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  * Copyright (c) 2015 by Chunwei Chen. All rights reserved.
  24  */
  25
  26
  27 #ifdef CONFIG_COMPAT
  28 #include <linux/compat.h>
  29 #endif
  30 #include <sys/dmu_objset.h>
  31 #include <sys/zfs_vfsops.h>
  32 #include <sys/zfs_vnops.h>
  33 #include <sys/zfs_znode.h>
  34 #include <sys/zpl.h>
  35
  36
  37 static int
  38 zpl_open(struct inode *ip, struct file *filp)
  39 {
  40         cred_t *cr = CRED();
  41         int error;
  42         fstrans_cookie_t cookie;
  43
  44         error = generic_file_open(ip, filp);
  45         if (error)
  46                 return (error);
  47
  48         crhold(cr);
  49         cookie = spl_fstrans_mark();
  50         error = -zfs_open(ip, filp->f_mode, filp->f_flags, cr);
  51         spl_fstrans_unmark(cookie);
  52         crfree(cr);
  53         ASSERT3S(error, <=, 0);
  54
  55         return (error);
  56 }
  57
  58 static int
  59 zpl_release(struct inode *ip, struct file *filp)
  60 {
  61         cred_t *cr = CRED();
  62         int error;
  63         fstrans_cookie_t cookie;
  64
  65         cookie = spl_fstrans_mark();
  66         if (ITOZ(ip)->z_atime_dirty)
  67                 zfs_mark_inode_dirty(ip);
  68
  69         crhold(cr);
  70         error = -zfs_close(ip, filp->f_flags, cr);
  71         spl_fstrans_unmark(cookie);
  72         crfree(cr);
  73         ASSERT3S(error, <=, 0);
  74
  75         return (error);
  76 }
  77
  78 static int
  79 zpl_iterate(struct file *filp, struct dir_context *ctx)
  80 {
  81         cred_t *cr = CRED();
  82         int error;
  83         fstrans_cookie_t cookie;
  84
  85         crhold(cr);
  86         cookie = spl_fstrans_mark();
  87         error = -zfs_readdir(file_inode(filp), ctx, cr);
  88         spl_fstrans_unmark(cookie);
  89         crfree(cr);
  90         ASSERT3S(error, <=, 0);
  91
  92         return (error);
  93 }
  94
  95 #if !defined(HAVE_VFS_ITERATE) && !defined(HAVE_VFS_ITERATE_SHARED)
  96 static int
  97 zpl_readdir(struct file *filp, void *dirent, filldir_t filldir)
  98 {
  99         struct dir_context ctx = DIR_CONTEXT_INIT(dirent, filldir, filp->f_pos);
 100         int error;
 101
 102         error = zpl_iterate(filp, &ctx);
 103         filp->f_pos = ctx.pos;
 104
 105         return (error);
 106 }
 107 #endif /* HAVE_VFS_ITERATE */
 108
 109 #if defined(HAVE_FSYNC_WITH_DENTRY)
 110 /*
 111  * Linux 2.6.x - 2.6.34 API,
 112  * Through 2.6.34 the nfsd kernel server would pass a NULL 'file struct *'
 113  * to the fops->fsync() hook.  For this reason, we must be careful not to
 114  * use filp unconditionally.
 115  */
 116 static int
 117 zpl_fsync(struct file *filp, struct dentry *dentry, int datasync)
 118 {
 119         cred_t *cr = CRED();
 120         int error;
 121         fstrans_cookie_t cookie;
 122
 123         crhold(cr);
 124         cookie = spl_fstrans_mark();
 125         error = -zfs_fsync(dentry->d_inode, datasync, cr);
 126         spl_fstrans_unmark(cookie);
 127         crfree(cr);
 128         ASSERT3S(error, <=, 0);
 129
 130         return (error);
 131 }
 132
 133 static int
 134 zpl_aio_fsync(struct kiocb *kiocb, int datasync)
 135 {
 136         struct file *filp = kiocb->ki_filp;
 137         return (zpl_fsync(filp, file_dentry(filp), datasync));
 138 }
 139 #elif defined(HAVE_FSYNC_WITHOUT_DENTRY)
 140 /*
 141  * Linux 2.6.35 - 3.0 API,
 142  * As of 2.6.35 the dentry argument to the fops->fsync() hook was deemed
 143  * redundant.  The dentry is still accessible via filp->f_path.dentry,
 144  * and we are guaranteed that filp will never be NULL.
