module/zfs/zpl_file.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2011, Lawrence Livermore National Security, LLC.
  23  */
  24
  25
  26 #include <sys/zfs_vfsops.h>
  27 #include <sys/zfs_vnops.h>
  28 #include <sys/zfs_znode.h>
  29 #include <sys/zpl.h>
  30
  31
  32 static int
  33 zpl_open(struct inode *ip, struct file *filp)
  34 {
  35         cred_t *cr;
  36         int error;
  37
  38         cr = (cred_t *)get_current_cred();
  39         error = -zfs_open(ip, filp->f_mode, filp->f_flags, cr);
  40         put_cred(cr);
  41         ASSERT3S(error, <=, 0);
  42
  43         if (error)
  44                 return (error);
  45
  46         return generic_file_open(ip, filp);
  47 }
  48
  49 static int
  50 zpl_release(struct inode *ip, struct file *filp)
  51 {
  52         cred_t *cr;
  53         int error;
  54
  55         cr = (cred_t *)get_current_cred();
  56         error = -zfs_close(ip, filp->f_flags, cr);
  57         put_cred(cr);
  58         ASSERT3S(error, <=, 0);
  59
  60         return (error);
  61 }
  62
  63 static int
  64 zpl_readdir(struct file *filp, void *dirent, filldir_t filldir)
  65 {
  66         struct dentry *dentry = filp->f_path.dentry;
  67         cred_t *cr;
  68         int error;
  69
  70         cr = (cred_t *)get_current_cred();
  71         error = -zfs_readdir(dentry->d_inode, dirent, filldir,
  72             &filp->f_pos, cr);
  73         put_cred(cr);
  74         ASSERT3S(error, <=, 0);
  75
  76         return (error);
  77 }
  78
  79 ZPL_FSYNC_PROTO(zpl_fsync, filp, unused_dentry, datasync)
  80 {
  81         cred_t *cr;
  82         int error;
  83
  84         cr = (cred_t *)get_current_cred();
  85         error = -zfs_fsync(filp->f_path.dentry->d_inode, datasync, cr);
  86         put_cred(cr);
  87         ASSERT3S(error, <=, 0);
  88
  89         return (error);
  90 }
  91
  92 ssize_t
  93 zpl_read_common(struct inode *ip, const char *buf, size_t len, loff_t pos,
  94      uio_seg_t segment, int flags, cred_t *cr)
  95 {
  96         int error;
  97         struct iovec iov;
  98         uio_t uio;
  99
 100         iov.iov_base = (void *)buf;
 101         iov.iov_len = len;
 102
 103         uio.uio_iov = &iov;
 104         uio.uio_resid = len;
 105         uio.uio_iovcnt = 1;
 106         uio.uio_loffset = pos;
 107         uio.uio_limit = MAXOFFSET_T;
 108         uio.uio_segflg = segment;
 109
 110         error = -zfs_read(ip, &uio, flags, cr);
 111         if (error < 0)
 112                 return (error);
 113
 114         return (len - uio.uio_resid);
 115 }
 116
 117 static ssize_t
 118 zpl_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 119 {
 120         cred_t *cr;
 121         ssize_t read;
 122
 123         cr = (cred_t *)get_current_cred();
 124         read = zpl_read_common(filp->f_mapping->host, buf, len, *ppos,
 125             UIO_USERSPACE, filp->f_flags, cr);
 126         put_cred(cr);
 127
 128         if (read < 0)
 129                 return (read);
 130
 131         *ppos += read;
 132         return (read);
 133 }
 134
 135 ssize_t
 136 zpl_write_common(struct inode *ip, const char *buf, size_t len, loff_t pos,
 137     uio_seg_t segment, int flags, cred_t *cr)
 138 {
 139         int error;
 140         struct iovec iov;
 141         uio_t uio;
 142
 143         iov.iov_base = (void *)buf;
 144         iov.iov_len = len;
 145
 146         uio.uio_iov = &iov;
 147         uio.uio_resid = len,
 148         uio.uio_iovcnt = 1;
 149         uio.uio_loffset = pos;
 150         uio.uio_limit = MAXOFFSET_T;
 151         uio.uio_segflg = segment;
 152
 153         error = -zfs_write(ip, &uio, flags, cr);
 154         if (error < 0)
 155                 return (error);
 156
 157         return (len - uio.uio_resid);
 158 }
 159
 160 static ssize_t
 161 zpl_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
 162 {
 163         cred_t *cr;
 164         ssize_t wrote;
 165
 166         cr = (cred_t *)get_current_cred();
 167         wrote = zpl_write_common(filp->f_mapping->host, buf, len, *ppos,
 168             UIO_USERSPACE, filp->f_flags, cr);
 169         put_cred(cr);
 170
 171         if (wrote < 0)
 172                 return (wrote);
 173
 174         *ppos += wrote;
 175         return (wrote);
 176 }
 177
 178 /*
 179  * It's worth taking a moment to describe how mmap is implemented
 180  * for zfs because it differs considerably from other Linux filesystems.
