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