fs/ext4/file.c

   1 /*
   2  *  linux/fs/ext4/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  *  ext4 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/fs.h>
  23 #include <linux/jbd2.h>
  24 #include <linux/mount.h>
  25 #include <linux/path.h>
  26 #include <linux/quotaops.h>
  27 #include "ext4.h"
  28 #include "ext4_jbd2.h"
  29 #include "xattr.h"
  30 #include "acl.h"
  31 #include "richacl.h"
  32
  33 /*
  34  * Called when an inode is released. Note that this is different
  35  * from ext4_file_open: open gets called at every open, but release
  36  * gets called only when /all/ the files are closed.
  37  */
  38 static int ext4_release_file(struct inode *inode, struct file *filp)
  39 {
  40         if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
  41                 ext4_alloc_da_blocks(inode);
  42                 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
  43         }
  44         /* if we are the last writer on the inode, drop the block reservation */
  45         if ((filp->f_mode & FMODE_WRITE) &&
  46                         (atomic_read(&inode->i_writecount) == 1) &&
  47                         !EXT4_I(inode)->i_reserved_data_blocks)
  48         {
  49                 down_write(&EXT4_I(inode)->i_data_sem);
  50                 ext4_discard_preallocations(inode);
  51                 up_write(&EXT4_I(inode)->i_data_sem);
  52         }
  53         if (is_dx(inode) && filp->private_data)
  54                 ext4_htree_free_dir_info(filp->private_data);
  55
  56         return 0;
  57 }
  58
  59 static void ext4_aiodio_wait(struct inode *inode)
  60 {
  61         wait_queue_head_t *wq = ext4_ioend_wq(inode);
  62
  63         wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_aiodio_unwritten) == 0));
  64 }
  65
  66 /*
  67  * This tests whether the IO in question is block-aligned or not.
  68  * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
  69  * are converted to written only after the IO is complete.  Until they are
  70  * mapped, these blocks appear as holes, so dio_zero_block() will assume that
  71  * it needs to zero out portions of the start and/or end block.  If 2 AIO
  72  * threads are at work on the same unwritten block, they must be synchronized
  73  * or one thread will zero the other's data, causing corruption.
  74  */
  75 static int
  76 ext4_unaligned_aio(struct inode *inode, const struct iovec *iov,
  77                    unsigned long nr_segs, loff_t pos)
  78 {
  79         struct super_block *sb = inode->i_sb;
  80         int blockmask = sb->s_blocksize - 1;
  81         size_t count = iov_length(iov, nr_segs);
  82         loff_t final_size = pos + count;
  83
  84         if (pos >= inode->i_size)
  85                 return 0;
  86
  87         if ((pos & blockmask) || (final_size & blockmask))
  88                 return 1;
  89
  90         return 0;
  91 }
  92
  93 static ssize_t
  94 ext4_file_write(struct kiocb *iocb, const struct iovec *iov,
  95                 unsigned long nr_segs, loff_t pos)
  96 {
  97         struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
  98         int unaligned_aio = 0;
  99         int ret;
 100
 101         /*
 102          * If we have encountered a bitmap-format file, the size limit
 103          * is smaller than s_maxbytes, which is for extent-mapped files.
 104          */
 105
 106         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
 107                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 108                 size_t length = iov_length(iov, nr_segs);
 109
 110                 if ((pos > sbi->s_bitmap_maxbytes ||
 111                     (pos == sbi->s_bitmap_maxbytes && length > 0)))
 112                         return -EFBIG;
 113
 114                 if (pos + length > sbi->s_bitmap_maxbytes) {
 115                         nr_segs = iov_shorten((struct iovec *)iov, nr_segs,
 116                                               sbi->s_bitmap_maxbytes - pos);
 117                 }
 118         } else if (unlikely((iocb->ki_filp->f_flags & O_DIRECT) &&
 119                    !is_sync_kiocb(iocb))) {
 120                 unaligned_aio = ext4_unaligned_aio(inode, iov, nr_segs, pos);
 121         }
 122
 123         /* Unaligned direct AIO must be serialized; see comment above */
 124         if (unaligned_aio) {
 125                 static unsigned long unaligned_warn_time;
 126
 127                 /* Warn about this once per day */
 128                 if (printk_timed_ratelimit(&unaligned_warn_time, 60*60*24*HZ))
 129                         ext4_msg(inode->i_sb, KERN_WARNING,
 130                                  "Unaligned AIO/DIO on inode %ld by %s; "
 131                                  "performance will be poor.",
 132                                  inode->i_ino, current->comm);
 133                 mutex_lock(ext4_aio_mutex(inode));
 134                 ext4_aiodio_wait(inode);
 135         }
 136
 137         ret = generic_file_aio_write(iocb, iov, nr_segs, pos);
 138
 139         if (unaligned_aio)
 140                 mutex_unlock(ext4_aio_mutex(inode));
 141
 142         return ret;
 143 }
 144
 145 static const struct vm_operations_struct ext4_file_vm_ops = {
 146         .fault          = filemap_fault,
 147         .page_mkwrite   = ext4_page_mkwrite,
 148 };
 149
 150 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
 151 {
 152         struct address_space *mapping = file->f_mapping;
 153
 154         if (!mapping->a_ops->readpage)
 155                 return -ENOEXEC;
 156         file_accessed(file);
 157         vma->vm_ops = &ext4_file_vm_ops;
 158         vma->vm_flags |= VM_CAN_NONLINEAR;
 159         return 0;
 160 }
 161
 162 static int ext4_file_open(struct inode * inode, struct file * filp)
 163 {
 164         struct super_block *sb = inode->i_sb;
 165         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 166         struct ext4_inode_info *ei = EXT4_I(inode);
 167         struct vfsmount *mnt = filp->f_path.mnt;
 168         struct path path;
 169         char buf[64], *cp;
 170
 171         if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
 172                      !(sb->s_flags & MS_RDONLY))) {
 173                 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
 174                 /*
 175                  * Sample where the filesystem has been mounted and
 176                  * store it in the superblock for sysadmin convenience
 177                  * when trying to sort through large numbers of block
 178                  * devices or filesystem images.