 145  */
 146 static int
 147 zpl_fsync(struct file *filp, int datasync)
 148 {
 149         struct inode *inode = filp->f_mapping->host;
 150         cred_t *cr = CRED();
 151         int error;
 152         fstrans_cookie_t cookie;
 153
 154         crhold(cr);
 155         cookie = spl_fstrans_mark();
 156         error = -zfs_fsync(inode, datasync, cr);
 157         spl_fstrans_unmark(cookie);
 158         crfree(cr);
 159         ASSERT3S(error, <=, 0);
 160
 161         return (error);
 162 }
 163
 164 static int
 165 zpl_aio_fsync(struct kiocb *kiocb, int datasync)
 166 {
 167         return (zpl_fsync(kiocb->ki_filp, datasync));
 168 }
 169 #elif defined(HAVE_FSYNC_RANGE)
 170 /*
 171  * Linux 3.1 - 3.x API,
 172  * As of 3.1 the responsibility to call filemap_write_and_wait_range() has
 173  * been pushed down in to the .fsync() vfs hook.  Additionally, the i_mutex
 174  * lock is no longer held by the caller, for zfs we don't require the lock
 175  * to be held so we don't acquire it.
 176  */
 177 static int
 178 zpl_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
 179 {
 180         struct inode *inode = filp->f_mapping->host;
 181         cred_t *cr = CRED();
 182         int error;
 183         fstrans_cookie_t cookie;
 184
 185         error = filemap_write_and_wait_range(inode->i_mapping, start, end);
 186         if (error)
 187                 return (error);
 188
 189         crhold(cr);
 190         cookie = spl_fstrans_mark();
 191         error = -zfs_fsync(inode, datasync, cr);
 192         spl_fstrans_unmark(cookie);
 193         crfree(cr);
 194         ASSERT3S(error, <=, 0);
 195
 196         return (error);
 197 }
 198
 199 static int
 200 zpl_aio_fsync(struct kiocb *kiocb, int datasync)
 201 {
 202         return (zpl_fsync(kiocb->ki_filp, kiocb->ki_pos, -1, datasync));
 203 }
 204 #else
 205 #error "Unsupported fops->fsync() implementation"
 206 #endif
 207
 208 static ssize_t
 209 zpl_read_common_iovec(struct inode *ip, const struct iovec *iovp, size_t count,
 210     unsigned long nr_segs, loff_t *ppos, uio_seg_t segment, int flags,
 211     cred_t *cr, size_t skip)
 212 {
 213         ssize_t read;
 214         uio_t uio;
 215         int error;
 216         fstrans_cookie_t cookie;
 217
 218         uio.uio_iov = iovp;
 219         uio.uio_skip = skip;
 220         uio.uio_resid = count;
 221         uio.uio_iovcnt = nr_segs;
 222         uio.uio_loffset = *ppos;
 223         uio.uio_limit = MAXOFFSET_T;
 224         uio.uio_segflg = segment;
 225
 226         cookie = spl_fstrans_mark();
 227         error = -zfs_read(ip, &uio, flags, cr);
 228         spl_fstrans_unmark(cookie);
 229         if (error < 0)
 230                 return (error);
 231
 232         read = count - uio.uio_resid;
 233         *ppos += read;
 234         task_io_account_read(read);
 235
 236         return (read);
 237 }
 238
 239 inline ssize_t
 240 zpl_read_common(struct inode *ip, const char *buf, size_t len, loff_t *ppos,
 241     uio_seg_t segment, int flags, cred_t *cr)
 242 {
 243         struct iovec iov;
 244
 245         iov.iov_base = (void *)buf;
 246         iov.iov_len = len;
 247
 248         return (zpl_read_common_iovec(ip, &iov, len, 1, ppos, segment,
 249             flags, cr, 0));
 250 }
 251
 252 static ssize_t
 253 zpl_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 254 {
 255         cred_t *cr = CRED();
 256         ssize_t read;
 257
 258         crhold(cr);
 259         read = zpl_read_common(filp->f_mapping->host, buf, len, ppos,
