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
  32 /*
  33  * Called when an inode is released. Note that this is different
  34  * from ext4_file_open: open gets called at every open, but release
  35  * gets called only when /all/ the files are closed.
  36  */
  37 static int ext4_release_file(struct inode *inode, struct file *filp)
  38 {
  39         if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
  40                 ext4_alloc_da_blocks(inode);
  41                 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
  42         }
  43         /* if we are the last writer on the inode, drop the block reservation */
  44         if ((filp->f_mode & FMODE_WRITE) &&
  45                         (atomic_read(&inode->i_writecount) == 1) &&
  46                         !EXT4_I(inode)->i_reserved_data_blocks)
  47         {
  48                 down_write(&EXT4_I(inode)->i_data_sem);
  49                 ext4_discard_preallocations(inode);
  50                 up_write(&EXT4_I(inode)->i_data_sem);
  51         }
  52         if (is_dx(inode) && filp->private_data)
  53                 ext4_htree_free_dir_info(filp->private_data);
  54
  55         return 0;
  56 }
  57
  58 static void ext4_aiodio_wait(struct inode *inode)
  59 {
  60         wait_queue_head_t *wq = ext4_ioend_wq(inode);
  61
  62         wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_aiodio_unwritten) == 0));
  63 }
  64
  65 /*
  66  * This tests whether the IO in question is block-aligned or not.
  67  * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
  68  * are converted to written only after the IO is complete.  Until they are
  69  * mapped, these blocks appear as holes, so dio_zero_block() will assume that
  70  * it needs to zero out portions of the start and/or end block.  If 2 AIO
  71  * threads are at work on the same unwritten block, they must be synchronized
  72  * or one thread will zero the other's data, causing corruption.
  73  */
  74 static int
  75 ext4_unaligned_aio(struct inode *inode, const struct iovec *iov,
  76                    unsigned long nr_segs, loff_t pos)
  77 {
  78         struct super_block *sb = inode->i_sb;
  79         int blockmask = sb->s_blocksize - 1;
  80         size_t count = iov_length(iov, nr_segs);
  81         loff_t final_size = pos + count;
  82
  83         if (pos >= inode->i_size)
  84                 return 0;
  85
  86         if ((pos & blockmask) || (final_size & blockmask))
  87                 return 1;
  88
  89         return 0;
  90 }
  91
  92 static ssize_t
  93 ext4_file_dio_write(struct kiocb *iocb, const struct iovec *iov,
  94                     unsigned long nr_segs, loff_t pos)
  95 {
  96         struct file *file = iocb->ki_filp;
  97         struct inode *inode = file->f_mapping->host;
  98         struct blk_plug plug;
  99         int unaligned_aio = 0;
 100         ssize_t ret;
 101         int overwrite = 0;
 102         size_t length = iov_length(iov, nr_segs);
 103
 104         if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
 105             !is_sync_kiocb(iocb))
 106                 unaligned_aio = ext4_unaligned_aio(inode, iov, nr_segs, pos);
 107
 108         /* Unaligned direct AIO must be serialized; see comment above */
 109         if (unaligned_aio) {
 110                 static unsigned long unaligned_warn_time;
 111
 112                 /* Warn about this once per day */
 113                 if (printk_timed_ratelimit(&unaligned_warn_time, 60*60*24*HZ))
 114                         ext4_msg(inode->i_sb, KERN_WARNING,
 115                                  "Unaligned AIO/DIO on inode %ld by %s; "
 116                                  "performance will be poor.",
 117                                  inode->i_ino, current->comm);
 118                 mutex_lock(ext4_aio_mutex(inode));
 119                 ext4_aiodio_wait(inode);
 120         }
 121
 122         BUG_ON(iocb->ki_pos != pos);
 123
 124         mutex_lock(&inode->i_mutex);
 125         blk_start_plug(&plug);
 126
 127         iocb->private = &overwrite;
 128
 129         /* check whether we do a DIO overwrite or not */
 130         if (ext4_should_dioread_nolock(inode) && !unaligned_aio &&
 131             !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
 132                 struct ext4_map_blocks map;
 133                 unsigned int blkbits = inode->i_blkbits;
 134                 int err, len;
 135
 136                 map.m_lblk = pos >> blkbits;
 137                 map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
 138                         - map.m_lblk;
 139                 len = map.m_len;
 140
 141                 err = ext4_map_blocks(NULL, inode, &map, 0);
 142                 /*
 143                  * 'err==len' means that all of blocks has been preallocated no
 144                  * matter they are initialized or not.  For excluding
 145                  * uninitialized extents, we need to check m_flags.  There are
 146                  * two conditions that indicate for initialized extents.