 179                  */
 180                 memset(buf, 0, sizeof(buf));
 181                 path.mnt = mnt;
 182                 path.dentry = mnt->mnt_root;
 183                 cp = d_path(&path, buf, sizeof(buf));
 184                 if (!IS_ERR(cp)) {
 185                         strlcpy(sbi->s_es->s_last_mounted, cp,
 186                                 sizeof(sbi->s_es->s_last_mounted));
 187                         ext4_mark_super_dirty(sb);
 188                 }
 189         }
 190         /*
 191          * Set up the jbd2_inode if we are opening the inode for
 192          * writing and the journal is present
 193          */
 194         if (sbi->s_journal && !ei->jinode && (filp->f_mode & FMODE_WRITE)) {
 195                 struct jbd2_inode *jinode = jbd2_alloc_inode(GFP_KERNEL);
 196
 197                 spin_lock(&inode->i_lock);
 198                 if (!ei->jinode) {
 199                         if (!jinode) {
 200                                 spin_unlock(&inode->i_lock);
 201                                 return -ENOMEM;
 202                         }
 203                         ei->jinode = jinode;
 204                         jbd2_journal_init_jbd_inode(ei->jinode, inode);
 205                         jinode = NULL;
 206                 }
 207                 spin_unlock(&inode->i_lock);
 208                 if (unlikely(jinode != NULL))
 209                         jbd2_free_inode(jinode);
 210         }
 211         return dquot_file_open(inode, filp);
 212 }
 213
 214 /*
 215  * ext4_llseek() copied from generic_file_llseek() to handle both
 216  * block-mapped and extent-mapped maxbytes values. This should
 217  * otherwise be identical with generic_file_llseek().
 218  */
 219 loff_t ext4_llseek(struct file *file, loff_t offset, int origin)
 220 {
 221         struct inode *inode = file->f_mapping->host;
 222         loff_t maxbytes;
 223
 224         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 225                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
 226         else
 227                 maxbytes = inode->i_sb->s_maxbytes;
 228
 229         return generic_file_llseek_size(file, offset, origin, maxbytes);
 230 }
 231
 232 const struct file_operations ext4_file_operations = {
 233         .llseek         = ext4_llseek,
 234         .read           = do_sync_read,
 235         .write          = do_sync_write,
 236         .aio_read       = generic_file_aio_read,
 237         .aio_write      = ext4_file_write,
 238         .unlocked_ioctl = ext4_ioctl,
 239 #ifdef CONFIG_COMPAT
 240         .compat_ioctl   = ext4_compat_ioctl,
 241 #endif
 242         .mmap           = ext4_file_mmap,
 243         .open           = ext4_file_open,
 244         .release        = ext4_release_file,
 245         .fsync          = ext4_sync_file,
 246         .splice_read    = generic_file_splice_read,
 247         .splice_write   = generic_file_splice_write,
 248         .fallocate      = ext4_fallocate,
 249 };
 250
 251 const struct inode_operations ext4_file_inode_operations = {
 252         .setattr        = ext4_setattr,
 253         .getattr        = ext4_getattr,
 254 #ifdef CONFIG_EXT4_FS_XATTR
 255         .setxattr       = generic_setxattr,
 256         .getxattr       = generic_getxattr,
 257         .listxattr      = ext4_listxattr,
 258         .removexattr    = generic_removexattr,
 259 #endif
 260         .get_acl        = ext4_get_acl,
 261         .fiemap         = ext4_fiemap,
 262         .permission     = ext4_permission,
 263         .may_create     = ext4_may_create,
 264         .may_delete     = ext4_may_delete,
 265 };
 266