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/file.h>
  31 #include <sys/dmu_objset.h>
  32 #include <sys/zfs_vfsops.h>
  33 #include <sys/zfs_vnops.h>
  34 #include <sys/zfs_znode.h>
  35 #include <sys/zfs_project.h>
  36
  37
  38 static int
  39 zpl_open(struct inode *ip, struct file *filp)
  40 {
  41         cred_t *cr = CRED();
  42         int error;
  43         fstrans_cookie_t cookie;
  44
  45         error = generic_file_open(ip, filp);
  46         if (error)
  47                 return (error);
  48
  49         crhold(cr);
  50         cookie = spl_fstrans_mark();
  51         error = -zfs_open(ip, filp->f_mode, filp->f_flags, cr);
  52         spl_fstrans_unmark(cookie);
  53         crfree(cr);
  54         ASSERT3S(error, <=, 0);
  55
  56         return (error);
  57 }
  58
  59 static int
  60 zpl_release(struct inode *ip, struct file *filp)
  61 {
  62         cred_t *cr = CRED();
  63         int error;
  64         fstrans_cookie_t cookie;
  65
  66         cookie = spl_fstrans_mark();
  67         if (ITOZ(ip)->z_atime_dirty)
  68                 zfs_mark_inode_dirty(ip);
  69
  70         crhold(cr);
  71         error = -zfs_close(ip, filp->f_flags, cr);
  72         spl_fstrans_unmark(cookie);
  73         crfree(cr);
  74         ASSERT3S(error, <=, 0);
  75
  76         return (error);
  77 }
  78
  79 static int
  80 zpl_iterate(struct file *filp, zpl_dir_context_t *ctx)
  81 {
  82         cred_t *cr = CRED();
  83         int error;
  84         fstrans_cookie_t cookie;
  85
  86         crhold(cr);
  87         cookie = spl_fstrans_mark();
  88         error = -zfs_readdir(file_inode(filp), ctx, cr);
  89         spl_fstrans_unmark(cookie);
  90         crfree(cr);
  91         ASSERT3S(error, <=, 0);
  92
  93         return (error);
  94 }
  95
  96 #if !defined(HAVE_VFS_ITERATE) && !defined(HAVE_VFS_ITERATE_SHARED)
  97 static int
  98 zpl_readdir(struct file *filp, void *dirent, filldir_t filldir)
  99 {
 100         zpl_dir_context_t ctx =
 101             ZPL_DIR_CONTEXT_INIT(dirent, filldir, filp->f_pos);
 102         int error;
 103
 104         error = zpl_iterate(filp, &ctx);
 105         filp->f_pos = ctx.pos;
 106
 107         return (error);
 108 }
 109 #endif /* !HAVE_VFS_ITERATE && !HAVE_VFS_ITERATE_SHARED */
 110
 111 #if defined(HAVE_FSYNC_WITH_DENTRY)
 112 /*
 113  * Linux 2.6.x - 2.6.34 API,
 114  * Through 2.6.34 the nfsd kernel server would pass a NULL 'file struct *'
 115  * to the fops->fsync() hook.  For this reason, we must be careful not to
 116  * use filp unconditionally.
 117  */
 118 static int
 119 zpl_fsync(struct file *filp, struct dentry *dentry, int datasync)
 120 {
 121         cred_t *cr = CRED();
 122         int error;
 123         fstrans_cookie_t cookie;
 124
 125         crhold(cr);
 126         cookie = spl_fstrans_mark();
 127         error = -zfs_fsync(dentry->d_inode, datasync, cr);
 128         spl_fstrans_unmark(cookie);
 129         crfree(cr);
 130         ASSERT3S(error, <=, 0);
 131
 132         return (error);
 133 }
 134
 135 #ifdef HAVE_FILE_AIO_FSYNC
 136 static int
 137 zpl_aio_fsync(struct kiocb *kiocb, int datasync)
 138 {
 139         struct file *filp = kiocb->ki_filp;
 140         return (zpl_fsync(filp, file_dentry(filp), datasync));
 141 }
 142 #endif
 143
 144 #elif defined(HAVE_FSYNC_WITHOUT_DENTRY)
 145 /*
 146  * Linux 2.6.35 - 3.0 API,
 147  * As of 2.6.35 the dentry argument to the fops->fsync() hook was deemed
 148  * redundant.  The dentry is still accessible via filp->f_path.dentry,
 149  * and we are guaranteed that filp will never be NULL.
