static int __bprm_mm_init(struct linux_binprm 347 static int __bprm_mm_init(struct linux_binprm *bprm)
360 { 348 {
361 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - 349 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
362 return 0; 350 return 0;
363 } 351 }
364 352
365 static bool valid_arg_len(struct linux_binprm 353 static bool valid_arg_len(struct linux_binprm *bprm, long len)
366 { 354 {
367 return len <= bprm->p; 355 return len <= bprm->p;
368 } 356 }
369 357
370 #endif /* CONFIG_MMU */ 358 #endif /* CONFIG_MMU */
371 359
372 /* 360 /*
373 * Create a new mm_struct and populate it with 361 * Create a new mm_struct and populate it with a temporary stack
374 * vm_area_struct. We don't have enough conte 362 * 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 363 * flags, permissions, and offset, so we use temporary values. We'll update
376 * them later in setup_arg_pages(). 364 * them later in setup_arg_pages().
377 */ 365 */
378 static int bprm_mm_init(struct linux_binprm *b 366 static int bprm_mm_init(struct linux_binprm *bprm)
379 { 367 {
380 int err; 368 int err;
381 struct mm_struct *mm = NULL; 369 struct mm_struct *mm = NULL;
382 370
383 bprm->mm = mm = mm_alloc(); 371 bprm->mm = mm = mm_alloc();
384 err = -ENOMEM; 372 err = -ENOMEM;
385 if (!mm) 373 if (!mm)
386 goto err; 374 goto err;
387 375
388 /* Save current stack limit for all ca 376 /* Save current stack limit for all calculations made during exec. */
389 task_lock(current->group_leader); 377 task_lock(current->group_leader);
390 bprm->rlim_stack = current->signal->rl 378 bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
391 task_unlock(current->group_leader); 379 task_unlock(current->group_leader);
392 380
393 err = __bprm_mm_init(bprm); 381 err = __bprm_mm_init(bprm);
394 if (err) 382 if (err)
395 goto err; 383 goto err;
396 384
397 return 0; 385 return 0;
398 386
399 err: 387 err:
400 if (mm) { 388 if (mm) {
401 bprm->mm = NULL; 389 bprm->mm = NULL;
402 mmdrop(mm); 390 mmdrop(mm);
403 } 391 }
404 392
405 return err; 393 return err;
406 } 394 }
407 395
408 struct user_arg_ptr { 396 struct user_arg_ptr {
409 #ifdef CONFIG_COMPAT 397 #ifdef CONFIG_COMPAT
410 bool is_compat; 398 bool is_compat;
411 #endif 399 #endif
412 union { 400 union {
413 const char __user *const __use 401 const char __user *const __user *native;
414 #ifdef CONFIG_COMPAT 402 #ifdef CONFIG_COMPAT
415 const compat_uptr_t __user *co 403 const compat_uptr_t __user *compat;
416 #endif 404 #endif
417 } ptr; 405 } ptr;
418 }; 406 };
419 407
420 static const char __user *get_user_arg_ptr(str 408 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
421 { 409 {
422 const char __user *native; 410 const char __user *native;
423 411
424 #ifdef CONFIG_COMPAT 412 #ifdef CONFIG_COMPAT
425 if (unlikely(argv.is_compat)) { 413 if (unlikely(argv.is_compat)) {
426 compat_uptr_t compat; 414 compat_uptr_t compat;
427 415
428 if (get_user(compat, argv.ptr. 416 if (get_user(compat, argv.ptr.compat + nr))
429 return ERR_PTR(-EFAULT 417 return ERR_PTR(-EFAULT);
430 418
431 return compat_ptr(compat); 419 return compat_ptr(compat);
432 } 420 }
433 #endif 421 #endif
434 422
435 if (get_user(native, argv.ptr.native + 423 if (get_user(native, argv.ptr.native + nr))
436 return ERR_PTR(-EFAULT); 424 return ERR_PTR(-EFAULT);
437 425
438 return native; 426 return native;
439 } 427 }
440 428
441 /* 429 /*
442 * count() counts the number of strings in arr 430 * count() counts the number of strings in array ARGV.
443 */ 431 */
444 static int count(struct user_arg_ptr argv, int 432 static int count(struct user_arg_ptr argv, int max)
445 { 433 {
446 int i = 0; 434 int i = 0;
447 435
448 if (argv.ptr.native != NULL) { 436 if (argv.ptr.native != NULL) {
449 for (;;) { 437 for (;;) {
450 const char __user *p = 438 const char __user *p = get_user_arg_ptr(argv, i);
451 439
452 if (!p) 440 if (!p)
453 break; 441 break;
454 442
455 if (IS_ERR(p)) 443 if (IS_ERR(p))
456 return -EFAULT 444 return -EFAULT;
457 445
458 if (i >= max) 446 if (i >= max)
459 return -E2BIG; 447 return -E2BIG;
460 ++i; 448 ++i;
461 449
462 if (fatal_signal_pendi 450 if (fatal_signal_pending(current))
463 return -ERESTA 451 return -ERESTARTNOHAND;
464 cond_resched(); 452 cond_resched();
465 } 453 }
466 } 454 }
467 return i; 455 return i;
468 } 456 }
469 457
470 static int count_strings_kernel(const char *co 458 static int count_strings_kernel(const char *const *argv)
471 { 459 {
472 int i; 460 int i;
473 461
474 if (!argv) 462 if (!argv)
475 return 0; 463 return 0;
476 464
477 for (i = 0; argv[i]; ++i) { 465 for (i = 0; argv[i]; ++i) {
478 if (i >= MAX_ARG_STRINGS) 466 if (i >= MAX_ARG_STRINGS)
479 return -E2BIG; 467 return -E2BIG;
480 if (fatal_signal_pending(curre 468 if (fatal_signal_pending(current))
481 return -ERESTARTNOHAND 469 return -ERESTARTNOHAND;
482 cond_resched(); 470 cond_resched();
483 } 471 }
484 return i; 472 return i;
485 } 473 }
486 474
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 475 static int bprm_stack_limits(struct linux_binprm *bprm)
517 { 476 {
518 unsigned long limit, ptr_size; 477 unsigned long limit, ptr_size;
519 478
520 /* 479 /*
521 * Limit to 1/4 of the max stack size 480 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
522 * (whichever is smaller) for the argv 481 * (whichever is smaller) for the argv+env strings.
523 * This ensures that: 482 * This ensures that:
524 * - the remaining binfmt code will n 483 * - the remaining binfmt code will not run out of stack space,
525 * - the program will have a reasonab 484 * - the program will have a reasonable amount of stack left
526 * to work from. 485 * to work from.
527 */ 486 */
528 limit = _STK_LIM / 4 * 3; 487 limit = _STK_LIM / 4 * 3;
529 limit = min(limit, bprm->rlim_stack.rl 488 limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
530 /* 489 /*
531 * We've historically supported up to 490 * We've historically supported up to 32 pages (ARG_MAX)
532 * of argument strings even with small 491 * of argument strings even with small stacks
533 */ 492 */
534 limit = max_t(unsigned long, limit, AR 493 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 /* 494 /*
539 * We must account for the size of all 495 * We must account for the size of all the argv and envp pointers to
540 * the argv and envp strings, since th 496 * the argv and envp strings, since they will also take up space in
541 * the stack. They aren't stored until 497 * the stack. They aren't stored until much later when we can't
542 * signal to the parent that the child 498 * signal to the parent that the child has run out of stack space.
543 * Instead, calculate it here so it's 499 * Instead, calculate it here so it's possible to fail gracefully.
544 * 500 *
545 * In the case of argc = 0, make sure 501 * In the case of argc = 0, make sure there is space for adding a
546 * empty string (which will bump argc 502 * empty string (which will bump argc to 1), to ensure confused
547 * userspace programs don't start proc 503 * userspace programs don't start processing from argv[1], thinking
548 * argc can never be 0, to keep them f 504 * argc can never be 0, to keep them from walking envp by accident.
549 * See do_execveat_common(). 505 * See do_execveat_common().
550 */ 506 */
551 if (check_add_overflow(max(bprm->argc, !! 507 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) 508 if (limit <= ptr_size)
555 return -E2BIG; 509 return -E2BIG;
556 limit -= ptr_size; 510 limit -= ptr_size;
557 511
558 return bprm_set_stack_limit(bprm, limi !! 512 bprm->argmin = bprm->p - limit;
>> 513 return 0;
559 } 514 }
560 515
561 /* 516 /*
562 * 'copy_strings()' copies argument/environmen 517 * 'copy_strings()' copies argument/environment strings from the old
563 * processes's memory to the new process's sta 518 * processes's memory to the new process's stack. The call to get_user_pages()
564 * ensures the destination page is created and 519 * ensures the destination page is created and not swapped out.
565 */ 520 */
566 static int copy_strings(int argc, struct user_ 521 static int copy_strings(int argc, struct user_arg_ptr argv,
567 struct linux_binprm *b 522 struct linux_binprm *bprm)
568 { 523 {
569 struct page *kmapped_page = NULL; 524 struct page *kmapped_page = NULL;
570 char *kaddr = NULL; 525 char *kaddr = NULL;
571 unsigned long kpos = 0; 526 unsigned long kpos = 0;
572 int ret; 527 int ret;
573 528
574 while (argc-- > 0) { 529 while (argc-- > 0) {
575 const char __user *str; 530 const char __user *str;
576 int len; 531 int len;
577 unsigned long pos; 532 unsigned long pos;
578 533
579 ret = -EFAULT; 534 ret = -EFAULT;
580 str = get_user_arg_ptr(argv, a 535 str = get_user_arg_ptr(argv, argc);
581 if (IS_ERR(str)) 536 if (IS_ERR(str))
582 goto out; 537 goto out;
583 538
584 len = strnlen_user(str, MAX_AR 539 len = strnlen_user(str, MAX_ARG_STRLEN);
585 if (!len) 540 if (!len)
586 goto out; 541 goto out;
587 542
588 ret = -E2BIG; 543 ret = -E2BIG;
589 if (!valid_arg_len(bprm, len)) 544 if (!valid_arg_len(bprm, len))
590 goto out; 545 goto out;
591 546
592 /* We're going to work our way 547 /* We're going to work our way backwards. */
593 pos = bprm->p; 548 pos = bprm->p;
594 str += len; 549 str += len;
595 bprm->p -= len; 550 bprm->p -= len;
596 if (bprm_hit_stack_limit(bprm) !! 551 #ifdef CONFIG_MMU
>> 552 if (bprm->p < bprm->argmin)
597 goto out; 553 goto out;
>> 554 #endif
598 555
599 while (len > 0) { 556 while (len > 0) {
600 int offset, bytes_to_c 557 int offset, bytes_to_copy;
601 558
602 if (fatal_signal_pendi 559 if (fatal_signal_pending(current)) {
603 ret = -ERESTAR 560 ret = -ERESTARTNOHAND;
604 goto out; 561 goto out;
605 } 562 }
606 cond_resched(); 563 cond_resched();
607 564
608 offset = pos % PAGE_SI 565 offset = pos % PAGE_SIZE;
609 if (offset == 0) 566 if (offset == 0)
610 offset = PAGE_ 567 offset = PAGE_SIZE;
611 568
612 bytes_to_copy = offset 569 bytes_to_copy = offset;
613 if (bytes_to_copy > le 570 if (bytes_to_copy > len)
614 bytes_to_copy 571 bytes_to_copy = len;
615 572
616 offset -= bytes_to_cop 573 offset -= bytes_to_copy;
617 pos -= bytes_to_copy; 574 pos -= bytes_to_copy;
618 str -= bytes_to_copy; 575 str -= bytes_to_copy;
619 len -= bytes_to_copy; 576 len -= bytes_to_copy;
620 577
621 if (!kmapped_page || k 578 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
622 struct page *p 579 struct page *page;
623 580
624 page = get_arg 581 page = get_arg_page(bprm, pos, 1);
625 if (!page) { 582 if (!page) {
626 ret = 583 ret = -E2BIG;
627 goto o 584 goto out;
628 } 585 }
629 586
630 if (kmapped_pa 587 if (kmapped_page) {
631 flush_ 588 flush_dcache_page(kmapped_page);
632 kunmap 589 kunmap_local(kaddr);
633 put_ar 590 put_arg_page(kmapped_page);
634 } 591 }
635 kmapped_page = 592 kmapped_page = page;
636 kaddr = kmap_l 593 kaddr = kmap_local_page(kmapped_page);
637 kpos = pos & P 594 kpos = pos & PAGE_MASK;
638 flush_arg_page 595 flush_arg_page(bprm, kpos, kmapped_page);
639 } 596 }
640 if (copy_from_user(kad 597 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
641 ret = -EFAULT; 598 ret = -EFAULT;
642 goto out; 599 goto out;
643 } 600 }
644 } 601 }
645 } 602 }
646 ret = 0; 603 ret = 0;
647 out: 604 out:
648 if (kmapped_page) { 605 if (kmapped_page) {
649 flush_dcache_page(kmapped_page 606 flush_dcache_page(kmapped_page);
650 kunmap_local(kaddr); 607 kunmap_local(kaddr);
651 put_arg_page(kmapped_page); 608 put_arg_page(kmapped_page);
652 } 609 }
653 return ret; 610 return ret;
654 } 611 }
655 612
656 /* 613 /*
657 * Copy and argument/environment string from t 614 * Copy and argument/environment string from the kernel to the processes stack.
