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_iterate(struct file *filp, struct dir_context *ctx)
  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, ctx, cr);
  75         crfree(cr);
  76         ASSERT3S(error, <=, 0);
  77
  78         return (error);
  79 }
  80
  81 #if !defined(HAVE_VFS_ITERATE)
  82 static int
  83 zpl_readdir(struct file *filp, void *dirent, filldir_t filldir)
  84 {
  85         struct dir_context ctx = DIR_CONTEXT_INIT(dirent, filldir, filp->f_pos);
  86         int error;
  87
  88         error = zpl_iterate(filp, &ctx);
  89         filp->f_pos = ctx.pos;
  90
  91         return (error);
  92 }
  93 #endif /* HAVE_VFS_ITERATE */
  94
  95 #if defined(HAVE_FSYNC_WITH_DENTRY)
  96 /*
  97  * Linux 2.6.x - 2.6.34 API,
  98  * Through 2.6.34 the nfsd kernel server would pass a NULL 'file struct *'
  99  * to the fops->fsync() hook.  For this reason, we must be careful not to
 100  * use filp unconditionally.
 101  */
 102 static int
 103 zpl_fsync(struct file *filp, struct dentry *dentry, int datasync)
 104 {
 105         cred_t *cr = CRED();
 106         int error;
 107
 108         crhold(cr);
 109         error = -zfs_fsync(dentry->d_inode, datasync, cr);
 110         crfree(cr);
 111         ASSERT3S(error, <=, 0);
 112
 113         return (error);
 114 }
 115
 116 #elif defined(HAVE_FSYNC_WITHOUT_DENTRY)
 117 /*
 118  * Linux 2.6.35 - 3.0 API,
 119  * As of 2.6.35 the dentry argument to the fops->fsync() hook was deemed
 120  * redundant.  The dentry is still accessible via filp->f_path.dentry,
 121  * and we are guaranteed that filp will never be NULL.
 122  */
 123 static int
 124 zpl_fsync(struct file *filp, int datasync)
 125 {
 126         struct inode *inode = filp->f_mapping->host;
 127         cred_t *cr = CRED();
 128         int error;
 129
 130         crhold(cr);
 131         error = -zfs_fsync(inode, datasync, cr);
 132         crfree(cr);
 133         ASSERT3S(error, <=, 0);
 134
 135         return (error);
 136 }
 137
 138 #elif defined(HAVE_FSYNC_RANGE)
 139 /*
 140  * Linux 3.1 - 3.x API,
 141  * As of 3.1 the responsibility to call filemap_write_and_wait_range() has
 142  * been pushed down in to the .fsync() vfs hook.  Additionally, the i_mutex
 143  * lock is no longer held by the caller, for zfs we don't require the lock
 144  * to be held so we don't acquire it.
 145  */
 146 static int
 147 zpl_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
 148 {
 149         struct inode *inode = filp->f_mapping->host;
 150         cred_t *cr = CRED();
 151         int error;
 152
 153         error = filemap_write_and_wait_range(inode->i_mapping, start, end);
 154         if (error)
 155                 return (error);
 156
 157         crhold(cr);
 158         error = -zfs_fsync(inode, datasync, cr);
 159         crfree(cr);
 160         ASSERT3S(error, <=, 0);
 161
 162         return (error);
 163 }
 164 #else
 165 #error "Unsupported fops->fsync() implementation"
 166 #endif
 167
 168 ssize_t
 169 zpl_read_common(struct inode *ip, const char *buf, size_t len, loff_t pos,
 170      uio_seg_t segment, int flags, cred_t *cr)
 171 {
 172         int error;
 173         ssize_t read;
 174         struct iovec iov;
 175         uio_t uio;
 176
 177         iov.iov_base = (void *)buf;
 178         iov.iov_len = len;
 179
 180         uio.uio_iov = &iov;
 181         uio.uio_resid = len;
 182         uio.uio_iovcnt = 1;
 183         uio.uio_loffset = pos;
 184         uio.uio_limit = MAXOFFSET_T;
 185         uio.uio_segflg = segment;
 186
 187         error = -zfs_read(ip, &uio, flags, cr);
 188         if (error < 0)
 189                 return (error);
 190
 191         read = len - uio.uio_resid;
 192         task_io_account_read(read);
 193
 194         return (read);
 195 }
 196
 197 static ssize_t
 198 zpl_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 199 {
 200         cred_t *cr = CRED();
 201         ssize_t read;
 202
 203         crhold(cr);