 260             UIO_USERSPACE, filp->f_flags, cr);
 261         crfree(cr);
 262
 263         file_accessed(filp);
 264         return (read);
 265 }
 266
 267 static ssize_t
 268 zpl_iter_read_common(struct kiocb *kiocb, const struct iovec *iovp,
 269     unsigned long nr_segs, size_t count, uio_seg_t seg, size_t skip)
 270 {
 271         cred_t *cr = CRED();
 272         struct file *filp = kiocb->ki_filp;
 273         ssize_t read;
 274
 275         crhold(cr);
 276         read = zpl_read_common_iovec(filp->f_mapping->host, iovp, count,
 277             nr_segs, &kiocb->ki_pos, seg, filp->f_flags, cr, skip);
 278         crfree(cr);
 279
 280         file_accessed(filp);
 281         return (read);
 282 }
 283
 284 #if defined(HAVE_VFS_RW_ITERATE)
 285 static ssize_t
 286 zpl_iter_read(struct kiocb *kiocb, struct iov_iter *to)
 287 {
 288         ssize_t ret;
 289         uio_seg_t seg = UIO_USERSPACE;
 290         if (to->type & ITER_KVEC)
 291                 seg = UIO_SYSSPACE;
 292         if (to->type & ITER_BVEC)
 293                 seg = UIO_BVEC;
 294         ret = zpl_iter_read_common(kiocb, to->iov, to->nr_segs,
 295             iov_iter_count(to), seg, to->iov_offset);
 296         if (ret > 0)
 297                 iov_iter_advance(to, ret);
 298         return (ret);
 299 }
 300 #else
 301 static ssize_t
 302 zpl_aio_read(struct kiocb *kiocb, const struct iovec *iovp,
 303     unsigned long nr_segs, loff_t pos)
 304 {
 305         return (zpl_iter_read_common(kiocb, iovp, nr_segs, kiocb->ki_nbytes,
 306             UIO_USERSPACE, 0));
 307 }
 308 #endif /* HAVE_VFS_RW_ITERATE */
 309
 310 static ssize_t
 311 zpl_write_common_iovec(struct inode *ip, const struct iovec *iovp, size_t count,
 312     unsigned long nr_segs, loff_t *ppos, uio_seg_t segment, int flags,
 313     cred_t *cr, size_t skip)
 314 {
 315         ssize_t wrote;
 316         uio_t uio;
 317         int error;
 318         fstrans_cookie_t cookie;
 319
 320         if (flags & O_APPEND)
 321                 *ppos = i_size_read(ip);
 322
 323         uio.uio_iov = iovp;
 324         uio.uio_skip = skip;
 325         uio.uio_resid = count;
 326         uio.uio_iovcnt = nr_segs;
 327         uio.uio_loffset = *ppos;
 328         uio.uio_limit = MAXOFFSET_T;
 329         uio.uio_segflg = segment;
 330
 331         cookie = spl_fstrans_mark();
 332         error = -zfs_write(ip, &uio, flags, cr);
 333         spl_fstrans_unmark(cookie);
 334         if (error < 0)
 335                 return (error);
 336
 337         wrote = count - uio.uio_resid;
 338         *ppos += wrote;
 339         task_io_account_write(wrote);
 340
 341         return (wrote);
 342 }
 343 inline ssize_t
 344 zpl_write_common(struct inode *ip, const char *buf, size_t len, loff_t *ppos,
 345     uio_seg_t segment, int flags, cred_t *cr)
 346 {
 347         struct iovec iov;
 348
 349         iov.iov_base = (void *)buf;
 350         iov.iov_len = len;
 351
 352         return (zpl_write_common_iovec(ip, &iov, len, 1, ppos, segment,
 353             flags, cr, 0));
 354 }
 355
 356 static ssize_t
 357 zpl_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
 358 {
 359         cred_t *cr = CRED();
 360         ssize_t wrote;
 361
 362         crhold(cr);
 363         wrote = zpl_write_common(filp->f_mapping->host, buf, len, ppos,
 364             UIO_USERSPACE, filp->f_flags, cr);
 365         crfree(cr);
 366
 367         return (wrote);
 368 }
 369