658 */ 615 */
659 int copy_string_kernel(const char *arg, struct 616 int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
660 { 617 {
661 int len = strnlen(arg, MAX_ARG_STRLEN) 618 int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
662 unsigned long pos = bprm->p; 619 unsigned long pos = bprm->p;
663 620
664 if (len == 0) 621 if (len == 0)
665 return -EFAULT; 622 return -EFAULT;
666 if (!valid_arg_len(bprm, len)) 623 if (!valid_arg_len(bprm, len))
667 return -E2BIG; 624 return -E2BIG;
668 625
669 /* We're going to work our way backwar 626 /* We're going to work our way backwards. */
670 arg += len; 627 arg += len;
671 bprm->p -= len; 628 bprm->p -= len;
672 if (bprm_hit_stack_limit(bprm)) !! 629 if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
673 return -E2BIG; 630 return -E2BIG;
674 631
675 while (len > 0) { 632 while (len > 0) {
676 unsigned int bytes_to_copy = m 633 unsigned int bytes_to_copy = min_t(unsigned int, len,
677 min_not_zero(o 634 min_not_zero(offset_in_page(pos), PAGE_SIZE));
678 struct page *page; 635 struct page *page;
679 636
680 pos -= bytes_to_copy; 637 pos -= bytes_to_copy;
681 arg -= bytes_to_copy; 638 arg -= bytes_to_copy;
682 len -= bytes_to_copy; 639 len -= bytes_to_copy;
683 640
684 page = get_arg_page(bprm, pos, 641 page = get_arg_page(bprm, pos, 1);
685 if (!page) 642 if (!page)
686 return -E2BIG; 643 return -E2BIG;
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_to_page(page, offset_in_page(pos), arg, bytes_to_copy);
689 put_arg_page(page); 646 put_arg_page(page);
690 } 647 }
691 648
692 return 0; 649 return 0;
693 } 650 }
694 EXPORT_SYMBOL(copy_string_kernel); 651 EXPORT_SYMBOL(copy_string_kernel);
695 652
696 static int copy_strings_kernel(int argc, const 653 static int copy_strings_kernel(int argc, const char *const *argv,
697 struct linux_bi 654 struct linux_binprm *bprm)
698 { 655 {
699 while (argc-- > 0) { 656 while (argc-- > 0) {
700 int ret = copy_string_kernel(a 657 int ret = copy_string_kernel(argv[argc], bprm);
701 if (ret < 0) 658 if (ret < 0)
702 return ret; 659 return ret;
703 if (fatal_signal_pending(curre 660 if (fatal_signal_pending(current))
704 return -ERESTARTNOHAND 661 return -ERESTARTNOHAND;
705 cond_resched(); 662 cond_resched();
706 } 663 }
707 return 0; 664 return 0;
708 } 665 }
709 666
710 #ifdef CONFIG_MMU 667 #ifdef CONFIG_MMU
711 668
712 /* 669 /*
>> 670 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
>> 671 * the binfmt code determines where the new stack should reside, we shift it to
>> 672 * its final location. The process proceeds as follows:
>> 673 *
>> 674 * 1) Use shift to calculate the new vma endpoints.
>> 675 * 2) Extend vma to cover both the old and new ranges. This ensures the
>> 676 * arguments passed to subsequent functions are consistent.
>> 677 * 3) Move vma's page tables to the new range.
>> 678 * 4) Free up any cleared pgd range.
>> 679 * 5) Shrink the vma to cover only the new range.
>> 680 */
>> 681 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
>> 682 {
>> 683 struct mm_struct *mm = vma->vm_mm;
>> 684 unsigned long old_start = vma->vm_start;
>> 685 unsigned long old_end = vma->vm_end;
>> 686 unsigned long length = old_end - old_start;
>> 687 unsigned long new_start = old_start - shift;
>> 688 unsigned long new_end = old_end - shift;
>> 689 VMA_ITERATOR(vmi, mm, new_start);
>> 690 struct vm_area_struct *next;
>> 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 != vma_next(&vmi))
>> 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 next = vma_next(&vmi);
>> 719 if (new_end > old_start) {
>> 720 /*
>> 721 * when the old and new regions overlap clear from new_end.
>> 722 */
>> 723 free_pgd_range(&tlb, new_end, old_end, new_end,
>> 724 next ? next->vm_start : USER_PGTABLES_CEILING);
>> 725 } else {
>> 726 /*
>> 727 * otherwise, clean from old_start; this is done to not touch
>> 728 * the address space in [new_end, old_start) some architectures
>> 729 * have constraints on va-space that make this illegal (IA64) -
>> 730 * for the others its just a little faster.
>> 731 */
>> 732 free_pgd_range(&tlb, old_start, old_end, new_end,
>> 733 next ? next->vm_start : USER_PGTABLES_CEILING);
>> 734 }
>> 735 tlb_finish_mmu(&tlb);
>> 736
>> 737 /*
>> 738 * Shrink the vma to just the new range. Always succeeds.
>> 739 */
>> 740 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
>> 741
>> 742 return 0;
>> 743 }
>> 744
>> 745 /*
713 * Finalizes the stack vm_area_struct. The fla 746 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
714 * the stack is optionally relocated, and some 747 * the stack is optionally relocated, and some extra space is added.
715 */ 748 */
716 int setup_arg_pages(struct linux_binprm *bprm, 749 int setup_arg_pages(struct linux_binprm *bprm,
717 unsigned long stack_top, 750 unsigned long stack_top,
718 int executable_stack) 751 int executable_stack)
719 { 752 {
720 unsigned long ret; 753 unsigned long ret;
721 unsigned long stack_shift; 754 unsigned long stack_shift;
722 struct mm_struct *mm = current->mm; 755 struct mm_struct *mm = current->mm;
723 struct vm_area_struct *vma = bprm->vma 756 struct vm_area_struct *vma = bprm->vma;
724 struct vm_area_struct *prev = NULL; 757 struct vm_area_struct *prev = NULL;
725 unsigned long vm_flags; 758 unsigned long vm_flags;
726 unsigned long stack_base; 759 unsigned long stack_base;
727 unsigned long stack_size; 760 unsigned long stack_size;
728 unsigned long stack_expand; 761 unsigned long stack_expand;
729 unsigned long rlim_stack; 762 unsigned long rlim_stack;
730 struct mmu_gather tlb; 763 struct mmu_gather tlb;
731 struct vma_iterator vmi; <<
732 764
733 #ifdef CONFIG_STACK_GROWSUP 765 #ifdef CONFIG_STACK_GROWSUP
734 /* Limit stack size */ 766 /* Limit stack size */
735 stack_base = bprm->rlim_stack.rlim_max 767 stack_base = bprm->rlim_stack.rlim_max;
736 768
737 stack_base = calc_max_stack_size(stack 769 stack_base = calc_max_stack_size(stack_base);
738 770
739 /* Add space for stack randomization. 771 /* Add space for stack randomization. */
740 if (current->flags & PF_RANDOMIZE) 772 if (current->flags & PF_RANDOMIZE)
741 stack_base += (STACK_RND_MASK 773 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
742 774
743 /* Make sure we didn't let the argumen 775 /* Make sure we didn't let the argument array grow too large. */
744 if (vma->vm_end - vma->vm_start > stac 776 if (vma->vm_end - vma->vm_start > stack_base)
745 return -ENOMEM; 777 return -ENOMEM;
746 778
747 stack_base = PAGE_ALIGN(stack_top - st 779 stack_base = PAGE_ALIGN(stack_top - stack_base);
748 780
749 stack_shift = vma->vm_start - stack_ba 781 stack_shift = vma->vm_start - stack_base;
750 mm->arg_start = bprm->p - stack_shift; 782 mm->arg_start = bprm->p - stack_shift;
751 bprm->p = vma->vm_end - stack_shift; 783 bprm->p = vma->vm_end - stack_shift;
752 #else 784 #else
753 stack_top = arch_align_stack(stack_top 785 stack_top = arch_align_stack(stack_top);
754 stack_top = PAGE_ALIGN(stack_top); 786 stack_top = PAGE_ALIGN(stack_top);
755 787
756 if (unlikely(stack_top < mmap_min_addr 788 if (unlikely(stack_top < mmap_min_addr) ||
757 unlikely(vma->vm_end - vma->vm_sta 789 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
758 return -ENOMEM; 790 return -ENOMEM;
759 791
760 stack_shift = vma->vm_end - stack_top; 792 stack_shift = vma->vm_end - stack_top;
761 793
762 bprm->p -= stack_shift; 794 bprm->p -= stack_shift;
763 mm->arg_start = bprm->p; 795 mm->arg_start = bprm->p;
764 #endif 796 #endif
765 797
766 if (bprm->loader) 798 if (bprm->loader)
767 bprm->loader -= stack_shift; 799 bprm->loader -= stack_shift;
768 bprm->exec -= stack_shift; 800 bprm->exec -= stack_shift;
769 801
770 if (mmap_write_lock_killable(mm)) 802 if (mmap_write_lock_killable(mm))
771 return -EINTR; 803 return -EINTR;
772 804
773 vm_flags = VM_STACK_FLAGS; 805 vm_flags = VM_STACK_FLAGS;
774 806
775 /* 807 /*
776 * Adjust stack execute permissions; e 808 * Adjust stack execute permissions; explicitly enable for
777 * EXSTACK_ENABLE_X, disable for EXSTA 809 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
778 * (arch default) otherwise. 810 * (arch default) otherwise.
779 */ 811 */
780 if (unlikely(executable_stack == EXSTA 812 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
781 vm_flags |= VM_EXEC; 813 vm_flags |= VM_EXEC;
782 else if (executable_stack == EXSTACK_D 814 else if (executable_stack == EXSTACK_DISABLE_X)
783 vm_flags &= ~VM_EXEC; 815 vm_flags &= ~VM_EXEC;
784 vm_flags |= mm->def_flags; 816 vm_flags |= mm->def_flags;
785 vm_flags |= VM_STACK_INCOMPLETE_SETUP; 817 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
786 818
787 vma_iter_init(&vmi, mm, vma->vm_start) <<
788 <<
789 tlb_gather_mmu(&tlb, mm); 819 tlb_gather_mmu(&tlb, mm);
790 ret = mprotect_fixup(&vmi, &tlb, vma, !! 820 ret = mprotect_fixup(&tlb, vma, &prev, vma->vm_start, vma->vm_end,
791 vm_flags); 821 vm_flags);
792 tlb_finish_mmu(&tlb); 822 tlb_finish_mmu(&tlb);
793 823
794 if (ret) 824 if (ret)
795 goto out_unlock; 825 goto out_unlock;
796 BUG_ON(prev != vma); 826 BUG_ON(prev != vma);
797 827
798 if (unlikely(vm_flags & VM_EXEC)) { 828 if (unlikely(vm_flags & VM_EXEC)) {
799 pr_warn_once("process '%pD4' s 829 pr_warn_once("process '%pD4' started with executable stack\n",
800 bprm->file); 830 bprm->file);
801 } 831 }
802 832
803 /* Move stack pages down in memory. */ 833 /* Move stack pages down in memory. */
804 if (stack_shift) { 834 if (stack_shift) {
805 /* !! 835 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) 836 if (ret)
812 goto out_unlock; 837 goto out_unlock;
813 } 838 }
814 839
815 /* mprotect_fixup is overkill to remov 840 /* mprotect_fixup is overkill to remove the temporary stack flags */
816 vm_flags_clear(vma, VM_STACK_INCOMPLET !! 841 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
817 842
818 stack_expand = 131072UL; /* randomly 3 843 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
819 stack_size = vma->vm_end - vma->vm_sta 844 stack_size = vma->vm_end - vma->vm_start;
820 /* 845 /*
821 * Align this down to a page boundary 846 * Align this down to a page boundary as expand_stack
822 * will align it up. 847 * will align it up.
823 */ 848 */
824 rlim_stack = bprm->rlim_stack.rlim_cur 849 rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
825 <<
826 stack_expand = min(rlim_stack, stack_s <<
827 <<
828 #ifdef CONFIG_STACK_GROWSUP 850 #ifdef CONFIG_STACK_GROWSUP
829 stack_base = vma->vm_start + stack_exp !! 851 if (stack_size + stack_expand > rlim_stack)
>> 852 stack_base = vma->vm_start + rlim_stack;
>> 853 else
>> 854 stack_base = vma->vm_end + stack_expand;
830 #else 855 #else
831 stack_base = vma->vm_end - stack_expan !! 856 if (stack_size + stack_expand > rlim_stack)
>> 857 stack_base = vma->vm_end - rlim_stack;
>> 858 else
>> 859 stack_base = vma->vm_start - stack_expand;
832 #endif 860 #endif
833 current->mm->start_stack = bprm->p; 861 current->mm->start_stack = bprm->p;
834 ret = expand_stack_locked(vma, stack_b 862 ret = expand_stack_locked(vma, stack_base);
835 if (ret) 863 if (ret)
836 ret = -EFAULT; 864 ret = -EFAULT;
837 865
838 out_unlock: 866 out_unlock:
839 mmap_write_unlock(mm); 867 mmap_write_unlock(mm);
840 return ret; 868 return ret;
841 } 869 }
842 EXPORT_SYMBOL(setup_arg_pages); 870 EXPORT_SYMBOL(setup_arg_pages);
843 871
844 #else 872 #else
845 873
846 /* 874 /*
847 * Transfer the program arguments and environm 875 * Transfer the program arguments and environment from the holding pages
848 * onto the stack. The provided stack pointer 876 * onto the stack. The provided stack pointer is adjusted accordingly.
