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 = CRED();
  36         int error;
  37
  38         crhold(cr);
  39         error = -zfs_open(ip, filp->f_mode, filp->f_flags, cr);
  40         crfree(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 = CRED();
  53         int error;
  54
  55         crhold(cr);
  56         error = -zfs_close(ip, filp->f_flags, cr);
  57         crfree(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 = CRED();
  68         int error;
  69
  70         crhold(cr);
  71         error = -zfs_readdir(dentry->d_inode, dirent, filldir,
  72             &filp->f_pos, cr);
  73         crfree(cr);
  74         ASSERT3S(error, <=, 0);
  75
  76         return (error);
  77 }
  78
  79 /*
  80  * 2.6.35 API change,
  81  * As of 2.6.35 the dentry argument to the .fsync() vfs hook was deemed
  82  * redundant.  The dentry is still accessible via filp->f_path.dentry,
  83  * and we are guaranteed that filp will never be NULL.
  84  *
  85  * 2.6.34 API change,
  86  * Prior to 2.6.34 the nfsd kernel server would pass a NULL file struct *
  87  * to the .fsync() hook.  For this reason, we must be careful not to use
  88  * filp unconditionally in the 3 argument case.
  89  */
  90 #ifdef HAVE_2ARGS_FSYNC
  91 static int
  92 zpl_fsync(struct file *filp, int datasync)
  93 {
  94         struct dentry *dentry = filp->f_path.dentry;
  95 #else
  96 static int
  97 zpl_fsync(struct file *filp, struct dentry *dentry, int datasync)
  98 {
  99 #endif /* HAVE_2ARGS_FSYNC */
 100         cred_t *cr = CRED();
 101         int error;
 102
 103         crhold(cr);
 104         error = -zfs_fsync(dentry->d_inode, datasync, cr);
 105         crfree(cr);
 106         ASSERT3S(error, <=, 0);
 107
 108         return (error);
 109 }
 110
 111 ssize_t
 112 zpl_read_common(struct inode *ip, const char *buf, size_t len, loff_t pos,
 113      uio_seg_t segment, int flags, cred_t *cr)
 114 {
 115         int error;
 116         struct iovec iov;
 117         uio_t uio;
 118
 119         iov.iov_base = (void *)buf;
 120         iov.iov_len = len;
 121
 122         uio.uio_iov = &iov;
 123         uio.uio_resid = len;
 124         uio.uio_iovcnt = 1;
 125         uio.uio_loffset = pos;
 126         uio.uio_limit = MAXOFFSET_T;
 127         uio.uio_segflg = segment;
 128
 129         error = -zfs_read(ip, &uio, flags, cr);
 130         if (error < 0)
 131                 return (error);
 132
 133         return (len - uio.uio_resid);
 134 }
 135
 136 static ssize_t
 137 zpl_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 138 {
 139         cred_t *cr = CRED();
 140         ssize_t read;
 141
 142         crhold(cr);
 143         read = zpl_read_common(filp->f_mapping->host, buf, len, *ppos,
 144             UIO_USERSPACE, filp->f_flags, cr);
 145         crfree(cr);
 146
 147         if (read < 0)
 148                 return (read);
 149
 150         *ppos += read;
 151         return (read);
 152 }
 153
 154 ssize_t
 155 zpl_write_common(struct inode *ip, const char *buf, size_t len, loff_t pos,
 156     uio_seg_t segment, int flags, cred_t *cr)
 157 {
 158         int error;
 159         struct iovec iov;
 160         uio_t uio;
 161
 162         iov.iov_base = (void *)buf;
 163         iov.iov_len = len;
 164
 165         uio.uio_iov = &iov;
 166         uio.uio_resid = len,
 167         uio.uio_iovcnt = 1;
 168         uio.uio_loffset = pos;
 169         uio.uio_limit = MAXOFFSET_T;
 170         uio.uio_segflg = segment;
 171
 172         error = -zfs_write(ip, &uio, flags, cr);
 173         if (error < 0)
 174                 return (error);
 175
 176         return (len - uio.uio_resid);
 177 }
 178
 179 static ssize_t
 180 zpl_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
 181 {
 182         cred_t *cr = CRED();
 183         ssize_t wrote;
 184
 185         crhold(cr);
 186         wrote = zpl_write_common(filp->f_mapping->host, buf, len, *ppos,
 187             UIO_USERSPACE, filp->f_flags, cr);
 188         crfree(cr);
 189
 190         if (wrote < 0)
 191                 return (wrote);
 192
 193         *ppos += wrote;
 194         return (wrote);
 195 }
 196
 197 /*
 198  * It's worth taking a moment to describe how mmap is implemented
 199  * for zfs because it differs considerably from other Linux filesystems.
