brintos

brintos / linux-shallow public Read only

0
0
Text · 52.3 KiB · 6a2736c Raw
2188 lines · c
1// SPDX-License-Identifier: GPL-2.0-only2/*3 *  linux/fs/exec.c4 *5 *  Copyright (C) 1991, 1992  Linus Torvalds6 */7 8/*9 * #!-checking implemented by tytso.10 */11/*12 * Demand-loading implemented 01.12.91 - no need to read anything but13 * the header into memory. The inode of the executable is put into14 * "current->executable", and page faults do the actual loading. Clean.15 *16 * Once more I can proudly say that linux stood up to being changed: it17 * was less than 2 hours work to get demand-loading completely implemented.18 *19 * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,20 * current->executable is only used by the procfs.  This allows a dispatch21 * table to check for several different types  of binary formats.  We keep22 * trying until we recognize the file or we run out of supported binary23 * formats.24 */25 26#include <linux/kernel_read_file.h>27#include <linux/slab.h>28#include <linux/file.h>29#include <linux/fdtable.h>30#include <linux/mm.h>31#include <linux/stat.h>32#include <linux/fcntl.h>33#include <linux/swap.h>34#include <linux/string.h>35#include <linux/init.h>36#include <linux/sched/mm.h>37#include <linux/sched/coredump.h>38#include <linux/sched/signal.h>39#include <linux/sched/numa_balancing.h>40#include <linux/sched/task.h>41#include <linux/pagemap.h>42#include <linux/perf_event.h>43#include <linux/highmem.h>44#include <linux/spinlock.h>45#include <linux/key.h>46#include <linux/personality.h>47#include <linux/binfmts.h>48#include <linux/utsname.h>49#include <linux/pid_namespace.h>50#include <linux/module.h>51#include <linux/namei.h>52#include <linux/mount.h>53#include <linux/security.h>54#include <linux/syscalls.h>55#include <linux/tsacct_kern.h>56#include <linux/cn_proc.h>57#include <linux/audit.h>58#include <linux/kmod.h>59#include <linux/fsnotify.h>60#include <linux/fs_struct.h>61#include <linux/oom.h>62#include <linux/compat.h>63#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 72#include <linux/uaccess.h>73#include <asm/mmu_context.h>74#include <asm/tlb.h>75 76#include <trace/events/task.h>77#include "internal.h"78 79#include <trace/events/sched.h>80 81static int bprm_creds_from_file(struct linux_binprm *bprm);82 83int suid_dumpable = 0;84 85static LIST_HEAD(formats);86static DEFINE_RWLOCK(binfmt_lock);87 88void __register_binfmt(struct linux_binfmt * fmt, int insert)89{90	write_lock(&binfmt_lock);91	insert ? list_add(&fmt->lh, &formats) :92		 list_add_tail(&fmt->lh, &formats);93	write_unlock(&binfmt_lock);94}95 96EXPORT_SYMBOL(__register_binfmt);97 98void unregister_binfmt(struct linux_binfmt * fmt)99{100	write_lock(&binfmt_lock);101	list_del(&fmt->lh);102	write_unlock(&binfmt_lock);103}104 105EXPORT_SYMBOL(unregister_binfmt);106 107static inline void put_binfmt(struct linux_binfmt * fmt)108{109	module_put(fmt->module);110}111 112bool path_noexec(const struct path *path)113{114	return (path->mnt->mnt_flags & MNT_NOEXEC) ||115	       (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);116}117 118#ifdef CONFIG_USELIB119/*120 * Note that a shared library must be both readable and executable due to121 * security reasons.122 *123 * Also note that we take the address to load from the file itself.124 */125SYSCALL_DEFINE1(uselib, const char __user *, library)126{127	struct linux_binfmt *fmt;128	struct file *file;129	struct filename *tmp = getname(library);130	int error = PTR_ERR(tmp);131	static const struct open_flags uselib_flags = {132		.open_flag = O_LARGEFILE | O_RDONLY,133		.acc_mode = MAY_READ | MAY_EXEC,134		.intent = LOOKUP_OPEN,135		.lookup_flags = LOOKUP_FOLLOW,136	};137 138	if (IS_ERR(tmp))139		goto out;140 141	file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);142	putname(tmp);143	error = PTR_ERR(file);144	if (IS_ERR(file))145		goto out;146 147	/*148	 * Check do_open_execat() for an explanation.149	 */150	error = -EACCES;151	if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode)) ||152	    path_noexec(&file->f_path))153		goto exit;154 155	error = -ENOEXEC;156 157	read_lock(&binfmt_lock);158	list_for_each_entry(fmt, &formats, lh) {159		if (!fmt->load_shlib)160			continue;161		if (!try_module_get(fmt->module))162			continue;163		read_unlock(&binfmt_lock);164		error = fmt->load_shlib(file);165		read_lock(&binfmt_lock);166		put_binfmt(fmt);167		if (error != -ENOEXEC)168			break;169	}170	read_unlock(&binfmt_lock);171exit:172	fput(file);173out:174	return error;175}176#endif /* #ifdef CONFIG_USELIB */177 178#ifdef CONFIG_MMU179/*180 * The nascent bprm->mm is not visible until exec_mmap() but it can181 * use a lot of memory, account these pages in current->mm temporary182 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we183 * change the counter back via acct_arg_size(0).184 */185static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)186{187	struct mm_struct *mm = current->mm;188	long diff = (long)(pages - bprm->vma_pages);189 190	if (!mm || !diff)191		return;192 193	bprm->vma_pages = pages;194	add_mm_counter(mm, MM_ANONPAGES, diff);195}196 197static HWJS_SUSPENDS struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,198		int write)199{200	struct page *page;201	struct vm_area_struct *vma = bprm->vma;202	struct mm_struct *mm = bprm->mm;203	int ret;204 205	/*206	 * Avoid relying on expanding the stack down in GUP (which207	 * does not work for STACK_GROWSUP anyway), and just do it208	 * by hand ahead of time.209	 */210	if (write && pos < vma->vm_start) {211		mmap_write_lock(mm);212		ret = expand_downwards(vma, pos);213		if (unlikely(ret < 0)) {214			mmap_write_unlock(mm);215			return NULL;216		}217		mmap_write_downgrade(mm);218	} else219		mmap_read_lock(mm);220 221	/*222	 * We are doing an exec().  'current' is the process223	 * doing the exec and 'mm' is the new process's mm.224	 */225	ret = get_user_pages_remote(mm, pos, 1,226			write ? FOLL_WRITE : 0,227			&page, NULL);228	mmap_read_unlock(mm);229	if (ret <= 0)230		return NULL;231 232	if (write)233		acct_arg_size(bprm, vma_pages(vma));234 235	return page;236}237 238static void put_arg_page(struct page *page)239{240	put_page(page);241}242 243static void free_arg_pages(struct linux_binprm *bprm)244{245}246 247static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,248		struct page *page)249{250	flush_cache_page(bprm->vma, pos, page_to_pfn(page));251}252 253static HWJS_SUSPENDS int __bprm_mm_init(struct linux_binprm *bprm)254{255	int err;256	struct vm_area_struct *vma = NULL;257	struct mm_struct *mm = bprm->mm;258 259	bprm->vma = vma = vm_area_alloc(mm);260	if (!vma)261		return -ENOMEM;262	vma_set_anonymous(vma);263 264	if (mmap_write_lock_killable(mm)) {265		err = -EINTR;266		goto err_free;267	}268 269	/*270	 * Need to be called with mmap write lock271	 * held, to avoid race with ksmd.272	 */273	err = ksm_execve(mm);274	if (err)275		goto err_ksm;276 277	/*278	 * Place the stack at the largest stack address the architecture279	 * supports. Later, we'll move this to an appropriate place. We don't280	 * use STACK_TOP because that can depend on attributes which aren't281	 * configured yet.282	 */283	BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);284	vma->vm_end = STACK_TOP_MAX;285	vma->vm_start = vma->vm_end - PAGE_SIZE;286	