849 */ 877 */
850 int transfer_args_to_stack(struct linux_binprm 878 int transfer_args_to_stack(struct linux_binprm *bprm,
851 unsigned long *sp_l 879 unsigned long *sp_location)
852 { 880 {
853 unsigned long index, stop, sp; 881 unsigned long index, stop, sp;
854 int ret = 0; 882 int ret = 0;
855 883
856 stop = bprm->p >> PAGE_SHIFT; 884 stop = bprm->p >> PAGE_SHIFT;
857 sp = *sp_location; 885 sp = *sp_location;
858 886
859 for (index = MAX_ARG_PAGES - 1; index 887 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
860 unsigned int offset = index == 888 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
861 char *src = kmap_local_page(bp 889 char *src = kmap_local_page(bprm->page[index]) + offset;
862 sp -= PAGE_SIZE - offset; 890 sp -= PAGE_SIZE - offset;
863 if (copy_to_user((void *) sp, 891 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
864 ret = -EFAULT; 892 ret = -EFAULT;
865 kunmap_local(src); 893 kunmap_local(src);
866 if (ret) 894 if (ret)
867 goto out; 895 goto out;
868 } 896 }
869 897
870 bprm->exec += *sp_location - MAX_ARG_P 898 bprm->exec += *sp_location - MAX_ARG_PAGES * PAGE_SIZE;
871 *sp_location = sp; 899 *sp_location = sp;
872 900
873 out: 901 out:
874 return ret; 902 return ret;
875 } 903 }
876 EXPORT_SYMBOL(transfer_args_to_stack); 904 EXPORT_SYMBOL(transfer_args_to_stack);
877 905
878 #endif /* CONFIG_MMU */ 906 #endif /* CONFIG_MMU */
879 907
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 908 static struct file *do_open_execat(int fd, struct filename *name, int flags)
885 { 909 {
886 struct file *file; 910 struct file *file;
>> 911 int err;
887 struct open_flags open_exec_flags = { 912 struct open_flags open_exec_flags = {
888 .open_flag = O_LARGEFILE | O_R 913 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
889 .acc_mode = MAY_EXEC, 914 .acc_mode = MAY_EXEC,
890 .intent = LOOKUP_OPEN, 915 .intent = LOOKUP_OPEN,
891 .lookup_flags = LOOKUP_FOLLOW, 916 .lookup_flags = LOOKUP_FOLLOW,
892 }; 917 };
893 918
894 if ((flags & ~(AT_SYMLINK_NOFOLLOW | A 919 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
895 return ERR_PTR(-EINVAL); 920 return ERR_PTR(-EINVAL);
896 if (flags & AT_SYMLINK_NOFOLLOW) 921 if (flags & AT_SYMLINK_NOFOLLOW)
897 open_exec_flags.lookup_flags & 922 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
898 if (flags & AT_EMPTY_PATH) 923 if (flags & AT_EMPTY_PATH)
899 open_exec_flags.lookup_flags | 924 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
900 925
901 file = do_filp_open(fd, name, &open_ex 926 file = do_filp_open(fd, name, &open_exec_flags);
902 if (IS_ERR(file)) 927 if (IS_ERR(file))
903 return file; 928 return file;
904 929
905 /* 930 /*
906 * In the past the regular type check 931 * In the past the regular type check was here. It moved to may_open() in
907 * 633fb6ac3980 ("exec: move S_ISREG() 932 * 633fb6ac3980 ("exec: move S_ISREG() check earlier"). Since then it is
908 * an invariant that all non-regular f 933 * an invariant that all non-regular files error out before we get here.
909 */ 934 */
>> 935 err = -EACCES;
910 if (WARN_ON_ONCE(!S_ISREG(file_inode(f 936 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode)) ||
911 path_noexec(&file->f_path)) { !! 937 path_noexec(&file->f_path))
912 fput(file); !! 938 goto exit;
913 return ERR_PTR(-EACCES); !! 939
914 } !! 940 err = deny_write_access(file);
>> 941 if (err)
>> 942 goto exit;
>> 943
>> 944 if (name->name[0] != '\0')
>> 945 fsnotify_open(file);
915 946
916 return file; 947 return file;
>> 948
>> 949 exit:
>> 950 fput(file);
>> 951 return ERR_PTR(err);
917 } 952 }
918 953
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) 954 struct file *open_exec(const char *name)
930 { 955 {
931 struct filename *filename = getname_ke 956 struct filename *filename = getname_kernel(name);
932 struct file *f = ERR_CAST(filename); 957 struct file *f = ERR_CAST(filename);
933 958
934 if (!IS_ERR(filename)) { 959 if (!IS_ERR(filename)) {
935 f = do_open_execat(AT_FDCWD, f 960 f = do_open_execat(AT_FDCWD, filename, 0);
936 putname(filename); 961 putname(filename);
937 } 962 }
938 return f; 963 return f;
939 } 964 }
940 EXPORT_SYMBOL(open_exec); 965 EXPORT_SYMBOL(open_exec);
941 966
942 #if defined(CONFIG_BINFMT_FLAT) || defined(CON 967 #if defined(CONFIG_BINFMT_FLAT) || defined(CONFIG_BINFMT_ELF_FDPIC)
943 ssize_t read_code(struct file *file, unsigned 968 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
944 { 969 {
945 ssize_t res = vfs_read(file, (void __u 970 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
946 if (res > 0) 971 if (res > 0)
947 flush_icache_user_range(addr, 972 flush_icache_user_range(addr, addr + len);
948 return res; 973 return res;
949 } 974 }
950 EXPORT_SYMBOL(read_code); 975 EXPORT_SYMBOL(read_code);
951 #endif 976 #endif
952 977
953 /* 978 /*
954 * Maps the mm_struct mm into the current task 979 * Maps the mm_struct mm into the current task struct.
955 * On success, this function returns with exec 980 * On success, this function returns with exec_update_lock
956 * held for writing. 981 * held for writing.
957 */ 982 */
958 static int exec_mmap(struct mm_struct *mm) 983 static int exec_mmap(struct mm_struct *mm)
959 { 984 {
960 struct task_struct *tsk; 985 struct task_struct *tsk;
961 struct mm_struct *old_mm, *active_mm; 986 struct mm_struct *old_mm, *active_mm;
962 int ret; 987 int ret;
963 988
964 /* Notify parent that we're no longer 989 /* Notify parent that we're no longer interested in the old VM */
965 tsk = current; 990 tsk = current;
966 old_mm = current->mm; 991 old_mm = current->mm;
967 exec_mm_release(tsk, old_mm); 992 exec_mm_release(tsk, old_mm);
>> 993 if (old_mm)
>> 994 sync_mm_rss(old_mm);
968 995
969 ret = down_write_killable(&tsk->signal 996 ret = down_write_killable(&tsk->signal->exec_update_lock);
970 if (ret) 997 if (ret)
971 return ret; 998 return ret;
972 999
973 if (old_mm) { 1000 if (old_mm) {
974 /* 1001 /*
975 * If there is a pending fatal 1002 * If there is a pending fatal signal perhaps a signal
976 * whose default action is to 1003 * whose default action is to create a coredump get
977 * out and die instead of goin 1004 * out and die instead of going through with the exec.
978 */ 1005 */
979 ret = mmap_read_lock_killable( 1006 ret = mmap_read_lock_killable(old_mm);
980 if (ret) { 1007 if (ret) {
981 up_write(&tsk->signal- 1008 up_write(&tsk->signal->exec_update_lock);
982 return ret; 1009 return ret;
983 } 1010 }
984 } 1011 }
985 1012
986 task_lock(tsk); 1013 task_lock(tsk);
987 membarrier_exec_mmap(mm); 1014 membarrier_exec_mmap(mm);
988 1015
989 local_irq_disable(); 1016 local_irq_disable();
990 active_mm = tsk->active_mm; 1017 active_mm = tsk->active_mm;
991 tsk->active_mm = mm; 1018 tsk->active_mm = mm;
992 tsk->mm = mm; 1019 tsk->mm = mm;
993 mm_init_cid(mm); <<
994 /* 1020 /*
995 * This prevents preemption while acti 1021 * This prevents preemption while active_mm is being loaded and
996 * it and mm are being updated, which 1022 * it and mm are being updated, which could cause problems for
997 * lazy tlb mm refcounting when these 1023 * lazy tlb mm refcounting when these are updated by context
998 * switches. Not all architectures can 1024 * switches. Not all architectures can handle irqs off over
999 * activate_mm yet. 1025 * activate_mm yet.
1000 */ 1026 */
1001 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS 1027 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1002 local_irq_enable(); 1028 local_irq_enable();
1003 activate_mm(active_mm, mm); 1029 activate_mm(active_mm, mm);
1004 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_ 1030 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1005 local_irq_enable(); 1031 local_irq_enable();
1006 lru_gen_add_mm(mm); 1032 lru_gen_add_mm(mm);
1007 task_unlock(tsk); 1033 task_unlock(tsk);
1008 lru_gen_use_mm(mm); 1034 lru_gen_use_mm(mm);
1009 if (old_mm) { 1035 if (old_mm) {
1010 mmap_read_unlock(old_mm); 1036 mmap_read_unlock(old_mm);
1011 BUG_ON(active_mm != old_mm); 1037 BUG_ON(active_mm != old_mm);
1012 setmax_mm_hiwater_rss(&tsk->s 1038 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1013 mm_update_next_owner(old_mm); 1039 mm_update_next_owner(old_mm);
1014 mmput(old_mm); 1040 mmput(old_mm);
1015 return 0; 1041 return 0;
1016 } 1042 }
1017 mmdrop_lazy_tlb(active_mm); !! 1043 mmdrop(active_mm);
1018 return 0; 1044 return 0;
1019 } 1045 }
1020 1046
1021 static int de_thread(struct task_struct *tsk) 1047 static int de_thread(struct task_struct *tsk)
1022 { 1048 {
1023 struct signal_struct *sig = tsk->sign 1049 struct signal_struct *sig = tsk->signal;
1024 struct sighand_struct *oldsighand = t 1050 struct sighand_struct *oldsighand = tsk->sighand;
1025 spinlock_t *lock = &oldsighand->siglo 1051 spinlock_t *lock = &oldsighand->siglock;
1026 1052
1027 if (thread_group_empty(tsk)) 1053 if (thread_group_empty(tsk))
1028 goto no_thread_group; 1054 goto no_thread_group;
1029 1055
1030 /* 1056 /*
1031 * Kill all other threads in the thre 1057 * Kill all other threads in the thread group.
1032 */ 1058 */
1033 spin_lock_irq(lock); 1059 spin_lock_irq(lock);
1034 if ((sig->flags & SIGNAL_GROUP_EXIT) 1060 if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) {
1035 /* 1061 /*
1036 * Another group action in pr 1062 * Another group action in progress, just
1037 * return so that the signal 1063 * return so that the signal is processed.
1038 */ 1064 */
1039 spin_unlock_irq(lock); 1065 spin_unlock_irq(lock);
1040 return -EAGAIN; 1066 return -EAGAIN;
1041 } 1067 }
1042 1068
1043 sig->group_exec_task = tsk; 1069 sig->group_exec_task = tsk;
1044 sig->notify_count = zap_other_threads 1070 sig->notify_count = zap_other_threads(tsk);
1045 if (!thread_group_leader(tsk)) 1071 if (!thread_group_leader(tsk))
1046 sig->notify_count--; 1072 sig->notify_count--;
1047 1073
1048 while (sig->notify_count) { 1074 while (sig->notify_count) {
1049 __set_current_state(TASK_KILL 1075 __set_current_state(TASK_KILLABLE);
1050 spin_unlock_irq(lock); 1076 spin_unlock_irq(lock);
1051 schedule(); 1077 schedule();
1052 if (__fatal_signal_pending(ts 1078 if (__fatal_signal_pending(tsk))
1053 goto killed; 1079 goto killed;
1054 spin_lock_irq(lock); 1080 spin_lock_irq(lock);
1055 } 1081 }
1056 spin_unlock_irq(lock); 1082 spin_unlock_irq(lock);
1057 1083
1058 /* 1084 /*
1059 * At this point all other threads ha 1085 * At this point all other threads have exited, all we have to
1060 * do is to wait for the thread group 1086 * do is to wait for the thread group leader to become inactive,
1061 * and to assume its PID: 1087 * and to assume its PID:
1062 */ 1088 */
1063 if (!thread_group_leader(tsk)) { 1089 if (!thread_group_leader(tsk)) {
1064 struct task_struct *leader = 1090 struct task_struct *leader = tsk->group_leader;
1065 1091
1066 for (;;) { 1092 for (;;) {
1067 cgroup_threadgroup_ch 1093 cgroup_threadgroup_change_begin(tsk);
1068 write_lock_irq(&taskl 1094 write_lock_irq(&tasklist_lock);
1069 /* 1095 /*
1070 * Do this under task 1096 * Do this under tasklist_lock to ensure that
1071 * exit_notify() can' 1097 * exit_notify() can't miss ->group_exec_task
1072 */ 1098 */
1073 sig->notify_count = - 1099 sig->notify_count = -1;
1074 if (likely(leader->ex 1100 if (likely(leader->exit_state))
1075 break; 1101 break;
1076 __set_current_state(T 1102 __set_current_state(TASK_KILLABLE);
1077 write_unlock_irq(&tas 1103 write_unlock_irq(&tasklist_lock);
1078 cgroup_threadgroup_ch 1104 cgroup_threadgroup_change_end(tsk);
1079 schedule(); 1105 schedule();
1080 if (__fatal_signal_pe 1106 if (__fatal_signal_pending(tsk))
1081 goto killed; 1107 goto killed;
1082 } 1108 }
1083 1109
1084 /* 1110 /*
1085 * The only record we have of 1111 * The only record we have of the real-time age of a
1086 * process, regardless of exe 1112 * process, regardless of execs it's done, is start_time.