 370 static ssize_t
 371 zpl_iter_write_common(struct kiocb *kiocb, const struct iovec *iovp,
 372     unsigned long nr_segs, size_t count, uio_seg_t seg, size_t skip)
 373 {
 374         cred_t *cr = CRED();
 375         struct file *filp = kiocb->ki_filp;
 376         ssize_t wrote;
 377
 378         crhold(cr);
 379         wrote = zpl_write_common_iovec(filp->f_mapping->host, iovp, count,
 380             nr_segs, &kiocb->ki_pos, seg, filp->f_flags, cr, skip);
 381         crfree(cr);
 382
 383         return (wrote);
 384 }
 385
 386 #if defined(HAVE_VFS_RW_ITERATE)
 387 static ssize_t
 388 zpl_iter_write(struct kiocb *kiocb, struct iov_iter *from)
 389 {
 390         ssize_t ret;
 391         uio_seg_t seg = UIO_USERSPACE;
 392         if (from->type & ITER_KVEC)
 393                 seg = UIO_SYSSPACE;
 394         if (from->type & ITER_BVEC)
 395                 seg = UIO_BVEC;
 396         ret = zpl_iter_write_common(kiocb, from->iov, from->nr_segs,
 397             iov_iter_count(from), seg, from->iov_offset);
 398         if (ret > 0)
 399                 iov_iter_advance(from, ret);
 400         return (ret);
 401 }
 402 #else
 403 static ssize_t
 404 zpl_aio_write(struct kiocb *kiocb, const struct iovec *iovp,
 405     unsigned long nr_segs, loff_t pos)
 406 {
 407         return (zpl_iter_write_common(kiocb, iovp, nr_segs, kiocb->ki_nbytes,
 408             UIO_USERSPACE, 0));
 409 }
 410 #endif /* HAVE_VFS_RW_ITERATE */
 411
 412 static loff_t
 413 zpl_llseek(struct file *filp, loff_t offset, int whence)
 414 {
 415 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
 416         fstrans_cookie_t cookie;
 417
 418         if (whence == SEEK_DATA || whence == SEEK_HOLE) {
 419                 struct inode *ip = filp->f_mapping->host;
 420                 loff_t maxbytes = ip->i_sb->s_maxbytes;
 421                 loff_t error;
 422
 423                 spl_inode_lock_shared(ip);
 424                 cookie = spl_fstrans_mark();
 425                 error = -zfs_holey(ip, whence, &offset);
 426                 spl_fstrans_unmark(cookie);
 427                 if (error == 0)
 428                         error = lseek_execute(filp, ip, offset, maxbytes);
 429                 spl_inode_unlock_shared(ip);
 430
 431                 return (error);
 432         }
 433 #endif /* SEEK_HOLE && SEEK_DATA */
 434
 435         return (generic_file_llseek(filp, offset, whence));
 436 }
 437
 438 /*
 439  * It's worth taking a moment to describe how mmap is implemented
 440  * for zfs because it differs considerably from other Linux filesystems.
 441  * However, this issue is handled the same way under OpenSolaris.
 442  *
 443  * The issue is that by design zfs bypasses the Linux page cache and
 444  * leaves all caching up to the ARC.  This has been shown to work
 445  * well for the common read(2)/write(2) case.  However, mmap(2)
 446  * is problem because it relies on being tightly integrated with the
 447  * page cache.  To handle this we cache mmap'ed files twice, once in
 448  * the ARC and a second time in the page cache.  The code is careful
 449  * to keep both copies synchronized.
 450  *
 451  * When a file with an mmap'ed region is written to using write(2)
 452  * both the data in the ARC and existing pages in the page cache
 453  * are updated.  For a read(2) data will be read first from the page
 454  * cache then the ARC if needed.  Neither a write(2) or read(2) will
 455  * will ever result in new pages being added to the page cache.
