fs/sync.c

   1 /*
   2  * High-level sync()-related operations
   3  */
   4
   5 #include <linux/kernel.h>
   6 #include <linux/file.h>
   7 #include <linux/fs.h>
   8 #include <linux/module.h>
   9 #include <linux/sched.h>
  10 #include <linux/writeback.h>
  11 #include <linux/syscalls.h>
  12 #include <linux/linkage.h>
  13 #include <linux/pagemap.h>
  14 #include <linux/quotaops.h>
  15 #include <linux/buffer_head.h>
  16 #include "internal.h"
  17
  18 #define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
  19                         SYNC_FILE_RANGE_WAIT_AFTER)
  20
  21 /*
  22  * sync everything.  Start out by waking pdflush, because that writes back
  23  * all queues in parallel.
  24  */
  25 static void do_sync(unsigned long wait)
  26 {
  27         wakeup_pdflush(0);
  28         sync_inodes(0);         /* All mappings, inodes and their blockdevs */
  29         vfs_dq_sync(NULL);
  30         sync_inodes(wait);      /* Mappings, inodes and blockdevs, again. */
  31         sync_supers();          /* Write the superblocks */
  32         sync_filesystems(0);    /* Start syncing the filesystems */
  33         sync_filesystems(wait); /* Waitingly sync the filesystems */
  34         sync_blockdevs();
  35         if (!wait)
  36                 printk("Emergency Sync complete\n");
  37         if (unlikely(laptop_mode))
  38                 laptop_sync_completion();
  39 }
  40
  41 SYSCALL_DEFINE0(sync)
  42 {
  43         do_sync(1);
  44         return 0;
  45 }
  46
  47 static void do_sync_work(struct work_struct *work)
  48 {
  49         do_sync(0);
  50         kfree(work);
  51 }
  52
  53 void emergency_sync(void)
  54 {
  55         struct work_struct *work;
  56
  57         work = kmalloc(sizeof(*work), GFP_ATOMIC);
  58         if (work) {
  59                 INIT_WORK(work, do_sync_work);
  60                 schedule_work(work);
  61         }
  62 }
  63
  64 /*
  65  * Generic function to fsync a file.
  66  *
  67  * filp may be NULL if called via the msync of a vma.
  68  */
  69 int file_fsync(struct file *filp, struct dentry *dentry, int datasync)
  70 {
  71         struct inode * inode = dentry->d_inode;
  72         struct super_block * sb;
  73         int ret, err;
  74
  75         /* sync the inode to buffers */
  76         ret = write_inode_now(inode, 0);
  77
  78         /* sync the superblock to buffers */
  79         sb = inode->i_sb;
  80         lock_super(sb);
  81         if (sb->s_dirt && sb->s_op->write_super)
  82                 sb->s_op->write_super(sb);
  83         unlock_super(sb);
  84
  85         /* .. finally sync the buffers to disk */
  86         err = sync_blockdev(sb->s_bdev);
  87         if (!ret)
  88                 ret = err;
  89         return ret;
  90 }
  91
  92 /**
  93  * vfs_fsync - perform a fsync or fdatasync on a file
  94  * @file:               file to sync
  95  * @dentry:             dentry of @file
  96  * @data:               only perform a fdatasync operation
  97  *
  98  * Write back data and metadata for @file to disk.  If @datasync is
  99  * set only metadata needed to access modified file data is written.
 100  *
 101  * In case this function is called from nfsd @file may be %NULL and
 102  * only @dentry is set.  This can only happen when the filesystem
 103  * implements the export_operations API.
 104  */
 105 int vfs_fsync(struct file *file, struct dentry *dentry, int datasync)
 106 {
 107         const struct file_operations *fop;
 108         struct address_space *mapping;
 109         int err, ret;
 110
 111         /*
 112          * Get mapping and operations from the file in case we have
 113          * as file, or get the default values for them in case we
 114          * don't have a struct file available.  Damn nfsd..
