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 #ifdef HAVE_FILE_AIO_FSYNC
 134 static int
 135 zpl_aio_fsync(struct kiocb *kiocb, int datasync)
 136 {
 137         struct file *filp = kiocb->ki_filp;
 138         return (zpl_fsync(filp, file_dentry(filp), datasync));
 139 }
 140 #endif
 141
 142 #elif defined(HAVE_FSYNC_WITHOUT_DENTRY)
 143 /*
 144  * Linux 2.6.35 - 3.0 API,
 145  * As of 2.6.35 the dentry argument to the fops->fsync() hook was deemed
 146  * redundant.  The dentry is still accessible via filp->f_path.dentry,
 147  * and we are guaranteed that filp will never be NULL.
 148  */
 149 static int
 150 zpl_fsync(struct file *filp, int datasync)
 151 {
 152         struct inode *inode = filp->f_mapping->host;
 153         cred_t *cr = CRED();
 154         int error;
 155         fstrans_cookie_t cookie;
 156
 157         crhold(cr);
 158         cookie = spl_fstrans_mark();
 159         error = -zfs_fsync(inode, datasync, cr);
 160         spl_fstrans_unmark(cookie);
 161         crfree(cr);
 162         ASSERT3S(error, <=, 0);
 163
 164         return (error);
 165 }
 166
 167 #ifdef HAVE_FILE_AIO_FSYNC
 168 static int
 169 zpl_aio_fsync(struct kiocb *kiocb, int datasync)
 170 {
 171         return (zpl_fsync(kiocb->ki_filp, datasync));
 172 }
 173 #endif
 174
 175 #elif defined(HAVE_FSYNC_RANGE)
 176 /*
 177  * Linux 3.1 - 3.x API,
 178  * As of 3.1 the responsibility to call filemap_write_and_wait_range() has
 179  * been pushed down in to the .fsync() vfs hook.  Additionally, the i_mutex
 180  * lock is no longer held by the caller, for zfs we don't require the lock
 181  * to be held so we don't acquire it.
 182  */
 183 static int
 184 zpl_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
 185 {
 186         struct inode *inode = filp->f_mapping->host;
 187         cred_t *cr = CRED();
 188         int error;
 189         fstrans_cookie_t cookie;
 190
 191         error = filemap_write_and_wait_range(inode->i_mapping, start, end);
 192         if (error)
 193                 return (error);
 194
 195         crhold(cr);
 196         cookie = spl_fstrans_mark();
 197         error = -zfs_fsync(inode, datasync, cr);
 198         spl_fstrans_unmark(cookie);
 199         crfree(cr);
 200         ASSERT3S(error, <=, 0);
 201
 202         return (error);
 203 }
 204
 205 #ifdef HAVE_FILE_AIO_FSYNC
 206 static int
 207 zpl_aio_fsync(struct kiocb *kiocb, int datasync)
 208 {
 209         return (zpl_fsync(kiocb->ki_filp, kiocb->ki_pos, -1, datasync));
 210 }
 211 #endif
 212
 213 #else
 214 #error "Unsupported fops->fsync() implementation"
 215 #endif
 216
 217 static ssize_t
 218 zpl_read_common_iovec(struct inode *ip, const struct iovec *iovp, size_t count,
 219     unsigned long nr_segs, loff_t *ppos, uio_seg_t segment, int flags,
 220     cred_t *cr, size_t skip)
 221 {
 222         ssize_t read;
 223         uio_t uio;
 224         int error;
 225         fstrans_cookie_t cookie;
 226
 227         uio.uio_iov = iovp;
 228         uio.uio_skip = skip;
 229         uio.uio_resid = count;
 230         uio.uio_iovcnt = nr_segs;
 231         uio.uio_loffset = *ppos;
 232         uio.uio_limit = MAXOFFSET_T;
 233         uio.uio_segflg = segment;
 234
 235         cookie = spl_fstrans_mark();
 236         error = -zfs_read(ip, &uio, flags, cr);
 237         spl_fstrans_unmark(cookie);
 238         if (error < 0)
 239                 return (error);
 240
 241         read = count - uio.uio_resid;
 242         *ppos += read;
 243         task_io_account_read(read);
 244
 245         return (read);
 246 }
 247
 248 inline ssize_t
