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         if (ITOZ(ip)->z_atime_dirty)
  56                 mark_inode_dirty(ip);
  57
  58         crhold(cr);
  59         error = -zfs_close(ip, filp->f_flags, cr);
  60         crfree(cr);
  61         ASSERT3S(error, <=, 0);
  62
  63         return (error);
  64 }
  65
  66 static int
  67 zpl_readdir(struct file *filp, void *dirent, filldir_t filldir)
  68 {
  69         struct dentry *dentry = filp->f_path.dentry;
  70         cred_t *cr = CRED();
  71         int error;
  72
  73         crhold(cr);
  74         error = -zfs_readdir(dentry->d_inode, dirent, filldir,
  75             &filp->f_pos, cr);
  76         crfree(cr);
  77         ASSERT3S(error, <=, 0);
  78
  79         return (error);
  80 }
  81
  82 #if defined(HAVE_FSYNC_WITH_DENTRY)
  83 /*
  84  * Linux 2.6.x - 2.6.34 API,
  85  * Through 2.6.34 the nfsd kernel server would pass a NULL 'file struct *'
  86  * to the fops->fsync() hook.  For this reason, we must be careful not to
  87  * use filp unconditionally.
  88  */
  89 static int
  90 zpl_fsync(struct file *filp, struct dentry *dentry, int datasync)
  91 {
  92         cred_t *cr = CRED();
  93         int error;
  94
  95         crhold(cr);
  96         error = -zfs_fsync(dentry->d_inode, datasync, cr);
  97         crfree(cr);
  98         ASSERT3S(error, <=, 0);
  99
 100         return (error);
 101 }
 102
 103 #elif defined(HAVE_FSYNC_WITHOUT_DENTRY)
 104 /*
 105  * Linux 2.6.35 - 3.0 API,
 106  * As of 2.6.35 the dentry argument to the fops->fsync() hook was deemed
 107  * redundant.  The dentry is still accessible via filp->f_path.dentry,
 108  * and we are guaranteed that filp will never be NULL.
 109  */
 110 static int
 111 zpl_fsync(struct file *filp, int datasync)
 112 {
 113         struct inode *inode = filp->f_mapping->host;
 114         cred_t *cr = CRED();
 115         int error;
 116
 117         crhold(cr);
 118         error = -zfs_fsync(inode, datasync, cr);
 119         crfree(cr);
 120         ASSERT3S(error, <=, 0);
 121
 122         return (error);
 123 }
 124
 125 #elif defined(HAVE_FSYNC_RANGE)
 126 /*
 127  * Linux 3.1 - 3.x API,
 128  * As of 3.1 the responsibility to call filemap_write_and_wait_range() has
 129  * been pushed down in to the .fsync() vfs hook.  Additionally, the i_mutex
 130  * lock is no longer held by the caller, for zfs we don't require the lock
 131  * to be held so we don't acquire it.
 132  */
 133 static int
 134 zpl_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
 135 {
 136         struct inode *inode = filp->f_mapping->host;
 137         cred_t *cr = CRED();
 138         int error;
 139
 140         error = filemap_write_and_wait_range(inode->i_mapping, start, end);
 141         if (error)
 142                 return (error);
 143
 144         crhold(cr);
 145         error = -zfs_fsync(inode, datasync, cr);
 146         crfree(cr);
 147         ASSERT3S(error, <=, 0);
 148
 149         return (error);
 150 }
 151 #else
 152 #error "Unsupported fops->fsync() implementation"
 153 #endif
 154
 155 ssize_t
 156 zpl_read_common(struct inode *ip, const char *buf, size_t len, loff_t pos,
 157      uio_seg_t segment, int flags, cred_t *cr)
 158 {
 159         int error;
 160         struct iovec iov;
 161         uio_t uio;
 162
 163         iov.iov_base = (void *)buf;
 164         iov.iov_len = len;
 165
 166         uio.uio_iov = &iov;
 167         uio.uio_resid = len;
 168         uio.uio_iovcnt = 1;
 169         uio.uio_loffset = pos;
 170         uio.uio_limit = MAXOFFSET_T;
 171         uio.uio_segflg = segment;
 172
 173         error = -zfs_read(ip, &uio, flags, cr);
 174         if (error < 0)
 175                 return (error);
 176
 177         return (len - uio.uio_resid);
 178 }
 179
 180 static ssize_t
 181 zpl_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 182 {
 183         cred_t *cr = CRED();
 184         ssize_t read;
 185