1087 * All the past CPU time is a 1113 * All the past CPU time is accumulated in signal_struct
1088 * from sister threads now de 1114 * from sister threads now dead. But in this non-leader
1089 * exec, nothing survives fro 1115 * exec, nothing survives from the original leader thread,
1090 * whose birth marks the true 1116 * whose birth marks the true age of this process now.
1091 * When we take on its identi 1117 * When we take on its identity by switching to its PID, we
1092 * also take its birthdate (a 1118 * also take its birthdate (always earlier than our own).
1093 */ 1119 */
1094 tsk->start_time = leader->sta 1120 tsk->start_time = leader->start_time;
1095 tsk->start_boottime = leader- 1121 tsk->start_boottime = leader->start_boottime;
1096 1122
1097 BUG_ON(!same_thread_group(lea 1123 BUG_ON(!same_thread_group(leader, tsk));
1098 /* 1124 /*
1099 * An exec() starts a new thr 1125 * An exec() starts a new thread group with the
1100 * TGID of the previous threa 1126 * TGID of the previous thread group. Rehash the
1101 * two threads with a switche 1127 * two threads with a switched PID, and release
1102 * the former thread group le 1128 * the former thread group leader:
1103 */ 1129 */
1104 1130
1105 /* Become a process group lea 1131 /* Become a process group leader with the old leader's pid.
1106 * The old leader becomes a t 1132 * The old leader becomes a thread of the this thread group.
1107 */ 1133 */
1108 exchange_tids(tsk, leader); 1134 exchange_tids(tsk, leader);
1109 transfer_pid(leader, tsk, PID 1135 transfer_pid(leader, tsk, PIDTYPE_TGID);
1110 transfer_pid(leader, tsk, PID 1136 transfer_pid(leader, tsk, PIDTYPE_PGID);
1111 transfer_pid(leader, tsk, PID 1137 transfer_pid(leader, tsk, PIDTYPE_SID);
1112 1138
1113 list_replace_rcu(&leader->tas 1139 list_replace_rcu(&leader->tasks, &tsk->tasks);
1114 list_replace_init(&leader->si 1140 list_replace_init(&leader->sibling, &tsk->sibling);
1115 1141
1116 tsk->group_leader = tsk; 1142 tsk->group_leader = tsk;
1117 leader->group_leader = tsk; 1143 leader->group_leader = tsk;
1118 1144
1119 tsk->exit_signal = SIGCHLD; 1145 tsk->exit_signal = SIGCHLD;
1120 leader->exit_signal = -1; 1146 leader->exit_signal = -1;
1121 1147
1122 BUG_ON(leader->exit_state != 1148 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1123 leader->exit_state = EXIT_DEA 1149 leader->exit_state = EXIT_DEAD;
>> 1150
1124 /* 1151 /*
1125 * We are going to release_ta 1152 * We are going to release_task()->ptrace_unlink() silently,
1126 * the tracer can sleep in do 1153 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1127 * the tracer won't block aga 1154 * the tracer won't block again waiting for this thread.
1128 */ 1155 */
1129 if (unlikely(leader->ptrace)) 1156 if (unlikely(leader->ptrace))
1130 __wake_up_parent(lead 1157 __wake_up_parent(leader, leader->parent);
1131 write_unlock_irq(&tasklist_lo 1158 write_unlock_irq(&tasklist_lock);
1132 cgroup_threadgroup_change_end 1159 cgroup_threadgroup_change_end(tsk);
1133 1160
1134 release_task(leader); 1161 release_task(leader);
1135 } 1162 }
1136 1163
1137 sig->group_exec_task = NULL; 1164 sig->group_exec_task = NULL;
1138 sig->notify_count = 0; 1165 sig->notify_count = 0;
1139 1166
1140 no_thread_group: 1167 no_thread_group:
1141 /* we have changed execution domain * 1168 /* we have changed execution domain */
1142 tsk->exit_signal = SIGCHLD; 1169 tsk->exit_signal = SIGCHLD;
1143 1170
1144 BUG_ON(!thread_group_leader(tsk)); 1171 BUG_ON(!thread_group_leader(tsk));
1145 return 0; 1172 return 0;
1146 1173
1147 killed: 1174 killed:
1148 /* protects against exit_notify() and 1175 /* protects against exit_notify() and __exit_signal() */
1149 read_lock(&tasklist_lock); 1176 read_lock(&tasklist_lock);
1150 sig->group_exec_task = NULL; 1177 sig->group_exec_task = NULL;
1151 sig->notify_count = 0; 1178 sig->notify_count = 0;
1152 read_unlock(&tasklist_lock); 1179 read_unlock(&tasklist_lock);
1153 return -EAGAIN; 1180 return -EAGAIN;
1154 } 1181 }
1155 1182
1156 1183
1157 /* 1184 /*
1158 * This function makes sure the current proce 1185 * This function makes sure the current process has its own signal table,
1159 * so that flush_signal_handlers can later re 1186 * so that flush_signal_handlers can later reset the handlers without
1160 * disturbing other processes. (Other proces 1187 * disturbing other processes. (Other processes might share the signal
1161 * table via the CLONE_SIGHAND option to clon 1188 * table via the CLONE_SIGHAND option to clone().)
1162 */ 1189 */
1163 static int unshare_sighand(struct task_struct 1190 static int unshare_sighand(struct task_struct *me)
1164 { 1191 {
1165 struct sighand_struct *oldsighand = m 1192 struct sighand_struct *oldsighand = me->sighand;
1166 1193
1167 if (refcount_read(&oldsighand->count) 1194 if (refcount_read(&oldsighand->count) != 1) {
1168 struct sighand_struct *newsig 1195 struct sighand_struct *newsighand;
1169 /* 1196 /*
1170 * This ->sighand is shared w 1197 * This ->sighand is shared with the CLONE_SIGHAND
1171 * but not CLONE_THREAD task, 1198 * but not CLONE_THREAD task, switch to the new one.
1172 */ 1199 */
1173 newsighand = kmem_cache_alloc 1200 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1174 if (!newsighand) 1201 if (!newsighand)
1175 return -ENOMEM; 1202 return -ENOMEM;
1176 1203
1177 refcount_set(&newsighand->cou 1204 refcount_set(&newsighand->count, 1);
1178 1205
1179 write_lock_irq(&tasklist_lock 1206 write_lock_irq(&tasklist_lock);
1180 spin_lock(&oldsighand->sigloc 1207 spin_lock(&oldsighand->siglock);
1181 memcpy(newsighand->action, ol 1208 memcpy(newsighand->action, oldsighand->action,
1182 sizeof(newsighand->act 1209 sizeof(newsighand->action));
1183 rcu_assign_pointer(me->sighan 1210 rcu_assign_pointer(me->sighand, newsighand);
1184 spin_unlock(&oldsighand->sigl 1211 spin_unlock(&oldsighand->siglock);
1185 write_unlock_irq(&tasklist_lo 1212 write_unlock_irq(&tasklist_lock);
1186 1213
1187 __cleanup_sighand(oldsighand) 1214 __cleanup_sighand(oldsighand);
1188 } 1215 }
1189 return 0; 1216 return 0;
1190 } 1217 }
1191 1218
1192 char *__get_task_comm(char *buf, size_t buf_s 1219 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1193 { 1220 {
1194 task_lock(tsk); 1221 task_lock(tsk);
1195 /* Always NUL terminated and zero-pad 1222 /* Always NUL terminated and zero-padded */
1196 strscpy_pad(buf, tsk->comm, buf_size) 1223 strscpy_pad(buf, tsk->comm, buf_size);
1197 task_unlock(tsk); 1224 task_unlock(tsk);
1198 return buf; 1225 return buf;
1199 } 1226 }
1200 EXPORT_SYMBOL_GPL(__get_task_comm); 1227 EXPORT_SYMBOL_GPL(__get_task_comm);
1201 1228
1202 /* 1229 /*
1203 * These functions flushes out all traces of 1230 * These functions flushes out all traces of the currently running executable
1204 * so that a new one can be started 1231 * so that a new one can be started
1205 */ 1232 */
1206 1233
1207 void __set_task_comm(struct task_struct *tsk, 1234 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1208 { 1235 {
1209 task_lock(tsk); 1236 task_lock(tsk);
1210 trace_task_rename(tsk, buf); 1237 trace_task_rename(tsk, buf);
1211 strscpy_pad(tsk->comm, buf, sizeof(ts 1238 strscpy_pad(tsk->comm, buf, sizeof(tsk->comm));
1212 task_unlock(tsk); 1239 task_unlock(tsk);
1213 perf_event_comm(tsk, exec); 1240 perf_event_comm(tsk, exec);
1214 } 1241 }
1215 1242
1216 /* 1243 /*
1217 * Calling this is the point of no return. No 1244 * Calling this is the point of no return. None of the failures will be
1218 * seen by userspace since either the process 1245 * seen by userspace since either the process is already taking a fatal
1219 * signal (via de_thread() or coredump), or w 1246 * signal (via de_thread() or coredump), or will have SEGV raised
1220 * (after exec_mmap()) by search_binary_handl 1247 * (after exec_mmap()) by search_binary_handler (see below).
1221 */ 1248 */
1222 int begin_new_exec(struct linux_binprm * bprm 1249 int begin_new_exec(struct linux_binprm * bprm)
1223 { 1250 {
1224 struct task_struct *me = current; 1251 struct task_struct *me = current;
1225 int retval; 1252 int retval;
1226 1253
1227 /* Once we are committed compute the 1254 /* Once we are committed compute the creds */
1228 retval = bprm_creds_from_file(bprm); 1255 retval = bprm_creds_from_file(bprm);
1229 if (retval) 1256 if (retval)
1230 return retval; 1257 return retval;
1231 1258
1232 /* 1259 /*
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 1260 * Ensure all future errors are fatal.
1242 */ 1261 */
1243 bprm->point_of_no_return = true; 1262 bprm->point_of_no_return = true;
1244 1263
1245 /* 1264 /*
1246 * Make this the only thread in the t 1265 * Make this the only thread in the thread group.
1247 */ 1266 */
1248 retval = de_thread(me); 1267 retval = de_thread(me);
1249 if (retval) 1268 if (retval)
1250 goto out; 1269 goto out;
1251 1270
1252 /* 1271 /*
1253 * Cancel any io_uring activity acros 1272 * Cancel any io_uring activity across execve
1254 */ 1273 */
1255 io_uring_task_cancel(); 1274 io_uring_task_cancel();
1256 1275
1257 /* Ensure the files table is not shar 1276 /* Ensure the files table is not shared. */
1258 retval = unshare_files(); 1277 retval = unshare_files();
1259 if (retval) 1278 if (retval)
1260 goto out; 1279 goto out;
1261 1280
1262 /* 1281 /*
1263 * Must be called _before_ exec_mmap( 1282 * Must be called _before_ exec_mmap() as bprm->mm is
1264 * not visible until then. Doing it h !! 1283 * not visible until then. This also enables the update
1265 * we don't race against replace_mm_e !! 1284 * to be lockless.
1266 */ 1285 */
1267 retval = set_mm_exe_file(bprm->mm, bp 1286 retval = set_mm_exe_file(bprm->mm, bprm->file);
1268 if (retval) 1287 if (retval)
1269 goto out; 1288 goto out;
1270 1289
1271 /* If the binary is not readable then 1290 /* If the binary is not readable then enforce mm->dumpable=0 */
1272 would_dump(bprm, bprm->file); 1291 would_dump(bprm, bprm->file);
1273 if (bprm->have_execfd) 1292 if (bprm->have_execfd)
1274 would_dump(bprm, bprm->execut 1293 would_dump(bprm, bprm->executable);
1275 1294
1276 /* 1295 /*
1277 * Release all of the old mmap stuff 1296 * Release all of the old mmap stuff
1278 */ 1297 */
1279 acct_arg_size(bprm, 0); 1298 acct_arg_size(bprm, 0);
1280 retval = exec_mmap(bprm->mm); 1299 retval = exec_mmap(bprm->mm);
1281 if (retval) 1300 if (retval)
1282 goto out; 1301 goto out;
1283 1302
1284 bprm->mm = NULL; 1303 bprm->mm = NULL;
1285 1304
1286 retval = exec_task_namespaces(); <<
1287 if (retval) <<
1288 goto out_unlock; <<
1289 <<
1290 #ifdef CONFIG_POSIX_TIMERS 1305 #ifdef CONFIG_POSIX_TIMERS
1291 spin_lock_irq(&me->sighand->siglock); 1306 spin_lock_irq(&me->sighand->siglock);
1292 posix_cpu_timers_exit(me); 1307 posix_cpu_timers_exit(me);
1293 spin_unlock_irq(&me->sighand->siglock 1308 spin_unlock_irq(&me->sighand->siglock);
1294 exit_itimers(me); 1309 exit_itimers(me);
1295 flush_itimer_signals(); 1310 flush_itimer_signals();
1296 #endif 1311 #endif
1297 1312
1298 /* 1313 /*
1299 * Make the signal table private. 1314 * Make the signal table private.