 181  * However, this issue is handled the same way under OpenSolaris.
 182  *
 183  * The issue is that by design zfs bypasses the Linux page cache and
 184  * leaves all caching up to the ARC.  This has been shown to work
 185  * well for the common read(2)/write(2) case.  However, mmap(2)
 186  * is problem because it relies on being tightly integrated with the
 187  * page cache.  To handle this we cache mmap'ed files twice, once in
 188  * the ARC and a second time in the page cache.  The code is careful
 189  * to keep both copies synchronized.
 190  *
 191  * When a file with an mmap'ed region is written to using write(2)
 192  * both the data in the ARC and existing pages in the page cache
 193  * are updated.  For a read(2) data will be read first from the page
 194  * cache then the ARC if needed.  Neither a write(2) or read(2) will
 195  * will ever result in new pages being added to the page cache.
 196  *
 197  * New pages are added to the page cache only via .readpage() which
 198  * is called when the vfs needs to read a page off disk to back the
 199  * virtual memory region.  These pages may be modified without
 200  * notifying the ARC and will be written out periodically via
 201  * .writepage().  This will occur due to either a sync or the usual
 202  * page aging behavior.  Note because a read(2) of a mmap'ed file
 203  * will always check the page cache first even when the ARC is out
 204  * of date correct data will still be returned.
 205  *
 206  * While this implementation ensures correct behavior it does have
 207  * have some drawbacks.  The most obvious of which is that it
 208  * increases the required memory footprint when access mmap'ed
 209  * files.  It also adds additional complexity to the code keeping
 210  * both caches synchronized.
 211  *
 212  * Longer term it may be possible to cleanly resolve this wart by
 213  * mapping page cache pages directly on to the ARC buffers.  The
 214  * Linux address space operations are flexible enough to allow
 215  * selection of which pages back a particular index.  The trick
 216  * would be working out the details of which subsystem is in
 217  * charge, the ARC, the page cache, or both.  It may also prove
 218  * helpful to move the ARC buffers to a scatter-gather lists
 219  * rather than a vmalloc'ed region.
 220  */
 221 static int
 222 zpl_mmap(struct file *filp, struct vm_area_struct *vma)
 223 {
 224         znode_t *zp = ITOZ(filp->f_mapping->host);
 225         int error;
 226
 227         error = generic_file_mmap(filp, vma);
 228         if (error)
 229                 return (error);
 230
 231         mutex_enter(&zp->z_lock);
 232         zp->z_is_mapped = 1;
 233         mutex_exit(&zp->z_lock);
 234
 235         return (error);
 236 }
 237
 238 /*
 239  * Populate a page with data for the Linux page cache.  This function is
 240  * only used to support mmap(2).  There will be an identical copy of the
 241  * data in the ARC which is kept up to date via .write() and .writepage().
 242  *
 243  * Current this function relies on zpl_read_common() and the O_DIRECT
 244  * flag to read in a page.  This works but the more correct way is to
 245  * update zfs_fillpage() to be Linux friendly and use that interface.
