brintos

brintos / linux-shallow public Read only

0
0
Text · 43.7 KiB · 4aeb98c Raw
1848 lines · c
1// SPDX-License-Identifier: GPL-2.02/*3 *  linux/fs/read_write.c4 *5 *  Copyright (C) 1991, 1992  Linus Torvalds6 */7 8#include <linux/slab.h>9#include <linux/stat.h>10#include <linux/sched/xacct.h>11#include <linux/fcntl.h>12#include <linux/file.h>13#include <linux/uio.h>14#include <linux/fsnotify.h>15#include <linux/security.h>16#include <linux/export.h>17#include <linux/syscalls.h>18#include <linux/pagemap.h>19#include <linux/splice.h>20#include <linux/compat.h>21#include <linux/mount.h>22#include <linux/fs.h>23#include "internal.h"24 25#include <linux/uaccess.h>26#include <asm/unistd.h>27 28const struct file_operations generic_ro_fops = {29	.llseek		= generic_file_llseek,30	.read_iter	= generic_file_read_iter,31	.mmap		= generic_file_readonly_mmap,32	.splice_read	= filemap_splice_read,33};34 35EXPORT_SYMBOL(generic_ro_fops);36 37static inline bool unsigned_offsets(struct file *file)38{39	return file->f_op->fop_flags & FOP_UNSIGNED_OFFSET;40}41 42/**43 * vfs_setpos_cookie - update the file offset for lseek and reset cookie44 * @file:	file structure in question45 * @offset:	file offset to seek to46 * @maxsize:	maximum file size47 * @cookie:	cookie to reset48 *49 * Update the file offset to the value specified by @offset if the given50 * offset is valid and it is not equal to the current file offset and51 * reset the specified cookie to indicate that a seek happened.52 *53 * Return the specified offset on success and -EINVAL on invalid offset.54 */55static loff_t vfs_setpos_cookie(struct file *file, loff_t offset,56				loff_t maxsize, u64 *cookie)57{58	if (offset < 0 && !unsigned_offsets(file))59		return -EINVAL;60	if (offset > maxsize)61		return -EINVAL;62 63	if (offset != file->f_pos) {64		file->f_pos = offset;65		if (cookie)66			*cookie = 0;67	}68	return offset;69}70 71/**72 * vfs_setpos - update the file offset for lseek73 * @file:	file structure in question74 * @offset:	file offset to seek to75 * @maxsize:	maximum file size76 *77 * This is a low-level filesystem helper for updating the file offset to78 * the value specified by @offset if the given offset is valid and it is79 * not equal to the current file offset.80 *81 * Return the specified offset on success and -EINVAL on invalid offset.82 */83loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)84{85	return vfs_setpos_cookie(file, offset, maxsize, NULL);86}87EXPORT_SYMBOL(vfs_setpos);88 89/**90 * must_set_pos - check whether f_pos has to be updated91 * @file: file to seek on92 * @offset: offset to use93 * @whence: type of seek operation94 * @eof: end of file95 *96 * Check whether f_pos needs to be updated and update @offset according97 * to @whence.98 *99 * Return: 0 if f_pos doesn't need to be updated, 1 if f_pos has to be100 * updated, and negative error code on failure.101 */102static int must_set_pos(struct file *file, loff_t *offset, int whence, loff_t eof)103{104	switch (whence) {105	case SEEK_END:106		*offset += eof;107		break;108	case SEEK_CUR:109		/*110		 * Here we special-case the lseek(fd, 0, SEEK_CUR)111		 * position-querying operation.  Avoid rewriting the "same"112		 * f_pos value back to the file because a concurrent read(),113		 * write() or lseek() might have altered it114		 */115		if (*offset == 0) {116			*offset = file->f_pos;117			return 0;118		}119		break;120	case SEEK_DATA:121		/*122		 * In the generic case the entire file is data, so as long as123		 * offset isn't at the end of the file then the offset is data.124		 */125		if ((unsigned long long)*offset >= eof)126			return -ENXIO;127		break;128	case SEEK_HOLE:129		/*130		 * There is a virtual hole at the end of the file, so as long as131		 * offset isn't i_size or larger, return i_size.132		 */133		if ((unsigned long long)*offset >= eof)134			return -ENXIO;135		*offset = eof;136		break;137	}138 139	return 1;140}141 142/**143 * generic_file_llseek_size - generic llseek implementation for regular files144 * @file:	file structure to seek on145 * @offset:	file offset to seek to146 * @whence:	type of seek147 * @maxsize:	max size of this file in file system148 * @eof:	offset used for SEEK_END position149 *150 * This is a variant of generic_file_llseek that allows passing in a custom151 * maximum file size and a custom EOF position, for e.g. hashed directories152 *153 * Synchronization:154 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)155 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.156 * read/writes behave like SEEK_SET against seeks.157 */158loff_t159generic_file_llseek_size(struct file *file, loff_t offset, int whence,160		loff_t maxsize, loff_t eof)161{162	int ret;163 164	ret = must_set_pos(file, &offset, whence, eof);165	if (ret < 0)166		return ret;167	if (ret == 0)168		return offset;169 170	if (whence == SEEK_CUR) {171		/*172		 * f_lock protects against read/modify/write race with173		 * other SEEK_CURs. Note that parallel writes and reads174		 * behave like SEEK_SET.175		 */176		guard(spinlock)(&file->f_lock);177		return vfs_setpos(file, file->f_pos + offset, maxsize);178	}179 180	return vfs_setpos(file, offset, maxsize);181}182EXPORT_SYMBOL(generic_file_llseek_size);183 184/**185 * generic_llseek_cookie - versioned llseek implementation186 * @file:	file structure to seek on187 * @offset:	file offset to seek to188 * @whence:	type of seek189 * @cookie:	cookie to update190 *191 * See generic_file_llseek for a general description and locking assumptions.192 *193 * In contrast to generic_file_llseek, this function also resets a194 * specified cookie to indicate a seek took place.195 */196loff_t generic_llseek_cookie(struct file *file, loff_t offset, int whence,197			     u64 *cookie)198{199	struct inode *inode = file->f_mapping->host;200	loff_t maxsize = inode->i_sb->s_maxbytes;201	loff_t eof = i_size_read(inode);202	int ret;203 204	if (WARN_ON_ONCE(!cookie))205		return -EINVAL;206 207	/*208	 * Require that this is only used for directories that guarantee209	 * synchronization between readdir and seek so that an update to210	 * @cookie is correctly synchronized with concurrent readdir.211	 */212	if (WARN_ON_ONCE(!