1300 */ 1315 */
1301 retval = unshare_sighand(me); 1316 retval = unshare_sighand(me);
1302 if (retval) 1317 if (retval)
1303 goto out_unlock; 1318 goto out_unlock;
1304 1319
1305 me->flags &= ~(PF_RANDOMIZE | PF_FORK 1320 me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC |
1306 PF_NO 1321 PF_NOFREEZE | PF_NO_SETAFFINITY);
1307 flush_thread(); 1322 flush_thread();
1308 me->personality &= ~bprm->per_clear; 1323 me->personality &= ~bprm->per_clear;
1309 1324
1310 clear_syscall_work_syscall_user_dispa 1325 clear_syscall_work_syscall_user_dispatch(me);
1311 1326
1312 /* 1327 /*
1313 * We have to apply CLOEXEC before we 1328 * We have to apply CLOEXEC before we change whether the process is
1314 * dumpable (in setup_new_exec) to av 1329 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1315 * trying to access the should-be-clo 1330 * trying to access the should-be-closed file descriptors of a process
1316 * undergoing exec(2). 1331 * undergoing exec(2).
1317 */ 1332 */
1318 do_close_on_exec(me->files); 1333 do_close_on_exec(me->files);
1319 1334
1320 if (bprm->secureexec) { 1335 if (bprm->secureexec) {
1321 /* Make sure parent cannot si 1336 /* Make sure parent cannot signal privileged process. */
1322 me->pdeath_signal = 0; 1337 me->pdeath_signal = 0;
1323 1338
1324 /* 1339 /*
1325 * For secureexec, reset the 1340 * For secureexec, reset the stack limit to sane default to
1326 * avoid bad behavior from th 1341 * avoid bad behavior from the prior rlimits. This has to
1327 * happen before arch_pick_mm 1342 * happen before arch_pick_mmap_layout(), which examines
1328 * RLIMIT_STACK, but after th 1343 * RLIMIT_STACK, but after the point of no return to avoid
1329 * needing to clean up the ch 1344 * needing to clean up the change on failure.
1330 */ 1345 */
1331 if (bprm->rlim_stack.rlim_cur 1346 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1332 bprm->rlim_stack.rlim 1347 bprm->rlim_stack.rlim_cur = _STK_LIM;
1333 } 1348 }
1334 1349
1335 me->sas_ss_sp = me->sas_ss_size = 0; 1350 me->sas_ss_sp = me->sas_ss_size = 0;
1336 1351
1337 /* 1352 /*
1338 * Figure out dumpability. Note that 1353 * Figure out dumpability. Note that this checking only of current
1339 * is wrong, but userspace depends on 1354 * is wrong, but userspace depends on it. This should be testing
1340 * bprm->secureexec instead. 1355 * bprm->secureexec instead.
1341 */ 1356 */
1342 if (bprm->interp_flags & BINPRM_FLAGS 1357 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1343 !(uid_eq(current_euid(), current_ 1358 !(uid_eq(current_euid(), current_uid()) &&
1344 gid_eq(current_egid(), current_ 1359 gid_eq(current_egid(), current_gid())))
1345 set_dumpable(current->mm, sui 1360 set_dumpable(current->mm, suid_dumpable);
1346 else 1361 else
1347 set_dumpable(current->mm, SUI 1362 set_dumpable(current->mm, SUID_DUMP_USER);
1348 1363
1349 perf_event_exec(); 1364 perf_event_exec();
1350 __set_task_comm(me, kbasename(bprm->f 1365 __set_task_comm(me, kbasename(bprm->filename), true);
1351 1366
1352 /* An exec changes our domain. We are 1367 /* An exec changes our domain. We are no longer part of the thread
1353 group */ 1368 group */
1354 WRITE_ONCE(me->self_exec_id, me->self 1369 WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1355 flush_signal_handlers(me, 0); 1370 flush_signal_handlers(me, 0);
1356 1371
1357 retval = set_cred_ucounts(bprm->cred) 1372 retval = set_cred_ucounts(bprm->cred);
1358 if (retval < 0) 1373 if (retval < 0)
1359 goto out_unlock; 1374 goto out_unlock;
1360 1375
1361 /* 1376 /*
1362 * install the new credentials for th 1377 * install the new credentials for this executable
1363 */ 1378 */
1364 security_bprm_committing_creds(bprm); 1379 security_bprm_committing_creds(bprm);
1365 1380
1366 commit_creds(bprm->cred); 1381 commit_creds(bprm->cred);
1367 bprm->cred = NULL; 1382 bprm->cred = NULL;
1368 1383
1369 /* 1384 /*
1370 * Disable monitoring for regular use 1385 * Disable monitoring for regular users
1371 * when executing setuid binaries. Mu 1386 * when executing setuid binaries. Must
1372 * wait until new credentials are com 1387 * wait until new credentials are committed
1373 * by commit_creds() above 1388 * by commit_creds() above
1374 */ 1389 */
1375 if (get_dumpable(me->mm) != SUID_DUMP 1390 if (get_dumpable(me->mm) != SUID_DUMP_USER)
1376 perf_event_exit_task(me); 1391 perf_event_exit_task(me);
1377 /* 1392 /*
1378 * cred_guard_mutex must be held at l 1393 * cred_guard_mutex must be held at least to this point to prevent
1379 * ptrace_attach() from altering our 1394 * ptrace_attach() from altering our determination of the task's
1380 * credentials; any time after this i 1395 * credentials; any time after this it may be unlocked.
1381 */ 1396 */
1382 security_bprm_committed_creds(bprm); 1397 security_bprm_committed_creds(bprm);
1383 1398
1384 /* Pass the opened binary to the inte 1399 /* Pass the opened binary to the interpreter. */
1385 if (bprm->have_execfd) { 1400 if (bprm->have_execfd) {
1386 retval = get_unused_fd_flags( 1401 retval = get_unused_fd_flags(0);
1387 if (retval < 0) 1402 if (retval < 0)
1388 goto out_unlock; 1403 goto out_unlock;
1389 fd_install(retval, bprm->exec 1404 fd_install(retval, bprm->executable);
1390 bprm->executable = NULL; 1405 bprm->executable = NULL;
1391 bprm->execfd = retval; 1406 bprm->execfd = retval;
1392 } 1407 }
1393 return 0; 1408 return 0;
1394 1409
1395 out_unlock: 1410 out_unlock:
1396 up_write(&me->signal->exec_update_loc 1411 up_write(&me->signal->exec_update_lock);
1397 if (!bprm->cred) 1412 if (!bprm->cred)
1398 mutex_unlock(&me->signal->cre 1413 mutex_unlock(&me->signal->cred_guard_mutex);
1399 1414
1400 out: 1415 out:
1401 return retval; 1416 return retval;
1402 } 1417 }
1403 EXPORT_SYMBOL(begin_new_exec); 1418 EXPORT_SYMBOL(begin_new_exec);
1404 1419
1405 void would_dump(struct linux_binprm *bprm, st 1420 void would_dump(struct linux_binprm *bprm, struct file *file)
1406 { 1421 {
1407 struct inode *inode = file_inode(file 1422 struct inode *inode = file_inode(file);
1408 struct mnt_idmap *idmap = file_mnt_id !! 1423 struct user_namespace *mnt_userns = file_mnt_user_ns(file);
1409 if (inode_permission(idmap, inode, MA !! 1424 if (inode_permission(mnt_userns, inode, MAY_READ) < 0) {
1410 struct user_namespace *old, * 1425 struct user_namespace *old, *user_ns;
1411 bprm->interp_flags |= BINPRM_ 1426 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1412 1427
1413 /* Ensure mm->user_ns contain 1428 /* Ensure mm->user_ns contains the executable */
1414 user_ns = old = bprm->mm->use 1429 user_ns = old = bprm->mm->user_ns;
1415 while ((user_ns != &init_user 1430 while ((user_ns != &init_user_ns) &&
1416 !privileged_wrt_inode_ !! 1431 !privileged_wrt_inode_uidgid(user_ns, mnt_userns, inode))
1417 user_ns = user_ns->pa 1432 user_ns = user_ns->parent;
1418 1433
1419 if (old != user_ns) { 1434 if (old != user_ns) {
1420 bprm->mm->user_ns = g 1435 bprm->mm->user_ns = get_user_ns(user_ns);
1421 put_user_ns(old); 1436 put_user_ns(old);
1422 } 1437 }
1423 } 1438 }
1424 } 1439 }
1425 EXPORT_SYMBOL(would_dump); 1440 EXPORT_SYMBOL(would_dump);
1426 1441
1427 void setup_new_exec(struct linux_binprm * bpr 1442 void setup_new_exec(struct linux_binprm * bprm)
1428 { 1443 {
1429 /* Setup things that can depend upon 1444 /* Setup things that can depend upon the personality */
1430 struct task_struct *me = current; 1445 struct task_struct *me = current;
1431 1446
1432 arch_pick_mmap_layout(me->mm, &bprm-> 1447 arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1433 1448
1434 arch_setup_new_exec(); 1449 arch_setup_new_exec();
1435 1450
1436 /* Set the new mm task size. We have 1451 /* Set the new mm task size. We have to do that late because it may
1437 * depend on TIF_32BIT which is only 1452 * depend on TIF_32BIT which is only updated in flush_thread() on
1438 * some architectures like powerpc 1453 * some architectures like powerpc
1439 */ 1454 */
1440 me->mm->task_size = TASK_SIZE; 1455 me->mm->task_size = TASK_SIZE;
1441 up_write(&me->signal->exec_update_loc 1456 up_write(&me->signal->exec_update_lock);
1442 mutex_unlock(&me->signal->cred_guard_ 1457 mutex_unlock(&me->signal->cred_guard_mutex);
1443 } 1458 }
1444 EXPORT_SYMBOL(setup_new_exec); 1459 EXPORT_SYMBOL(setup_new_exec);
1445 1460
1446 /* Runs immediately before start_thread() tak 1461 /* Runs immediately before start_thread() takes over. */
1447 void finalize_exec(struct linux_binprm *bprm) 1462 void finalize_exec(struct linux_binprm *bprm)
1448 { 1463 {
1449 /* Store any stack rlimit changes bef 1464 /* Store any stack rlimit changes before starting thread. */
1450 task_lock(current->group_leader); 1465 task_lock(current->group_leader);
1451 current->signal->rlim[RLIMIT_STACK] = 1466 current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1452 task_unlock(current->group_leader); 1467 task_unlock(current->group_leader);
1453 } 1468 }
1454 EXPORT_SYMBOL(finalize_exec); 1469 EXPORT_SYMBOL(finalize_exec);
1455 1470
1456 /* 1471 /*
1457 * Prepare credentials and lock ->cred_guard_ 1472 * Prepare credentials and lock ->cred_guard_mutex.
1458 * setup_new_exec() commits the new creds and 1473 * setup_new_exec() commits the new creds and drops the lock.
1459 * Or, if exec fails before, free_bprm() shou 1474 * Or, if exec fails before, free_bprm() should release ->cred
1460 * and unlock. 1475 * and unlock.