 200  * However, this issue is handled the same way under OpenSolaris.
 201  *
 202  * The issue is that by design zfs bypasses the Linux page cache and
 203  * leaves all caching up to the ARC.  This has been shown to work
 204  * well for the common read(2)/write(2) case.  However, mmap(2)
 205  * is problem because it relies on being tightly integrated with the
 206  * page cache.  To handle this we cache mmap'ed files twice, once in
 207  * the ARC and a second time in the page cache.  The code is careful
 208  * to keep both copies synchronized.
 209  *
 210  * When a file with an mmap'ed region is written to using write(2)
 211  * both the data in the ARC and existing pages in the page cache
 212  * are updated.  For a read(2) data will be read first from the page
 213  * cache then the ARC if needed.  Neither a write(2) or read(2) will
 214  * will ever result in new pages being added to the page cache.
 215  *
 216  * New pages are added to the page cache only via .readpage() which
 217  * is called when the vfs needs to read a page off disk to back the
 218  * virtual memory region.  These pages may be modified without
 219  * notifying the ARC and will be written out periodically via
 220  * .writepage().  This will occur due to either a sync or the usual
 221  * page aging behavior.  Note because a read(2) of a mmap'ed file
 222  * will always check the page cache first even when the ARC is out
 223  * of date correct data will still be returned.
 224  *
 225  * While this implementation ensures correct behavior it does have
 226  * have some drawbacks.  The most obvious of which is that it
 227  * increases the required memory footprint when access mmap'ed
 228  * files.  It also adds additional complexity to the code keeping
 229  * both caches synchronized.
 230  *
 231  * Longer term it may be possible to cleanly resolve this wart by
 232  * mapping page cache pages directly on to the ARC buffers.  The
 233  * Linux address space operations are flexible enough to allow
 234  * selection of which pages back a particular index.  The trick
 235  * would be working out the details of which subsystem is in
 236  * charge, the ARC, the page cache, or both.  It may also prove
 237  * helpful to move the ARC buffers to a scatter-gather lists
 238  * rather than a vmalloc'ed region.
 239  */
 240 static int
 241 zpl_mmap(struct file *filp, struct vm_area_struct *vma)
 242 {
 243         struct inode *ip = filp->f_mapping->host;
 244         znode_t *zp = ITOZ(ip);
 245         int error;
 246
 247         error = -zfs_map(ip, vma->vm_pgoff, (caddr_t *)vma->vm_start,
 248             (size_t)(vma->vm_end - vma->vm_start), vma->vm_flags);
 249         if (error)
 250                 return (error);
 251
 252         error = generic_file_mmap(filp, vma);
 253         if (error)
 254                 return (error);
 255
 256         mutex_enter(&zp->z_lock);
 257         zp->z_is_mapped = 1;
 258         mutex_exit(&zp->z_lock);
 259
 260         return (error);
 261 }
 262
 263 static struct page **
 264 pages_vector_from_list(struct list_head *pages, unsigned nr_pages)
 265 {
 266         struct page **pl;
 267         struct page *t;
 268         unsigned page_idx;
 269
 270         pl = kmalloc(sizeof(*pl) * nr_pages, GFP_NOFS);
 271         if (!pl)
 272                 return ERR_PTR(-ENOMEM);
 273
 274         page_idx = 0;
 275         list_for_each_entry_reverse(t, pages, lru) {
 276                 pl[page_idx] = t;
 277                 page_idx++;
 278         }
 279
 280         return pl;
 281 }
 282
 283 static int
 284 zpl_readpages(struct file *file, struct address_space *mapping,
 285         struct list_head *pages, unsigned nr_pages)
 286 {
 287         struct inode *ip;
 288         struct page  **pl;
 289         struct page  *p, *n;
 290         int          error;
 291
 292         ip = mapping->host;
 293
 294         pl = pages_vector_from_list(pages, nr_pages);
 295         if (IS_ERR(pl))
 296                 return PTR_ERR(pl);
 297
 298         error = -zfs_getpage(ip, pl, nr_pages);
 299         if (error)
 300                 goto error;
 301
 302         list_for_each_entry_safe_reverse(p, n, pages, lru) {
 303
 304                 list_del(&p->lru);
 305
 306                 flush_dcache_page(p);
 307                 SetPageUptodate(p);
 308                 unlock_page(p);
 309                 page_cache_release(p);
 310         }
 311
 312 error:
 313         kfree(pl);
 314         return error;
 315 }
 316
 317 /*
 318  * Populate a page with data for the Linux page cache.  This function is
 319  * only used to support mmap(2).  There will be an identical copy of the
 320  * data in the ARC which is kept up to date via .write() and .writepage().