(file->f_mode & FMODE_ATOMIC_POS)))213		return -EINVAL;214 215	ret = must_set_pos(file, &offset, whence, eof);216	if (ret < 0)217		return ret;218	if (ret == 0)219		return offset;220 221	/* No need to hold f_lock because we know that f_pos_lock is held. */222	if (whence == SEEK_CUR)223		return vfs_setpos_cookie(file, file->f_pos + offset, maxsize, cookie);224 225	return vfs_setpos_cookie(file, offset, maxsize, cookie);226}227EXPORT_SYMBOL(generic_llseek_cookie);228 229/**230 * generic_file_llseek - generic llseek implementation for regular files231 * @file:	file structure to seek on232 * @offset:	file offset to seek to233 * @whence:	type of seek234 *235 * This is a generic implemenation of ->llseek useable for all normal local236 * filesystems.  It just updates the file offset to the value specified by237 * @offset and @whence.238 */239loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)240{241	struct inode *inode = file->f_mapping->host;242 243	return generic_file_llseek_size(file, offset, whence,244					inode->i_sb->s_maxbytes,245					i_size_read(inode));246}247EXPORT_SYMBOL(generic_file_llseek);248 249/**250 * fixed_size_llseek - llseek implementation for fixed-sized devices251 * @file:	file structure to seek on252 * @offset:	file offset to seek to253 * @whence:	type of seek254 * @size:	size of the file255 *256 */257loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)258{259	switch (whence) {260	case SEEK_SET: case SEEK_CUR: case SEEK_END:261		return generic_file_llseek_size(file, offset, whence,262						size, size);263	default:264		return -EINVAL;265	}266}267EXPORT_SYMBOL(fixed_size_llseek);268 269/**270 * no_seek_end_llseek - llseek implementation for fixed-sized devices271 * @file:	file structure to seek on272 * @offset:	file offset to seek to273 * @whence:	type of seek274 *275 */276loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence)277{278	switch (whence) {279	case SEEK_SET: case SEEK_CUR:280		return generic_file_llseek_size(file, offset, whence,281						OFFSET_MAX, 0);282	default:283		return -EINVAL;284	}285}286EXPORT_SYMBOL(no_seek_end_llseek);287 288/**289 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices290 * @file:	file structure to seek on291 * @offset:	file offset to seek to292 * @whence:	type of seek293 * @size:	maximal offset allowed294 *295 */296loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size)297{298	switch (whence) {299	case SEEK_SET: case SEEK_CUR:300		return generic_file_llseek_size(file, offset, whence,301						size, 0);302	default:303		return -EINVAL;304	}305}306EXPORT_SYMBOL(no_seek_end_llseek_size);307 308/**309 * noop_llseek - No Operation Performed llseek implementation310 * @file:	file structure to seek on311 * @offset:	file offset to seek to312 * @whence:	type of seek313 *314 * This is an implementation of ->llseek useable for the rare special case when315 * userspace expects the seek to succeed but the (device) file is actually not316 * able to perform the seek. In this case you use noop_llseek() instead of317 * falling back to the default implementation of ->llseek.318 */319loff_t noop_llseek(struct file *file, loff_t offset, int whence)320{321	return file->f_pos;322}323EXPORT_SYMBOL(noop_llseek);324 325loff_t default_llseek(struct file *file, loff_t offset, int whence)326{327	struct inode *inode = file_inode(file);328	loff_t retval;329 330	inode_lock(inode);331	switch (whence) {332		case SEEK_END:333			offset += i_size_read(inode);334			break;335		case SEEK_CUR:336			if (offset == 0) {337				retval = file->f_pos;338				goto out;339			}340			offset += file->f_pos;341			break;342		case SEEK_DATA:343			/*344			 * In the generic case the entire file is data, so as345			 * long as offset isn't at the end of the file then the346			 * offset is data.347			 */348			if (offset >= inode->i_size) {349				retval = -ENXIO;350				goto out;351			}352			break;353		case SEEK_HOLE:354			/*355			 * There is a virtual hole at the end of the file, so356			 * as long as offset isn't i_size or larger, return357			 * i_size.358			 */359			if (offset >= inode->i_size) {360				retval = -ENXIO;361				goto out;362			}363			offset = inode->i_size;364			break;365	}366	retval = -EINVAL;367	if (offset >= 0 || unsigned_offsets(file)) {368		if (offset != file->f_pos)369			file->f_pos = offset;370		retval = offset;371	}372out:373	inode_unlock(inode);374	return retval;375}376EXPORT_SYMBOL(default_llseek);377 378loff_t vfs_llseek(struct file *file, loff_t offset, int whence)379{380	if (!(file->f_mode & FMODE_LSEEK))381		return -ESPIPE;382	return file->f_op->llseek(file, offset, whence);383}384EXPORT_SYMBOL(vfs_llseek);385 386static off_t ksys_lseek(unsigned int fd, off_t offset, unsigned int whence)387{388	off_t retval;389	struct fd f = fdget_pos(fd);390	if (!fd_file(f))391		return -EBADF;392 393	retval = -EINVAL;394	if (whence <= SEEK_MAX) {395		loff_t res = vfs_llseek(fd_file(f), offset, whence);396		retval = res;397		if (res != (loff_t)retval)398			retval = -EOVERFLOW;	/* LFS: should only happen on 32 bit platforms */399	}400	fdput_pos(f);401	return retval;402}403 404SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)405{406	return ksys_lseek(fd, offset, whence);407}408 409#ifdef CONFIG_COMPAT410COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)411{412	return ksys_lseek(fd, offset, whence);413}414#endif415 416#if !defined(CONFIG_64BIT) || defined(CONFIG_COMPAT) || \417	defined(__ARCH_WANT_SYS_LLSEEK)418SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,419		unsigned long, offset_low, loff_t __user *, result,420		unsigned int, whence)421{422	int retval;423	struct fd f = fdget_pos(fd);424	loff_t offset;425 426	if (!fd_file(f))427		return -EBADF;428 429	retval = -EINVAL;430	if (whence > SEEK_MAX)431		goto out_putf;432 433	offset = vfs_llseek(fd_file(f), ((loff_t) offset_high << 32) | offset_low,434			whence);435 436	retval = (int)offset;437	if (offset >= 0) {438		retval = -EFAULT;439		if (!copy_to_user(result, &offset, sizeof(offset)))440			