 150  */
 151 static int
 152 zpl_fsync(struct file *filp, int datasync)
 153 {
 154         struct inode *inode = filp->f_mapping->host;
 155         cred_t *cr = CRED();
 156         int error;
 157         fstrans_cookie_t cookie;
 158
 159         crhold(cr);
 160         cookie = spl_fstrans_mark();
 161         error = -zfs_fsync(inode, datasync, cr);
 162         spl_fstrans_unmark(cookie);
 163         crfree(cr);
 164         ASSERT3S(error, <=, 0);
 165
 166         return (error);
 167 }
 168
 169 #ifdef HAVE_FILE_AIO_FSYNC
 170 static int
 171 zpl_aio_fsync(struct kiocb *kiocb, int datasync)
 172 {
 173         return (zpl_fsync(kiocb->ki_filp, datasync));
 174 }
 175 #endif
 176
 177 #elif defined(HAVE_FSYNC_RANGE)
 178 /*
 179  * Linux 3.1 - 3.x API,
 180  * As of 3.1 the responsibility to call filemap_write_and_wait_range() has
 181  * been pushed down in to the .fsync() vfs hook.  Additionally, the i_mutex
 182  * lock is no longer held by the caller, for zfs we don't require the lock
 183  * to be held so we don't acquire it.
 184  */
 185 static int
 186 zpl_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
 187 {
 188         struct inode *inode = filp->f_mapping->host;
 189         cred_t *cr = CRED();
 190         int error;
 191         fstrans_cookie_t cookie;
 192
 193         error = filemap_write_and_wait_range(inode->i_mapping, start, end);
 194         if (error)
 195                 return (error);
 196
 197         crhold(cr);
 198         cookie = spl_fstrans_mark();
 199         error = -zfs_fsync(inode, datasync, cr);
 200         spl_fstrans_unmark(cookie);
 201         crfree(cr);
 202         ASSERT3S(error, <=, 0);
 203
 204         return (error);
 205 }
 206
 207 #ifdef HAVE_FILE_AIO_FSYNC
 208 static int
 209 zpl_aio_fsync(struct kiocb *kiocb, int datasync)
 210 {
 211         return (zpl_fsync(kiocb->ki_filp, kiocb->ki_pos, -1, datasync));
 212 }
 213 #endif
 214
 215 #else
 216 #error "Unsupported fops->fsync() implementation"
 217 #endif
 218
 219 static inline int
 220 zfs_io_flags(struct kiocb *kiocb)
 221 {
 222         int flags = 0;
 223
 224 #if defined(IOCB_DSYNC)
 225         if (kiocb->ki_flags & IOCB_DSYNC)
 226                 flags |= FDSYNC;
 227 #endif
 228 #if defined(IOCB_SYNC)
 229         if (kiocb->ki_flags & IOCB_SYNC)
 230                 flags |= FSYNC;
 231 #endif
 232 #if defined(IOCB_APPEND)
 233         if (kiocb->ki_flags & IOCB_APPEND)
 234                 flags |= FAPPEND;
 235 #endif
 236 #if defined(IOCB_DIRECT)
 237         if (kiocb->ki_flags & IOCB_DIRECT)
 238                 flags |= FDIRECT;
 239 #endif
 240         return (flags);
 241 }
 242
 243 static ssize_t
 244 zpl_read_common_iovec(struct inode *ip, const struct iovec *iovp, size_t count,
 245     unsigned long nr_segs, loff_t *ppos, uio_seg_t segment, int flags,
 246     cred_t *cr, size_t skip)
 247 {
 248         ssize_t read;
 249         uio_t uio;
 250         int error;
 251         fstrans_cookie_t cookie;
 252
 253         uio.uio_iov = iovp;
 254         uio.uio_skip = skip;
 255         uio.uio_resid = count;
 256         uio.uio_iovcnt = nr_segs;
 257         uio.uio_loffset = *ppos;
 258         uio.uio_limit = MAXOFFSET_T;
 259         uio.uio_segflg = segment;
 260
 261         cookie = spl_fstrans_mark();
 262         error = -zfs_read(ip, &uio, flags, cr);
 263         spl_fstrans_unmark(cookie);
 264         if (error < 0)
 265                 return (error);
 266
 267         read = count - uio.uio_resid;
 268         *ppos += read;