 249 zpl_read_common(struct inode *ip, const char *buf, size_t len, loff_t *ppos,
 250     uio_seg_t segment, int flags, cred_t *cr)
 251 {
 252         struct iovec iov;
 253
 254         iov.iov_base = (void *)buf;
 255         iov.iov_len = len;
 256
 257         return (zpl_read_common_iovec(ip, &iov, len, 1, ppos, segment,
 258             flags, cr, 0));
 259 }
 260
 261 static ssize_t
 262 zpl_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 263 {
 264         cred_t *cr = CRED();
 265         ssize_t read;
 266
 267         crhold(cr);
 268         read = zpl_read_common(filp->f_mapping->host, buf, len, ppos,
 269             UIO_USERSPACE, filp->f_flags, cr);
 270         crfree(cr);
 271
 272         file_accessed(filp);
 273         return (read);
 274 }
 275
 276 static ssize_t
 277 zpl_iter_read_common(struct kiocb *kiocb, const struct iovec *iovp,
 278     unsigned long nr_segs, size_t count, uio_seg_t seg, size_t skip)
 279 {
 280         cred_t *cr = CRED();
 281         struct file *filp = kiocb->ki_filp;
 282         ssize_t read;
 283
 284         crhold(cr);
 285         read = zpl_read_common_iovec(filp->f_mapping->host, iovp, count,
 286             nr_segs, &kiocb->ki_pos, seg, filp->f_flags, cr, skip);
 287         crfree(cr);
 288
 289         file_accessed(filp);
 290         return (read);
 291 }
 292
 293 #if defined(HAVE_VFS_RW_ITERATE)
 294 static ssize_t
 295 zpl_iter_read(struct kiocb *kiocb, struct iov_iter *to)
 296 {
 297         ssize_t ret;
 298         uio_seg_t seg = UIO_USERSPACE;
 299         if (to->type & ITER_KVEC)
 300                 seg = UIO_SYSSPACE;
 301         if (to->type & ITER_BVEC)
 302                 seg = UIO_BVEC;
 303         ret = zpl_iter_read_common(kiocb, to->iov, to->nr_segs,
 304             iov_iter_count(to), seg, to->iov_offset);
 305         if (ret > 0)
 306                 iov_iter_advance(to, ret);
 307         return (ret);
 308 }
 309 #else
 310 static ssize_t
 311 zpl_aio_read(struct kiocb *kiocb, const struct iovec *iovp,
 312     unsigned long nr_segs, loff_t pos)
 313 {
 314         return (zpl_iter_read_common(kiocb, iovp, nr_segs, kiocb->ki_nbytes,
 315             UIO_USERSPACE, 0));
 316 }
 317 #endif /* HAVE_VFS_RW_ITERATE */
 318
 319 static ssize_t
 320 zpl_write_common_iovec(struct inode *ip, const struct iovec *iovp, size_t count,
 321     unsigned long nr_segs, loff_t *ppos, uio_seg_t segment, int flags,
 322     cred_t *cr, size_t skip)
 323 {
 324         ssize_t wrote;
 325         uio_t uio;
 326         int error;
 327         fstrans_cookie_t cookie;
 328
 329         if (flags & O_APPEND)
 330                 *ppos = i_size_read(ip);
 331
 332         uio.uio_iov = iovp;
 333         uio.uio_skip = skip;
 334         uio.uio_resid = count;
 335         uio.uio_iovcnt = nr_segs;
 336         uio.uio_loffset = *ppos;
 337         uio.uio_limit = MAXOFFSET_T;
 338         uio.uio_segflg = segment;
 339
 340         cookie = spl_fstrans_mark();
 341         error = -zfs_write(ip, &uio, flags, cr);
 342         spl_fstrans_unmark(cookie);
 343         if (error < 0)
 344                 return (error);
 345
 346         wrote = count - uio.uio_resid;
 347         *ppos += wrote;
 348         task_io_account_write(wrote);
 349
 350         return (wrote);
 351 }
 352 inline ssize_t
 353 zpl_write_common(struct inode *ip, const char *buf, size_t len, loff_t *ppos,
 354     uio_seg_t segment, int flags, cred_t *cr)
 355 {
 356         struct iovec iov;
 357
 358         iov.iov_base = (void *)buf;
 359         iov.iov_len = len;
 360