retval = 0;441	}442out_putf:443	fdput_pos(f);444	return retval;445}446#endif447 448int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)449{450	int mask = read_write == READ ? MAY_READ : MAY_WRITE;451	int ret;452 453	if (unlikely((ssize_t) count < 0))454		return -EINVAL;455 456	if (ppos) {457		loff_t pos = *ppos;458 459		if (unlikely(pos < 0)) {460			if (!unsigned_offsets(file))461				return -EINVAL;462			if (count >= -pos) /* both values are in 0..LLONG_MAX */463				return -EOVERFLOW;464		} else if (unlikely((loff_t) (pos + count) < 0)) {465			if (!unsigned_offsets(file))466				return -EINVAL;467		}468	}469 470	ret = security_file_permission(file, mask);471	if (ret)472		return ret;473 474	return fsnotify_file_area_perm(file, mask, ppos, count);475}476EXPORT_SYMBOL(rw_verify_area);477 478static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)479{480	struct kiocb kiocb;481	struct iov_iter iter;482	ssize_t ret;483 484	init_sync_kiocb(&kiocb, filp);485	kiocb.ki_pos = (ppos ? *ppos : 0);486	iov_iter_ubuf(&iter, ITER_DEST, buf, len);487 488	ret = filp->f_op->read_iter(&kiocb, &iter);489	BUG_ON(ret == -EIOCBQUEUED);490	if (ppos)491		*ppos = kiocb.ki_pos;492	return ret;493}494 495static int warn_unsupported(struct file *file, const char *op)496{497	pr_warn_ratelimited(498		"kernel %s not supported for file %pD4 (pid: %d comm: %.20s)\n",499		op, file, current->pid, current->comm);500	return -EINVAL;501}502 503ssize_t __kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)504{505	struct kvec iov = {506		.iov_base	= buf,507		.iov_len	= min_t(size_t, count, MAX_RW_COUNT),508	};509	struct kiocb kiocb;510	struct iov_iter iter;511	ssize_t ret;512 513	if (WARN_ON_ONCE(!(file->f_mode & FMODE_READ)))514		return -EINVAL;515	if (!(file->f_mode & FMODE_CAN_READ))516		return -EINVAL;517	/*518	 * Also fail if ->read_iter and ->read are both wired up as that519	 * implies very convoluted semantics.520	 */521	if (unlikely(!file->f_op->read_iter || file->f_op->read))522		return warn_unsupported(file, "read");523 524	init_sync_kiocb(&kiocb, file);525	kiocb.ki_pos = pos ? *pos : 0;526	iov_iter_kvec(&iter, ITER_DEST, &iov, 1, iov.iov_len);527	ret = file->f_op->read_iter(&kiocb, &iter);528	if (ret > 0) {529		if (pos)530			*pos = kiocb.ki_pos;531		fsnotify_access(file);532		add_rchar(current, ret);533	}534	inc_syscr(current);535	return ret;536}537 538ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)539{540	ssize_t ret;541 542	ret = rw_verify_area(READ, file, pos, count);543	if (ret)544		return ret;545	return __kernel_read(file, buf, count, pos);546}547EXPORT_SYMBOL(kernel_read);548 549ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)550{551	ssize_t ret;552 553	if (!(file->f_mode & FMODE_READ))554		return -EBADF;555	if (!(file->f_mode & FMODE_CAN_READ))556		return -EINVAL;557	if (unlikely(!access_ok(buf, count)))558		return -EFAULT;559 560	ret = rw_verify_area(READ, file, pos, count);561	if (ret)562		return ret;563	if (count > MAX_RW_COUNT)564		count =  MAX_RW_COUNT;565 566	if (file->f_op->read)567		ret = file->f_op->read(file, buf, count, pos);568	else if (file->f_op->read_iter)569		ret = new_sync_read(file, buf, count, pos);570	else571		ret = -EINVAL;572	if (ret > 0) {573		fsnotify_access(file);574		add_rchar(current, ret);575	}576	inc_syscr(current);577	return ret;578}579 580static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)581{582	struct kiocb kiocb;583	struct iov_iter iter;584	ssize_t ret;585 586	init_sync_kiocb(&kiocb, filp);587	kiocb.ki_pos = (ppos ? *ppos : 0);588	iov_iter_ubuf(&iter, ITER_SOURCE, (void __user *)buf, len);589 590	ret = filp->f_op->write_iter(&kiocb, &iter);591	BUG_ON(ret == -EIOCBQUEUED);592	if (ret > 0 && ppos)593		*ppos = kiocb.ki_pos;594	return ret;595}596 597/* caller is responsible for file_start_write/file_end_write */598ssize_t __kernel_write_iter(struct file *file, struct iov_iter *from, loff_t *pos)599{600	struct kiocb kiocb;601	ssize_t ret;602 603	if (WARN_ON_ONCE(!(file->f_mode & FMODE_WRITE)))604		return -EBADF;605	if (!(file->f_mode & FMODE_CAN_WRITE))606		return -EINVAL;607	/*608	 * Also fail if ->write_iter and ->write are both wired up as that609	 * implies very convoluted semantics.610	 */611	if (unlikely(!file->f_op->write_iter || file->f_op->write))612		return warn_unsupported(file, "write");613 614	init_sync_kiocb(&kiocb, file);615	kiocb.ki_pos = pos ? *pos : 0;616	ret = file->f_op->write_iter(&kiocb, from);617	if (ret > 0) {618		if (pos)619			*pos = kiocb.ki_pos;620		fsnotify_modify(file);621		add_wchar(current, ret);622	}623	inc_syscw(current);624	return ret;625}626 627/* caller is responsible for file_start_write/file_end_write */628ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)629{630	struct kvec iov = {631		.iov_base	= (void *)buf,632		.iov_len	= min_t(size_t, count, MAX_RW_COUNT),633	};634	struct iov_iter iter;635	iov_iter_kvec(&iter, ITER_SOURCE, &iov, 1, iov.iov_len);636	return __kernel_write_iter(file, &iter, pos);637}638/*639 * This "EXPORT_SYMBOL_GPL()" is more of a "EXPORT_SYMBOL_DONTUSE()",640 * but autofs is one of the few internal kernel users that actually641 * wants this _and_ can be built as a module. So we need to export642 * this symbol for autofs, even though it really isn't appropriate643 * for any other kernel modules.644 */645EXPORT_SYMBOL_GPL(__kernel_write);646 647ssize_t kernel_write(struct file *file, const void *buf, size_t count,648			    loff_t *pos)649{650	ssize_t ret;651 652	ret = rw_verify_area(WRITE, file, pos, count);653	if (ret)654		return ret;655 656	file_start_write(file);657	ret =  __kernel_write(file, buf, count, pos);658	file_end_write(file);659	return ret;660}661EXPORT_SYMBOL(kernel_write);662 663ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)664{665	ssize_t ret;666 667	if (!(file->f_mode & FMODE_WRITE))668		return -EBADF;669	if (!