 269
 270         return (read);
 271 }
 272
 273 inline ssize_t
 274 zpl_read_common(struct inode *ip, const char *buf, size_t len, loff_t *ppos,
 275     uio_seg_t segment, int flags, cred_t *cr)
 276 {
 277         struct iovec iov;
 278
 279         iov.iov_base = (void *)buf;
 280         iov.iov_len = len;
 281
 282         return (zpl_read_common_iovec(ip, &iov, len, 1, ppos, segment,
 283             flags, cr, 0));
 284 }
 285
 286 static ssize_t
 287 zpl_iter_read_common(struct kiocb *kiocb, const struct iovec *iovp,
 288     unsigned long nr_segs, size_t count, uio_seg_t seg, size_t skip)
 289 {
 290         cred_t *cr = CRED();
 291         struct file *filp = kiocb->ki_filp;
 292         ssize_t read;
 293         unsigned int f_flags = filp->f_flags;
 294
 295         f_flags |= zfs_io_flags(kiocb);
 296         crhold(cr);
 297         read = zpl_read_common_iovec(filp->f_mapping->host, iovp, count,
 298             nr_segs, &kiocb->ki_pos, seg, f_flags, cr, skip);
 299         crfree(cr);
 300
 301         file_accessed(filp);
 302         return (read);
 303 }
 304
 305 #if defined(HAVE_VFS_RW_ITERATE)
 306 static ssize_t
 307 zpl_iter_read(struct kiocb *kiocb, struct iov_iter *to)
 308 {
 309         ssize_t ret;
 310         uio_seg_t seg = UIO_USERSPACE;
 311         if (to->type & ITER_KVEC)
 312                 seg = UIO_SYSSPACE;
 313         if (to->type & ITER_BVEC)
 314                 seg = UIO_BVEC;
 315         ret = zpl_iter_read_common(kiocb, to->iov, to->nr_segs,
 316             iov_iter_count(to), seg, to->iov_offset);
 317         if (ret > 0)
 318                 iov_iter_advance(to, ret);
 319         return (ret);
 320 }
 321 #else
 322 static ssize_t
 323 zpl_aio_read(struct kiocb *kiocb, const struct iovec *iovp,
 324     unsigned long nr_segs, loff_t pos)
 325 {
 326         ssize_t ret;
 327         size_t count;
 328
 329         ret = generic_segment_checks(iovp, &nr_segs, &count, VERIFY_WRITE);
 330         if (ret)
 331                 return (ret);
 332
 333         return (zpl_iter_read_common(kiocb, iovp, nr_segs, count,
 334             UIO_USERSPACE, 0));
 335 }
 336 #endif /* HAVE_VFS_RW_ITERATE */
 337
 338 static ssize_t
 339 zpl_write_common_iovec(struct inode *ip, const struct iovec *iovp, size_t count,
 340     unsigned long nr_segs, loff_t *ppos, uio_seg_t segment, int flags,
 341     cred_t *cr, size_t skip)
 342 {
 343         ssize_t wrote;
 344         uio_t uio;
 345         int error;
 346         fstrans_cookie_t cookie;
 347
 348         if (flags & O_APPEND)
 349                 *ppos = i_size_read(ip);
 350
 351         uio.uio_iov = iovp;
 352         uio.uio_skip = skip;
 353         uio.uio_resid = count;
 354         uio.uio_iovcnt = nr_segs;
 355         uio.uio_loffset = *ppos;
 356         uio.uio_limit = MAXOFFSET_T;
 357         uio.uio_segflg = segment;
 358
 359         cookie = spl_fstrans_mark();
 360         error = -zfs_write(ip, &uio, flags, cr);
 361         spl_fstrans_unmark(cookie);
 362         if (error < 0)
 363                 return (error);
 364
 365         wrote = count - uio.uio_resid;
 366         *ppos += wrote;
 367
 368         return (wrote);
 369 }
 370
 371 inline ssize_t
 372 zpl_write_common(struct inode *ip, const char *buf, size_t len, loff_t *ppos,
 373     uio_seg_t segment, int flags, cred_t *cr)
 374 {
 375         struct iovec iov;
 376
 377         iov.iov_base = (void *)buf;
 378         iov.iov_len = len;
 379
 380         return (zpl_write_common_iovec(ip, &iov, len, 1, ppos, segment,