 321  *
 322  * Current this function relies on zpl_read_common() and the O_DIRECT
 323  * flag to read in a page.  This works but the more correct way is to
 324  * update zfs_fillpage() to be Linux friendly and use that interface.
 325  */
 326 static int
 327 zpl_readpage(struct file *filp, struct page *pp)
 328 {
 329         struct inode *ip;
 330         struct page *pl[1];
 331         int error = 0;
 332
 333         ASSERT(PageLocked(pp));
 334         ip = pp->mapping->host;
 335         pl[0] = pp;
 336
 337         error = -zfs_getpage(ip, pl, 1);
 338
 339         if (error) {
 340                 SetPageError(pp);
 341                 ClearPageUptodate(pp);
 342         } else {
 343                 ClearPageError(pp);
 344                 SetPageUptodate(pp);
 345                 flush_dcache_page(pp);
 346         }
 347
 348         unlock_page(pp);
 349         return error;
 350 }
 351
 352 int
 353 zpl_putpage(struct page *pp, struct writeback_control *wbc, void *data)
 354 {
 355         int error;
 356
 357         error = -zfs_putpage(pp, wbc, data);
 358
 359         if (error) {
 360                 SetPageError(pp);
 361                 ClearPageUptodate(pp);
 362         } else {
 363                 ClearPageError(pp);
 364                 SetPageUptodate(pp);
 365                 flush_dcache_page(pp);
 366         }
 367
 368         unlock_page(pp);
 369         return error;
 370 }
 371
 372 static int
 373 zpl_writepages(struct address_space *mapping, struct writeback_control *wbc)
 374 {
 375         return write_cache_pages(mapping, wbc, zpl_putpage, mapping);
 376 }
 377
 378 /*
 379  * Write out dirty pages to the ARC, this function is only required to
 380  * support mmap(2).  Mapped pages may be dirtied by memory operations
 381  * which never call .write().  These dirty pages are kept in sync with
 382  * the ARC buffers via this hook.
 383  */
 384 static int
 385 zpl_writepage(struct page *pp, struct writeback_control *wbc)
 386 {
 387         return zpl_putpage(pp, wbc, pp->mapping);
 388 }
 389
 390 const struct address_space_operations zpl_address_space_operations = {
 391         .readpages      = zpl_readpages,
 392         .readpage       = zpl_readpage,
 393         .writepage      = zpl_writepage,
 394         .writepages     = zpl_writepages,
 395 };
 396
 397 const struct file_operations zpl_file_operations = {
 398         .open           = zpl_open,
 399         .release        = zpl_release,
 400         .llseek         = generic_file_llseek,
 401         .read           = zpl_read,
 402         .write          = zpl_write,
 403         .readdir        = zpl_readdir,
 404         .mmap           = zpl_mmap,
 405         .fsync          = zpl_fsync,
 406 };
 407
 408 const struct file_operations zpl_dir_file_operations = {
 409         .llseek         = generic_file_llseek,
 410         .read           = generic_read_dir,
 411         .readdir        = zpl_readdir,
 412         .fsync          = zpl_fsync,
 413 };