1461 */ 1476 */
1462 static int prepare_bprm_creds(struct linux_bi 1477 static int prepare_bprm_creds(struct linux_binprm *bprm)
1463 { 1478 {
1464 if (mutex_lock_interruptible(¤t 1479 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1465 return -ERESTARTNOINTR; 1480 return -ERESTARTNOINTR;
1466 1481
1467 bprm->cred = prepare_exec_creds(); 1482 bprm->cred = prepare_exec_creds();
1468 if (likely(bprm->cred)) 1483 if (likely(bprm->cred))
1469 return 0; 1484 return 0;
1470 1485
1471 mutex_unlock(¤t->signal->cred_g 1486 mutex_unlock(¤t->signal->cred_guard_mutex);
1472 return -ENOMEM; 1487 return -ENOMEM;
1473 } 1488 }
1474 1489
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 1490 static void free_bprm(struct linux_binprm *bprm)
1483 { 1491 {
1484 if (bprm->mm) { 1492 if (bprm->mm) {
1485 acct_arg_size(bprm, 0); 1493 acct_arg_size(bprm, 0);
1486 mmput(bprm->mm); 1494 mmput(bprm->mm);
1487 } 1495 }
1488 free_arg_pages(bprm); 1496 free_arg_pages(bprm);
1489 if (bprm->cred) { 1497 if (bprm->cred) {
1490 mutex_unlock(¤t->signal 1498 mutex_unlock(¤t->signal->cred_guard_mutex);
1491 abort_creds(bprm->cred); 1499 abort_creds(bprm->cred);
1492 } 1500 }
1493 do_close_execat(bprm->file); !! 1501 if (bprm->file) {
>> 1502 allow_write_access(bprm->file);
>> 1503 fput(bprm->file);
>> 1504 }
1494 if (bprm->executable) 1505 if (bprm->executable)
1495 fput(bprm->executable); 1506 fput(bprm->executable);
1496 /* If a binfmt changed the interp, fr 1507 /* If a binfmt changed the interp, free it. */
1497 if (bprm->interp != bprm->filename) 1508 if (bprm->interp != bprm->filename)
1498 kfree(bprm->interp); 1509 kfree(bprm->interp);
1499 kfree(bprm->fdpath); 1510 kfree(bprm->fdpath);
1500 kfree(bprm); 1511 kfree(bprm);
1501 } 1512 }
1502 1513
1503 static struct linux_binprm *alloc_bprm(int fd !! 1514 static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1504 { 1515 {
1505 struct linux_binprm *bprm; !! 1516 struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1506 struct file *file; <<
1507 int retval = -ENOMEM; 1517 int retval = -ENOMEM;
1508 !! 1518 if (!bprm)
1509 file = do_open_execat(fd, filename, f !! 1519 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 1520
1521 if (fd == AT_FDCWD || filename->name[ 1521 if (fd == AT_FDCWD || filename->name[0] == '/') {
1522 bprm->filename = filename->na 1522 bprm->filename = filename->name;
1523 } else { 1523 } else {
1524 if (filename->name[0] == '\0' 1524 if (filename->name[0] == '\0')
1525 bprm->fdpath = kaspri 1525 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1526 else 1526 else
1527 bprm->fdpath = kaspri 1527 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1528 1528 fd, filename->name);
1529 if (!bprm->fdpath) 1529 if (!bprm->fdpath)
1530 goto out_free; 1530 goto out_free;
1531 1531
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 1532 bprm->filename = bprm->fdpath;
1545 } 1533 }
1546 bprm->interp = bprm->filename; 1534 bprm->interp = bprm->filename;
1547 1535
1548 retval = bprm_mm_init(bprm); 1536 retval = bprm_mm_init(bprm);
1549 if (!retval) !! 1537 if (retval)
1550 return bprm; !! 1538 goto out_free;
>> 1539 return bprm;
1551 1540
1552 out_free: 1541 out_free:
1553 free_bprm(bprm); 1542 free_bprm(bprm);
>> 1543 out:
1554 return ERR_PTR(retval); 1544 return ERR_PTR(retval);
1555 } 1545 }
1556 1546
1557 int bprm_change_interp(const char *interp, st 1547 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1558 { 1548 {
1559 /* If a binfmt changed the interp, fr 1549 /* If a binfmt changed the interp, free it first. */
1560 if (bprm->interp != bprm->filename) 1550 if (bprm->interp != bprm->filename)
1561 kfree(bprm->interp); 1551 kfree(bprm->interp);
1562 bprm->interp = kstrdup(interp, GFP_KE 1552 bprm->interp = kstrdup(interp, GFP_KERNEL);
1563 if (!bprm->interp) 1553 if (!bprm->interp)
1564 return -ENOMEM; 1554 return -ENOMEM;
1565 return 0; 1555 return 0;
1566 } 1556 }
1567 EXPORT_SYMBOL(bprm_change_interp); 1557 EXPORT_SYMBOL(bprm_change_interp);
1568 1558
1569 /* 1559 /*
1570 * determine how safe it is to execute the pr 1560 * determine how safe it is to execute the proposed program
1571 * - the caller must hold ->cred_guard_mutex 1561 * - the caller must hold ->cred_guard_mutex to protect against
1572 * PTRACE_ATTACH or seccomp thread-sync 1562 * PTRACE_ATTACH or seccomp thread-sync
1573 */ 1563 */
1574 static void check_unsafe_exec(struct linux_bi 1564 static void check_unsafe_exec(struct linux_binprm *bprm)
1575 { 1565 {
1576 struct task_struct *p = current, *t; 1566 struct task_struct *p = current, *t;
1577 unsigned n_fs; 1567 unsigned n_fs;
1578 1568
1579 if (p->ptrace) 1569 if (p->ptrace)
1580 bprm->unsafe |= LSM_UNSAFE_PT 1570 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1581 1571
1582 /* 1572 /*
1583 * This isn't strictly necessary, but 1573 * This isn't strictly necessary, but it makes it harder for LSMs to
1584 * mess up. 1574 * mess up.
1585 */ 1575 */
1586 if (task_no_new_privs(current)) 1576 if (task_no_new_privs(current))
1587 bprm->unsafe |= LSM_UNSAFE_NO 1577 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1588 1578
1589 /* !! 1579 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; 1580 n_fs = 1;
1596 spin_lock(&p->fs->lock); 1581 spin_lock(&p->fs->lock);
1597 rcu_read_lock(); 1582 rcu_read_lock();
1598 for_other_threads(p, t) { !! 1583 while_each_thread(p, t) {
1599 if (t->fs == p->fs) 1584 if (t->fs == p->fs)
1600 n_fs++; 1585 n_fs++;
1601 } 1586 }
1602 rcu_read_unlock(); 1587 rcu_read_unlock();
1603 1588
1604 /* "users" and "in_exec" locked for c <<
1605 if (p->fs->users > n_fs) 1589 if (p->fs->users > n_fs)
1606 bprm->unsafe |= LSM_UNSAFE_SH 1590 bprm->unsafe |= LSM_UNSAFE_SHARE;
1607 else 1591 else
1608 p->fs->in_exec = 1; 1592 p->fs->in_exec = 1;
1609 spin_unlock(&p->fs->lock); 1593 spin_unlock(&p->fs->lock);
1610 } 1594 }
1611 1595
1612 static void bprm_fill_uid(struct linux_binprm 1596 static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1613 { 1597 {
1614 /* Handle suid and sgid on files */ 1598 /* Handle suid and sgid on files */
1615 struct mnt_idmap *idmap; !! 1599 struct user_namespace *mnt_userns;
1616 struct inode *inode = file_inode(file 1600 struct inode *inode = file_inode(file);
1617 unsigned int mode; 1601 unsigned int mode;
1618 vfsuid_t vfsuid; !! 1602 kuid_t uid;
1619 vfsgid_t vfsgid; !! 1603 kgid_t gid;
1620 int err; 1604 int err;
1621 1605
1622 if (!mnt_may_suid(file->f_path.mnt)) 1606 if (!mnt_may_suid(file->f_path.mnt))
1623 return; 1607 return;
1624 1608
1625 if (task_no_new_privs(current)) 1609 if (task_no_new_privs(current))
1626 return; 1610 return;
1627 1611
1628 mode = READ_ONCE(inode->i_mode); 1612 mode = READ_ONCE(inode->i_mode);
1629 if (!(mode & (S_ISUID|S_ISGID))) 1613 if (!(mode & (S_ISUID|S_ISGID)))
1630 return; 1614 return;
1631 1615
1632 idmap = file_mnt_idmap(file); !! 1616 mnt_userns = file_mnt_user_ns(file);
1633 1617
1634 /* Be careful if suid/sgid is set */ 1618 /* Be careful if suid/sgid is set */
1635 inode_lock(inode); 1619 inode_lock(inode);
1636 1620
1637 /* Atomically reload and check mode/u 1621 /* Atomically reload and check mode/uid/gid now that lock held. */
1638 mode = inode->i_mode; 1622 mode = inode->i_mode;
1639 vfsuid = i_uid_into_vfsuid(idmap, ino !! 1623 uid = i_uid_into_mnt(mnt_userns, inode);
1640 vfsgid = i_gid_into_vfsgid(idmap, ino !! 1624 gid = i_gid_into_mnt(mnt_userns, inode);
1641 err = inode_permission(idmap, inode, !! 1625 err = inode_permission(mnt_userns, inode, MAY_EXEC);
1642 inode_unlock(inode); 1626 inode_unlock(inode);
1643 1627
1644 /* Did the exec bit vanish out from u 1628 /* Did the exec bit vanish out from under us? Give up. */
1645 if (err) 1629 if (err)
1646 return; 1630 return;
1647 1631
1648 /* We ignore suid/sgid if there are n 1632 /* We ignore suid/sgid if there are no mappings for them in the ns */
1649 if (!vfsuid_has_mapping(bprm->cred->u !! 1633 if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1650 !vfsgid_has_mapping(bprm->cred->u !! 1634 !kgid_has_mapping(bprm->cred->user_ns, gid))
1651 return; 1635 return;
1652 1636
1653 if (mode & S_ISUID) { 1637 if (mode & S_ISUID) {
1654 bprm->per_clear |= PER_CLEAR_ 1638 bprm->per_clear |= PER_CLEAR_ON_SETID;
1655 bprm->cred->euid = vfsuid_int !! 1639 bprm->cred->euid = uid;
1656 } 1640 }
1657 1641
1658 if ((mode & (S_ISGID | S_IXGRP)) == ( 1642 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1659 bprm->per_clear |= PER_CLEAR_ 1643 bprm->per_clear |= PER_CLEAR_ON_SETID;
1660 bprm->cred->egid = vfsgid_int !! 1644 bprm->cred->egid = gid;
1661 } 1645 }
1662 } 1646 }
1663 1647
1664 /* 1648 /*
1665 * Compute brpm->cred based upon the final bi 1649 * Compute brpm->cred based upon the final binary.
1666 */ 1650 */
1667 static int bprm_creds_from_file(struct linux_ 1651 static int bprm_creds_from_file(struct linux_binprm *bprm)
1668 { 1652 {
1669 /* Compute creds based on which file? 1653 /* Compute creds based on which file? */
1670 struct file *file = bprm->execfd_cred 1654 struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1671 1655
1672 bprm_fill_uid(bprm, file); 1656 bprm_fill_uid(bprm, file);
1673 return security_bprm_creds_from_file( 1657 return security_bprm_creds_from_file(bprm, file);
1674 } 1658 }
1675 1659
1676 /* 1660 /*
1677 * Fill the binprm structure from the inode. 1661 * Fill the binprm structure from the inode.
1678 * Read the first BINPRM_BUF_SIZE bytes 1662 * Read the first BINPRM_BUF_SIZE bytes
1679 * 1663 *
1680 * This may be called multiple times for bina 1664 * This may be called multiple times for binary chains (scripts for example).
1681 */ 1665 */
1682 static int prepare_binprm(struct linux_binprm 1666 static int prepare_binprm(struct linux_binprm *bprm)
1683 { 1667 {
1684 loff_t pos = 0; 1668 loff_t pos = 0;
1685 1669
1686 memset(bprm->buf, 0, BINPRM_BUF_SIZE) 1670 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1687 return kernel_read(bprm->file, bprm-> 1671 return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1688 } 1672 }
1689 1673
1690 /* 1674 /*
1691 * Arguments are '\0' separated strings found 1675 * Arguments are '\0' separated strings found at the location bprm->p
1692 * points to; chop off the first by relocatin 1676 * points to; chop off the first by relocating brpm->p to right after
1693 * the first '\0' encountered. 1677 * the first '\0' encountered.
