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