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