1694 */ 1678 */
1695 int remove_arg_zero(struct linux_binprm *bprm 1679 int remove_arg_zero(struct linux_binprm *bprm)
1696 { 1680 {
>> 1681 int ret = 0;
1697 unsigned long offset; 1682 unsigned long offset;
1698 char *kaddr; 1683 char *kaddr;
1699 struct page *page; 1684 struct page *page;
1700 1685
1701 if (!bprm->argc) 1686 if (!bprm->argc)
1702 return 0; 1687 return 0;
1703 1688
1704 do { 1689 do {
1705 offset = bprm->p & ~PAGE_MASK 1690 offset = bprm->p & ~PAGE_MASK;
1706 page = get_arg_page(bprm, bpr 1691 page = get_arg_page(bprm, bprm->p, 0);
1707 if (!page) !! 1692 if (!page) {
1708 return -EFAULT; !! 1693 ret = -EFAULT;
>> 1694 goto out;
>> 1695 }
1709 kaddr = kmap_local_page(page) 1696 kaddr = kmap_local_page(page);
1710 1697
1711 for (; offset < PAGE_SIZE && 1698 for (; offset < PAGE_SIZE && kaddr[offset];
1712 offset++, bpr 1699 offset++, bprm->p++)
1713 ; 1700 ;
1714 1701
1715 kunmap_local(kaddr); 1702 kunmap_local(kaddr);
1716 put_arg_page(page); 1703 put_arg_page(page);
1717 } while (offset == PAGE_SIZE); 1704 } while (offset == PAGE_SIZE);
1718 1705
1719 bprm->p++; 1706 bprm->p++;
1720 bprm->argc--; 1707 bprm->argc--;
>> 1708 ret = 0;
1721 1709
1722 return 0; !! 1710 out:
>> 1711 return ret;
1723 } 1712 }
1724 EXPORT_SYMBOL(remove_arg_zero); 1713 EXPORT_SYMBOL(remove_arg_zero);
1725 1714
1726 #define printable(c) (((c)=='\t') || ((c)=='\ 1715 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1727 /* 1716 /*
1728 * cycle the list of binary formats handler, 1717 * cycle the list of binary formats handler, until one recognizes the image
1729 */ 1718 */
1730 static int search_binary_handler(struct linux 1719 static int search_binary_handler(struct linux_binprm *bprm)
1731 { 1720 {
1732 bool need_retry = IS_ENABLED(CONFIG_M 1721 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1733 struct linux_binfmt *fmt; 1722 struct linux_binfmt *fmt;
1734 int retval; 1723 int retval;
1735 1724
1736 retval = prepare_binprm(bprm); 1725 retval = prepare_binprm(bprm);
1737 if (retval < 0) 1726 if (retval < 0)
1738 return retval; 1727 return retval;
1739 1728
1740 retval = security_bprm_check(bprm); 1729 retval = security_bprm_check(bprm);
1741 if (retval) 1730 if (retval)
1742 return retval; 1731 return retval;
1743 1732
1744 retval = -ENOENT; 1733 retval = -ENOENT;
1745 retry: 1734 retry:
1746 read_lock(&binfmt_lock); 1735 read_lock(&binfmt_lock);
1747 list_for_each_entry(fmt, &formats, lh 1736 list_for_each_entry(fmt, &formats, lh) {
1748 if (!try_module_get(fmt->modu 1737 if (!try_module_get(fmt->module))
1749 continue; 1738 continue;
1750 read_unlock(&binfmt_lock); 1739 read_unlock(&binfmt_lock);
1751 1740
1752 retval = fmt->load_binary(bpr 1741 retval = fmt->load_binary(bprm);
1753 1742
1754 read_lock(&binfmt_lock); 1743 read_lock(&binfmt_lock);
1755 put_binfmt(fmt); 1744 put_binfmt(fmt);
1756 if (bprm->point_of_no_return 1745 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1757 read_unlock(&binfmt_l 1746 read_unlock(&binfmt_lock);
1758 return retval; 1747 return retval;
1759 } 1748 }
1760 } 1749 }
1761 read_unlock(&binfmt_lock); 1750 read_unlock(&binfmt_lock);
1762 1751
1763 if (need_retry) { 1752 if (need_retry) {
1764 if (printable(bprm->buf[0]) & 1753 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1765 printable(bprm->buf[2]) & 1754 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1766 return retval; 1755 return retval;
1767 if (request_module("binfmt-%0 1756 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1768 return retval; 1757 return retval;
1769 need_retry = false; 1758 need_retry = false;
1770 goto retry; 1759 goto retry;
1771 } 1760 }
1772 1761
1773 return retval; 1762 return retval;
1774 } 1763 }
1775 1764
1776 /* binfmt handlers will call back into begin_ <<
1777 static int exec_binprm(struct linux_binprm *b 1765 static int exec_binprm(struct linux_binprm *bprm)
1778 { 1766 {
1779 pid_t old_pid, old_vpid; 1767 pid_t old_pid, old_vpid;
1780 int ret, depth; 1768 int ret, depth;
1781 1769
1782 /* Need to fetch pid before load_bina 1770 /* Need to fetch pid before load_binary changes it */
1783 old_pid = current->pid; 1771 old_pid = current->pid;
1784 rcu_read_lock(); 1772 rcu_read_lock();
1785 old_vpid = task_pid_nr_ns(current, ta 1773 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1786 rcu_read_unlock(); 1774 rcu_read_unlock();
1787 1775
1788 /* This allows 4 levels of binfmt rew 1776 /* This allows 4 levels of binfmt rewrites before failing hard. */
1789 for (depth = 0;; depth++) { 1777 for (depth = 0;; depth++) {
1790 struct file *exec; 1778 struct file *exec;
1791 if (depth > 5) 1779 if (depth > 5)
1792 return -ELOOP; 1780 return -ELOOP;
1793 1781
1794 ret = search_binary_handler(b 1782 ret = search_binary_handler(bprm);
1795 if (ret < 0) 1783 if (ret < 0)
1796 return ret; 1784 return ret;
1797 if (!bprm->interpreter) 1785 if (!bprm->interpreter)
1798 break; 1786 break;
1799 1787
1800 exec = bprm->file; 1788 exec = bprm->file;
1801 bprm->file = bprm->interprete 1789 bprm->file = bprm->interpreter;
1802 bprm->interpreter = NULL; 1790 bprm->interpreter = NULL;
1803 1791
>> 1792 allow_write_access(exec);
1804 if (unlikely(bprm->have_execf 1793 if (unlikely(bprm->have_execfd)) {
1805 if (bprm->executable) 1794 if (bprm->executable) {
1806 fput(exec); 1795 fput(exec);
1807 return -ENOEX 1796 return -ENOEXEC;
1808 } 1797 }
1809 bprm->executable = ex 1798 bprm->executable = exec;
1810 } else 1799 } else
1811 fput(exec); 1800 fput(exec);
1812 } 1801 }
1813 1802
1814 audit_bprm(bprm); 1803 audit_bprm(bprm);
1815 trace_sched_process_exec(current, old 1804 trace_sched_process_exec(current, old_pid, bprm);
1816 ptrace_event(PTRACE_EVENT_EXEC, old_v 1805 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1817 proc_exec_connector(current); 1806 proc_exec_connector(current);
1818 return 0; 1807 return 0;
1819 } 1808 }
1820 1809
1821 static int bprm_execve(struct linux_binprm *b !! 1810 /*
>> 1811 * sys_execve() executes a new program.
>> 1812 */
>> 1813 static int bprm_execve(struct linux_binprm *bprm,
>> 1814 int fd, struct filename *filename, int flags)
1822 { 1815 {
>> 1816 struct file *file;
1823 int retval; 1817 int retval;
1824 1818
1825 retval = prepare_bprm_creds(bprm); 1819 retval = prepare_bprm_creds(bprm);
1826 if (retval) 1820 if (retval)
1827 return retval; 1821 return retval;
1828 1822
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); 1823 check_unsafe_exec(bprm);
1835 current->in_execve = 1; 1824 current->in_execve = 1;
1836 sched_mm_cid_before_execve(current); !! 1825
>> 1826 file = do_open_execat(fd, filename, flags);
>> 1827 retval = PTR_ERR(file);
>> 1828 if (IS_ERR(file))
>> 1829 goto out_unmark;
1837 1830
1838 sched_exec(); 1831 sched_exec();
1839 1832
>> 1833 bprm->file = file;
>> 1834 /*
>> 1835 * Record that a name derived from an O_CLOEXEC fd will be
>> 1836 * inaccessible after exec. This allows the code in exec to
>> 1837 * choose to fail when the executable is not mmaped into the
>> 1838 * interpreter and an open file descriptor is not passed to
>> 1839 * the interpreter. This makes for a better user experience
>> 1840 * than having the interpreter start and then immediately fail
>> 1841 * when it finds the executable is inaccessible.
>> 1842 */
>> 1843 if (bprm->fdpath && get_close_on_exec(fd))
>> 1844 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
>> 1845
1840 /* Set the unchanging part of bprm->c 1846 /* Set the unchanging part of bprm->cred */
1841 retval = security_bprm_creds_for_exec 1847 retval = security_bprm_creds_for_exec(bprm);
1842 if (retval) 1848 if (retval)
1843 goto out; 1849 goto out;
1844 1850
1845 retval = ccs_exec_binprm(bprm); 1851 retval = ccs_exec_binprm(bprm);
1846 if (retval < 0) 1852 if (retval < 0)
1847 goto out; 1853 goto out;
1848 1854
1849 sched_mm_cid_after_execve(current); <<
1850 /* execve succeeded */ 1855 /* execve succeeded */
1851 current->fs->in_exec = 0; 1856 current->fs->in_exec = 0;
1852 current->in_execve = 0; 1857 current->in_execve = 0;
1853 rseq_execve(current); 1858 rseq_execve(current);
1854 user_events_execve(current); <<
1855 acct_update_integrals(current); 1859 acct_update_integrals(current);
1856 task_numa_free(current, false); 1860 task_numa_free(current, false);
1857 return retval; 1861 return retval;
1858 1862
1859 out: 1863 out:
1860 /* 1864 /*
1861 * If past the point of no return ens 1865 * If past the point of no return ensure the code never
1862 * returns to the userspace process. 1866 * returns to the userspace process. Use an existing fatal
1863 * signal if present otherwise termin 1867 * signal if present otherwise terminate the process with
1864 * SIGSEGV. 1868 * SIGSEGV.
1865 */ 1869 */
1866 if (bprm->point_of_no_return && !fata 1870 if (bprm->point_of_no_return && !fatal_signal_pending(current))
1867 force_fatal_sig(SIGSEGV); 1871 force_fatal_sig(SIGSEGV);
1868 1872
1869 sched_mm_cid_after_execve(current); !! 1873 out_unmark:
1870 current->fs->in_exec = 0; 1874 current->fs->in_exec = 0;
1871 current->in_execve = 0; 1875 current->in_execve = 0;
1872 1876
1873 return retval; 1877 return retval;
1874 } 1878 }
1875 1879
1876 static int do_execveat_common(int fd, struct 1880 static int do_execveat_common(int fd, struct filename *filename,
1877 struct user_arg 1881 struct user_arg_ptr argv,
1878 struct user_arg 1882 struct user_arg_ptr envp,
1879 int flags) 1883 int flags)
1880 { 1884 {
1881 struct linux_binprm *bprm; 1885 struct linux_binprm *bprm;
1882 int retval; 1886 int retval;
1883 1887
1884 if (IS_ERR(filename)) 1888 if (IS_ERR(filename))
1885 return PTR_ERR(filename); 1889 return PTR_ERR(filename);
1886 1890
1887 /* 1891 /*
1888 * We move the actual failure in case 1892 * We move the actual failure in case of RLIMIT_NPROC excess from
1889 * set*uid() to execve() because too 1893 * set*uid() to execve() because too many poorly written programs
1890 * don't check setuid() return code. 1894 * don't check setuid() return code. Here we additionally recheck
1891 * whether NPROC limit is still excee 1895 * whether NPROC limit is still exceeded.
1892 */ 1896 */
1893 if ((current->flags & PF_NPROC_EXCEED 1897 if ((current->flags & PF_NPROC_EXCEEDED) &&
1894 is_rlimit_overlimit(current_ucoun 1898 is_rlimit_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1895 retval = -EAGAIN; 1899 retval = -EAGAIN;
1896 goto out_ret; 1900 goto out_ret;
1897 } 1901 }
1898 1902
1899 /* We're below the limit (still or ag 1903 /* We're below the limit (still or again), so we don't want to make
1900 * further execve() calls fail. */ 1904 * further execve() calls fail. */
1901 current->flags &= ~PF_NPROC_EXCEEDED; 1905 current->flags &= ~PF_NPROC_EXCEEDED;
1902 1906
1903 bprm = alloc_bprm(fd, filename, flags !! 1907 bprm = alloc_bprm(fd, filename);
1904 if (IS_ERR(bprm)) { 1908 if (IS_ERR(bprm)) {
1905 retval = PTR_ERR(bprm); 1909 retval = PTR_ERR(bprm);
1906 goto out_ret; 1910 goto out_ret;
1907 } 1911 }
1908 1912
1909 retval = count(argv, MAX_ARG_STRINGS) 1913 retval = count(argv, MAX_ARG_STRINGS);
1910 if (retval == 0) 1914 if (retval == 0)
1911 pr_warn_once("process '%s' la 1915 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1912 current->comm, b 1916 current->comm, bprm->filename);
1913 if (retval < 0) 1917 if (retval < 0)
1914 goto out_free; 1918 goto out_free;
1915 bprm->argc = retval; 1919 bprm->argc = retval;
1916 1920
1917 retval = count(envp, MAX_ARG_STRINGS) 1921 retval = count(envp, MAX_ARG_STRINGS);
1918 if (retval < 0) 1922 if (retval < 0)
1919 goto out_free; 1923 goto out_free;
1920 bprm->envc = retval; 1924 bprm->envc = retval;
1921 1925
1922 retval = bprm_stack_limits(bprm); 1926 retval = bprm_stack_limits(bprm);
1923 if (retval < 0) 1927 if (retval < 0)
1924 goto out_free; 1928 goto out_free;
1925 1929
1926 retval = copy_string_kernel(bprm->fil 1930 retval = copy_string_kernel(bprm->filename, bprm);
1927 if (retval < 0) 1931 if (retval < 0)
1928 goto out_free; 1932 goto out_free;
1929 bprm->exec = bprm->p; 1933 bprm->exec = bprm->p;
1930 1934
1931 retval = copy_strings(bprm->envc, env 1935 retval = copy_strings(bprm->envc, envp, bprm);
1932 if (retval < 0) 1936 if (retval < 0)
1933 goto out_free; 1937 goto out_free;
1934 1938
1935 retval = copy_strings(bprm->argc, arg 1939 retval = copy_strings(bprm->argc, argv, bprm);
1936 if (retval < 0) 1940 if (retval < 0)
1937 goto out_free; 1941 goto out_free;
1938 1942
1939 /* 1943 /*
1940 * When argv is empty, add an empty s 1944 * When argv is empty, add an empty string ("") as argv[0] to
1941 * ensure confused userspace programs 1945 * ensure confused userspace programs that start processing
1942 * from argv[1] won't end up walking 1946 * from argv[1] won't end up walking envp. See also
1943 * bprm_stack_limits(). 1947 * bprm_stack_limits().