 186         crhold(cr);
 187         read = zpl_read_common(filp->f_mapping->host, buf, len, *ppos,
 188             UIO_USERSPACE, filp->f_flags, cr);
 189         crfree(cr);
 190
 191         if (read < 0)
 192                 return (read);
 193
 194         *ppos += read;
 195         return (read);
 196 }
 197
 198 ssize_t
 199 zpl_write_common(struct inode *ip, const char *buf, size_t len, loff_t pos,
 200     uio_seg_t segment, int flags, cred_t *cr)
 201 {
 202         int error;
 203         struct iovec iov;
 204         uio_t uio;
 205
 206         iov.iov_base = (void *)buf;
 207         iov.iov_len = len;
 208
 209         uio.uio_iov = &iov;
 210         uio.uio_resid = len,
 211         uio.uio_iovcnt = 1;
 212         uio.uio_loffset = pos;
 213         uio.uio_limit = MAXOFFSET_T;
 214         uio.uio_segflg = segment;
 215
 216         error = -zfs_write(ip, &uio, flags, cr);
 217         if (error < 0)
 218                 return (error);
 219
 220         return (len - uio.uio_resid);
 221 }
 222
 223 static ssize_t
 224 zpl_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
 225 {
 226         cred_t *cr = CRED();
 227         ssize_t wrote;
 228
 229         crhold(cr);
 230         wrote = zpl_write_common(filp->f_mapping->host, buf, len, *ppos,
 231             UIO_USERSPACE, filp->f_flags, cr);
 232         crfree(cr);
 233
 234         if (wrote < 0)
 235                 return (wrote);
 236
 237         *ppos += wrote;
 238         return (wrote);
 239 }
 240
 241 static loff_t
 242 zpl_llseek(struct file *filp, loff_t offset, int whence)
 243 {
 244 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
 245         if (whence == SEEK_DATA || whence == SEEK_HOLE) {
 246                 struct inode *ip = filp->f_mapping->host;
 247                 loff_t maxbytes = ip->i_sb->s_maxbytes;
 248                 loff_t error;
 249
 250                 spl_inode_lock(ip);
 251                 error = -zfs_holey(ip, whence, &offset);
 252                 if (error == 0)
 253                         error = lseek_execute(filp, ip, offset, maxbytes);
 254                 spl_inode_unlock(ip);
 255
 256                 return (error);
 257         }
 258 #endif /* SEEK_HOLE && SEEK_DATA */
 259
 260         return generic_file_llseek(filp, offset, whence);
 261 }
 262
 263 /*
 264  * It's worth taking a moment to describe how mmap is implemented
 265  * for zfs because it differs considerably from other Linux filesystems.
 266  * However, this issue is handled the same way under OpenSolaris.
 267  *
 268  * The issue is that by design zfs bypasses the Linux page cache and
 269  * leaves all caching up to the ARC.  This has been shown to work
 270  * well for the common read(2)/write(2) case.  However, mmap(2)
 271  * is problem because it relies on being tightly integrated with the
 272  * page cache.  To handle this we cache mmap'ed files twice, once in
 273  * the ARC and a second time in the page cache.  The code is careful
 274  * to keep both copies synchronized.
 275  *
 276  * When a file with an mmap'ed region is written to using write(2)
 277  * both the data in the ARC and existing pages in the page cache
 278  * are updated.  For a read(2) data will be read first from the page
 279  * cache then the ARC if needed.  Neither a write(2) or read(2) will
 280  * will ever result in new pages being added to the page cache.
 281  *
 282  * New pages are added to the page cache only via .readpage() which
 283  * is called when the vfs needs to read a page off disk to back the
 284  * virtual memory region.  These pages may be modified without
 285  * notifying the ARC and will be written out periodically via
 286  * .writepage().  This will occur due to either a sync or the usual
 287  * page aging behavior.  Note because a read(2) of a mmap'ed file
 288  * will always check the page cache first even when the ARC is out
 289  * of date correct data will still be returned.