 381             flags, cr, 0));
 382 }
 383
 384 static ssize_t
 385 zpl_iter_write_common(struct kiocb *kiocb, const struct iovec *iovp,
 386     unsigned long nr_segs, size_t count, uio_seg_t seg, size_t skip)
 387 {
 388         cred_t *cr = CRED();
 389         struct file *filp = kiocb->ki_filp;
 390         ssize_t wrote;
 391         unsigned int f_flags = filp->f_flags;
 392
 393         f_flags |= zfs_io_flags(kiocb);
 394         crhold(cr);
 395         wrote = zpl_write_common_iovec(filp->f_mapping->host, iovp, count,
 396             nr_segs, &kiocb->ki_pos, seg, f_flags, cr, skip);
 397         crfree(cr);
 398
 399         return (wrote);
 400 }
 401
 402 #if defined(HAVE_VFS_RW_ITERATE)
 403 static ssize_t
 404 zpl_iter_write(struct kiocb *kiocb, struct iov_iter *from)
 405 {
 406         size_t count;
 407         ssize_t ret;
 408         uio_seg_t seg = UIO_USERSPACE;
 409
 410 #ifndef HAVE_GENERIC_WRITE_CHECKS_KIOCB
 411         struct file *file = kiocb->ki_filp;
 412         struct address_space *mapping = file->f_mapping;
 413         struct inode *ip = mapping->host;
 414         int isblk = S_ISBLK(ip->i_mode);
 415
 416         count = iov_iter_count(from);
 417         ret = generic_write_checks(file, &kiocb->ki_pos, &count, isblk);
 418         if (ret)
 419                 return (ret);
 420 #else
 421         /*
 422          * XXX - ideally this check should be in the same lock region with
 423          * write operations, so that there's no TOCTTOU race when doing
 424          * append and someone else grow the file.
 425          */
 426         ret = generic_write_checks(kiocb, from);
 427         if (ret <= 0)
 428                 return (ret);
 429         count = ret;
 430 #endif
 431
 432         if (from->type & ITER_KVEC)
 433                 seg = UIO_SYSSPACE;
 434         if (from->type & ITER_BVEC)
 435                 seg = UIO_BVEC;
 436
 437         ret = zpl_iter_write_common(kiocb, from->iov, from->nr_segs,
 438             count, seg, from->iov_offset);
 439         if (ret > 0)
 440                 iov_iter_advance(from, ret);
 441
 442         return (ret);
 443 }
 444 #else
 445 static ssize_t
 446 zpl_aio_write(struct kiocb *kiocb, const struct iovec *iovp,
 447     unsigned long nr_segs, loff_t pos)
 448 {
 449         struct file *file = kiocb->ki_filp;
 450         struct address_space *mapping = file->f_mapping;
 451         struct inode *ip = mapping->host;
 452         int isblk = S_ISBLK(ip->i_mode);
 453         size_t count;
 454         ssize_t ret;
 455
 456         ret = generic_segment_checks(iovp, &nr_segs, &count, VERIFY_READ);
 457         if (ret)
 458                 return (ret);
 459
 460         ret = generic_write_checks(file, &pos, &count, isblk);
 461         if (ret)
 462                 return (ret);
 463
 464         return (zpl_iter_write_common(kiocb, iovp, nr_segs, count,
 465             UIO_USERSPACE, 0));
 466 }
 467 #endif /* HAVE_VFS_RW_ITERATE */
 468
 469 #if defined(HAVE_VFS_RW_ITERATE)
 470 static ssize_t
 471 zpl_direct_IO_impl(int rw, struct kiocb *kiocb, struct iov_iter *iter)
 472 {
 473         if (rw == WRITE)
 474                 return (zpl_iter_write(kiocb, iter));
 475         else
 476                 return (zpl_iter_read(kiocb, iter));
 477 }
 478 #if defined(HAVE_VFS_DIRECT_IO_ITER)
 479 static ssize_t
 480 zpl_direct_IO(struct kiocb *kiocb, struct iov_iter *iter)
 481 {
 482         return (zpl_direct_IO_impl(iov_iter_rw(iter), kiocb, iter));
 483 }
 484 #elif defined(HAVE_VFS_DIRECT_IO_ITER_OFFSET)