 246  */
 247 static int
 248 zpl_readpage(struct file *filp, struct page *pp)
 249 {
 250         struct inode *ip;
 251         loff_t off, i_size;
 252         size_t len, wrote;
 253         cred_t *cr;
 254         void *pb;
 255         int error = 0;
 256
 257         ASSERT(PageLocked(pp));
 258         ip = pp->mapping->host;
 259         off = page_offset(pp);
 260         i_size = i_size_read(ip);
 261         ASSERT3S(off, <, i_size);
 262
 263         cr = (cred_t *)get_current_cred();
 264         len = MIN(PAGE_CACHE_SIZE, i_size - off);
 265
 266         pb = kmap(pp);
 267
 268         /* O_DIRECT is passed to bypass the page cache and avoid deadlock. */
 269         wrote = zpl_read_common(ip, pb, len, off, UIO_SYSSPACE, O_DIRECT, cr);
 270         if (wrote != len)
 271                 error = -EIO;
 272
 273         if (!error && (len < PAGE_CACHE_SIZE))
 274                 memset(pb + len, 0, PAGE_CACHE_SIZE - len);
 275
 276         kunmap(pp);
 277         put_cred(cr);
 278
 279         if (error) {
 280                 SetPageError(pp);
 281                 ClearPageUptodate(pp);
 282         } else {
 283                 ClearPageError(pp);
 284                 SetPageUptodate(pp);
 285                 flush_dcache_page(pp);
 286         }
 287
 288         unlock_page(pp);
 289
 290         return (error);
 291 }
 292
 293 /*
 294  * Write out dirty pages to the ARC, this function is only required to
 295  * support mmap(2).  Mapped pages may be dirtied by memory operations
 296  * which never call .write().  These dirty pages are kept in sync with
 297  * the ARC buffers via this hook.
 298  *
 299  * Currently this function relies on zpl_write_common() and the O_DIRECT
 300  * flag to push out the page.  This works but the more correct way is
 301  * to update zfs_putapage() to be Linux friendly and use that interface.
 302  */
 303 static int
 304 zpl_writepage(struct page *pp, struct writeback_control *wbc)
 305 {
 306         struct inode *ip;
 307         loff_t off, i_size;
 308         size_t len, read;
 309         cred_t *cr;
 310         void *pb;
 311         int error = 0;
 312
 313         ASSERT(PageLocked(pp));
 314         ip = pp->mapping->host;
 315         off = page_offset(pp);
 316         i_size = i_size_read(ip);
 317
 318         cr = (cred_t *)get_current_cred();
 319         len = MIN(PAGE_CACHE_SIZE, i_size - off);
 320
 321         pb = kmap(pp);
 322
 323         /* O_DIRECT is passed to bypass the page cache and avoid deadlock. */
 324         read = zpl_write_common(ip, pb, len, off, UIO_SYSSPACE, O_DIRECT, cr);
 325         if (read != len)
 326                 error = -EIO;
 327
 328         kunmap(pp);
 329         put_cred(cr);
 330
 331         if (error) {
 332                 SetPageError(pp);
 333                 ClearPageUptodate(pp);
 334         } else {
 335                 ClearPageError(pp);
 336                 SetPageUptodate(pp);
 337         }
 338
 339         unlock_page(pp);
 340
 341         return (error);
 342 }
 343
 344 const struct address_space_operations zpl_address_space_operations = {
 345         .readpage       = zpl_readpage,
 346         .writepage      = zpl_writepage,
 347 };
 348
 349 const struct file_operations zpl_file_operations = {
 350         .open           = zpl_open,
 351         .release        = zpl_release,
 352         .llseek         = generic_file_llseek,
 353         .read           = zpl_read,
 354         .write          = zpl_write,
 355         .readdir        = zpl_readdir,
 356         .mmap           = zpl_mmap,
 357         .fsync          = zpl_fsync,
 358 };
 359
 360 const struct file_operations zpl_dir_file_operations = {
 361         .llseek         = generic_file_llseek,
 362         .read           = generic_read_dir,
 363         .readdir        = zpl_readdir,
 364         .fsync          = zpl_fsync,
 365 };