 147                  * 1) If we hit extent cache, EXT4_MAP_MAPPED flag is returned;
 148                  * 2) If we do a real lookup, non-flags are returned.
 149                  * So we should check these two conditions.
 150                  */
 151                 if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
 152                         overwrite = 1;
 153         }
 154
 155         ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
 156         mutex_unlock(&inode->i_mutex);
 157
 158         if (ret > 0 || ret == -EIOCBQUEUED) {
 159                 ssize_t err;
 160
 161                 err = generic_write_sync(file, pos, ret);
 162                 if (err < 0 && ret > 0)
 163                         ret = err;
 164         }
 165         blk_finish_plug(&plug);
 166
 167         if (unaligned_aio)
 168                 mutex_unlock(ext4_aio_mutex(inode));
 169
 170         return ret;
 171 }
 172
 173 static ssize_t
 174 ext4_file_write(struct kiocb *iocb, const struct iovec *iov,
 175                 unsigned long nr_segs, loff_t pos)
 176 {
 177         struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
 178         ssize_t ret;
 179
 180         /*
 181          * If we have encountered a bitmap-format file, the size limit
 182          * is smaller than s_maxbytes, which is for extent-mapped files.
 183          */
 184
 185         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
 186                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 187                 size_t length = iov_length(iov, nr_segs);
 188
 189                 if ((pos > sbi->s_bitmap_maxbytes ||
 190                     (pos == sbi->s_bitmap_maxbytes && length > 0)))
 191                         return -EFBIG;
 192
 193                 if (pos + length > sbi->s_bitmap_maxbytes) {
 194                         nr_segs = iov_shorten((struct iovec *)iov, nr_segs,
 195                                               sbi->s_bitmap_maxbytes - pos);
 196                 }
 197         }
 198
 199         if (unlikely(iocb->ki_filp->f_flags & O_DIRECT))
 200                 ret = ext4_file_dio_write(iocb, iov, nr_segs, pos);
 201         else
 202                 ret = generic_file_aio_write(iocb, iov, nr_segs, pos);
 203
 204         return ret;
 205 }
 206
 207 static const struct vm_operations_struct ext4_file_vm_ops = {
 208         .fault          = filemap_fault,
 209         .page_mkwrite   = ext4_page_mkwrite,
 210 };
 211
 212 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
 213 {
 214         struct address_space *mapping = file->f_mapping;
 215
 216         if (!mapping->a_ops->readpage)
 217                 return -ENOEXEC;
 218         file_accessed(file);
 219         vma->vm_ops = &ext4_file_vm_ops;
 220         vma->vm_flags |= VM_CAN_NONLINEAR;
 221         return 0;
 222 }
 223
 224 static int ext4_file_open(struct inode * inode, struct file * filp)
 225 {
 226         struct super_block *sb = inode->i_sb;
 227         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 228         struct ext4_inode_info *ei = EXT4_I(inode);
 229         struct vfsmount *mnt = filp->f_path.mnt;
 230         struct path path;
 231         char buf[64], *cp;
 232
 233         if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
 234                      !(sb->s_flags & MS_RDONLY))) {
 235                 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
 236                 /*
 237                  * Sample where the filesystem has been mounted and
 238                  * store it in the superblock for sysadmin convenience
 239                  * when trying to sort through large numbers of block
 240                  * devices or filesystem images.