vm_flags_init(vma, VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP);287	vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);288 289	err = insert_vm_struct(mm, vma);290	if (err)291		goto err;292 293	mm->stack_vm = mm->total_vm = 1;294	mmap_write_unlock(mm);295	bprm->p = vma->vm_end - sizeof(void *);296	return 0;297err:298	ksm_exit(mm);299err_ksm:300	mmap_write_unlock(mm);301err_free:302	bprm->vma = NULL;303	vm_area_free(vma);304	return err;305}306 307static bool valid_arg_len(struct linux_binprm *bprm, long len)308{309	return len <= MAX_ARG_STRLEN;310}311 312#else313 314static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)315{316}317 318static HWJS_SUSPENDS struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,319		int write)320{321	struct page *page;322 323	page = bprm->page[pos / PAGE_SIZE];324	if (!page && write) {325		page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);326		if (!page)327			return NULL;328		bprm->page[pos / PAGE_SIZE] = page;329	}330 331	return page;332}333 334static void put_arg_page(struct page *page)335{336}337 338static void free_arg_page(struct linux_binprm *bprm, int i)339{340	if (bprm->page[i]) {341		__free_page(bprm->page[i]);342		bprm->page[i] = NULL;343	}344}345 346static void free_arg_pages(struct linux_binprm *bprm)347{348	int i;349 350	for (i = 0; i < MAX_ARG_PAGES; i++)351		free_arg_page(bprm, i);352}353 354static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,355		struct page *page)356{357}358 359static HWJS_SUSPENDS int __bprm_mm_init(struct linux_binprm *bprm)360{361	bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);362	return 0;363}364 365static bool valid_arg_len(struct linux_binprm *bprm, long len)366{367	return len <= bprm->p;368}369 370#endif /* CONFIG_MMU */371 372/*373 * Create a new mm_struct and populate it with a temporary stack374 * vm_area_struct.  We don't have enough context at this point to set the stack375 * flags, permissions, and offset, so we use temporary values.  We'll update376 * them later in setup_arg_pages().377 */378static int bprm_mm_init(struct linux_binprm *bprm)379{380	int err;381	struct mm_struct *mm = NULL;382 383	bprm->mm = mm = mm_alloc();384	err = -ENOMEM;385	if (!mm)386		goto err;387 388	/* Save current stack limit for all calculations made during exec. */389	task_lock(current->group_leader);390	bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];391	task_unlock(current->group_leader);392 393	err = __bprm_mm_init(bprm);394	if (err)395		goto err;396 397	return 0;398 399err:400	if (mm) {401		bprm->mm = NULL;402		mmdrop(mm);403	}404 405	return err;406}407 408struct user_arg_ptr {409#ifdef CONFIG_COMPAT410	bool is_compat;411#endif412	union {413		const char __user *const __user *native;414#ifdef CONFIG_COMPAT415		const compat_uptr_t __user *compat;416#endif417	} ptr;418};419 420static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)421{422	const char __user *native;423 424#ifdef CONFIG_COMPAT425	if (unlikely(argv.is_compat)) {426		compat_uptr_t compat;427 428		if (get_user(compat, argv.ptr.compat + nr))429			return ERR_PTR(-EFAULT);430 431		return compat_ptr(compat);432	}433#endif434 435	if (get_user(native, argv.ptr.native + nr))436		return ERR_PTR(-EFAULT);437 438	return native;439}440 441/*442 * count() counts the number of strings in array ARGV.443 */444static HWJS_SUSPENDS int count(struct user_arg_ptr argv, int max)445{446	int i = 0;447 448	if (argv.ptr.native != NULL) {449		for (;;) {450			const char __user *p = get_user_arg_ptr(argv, i);451 452			if (!p)453				break;454 455			if (IS_ERR(p))456				return -EFAULT;457 458			if (i >= max)459				return -E2BIG;460			++i;461 462			if (fatal_signal_pending(current))463				return -ERESTARTNOHAND;464			cond_resched();465		}466	}467	return i;468}469 470static HWJS_SUSPENDS int count_strings_kernel(const char *const *argv)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(current))481			return -ERESTARTNOHAND;482		cond_resched();483	}484	return i;485}486 487static inline int bprm_set_stack_limit(struct linux_binprm *bprm,488				       unsigned long limit)489{490#ifdef CONFIG_MMU491	/* Avoid a pathological bprm->p. */492	if (bprm->p < limit)493		return -E2BIG;494	bprm->argmin = bprm->p - limit;495#endif496	return 0;497}498static inline bool bprm_hit_stack_limit(struct linux_binprm *bprm)499{500#ifdef CONFIG_MMU501	return bprm->p < bprm->argmin;502#else503	return false;504#endif505}506 507/*508 * Calculate bprm->argmin from:509 * - _STK_LIM510 * - ARG_MAX511 * - bprm->rlim_stack.rlim_cur512 * - bprm->argc513 * - bprm->envc514 * - bprm->p515 */516static int bprm_stack_limits(struct linux_binprm *bprm)517{518	unsigned long limit, ptr_size;519 520	/*521	 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM522	 * (whichever is smaller) for the argv+env strings.523	 * This ensures that:524	 *  - the remaining binfmt code will not run out of stack space,525	 *  - the program will have a reasonable amount of stack left526	 *    to work from.527	 */528	limit = _STK_LIM / 4 * 3;529	limit = min(limit, bprm->rlim_stack.rlim_cur / 4);530	/*531	 * We've historically supported up to 32 pages (ARG_MAX)532	 * of argument strings even with small stacks533	 */534	limit = max_t(unsigned long, limit, ARG_MAX);535	/* Reject totally pathological counts. */536	if (bprm->argc < 0 || bprm->envc < 0)537		return -E2BIG;538	/*539	 * We must account for the size of all the argv and envp pointers to540	 * the argv and envp strings, since they will also take up space in541	 * the stack. They aren't stored until much later when we can't542	 * signal to the parent that the child has run out of stack space.543	 * Instead, calculate it here so it's possible to fail gracefully.544	 *545	 * In the case of argc = 0, make sure there is space for adding a546	 * empty string (which will bump argc to 1), to ensure confused547	 * userspace programs don't start processing from argv[1], thinking548	 * argc can never be 0, to keep them from walking envp by accident.549	 * See do_execveat_common().550	 */551	if (check_add_overflow(max(bprm->argc, 1), bprm->envc, &ptr_size) ||552	    check_mul_overflow(ptr_size, sizeof(void *), &ptr_size))553		return -E2BIG;554	if (limit <= ptr_size)555		return -E2BIG;556	limit -= ptr_size;557 558	return bprm_set_stack_limit(bprm, limit);559}560 561/*562 * 'copy_strings()' copies argument/environment strings from the old563 * processes's memory to the new process's stack.  