(file->f_mode & FMODE_CAN_WRITE))670		return -EINVAL;671	if (unlikely(!access_ok(buf, count)))672		return -EFAULT;673 674	ret = rw_verify_area(WRITE, file, pos, count);675	if (ret)676		return ret;677	if (count > MAX_RW_COUNT)678		count =  MAX_RW_COUNT;679	file_start_write(file);680	if (file->f_op->write)681		ret = file->f_op->write(file, buf, count, pos);682	else if (file->f_op->write_iter)683		ret = new_sync_write(file, buf, count, pos);684	else685		ret = -EINVAL;686	if (ret > 0) {687		fsnotify_modify(file);688		add_wchar(current, ret);689	}690	inc_syscw(current);691	file_end_write(file);692	return ret;693}694 695/* file_ppos returns &file->f_pos or NULL if file is stream */696static inline loff_t *file_ppos(struct file *file)697{698	return file->f_mode & FMODE_STREAM ? NULL : &file->f_pos;699}700 701ssize_t ksys_read(unsigned int fd, char __user *buf, size_t count)702{703	struct fd f = fdget_pos(fd);704	ssize_t ret = -EBADF;705 706	if (fd_file(f)) {707		loff_t pos, *ppos = file_ppos(fd_file(f));708		if (ppos) {709			pos = *ppos;710			ppos = &pos;711		}712		ret = vfs_read(fd_file(f), buf, count, ppos);713		if (ret >= 0 && ppos)714			fd_file(f)->f_pos = pos;715		fdput_pos(f);716	}717	return ret;718}719 720SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)721{722	return ksys_read(fd, buf, count);723}724 725ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count)726{727	struct fd f = fdget_pos(fd);728	ssize_t ret = -EBADF;729 730	if (fd_file(f)) {731		loff_t pos, *ppos = file_ppos(fd_file(f));732		if (ppos) {733			pos = *ppos;734			ppos = &pos;735		}736		ret = vfs_write(fd_file(f), buf, count, ppos);737		if (ret >= 0 && ppos)738			fd_file(f)->f_pos = pos;739		fdput_pos(f);740	}741 742	return ret;743}744 745SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,746		size_t, count)747{748	return ksys_write(fd, buf, count);749}750 751ssize_t ksys_pread64(unsigned int fd, char __user *buf, size_t count,752		     loff_t pos)753{754	struct fd f;755	ssize_t ret = -EBADF;756 757	if (pos < 0)758		return -EINVAL;759 760	f = fdget(fd);761	if (fd_file(f)) {762		ret = -ESPIPE;763		if (fd_file(f)->f_mode & FMODE_PREAD)764			ret = vfs_read(fd_file(f), buf, count, &pos);765		fdput(f);766	}767 768	return ret;769}770 771SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,772			size_t, count, loff_t, pos)773{774	return ksys_pread64(fd, buf, count, pos);775}776 777#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_PREAD64)778COMPAT_SYSCALL_DEFINE5(pread64, unsigned int, fd, char __user *, buf,779		       size_t, count, compat_arg_u64_dual(pos))780{781	return ksys_pread64(fd, buf, count, compat_arg_u64_glue(pos));782}783#endif784 785ssize_t ksys_pwrite64(unsigned int fd, const char __user *buf,786		      size_t count, loff_t pos)787{788	struct fd f;789	ssize_t ret = -EBADF;790 791	if (pos < 0)792		return -EINVAL;793 794	f = fdget(fd);795	if (fd_file(f)) {796		ret = -ESPIPE;797		if (fd_file(f)->f_mode & FMODE_PWRITE)798			ret = vfs_write(fd_file(f), buf, count, &pos);799		fdput(f);800	}801 802	return ret;803}804 805SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,806			 size_t, count, loff_t, pos)807{808	return ksys_pwrite64(fd, buf, count, pos);809}810 811#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_PWRITE64)812COMPAT_SYSCALL_DEFINE5(pwrite64, unsigned int, fd, const char __user *, buf,813		       size_t, count, compat_arg_u64_dual(pos))814{815	return ksys_pwrite64(fd, buf, count, compat_arg_u64_glue(pos));816}817#endif818 819static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,820		loff_t *ppos, int type, rwf_t flags)821{822	struct kiocb kiocb;823	ssize_t ret;824 825	init_sync_kiocb(&kiocb, filp);826	ret = kiocb_set_rw_flags(&kiocb, flags, type);827	if (ret)828		return ret;829	kiocb.ki_pos = (ppos ? *ppos : 0);830 831	if (type == READ)832		ret = filp->f_op->read_iter(&kiocb, iter);833	else834		ret = filp->f_op->write_iter(&kiocb, iter);835	BUG_ON(ret == -EIOCBQUEUED);836	if (ppos)837		*ppos = kiocb.ki_pos;838	return ret;839}840 841/* Do it by hand, with file-ops */842static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,843		loff_t *ppos, int type, rwf_t flags)844{845	ssize_t ret = 0;846 847	if (flags & ~RWF_HIPRI)848		return -EOPNOTSUPP;849 850	while (iov_iter_count(iter)) {851		ssize_t nr;852 853		if (type == READ) {854			nr = filp->f_op->read(filp, iter_iov_addr(iter),855						iter_iov_len(iter), ppos);856		} else {857			nr = filp->f_op->write(filp, iter_iov_addr(iter),858						iter_iov_len(iter), ppos);859		}860 861		if (nr < 0) {862			if (!ret)863				ret = nr;864			break;865		}866		ret += nr;867		if (nr != iter_iov_len(iter))868			break;869		iov_iter_advance(iter, nr);870	}871 872	return ret;873}874 875ssize_t vfs_iocb_iter_read(struct file *file, struct kiocb *iocb,876			   struct iov_iter *iter)877{878	size_t tot_len;879	ssize_t ret = 0;880 881	if (!file->f_op->read_iter)882		return -EINVAL;883	if (!(file->f_mode & FMODE_READ))884		return -EBADF;885	if (!(file->f_mode & FMODE_CAN_READ))886		return -EINVAL;887 888	tot_len = iov_iter_count(iter);889	if (!tot_len)890		goto out;891	ret = rw_verify_area(READ, file, &iocb->ki_pos, tot_len);892	if (ret < 0)893		return ret;894 895	ret = file->f_op->read_iter(iocb, iter);896out:897	if (ret >= 0)898		fsnotify_access(file);899	return ret;900}901EXPORT_SYMBOL(vfs_iocb_iter_read);902 903ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,904		      rwf_t flags)905{906	size_t tot_len;907	ssize_t ret = 0;908 909	if (!file->f_op->read_iter)910		return -EINVAL;911	if (!(file->f_mode & FMODE_READ))912		return -EBADF;913	if (!(file->f_mode & FMODE_CAN_READ))914		return -EINVAL;915 916	tot_len = iov_iter_count(iter);917	if (!tot_len)918		goto out;919	ret = rw_verify_area(READ, file, ppos, tot_len);920	if (ret < 0)921		return ret;922 923	ret = do_iter_readv_writev(file, iter, ppos, READ, flags);924out:925	if (ret >= 0)926		fsnotify_access(file);927	return ret;928}929EXPORT_SYMBOL(vfs_iter_read);930 931/*932 * Caller is responsible for calling kiocb_end_write() on completion933 * if async iocb was queued.934 */935ssize_t vfs_iocb_iter_write(struct file *file, struct kiocb *iocb,936			    struct iov_iter *iter)937{938	size_t tot_len;939	ssize_t ret = 0;940 941	if (!file->f_op->write_iter)942		return -EINVAL;943	if (!