1944 */ 1948 */
1945 if (bprm->argc == 0) { 1949 if (bprm->argc == 0) {
1946 retval = copy_string_kernel(" 1950 retval = copy_string_kernel("", bprm);
1947 if (retval < 0) 1951 if (retval < 0)
1948 goto out_free; 1952 goto out_free;
1949 bprm->argc = 1; 1953 bprm->argc = 1;
1950 } 1954 }
1951 1955
1952 retval = bprm_execve(bprm); !! 1956 retval = bprm_execve(bprm, fd, filename, flags);
1953 out_free: 1957 out_free:
1954 free_bprm(bprm); 1958 free_bprm(bprm);
1955 1959
1956 out_ret: 1960 out_ret:
1957 putname(filename); 1961 putname(filename);
1958 return retval; 1962 return retval;
1959 } 1963 }
1960 1964
1961 int kernel_execve(const char *kernel_filename 1965 int kernel_execve(const char *kernel_filename,
1962 const char *const *argv, co 1966 const char *const *argv, const char *const *envp)
1963 { 1967 {
1964 struct filename *filename; 1968 struct filename *filename;
1965 struct linux_binprm *bprm; 1969 struct linux_binprm *bprm;
1966 int fd = AT_FDCWD; 1970 int fd = AT_FDCWD;
1967 int retval; 1971 int retval;
1968 1972
1969 /* It is non-sense for kernel threads 1973 /* It is non-sense for kernel threads to call execve */
1970 if (WARN_ON_ONCE(current->flags & PF_ 1974 if (WARN_ON_ONCE(current->flags & PF_KTHREAD))
1971 return -EINVAL; 1975 return -EINVAL;
1972 1976
1973 filename = getname_kernel(kernel_file 1977 filename = getname_kernel(kernel_filename);
1974 if (IS_ERR(filename)) 1978 if (IS_ERR(filename))
1975 return PTR_ERR(filename); 1979 return PTR_ERR(filename);
1976 1980
1977 bprm = alloc_bprm(fd, filename, 0); !! 1981 bprm = alloc_bprm(fd, filename);
1978 if (IS_ERR(bprm)) { 1982 if (IS_ERR(bprm)) {
1979 retval = PTR_ERR(bprm); 1983 retval = PTR_ERR(bprm);
1980 goto out_ret; 1984 goto out_ret;
1981 } 1985 }
1982 1986
1983 retval = count_strings_kernel(argv); 1987 retval = count_strings_kernel(argv);
1984 if (WARN_ON_ONCE(retval == 0)) 1988 if (WARN_ON_ONCE(retval == 0))
1985 retval = -EINVAL; 1989 retval = -EINVAL;
1986 if (retval < 0) 1990 if (retval < 0)
1987 goto out_free; 1991 goto out_free;
1988 bprm->argc = retval; 1992 bprm->argc = retval;
1989 1993
1990 retval = count_strings_kernel(envp); 1994 retval = count_strings_kernel(envp);
1991 if (retval < 0) 1995 if (retval < 0)
1992 goto out_free; 1996 goto out_free;
1993 bprm->envc = retval; 1997 bprm->envc = retval;
1994 1998
1995 retval = bprm_stack_limits(bprm); 1999 retval = bprm_stack_limits(bprm);
1996 if (retval < 0) 2000 if (retval < 0)
1997 goto out_free; 2001 goto out_free;
1998 2002
1999 retval = copy_string_kernel(bprm->fil 2003 retval = copy_string_kernel(bprm->filename, bprm);
2000 if (retval < 0) 2004 if (retval < 0)
2001 goto out_free; 2005 goto out_free;
2002 bprm->exec = bprm->p; 2006 bprm->exec = bprm->p;
2003 2007
2004 retval = copy_strings_kernel(bprm->en 2008 retval = copy_strings_kernel(bprm->envc, envp, bprm);
2005 if (retval < 0) 2009 if (retval < 0)
2006 goto out_free; 2010 goto out_free;
2007 2011
2008 retval = copy_strings_kernel(bprm->ar 2012 retval = copy_strings_kernel(bprm->argc, argv, bprm);
2009 if (retval < 0) 2013 if (retval < 0)
2010 goto out_free; 2014 goto out_free;
2011 2015
2012 retval = bprm_execve(bprm); !! 2016 retval = bprm_execve(bprm, fd, filename, 0);
2013 out_free: 2017 out_free:
2014 free_bprm(bprm); 2018 free_bprm(bprm);
2015 out_ret: 2019 out_ret:
2016 putname(filename); 2020 putname(filename);
2017 return retval; 2021 return retval;
2018 } 2022 }
2019 2023
2020 static int do_execve(struct filename *filenam 2024 static int do_execve(struct filename *filename,
2021 const char __user *const __user *__ar 2025 const char __user *const __user *__argv,
2022 const char __user *const __user *__en 2026 const char __user *const __user *__envp)
2023 { 2027 {
2024 struct user_arg_ptr argv = { .ptr.nat 2028 struct user_arg_ptr argv = { .ptr.native = __argv };
2025 struct user_arg_ptr envp = { .ptr.nat 2029 struct user_arg_ptr envp = { .ptr.native = __envp };
2026 return do_execveat_common(AT_FDCWD, f 2030 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2027 } 2031 }
2028 2032
2029 static int do_execveat(int fd, struct filenam 2033 static int do_execveat(int fd, struct filename *filename,
2030 const char __user *const __us 2034 const char __user *const __user *__argv,
2031 const char __user *const __us 2035 const char __user *const __user *__envp,
2032 int flags) 2036 int flags)
2033 { 2037 {
2034 struct user_arg_ptr argv = { .ptr.nat 2038 struct user_arg_ptr argv = { .ptr.native = __argv };
2035 struct user_arg_ptr envp = { .ptr.nat 2039 struct user_arg_ptr envp = { .ptr.native = __envp };
2036 2040
2037 return do_execveat_common(fd, filenam 2041 return do_execveat_common(fd, filename, argv, envp, flags);
2038 } 2042 }
2039 2043
2040 #ifdef CONFIG_COMPAT 2044 #ifdef CONFIG_COMPAT
2041 static int compat_do_execve(struct filename * 2045 static int compat_do_execve(struct filename *filename,
2042 const compat_uptr_t __user *__argv, 2046 const compat_uptr_t __user *__argv,
2043 const compat_uptr_t __user *__envp) 2047 const compat_uptr_t __user *__envp)
2044 { 2048 {
2045 struct user_arg_ptr argv = { 2049 struct user_arg_ptr argv = {
2046 .is_compat = true, 2050 .is_compat = true,
2047 .ptr.compat = __argv, 2051 .ptr.compat = __argv,
2048 }; 2052 };
2049 struct user_arg_ptr envp = { 2053 struct user_arg_ptr envp = {
2050 .is_compat = true, 2054 .is_compat = true,
2051 .ptr.compat = __envp, 2055 .ptr.compat = __envp,
2052 }; 2056 };
2053 return do_execveat_common(AT_FDCWD, f 2057 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2054 } 2058 }
2055 2059
2056 static int compat_do_execveat(int fd, struct 2060 static int compat_do_execveat(int fd, struct filename *filename,
2057 const compat_up 2061 const compat_uptr_t __user *__argv,
2058 const compat_up 2062 const compat_uptr_t __user *__envp,
2059 int flags) 2063 int flags)
2060 { 2064 {
2061 struct user_arg_ptr argv = { 2065 struct user_arg_ptr argv = {
2062 .is_compat = true, 2066 .is_compat = true,
2063 .ptr.compat = __argv, 2067 .ptr.compat = __argv,
2064 }; 2068 };
2065 struct user_arg_ptr envp = { 2069 struct user_arg_ptr envp = {
2066 .is_compat = true, 2070 .is_compat = true,
2067 .ptr.compat = __envp, 2071 .ptr.compat = __envp,
2068 }; 2072 };
2069 return do_execveat_common(fd, filenam 2073 return do_execveat_common(fd, filename, argv, envp, flags);
2070 } 2074 }
2071 #endif 2075 #endif
2072 2076
2073 void set_binfmt(struct linux_binfmt *new) 2077 void set_binfmt(struct linux_binfmt *new)
2074 { 2078 {
2075 struct mm_struct *mm = current->mm; 2079 struct mm_struct *mm = current->mm;
2076 2080
2077 if (mm->binfmt) 2081 if (mm->binfmt)
2078 module_put(mm->binfmt->module 2082 module_put(mm->binfmt->module);
2079 2083
2080 mm->binfmt = new; 2084 mm->binfmt = new;
2081 if (new) 2085 if (new)
2082 __module_get(new->module); 2086 __module_get(new->module);
2083 } 2087 }
2084 EXPORT_SYMBOL(set_binfmt); 2088 EXPORT_SYMBOL(set_binfmt);
2085 2089
2086 /* 2090 /*
2087 * set_dumpable stores three-value SUID_DUMP_ 2091 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2088 */ 2092 */
2089 void set_dumpable(struct mm_struct *mm, int v 2093 void set_dumpable(struct mm_struct *mm, int value)
2090 { 2094 {
2091 if (WARN_ON((unsigned)value > SUID_DU 2095 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2092 return; 2096 return;
2093 2097
2094 set_mask_bits(&mm->flags, MMF_DUMPABL 2098 set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2095 } 2099 }
2096 2100
2097 SYSCALL_DEFINE3(execve, 2101 SYSCALL_DEFINE3(execve,
2098 const char __user *, filename 2102 const char __user *, filename,
2099 const char __user *const __us 2103 const char __user *const __user *, argv,
2100 const char __user *const __us 2104 const char __user *const __user *, envp)
2101 { 2105 {
2102 return do_execve(getname(filename), a 2106 return do_execve(getname(filename), argv, envp);
2103 } 2107 }
2104 2108
2105 SYSCALL_DEFINE5(execveat, 2109 SYSCALL_DEFINE5(execveat,
2106 int, fd, const char __user *, 2110 int, fd, const char __user *, filename,
2107 const char __user *const __us 2111 const char __user *const __user *, argv,
2108 const char __user *const __us 2112 const char __user *const __user *, envp,
2109 int, flags) 2113 int, flags)
2110 { 2114 {
2111 return do_execveat(fd, 2115 return do_execveat(fd,
2112 getname_uflags(fil 2116 getname_uflags(filename, flags),
2113 argv, envp, flags) 2117 argv, envp, flags);
2114 } 2118 }
2115 2119
2116 #ifdef CONFIG_COMPAT 2120 #ifdef CONFIG_COMPAT
2117 COMPAT_SYSCALL_DEFINE3(execve, const char __u 2121 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2118 const compat_uptr_t __user *, argv, 2122 const compat_uptr_t __user *, argv,
2119 const compat_uptr_t __user *, envp) 2123 const compat_uptr_t __user *, envp)
2120 { 2124 {
2121 return compat_do_execve(getname(filen 2125 return compat_do_execve(getname(filename), argv, envp);
2122 } 2126 }
2123 2127
2124 COMPAT_SYSCALL_DEFINE5(execveat, int, fd, 2128 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2125 const char __user *, f 2129 const char __user *, filename,
2126 const compat_uptr_t __ 2130 const compat_uptr_t __user *, argv,
2127 const compat_uptr_t __ 2131 const compat_uptr_t __user *, envp,
2128 int, flags) 2132 int, flags)
2129 { 2133 {
2130 return compat_do_execveat(fd, 2134 return compat_do_execveat(fd,
2131 getname_ufl 2135 getname_uflags(filename, flags),
2132 argv, envp, 2136 argv, envp, flags);
2133 } 2137 }
2134 #endif 2138 #endif
2135 2139
2136 #ifdef CONFIG_SYSCTL 2140 #ifdef CONFIG_SYSCTL
2137 2141
2138 static int proc_dointvec_minmax_coredump(cons !! 2142 static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
2139 void *buffer, size_t *lenp, l 2143 void *buffer, size_t *lenp, loff_t *ppos)
2140 { 2144 {
2141 int error = proc_dointvec_minmax(tabl 2145 int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2142 2146
2143 if (!error) 2147 if (!error)
2144 validate_coredump_safety(); 2148 validate_coredump_safety();
2145 return error; 2149 return error;
2146 } 2150 }
2147 2151
2148 static struct ctl_table fs_exec_sysctls[] = { 2152 static struct ctl_table fs_exec_sysctls[] = {
2149 { 2153 {
2150 .procname = "suid_dumpa 2154 .procname = "suid_dumpable",
2151 .data = &suid_dumpa 2155 .data = &suid_dumpable,
2152 .maxlen = sizeof(int) 2156 .maxlen = sizeof(int),
2153 .mode = 0644, 2157 .mode = 0644,
2154 .proc_handler = proc_dointv 2158 .proc_handler = proc_dointvec_minmax_coredump,
2155 .extra1 = SYSCTL_ZERO 2159 .extra1 = SYSCTL_ZERO,
2156 .extra2 = SYSCTL_TWO, 2160 .extra2 = SYSCTL_TWO,
2157 }, 2161 },
>> 2162 { }
2158 }; 2163 };
2159 2164
2160 static int __init init_fs_exec_sysctls(void) 2165 static int __init init_fs_exec_sysctls(void)
2161 { 2166 {
2162 register_sysctl_init("fs", fs_exec_sy 2167 register_sysctl_init("fs", fs_exec_sysctls);
2163 return 0; 2168 return 0;
2164 } 2169 }
2165 2170
2166 fs_initcall(init_fs_exec_sysctls); 2171 fs_initcall(init_fs_exec_sysctls);
2167 #endif /* CONFIG_SYSCTL */ 2172 #endif /* CONFIG_SYSCTL */
2168 <<
2169 #ifdef CONFIG_EXEC_KUNIT_TEST <<
2170 #include "tests/exec_kunit.c" <<
2171 #endif <<
2172 2173
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.