1b2441318SGreg Kroah-Hartman // SPDX-License-Identifier: GPL-2.0 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * linux/fs/ext2/file.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * Copyright (C) 1992, 1993, 1994, 1995 61da177e4SLinus Torvalds * Remy Card (card@masi.ibp.fr) 71da177e4SLinus Torvalds * Laboratoire MASI - Institut Blaise Pascal 81da177e4SLinus Torvalds * Universite Pierre et Marie Curie (Paris VI) 91da177e4SLinus Torvalds * 101da177e4SLinus Torvalds * from 111da177e4SLinus Torvalds * 121da177e4SLinus Torvalds * linux/fs/minix/file.c 131da177e4SLinus Torvalds * 141da177e4SLinus Torvalds * Copyright (C) 1991, 1992 Linus Torvalds 151da177e4SLinus Torvalds * 161da177e4SLinus Torvalds * ext2 fs regular file handling primitives 171da177e4SLinus Torvalds * 181da177e4SLinus Torvalds * 64-bit file support on 64-bit platforms by Jakub Jelinek 191da177e4SLinus Torvalds * (jj@sunsite.ms.mff.cuni.cz) 201da177e4SLinus Torvalds */ 211da177e4SLinus Torvalds 221da177e4SLinus Torvalds #include <linux/time.h> 2348bde86dSJan Kara #include <linux/pagemap.h> 24c94c2acfSMatthew Wilcox #include <linux/dax.h> 25871a2931SChristoph Hellwig #include <linux/quotaops.h> 2625f4e702SChristoph Hellwig #include <linux/iomap.h> 2725f4e702SChristoph Hellwig #include <linux/uio.h> 28d0530704SRitesh Harjani (IBM) #include <linux/buffer_head.h> 291da177e4SLinus Torvalds #include "ext2.h" 301da177e4SLinus Torvalds #include "xattr.h" 311da177e4SLinus Torvalds #include "acl.h" 321da177e4SLinus Torvalds 336cd176a5SMatthew Wilcox #ifdef CONFIG_FS_DAX 3425f4e702SChristoph Hellwig static ssize_t ext2_dax_read_iter(struct kiocb *iocb, struct iov_iter *to) 3525f4e702SChristoph Hellwig { 3625f4e702SChristoph Hellwig struct inode *inode = iocb->ki_filp->f_mapping->host; 3725f4e702SChristoph Hellwig ssize_t ret; 3825f4e702SChristoph Hellwig 3925f4e702SChristoph Hellwig if (!iov_iter_count(to)) 4025f4e702SChristoph Hellwig return 0; /* skip atime */ 4125f4e702SChristoph Hellwig 4225f4e702SChristoph Hellwig inode_lock_shared(inode); 4311c59c92SRoss Zwisler ret = dax_iomap_rw(iocb, to, &ext2_iomap_ops); 4425f4e702SChristoph Hellwig inode_unlock_shared(inode); 4525f4e702SChristoph Hellwig 4625f4e702SChristoph Hellwig file_accessed(iocb->ki_filp); 4725f4e702SChristoph Hellwig return ret; 4825f4e702SChristoph Hellwig } 4925f4e702SChristoph Hellwig 5025f4e702SChristoph Hellwig static ssize_t ext2_dax_write_iter(struct kiocb *iocb, struct iov_iter *from) 5125f4e702SChristoph Hellwig { 5225f4e702SChristoph Hellwig struct file *file = iocb->ki_filp; 5325f4e702SChristoph Hellwig struct inode *inode = file->f_mapping->host; 5425f4e702SChristoph Hellwig ssize_t ret; 5525f4e702SChristoph Hellwig 5625f4e702SChristoph Hellwig inode_lock(inode); 5725f4e702SChristoph Hellwig ret = generic_write_checks(iocb, from); 5825f4e702SChristoph Hellwig if (ret <= 0) 5925f4e702SChristoph Hellwig goto out_unlock; 6025f4e702SChristoph