 456  *
 457  * New pages are added to the page cache only via .readpage() which
 458  * is called when the vfs needs to read a page off disk to back the
 459  * virtual memory region.  These pages may be modified without
 460  * notifying the ARC and will be written out periodically via
 461  * .writepage().  This will occur due to either a sync or the usual
 462  * page aging behavior.  Note because a read(2) of a mmap'ed file
 463  * will always check the page cache first even when the ARC is out
 464  * of date correct data will still be returned.
 465  *
 466  * While this implementation ensures correct behavior it does have
 467  * have some drawbacks.  The most obvious of which is that it
 468  * increases the required memory footprint when access mmap'ed
 469  * files.  It also adds additional complexity to the code keeping
 470  * both caches synchronized.
 471  *
 472  * Longer term it may be possible to cleanly resolve this wart by
 473  * mapping page cache pages directly on to the ARC buffers.  The
 474  * Linux address space operations are flexible enough to allow
 475  * selection of which pages back a particular index.  The trick
 476  * would be working out the details of which subsystem is in
 477  * charge, the ARC, the page cache, or both.  It may also prove
 478  * helpful to move the ARC buffers to a scatter-gather lists
 479  * rather than a vmalloc'ed region.
 480  */
 481 static int
 482 zpl_mmap(struct file *filp, struct vm_area_struct *vma)
 483 {
 484         struct inode *ip = filp->f_mapping->host;
 485         znode_t *zp = ITOZ(ip);
 486         int error;
 487         fstrans_cookie_t cookie;
 488
 489         cookie = spl_fstrans_mark();
 490         error = -zfs_map(ip, vma->vm_pgoff, (caddr_t *)vma->vm_start,
 491             (size_t)(vma->vm_end - vma->vm_start), vma->vm_flags);
 492         spl_fstrans_unmark(cookie);
 493         if (error)
 494                 return (error);
 495
 496         error = generic_file_mmap(filp, vma);
 497         if (error)
 498                 return (error);
 499
 500         mutex_enter(&zp->z_lock);
 501         zp->z_is_mapped = 1;
 502         mutex_exit(&zp->z_lock);
 503
 504         return (error);
 505 }
 506
 507 /*
 508  * Populate a page with data for the Linux page cache.  This function is
 509  * only used to support mmap(2).  There will be an identical copy of the
 510  * data in the ARC which is kept up to date via .write() and .writepage().
 511  *
 512  * Current this function relies on zpl_read_common() and the O_DIRECT
 513  * flag to read in a page.  This works but the more correct way is to
 514  * update zfs_fillpage() to be Linux friendly and use that interface.
 515  */
 516 static int
 517 zpl_readpage(struct file *filp, struct page *pp)
 518 {
 519         struct inode *ip;
 520         struct page *pl[1];
 521         int error = 0;
 522         fstrans_cookie_t cookie;
 523
 524         ASSERT(PageLocked(pp));
 525         ip = pp->mapping->host;
 526         pl[0] = pp;
 527
 528         cookie = spl_fstrans_mark();
 529         error = -zfs_getpage(ip, pl, 1);
 530         spl_fstrans_unmark(cookie);
 531
 532         if (error) {
 533                 SetPageError(pp);
 534                 ClearPageUptodate(pp);
 535         } else {
 536                 ClearPageError(pp);
 537                 SetPageUptodate(pp);
 538                 flush_dcache_page(pp);
 539         }
 540
 541         unlock_page(pp);
 542         return (error);
 543 }
 544
 545 /*
 546  * Populate a set of pages with data for the Linux page cache.  This
 547  * function will only be called for read ahead and never for demand
 548  * paging.  For simplicity, the code relies on read_cache_pages() to
 549  * correctly lock each page for IO and call zpl_readpage().