 290  *
 291  * While this implementation ensures correct behavior it does have
 292  * have some drawbacks.  The most obvious of which is that it
 293  * increases the required memory footprint when access mmap'ed
 294  * files.  It also adds additional complexity to the code keeping
 295  * both caches synchronized.
 296  *
 297  * Longer term it may be possible to cleanly resolve this wart by
 298  * mapping page cache pages directly on to the ARC buffers.  The
 299  * Linux address space operations are flexible enough to allow
 300  * selection of which pages back a particular index.  The trick
 301  * would be working out the details of which subsystem is in
 302  * charge, the ARC, the page cache, or both.  It may also prove
 303  * helpful to move the ARC buffers to a scatter-gather lists
 304  * rather than a vmalloc'ed region.
 305  */
 306 static int
 307 zpl_mmap(struct file *filp, struct vm_area_struct *vma)
 308 {
 309         struct inode *ip = filp->f_mapping->host;
 310         znode_t *zp = ITOZ(ip);
 311         int error;
 312
 313         error = -zfs_map(ip, vma->vm_pgoff, (caddr_t *)vma->vm_start,
 314             (size_t)(vma->vm_end - vma->vm_start), vma->vm_flags);
 315         if (error)
 316                 return (error);
 317
 318         error = generic_file_mmap(filp, vma);
 319         if (error)
 320                 return (error);
 321
 322         mutex_enter(&zp->z_lock);
 323         zp->z_is_mapped = 1;
 324         mutex_exit(&zp->z_lock);
 325
 326         return (error);
 327 }
 328
 329 /*
 330  * Populate a page with data for the Linux page cache.  This function is
 331  * only used to support mmap(2).  There will be an identical copy of the
 332  * data in the ARC which is kept up to date via .write() and .writepage().
 333  *
 334  * Current this function relies on zpl_read_common() and the O_DIRECT
 335  * flag to read in a page.  This works but the more correct way is to
 336  * update zfs_fillpage() to be Linux friendly and use that interface.
 337  */
 338 static int
 339 zpl_readpage(struct file *filp, struct page *pp)
 340 {
 341         struct inode *ip;
 342         struct page *pl[1];
 343         int error = 0;
 344
 345         ASSERT(PageLocked(pp));
 346         ip = pp->mapping->host;
 347         pl[0] = pp;
 348
 349         error = -zfs_getpage(ip, pl, 1);
 350
 351         if (error) {
 352                 SetPageError(pp);
 353                 ClearPageUptodate(pp);
 354         } else {
 355                 ClearPageError(pp);
 356                 SetPageUptodate(pp);
 357                 flush_dcache_page(pp);
 358         }
 359
 360         unlock_page(pp);
 361         return error;
 362 }
 363
 364 /*
 365  * Populate a set of pages with data for the Linux page cache.  This
 366  * function will only be called for read ahead and never for demand
 367  * paging.  For simplicity, the code relies on read_cache_pages() to
 368  * correctly lock each page for IO and call zpl_readpage().
 369  */
 370 static int
 371 zpl_readpages(struct file *filp, struct address_space *mapping,
 372         struct list_head *pages, unsigned nr_pages)
 373 {
 374         return (read_cache_pages(mapping, pages,
 375             (filler_t *)zpl_readpage, filp));
 376 }
 377
 378 int
 379 zpl_putpage(struct page *pp, struct writeback_control *wbc, void *data)
 380 {
 381         struct address_space *mapping = data;
 382
 383         ASSERT(PageLocked(pp));
 384         ASSERT(!PageWriteback(pp));
 385         ASSERT(!(current->flags & PF_NOFS));
 386
 387         /*
 388          * Annotate this call path with a flag that indicates that it is
 389          * unsafe to use KM_SLEEP during memory allocations due to the
 390          * potential for a deadlock.  KM_PUSHPAGE should be used instead.