 204         read = zpl_read_common(filp->f_mapping->host, buf, len, *ppos,
 205             UIO_USERSPACE, filp->f_flags, cr);
 206         crfree(cr);
 207
 208         if (read < 0)
 209                 return (read);
 210
 211         *ppos += read;
 212         return (read);
 213 }
 214
 215 ssize_t
 216 zpl_write_common(struct inode *ip, const char *buf, size_t len, loff_t pos,
 217     uio_seg_t segment, int flags, cred_t *cr)
 218 {
 219         int error;
 220         ssize_t wrote;
 221         struct iovec iov;
 222         uio_t uio;
 223
 224         iov.iov_base = (void *)buf;
 225         iov.iov_len = len;
 226
 227         uio.uio_iov = &iov;
 228         uio.uio_resid = len,
 229         uio.uio_iovcnt = 1;
 230         uio.uio_loffset = pos;
 231         uio.uio_limit = MAXOFFSET_T;
 232         uio.uio_segflg = segment;
 233
 234         error = -zfs_write(ip, &uio, flags, cr);
 235         if (error < 0)
 236                 return (error);
 237
 238         wrote = len - uio.uio_resid;
 239         task_io_account_write(wrote);
 240
 241         return (wrote);
 242 }
 243
 244 static ssize_t
 245 zpl_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
 246 {
 247         cred_t *cr = CRED();
 248         ssize_t wrote;
 249
 250         crhold(cr);
 251         wrote = zpl_write_common(filp->f_mapping->host, buf, len, *ppos,
 252             UIO_USERSPACE, filp->f_flags, cr);
 253         crfree(cr);
 254
 255         if (wrote < 0)
 256                 return (wrote);
 257
 258         *ppos += wrote;
 259         return (wrote);
 260 }
 261
 262 static loff_t
 263 zpl_llseek(struct file *filp, loff_t offset, int whence)
 264 {
 265 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
 266         if (whence == SEEK_DATA || whence == SEEK_HOLE) {
 267                 struct inode *ip = filp->f_mapping->host;
 268                 loff_t maxbytes = ip->i_sb->s_maxbytes;
 269                 loff_t error;
 270
 271                 spl_inode_lock(ip);
 272                 error = -zfs_holey(ip, whence, &offset);
 273                 if (error == 0)
 274                         error = lseek_execute(filp, ip, offset, maxbytes);
 275                 spl_inode_unlock(ip);
 276
 277                 return (error);
 278         }
 279 #endif /* SEEK_HOLE && SEEK_DATA */
 280
 281         return generic_file_llseek(filp, offset, whence);
 282 }
 283
 284 /*
 285  * It's worth taking a moment to describe how mmap is implemented
 286  * for zfs because it differs considerably from other Linux filesystems.
 287  * However, this issue is handled the same way under OpenSolaris.
 288  *
 289  * The issue is that by design zfs bypasses the Linux page cache and
 290  * leaves all caching up to the ARC.  This has been shown to work
 291  * well for the common read(2)/write(2) case.  However, mmap(2)
 292  * is problem because it relies on being tightly integrated with the
 293  * page cache.  To handle this we cache mmap'ed files twice, once in
 294  * the ARC and a second time in the page cache.  The code is careful
 295  * to keep both copies synchronized.
 296  *
 297  * When a file with an mmap'ed region is written to using write(2)
 298  * both the data in the ARC and existing pages in the page cache
 299  * are updated.  For a read(2) data will be read first from the page
 300  * cache then the ARC if needed.  Neither a write(2) or read(2) will
 301  * will ever result in new pages being added to the page cache.
 302  *
 303  * New pages are added to the page cache only via .readpage() which
 304  * is called when the vfs needs to read a page off disk to back the
 305  * virtual memory region.  These pages may be modified without
 306  * notifying the ARC and will be written out periodically via
 307  * .writepage().  This will occur due to either a sync or the usual
 308  * page aging behavior.  Note because a read(2) of a mmap'ed file
 309  * will always check the page cache first even when the ARC is out
 310  * of date correct data will still be returned.