 361         return (zpl_write_common_iovec(ip, &iov, len, 1, ppos, segment,
 362             flags, cr, 0));
 363 }
 364
 365 static ssize_t
 366 zpl_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
 367 {
 368         cred_t *cr = CRED();
 369         ssize_t wrote;
 370
 371         crhold(cr);
 372         wrote = zpl_write_common(filp->f_mapping->host, buf, len, ppos,
 373             UIO_USERSPACE, filp->f_flags, cr);
 374         crfree(cr);
 375
 376         return (wrote);
 377 }
 378
 379 static ssize_t
 380 zpl_iter_write_common(struct kiocb *kiocb, const struct iovec *iovp,
 381     unsigned long nr_segs, size_t count, uio_seg_t seg, size_t skip)
 382 {
 383         cred_t *cr = CRED();
 384         struct file *filp = kiocb->ki_filp;
 385         ssize_t wrote;
 386
 387         crhold(cr);
 388         wrote = zpl_write_common_iovec(filp->f_mapping->host, iovp, count,
 389             nr_segs, &kiocb->ki_pos, seg, filp->f_flags, cr, skip);
 390         crfree(cr);
 391
 392         return (wrote);
 393 }
 394
 395 #if defined(HAVE_VFS_RW_ITERATE)
 396 static ssize_t
 397 zpl_iter_write(struct kiocb *kiocb, struct iov_iter *from)
 398 {
 399         ssize_t ret;
 400         uio_seg_t seg = UIO_USERSPACE;
 401         if (from->type & ITER_KVEC)
 402                 seg = UIO_SYSSPACE;
 403         if (from->type & ITER_BVEC)
 404                 seg = UIO_BVEC;
 405         ret = zpl_iter_write_common(kiocb, from->iov, from->nr_segs,
 406             iov_iter_count(from), seg, from->iov_offset);
 407         if (ret > 0)
 408                 iov_iter_advance(from, ret);
 409         return (ret);
 410 }
 411 #else
 412 static ssize_t
 413 zpl_aio_write(struct kiocb *kiocb, const struct iovec *iovp,
 414     unsigned long nr_segs, loff_t pos)
 415 {
 416         return (zpl_iter_write_common(kiocb, iovp, nr_segs, kiocb->ki_nbytes,
 417             UIO_USERSPACE, 0));
 418 }
 419 #endif /* HAVE_VFS_RW_ITERATE */
 420
 421 static loff_t
 422 zpl_llseek(struct file *filp, loff_t offset, int whence)
 423 {
 424 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
 425         fstrans_cookie_t cookie;
 426
 427         if (whence == SEEK_DATA || whence == SEEK_HOLE) {
 428                 struct inode *ip = filp->f_mapping->host;
 429                 loff_t maxbytes = ip->i_sb->s_maxbytes;
 430                 loff_t error;
 431
 432                 spl_inode_lock_shared(ip);
 433                 cookie = spl_fstrans_mark();
 434                 error = -zfs_holey(ip, whence, &offset);
 435                 spl_fstrans_unmark(cookie);
 436                 if (error == 0)
 437                         error = lseek_execute(filp, ip, offset, maxbytes);
 438                 spl_inode_unlock_shared(ip);
 439
 440                 return (error);
 441         }
 442 #endif /* SEEK_HOLE && SEEK_DATA */
 443
 444         return (generic_file_llseek(filp, offset, whence));
 445 }
 446
 447 /*
 448  * It's worth taking a moment to describe how mmap is implemented
 449  * for zfs because it differs considerably from other Linux filesystems.
 450  * However, this issue is handled the same way under OpenSolaris.
 451  *
 452  * The issue is that by design zfs bypasses the Linux page cache and
 453  * leaves all caching up to the ARC.  This has been shown to work
 454  * well for the common read(2)/write(2) case.  However, mmap(2)
 455  * is problem because it relies on being tightly integrated with the
 456  * page cache.  To handle this we cache mmap'ed files twice, once in
 457  * the ARC and a second time in the page cache.  The code is careful
 458  * to keep both copies synchronized.