Hellwig ret = file_remove_privs(file); 6125f4e702SChristoph Hellwig if (ret) 6225f4e702SChristoph Hellwig goto out_unlock; 6325f4e702SChristoph Hellwig ret = file_update_time(file); 6425f4e702SChristoph Hellwig if (ret) 6525f4e702SChristoph Hellwig goto out_unlock; 6625f4e702SChristoph Hellwig 6711c59c92SRoss Zwisler ret = dax_iomap_rw(iocb, from, &ext2_iomap_ops); 6825f4e702SChristoph Hellwig if (ret > 0 && iocb->ki_pos > i_size_read(inode)) { 6925f4e702SChristoph Hellwig i_size_write(inode, iocb->ki_pos); 7025f4e702SChristoph Hellwig mark_inode_dirty(inode); 7125f4e702SChristoph Hellwig } 7225f4e702SChristoph Hellwig 7325f4e702SChristoph Hellwig out_unlock: 7425f4e702SChristoph Hellwig inode_unlock(inode); 7525f4e702SChristoph Hellwig if (ret > 0) 7625f4e702SChristoph Hellwig ret = generic_write_sync(iocb, ret); 7725f4e702SChristoph Hellwig return ret; 7825f4e702SChristoph Hellwig } 7925f4e702SChristoph Hellwig 805726b27bSRoss Zwisler /* 815726b27bSRoss Zwisler * The lock ordering for ext2 DAX fault paths is: 825726b27bSRoss Zwisler * 83c1e8d7c6SMichel Lespinasse * mmap_lock (MM) 845726b27bSRoss Zwisler * sb_start_pagefault (vfs, freeze) 8570f3bad8SJan Kara * address_space->invalidate_lock 865726b27bSRoss Zwisler * address_space->i_mmap_rwsem or page_lock (mutually exclusive in DAX) 875726b27bSRoss Zwisler * ext2_inode_info->truncate_mutex 885726b27bSRoss Zwisler * 895726b27bSRoss Zwisler * The default page_lock and i_size verification done by non-DAX fault paths 905726b27bSRoss Zwisler * is sufficient because ext2 doesn't support hole punching. 915726b27bSRoss Zwisler */ 9206856938SSouptick Joarder static vm_fault_t ext2_dax_fault(struct vm_fault *vmf) 93f7ca90b1SMatthew Wilcox { 9411bac800SDave Jiang struct inode *inode = file_inode(vmf->vma->vm_file); 9506856938SSouptick Joarder vm_fault_t ret; 961ef6ea0eSMikulas Patocka bool write = (vmf->flags & FAULT_FLAG_WRITE) && 971ef6ea0eSMikulas Patocka (vmf->vma->vm_flags & VM_SHARED); 985726b27bSRoss Zwisler 991ef6ea0eSMikulas Patocka if (write) { 1005726b27bSRoss Zwisler sb_start_pagefault(inode->i_sb); 10111bac800SDave Jiang file_update_time(vmf->vma->vm_file); 1025726b27bSRoss Zwisler } 10370f3bad8SJan Kara filemap_invalidate_lock_shared(inode->i_mapping); 1045726b27bSRoss Zwisler 105c0b24625SJan Kara ret = dax_iomap_fault(vmf, PE_SIZE_PTE, NULL, NULL, &ext2_iomap_ops); 1065726b27bSRoss Zwisler 10770f3bad8SJan Kara filemap_invalidate_unlock_shared(inode->i_mapping); 1081ef6ea0eSMikulas Patocka if (write) 1095726b27bSRoss Zwisler sb_end_pagefault(inode->i_sb); 1105726b27bSRoss Zwisler return ret; 111f7ca90b1SMatthew Wilcox } 112f7ca90b1SMatthew Wilcox 113f7ca90b1SMatthew Wilcox static const struct vm_operations_struct ext2_dax_vm_ops = { 114f7ca90b1SMatthew Wilcox .fault = ext2_dax_fault, 11503e0990fSRoss Zwisler /* 116a2d58167SDave Jiang * .huge_fault is not supported for DAX because allocation in ext2 11703e0990fSRoss Zwisler * cannot be reliably aligned to huge page sizes and so pmd faults 11803e0990fSRoss Zwisler * will always fail and fail back to regular faults. 