The call to get_user_pages()564 * ensures the destination page is created and not swapped out.565 */566static HWJS_SUSPENDS int copy_strings(int argc, struct user_arg_ptr argv,567			struct linux_binprm *bprm)568{569	struct page *kmapped_page = NULL;570	char *kaddr = NULL;571	unsigned long kpos = 0;572	int ret;573 574	while (argc-- > 0) {575		const char __user *str;576		int len;577		unsigned long pos;578 579		ret = -EFAULT;580		str = get_user_arg_ptr(argv, argc);581		if (IS_ERR(str))582			goto out;583 584		len = strnlen_user(str, MAX_ARG_STRLEN);585		if (!len)586			goto out;587 588		ret = -E2BIG;589		if (!valid_arg_len(bprm, len))590			goto out;591 592		/* We're going to work our way backwards. */593		pos = bprm->p;594		str += len;595		bprm->p -= len;596		if (bprm_hit_stack_limit(bprm))597			goto out;598 599		while (len > 0) {600			int offset, bytes_to_copy;601 602			if (fatal_signal_pending(current)) {603				ret = -ERESTARTNOHAND;604				goto out;605			}606			cond_resched();607 608			offset = pos % PAGE_SIZE;609			if (offset == 0)610				offset = PAGE_SIZE;611 612			bytes_to_copy = offset;613			if (bytes_to_copy > len)614				bytes_to_copy = len;615 616			offset -= bytes_to_copy;617			pos -= bytes_to_copy;618			str -= bytes_to_copy;619			len -= bytes_to_copy;620 621			if (!kmapped_page || kpos != (pos & PAGE_MASK)) {622				struct page *page;623 624				page = get_arg_page(bprm, pos, 1);625				if (!page) {626					ret = -E2BIG;627					goto out;628				}629 630				if (kmapped_page) {631					flush_dcache_page(kmapped_page);632					kunmap_local(kaddr);633					put_arg_page(kmapped_page);634				}635				kmapped_page = page;636				kaddr = kmap_local_page(kmapped_page);637				kpos = pos & PAGE_MASK;638				flush_arg_page(bprm, kpos, kmapped_page);639			}640			if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {641				ret = -EFAULT;642				goto out;643			}644		}645	}646	ret = 0;647out:648	if (kmapped_page) {649		flush_dcache_page(kmapped_page);650		kunmap_local(kaddr);651		put_arg_page(kmapped_page);652	}653	return ret;654}655 656/*657 * Copy and argument/environment string from the kernel to the processes stack.658 */659int copy_string_kernel(const char *arg, struct linux_binprm *bprm)660{661	int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;662	unsigned long pos = bprm->p;663 664	if (len == 0)665		return -EFAULT;666	if (!valid_arg_len(bprm, len))667		return -E2BIG;668 669	/* We're going to work our way backwards. */670	arg += len;671	bprm->p -= len;672	if (bprm_hit_stack_limit(bprm))673		return -E2BIG;674 675	while (len > 0) {676		unsigned int bytes_to_copy = min_t(unsigned int, len,677				min_not_zero(offset_in_page(pos), PAGE_SIZE));678		struct page *page;679 680		pos -= bytes_to_copy;681		arg -= bytes_to_copy;682		len -= bytes_to_copy;683 684		page = get_arg_page(bprm, pos, 1);685		if (!page)686			return -E2BIG;687		flush_arg_page(bprm, pos & PAGE_MASK, page);688		memcpy_to_page(page, offset_in_page(pos), arg, bytes_to_copy);689		put_arg_page(page);690	}691 692	return 0;693}694EXPORT_SYMBOL(copy_string_kernel);695 696static HWJS_SUSPENDS int copy_strings_kernel(int argc, const char *const *argv,697			       struct linux_binprm *bprm)698{699	while (argc-- > 0) {700		int ret = copy_string_kernel(argv[argc], bprm);701		if (ret < 0)702			return ret;703		if (fatal_signal_pending(current))704			return -ERESTARTNOHAND;705		cond_resched();706	}707	return 0;708}709 710#ifdef CONFIG_MMU711 712/*713 * Finalizes the stack vm_area_struct. The flags and permissions are updated,714 * the stack is optionally relocated, and some extra space is added.715 */716int setup_arg_pages(struct linux_binprm *bprm,717		    unsigned long stack_top,718		    int executable_stack)719{720	unsigned long ret;721	unsigned long stack_shift;722	struct mm_struct *mm = current->mm;723	struct vm_area_struct *vma = bprm->vma;724	struct vm_area_struct *prev = NULL;725	unsigned long vm_flags;726	unsigned long stack_base;727	unsigned long stack_size;728	unsigned long stack_expand;729	unsigned long rlim_stack;730	struct mmu_gather tlb;731	struct vma_iterator vmi;732 733#ifdef CONFIG_STACK_GROWSUP734	/* Limit stack size */735	stack_base = bprm->rlim_stack.rlim_max;736 737	stack_base = calc_max_stack_size(stack_base);738 739	/* Add space for stack randomization. */740	if (current->flags & PF_RANDOMIZE)741		stack_base += (STACK_RND_MASK << PAGE_SHIFT);742 743	/* Make sure we didn't let the argument array grow too large. */744	if (vma->vm_end - vma->vm_start > stack_base)745		return -ENOMEM;746 747	stack_base = PAGE_ALIGN(stack_top - stack_base);748 749	stack_shift = vma->vm_start - stack_base;750	mm->arg_start = bprm->p - stack_shift;751	bprm->p = vma->vm_end - stack_shift;752#else753	stack_top = arch_align_stack(stack_top);754	stack_top = PAGE_ALIGN(stack_top);755 756	if (unlikely(stack_top < mmap_min_addr) ||757	    unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))758		return -ENOMEM;759 760	stack_shift = vma->vm_end - stack_top;761 762	bprm->p -= stack_shift;763	mm->arg_start = bprm->p;764#endif765 766	if (bprm->loader)767		bprm->loader -= stack_shift;768	bprm->exec -= stack_shift;769 770	if (mmap_write_lock_killable(mm))771		return -EINTR;772 773	vm_flags = VM_STACK_FLAGS;774 775	/*776	 * Adjust stack execute permissions; explicitly enable for777	 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone778	 * (arch default) otherwise.779	 */780	if (unlikely(executable_stack == EXSTACK_ENABLE_X))781		vm_flags |= VM_EXEC;782	else if (executable_stack == EXSTACK_DISABLE_X)783		vm_flags &= ~VM_EXEC;784	vm_flags |= mm->def_flags;785	vm_flags |= VM_STACK_INCOMPLETE_SETUP;786 787	vma_iter_init(&vmi, mm, vma->vm_start);788 789	tlb_gather_mmu(&tlb, mm);790	ret = mprotect_fixup(&vmi, &tlb, vma, &prev, vma->vm_start, vma->vm_end,791			vm_flags);792	tlb_finish_mmu(&tlb);793 794	if (ret)795		goto out_unlock;796	BUG_ON(prev != vma);797 798	if (unlikely(vm_flags & VM_EXEC)) {799		pr_warn_once("process '%pD4' started with executable stack\n",800			     bprm->file);801	}802 803	/* Move stack pages down in memory. */804	if (stack_shift) {805		/*806		 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once807		 * the binfmt code determines where the new stack should reside, we shift it to808		 * its final location.809		 */810		ret = relocate_vma_down(vma, stack_shift);811		if (ret)812			goto out_unlock;813	}814 815	/* mprotect_fixup is overkill to remove the temporary stack flags */816	vm_flags_clear(vma, VM_STACK_INCOMPLETE_SETUP);817 818	stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */819	stack_size = vma->vm_end - vma->vm_start;820	/*821	 * Align this down to a page boundary as expand_stack822	 * will align it up.823	 */824	rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;825 826	stack_expand = min(rlim_stack, stack_size + stack_expand);827 828#ifdef CONFIG_STACK_GROWSUP829	stack_base = vma->vm_start + stack_expand;830#else831	stack_base = vma->vm_end - stack_expand;832#endif833	current->mm->start_stack = bprm->p;834	ret = expand_stack_locked(vma, stack_base);835	if (ret)836		ret = -EFAULT;837 838out_unlock:839	mmap_write_unlock(mm);840	return ret;841}842EXPORT_SYMBOL(setup_arg_pages);843 844#else845 846/*847 * Transfer the program arguments and environment from the holding pages848 * onto the stack. The provided stack pointer is adjusted accordingly.849 */850int transfer_args_to_stack(struct linux_binprm *bprm,851			   unsigned long *sp_location)852{853	unsigned long index, stop, sp;854	int ret = 0;855 856	stop = bprm->p >> PAGE_SHIFT;857	sp = *sp_location;858 859	for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {860		unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;861		char *src = kmap_local_page(bprm->page[index]) + offset;862		sp -= PAGE_SIZE - offset;863		if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)864			ret = -EFAULT;865		kunmap_local(src);866		if (ret)867			goto out;868	}869 870	bprm->exec += *sp_location - MAX_ARG_PAGES * PAGE_SIZE;871	*sp_location = sp;872 873out:874	return ret;875}876EXPORT_SYMBOL(transfer_args_to_stack);877 878#endif /* CONFIG_MMU */879 880/*881 * On success, caller must call do_close_execat() on the returned882 * struct file to close it.883 */884static struct file *do_open_execat(int fd, struct filename *name, int flags)885{886	struct file *file;887	struct open_flags open_exec_flags = {888		.open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,889		.acc_mode = MAY_EXEC,890		.intent = LOOKUP_OPEN,891		.lookup_flags = LOOKUP_FOLLOW,892	};893 894	if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)895		return ERR_PTR(-EINVAL);896	if (flags & AT_SYMLINK_NOFOLLOW)897		open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;898	if (flags & AT_EMPTY_PATH)899		open_exec_flags.lookup_flags |= LOOKUP_EMPTY;900 901	file = do_filp_open(fd, name, &open_exec_flags);902	if (IS_ERR(file))903		return file;904 905	/*906	 * In the past the regular type check was here. It moved to may_open() in907	 * 633fb6ac3980 ("exec: move S_ISREG() check earlier"). Since then it is908	 * an invariant that all non-regular files error out before we get here.909	 */910	if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode)) ||911	    path_noexec(&file->f_path)) {912		fput(file);913		return ERR_PTR(-EACCES);914	}915 916	return file;917}918 919/**920 * open_exec - Open a path name for execution921 *922 * @name: path name to open with the intent of executing it.923 *924 * Returns ERR_PTR on failure or allocated struct file on success.925 *926 * As this is a wrapper for the internal do_open_execat(). Also see927 * do_close_execat().928 */929struct file *open_exec(const char *name)930{931	struct filename *filename = getname_kernel(name);932	struct file *f = ERR_CAST(filename);933 934	if (!IS_ERR(filename)) {935		f = do_open_execat(AT_FDCWD, filename, 0);936		putname(filename);937	}938	return f;939}940EXPORT_SYMBOL(open_exec);941 942#if defined(CONFIG_BINFMT_FLAT) || defined(CONFIG_BINFMT_ELF_FDPIC)943ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)944{945	ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);946	if (res > 0)947		flush_icache_user_range(addr, addr + len);948	return res;949}950EXPORT_SYMBOL(read_code);951#endif952 953/*954 * Maps the mm_struct mm into the current task struct.955 * On success, this function returns with exec_update_lock956 * held for writing.957 */958static int exec_mmap(struct mm_struct *mm)959{960	struct task_struct *tsk;961	struct mm_struct *old_mm, *active_mm;962	int ret;963 964	/* Notify parent that we're no longer interested in the old VM */965	tsk = current;966	old_mm = current->mm;967	exec_mm_release(tsk, old_mm);968 969	ret = down_write_killable(&tsk->signal->exec_update_lock);970	if (ret)971		return ret;972 973	if (old_mm) {974		/*975		 * If there is a pending fatal signal perhaps a signal976		 * whose default action is to create a coredump get977		 * out and die instead of going through with the exec.978		 */979		ret = mmap_read_lock_killable(old_mm);980		if (ret) {981			up_write(&tsk->signal->exec_update_lock);982			return ret;983		}984	}985 986	task_lock(tsk);987	membarrier_exec_mmap(mm);988 989	local_irq_disable();990	active_mm = tsk->active_mm;991	tsk->active_mm = mm;992	tsk->mm = mm;993	mm_init_cid(mm);994	/*995	 * This prevents preemption while active_mm is being loaded and996	 * it and mm are being updated, which could cause problems for997	 * lazy tlb mm refcounting when these are updated by context998	 * switches. Not all architectures can handle irqs off over999	 * activate_mm yet.1000	 */1001	if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))1002		local_irq_enable();1003	activate_mm(active_mm, mm);1004	if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))1005		local_irq_enable();1006	lru_gen_add_mm(mm);1007	task_unlock(tsk);1008	lru_gen_use_mm(mm);1009	if (old_mm) {1010		mmap_read_unlock(old_mm);1011		BUG_ON(active_mm != old_mm);1012		setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);1013		mm_update_next_owner(old_mm);1014		mmput(old_mm);1015		return 0;1016	}1017	mmdrop_lazy_tlb(active_mm);1018	return 0;1019}1020 1021static HWJS_SUSPENDS int de_thread(struct task_struct *tsk)1022{1023	struct signal_struct *sig = tsk->signal;1024	struct sighand_struct *oldsighand = tsk->sighand;1025	spinlock_t *lock = &oldsighand->siglock;1026 1027	if (thread_group_empty(tsk))1028		goto no_thread_group;1029 1030	/*1031	 * Kill all other threads in the thread group.1032	 */1033	spin_lock_irq(lock);1034	if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) {1035		/*1036		 * Another group action in progress, just1037		 * return so that the signal is processed.1038		 */1039		spin_unlock_irq(lock);1040		return -EAGAIN;1041	}1042 1043	sig->group_exec_task = tsk;1044	sig->notify_count = zap_other_threads(tsk);1045	if (!thread_group_leader(tsk))1046		sig->notify_count--;1047 1048	while (sig->notify_count) {1049		__set_current_state(TASK_KILLABLE);1050		spin_unlock_irq(lock);1051		schedule();1052		if (__fatal_signal_pending(tsk))1053			goto killed;1054		spin_lock_irq(lock);1055	}1056	spin_unlock_irq(lock);1057 1058	/*1059	 * At this point all other threads have exited, all we have to1060	 * do is to wait for the thread group leader to become inactive,1061	 * and to assume its PID:1062	 */1063	if (!thread_group_leader(tsk)) {1064		struct task_struct *leader = tsk->group_leader;1065 1066		for (;;) {1067			cgroup_threadgroup_change_begin(tsk);1068			write_lock_irq(&tasklist_lock);1069			/*1070			 * Do this under tasklist_lock to ensure that1071			 * exit_notify() can't miss ->group_exec_task1072			 */1073			sig->notify_count = -1;1074			if (likely(leader->exit_state))1075				break;1076			__set_current_state(TASK_KILLABLE);1077			write_unlock_irq(&tasklist_lock);1078			cgroup_threadgroup_change_end(tsk);1079			schedule();1080			if (__fatal_signal_pending(tsk))1081				goto killed;1082		}1083 1084		/*1085		 * The only record we have of the real-time age of a1086		 * process, regardless of execs it's done, is start_time.1087		 * All the past CPU time is accumulated in signal_struct1088		 * from sister threads now dead.  But in this non-leader1089		 * exec, nothing survives from the original leader thread,1090		 * whose birth marks the true age of this process now.1091		 * When we take on its identity by switching to its PID, we1092		 * also take its birthdate (always earlier than our own).1093		 */1094		tsk->start_time = leader->start_time;1095		tsk->start_boottime = leader->start_boottime;1096 1097		BUG_ON(!same_thread_group(leader, tsk));1098		/*1099		 * An exec() starts a new thread group with the1100		 * TGID of the previous thread group. Rehash the1101		 * two threads with a switched PID, and release1102		 * the former thread group leader:1103		 */1104 1105		/* Become a process group leader with the old leader's pid.1106		 * The old leader becomes a thread of the this thread group.1107		 */1108		exchange_tids(tsk, leader);1109		transfer_pid(leader, tsk, PIDTYPE_TGID);1110		transfer_pid(leader, tsk, PIDTYPE_PGID);1111		transfer_pid(leader, tsk, PIDTYPE_SID);1112 1113		list_replace_rcu(&leader->tasks, &tsk->tasks);1114		list_replace_init(&leader->sibling, &tsk->sibling);1115 1116		tsk->group_leader = tsk;1117		leader->group_leader = tsk;1118 1119		tsk->exit_signal = SIGCHLD;1120		leader->exit_signal = -1;1121 1122		BUG_ON(leader->exit_state != EXIT_ZOMBIE);1123		leader->exit_state = EXIT_DEAD;1124		/*1125		 * We are going to release_task()->ptrace_unlink() silently,1126		 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees1127		 * the tracer won't block again waiting for this thread.1128		 */1129		if (unlikely(leader->ptrace))1130			__wake_up_parent(leader, leader->parent);1131		write_unlock_irq(&tasklist_lock);1132		cgroup_threadgroup_change_end(tsk);1133 1134		release_task(leader);1135	}1136 1137	sig->group_exec_task = NULL;1138	sig->notify_count = 0;1139 1140no_thread_group:1141	/* we have changed execution domain */1142	tsk->exit_signal = SIGCHLD;1143 1144	BUG_ON(!thread_group_leader(tsk));1145	return 0;1146 1147killed:1148	/* protects against exit_notify() and __exit_signal() */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 process has its own signal table,1159 * so that flush_signal_handlers can later reset the handlers without1160 * disturbing other processes.  (Other processes might share the signal1161 * table via the CLONE_SIGHAND option to clone().)1162 */1163static int unshare_sighand(struct task_struct *me)1164{1165	struct sighand_struct *oldsighand = me->sighand;1166 1167	if (refcount_read(&oldsighand->count) != 1) {1168		struct sighand_struct *newsighand;1169		/*1170		 * This ->sighand is shared with the CLONE_SIGHAND1171		 * but not CLONE_THREAD task, switch to the new one.1172		 */1173		newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);1174		if (!newsighand)1175			return -ENOMEM;1176 1177		refcount_set(&newsighand->count, 1);1178 1179		write_lock_irq(&tasklist_lock);1180		spin_lock(&oldsighand->siglock);1181		memcpy(newsighand->action, oldsighand->action,1182		       sizeof(newsighand->action));1183		rcu_assign_pointer(me->sighand, newsighand);1184		spin_unlock(&oldsighand->siglock);1185		write_unlock_irq(&tasklist_lock);1186 1187		__cleanup_sighand(oldsighand);1188	}1189	return 0;1190}1191 1192char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)1193{1194	task_lock(tsk);1195	/* Always NUL terminated and zero-padded */1196	strscpy_pad(buf, tsk->comm, buf_size);1197	task_unlock(tsk);1198	return buf;1199}1200EXPORT_SYMBOL_GPL(__get_task_comm);1201 1202/*1203 * These functions flushes out all traces of the currently running executable1204 * so that a new one can be started1205 */1206 1207void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)1208{1209	task_lock(tsk);1210	trace_task_rename(tsk, buf);1211	strscpy_pad(tsk->comm, buf, sizeof(tsk->comm));1212	task_unlock(tsk);1213	perf_event_comm(tsk, exec);1214}1215 1216/*1217 * Calling this is the point of no return. None of the failures will be1218 * seen by userspace since either the process is already taking a fatal1219 * signal (via de_thread() or coredump), or will have SEGV raised1220 * (after exec_mmap()) by search_binary_handler (see below).1221 */1222int begin_new_exec(struct linux_binprm * bprm)1223{1224	struct task_struct *me = current;1225	int retval;1226 1227	/* Once we are committed compute the creds */1228	retval = bprm_creds_from_file(bprm);1229	if (retval)1230		return retval;1231 1232	/*1233	 * This tracepoint marks the point before flushing the old exec where1234	 * the current task is still unchanged, but errors are fatal (point of1235	 * no return). The later "sched_process_exec" tracepoint is called after1236	 * the current task has successfully switched to the new exec.1237	 */1238	trace_sched_prepare_exec(current, bprm);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 thread group.1247	 */1248	retval = de_thread(me);1249	if (retval)1250		goto out;1251 1252	/*1253	 * Cancel any io_uring activity across execve1254	 */1255	io_uring_task_cancel();1256 1257	/* Ensure the files table is not shared. */1258	retval = unshare_files();1259	if (retval)1260		goto out;1261 1262	/*1263	 * Must be called _before_ exec_mmap() as bprm->mm is1264	 * not visible until then. Doing it here also ensures1265	 * we don't race against replace_mm_exe_file().1266	 */1267	retval = set_mm_exe_file(bprm->mm, bprm->file);1268	if (retval)1269		goto out;1270 1271	/* If the binary is not readable then enforce mm->dumpable=0 */1272	would_dump(bprm, bprm->file);1273	if (bprm->have_execfd)1274		would_dump(bprm, bprm->executable);1275 1276	/*1277	 * Release all of the old mmap stuff1278	 */1279	acct_arg_size(bprm, 0);1280	retval = exec_mmap(bprm->mm);1281	if (retval)1282		goto out;1283 1284	bprm->mm = NULL;1285 1286	retval = exec_task_namespaces();1287	if (retval)1288		goto out_unlock;1289 1290#ifdef CONFIG_POSIX_TIMERS1291	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#endif1297 1298	/*1299	 * Make the signal table private.1300	 */1301	retval = unshare_sighand(me);1302	if (retval)1303		goto out_unlock;1304 1305	me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC |1306					PF_NOFREEZE | PF_NO_SETAFFINITY);1307	flush_thread();1308	me->personality &= ~bprm->per_clear;1309 1310	clear_syscall_work_syscall_user_dispatch(me);1311 1312	/*1313	 * We have to apply CLOEXEC before we change whether the process is1314	 * dumpable (in setup_new_exec) to avoid a race with a process in userspace1315	 * trying to access the should-be-closed file descriptors of a process1316	 * undergoing exec(2).1317	 */1318	do_close_on_exec(me->files);1319 1320	if (bprm->secureexec) {1321		/* Make sure parent cannot signal privileged process. */1322		me->pdeath_signal = 0;1323 1324		/*1325		 * For secureexec, reset the stack limit to sane default to1326		 * avoid bad behavior from the prior rlimits. This has to1327		 * happen before arch_pick_mmap_layout(), which examines1328		 * RLIMIT_STACK, but after the point of no return to avoid1329		 * needing to clean up the change on failure.1330		 */1331		if (bprm->rlim_stack.rlim_cur > _STK_LIM)1332			bprm->rlim_stack.rlim_cur = _STK_LIM;1333	}1334 1335	me->sas_ss_sp = me->sas_ss_size = 0;1336 1337	/*1338	 * Figure out dumpability. Note that this checking only of current1339	 * is wrong, but userspace depends on it. This should be testing1340	 * bprm->secureexec instead.1341	 */1342	if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||1343	    !(uid_eq(current_euid(), current_uid()) &&1344	      gid_eq(current_egid(), current_gid())))1345		set_dumpable(current->mm, suid_dumpable);1346	else1347		set_dumpable(current->mm, SUID_DUMP_USER);1348 1349	perf_event_exec();1350	__set_task_comm(me, kbasename(bprm->filename), true);1351 1352	/* An exec changes our domain. We are no longer part of the thread1353	   group */1354	WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);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 this executable1363	 */1364	security_bprm_committing_creds(bprm);1365 1366	commit_creds(bprm->cred);1367	bprm->cred = NULL;1368 1369	/*1370	 * Disable monitoring for regular users1371	 * when executing setuid binaries. Must1372	 * wait until new credentials are committed1373	 * by commit_creds() above1374	 */1375	if (get_dumpable(me->mm) != SUID_DUMP_USER)1376		perf_event_exit_task(me);1377	/*1378	 * cred_guard_mutex must be held at least to this point to prevent1379	 * ptrace_attach() from altering our determination of the task's1380	 * credentials; any time after this it may be unlocked.1381	 */1382	security_bprm_committed_creds(bprm);1383 1384	/* Pass the opened binary