 459  *
 460  * When a file with an mmap'ed region is written to using write(2)
 461  * both the data in the ARC and existing pages in the page cache
 462  * are updated.  For a read(2) data will be read first from the page
 463  * cache then the ARC if needed.  Neither a write(2) or read(2) will
 464  * will ever result in new pages being added to the page cache.
 465  *
 466  * New pages are added to the page cache only via .readpage() which
 467  * is called when the vfs needs to read a page off disk to back the
 468  * virtual memory region.  These pages may be modified without
 469  * notifying the ARC and will be written out periodically via
 470  * .writepage().  This will occur due to either a sync or the usual
 471  * page aging behavior.  Note because a read(2) of a mmap'ed file
 472  * will always check the page cache first even when the ARC is out
 473  * of date correct data will still be returned.
 474  *
 475  * While this implementation ensures correct behavior it does have
 476  * have some drawbacks.  The most obvious of which is that it
 477  * increases the required memory footprint when access mmap'ed
 478  * files.  It also adds additional complexity to the code keeping
 479  * both caches synchronized.
 480  *
 481  * Longer term it may be possible to cleanly resolve this wart by
 482  * mapping page cache pages directly on to the ARC buffers.  The
 483  * Linux address space operations are flexible enough to allow
 484  * selection of which pages back a particular index.  The trick
 485  * would be working out the details of which subsystem is in
 486  * charge, the ARC, the page cache, or both.  It may also prove
 487  * helpful to move the ARC buffers to a scatter-gather lists
 488  * rather than a vmalloc'ed region.
 489  */
 490 static int
 491 zpl_mmap(struct file *filp, struct vm_area_struct *vma)
 492 {
 493         struct inode *ip = filp->f_mapping->host;
 494         znode_t *zp = ITOZ(ip);
 495         int error;
 496         fstrans_cookie_t cookie;
 497
 498         cookie = spl_fstrans_mark();
 499         error = -zfs_map(ip, vma->vm_pgoff, (caddr_t *)vma->vm_start,
 500             (size_t)(vma->vm_end - vma->vm_start), vma->vm_flags);
 501         spl_fstrans_unmark(cookie);
 502         if (error)
 503                 return (error);
 504
 505         error = generic_file_mmap(filp, vma);
 506         if (error)
 507                 return (error);
 508
 509         mutex_enter(&zp->z_lock);
 510         zp->z_is_mapped = 1;
 511         mutex_exit(&zp->z_lock);
 512
 513         return (error);
 514 }
 515
 516 /*
 517  * Populate a page with data for the Linux page cache.  This function is
 518  * only used to support mmap(2).  There will be an identical copy of the
 519  * data in the ARC which is kept up to date via .write() and .writepage().
 520  *
 521  * Current this function relies on zpl_read_common() and the O_DIRECT
 522  * flag to read in a page.  This works but the more correct way is to
 523  * update zfs_fillpage() to be Linux friendly and use that interface.
 524  */
 525 static int
 526 zpl_readpage(struct file *filp, struct page *pp)
 527 {
 528         struct inode *ip;
 529         struct page *pl[1];
 530         int error = 0;
 531         fstrans_cookie_t cookie;
 532
 533         ASSERT(PageLocked(pp));
 534         ip = pp->mapping->host;
 535         pl[0] = pp;
 536
 537         cookie = spl_fstrans_mark();
 538         error = -zfs_getpage(ip, pl, 1);
 539         spl_fstrans_unmark(cookie);
 540
 541         if (error) {
 542                 SetPageError(pp);
 543                 ClearPageUptodate(pp);
 544         } else {
 545                 ClearPageError(pp);
 546                 SetPageUptodate(pp);
 547                 flush_dcache_page(pp);
 548         }
 549
 550         unlock_page(pp);
 551         return (error);
 552 }
 553
 554 /*
 555  * Populate a set of pages with data for the Linux page cache.  This
 556  * function will only be called for read ahead and never for demand
 557  * paging.  For simplicity, the code relies on read_cache_pages() to
 558  * correctly lock each page for IO and call zpl_readpage().