(file->f_mode & FMODE_WRITE))944		return -EBADF;945	if (!(file->f_mode & FMODE_CAN_WRITE))946		return -EINVAL;947 948	tot_len = iov_iter_count(iter);949	if (!tot_len)950		return 0;951	ret = rw_verify_area(WRITE, file, &iocb->ki_pos, tot_len);952	if (ret < 0)953		return ret;954 955	kiocb_start_write(iocb);956	ret = file->f_op->write_iter(iocb, iter);957	if (ret != -EIOCBQUEUED)958		kiocb_end_write(iocb);959	if (ret > 0)960		fsnotify_modify(file);961 962	return ret;963}964EXPORT_SYMBOL(vfs_iocb_iter_write);965 966ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,967		       rwf_t flags)968{969	size_t tot_len;970	ssize_t ret;971 972	if (!(file->f_mode & FMODE_WRITE))973		return -EBADF;974	if (!(file->f_mode & FMODE_CAN_WRITE))975		return -EINVAL;976	if (!file->f_op->write_iter)977		return -EINVAL;978 979	tot_len = iov_iter_count(iter);980	if (!tot_len)981		return 0;982 983	ret = rw_verify_area(WRITE, file, ppos, tot_len);984	if (ret < 0)985		return ret;986 987	file_start_write(file);988	ret = do_iter_readv_writev(file, iter, ppos, WRITE, flags);989	if (ret > 0)990		fsnotify_modify(file);991	file_end_write(file);992 993	return ret;994}995EXPORT_SYMBOL(vfs_iter_write);996 997static ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,998			 unsigned long vlen, loff_t *pos, rwf_t flags)999{1000	struct iovec iovstack[UIO_FASTIOV];1001	struct iovec *iov = iovstack;1002	struct iov_iter iter;1003	size_t tot_len;1004	ssize_t ret = 0;1005 1006	if (!(file->f_mode & FMODE_READ))1007		return -EBADF;1008	if (!(file->f_mode & FMODE_CAN_READ))1009		return -EINVAL;1010 1011	ret = import_iovec(ITER_DEST, vec, vlen, ARRAY_SIZE(iovstack), &iov,1012			   &iter);1013	if (ret < 0)1014		return ret;1015 1016	tot_len = iov_iter_count(&iter);1017	if (!tot_len)1018		goto out;1019 1020	ret = rw_verify_area(READ, file, pos, tot_len);1021	if (ret < 0)1022		goto out;1023 1024	if (file->f_op->read_iter)1025		ret = do_iter_readv_writev(file, &iter, pos, READ, flags);1026	else1027		ret = do_loop_readv_writev(file, &iter, pos, READ, flags);1028out:1029	if (ret >= 0)1030		fsnotify_access(file);1031	kfree(iov);1032	return ret;1033}1034 1035static HWJS_SUSPENDS ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,1036			  unsigned long vlen, loff_t *pos, rwf_t flags)1037{1038	struct iovec iovstack[UIO_FASTIOV];1039	struct iovec *iov = iovstack;1040	struct iov_iter iter;1041	size_t tot_len;1042	ssize_t ret = 0;1043 1044	if (!(file->f_mode & FMODE_WRITE))1045		return -EBADF;1046	if (!(file->f_mode & FMODE_CAN_WRITE))1047		return -EINVAL;1048 1049	ret = import_iovec(ITER_SOURCE, vec, vlen, ARRAY_SIZE(iovstack), &iov,1050			   &iter);1051	if (ret < 0)1052		return ret;1053 1054	tot_len = iov_iter_count(&iter);1055	if (!tot_len)1056		goto out;1057 1058	ret = rw_verify_area(WRITE, file, pos, tot_len);1059	if (ret < 0)1060		goto out;1061 1062	file_start_write(file);1063	if (file->f_op->write_iter)1064		ret = do_iter_readv_writev(file, &iter, pos, WRITE, flags);1065	else1066		ret = do_loop_readv_writev(file, &iter, pos, WRITE, flags);1067	if (ret > 0)1068		fsnotify_modify(file);1069	file_end_write(file);1070out:1071	kfree(iov);1072	return ret;1073}1074 1075static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,1076			unsigned long vlen, rwf_t flags)1077{1078	struct fd f = fdget_pos(fd);1079	ssize_t ret = -EBADF;1080 1081	if (fd_file(f)) {1082		loff_t pos, *ppos = file_ppos(fd_file(f));1083		if (ppos) {1084			pos = *ppos;1085			ppos = &pos;1086		}1087		ret = vfs_readv(fd_file(f), vec, vlen, ppos, flags);1088		if (ret >= 0 && ppos)1089			fd_file(f)->f_pos = pos;1090		fdput_pos(f);1091	}1092 1093	if (ret > 0)1094		add_rchar(current, ret);1095	inc_syscr(current);1096	return ret;1097}1098 1099static HWJS_SUSPENDS ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,1100			 unsigned long vlen, rwf_t flags)1101{1102	struct fd f = fdget_pos(fd);1103	ssize_t ret = -EBADF;1104 1105	if (fd_file(f)) {1106		loff_t pos, *ppos = file_ppos(fd_file(f));1107		if (ppos) {1108			pos = *ppos;1109			ppos = &pos;1110		}1111		ret = vfs_writev(fd_file(f), vec, vlen, ppos, flags);1112		if (ret >= 0 && ppos)1113			fd_file(f)->f_pos = pos;1114		fdput_pos(f);1115	}1116 1117	if (ret > 0)1118		add_wchar(current, ret);1119	inc_syscw(current);1120	return ret;1121}1122 1123static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)1124{1125#define HALF_LONG_BITS (BITS_PER_LONG / 2)1126	return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;1127}1128 1129static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,1130			 unsigned long vlen, loff_t pos, rwf_t flags)1131{1132	struct fd f;1133	ssize_t ret = -EBADF;1134 1135	if (pos < 0)1136		return -EINVAL;1137 1138	f = fdget(fd);1139	if (fd_file(f)) {1140		ret = -ESPIPE;1141		if (fd_file(f)->f_mode & FMODE_PREAD)1142			ret = vfs_readv(fd_file(f), vec, vlen, &pos, flags);1143		fdput(f);1144	}1145 1146	if (ret > 0)1147		add_rchar(current, ret);1148	inc_syscr(current);1149	return ret;1150}1151 1152static HWJS_SUSPENDS ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,1153			  unsigned long vlen, loff_t pos, rwf_t flags)1154{1155	struct fd f;1156	ssize_t ret = -EBADF;1157 1158	if (pos < 0)1159		return -EINVAL;1160 1161	f = fdget(fd);1162	if (fd_file(f)) {1163		ret = -ESPIPE;1164		if (fd_file(f)->f_mode & FMODE_PWRITE)1165			ret = vfs_writev(fd_file(f), vec, vlen, &pos, flags);1166		fdput(f);1167	}1168 1169	if (ret > 0)1170		add_wchar(current, ret);1171	inc_syscw(current);1172	return ret;1173}1174 1175SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,1176		unsigned long, vlen)1177{1178	return do_readv(fd, vec, vlen, 0);1179}1180 1181SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,1182		unsigned long, vlen)1183{1184	return do_writev(fd, vec, vlen, 0);1185}1186 1187SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,1188		unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)1189{1190	loff_t pos = pos_from_hilo(pos_h, pos_l);1191 1192	return do_preadv(fd, vec, vlen, pos, 0);1193}1194 1195SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,1196		unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,1197		rwf_t, flags)1198{1199	loff_t pos = pos_from_hilo(pos_h, pos_l);1200 1201	if (pos == -1)1202		return do_readv(fd, vec, vlen, flags);1203 1204	return do_preadv(fd, vec, vlen, pos, flags);1205}1206 1207SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,1208		unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)1209{1210	loff_t pos = pos_from_hilo(pos_h, pos_l);1211 1212	return do_pwritev(fd, vec, vlen, pos, 0);1213}1214 1215SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,1216		unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,1217		rwf_t, flags)1218{1219	loff_t pos = pos_from_hilo(pos_h, pos_l);1220 1221	if (pos == -1)1222		return do_writev(fd, vec, vlen, flags);1223 1224	return do_pwritev(fd, vec, vlen, pos, flags);1225}1226 1227/*1228 * Various compat syscalls.  