fs/ext2/file.c

   1 /*
   2  *  linux/fs/ext2/file.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/file.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  ext2 fs regular file handling primitives
  16  *
  17  *  64-bit file support on 64-bit platforms by Jakub Jelinek
  18  *      (jj@sunsite.ms.mff.cuni.cz)
  19  */
  20
  21 #include <linux/time.h>
  22 #include <linux/pagemap.h>
  23 #include <linux/dax.h>
  24 #include <linux/quotaops.h>
  25 #include <linux/iomap.h>
  26 #include <linux/uio.h>
  27 #include "ext2.h"
  28 #include "xattr.h"
  29 #include "acl.h"
  30
  31 #ifdef CONFIG_FS_DAX
  32 static ssize_t ext2_dax_read_iter(struct kiocb *iocb, struct iov_iter *to)
  33 {
  34         struct inode *inode = iocb->ki_filp->f_mapping->host;
  35         ssize_t ret;
  36
  37         if (!iov_iter_count(to))
  38                 return 0; /* skip atime */
  39
  40         inode_lock_shared(inode);
  41         ret = dax_iomap_rw(iocb, to, &ext2_iomap_ops);
  42         inode_unlock_shared(inode);
  43
  44         file_accessed(iocb->ki_filp);
  45         return ret;
  46 }
  47
  48 static ssize_t ext2_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
  49 {
  50         struct file *file = iocb->ki_filp;
  51         struct inode *inode = file->f_mapping->host;
  52         ssize_t ret;
  53
  54         inode_lock(inode);
  55         ret = generic_write_checks(iocb, from);
  56         if (ret <= 0)
  57                 goto out_unlock;
  58         ret = file_remove_privs(file);
  59         if (ret)
  60                 goto out_unlock;
  61         ret = file_update_time(file);
  62         if (ret)
  63                 goto out_unlock;
  64
  65         ret = dax_iomap_rw(iocb, from, &ext2_iomap_ops);
  66         if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
  67                 i_size_write(inode, iocb->ki_pos);
  68                 mark_inode_dirty(inode);
  69         }
  70
  71 out_unlock:
  72         inode_unlock(inode);
  73         if (ret > 0)
  74                 ret = generic_write_sync(iocb, ret);
  75         return ret;
  76 }
  77
  78 /*
  79  * The lock ordering for ext2 DAX fault paths is:
  80  *
  81  * mmap_sem (MM)
  82  *   sb_start_pagefault (vfs, freeze)
  83  *     ext2_inode_info->dax_sem
  84  *       address_space->i_mmap_rwsem or page_lock (mutually exclusive in DAX)
  85  *         ext2_inode_info->truncate_mutex
  86  *
  87  * The default page_lock and i_size verification done by non-DAX fault paths
  88  * is sufficient because ext2 doesn't support hole punching.
  89  */
  90 static int ext2_dax_fault(struct vm_fault *vmf)
  91 {
  92         struct inode *inode = file_inode(vmf->vma->vm_file);
  93         struct ext2_inode_info *ei = EXT2_I(inode);
  94         int ret;
  95
  96         if (vmf->flags & FAULT_FLAG_WRITE) {
  97                 sb_start_pagefault(inode->i_sb);
  98                 file_update_time(vmf->vma->vm_file);
  99         }
 100         down_read(&ei->dax_sem);
 101
 102         ret = dax_iomap_fault(vmf, PE_SIZE_PTE, &ext2_iomap_ops);
 103
 104         up_read(&ei->dax_sem);
 105         if (vmf->flags & FAULT_FLAG_WRITE)
 106                 sb_end_pagefault(inode->i_sb);
 107         return ret;
 108 }
 109
 110 static int ext2_dax_pfn_mkwrite(struct vm_fault *vmf)
 111 {
 112         struct inode *inode = file_inode(vmf->vma->vm_file);
 113         struct ext2_inode_info *ei = EXT2_I(inode);
 114         loff_t size;
 115         int ret;
 116
 117         sb_start_pagefault(inode->i_sb);
 118         file_update_time(vmf->vma->vm_file);
 119         down_read(&ei->dax_sem);
 120
 121         /* check that the faulting page hasn't raced with truncate */
 122         size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
 123         if (vmf->pgoff >= size)
 124                 ret = VM_FAULT_SIGBUS;
 125         else
 126                 ret = dax_pfn_mkwrite(vmf);
 127
 128         up_read(&ei->dax_sem);
 129         sb_end_pagefault(inode->i_sb);
 130         return ret;
 131 }
 132
 133 static const struct vm_operations_struct ext2_dax_vm_ops = {
 134         .fault          = ext2_dax_fault,
 135         /*
 136          * .huge_fault is not supported for DAX because allocation in ext2
 137          * cannot be reliably aligned to huge page sizes and so pmd faults
 138          * will always fail and fail back to regular faults.