 559  */
 560 static int
 561 zpl_readpages(struct file *filp, struct address_space *mapping,
 562         struct list_head *pages, unsigned nr_pages)
 563 {
 564         return (read_cache_pages(mapping, pages,
 565             (filler_t *)zpl_readpage, filp));
 566 }
 567
 568 int
 569 zpl_putpage(struct page *pp, struct writeback_control *wbc, void *data)
 570 {
 571         struct address_space *mapping = data;
 572         fstrans_cookie_t cookie;
 573
 574         ASSERT(PageLocked(pp));
 575         ASSERT(!PageWriteback(pp));
 576
 577         cookie = spl_fstrans_mark();
 578         (void) zfs_putpage(mapping->host, pp, wbc);
 579         spl_fstrans_unmark(cookie);
 580
 581         return (0);
 582 }
 583
 584 static int
 585 zpl_writepages(struct address_space *mapping, struct writeback_control *wbc)
 586 {
 587         znode_t         *zp = ITOZ(mapping->host);
 588         zfs_sb_t        *zsb = ITOZSB(mapping->host);
 589         enum writeback_sync_modes sync_mode;
 590         int result;
 591
 592         ZFS_ENTER(zsb);
 593         if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
 594                 wbc->sync_mode = WB_SYNC_ALL;
 595         ZFS_EXIT(zsb);
 596         sync_mode = wbc->sync_mode;
 597
 598         /*
 599          * We don't want to run write_cache_pages() in SYNC mode here, because
 600          * that would make putpage() wait for a single page to be committed to
 601          * disk every single time, resulting in atrocious performance. Instead
 602          * we run it once in non-SYNC mode so that the ZIL gets all the data,
 603          * and then we commit it all in one go.
 604          */
 605         wbc->sync_mode = WB_SYNC_NONE;
 606         result = write_cache_pages(mapping, wbc, zpl_putpage, mapping);
 607         if (sync_mode != wbc->sync_mode) {
 608                 ZFS_ENTER(zsb);
 609                 ZFS_VERIFY_ZP(zp);
 610                 if (zsb->z_log != NULL)
 611                         zil_commit(zsb->z_log, zp->z_id);
 612                 ZFS_EXIT(zsb);
 613
 614                 /*
 615                  * We need to call write_cache_pages() again (we can't just
 616                  * return after the commit) because the previous call in
 617                  * non-SYNC mode does not guarantee that we got all the dirty
 618                  * pages (see the implementation of write_cache_pages() for
 619                  * details). That being said, this is a no-op in most cases.
 620                  */
 621                 wbc->sync_mode = sync_mode;
 622                 result = write_cache_pages(mapping, wbc, zpl_putpage, mapping);
 623         }
 624         return (result);
 625 }
 626
 627 /*
 628  * Write out dirty pages to the ARC, this function is only required to
 629  * support mmap(2).  Mapped pages may be dirtied by memory operations
 630  * which never call .write().  These dirty pages are kept in sync with
 631  * the ARC buffers via this hook.
 632  */
 633 static int
 634 zpl_writepage(struct page *pp, struct writeback_control *wbc)
 635 {
 636         if (ITOZSB(pp->mapping->host)->z_os->os_sync == ZFS_SYNC_ALWAYS)
 637                 wbc->sync_mode = WB_SYNC_ALL;
 638
 639         return (zpl_putpage(pp, wbc, pp->mapping));
 640 }
 641
 642 /*
 643  * The only flag combination which matches the behavior of zfs_space()
 644  * is FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE.  The FALLOC_FL_PUNCH_HOLE
 645  * flag was introduced in the 2.6.38 kernel.