Note that they all pretend to take a native1229 * iovec - import_iovec will properly treat those as compat_iovecs based on1230 * in_compat_syscall().1231 */1232#ifdef CONFIG_COMPAT1233#ifdef __ARCH_WANT_COMPAT_SYS_PREADV641234COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,1235		const struct iovec __user *, vec,1236		unsigned long, vlen, loff_t, pos)1237{1238	return do_preadv(fd, vec, vlen, pos, 0);1239}1240#endif1241 1242COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,1243		const struct iovec __user *, vec,1244		compat_ulong_t, vlen, u32, pos_low, u32, pos_high)1245{1246	loff_t pos = ((loff_t)pos_high << 32) | pos_low;1247 1248	return do_preadv(fd, vec, vlen, pos, 0);1249}1250 1251#ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V21252COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,1253		const struct iovec __user *, vec,1254		unsigned long, vlen, loff_t, pos, rwf_t, flags)1255{1256	if (pos == -1)1257		return do_readv(fd, vec, vlen, flags);1258	return do_preadv(fd, vec, vlen, pos, flags);1259}1260#endif1261 1262COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,1263		const struct iovec __user *, vec,1264		compat_ulong_t, vlen, u32, pos_low, u32, pos_high,1265		rwf_t, flags)1266{1267	loff_t pos = ((loff_t)pos_high << 32) | pos_low;1268 1269	if (pos == -1)1270		return do_readv(fd, vec, vlen, flags);1271	return do_preadv(fd, vec, vlen, pos, flags);1272}1273 1274#ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV641275COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,1276		const struct iovec __user *, vec,1277		unsigned long, vlen, loff_t, pos)1278{1279	return do_pwritev(fd, vec, vlen, pos, 0);1280}1281#endif1282 1283COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,1284		const struct iovec __user *,vec,1285		compat_ulong_t, vlen, u32, pos_low, u32, pos_high)1286{1287	loff_t pos = ((loff_t)pos_high << 32) | pos_low;1288 1289	return do_pwritev(fd, vec, vlen, pos, 0);1290}1291 1292#ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V21293COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,1294		const struct iovec __user *, vec,1295		unsigned long, vlen, loff_t, pos, rwf_t, flags)1296{1297	if (pos == -1)1298		return do_writev(fd, vec, vlen, flags);1299	return do_pwritev(fd, vec, vlen, pos, flags);1300}1301#endif1302 1303COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,1304		const struct iovec __user *,vec,1305		compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags)1306{1307	loff_t pos = ((loff_t)pos_high << 32) | pos_low;1308 1309	if (pos == -1)1310		return do_writev(fd, vec, vlen, flags);1311	return do_pwritev(fd, vec, vlen, pos, flags);1312}1313#endif /* CONFIG_COMPAT */1314 1315static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,1316			   size_t count, loff_t max)1317{1318	struct fd in, out;1319	struct inode *in_inode, *out_inode;1320	struct pipe_inode_info *opipe;1321	loff_t pos;1322	loff_t out_pos;1323	ssize_t retval;1324	int fl;1325 1326	/*1327	 * Get input file, and verify that it is ok..1328	 */1329	retval = -EBADF;1330	in = fdget(in_fd);1331	if (!fd_file(in))1332		goto out;1333	if (!(fd_file(in)->f_mode & FMODE_READ))1334		goto fput_in;1335	retval = -ESPIPE;1336	if (!ppos) {1337		pos = fd_file(in)->f_pos;1338	} else {1339		pos = *ppos;1340		if (!(fd_file(in)->f_mode & FMODE_PREAD))1341			goto fput_in;1342	}1343	retval = rw_verify_area(READ, fd_file(in), &pos, count);1344	if (retval < 0)1345		goto fput_in;1346	if (count > MAX_RW_COUNT)1347		count =  MAX_RW_COUNT;1348 1349	/*1350	 * Get output file, and verify that it is ok..1351	 */1352	retval = -EBADF;1353	out = fdget(out_fd);1354	if (!fd_file(out))1355		goto fput_in;1356	if (!(fd_file(out)->f_mode & FMODE_WRITE))1357		goto fput_out;1358	in_inode = file_inode(fd_file(in));1359	out_inode = file_inode(fd_file(out));1360	out_pos = fd_file(out)->f_pos;1361 1362	if (!max)1363		max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);1364 1365	if (unlikely(pos + count > max)) {1366		retval = -EOVERFLOW;1367		if (pos >= max)1368			goto fput_out;1369		count = max - pos;1370	}1371 1372	fl = 0;1373#if 01374	/*1375	 * We need to debate whether we can enable this or not. The1376	 * man page documents EAGAIN return for the output at least,1377	 * and the application is arguably buggy if it doesn't expect1378	 * EAGAIN on a non-blocking file descriptor.1379	 */1380	if (fd_file(in)->f_flags & O_NONBLOCK)1381		fl = SPLICE_F_NONBLOCK;1382#endif1383	opipe = get_pipe_info(fd_file(out), true);1384	if (!opipe) {1385		retval = rw_verify_area(WRITE, fd_file(out), &out_pos, count);1386		if (retval < 0)1387			goto fput_out;1388		retval = do_splice_direct(fd_file(in), &pos, fd_file(out), &out_pos,1389					  count, fl);1390	} else {1391		if (fd_file(out)->f_flags & O_NONBLOCK)1392			fl |= SPLICE_F_NONBLOCK;1393 1394		retval = splice_file_to_pipe(fd_file(in), opipe, &pos, count, fl);1395	}1396 1397	if (retval > 0) {1398		add_rchar(current, retval);1399		add_wchar(current, retval);1400		fsnotify_access(fd_file(in));1401		fsnotify_modify(fd_file(out));1402		fd_file(out)->f_pos = out_pos;1403		if (ppos)1404			*ppos = pos;1405		else1406			fd_file(in)->f_pos = pos;1407	}1408 1409	inc_syscr(current);1410	inc_syscw(current);1411	if (pos > max)1412		