 139          */
 140         .page_mkwrite   = ext2_dax_fault,
 141         .pfn_mkwrite    = ext2_dax_pfn_mkwrite,
 142 };
 143
 144 static int ext2_file_mmap(struct file *file, struct vm_area_struct *vma)
 145 {
 146         if (!IS_DAX(file_inode(file)))
 147                 return generic_file_mmap(file, vma);
 148
 149         file_accessed(file);
 150         vma->vm_ops = &ext2_dax_vm_ops;
 151         vma->vm_flags |= VM_MIXEDMAP;
 152         return 0;
 153 }
 154 #else
 155 #define ext2_file_mmap  generic_file_mmap
 156 #endif
 157
 158 /*
 159  * Called when filp is released. This happens when all file descriptors
 160  * for a single struct file are closed. Note that different open() calls
 161  * for the same file yield different struct file structures.
 162  */
 163 static int ext2_release_file (struct inode * inode, struct file * filp)
 164 {
 165         if (filp->f_mode & FMODE_WRITE) {
 166                 mutex_lock(&EXT2_I(inode)->truncate_mutex);
 167                 ext2_discard_reservation(inode);
 168                 mutex_unlock(&EXT2_I(inode)->truncate_mutex);
 169         }
 170         return 0;
 171 }
 172
 173 int ext2_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 174 {
 175         int ret;
 176         struct super_block *sb = file->f_mapping->host->i_sb;
 177         struct address_space *mapping = sb->s_bdev->bd_inode->i_mapping;
 178
 179         ret = generic_file_fsync(file, start, end, datasync);
 180         if (ret == -EIO || test_and_clear_bit(AS_EIO, &mapping->flags)) {
 181                 /* We don't really know where the IO error happened... */
 182                 ext2_error(sb, __func__,
 183                            "detected IO error when writing metadata buffers");
 184                 ret = -EIO;
 185         }
 186         return ret;
 187 }
 188
 189 static ssize_t ext2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
 190 {
 191 #ifdef CONFIG_FS_DAX
 192         if (IS_DAX(iocb->ki_filp->f_mapping->host))
 193                 return ext2_dax_read_iter(iocb, to);
 194 #endif
 195         return generic_file_read_iter(iocb, to);
 196 }
 197
 198 static ssize_t ext2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 199 {
 200 #ifdef CONFIG_FS_DAX
 201         if (IS_DAX(iocb->ki_filp->f_mapping->host))
 202                 return ext2_dax_write_iter(iocb, from);
 203 #endif
 204         return generic_file_write_iter(iocb, from);
 205 }
 206
 207 const struct file_operations ext2_file_operations = {
 208         .llseek         = generic_file_llseek,
 209         .read_iter      = ext2_file_read_iter,
 210         .write_iter     = ext2_file_write_iter,
 211         .unlocked_ioctl = ext2_ioctl,
 212 #ifdef CONFIG_COMPAT
 213         .compat_ioctl   = ext2_compat_ioctl,
 214 #endif
 215         .mmap           = ext2_file_mmap,
 216         .open           = dquot_file_open,
 217         .release        = ext2_release_file,
 218         .fsync          = ext2_fsync,
 219         .get_unmapped_area = thp_get_unmapped_area,
 220         .splice_read    = generic_file_splice_read,
 221         .splice_write   = iter_file_splice_write,
 222 };
 223
 224 const struct inode_operations ext2_file_inode_operations = {
 225 #ifdef CONFIG_EXT2_FS_XATTR
 226         .listxattr      = ext2_listxattr,
 227 #endif
 228         .setattr        = ext2_setattr,
 229         .get_acl        = ext2_get_acl,
 230         .set_acl        = ext2_set_acl,
 231         .fiemap         = ext2_fiemap,
 232 };