to the interpreter. */1385	if (bprm->have_execfd) {1386		retval = get_unused_fd_flags(0);1387		if (retval < 0)1388			goto out_unlock;1389		fd_install(retval, bprm->executable);1390		bprm->executable = NULL;1391		bprm->execfd = retval;1392	}1393	return 0;1394 1395out_unlock:1396	up_write(&me->signal->exec_update_lock);1397	if (!bprm->cred)1398		mutex_unlock(&me->signal->cred_guard_mutex);1399 1400out:1401	return retval;1402}1403EXPORT_SYMBOL(begin_new_exec);1404 1405void would_dump(struct linux_binprm *bprm, struct file *file)1406{1407	struct inode *inode = file_inode(file);1408	struct mnt_idmap *idmap = file_mnt_idmap(file);1409	if (inode_permission(idmap, inode, MAY_READ) < 0) {1410		struct user_namespace *old, *user_ns;1411		bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;1412 1413		/* Ensure mm->user_ns contains the executable */1414		user_ns = old = bprm->mm->user_ns;1415		while ((user_ns != &init_user_ns) &&1416		       !privileged_wrt_inode_uidgid(user_ns, idmap, inode))1417			user_ns = user_ns->parent;1418 1419		if (old != user_ns) {1420			bprm->mm->user_ns = get_user_ns(user_ns);1421			put_user_ns(old);1422		}1423	}1424}1425EXPORT_SYMBOL(would_dump);1426 1427void setup_new_exec(struct linux_binprm * bprm)1428{1429	/* Setup things that can depend upon the personality */1430	struct task_struct *me = current;1431 1432	arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);1433 1434	arch_setup_new_exec();1435 1436	/* Set the new mm task size. We have to do that late because it may1437	 * depend on TIF_32BIT which is only updated in flush_thread() on1438	 * some architectures like powerpc1439	 */1440	me->mm->task_size = TASK_SIZE;1441	up_write(&me->signal->exec_update_lock);1442	mutex_unlock(&me->signal->cred_guard_mutex);1443}1444EXPORT_SYMBOL(setup_new_exec);1445 1446/* Runs immediately before start_thread() takes over. */1447void finalize_exec(struct linux_binprm *bprm)1448{1449	/* Store any stack rlimit changes before starting thread. */1450	task_lock(current->group_leader);1451	current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;1452	task_unlock(current->group_leader);1453}1454EXPORT_SYMBOL(finalize_exec);1455 1456/*1457 * Prepare credentials and lock ->cred_guard_mutex.1458 * setup_new_exec() commits the new creds and drops the lock.1459 * Or, if exec fails before, free_bprm() should release ->cred1460 * and unlock.1461 */1462static HWJS_SUSPENDS int prepare_bprm_creds(struct linux_binprm *bprm)1463{1464	if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))1465		return -ERESTARTNOINTR;1466 1467	bprm->cred = prepare_exec_creds();1468	if (likely(bprm->cred))1469		return 0;1470 1471	mutex_unlock(&current->signal->cred_guard_mutex);1472	return -ENOMEM;1473}1474 1475/* Matches do_open_execat() */1476static void do_close_execat(struct file *file)1477{1478	if (file)1479		fput(file);1480}1481 1482static HWJS_SUSPENDS void free_bprm(struct linux_binprm *bprm)1483{1484	if (bprm->mm) {1485		acct_arg_size(bprm, 0);1486		mmput(bprm->mm);1487	}1488	free_arg_pages(bprm);1489	if (bprm->cred) {1490		mutex_unlock(&current->signal->cred_guard_mutex);1491		abort_creds(bprm->cred);1492	}1493	do_close_execat(bprm->file);1494	if (bprm->executable)1495		fput(bprm->executable);1496	/* If a binfmt changed the interp, free it. */1497	if (bprm->interp != bprm->filename)1498		kfree(bprm->interp);1499	kfree(bprm->fdpath);1500	kfree(bprm);1501}1502 1503static HWJS_SUSPENDS struct linux_binprm *alloc_bprm(int fd, struct filename *filename, int flags)1504{1505	struct linux_binprm *bprm;1506	struct file *file;1507	int retval = -ENOMEM;1508 1509	file = do_open_execat(fd, filename, flags);1510	if (IS_ERR(file))1511		return ERR_CAST(file);1512 1513	bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);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[0] == '/') {1522		bprm->filename = filename->name;1523	} else {1524		if (filename->name[0] == '\0')1525			bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);1526		else1527			bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",1528						  fd, filename->name);1529		if (!bprm->fdpath)1530			goto out_free;1531 1532		/*1533		 * Record that a name derived from an O_CLOEXEC fd will be1534		 * inaccessible after exec.  This allows the code in exec to1535		 * choose to fail when the executable is not mmaped into the1536		 * interpreter and an open file descriptor is not passed to1537		 * the interpreter.  This makes for a better user experience1538		 * than having the interpreter start and then immediately fail1539		 * when it finds the executable is inaccessible.1540		 */1541		if (get_close_on_exec(fd))1542			bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;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 1552out_free:1553	free_bprm(bprm);1554	return ERR_PTR(retval);1555}1556 1557int bprm_change_interp(const char *interp, struct linux_binprm *bprm)1558{1559	/* If a binfmt changed the interp, free it first. */1560	if (bprm->interp != bprm->filename)1561		kfree(bprm->interp);1562	bprm->interp = kstrdup(interp, GFP_KERNEL);1563	if (!bprm->interp)1564		return -ENOMEM;1565	return 0;1566}1567EXPORT_SYMBOL(bprm_change_interp);1568 1569/*1570 * determine how safe it is to execute the proposed program1571 * - the caller must hold ->cred_guard_mutex to protect against1572 *   PTRACE_ATTACH or seccomp thread-sync1573 */1574static void check_unsafe_exec(struct linux_binprm *bprm)1575{1576	struct task_struct *p = current, *t;1577	unsigned n_fs;1578 1579	if (p->ptrace)1580		bprm->unsafe |= LSM_UNSAFE_PTRACE;1581 1582	/*1583	 * This isn't strictly necessary, but it makes it harder for LSMs to1584	 * mess up.1585	 */1586	if (task_no_new_privs(current))1587		bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;1588 1589	/*1590	 * If another task is sharing our fs, we cannot safely1591	 * suid exec because the differently privileged task1592	 * will be able to manipulate the current directory, etc.1593	 * It would be nice to force an unshare instead...1594	 */1595	n_fs = 1;1596	spin_lock(&p->fs->lock);1597	rcu_read_lock();1598	for_other_threads(p, t) {1599		if (t->fs == p->fs)1600			n_fs++;1601	}1602	rcu_read_unlock();1603 1604	/* "users" and "in_exec" locked for copy_fs() */1605	if (p->fs->users > n_fs)1606		bprm->unsafe |= LSM_UNSAFE_SHARE;1607	else1608		p->fs->in_exec = 1;1609	spin_unlock(&p->fs->lock);1610}1611 1612static HWJS_SUSPENDS void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)1613{1614	/* Handle suid and sgid on files */1615	struct mnt_idmap *idmap;1616	struct inode *inode = file_inode(file);1617	unsigned int mode;1618	vfsuid_t vfsuid;1619	vfsgid_t vfsgid;1620	int err;1621 1622	if (!mnt_may_suid(file->f_path.mnt))1623		return;1624 1625	if (task_no_new_privs(current))1626		return;1627 1628	mode = READ_ONCE(inode->i_mode);1629	if (!