 311  *
 312  * While this implementation ensures correct behavior it does have
 313  * have some drawbacks.  The most obvious of which is that it
 314  * increases the required memory footprint when access mmap'ed
 315  * files.  It also adds additional complexity to the code keeping
 316  * both caches synchronized.
 317  *
 318  * Longer term it may be possible to cleanly resolve this wart by
 319  * mapping page cache pages directly on to the ARC buffers.  The
 320  * Linux address space operations are flexible enough to allow
 321  * selection of which pages back a particular index.  The trick
 322  * would be working out the details of which subsystem is in
 323  * charge, the ARC, the page cache, or both.  It may also prove
 324  * helpful to move the ARC buffers to a scatter-gather lists
 325  * rather than a vmalloc'ed region.
 326  */
 327 static int
 328 zpl_mmap(struct file *filp, struct vm_area_struct *vma)
 329 {
 330         struct inode *ip = filp->f_mapping->host;
 331         znode_t *zp = ITOZ(ip);
 332         int error;
 333
 334         error = -zfs_map(ip, vma->vm_pgoff, (caddr_t *)vma->vm_start,
 335             (size_t)(vma->vm_end - vma->vm_start), vma->vm_flags);
 336         if (error)
 337                 return (error);
 338
 339         error = generic_file_mmap(filp, vma);
 340         if (error)
 341                 return (error);
 342
 343         mutex_enter(&zp->z_lock);
 344         zp->z_is_mapped = 1;
 345         mutex_exit(&zp->z_lock);
 346
 347         return (error);
 348 }
 349
 350 /*
 351  * Populate a page with data for the Linux page cache.  This function is
 352  * only used to support mmap(2).  There will be an identical copy of the
 353  * data in the ARC which is kept up to date via .write() and .writepage().
 354  *
 355  * Current this function relies on zpl_read_common() and the O_DIRECT
 356  * flag to read in a page.  This works but the more correct way is to
 357  * update zfs_fillpage() to be Linux friendly and use that interface.
 358  */
 359 static int
 360 zpl_readpage(struct file *filp, struct page *pp)
 361 {
 362         struct inode *ip;
 363         struct page *pl[1];
 364         int error = 0;
 365
 366         ASSERT(PageLocked(pp));
 367         ip = pp->mapping->host;
 368         pl[0] = pp;
 369
 370         error = -zfs_getpage(ip, pl, 1);
 371
 372         if (error) {
 373                 SetPageError(pp);
 374                 ClearPageUptodate(pp);
 375         } else {
 376                 ClearPageError(pp);
 377                 SetPageUptodate(pp);
 378                 flush_dcache_page(pp);
 379         }
 380
 381         unlock_page(pp);
 382         return error;
 383 }
 384
 385 /*
 386  * Populate a set of pages with data for the Linux page cache.  This
 387  * function will only be called for read ahead and never for demand
 388  * paging.  For simplicity, the code relies on read_cache_pages() to
 389  * correctly lock each page for IO and call zpl_readpage().
 390  */
 391 static int
 392 zpl_readpages(struct file *filp, struct address_space *mapping,
 393         struct list_head *pages, unsigned nr_pages)
 394 {
 395         return (read_cache_pages(mapping, pages,
 396             (filler_t *)zpl_readpage, filp));
 397 }
 398
 399 int
 400 zpl_putpage(struct page *pp, struct writeback_control *wbc, void *data)
 401 {
 402         struct address_space *mapping = data;
 403
 404         ASSERT(PageLocked(pp));
 405         ASSERT(!PageWriteback(pp));
 406         ASSERT(!(current->flags & PF_NOFS));
 407
 408         /*
 409          * Annotate this call path with a flag that indicates that it is
 410          * unsafe to use KM_SLEEP during memory allocations due to the
 411          * potential for a deadlock.  KM_PUSHPAGE should be used instead.
 412          */
 413         current->flags |= PF_NOFS;
 414         (void) zfs_putpage(mapping->host, pp, wbc);
 415         current->flags &= ~PF_NOFS;
 416
 417         return (0);
 418 }
 419
 420 static int
 421 zpl_writepages(struct address_space *mapping, struct writeback_control *wbc)
 422 {
 423         return write_cache_pages(mapping, wbc, zpl_putpage, mapping);
 424 }
 425
 426 /*
 427  * Write out dirty pages to the ARC, this function is only required to
 428  * support mmap(2).  Mapped pages may be dirtied by memory operations
 429  * which never call .write().  These dirty pages are kept in sync with
 430  * the ARC buffers via this hook.