 115          */
 116         if (file) {
 117                 mapping = file->f_mapping;
 118                 fop = file->f_op;
 119         } else {
 120                 mapping = dentry->d_inode->i_mapping;
 121                 fop = dentry->d_inode->i_fop;
 122         }
 123
 124         if (!fop || !fop->fsync) {
 125                 ret = -EINVAL;
 126                 goto out;
 127         }
 128
 129         ret = filemap_fdatawrite(mapping);
 130
 131         /*
 132          * We need to protect against concurrent writers, which could cause
 133          * livelocks in fsync_buffers_list().
 134          */
 135         mutex_lock(&mapping->host->i_mutex);
 136         err = fop->fsync(file, dentry, datasync);
 137         if (!ret)
 138                 ret = err;
 139         mutex_unlock(&mapping->host->i_mutex);
 140         err = filemap_fdatawait(mapping);
 141         if (!ret)
 142                 ret = err;
 143 out:
 144         return ret;
 145 }
 146 EXPORT_SYMBOL(vfs_fsync);
 147
 148 static int do_fsync(unsigned int fd, int datasync)
 149 {
 150         struct file *file;
 151         int ret = -EBADF;
 152
 153         file = fget(fd);
 154         if (file) {
 155                 ret = vfs_fsync(file, file->f_path.dentry, datasync);
 156                 fput(file);
 157         }
 158         return ret;
 159 }
 160
 161 SYSCALL_DEFINE1(fsync, unsigned int, fd)
 162 {
 163         return do_fsync(fd, 0);
 164 }
 165
 166 SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
 167 {
 168         return do_fsync(fd, 1);
 169 }
 170
 171 /*
 172  * sys_sync_file_range() permits finely controlled syncing over a segment of
 173  * a file in the range offset .. (offset+nbytes-1) inclusive.  If nbytes is
 174  * zero then sys_sync_file_range() will operate from offset out to EOF.
 175  *
 176  * The flag bits are:
 177  *
 178  * SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
 179  * before performing the write.
 180  *
 181  * SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
 182  * range which are not presently under writeback. Note that this may block for
 183  * significant periods due to exhaustion of disk request structures.
 184  *
 185  * SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
 186  * after performing the write.
 187  *
 188  * Useful combinations of the flag bits are:
 189  *
 190  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
 191  * in the range which were dirty on entry to sys_sync_file_range() are placed
 192  * under writeout.  This is a start-write-for-data-integrity operation.
 193  *
 194  * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
 195  * are not presently under writeout.  This is an asynchronous flush-to-disk
 196  * operation.  Not suitable for data integrity operations.
 197  *
 198  * SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
 199  * completion of writeout of all pages in the range.  This will be used after an
 200  * earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
 201  * for that operation to complete and to return the result.
 202  *
 203  * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
 204  * a traditional sync() operation.  This is a write-for-data-integrity operation
 205  * which will ensure that all pages in the range which were dirty on entry to
 206  * sys_sync_file_range() are committed to disk.
 207  *
 208  *
 209  * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
 210  * I/O errors or ENOSPC conditions and will return those to the caller, after
 211  * clearing the EIO and ENOSPC flags in the address_space.
 212  *
 213  * It should be noted that none of these operations write out the file's
 214  * metadata.  So unless the application is strictly performing overwrites of
 215  * already-instantiated disk blocks, there are no guarantees here that the data
 216  * will be available after a crash.