(mode & (S_ISUID|S_ISGID)))1630		return;1631 1632	idmap = file_mnt_idmap(file);1633 1634	/* Be careful if suid/sgid is set */1635	inode_lock(inode);1636 1637	/* Atomically reload and check mode/uid/gid now that lock held. */1638	mode = inode->i_mode;1639	vfsuid = i_uid_into_vfsuid(idmap, inode);1640	vfsgid = i_gid_into_vfsgid(idmap, inode);1641	err = inode_permission(idmap, inode, MAY_EXEC);1642	inode_unlock(inode);1643 1644	/* Did the exec bit vanish out from under us? Give up. */1645	if (err)1646		return;1647 1648	/* We ignore suid/sgid if there are no mappings for them in the ns */1649	if (!vfsuid_has_mapping(bprm->cred->user_ns, vfsuid) ||1650	    !vfsgid_has_mapping(bprm->cred->user_ns, vfsgid))1651		return;1652 1653	if (mode & S_ISUID) {1654		bprm->per_clear |= PER_CLEAR_ON_SETID;1655		bprm->cred->euid = vfsuid_into_kuid(vfsuid);1656	}1657 1658	if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {1659		bprm->per_clear |= PER_CLEAR_ON_SETID;1660		bprm->cred->egid = vfsgid_into_kgid(vfsgid);1661	}1662}1663 1664/*1665 * Compute brpm->cred based upon the final binary.1666 */1667static HWJS_SUSPENDS int bprm_creds_from_file(struct linux_binprm *bprm)1668{1669	/* Compute creds based on which file? */1670	struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;1671 1672	bprm_fill_uid(bprm, file);1673	return security_bprm_creds_from_file(bprm, file);1674}1675 1676/*1677 * Fill the binprm structure from the inode.1678 * Read the first BINPRM_BUF_SIZE bytes1679 *1680 * This may be called multiple times for binary chains (scripts for example).1681 */1682static int prepare_binprm(struct linux_binprm *bprm)1683{1684	loff_t pos = 0;1685 1686	memset(bprm->buf, 0, BINPRM_BUF_SIZE);1687	return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);1688}1689 1690/*1691 * Arguments are '\0' separated strings found at the location bprm->p1692 * points to; chop off the first by relocating brpm->p to right after1693 * the first '\0' encountered.1694 */1695int remove_arg_zero(struct linux_binprm *bprm)1696{1697	unsigned long offset;1698	char *kaddr;1699	struct page *page;1700 1701	if (!bprm->argc)1702		return 0;1703 1704	do {1705		offset = bprm->p & ~PAGE_MASK;1706		page = get_arg_page(bprm, bprm->p, 0);1707		if (!page)1708			return -EFAULT;1709		kaddr = kmap_local_page(page);1710 1711		for (; offset < PAGE_SIZE && kaddr[offset];1712				offset++, bprm->p++)1713			;1714 1715		kunmap_local(kaddr);1716		put_arg_page(page);1717	} while (offset == PAGE_SIZE);1718 1719	bprm->p++;1720	bprm->argc--;1721 1722	return 0;1723}1724EXPORT_SYMBOL(remove_arg_zero);1725 1726#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))1727/*1728 * cycle the list of binary formats handler, until one recognizes the image1729 */1730static int search_binary_handler(struct linux_binprm *bprm)1731{1732	bool need_retry = IS_ENABLED(CONFIG_MODULES);1733	struct linux_binfmt *fmt;1734	int retval;1735 1736	retval = prepare_binprm(bprm);1737	if (retval < 0)1738		return retval;1739 1740	retval = security_bprm_check(bprm);1741	if (retval)1742		return retval;1743 1744	retval = -ENOENT;1745 retry:1746	read_lock(&binfmt_lock);1747	list_for_each_entry(fmt, &formats, lh) {1748		if (!try_module_get(fmt->module))1749			continue;1750		read_unlock(&binfmt_lock);1751 1752		retval = fmt->load_binary(bprm);1753 1754		read_lock(&binfmt_lock);1755		put_binfmt(fmt);1756		if (bprm->point_of_no_return || (retval != -ENOEXEC)) {1757			read_unlock(&binfmt_lock);1758			return retval;1759		}1760	}1761	read_unlock(&binfmt_lock);1762 1763	if (need_retry) {1764		if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&1765		    printable(bprm->buf[2]) && printable(bprm->buf[3]))1766			return retval;1767		if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)1768			return retval;1769		need_retry = false;1770		goto retry;1771	}1772 1773	return retval;1774}1775 1776/* arch hook; see end of file for the weak default + patch 0005 header */1777void arch_post_execve(struct linux_binprm *bprm);1778 1779/* binfmt handlers will call back into begin_new_exec() on success. */1780static HWJS_SUSPENDS int exec_binprm(struct linux_binprm *bprm)1781{1782	pid_t old_pid, old_vpid;1783	int ret, depth;1784 1785	/* Need to fetch pid before load_binary changes it */1786	old_pid = current->pid;1787	rcu_read_lock();1788	old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));1789	rcu_read_unlock();1790 1791	/* This allows 4 levels of binfmt rewrites before failing hard. */1792	for (depth = 0;; depth++) {1793		struct file *exec;1794		if (depth > 5)1795			return -ELOOP;1796 1797		ret = search_binary_handler(bprm);1798		if (ret < 0)1799			return ret;1800		if (!bprm->interpreter)1801			break;1802 1803		exec = bprm->file;1804		bprm->file = bprm->interpreter;1805		bprm->interpreter = NULL;1806 1807		if (unlikely(bprm->have_execfd)) {1808			if (bprm->executable) {1809				fput(exec);1810				return -ENOEXEC;1811			}1812			bprm->executable = exec;1813		} else1814			fput(exec);1815	}1816 1817	audit_bprm(bprm);1818	trace_sched_process_exec(current, old_pid, bprm);1819	ptrace_event(PTRACE_EVENT_EXEC, old_vpid);1820	proc_exec_connector(current);1821	return 0;1822}1823 1824static HWJS_SUSPENDS int bprm_execve(struct linux_binprm *bprm)1825{1826	int retval;1827 1828	retval = prepare_bprm_creds(bprm);1829	if (retval)1830		return retval;1831 1832	/*1833	 * Check for unsafe execution states before exec_binprm(), which1834	 * will call back into begin_new_exec(), into bprm_creds_from_file(),1835	 * where setuid-ness is evaluated.1836	 */1837	check_unsafe_exec(bprm);1838	current->in_execve = 1;1839	sched_mm_cid_before_execve(current);1840 1841	sched_exec();1842 1843	/* Set the unchanging part of bprm->cred */1844	retval = security_bprm_creds_for_exec(bprm);1845	if (retval)1846		goto out;1847 1848	retval = exec_binprm(bprm);1849	if (retval < 0)1850		goto out;1851 1852	sched_mm_cid_after_execve(current);1853	/* execve succeeded */1854	current->fs->in_exec = 0;1855	current->in_execve = 0;1856	rseq_execve(current);1857	user_events_execve(current);1858	acct_update_integrals(current);1859	task_numa_free(current, false);1860	arch_post_execve(bprm);1861	return retval;1862 1863out:1864	/*1865	 * If past the point of no return ensure the code never1866	 * returns to the userspace process.  Use an existing fatal1867	 * signal if present otherwise terminate the process with1868	 * SIGSEGV.1869	 */1870	if (bprm->point_of_no_return && !fatal_signal_pending(current))1871		force_fatal_sig(SIGSEGV);1872 1873	sched_mm_cid_after_execve(current);1874	current->fs->in_exec = 0;1875	current->in_execve = 0;1876 1877	return retval;1878}1879 1880static HWJS_SUSPENDS int do_execveat_common(int fd, struct filename *filename,1881			      struct user_arg_ptr argv,1882			      struct user_arg_ptr envp,1883			      int flags)1884{1885	struct linux_binprm *bprm;1886	int retval;1887 1888	if (IS_ERR(filename))1889		return PTR_ERR(filename);1890 1891	/*1892	 * We move the actual failure in case of RLIMIT_NPROC excess from1893	 * set*uid() to execve() because too many poorly written programs1894	 * don't check setuid() return code.  