retval = -EOVERFLOW;1413 1414fput_out:1415	fdput(out);1416fput_in:1417	fdput(in);1418out:1419	return retval;1420}1421 1422SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)1423{1424	loff_t pos;1425	off_t off;1426	ssize_t ret;1427 1428	if (offset) {1429		if (unlikely(get_user(off, offset)))1430			return -EFAULT;1431		pos = off;1432		ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);1433		if (unlikely(put_user(pos, offset)))1434			return -EFAULT;1435		return ret;1436	}1437 1438	return do_sendfile(out_fd, in_fd, NULL, count, 0);1439}1440 1441SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)1442{1443	loff_t pos;1444	ssize_t ret;1445 1446	if (offset) {1447		if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))1448			return -EFAULT;1449		ret = do_sendfile(out_fd, in_fd, &pos, count, 0);1450		if (unlikely(put_user(pos, offset)))1451			return -EFAULT;1452		return ret;1453	}1454 1455	return do_sendfile(out_fd, in_fd, NULL, count, 0);1456}1457 1458#ifdef CONFIG_COMPAT1459COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,1460		compat_off_t __user *, offset, compat_size_t, count)1461{1462	loff_t pos;1463	off_t off;1464	ssize_t ret;1465 1466	if (offset) {1467		if (unlikely(get_user(off, offset)))1468			return -EFAULT;1469		pos = off;1470		ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);1471		if (unlikely(put_user(pos, offset)))1472			return -EFAULT;1473		return ret;1474	}1475 1476	return do_sendfile(out_fd, in_fd, NULL, count, 0);1477}1478 1479COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,1480		compat_loff_t __user *, offset, compat_size_t, count)1481{1482	loff_t pos;1483	ssize_t ret;1484 1485	if (offset) {1486		if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))1487			return -EFAULT;1488		ret = do_sendfile(out_fd, in_fd, &pos, count, 0);1489		if (unlikely(put_user(pos, offset)))1490			return -EFAULT;1491		return ret;1492	}1493 1494	return do_sendfile(out_fd, in_fd, NULL, count, 0);1495}1496#endif1497 1498/*1499 * Performs necessary checks before doing a file copy1500 *1501 * Can adjust amount of bytes to copy via @req_count argument.1502 * Returns appropriate error code that caller should return or1503 * zero in case the copy should be allowed.1504 */1505static int generic_copy_file_checks(struct file *file_in, loff_t pos_in,1506				    struct file *file_out, loff_t pos_out,1507				    size_t *req_count, unsigned int flags)1508{1509	struct inode *inode_in = file_inode(file_in);1510	struct inode *inode_out = file_inode(file_out);1511	uint64_t count = *req_count;1512	loff_t size_in;1513	int ret;1514 1515	ret = generic_file_rw_checks(file_in, file_out);1516	if (ret)1517		return ret;1518 1519	/*1520	 * We allow some filesystems to handle cross sb copy, but passing1521	 * a file of the wrong filesystem type to filesystem driver can result1522	 * in an attempt to dereference the wrong type of ->private_data, so1523	 * avoid doing that until we really have a good reason.1524	 *1525	 * nfs and cifs define several different file_system_type structures1526	 * and several different sets of file_operations, but they all end up1527	 * using the same ->copy_file_range() function pointer.1528	 */1529	if (flags & COPY_FILE_SPLICE) {1530		/* cross sb splice is allowed */1531	} else if (file_out->f_op->copy_file_range) {1532		if (file_in->f_op->copy_file_range !=1533		    file_out->f_op->copy_file_range)1534			return -EXDEV;1535	} else if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb) {1536		return -EXDEV;1537	}1538 1539	/* Don't touch certain kinds of inodes */1540	if (IS_IMMUTABLE(inode_out))1541		return -EPERM;1542 1543	if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))1544		return -ETXTBSY;1545 1546	/* Ensure offsets don't wrap. */1547	if (pos_in + count < pos_in || pos_out + count < pos_out)1548		return -EOVERFLOW;1549 1550	/* Shorten the copy to EOF */1551	size_in = i_size_read(inode_in);1552	if (pos_in >= size_in)1553		count = 0;1554	else1555		count = min(count, size_in - (uint64_t)pos_in);1556 1557	ret = generic_write_check_limits(file_out, pos_out, &count);1558	if (ret)1559		return ret;1560 1561	/* Don't allow overlapped copying within the same file. */1562	if (inode_in == inode_out &&1563	    pos_out + count > pos_in &&1564	    pos_out < pos_in + count)1565		return -EINVAL;1566 1567	*req_count = count;1568	return 0;1569}1570 1571/*1572 * copy_file_range() differs from regular file read and write in that it1573 * specifically allows return partial success.  When it does so is up to1574 * the copy_file_range method.1575 */1576ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,1577			    struct file *file_out, loff_t pos_out,1578			    size_t len, unsigned int flags)1579{1580	ssize_t ret;1581	bool splice = flags & COPY_FILE_SPLICE;1582	bool samesb = file_inode(file_in)->i_sb == file_inode(file_out)->i_sb;1583 1584	if (flags & ~COPY_FILE_SPLICE)1585		return -EINVAL;1586 1587	ret = generic_copy_file_checks(file_in, pos_in, file_out, pos_out, &len,1588				       flags);1589	if (unlikely(ret))1590		return ret;1591 1592	ret = rw_verify_area(READ, file_in, &pos_in, len);1593	if (unlikely(ret))1594		return ret;1595 1596	ret = rw_verify_area(WRITE, file_out, &pos_out, len);1597	if (unlikely(ret))1598		return ret;1599 1600	if (len == 0)1601		return 0;1602 1603	file_start_write(file_out);1604 1605	/*1606	 * Cloning is supported by more file systems, so we implement copy on1607	 * same sb using clone, but for filesystems where both clone and copy1608	 * are supported (e.g. nfs,cifs), we only call the copy method.1609	 */1610	if (!splice && file_out->f_op->copy_file_range) {1611		ret = file_out->f_op->copy_file_range(file_in, pos_in,1612						      file_out, pos_out,1613						      len, flags);1614	} else if (!splice && file_in->f_op->remap_file_range && samesb) {1615		ret = file_in->f_op->remap_file_range(file_in, pos_in,1616				file_out, pos_out,1617				min_t(loff_t, MAX_RW_COUNT, len),1618				REMAP_FILE_CAN_SHORTEN);1619		/* fallback to splice */1620		if (ret <= 0)1621			splice = true;1622	} else if (samesb) {1623		/* Fallback