 550  */
 551 static int
 552 zpl_readpages(struct file *filp, struct address_space *mapping,
 553         struct list_head *pages, unsigned nr_pages)
 554 {
 555         return (read_cache_pages(mapping, pages,
 556             (filler_t *)zpl_readpage, filp));
 557 }
 558
 559 int
 560 zpl_putpage(struct page *pp, struct writeback_control *wbc, void *data)
 561 {
 562         struct address_space *mapping = data;
 563         fstrans_cookie_t cookie;
 564
 565         ASSERT(PageLocked(pp));
 566         ASSERT(!PageWriteback(pp));
 567
 568         cookie = spl_fstrans_mark();
 569         (void) zfs_putpage(mapping->host, pp, wbc);
 570         spl_fstrans_unmark(cookie);
 571
 572         return (0);
 573 }
 574
 575 static int
 576 zpl_writepages(struct address_space *mapping, struct writeback_control *wbc)
 577 {
 578         znode_t         *zp = ITOZ(mapping->host);
 579         zfs_sb_t        *zsb = ITOZSB(mapping->host);
 580         enum writeback_sync_modes sync_mode;
 581         int result;
 582
 583         ZFS_ENTER(zsb);
 584         if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
 585                 wbc->sync_mode = WB_SYNC_ALL;
 586         ZFS_EXIT(zsb);
 587         sync_mode = wbc->sync_mode;
 588
 589         /*
 590          * We don't want to run write_cache_pages() in SYNC mode here, because
 591          * that would make putpage() wait for a single page to be committed to
 592          * disk every single time, resulting in atrocious performance. Instead
 593          * we run it once in non-SYNC mode so that the ZIL gets all the data,
 594          * and then we commit it all in one go.
 595          */
 596         wbc->sync_mode = WB_SYNC_NONE;
 597         result = write_cache_pages(mapping, wbc, zpl_putpage, mapping);
 598         if (sync_mode != wbc->sync_mode) {
 599                 ZFS_ENTER(zsb);
 600                 ZFS_VERIFY_ZP(zp);
 601                 if (zsb->z_log != NULL)
 602                         zil_commit(zsb->z_log, zp->z_id);
 603                 ZFS_EXIT(zsb);
 604
 605                 /*
 606                  * We need to call write_cache_pages() again (we can't just
 607                  * return after the commit) because the previous call in
 608                  * non-SYNC mode does not guarantee that we got all the dirty
 609                  * pages (see the implementation of write_cache_pages() for
 610                  * details). That being said, this is a no-op in most cases.
 611                  */
 612                 wbc->sync_mode = sync_mode;
 613                 result = write_cache_pages(mapping, wbc, zpl_putpage, mapping);
 614         }
 615         return (result);
 616 }
 617
 618 /*
 619  * Write out dirty pages to the ARC, this function is only required to
 620  * support mmap(2).  Mapped pages may be dirtied by memory operations
 621  * which never call .write().  These dirty pages are kept in sync with
 622  * the ARC buffers via this hook.
 623  */
 624 static int
 625 zpl_writepage(struct page *pp, struct writeback_control *wbc)
 626 {
 627         if (ITOZSB(pp->mapping->host)->z_os->os_sync == ZFS_SYNC_ALWAYS)
 628                 wbc->sync_mode = WB_SYNC_ALL;
 629
 630         return (zpl_putpage(pp, wbc, pp->mapping));
 631 }
 632
 633 /*
 634  * The only flag combination which matches the behavior of zfs_space()
 635  * is FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE.  The FALLOC_FL_PUNCH_HOLE
 636  * flag was introduced in the 2.6.38 kernel.