 646  */
 647 #if defined(HAVE_FILE_FALLOCATE) || defined(HAVE_INODE_FALLOCATE)
 648 long
 649 zpl_fallocate_common(struct inode *ip, int mode, loff_t offset, loff_t len)
 650 {
 651         int error = -EOPNOTSUPP;
 652
 653 #if defined(FALLOC_FL_PUNCH_HOLE) && defined(FALLOC_FL_KEEP_SIZE)
 654         cred_t *cr = CRED();
 655         flock64_t bf;
 656         loff_t olen;
 657         fstrans_cookie_t cookie;
 658
 659         if (mode != (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
 660                 return (error);
 661
 662         if (offset < 0 || len <= 0)
 663                 return (-EINVAL);
 664
 665         spl_inode_lock(ip);
 666         olen = i_size_read(ip);
 667
 668         if (offset > olen) {
 669                 spl_inode_unlock(ip);
 670                 return (0);
 671         }
 672         if (offset + len > olen)
 673                 len = olen - offset;
 674         bf.l_type = F_WRLCK;
 675         bf.l_whence = 0;
 676         bf.l_start = offset;
 677         bf.l_len = len;
 678         bf.l_pid = 0;
 679
 680         crhold(cr);
 681         cookie = spl_fstrans_mark();
 682         error = -zfs_space(ip, F_FREESP, &bf, FWRITE, offset, cr);
 683         spl_fstrans_unmark(cookie);
 684         spl_inode_unlock(ip);
 685
 686         crfree(cr);
 687 #endif /* defined(FALLOC_FL_PUNCH_HOLE) && defined(FALLOC_FL_KEEP_SIZE) */
 688
 689         ASSERT3S(error, <=, 0);
 690         return (error);
 691 }
 692 #endif /* defined(HAVE_FILE_FALLOCATE) || defined(HAVE_INODE_FALLOCATE) */
 693
 694 #ifdef HAVE_FILE_FALLOCATE
 695 static long
 696 zpl_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
 697 {
 698         return zpl_fallocate_common(file_inode(filp),
 699             mode, offset, len);
 700 }
 701 #endif /* HAVE_FILE_FALLOCATE */
 702
 703 /*
 704  * Map zfs file z_pflags (xvattr_t) to linux file attributes. Only file
 705  * attributes common to both Linux and Solaris are mapped.
 706  */
 707 static int
 708 zpl_ioctl_getflags(struct file *filp, void __user *arg)
 709 {
 710         struct inode *ip = file_inode(filp);
 711         unsigned int ioctl_flags = 0;
 712         uint64_t zfs_flags = ITOZ(ip)->z_pflags;
 713         int error;
 714
 715         if (zfs_flags & ZFS_IMMUTABLE)
 716                 ioctl_flags |= FS_IMMUTABLE_FL;
 717
 718         if (zfs_flags & ZFS_APPENDONLY)
 719                 ioctl_flags |= FS_APPEND_FL;
 720
 721         if (zfs_flags & ZFS_NODUMP)
 722                 ioctl_flags |= FS_NODUMP_FL;
 723
 724         ioctl_flags &= FS_FL_USER_VISIBLE;
 725
 726         error = copy_to_user(arg, &ioctl_flags, sizeof (ioctl_flags));
 727
 728         return (error);
 729 }
 730
 731 /*
 732  * fchange() is a helper macro to detect if we have been asked to change a
 733  * flag. This is ugly, but the requirement that we do this is a consequence of
 734  * how the Linux file attribute interface was designed. Another consequence is
 735  * that concurrent modification of files suffers from a TOCTOU race. Neither
 736  * are things we can fix without modifying the kernel-userland interface, which
 737  * is outside of our jurisdiction.
 738  */
 739
 740 #define fchange(f0, f1, b0, b1) (!((f0) & (b0)) != !((f1) & (b1)))
 741
 742 static int
 743 zpl_ioctl_setflags(struct file *filp, void __user *arg)
 744 {
 745         struct inode    *ip = file_inode(filp);
 746         uint64_t        zfs_flags = ITOZ(ip)->z_pflags;
 747         unsigned int    ioctl_flags;
 748         cred_t          *cr = CRED();
 749         xvattr_t        xva;
 750         xoptattr_t      *xoap;
 751         int             error;
 752         fstrans_cookie_t cookie;
 753
 754         if (copy_from_user(&ioctl_flags, arg, sizeof (ioctl_flags)))
 755                 return (-EFAULT);
 756
 757         if ((ioctl_flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NODUMP_FL)))
 758                 return (-EOPNOTSUPP);
 759