to splice for same sb copy for backward compat */1624		splice = true;1625	}1626 1627	file_end_write(file_out);1628 1629	if (!splice)1630		goto done;1631 1632	/*1633	 * We can get here for same sb copy of filesystems that do not implement1634	 * ->copy_file_range() in case filesystem does not support clone or in1635	 * case filesystem supports clone but rejected the clone request (e.g.1636	 * because it was not block aligned).1637	 *1638	 * In both cases, fall back to kernel copy so we are able to maintain a1639	 * consistent story about which filesystems support copy_file_range()1640	 * and which filesystems do not, that will allow userspace tools to1641	 * make consistent desicions w.r.t using copy_file_range().1642	 *1643	 * We also get here if caller (e.g. nfsd) requested COPY_FILE_SPLICE1644	 * for server-side-copy between any two sb.1645	 *1646	 * In any case, we call do_splice_direct() and not splice_file_range(),1647	 * without file_start_write() held, to avoid possible deadlocks related1648	 * to splicing from input file, while file_start_write() is held on1649	 * the output file on a different sb.1650	 */1651	ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,1652			       min_t(size_t, len, MAX_RW_COUNT), 0);1653done:1654	if (ret > 0) {1655		fsnotify_access(file_in);1656		add_rchar(current, ret);1657		fsnotify_modify(file_out);1658		add_wchar(current, ret);1659	}1660 1661	inc_syscr(current);1662	inc_syscw(current);1663 1664	return ret;1665}1666EXPORT_SYMBOL(vfs_copy_file_range);1667 1668SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,1669		int, fd_out, loff_t __user *, off_out,1670		size_t, len, unsigned int, flags)1671{1672	loff_t pos_in;1673	loff_t pos_out;1674	struct fd f_in;1675	struct fd f_out;1676	ssize_t ret = -EBADF;1677 1678	f_in = fdget(fd_in);1679	if (!fd_file(f_in))1680		goto out2;1681 1682	f_out = fdget(fd_out);1683	if (!fd_file(f_out))1684		goto out1;1685 1686	ret = -EFAULT;1687	if (off_in) {1688		if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))1689			goto out;1690	} else {1691		pos_in = fd_file(f_in)->f_pos;1692	}1693 1694	if (off_out) {1695		if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))1696			goto out;1697	} else {1698		pos_out = fd_file(f_out)->f_pos;1699	}1700 1701	ret = -EINVAL;1702	if (flags != 0)1703		goto out;1704 1705	ret = vfs_copy_file_range(fd_file(f_in), pos_in, fd_file(f_out), pos_out, len,1706				  flags);1707	if (ret > 0) {1708		pos_in += ret;1709		pos_out += ret;1710 1711		if (off_in) {1712			if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))1713				ret = -EFAULT;1714		} else {1715			fd_file(f_in)->f_pos = pos_in;1716		}1717 1718		if (off_out) {1719			if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))1720				ret = -EFAULT;1721		} else {1722			fd_file(f_out)->f_pos = pos_out;1723		}1724	}1725 1726out:1727	fdput(f_out);1728out1:1729	fdput(f_in);1730out2:1731	return ret;1732}1733 1734/*1735 * Don't operate on ranges the page cache doesn't support, and don't exceed the1736 * LFS limits.  If pos is under the limit it becomes a short access.  If it1737 * exceeds the limit we return -EFBIG.1738 */1739int generic_write_check_limits(struct file *file, loff_t pos, loff_t *count)1740{1741	struct inode *inode = file->f_mapping->host;1742	loff_t max_size = inode->i_sb->s_maxbytes;1743	loff_t limit = rlimit(RLIMIT_FSIZE);1744 1745	if (limit != RLIM_INFINITY) {1746		if (pos >= limit) {1747			send_sig(SIGXFSZ, current, 0);1748			return -EFBIG;1749		}1750		*count = min(*count, limit - pos);1751	}1752 1753	if (!(file->f_flags & O_LARGEFILE))1754		max_size = MAX_NON_LFS;1755 1756	if (unlikely(pos >= max_size))1757		return -EFBIG;1758 1759	*count = min(*count, max_size - pos);1760 1761	return 0;1762}1763EXPORT_SYMBOL_GPL(generic_write_check_limits);1764 1765/* Like generic_write_checks(), but takes size of write instead of iter. */1766int generic_write_checks_count(struct kiocb *iocb, loff_t *count)1767{1768	struct file *file = iocb->ki_filp;1769	struct inode *inode = file->f_mapping->host;1770 1771	if (IS_SWAPFILE(inode))1772		return -ETXTBSY;1773 1774	if (!*count)1775		return 0;1776 1777	if (iocb->ki_flags & IOCB_APPEND)1778		iocb->ki_pos = i_size_read(inode);1779 1780	if ((iocb->ki_flags & IOCB_NOWAIT) &&1781	    !((iocb->ki_flags & IOCB_DIRECT) ||1782	      (file->f_op->fop_flags & FOP_BUFFER_WASYNC)))1783		return -EINVAL;1784 1785	return generic_write_check_limits(iocb->ki_filp, iocb->ki_pos, count);1786}1787EXPORT_SYMBOL(generic_write_checks_count);1788 1789/*1790 * Performs necessary checks before doing a write1791 *1792 * Can adjust writing position or amount of bytes to write.1793 * Returns appropriate error code that caller should return or1794 * zero in case that write should be allowed.1795 */1796ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)1797{1798	loff_t count = iov_iter_count(from);1799	int ret;1800 1801	ret = generic_write_checks_count(iocb, &count);1802	if (ret)1803		return ret;1804 1805	iov_iter_truncate(from, count);1806	return iov_iter_count(from);1807}1808EXPORT_SYMBOL(generic_write_checks);1809 1810/*1811 * Performs common checks before doing a file copy/clone1812 * from @file_in to @file_out.1813 */1814int generic_file_rw_checks(struct file *file_in, struct file *file_out)1815{1816	struct inode *inode_in = file_inode(file_in);1817	struct inode *inode_out = file_inode(file_out);1818 1819	/* Don't copy dirs, pipes, sockets... */1820	if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))1821		return -EISDIR;1822	if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))1823		return -EINVAL;1824 1825	if (!(file_in->f_mode & FMODE_READ) ||1826	    !(file_out->f_mode & FMODE_WRITE) ||1827	    (file_out->f_flags & O_APPEND))1828		return -EBADF;1829 1830	return 0;1831}1832 1833bool generic_atomic_write_valid(struct iov_iter *iter, loff_t pos)1834{1835	size_t len = iov_iter_count(iter);1836 1837	if (!iter_is_ubuf(iter))1838		return false;1839 1840	if (!is_power_of_2(len))1841		return false;1842 1843	if (!IS_ALIGNED(pos, len))1844		return false;1845 1846	return true;1847}1848