 637  */
 638 #if defined(HAVE_FILE_FALLOCATE) || defined(HAVE_INODE_FALLOCATE)
 639 long
 640 zpl_fallocate_common(struct inode *ip, int mode, loff_t offset, loff_t len)
 641 {
 642         int error = -EOPNOTSUPP;
 643
 644 #if defined(FALLOC_FL_PUNCH_HOLE) && defined(FALLOC_FL_KEEP_SIZE)
 645         cred_t *cr = CRED();
 646         flock64_t bf;
 647         loff_t olen;
 648         fstrans_cookie_t cookie;
 649
 650         if (mode != (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
 651                 return (error);
 652
 653         if (offset < 0 || len <= 0)
 654                 return (-EINVAL);
 655
 656         spl_inode_lock(ip);
 657         olen = i_size_read(ip);
 658
 659         if (offset > olen) {
 660                 spl_inode_unlock(ip);
 661                 return (0);
 662         }
 663         if (offset + len > olen)
 664                 len = olen - offset;
 665         bf.l_type = F_WRLCK;
 666         bf.l_whence = 0;
 667         bf.l_start = offset;
 668         bf.l_len = len;
 669         bf.l_pid = 0;
 670
 671         crhold(cr);
 672         cookie = spl_fstrans_mark();
 673         error = -zfs_space(ip, F_FREESP, &bf, FWRITE, offset, cr);
 674         spl_fstrans_unmark(cookie);
 675         spl_inode_unlock(ip);
 676
 677         crfree(cr);
 678 #endif /* defined(FALLOC_FL_PUNCH_HOLE) && defined(FALLOC_FL_KEEP_SIZE) */
 679
 680         ASSERT3S(error, <=, 0);
 681         return (error);
 682 }
 683 #endif /* defined(HAVE_FILE_FALLOCATE) || defined(HAVE_INODE_FALLOCATE) */
 684
 685 #ifdef HAVE_FILE_FALLOCATE
 686 static long
 687 zpl_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
 688 {
 689         return zpl_fallocate_common(file_inode(filp),
 690             mode, offset, len);
 691 }
 692 #endif /* HAVE_FILE_FALLOCATE */
 693
 694 /*
 695  * Map zfs file z_pflags (xvattr_t) to linux file attributes. Only file
 696  * attributes common to both Linux and Solaris are mapped.
 697  */
 698 static int
 699 zpl_ioctl_getflags(struct file *filp, void __user *arg)
 700 {
 701         struct inode *ip = file_inode(filp);
 702         unsigned int ioctl_flags = 0;
 703         uint64_t zfs_flags = ITOZ(ip)->z_pflags;
 704         int error;
 705
 706         if (zfs_flags & ZFS_IMMUTABLE)
 707                 ioctl_flags |= FS_IMMUTABLE_FL;
 708
 709         if (zfs_flags & ZFS_APPENDONLY)
 710                 ioctl_flags |= FS_APPEND_FL;
 711
 712         if (zfs_flags & ZFS_NODUMP)
 713                 ioctl_flags |= FS_NODUMP_FL;
 714
 715         ioctl_flags &= FS_FL_USER_VISIBLE;
 716
 717         error = copy_to_user(arg, &ioctl_flags, sizeof (ioctl_flags));
 718
 719         return (error);
 720 }
 721
 722 /*
 723  * fchange() is a helper macro to detect if we have been asked to change a
 724  * flag. This is ugly, but the requirement that we do this is a consequence of
 725  * how the Linux file attribute interface was designed. Another consequence is
 726  * that concurrent modification of files suffers from a TOCTOU race. Neither
 727  * are things we can fix without modifying the kernel-userland interface, which
 728  * is outside of our jurisdiction.
 729  */
 730
 731 #define fchange(f0, f1, b0, b1) ((((f0) & (b0)) == (b0)) != \
 732         (((b1) & (f1)) == (f1)))
 733
 734 static int
 735 zpl_ioctl_setflags(struct file *filp, void __user *arg)
 736 {
 737         struct inode    *ip = file_inode(filp);
 738         uint64_t        zfs_flags = ITOZ(ip)->z_pflags;
 739         unsigned int    ioctl_flags;
 740         cred_t          *cr = CRED();
 741         xvattr_t        xva;
 742         xoptattr_t      *xoap;
 743         int             error;
 744         fstrans_cookie_t cookie;
 745
 746         if (copy_from_user(&ioctl_flags, arg, sizeof (ioctl_flags)))
 747                 return (-EFAULT);
 748
 749         if ((ioctl_flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NODUMP_FL)))