 241                  */
 242                 memset(buf, 0, sizeof(buf));
 243                 path.mnt = mnt;
 244                 path.dentry = mnt->mnt_root;
 245                 cp = d_path(&path, buf, sizeof(buf));
 246                 if (!IS_ERR(cp)) {
 247                         handle_t *handle;
 248                         int err;
 249
 250                         handle = ext4_journal_start_sb(sb, 1);
 251                         if (IS_ERR(handle))
 252                                 return PTR_ERR(handle);
 253                         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
 254                         if (err) {
 255                                 ext4_journal_stop(handle);
 256                                 return err;
 257                         }
 258                         strlcpy(sbi->s_es->s_last_mounted, cp,
 259                                 sizeof(sbi->s_es->s_last_mounted));
 260                         ext4_handle_dirty_super(handle, sb);
 261                         ext4_journal_stop(handle);
 262                 }
 263         }
 264         /*
 265          * Set up the jbd2_inode if we are opening the inode for
 266          * writing and the journal is present
 267          */
 268         if (sbi->s_journal && !ei->jinode && (filp->f_mode & FMODE_WRITE)) {
 269                 struct jbd2_inode *jinode = jbd2_alloc_inode(GFP_KERNEL);
 270
 271                 spin_lock(&inode->i_lock);
 272                 if (!ei->jinode) {
 273                         if (!jinode) {
 274                                 spin_unlock(&inode->i_lock);
 275                                 return -ENOMEM;
 276                         }
 277                         ei->jinode = jinode;
 278                         jbd2_journal_init_jbd_inode(ei->jinode, inode);
 279                         jinode = NULL;
 280                 }
 281                 spin_unlock(&inode->i_lock);
 282                 if (unlikely(jinode != NULL))
 283                         jbd2_free_inode(jinode);
 284         }
 285         return dquot_file_open(inode, filp);
 286 }
 287
 288 /*
 289  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
 290  * by calling generic_file_llseek_size() with the appropriate maxbytes
 291  * value for each.
 292  */
 293 loff_t ext4_llseek(struct file *file, loff_t offset, int origin)
 294 {
 295         struct inode *inode = file->f_mapping->host;
 296         loff_t maxbytes;
 297
 298         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 299                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
 300         else
 301                 maxbytes = inode->i_sb->s_maxbytes;
 302
 303         return generic_file_llseek_size(file, offset, origin,
 304                                         maxbytes, i_size_read(inode));
 305 }
 306
 307 const struct file_operations ext4_file_operations = {
 308         .llseek         = ext4_llseek,
 309         .read           = do_sync_read,
 310         .write          = do_sync_write,
 311         .aio_read       = generic_file_aio_read,
 312         .aio_write      = ext4_file_write,
 313         .unlocked_ioctl = ext4_ioctl,
 314 #ifdef CONFIG_COMPAT
 315         .compat_ioctl   = ext4_compat_ioctl,
 316 #endif
 317         .mmap           = ext4_file_mmap,
 318         .open           = ext4_file_open,
 319         .release        = ext4_release_file,
 320         .fsync          = ext4_sync_file,
 321         .splice_read    = generic_file_splice_read,
 322         .splice_write   = generic_file_splice_write,
 323         .fallocate      = ext4_fallocate,
 324 };
 325
 326 const struct inode_operations ext4_file_inode_operations = {
 327         .setattr        = ext4_setattr,
 328         .getattr        = ext4_getattr,
 329 #ifdef CONFIG_EXT4_FS_XATTR
 330         .setxattr       = generic_setxattr,
 331         .getxattr       = generic_getxattr,
 332         .listxattr      = ext4_listxattr,
 333         .removexattr    = generic_removexattr,
 334 #endif
 335         .get_acl        = ext4_get_acl,
 336         .fiemap         = ext4_fiemap,
 337 };
 338