11903e0990fSRoss Zwisler */ 1201e9d180bSRoss Zwisler .page_mkwrite = ext2_dax_fault, 12191d25ba8SRoss Zwisler .pfn_mkwrite = ext2_dax_fault, 122f7ca90b1SMatthew Wilcox }; 123f7ca90b1SMatthew Wilcox 124f7ca90b1SMatthew Wilcox static int ext2_file_mmap(struct file *file, struct vm_area_struct *vma) 125f7ca90b1SMatthew Wilcox { 126f7ca90b1SMatthew Wilcox if (!IS_DAX(file_inode(file))) 127f7ca90b1SMatthew Wilcox return generic_file_mmap(file, vma); 128f7ca90b1SMatthew Wilcox 129f7ca90b1SMatthew Wilcox file_accessed(file); 130f7ca90b1SMatthew Wilcox vma->vm_ops = &ext2_dax_vm_ops; 131f7ca90b1SMatthew Wilcox return 0; 132f7ca90b1SMatthew Wilcox } 133f7ca90b1SMatthew Wilcox #else 134f7ca90b1SMatthew Wilcox #define ext2_file_mmap generic_file_mmap 135f7ca90b1SMatthew Wilcox #endif 136f7ca90b1SMatthew Wilcox 1371da177e4SLinus Torvalds /* 138a6739af8SJan Kara * Called when filp is released. This happens when all file descriptors 139a6739af8SJan Kara * for a single struct file are closed. Note that different open() calls 140a6739af8SJan Kara * for the same file yield different struct file structures. 1411da177e4SLinus Torvalds */ 1421da177e4SLinus Torvalds static int ext2_release_file (struct inode * inode, struct file * filp) 1431da177e4SLinus Torvalds { 144a686cd89SMartin J. Bligh if (filp->f_mode & FMODE_WRITE) { 145a686cd89SMartin J. Bligh mutex_lock(&EXT2_I(inode)->truncate_mutex); 146a686cd89SMartin J. Bligh ext2_discard_reservation(inode); 147a686cd89SMartin J. Bligh mutex_unlock(&EXT2_I(inode)->truncate_mutex); 148a686cd89SMartin J. Bligh } 1491da177e4SLinus Torvalds return 0; 1501da177e4SLinus Torvalds } 1511da177e4SLinus Torvalds 15202c24a82SJosef Bacik int ext2_fsync(struct file *file, loff_t start, loff_t end, int datasync) 15348bde86dSJan Kara { 15448bde86dSJan Kara int ret; 1557ea80859SChristoph Hellwig struct super_block *sb = file->f_mapping->host->i_sb; 15648bde86dSJan Kara 157d0530704SRitesh Harjani (IBM) ret = generic_buffers_fsync(file, start, end, datasync); 158dac257f7SJeff Layton if (ret == -EIO) 15948bde86dSJan Kara /* We don't really know where the IO error happened... */ 16048bde86dSJan Kara ext2_error(sb, __func__, 16148bde86dSJan Kara "detected IO error when writing metadata buffers"); 16248bde86dSJan Kara return ret; 16348bde86dSJan Kara } 16448bde86dSJan Kara 165*fb5de435SRitesh Harjani (IBM) static ssize_t ext2_dio_read_iter(struct kiocb *iocb, struct iov_iter *to) 166*fb5de435SRitesh Harjani (IBM) { 167*fb5de435SRitesh Harjani (IBM) struct file *file = iocb->ki_filp; 168*fb5de435SRitesh Harjani (IBM) struct inode *inode = file->f_mapping->host; 169*fb5de435SRitesh Harjani (IBM) ssize_t ret; 170*fb5de435SRitesh Harjani (IBM) 171*fb5de435SRitesh Harjani (IBM) inode_lock_shared(inode); 172*fb5de435SRitesh Harjani (IBM) ret = iomap_dio_rw(iocb, to, &ext2_iomap_ops, NULL, 0, NULL, 0); 173*fb5de435SRitesh Harjani (IBM) inode_unlock_shared(inode); 174*fb5de435SRitesh Harjani (IBM) 175*fb5de435SRitesh Harjani (IBM) return ret; 176*fb5de435SRitesh Harjani (IBM) } 177*fb5de435SRitesh Harjani (IBM) 178*fb5de435SRitesh Harjani (IBM) static int ext2_dio_write_end_io(struct kiocb *iocb, ssize_t size, 179*fb5de435SRitesh Harjani (IBM) int error, unsigned int flags) 180*fb5de435SRitesh Harjani (IBM) { 181*fb5de435SRitesh Harjani (IBM) loff_t pos = iocb->ki_pos; 182*fb5de435SRitesh Harjani (IBM) struct inode *inode = file_inode(iocb->ki_filp); 183*fb5de435SRitesh Harjani (IBM) 184*fb5de435SRitesh Harjani (IBM) if (error) 185*fb5de435SRitesh Harjani (IBM) goto out; 186*fb5de435SRitesh Harjani (IBM) 187*fb5de435SRitesh Harjani (IBM) /* 188*fb5de435SRitesh Harjani (IBM) * If we are extending the file, we have to update i_size here before 189*fb5de435SRitesh Harjani (IBM) * page cache gets invalidated in iomap_dio_rw(). This prevents racing 190*fb5de435SRitesh Harjani (IBM) * buffered reads from zeroing out too much from page cache pages. 191*fb5de435SRitesh Harjani (IBM) * Note that all extending writes always happens synchronously with 192*fb5de435SRitesh Harjani (IBM) * inode lock held by ext2_dio_write_iter(). So it is safe to update 193*fb5de435SRitesh Harjani (IBM) * inode size here for extending file writes. 194*fb5de435SRitesh Harjani (IBM) */ 195*fb5de435SRitesh Harjani (IBM) pos += size; 196*fb5de435SRitesh Harjani (IBM) if (pos > i_size_read(inode)) { 197*fb5de435SRitesh Harjani (IBM) i_size_write(inode, pos); 198*fb5de435SRitesh Harjani (IBM) mark_inode_dirty(inode); 199*fb5de435SRitesh Harjani (IBM) } 200*fb5de435SRitesh Harjani (IBM) out: 201*fb5de435SRitesh Harjani (IBM) return error; 202*fb5de435SRitesh Harjani (IBM) } 203*fb5de435SRitesh Harjani (IBM) 204*fb5de435SRitesh Harjani (IBM) static const struct iomap_dio_ops ext2_dio_write_ops = { 205*fb5de435SRitesh Harjani (IBM) .end_io = ext2_dio_write_end_io, 206*fb5de435SRitesh Harjani (IBM) }; 207*fb5de435SRitesh Harjani (IBM) 208*fb5de435SRitesh Harjani (IBM) static ssize_t ext2_dio_write_iter(struct kiocb *iocb, struct iov_iter *from) 209*fb5de435SRitesh Harjani (IBM) { 210*fb5de435SRitesh Harjani (IBM) struct file *file = iocb->ki_filp; 211*fb5de435SRitesh Harjani (IBM) struct inode *inode = file->f_mapping->host; 212*fb5de435SRitesh Harjani (IBM) ssize_t ret; 213*fb5de435SRitesh Harjani (IBM) unsigned int flags = 0; 214*fb5de435SRitesh Harjani (IBM) unsigned long blocksize = inode->i_sb->s_blocksize; 215*fb5de435SRitesh Harjani (IBM) loff_t offset = iocb->ki_pos; 216*fb5de435SRitesh Harjani (IBM) loff_t count = iov_iter_count(from); 217*fb5de435SRitesh Harjani (IBM) 218*fb5de435SRitesh Harjani (IBM) inode_lock(inode); 219*fb5de435SRitesh Harjani (IBM) ret = generic_write_checks(iocb, from); 220*fb5de435SRitesh Harjani (IBM) if (ret <= 0) 221*fb5de435SRitesh Harjani (IBM) goto out_unlock; 222*fb5de435SRitesh Harjani (IBM) 223*fb5de435SRitesh Harjani (IBM) ret = kiocb_modified(iocb); 224*fb5de435SRitesh Harjani (IBM) if (ret) 225*fb5de435SRitesh Harjani (IBM) goto out_unlock; 226*fb5de435SRitesh Harjani (IBM) 227*fb5de435SRitesh Harjani (IBM) /* use IOMAP_DIO_FORCE_WAIT for unaligned or extending writes */ 228*fb5de435SRitesh Harjani (IBM) if (iocb->ki_pos + iov_iter_count(from) > i_size_read(inode) || 229*fb5de435SRitesh Harjani (IBM) (!IS_ALIGNED(iocb->ki_pos | iov_iter_alignment(from), blocksize))) 230*fb5de435SRitesh Harjani (IBM) flags |= IOMAP_DIO_FORCE_WAIT; 231*fb5de435SRitesh Harjani (IBM) 232*fb5de435SRitesh Harjani (IBM) ret = iomap_dio_rw(iocb, from, &ext2_iomap_ops, &ext2_dio_write_ops, 233*fb5de435SRitesh Harjani (IBM) flags, NULL, 0); 234*fb5de435SRitesh Harjani (IBM) 235*fb5de435SRitesh Harjani (IBM) /* ENOTBLK is magic return value for fallback to buffered-io */ 236*fb5de435SRitesh Harjani (IBM) if (ret == -ENOTBLK) 237*fb5de435SRitesh Harjani (IBM) ret = 0; 238*fb5de435SRitesh Harjani (IBM) 239*fb5de435SRitesh Harjani (IBM) if (ret < 0 && ret != -EIOCBQUEUED) 240*fb5de435SRitesh Harjani (IBM) ext2_write_failed(inode->i_mapping, offset + count); 241*fb5de435SRitesh Harjani (IBM) 242*fb5de435SRitesh Harjani (IBM) /* handle case for partial write and for fallback to buffered write */ 243*fb5de435SRitesh Harjani (IBM) if (ret >= 0 && iov_iter_count(from)) { 244*fb5de435SRitesh Harjani (IBM) loff_t pos, endbyte; 245*fb5de435SRitesh Harjani (IBM) ssize_t status; 246*fb5de435SRitesh Harjani (IBM) int ret2; 247*fb5de435SRitesh Harjani (IBM) 248*fb5de435SRitesh Harjani (IBM) iocb->ki_flags &= ~IOCB_DIRECT; 249*fb5de435SRitesh Harjani (IBM) pos = iocb->ki_pos; 250*fb5de435SRitesh Harjani (IBM) status = generic_perform_write(iocb, from); 251*fb5de435SRitesh Harjani (IBM) if (unlikely(status < 0)) { 252*fb5de435SRitesh Harjani (IBM) ret = status; 253*fb5de435SRitesh Harjani (IBM) goto out_unlock; 254*fb5de435SRitesh Harjani (IBM) } 255*fb5de435SRitesh Harjani (IBM) 256*fb5de435SRitesh Harjani (IBM) iocb->ki_pos += status; 257*fb5de435SRitesh Harjani (IBM) ret += status; 258*fb5de435SRitesh Harjani (IBM) endbyte = pos + status - 1; 259*fb5de435SRitesh Harjani (IBM) ret2 = filemap_write_and_wait_range(inode->i_mapping, pos, 260*fb5de435SRitesh Harjani (IBM) endbyte); 261*fb5de435SRitesh Harjani (IBM) if (!ret2) 262*fb5de435SRitesh Harjani (IBM) invalidate_mapping_pages(inode->i_mapping, 263*fb5de435SRitesh Harjani (IBM) pos >> PAGE_SHIFT, 264*fb5de435SRitesh Harjani (IBM) endbyte >> PAGE_SHIFT); 265*fb5de435SRitesh Harjani (IBM) if (ret > 0) 266*fb5de435SRitesh Harjani (IBM) generic_write_sync(iocb, ret); 267*fb5de435SRitesh Harjani (IBM) } 268*fb5de435SRitesh Harjani (IBM) 269*fb5de435SRitesh Harjani (IBM) out_unlock: 270*fb5de435SRitesh Harjani (IBM) inode_unlock(inode); 271*fb5de435SRitesh Harjani (IBM) return ret; 272*fb5de435SRitesh Harjani (IBM) } 273*fb5de435SRitesh Harjani (IBM) 27425f4e702SChristoph Hellwig static ssize_t ext2_file_read_iter(struct kiocb *iocb, struct iov_iter *to) 27525f4e702SChristoph Hellwig { 27625f4e702SChristoph Hellwig #ifdef CONFIG_FS_DAX 27725f4e702SChristoph Hellwig if (IS_DAX(iocb->ki_filp->f_mapping->host)) 27825f4e702SChristoph Hellwig return ext2_dax_read_iter(iocb, to); 27925f4e702SChristoph Hellwig #endif 280*fb5de435SRitesh Harjani (IBM) if (iocb->ki_flags & IOCB_DIRECT) 281*fb5de435SRitesh Harjani (IBM) return ext2_dio_read_iter(iocb, to); 282*fb5de435SRitesh Harjani (IBM) 28325f4e702SChristoph Hellwig return generic_file_read_iter(iocb, to); 28425f4e702SChristoph Hellwig } 28525f4e702SChristoph Hellwig 28625f4e702SChristoph Hellwig static ssize_t ext2_file_write_iter(struct kiocb *iocb, struct iov_iter *from) 28725f4e702SChristoph Hellwig { 28825f4e702SChristoph Hellwig #ifdef CONFIG_FS_DAX 28925f4e702SChristoph Hellwig if (IS_DAX(iocb->ki_filp->f_mapping->host)) 29025f4e702SChristoph Hellwig return ext2_dax_write_iter(iocb, from); 29125f4e702SChristoph Hellwig #endif 292*fb5de435SRitesh Harjani (IBM) if (iocb->ki_flags & IOCB_DIRECT) 293*fb5de435SRitesh Harjani (IBM) return ext2_dio_write_iter(iocb, from); 294*fb5de435SRitesh Harjani (IBM) 29525f4e702SChristoph Hellwig return generic_file_write_iter(iocb, from); 29625f4e702SChristoph Hellwig } 29725f4e702SChristoph Hellwig 2984b6f5d20SArjan van de Ven const struct file_operations ext2_file_operations = { 2991da177e4SLinus Torvalds .llseek = generic_file_llseek, 30025f4e702SChristoph Hellwig .read_iter = ext2_file_read_iter, 30125f4e702SChristoph Hellwig .write_iter = ext2_file_write_iter, 30214f9f7b2SAndi Kleen .unlocked_ioctl = ext2_ioctl, 303e322ff07SDavid Howells #ifdef CONFIG_COMPAT 304e322ff07SDavid Howells .compat_ioctl = ext2_compat_ioctl, 305e322ff07SDavid Howells #endif 306f7ca90b1SMatthew Wilcox .mmap = ext2_file_mmap, 307907f4554SChristoph Hellwig .open = dquot_file_open, 3081da177e4SLinus Torvalds .release = ext2_release_file, 30948bde86dSJan Kara .fsync = ext2_fsync, 310dbe6ec81SToshi Kani .get_unmapped_area = thp_get_unmapped_area, 3115274f052SJens Axboe .splice_read = generic_file_splice_read, 3128d020765SAl Viro .splice_write = iter_file_splice_write, 3131da177e4SLinus Torvalds }; 3141da177e4SLinus Torvalds 315754661f1SArjan van de Ven const struct inode_operations ext2_file_inode_operations = { 3161da177e4SLinus Torvalds .listxattr = ext2_listxattr, 31793bc420eSyangerkun .getattr = ext2_getattr, 3181da177e4SLinus Torvalds .setattr = ext2_setattr, 319cac2f8b8SChristian Brauner .get_inode_acl = ext2_get_acl, 32064e178a7SChristoph Hellwig .set_acl = ext2_set_acl, 32168c9d702SJosef Bacik .fiemap = ext2_fiemap, 322aba405e3SMiklos Szeredi .fileattr_get = ext2_fileattr_get, 323aba405e3SMiklos Szeredi .fileattr_set = ext2_fileattr_set, 3241da177e4SLinus Torvalds }; 325