 750                 return (-EOPNOTSUPP);
 751
 752         if ((ioctl_flags & ~(FS_FL_USER_MODIFIABLE)))
 753                 return (-EACCES);
 754
 755         if ((fchange(ioctl_flags, zfs_flags, FS_IMMUTABLE_FL, ZFS_IMMUTABLE) ||
 756             fchange(ioctl_flags, zfs_flags, FS_APPEND_FL, ZFS_APPENDONLY)) &&
 757             !capable(CAP_LINUX_IMMUTABLE))
 758                 return (-EACCES);
 759
 760         if (!zpl_inode_owner_or_capable(ip))
 761                 return (-EACCES);
 762
 763         xva_init(&xva);
 764         xoap = xva_getxoptattr(&xva);
 765
 766         XVA_SET_REQ(&xva, XAT_IMMUTABLE);
 767         if (ioctl_flags & FS_IMMUTABLE_FL)
 768                 xoap->xoa_immutable = B_TRUE;
 769
 770         XVA_SET_REQ(&xva, XAT_APPENDONLY);
 771         if (ioctl_flags & FS_APPEND_FL)
 772                 xoap->xoa_appendonly = B_TRUE;
 773
 774         XVA_SET_REQ(&xva, XAT_NODUMP);
 775         if (ioctl_flags & FS_NODUMP_FL)
 776                 xoap->xoa_nodump = B_TRUE;
 777
 778         crhold(cr);
 779         cookie = spl_fstrans_mark();
 780         error = -zfs_setattr(ip, (vattr_t *)&xva, 0, cr);
 781         spl_fstrans_unmark(cookie);
 782         crfree(cr);
 783
 784         return (error);
 785 }
 786
 787 static long
 788 zpl_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 789 {
 790         switch (cmd) {
 791         case FS_IOC_GETFLAGS:
 792                 return (zpl_ioctl_getflags(filp, (void *)arg));
 793         case FS_IOC_SETFLAGS:
 794                 return (zpl_ioctl_setflags(filp, (void *)arg));
 795         default:
 796                 return (-ENOTTY);
 797         }
 798 }
 799
 800 #ifdef CONFIG_COMPAT
 801 static long
 802 zpl_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 803 {
 804         switch (cmd) {
 805         case FS_IOC32_GETFLAGS:
 806                 cmd = FS_IOC_GETFLAGS;
 807                 break;
 808         case FS_IOC32_SETFLAGS:
 809                 cmd = FS_IOC_SETFLAGS;
 810                 break;
 811         default:
 812                 return (-ENOTTY);
 813         }
 814         return (zpl_ioctl(filp, cmd, (unsigned long)compat_ptr(arg)));
 815 }
 816 #endif /* CONFIG_COMPAT */
 817
 818
 819 const struct address_space_operations zpl_address_space_operations = {
 820         .readpages      = zpl_readpages,
 821         .readpage       = zpl_readpage,
 822         .writepage      = zpl_writepage,
 823         .writepages     = zpl_writepages,
 824 };
 825
 826 const struct file_operations zpl_file_operations = {
 827         .open           = zpl_open,
 828         .release        = zpl_release,
 829         .llseek         = zpl_llseek,
 830         .read           = zpl_read,
 831         .write          = zpl_write,
 832 #ifdef HAVE_VFS_RW_ITERATE
 833         .read_iter      = zpl_iter_read,
 834         .write_iter     = zpl_iter_write,
 835 #else
 836         .aio_read       = zpl_aio_read,
 837         .aio_write      = zpl_aio_write,
 838 #endif
 839         .mmap           = zpl_mmap,
 840         .fsync          = zpl_fsync,
 841         .aio_fsync      = zpl_aio_fsync,
 842 #ifdef HAVE_FILE_FALLOCATE
 843         .fallocate      = zpl_fallocate,
 844 #endif /* HAVE_FILE_FALLOCATE */
 845         .unlocked_ioctl = zpl_ioctl,
 846 #ifdef CONFIG_COMPAT
 847         .compat_ioctl   = zpl_compat_ioctl,
 848 #endif
 849 };
 850
 851 const struct file_operations zpl_dir_file_operations = {
 852         .llseek         = generic_file_llseek,
 853         .read           = generic_read_dir,
 854 #ifdef HAVE_VFS_ITERATE_SHARED
 855         .iterate_shared = zpl_iterate,
 856 #elif defined(HAVE_VFS_ITERATE)
 857         .iterate        = zpl_iterate,
 858 #else
 859         .readdir        = zpl_readdir,
 860 #endif
 861         .fsync          = zpl_fsync,
 862         .unlocked_ioctl = zpl_ioctl,
 863 #ifdef CONFIG_COMPAT
 864         .compat_ioctl   = zpl_compat_ioctl,
 865 #endif
 866 };