 391          */
 392         current->flags |= PF_NOFS;
 393         (void) zfs_putpage(mapping->host, pp, wbc);
 394         current->flags &= ~PF_NOFS;
 395
 396         return (0);
 397 }
 398
 399 static int
 400 zpl_writepages(struct address_space *mapping, struct writeback_control *wbc)
 401 {
 402         return write_cache_pages(mapping, wbc, zpl_putpage, mapping);
 403 }
 404
 405 /*
 406  * Write out dirty pages to the ARC, this function is only required to
 407  * support mmap(2).  Mapped pages may be dirtied by memory operations
 408  * which never call .write().  These dirty pages are kept in sync with
 409  * the ARC buffers via this hook.
 410  */
 411 static int
 412 zpl_writepage(struct page *pp, struct writeback_control *wbc)
 413 {
 414         return zpl_putpage(pp, wbc, pp->mapping);
 415 }
 416
 417 /*
 418  * The only flag combination which matches the behavior of zfs_space()
 419  * is FALLOC_FL_PUNCH_HOLE.  This flag was introduced in the 2.6.38 kernel.
 420  */
 421 long
 422 zpl_fallocate_common(struct inode *ip, int mode, loff_t offset, loff_t len)
 423 {
 424         cred_t *cr = CRED();
 425         int error = -EOPNOTSUPP;
 426
 427         if (mode & FALLOC_FL_KEEP_SIZE)
 428                 return (-EOPNOTSUPP);
 429
 430         crhold(cr);
 431
 432 #ifdef FALLOC_FL_PUNCH_HOLE
 433         if (mode & FALLOC_FL_PUNCH_HOLE) {
 434                 flock64_t bf;
 435
 436                 bf.l_type = F_WRLCK;
 437                 bf.l_whence = 0;
 438                 bf.l_start = offset;
 439                 bf.l_len = len;
 440                 bf.l_pid = 0;
 441
 442                 error = -zfs_space(ip, F_FREESP, &bf, FWRITE, offset, cr);
 443         }
 444 #endif /* FALLOC_FL_PUNCH_HOLE */
 445
 446         crfree(cr);
 447
 448         ASSERT3S(error, <=, 0);
 449         return (error);
 450 }
 451
 452 #ifdef HAVE_FILE_FALLOCATE
 453 static long
 454 zpl_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
 455 {
 456         return zpl_fallocate_common(filp->f_path.dentry->d_inode,
 457             mode, offset, len);
 458 }
 459 #endif /* HAVE_FILE_FALLOCATE */
 460
 461 static long
 462 zpl_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 463 {
 464         switch (cmd) {
 465         case ZFS_IOC_GETFLAGS:
 466         case ZFS_IOC_SETFLAGS:
 467                 return (-EOPNOTSUPP);
 468         default:
 469                 return (-ENOTTY);
 470         }
 471 }
 472
 473 #ifdef CONFIG_COMPAT
 474 static long
 475 zpl_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 476 {
 477         return zpl_ioctl(filp, cmd, arg);
 478 }
 479 #endif /* CONFIG_COMPAT */
 480
 481
 482 const struct address_space_operations zpl_address_space_operations = {
 483         .readpages      = zpl_readpages,
 484         .readpage       = zpl_readpage,
 485         .writepage      = zpl_writepage,
 486         .writepages     = zpl_writepages,
 487 };
 488
 489 const struct file_operations zpl_file_operations = {
 490         .open           = zpl_open,
 491         .release        = zpl_release,
 492         .llseek         = zpl_llseek,
 493         .read           = zpl_read,
 494         .write          = zpl_write,
 495         .mmap           = zpl_mmap,
 496         .fsync          = zpl_fsync,
 497 #ifdef HAVE_FILE_FALLOCATE
 498         .fallocate      = zpl_fallocate,
 499 #endif /* HAVE_FILE_FALLOCATE */
 500         .unlocked_ioctl = zpl_ioctl,
 501 #ifdef CONFIG_COMPAT
 502         .compat_ioctl   = zpl_compat_ioctl,
 503 #endif
 504 };
 505
 506 const struct file_operations zpl_dir_file_operations = {
 507         .llseek         = generic_file_llseek,
 508         .read           = generic_read_dir,
 509         .readdir        = zpl_readdir,
 510         .fsync          = zpl_fsync,
 511         .unlocked_ioctl = zpl_ioctl,
 512 #ifdef CONFIG_COMPAT
 513         .compat_ioctl   = zpl_compat_ioctl,
 514 #endif
 515 };