Here we additionally recheck1895	 * whether NPROC limit is still exceeded.1896	 */1897	if ((current->flags & PF_NPROC_EXCEEDED) &&1898	    is_rlimit_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {1899		retval = -EAGAIN;1900		goto out_ret;1901	}1902 1903	/* We're below the limit (still or again), so we don't want to make1904	 * further execve() calls fail. */1905	current->flags &= ~PF_NPROC_EXCEEDED;1906 1907	bprm = alloc_bprm(fd, filename, flags);1908	if (IS_ERR(bprm)) {1909		retval = PTR_ERR(bprm);1910		goto out_ret;1911	}1912 1913	retval = count(argv, MAX_ARG_STRINGS);1914	if (retval == 0)1915		pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",1916			     current->comm, bprm->filename);1917	if (retval < 0)1918		goto out_free;1919	bprm->argc = retval;1920 1921	retval = count(envp, MAX_ARG_STRINGS);1922	if (retval < 0)1923		goto out_free;1924	bprm->envc = retval;1925 1926	retval = bprm_stack_limits(bprm);1927	if (retval < 0)1928		goto out_free;1929 1930	retval = copy_string_kernel(bprm->filename, bprm);1931	if (retval < 0)1932		goto out_free;1933	bprm->exec = bprm->p;1934 1935	retval = copy_strings(bprm->envc, envp, bprm);1936	if (retval < 0)1937		goto out_free;1938 1939	retval = copy_strings(bprm->argc, argv, bprm);1940	if (retval < 0)1941		goto out_free;1942 1943	/*1944	 * When argv is empty, add an empty string ("") as argv[0] to1945	 * ensure confused userspace programs that start processing1946	 * from argv[1] won't end up walking envp. See also1947	 * bprm_stack_limits().1948	 */1949	if (bprm->argc == 0) {1950		retval = copy_string_kernel("", bprm);1951		if (retval < 0)1952			goto out_free;1953		bprm->argc = 1;1954	}1955 1956	retval = bprm_execve(bprm);1957out_free:1958	free_bprm(bprm);1959 1960out_ret:1961	putname(filename);1962	return retval;1963}1964 1965int kernel_execve(const char *kernel_filename,1966		  const char *const *argv, const char *const *envp)1967{1968	struct filename *filename;1969	struct linux_binprm *bprm;1970	int fd = AT_FDCWD;1971	int retval;1972 1973	/* It is non-sense for kernel threads to call execve */1974	if (WARN_ON_ONCE(current->flags & PF_KTHREAD))1975		return -EINVAL;1976 1977	filename = getname_kernel(kernel_filename);1978	if (IS_ERR(filename))1979		return PTR_ERR(filename);1980 1981	bprm = alloc_bprm(fd, filename, 0);1982	if (IS_ERR(bprm)) {1983		retval = PTR_ERR(bprm);1984		goto out_ret;1985	}1986 1987	retval = count_strings_kernel(argv);1988	if (WARN_ON_ONCE(retval == 0))1989		retval = -EINVAL;1990	if (retval < 0)1991		goto out_free;1992	bprm->argc = retval;1993 1994	retval = count_strings_kernel(envp);1995	if (retval < 0)1996		goto out_free;1997	bprm->envc = retval;1998 1999	retval = bprm_stack_limits(bprm);2000	if (retval < 0)2001		goto out_free;2002 2003	retval = copy_string_kernel(bprm->filename, bprm);2004	if (retval < 0)2005		goto out_free;2006	bprm->exec = bprm->p;2007 2008	retval = copy_strings_kernel(bprm->envc, envp, bprm);2009	if (retval < 0)2010		goto out_free;2011 2012	retval = copy_strings_kernel(bprm->argc, argv, bprm);2013	if (retval < 0)2014		goto out_free;2015 2016	retval = bprm_execve(bprm);2017out_free:2018	free_bprm(bprm);2019out_ret:2020	putname(filename);2021	return retval;2022}2023 2024static HWJS_SUSPENDS int do_execve(struct filename *filename,2025	const char __user *const __user *__argv,2026	const char __user *const __user *__envp)2027{2028	struct user_arg_ptr argv = { .ptr.native = __argv };2029	struct user_arg_ptr envp = { .ptr.native = __envp };2030	return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);2031}2032 2033static HWJS_SUSPENDS int do_execveat(int fd, struct filename *filename,2034		const char __user *const __user *__argv,2035		const char __user *const __user *__envp,2036		int flags)2037{2038	struct user_arg_ptr argv = { .ptr.native = __argv };2039	struct user_arg_ptr envp = { .ptr.native = __envp };2040 2041	return do_execveat_common(fd, filename, argv, envp, flags);2042}2043 2044#ifdef CONFIG_COMPAT2045static int compat_do_execve(struct filename *filename,2046	const compat_uptr_t __user *__argv,2047	const compat_uptr_t __user *__envp)2048{2049	struct user_arg_ptr argv = {2050		.is_compat = true,2051		.ptr.compat = __argv,2052	};2053	struct user_arg_ptr envp = {2054		.is_compat = true,2055		.ptr.compat = __envp,2056	};2057	return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);2058}2059 2060static int compat_do_execveat(int fd, struct filename *filename,2061			      const compat_uptr_t __user *__argv,2062			      const compat_uptr_t __user *__envp,2063			      int flags)2064{2065	struct user_arg_ptr argv = {2066		.is_compat = true,2067		.ptr.compat = __argv,2068	};2069	struct user_arg_ptr envp = {2070		.is_compat = true,2071		.ptr.compat = __envp,2072	};2073	return do_execveat_common(fd, filename, argv, envp, flags);2074}2075#endif2076 2077void set_binfmt(struct linux_binfmt *new)2078{2079	struct mm_struct *mm = current->mm;2080 2081	if (mm->binfmt)2082		module_put(mm->binfmt->module);2083 2084	mm->binfmt = new;2085	if (new)2086		__module_get(new->module);2087}2088EXPORT_SYMBOL(set_binfmt);2089 2090/*2091 * arch_post_execve - arch-specific hook called at the tail of2092 * bprm_execve's success path, after the standard post-execve2093 * teardown has run. bprm->binfmt is the loader that matched.2094 *2095 * Default is no-op. Arches that need to react to a specific2096 * binfmt match — e.g. wasm32 parks the kthread when binfmt_wasm2097 * transferred execution to a user Worker — override the weak2098 * default. See Documentation/wasm/rootfs-model.rst §4 R7.2099 */2100void __weak arch_post_execve(struct linux_binprm *bprm) { }2101 2102/*2103 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.2104 */2105void set_dumpable(struct mm_struct *mm, int value)2106{2107	if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))2108		return;2109 2110	set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);2111}2112 2113SYSCALL_DEFINE3(execve,2114		const char __user *, filename,2115		const char __user *const __user *, argv,2116		const char __user *const __user *, envp)2117{2118	return do_execve(getname(filename), argv, envp);2119}2120 2121SYSCALL_DEFINE5(execveat,2122		int, fd, const char __user *, filename,2123		const char __user *const __user *, argv,2124		const char __user *const __user *, envp,2125		int, flags)2126{2127	return do_execveat(fd,2128			   getname_uflags(filename, flags),2129			   argv, envp, flags);2130}2131 2132#ifdef CONFIG_COMPAT2133COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,2134	const compat_uptr_t __user *, argv,2135	const compat_uptr_t __user *, envp)2136{2137	return compat_do_execve(getname(filename), argv, envp);2138}2139 2140COMPAT_SYSCALL_DEFINE5(execveat, int, fd,2141		       const char __user *, filename,2142		       const compat_uptr_t __user *, argv,2143		       const compat_uptr_t __user *, envp,2144		       int,  flags)2145{2146	return compat_do_execveat(fd,2147				  getname_uflags(filename, flags),2148				  argv, envp, flags);2149}2150#endif2151 2152#ifdef CONFIG_SYSCTL2153 2154static int proc_dointvec_minmax_coredump(const struct ctl_table *table, int write,2155		void *buffer, size_t *lenp, loff_t *ppos)2156{2157	int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);2158 2159	if (!error)2160		validate_coredump_safety();2161	return error;2162}2163 2164static struct ctl_table fs_exec_sysctls[] = {2165	{2166		.procname	= "suid_dumpable",2167		.data		= &suid_dumpable,2168		.maxlen		= sizeof(int),2169		.mode		= 0644,2170		.proc_handler	= proc_dointvec_minmax_coredump,2171		.extra1		= SYSCTL_ZERO,2172		.extra2		= SYSCTL_TWO,2173	},2174};2175 2176static int __init init_fs_exec_sysctls(void)2177{2178	register_sysctl_init("fs", fs_exec_sysctls);2179	return 0;2180}2181 2182fs_initcall(init_fs_exec_sysctls);2183#endif /* CONFIG_SYSCTL */2184 2185#ifdef CONFIG_EXEC_KUNIT_TEST2186#include "tests/exec_kunit.c"2187#endif2188