 431  */
 432 static int
 433 zpl_writepage(struct page *pp, struct writeback_control *wbc)
 434 {
 435         return zpl_putpage(pp, wbc, pp->mapping);
 436 }
 437
 438 /*
 439  * The only flag combination which matches the behavior of zfs_space()
 440  * is FALLOC_FL_PUNCH_HOLE.  This flag was introduced in the 2.6.38 kernel.
 441  */
 442 long
 443 zpl_fallocate_common(struct inode *ip, int mode, loff_t offset, loff_t len)
 444 {
 445         cred_t *cr = CRED();
 446         int error = -EOPNOTSUPP;
 447
 448         if (mode & FALLOC_FL_KEEP_SIZE)
 449                 return (-EOPNOTSUPP);
 450
 451         crhold(cr);
 452
 453 #ifdef FALLOC_FL_PUNCH_HOLE
 454         if (mode & FALLOC_FL_PUNCH_HOLE) {
 455                 flock64_t bf;
 456
 457                 bf.l_type = F_WRLCK;
 458                 bf.l_whence = 0;
 459                 bf.l_start = offset;
 460                 bf.l_len = len;
 461                 bf.l_pid = 0;
 462
 463                 error = -zfs_space(ip, F_FREESP, &bf, FWRITE, offset, cr);
 464         }
 465 #endif /* FALLOC_FL_PUNCH_HOLE */
 466
 467         crfree(cr);
 468
 469         ASSERT3S(error, <=, 0);
 470         return (error);
 471 }
 472
 473 #ifdef HAVE_FILE_FALLOCATE
 474 static long
 475 zpl_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
 476 {
 477         return zpl_fallocate_common(filp->f_path.dentry->d_inode,
 478             mode, offset, len);
 479 }
 480 #endif /* HAVE_FILE_FALLOCATE */
 481
 482 static long
 483 zpl_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 484 {
 485         switch (cmd) {
 486         case ZFS_IOC_GETFLAGS:
 487         case ZFS_IOC_SETFLAGS:
 488                 return (-EOPNOTSUPP);
 489         default:
 490                 return (-ENOTTY);
 491         }
 492 }
 493
 494 #ifdef CONFIG_COMPAT
 495 static long
 496 zpl_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 497 {
 498         return zpl_ioctl(filp, cmd, arg);
 499 }
 500 #endif /* CONFIG_COMPAT */
 501
 502
 503 const struct address_space_operations zpl_address_space_operations = {
 504         .readpages      = zpl_readpages,
 505         .readpage       = zpl_readpage,
 506         .writepage      = zpl_writepage,
 507         .writepages     = zpl_writepages,
 508 };
 509
 510 const struct file_operations zpl_file_operations = {
 511         .open           = zpl_open,
 512         .release        = zpl_release,
 513         .llseek         = zpl_llseek,
 514         .read           = zpl_read,
 515         .write          = zpl_write,
 516         .mmap           = zpl_mmap,
 517         .fsync          = zpl_fsync,
 518 #ifdef HAVE_FILE_FALLOCATE
 519         .fallocate      = zpl_fallocate,
 520 #endif /* HAVE_FILE_FALLOCATE */
 521         .unlocked_ioctl = zpl_ioctl,
 522 #ifdef CONFIG_COMPAT
 523         .compat_ioctl   = zpl_compat_ioctl,
 524 #endif
 525 };
 526
 527 const struct file_operations zpl_dir_file_operations = {
 528         .llseek         = generic_file_llseek,
 529         .read           = generic_read_dir,
 530 #ifdef HAVE_VFS_ITERATE
 531         .iterate        = zpl_iterate,
 532 #else
 533         .readdir        = zpl_readdir,
 534 #endif
 535         .fsync          = zpl_fsync,
 536         .unlocked_ioctl = zpl_ioctl,
 537 #ifdef CONFIG_COMPAT
 538         .compat_ioctl   = zpl_compat_ioctl,
 539 #endif
 540 };