 760         if ((ioctl_flags & ~(FS_FL_USER_MODIFIABLE)))
 761                 return (-EACCES);
 762
 763         if ((fchange(ioctl_flags, zfs_flags, FS_IMMUTABLE_FL, ZFS_IMMUTABLE) ||
 764             fchange(ioctl_flags, zfs_flags, FS_APPEND_FL, ZFS_APPENDONLY)) &&
 765             !capable(CAP_LINUX_IMMUTABLE))
 766                 return (-EACCES);
 767
 768         if (!zpl_inode_owner_or_capable(ip))
 769                 return (-EACCES);
 770
 771         xva_init(&xva);
 772         xoap = xva_getxoptattr(&xva);
 773
 774         XVA_SET_REQ(&xva, XAT_IMMUTABLE);
 775         if (ioctl_flags & FS_IMMUTABLE_FL)
 776                 xoap->xoa_immutable = B_TRUE;
 777
 778         XVA_SET_REQ(&xva, XAT_APPENDONLY);
 779         if (ioctl_flags & FS_APPEND_FL)
 780                 xoap->xoa_appendonly = B_TRUE;
 781
 782         XVA_SET_REQ(&xva, XAT_NODUMP);
 783         if (ioctl_flags & FS_NODUMP_FL)
 784                 xoap->xoa_nodump = B_TRUE;
 785
 786         crhold(cr);
 787         cookie = spl_fstrans_mark();
 788         error = -zfs_setattr(ip, (vattr_t *)&xva, 0, cr);
 789         spl_fstrans_unmark(cookie);
 790         crfree(cr);
 791
 792         return (error);
 793 }
 794
 795 static long
 796 zpl_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 797 {
 798         switch (cmd) {
 799         case FS_IOC_GETFLAGS:
 800                 return (zpl_ioctl_getflags(filp, (void *)arg));
 801         case FS_IOC_SETFLAGS:
 802                 return (zpl_ioctl_setflags(filp, (void *)arg));
 803         default:
 804                 return (-ENOTTY);
 805         }
 806 }
 807
 808 #ifdef CONFIG_COMPAT
 809 static long
 810 zpl_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 811 {
 812         switch (cmd) {
 813         case FS_IOC32_GETFLAGS:
 814                 cmd = FS_IOC_GETFLAGS;
 815                 break;
 816         case FS_IOC32_SETFLAGS:
 817                 cmd = FS_IOC_SETFLAGS;
 818                 break;
 819         default:
 820                 return (-ENOTTY);
 821         }
 822         return (zpl_ioctl(filp, cmd, (unsigned long)compat_ptr(arg)));
 823 }
 824 #endif /* CONFIG_COMPAT */
 825
 826
 827 const struct address_space_operations zpl_address_space_operations = {
 828         .readpages      = zpl_readpages,
 829         .readpage       = zpl_readpage,
 830         .writepage      = zpl_writepage,
 831         .writepages     = zpl_writepages,
 832 };
 833
 834 const struct file_operations zpl_file_operations = {
 835         .open           = zpl_open,
 836         .release        = zpl_release,
 837         .llseek         = zpl_llseek,
 838         .read           = zpl_read,
 839         .write          = zpl_write,
 840 #ifdef HAVE_VFS_RW_ITERATE
 841         .read_iter      = zpl_iter_read,
 842         .write_iter     = zpl_iter_write,
 843 #else
 844         .aio_read       = zpl_aio_read,
 845         .aio_write      = zpl_aio_write,
 846 #endif
 847         .mmap           = zpl_mmap,
 848         .fsync          = zpl_fsync,
 849 #ifdef HAVE_FILE_AIO_FSYNC
 850         .aio_fsync      = zpl_aio_fsync,
 851 #endif
 852 #ifdef HAVE_FILE_FALLOCATE
 853         .fallocate      = zpl_fallocate,
 854 #endif /* HAVE_FILE_FALLOCATE */
 855         .unlocked_ioctl = zpl_ioctl,
 856 #ifdef CONFIG_COMPAT
 857         .compat_ioctl   = zpl_compat_ioctl,
 858 #endif
 859 };
 860
 861 const struct file_operations zpl_dir_file_operations = {
 862         .llseek         = generic_file_llseek,
 863         .read           = generic_read_dir,
 864 #ifdef HAVE_VFS_ITERATE_SHARED
 865         .iterate_shared = zpl_iterate,
 866 #elif defined(HAVE_VFS_ITERATE)
 867         .iterate        = zpl_iterate,
 868 #else
 869         .readdir        = zpl_readdir,
 870 #endif
 871         .fsync          = zpl_fsync,
 872         .unlocked_ioctl = zpl_ioctl,
 873 #ifdef CONFIG_COMPAT
 874         .compat_ioctl   = zpl_compat_ioctl,
 875 #endif
 876 };