 217  */
 218 SYSCALL_DEFINE(sync_file_range)(int fd, loff_t offset, loff_t nbytes,
 219                                 unsigned int flags)
 220 {
 221         int ret;
 222         struct file *file;
 223         loff_t endbyte;                 /* inclusive */
 224         int fput_needed;
 225         umode_t i_mode;
 226
 227         ret = -EINVAL;
 228         if (flags & ~VALID_FLAGS)
 229                 goto out;
 230
 231         endbyte = offset + nbytes;
 232
 233         if ((s64)offset < 0)
 234                 goto out;
 235         if ((s64)endbyte < 0)
 236                 goto out;
 237         if (endbyte < offset)
 238                 goto out;
 239
 240         if (sizeof(pgoff_t) == 4) {
 241                 if (offset >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 242                         /*
 243                          * The range starts outside a 32 bit machine's
 244                          * pagecache addressing capabilities.  Let it "succeed"
 245                          */
 246                         ret = 0;
 247                         goto out;
 248                 }
 249                 if (endbyte >= (0x100000000ULL << PAGE_CACHE_SHIFT)) {
 250                         /*
 251                          * Out to EOF
 252                          */
 253                         nbytes = 0;
 254                 }
 255         }
 256
 257         if (nbytes == 0)
 258                 endbyte = LLONG_MAX;
 259         else
 260                 endbyte--;              /* inclusive */
 261
 262         ret = -EBADF;
 263         file = fget_light(fd, &fput_needed);
 264         if (!file)
 265                 goto out;
 266
 267         i_mode = file->f_path.dentry->d_inode->i_mode;
 268         ret = -ESPIPE;
 269         if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
 270                         !S_ISLNK(i_mode))
 271                 goto out_put;
 272
 273         ret = do_sync_mapping_range(file->f_mapping, offset, endbyte, flags);
 274 out_put:
 275         fput_light(file, fput_needed);
 276 out:
 277         return ret;
 278 }
 279 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 280 asmlinkage long SyS_sync_file_range(long fd, loff_t offset, loff_t nbytes,
 281                                     long flags)
 282 {
 283         return SYSC_sync_file_range((int) fd, offset, nbytes,
 284                                     (unsigned int) flags);
 285 }
 286 SYSCALL_ALIAS(sys_sync_file_range, SyS_sync_file_range);
 287 #endif
 288
 289 /* It would be nice if people remember that not all the world's an i386
 290    when they introduce new system calls */
 291 SYSCALL_DEFINE(sync_file_range2)(int fd, unsigned int flags,
 292                                  loff_t offset, loff_t nbytes)
 293 {
 294         return sys_sync_file_range(fd, offset, nbytes, flags);
 295 }
 296 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 297 asmlinkage long SyS_sync_file_range2(long fd, long flags,
 298                                      loff_t offset, loff_t nbytes)
 299 {
 300         return SYSC_sync_file_range2((int) fd, (unsigned int) flags,
 301                                      offset, nbytes);
 302 }
 303 SYSCALL_ALIAS(sys_sync_file_range2, SyS_sync_file_range2);
 304 #endif
 305
 306 /*
 307  * `endbyte' is inclusive
 308  */
 309 int do_sync_mapping_range(struct address_space *mapping, loff_t offset,
 310                           loff_t endbyte, unsigned int flags)
 311 {
 312         int ret;
 313
 314         if (!mapping) {
 315                 ret = -EINVAL;
 316                 goto out;
 317         }
 318
 319         ret = 0;
 320         if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
 321                 ret = wait_on_page_writeback_range(mapping,
 322                                         offset >> PAGE_CACHE_SHIFT,
 323                                         endbyte >> PAGE_CACHE_SHIFT);
 324                 if (ret < 0)
 325                         goto out;
 326         }
 327
 328         if (flags & SYNC_FILE_RANGE_WRITE) {
 329                 ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
 330                                                 WB_SYNC_ALL);
 331                 if (ret < 0)
 332                         goto out;
 333         }
 334
 335         if (flags & SYNC_FILE_RANGE_WAIT_AFTER) {
 336                 ret = wait_on_page_writeback_range(mapping,
 337                                         offset >> PAGE_CACHE_SHIFT,
 338                                         endbyte >> PAGE_CACHE_SHIFT);
 339         }
 340 out:
 341         return ret;
 342 }
 343 EXPORT_SYMBOL_GPL(do_sync_mapping_range);