 485 static ssize_t
 486 zpl_direct_IO(struct kiocb *kiocb, struct iov_iter *iter, loff_t pos)
 487 {
 488         ASSERT3S(pos, ==, kiocb->ki_pos);
 489         return (zpl_direct_IO_impl(iov_iter_rw(iter), kiocb, iter));
 490 }
 491 #elif defined(HAVE_VFS_DIRECT_IO_ITER_RW_OFFSET)
 492 static ssize_t
 493 zpl_direct_IO(int rw, struct kiocb *kiocb, struct iov_iter *iter, loff_t pos)
 494 {
 495         ASSERT3S(pos, ==, kiocb->ki_pos);
 496         return (zpl_direct_IO_impl(rw, kiocb, iter));
 497 }
 498 #else
 499 #error "Unknown direct IO interface"
 500 #endif
 501
 502 #else
 503
 504 #if defined(HAVE_VFS_DIRECT_IO_IOVEC)
 505 static ssize_t
 506 zpl_direct_IO(int rw, struct kiocb *kiocb, const struct iovec *iovp,
 507     loff_t pos, unsigned long nr_segs)
 508 {
 509         if (rw == WRITE)
 510                 return (zpl_aio_write(kiocb, iovp, nr_segs, pos));
 511         else
 512                 return (zpl_aio_read(kiocb, iovp, nr_segs, pos));
 513 }
 514 #else
 515 #error "Unknown direct IO interface"
 516 #endif
 517
 518 #endif /* HAVE_VFS_RW_ITERATE */
 519
 520 static loff_t
 521 zpl_llseek(struct file *filp, loff_t offset, int whence)
 522 {
 523 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
 524         fstrans_cookie_t cookie;
 525
 526         if (whence == SEEK_DATA || whence == SEEK_HOLE) {
 527                 struct inode *ip = filp->f_mapping->host;
 528                 loff_t maxbytes = ip->i_sb->s_maxbytes;
 529                 loff_t error;
 530
 531                 spl_inode_lock_shared(ip);
 532                 cookie = spl_fstrans_mark();
 533                 error = -zfs_holey(ip, whence, &offset);
 534                 spl_fstrans_unmark(cookie);
 535                 if (error == 0)
 536                         error = lseek_execute(filp, ip, offset, maxbytes);
 537                 spl_inode_unlock_shared(ip);
 538
 539                 return (error);
 540         }
 541 #endif /* SEEK_HOLE && SEEK_DATA */
 542
 543         return (generic_file_llseek(filp, offset, whence));
 544 }
 545
 546 /*
 547  * It's worth taking a moment to describe how mmap is implemented
 548  * for zfs because it differs considerably from other Linux filesystems.
 549  * However, this issue is handled the same way under OpenSolaris.
 550  *
 551  * The issue is that by design zfs bypasses the Linux page cache and
 552  * leaves all caching up to the ARC.  This has been shown to work
 553  * well for the common read(2)/write(2) case.  However, mmap(2)
 554  * is problem because it relies on being tightly integrated with the
 555  * page cache.  To handle this we cache mmap'ed files twice, once in
 556  * the ARC and a second time in the page cache.  The code is careful
 557  * to keep both copies synchronized.
 558  *
 559  * When a file with an mmap'ed region is written to using write(2)
 560  * both the data in the ARC and existing pages in the page cache
 561  * are updated.  For a read(2) data will be read first from the page
 562  * cache then the ARC if needed.  Neither a write(2) or read(2) will
 563  * will ever result in new pages being added to the page cache.
 564  *
 565  * New pages are added to the page cache only via .readpage() which
 566  * is called when the vfs needs to read a page off disk to back the
 567  * virtual memory region.  These pages may be modified without
 568  * notifying the ARC and will be written out periodically via
 569  * .writepage().  This will occur due to either a sync or the usual
 570  * page aging behavior.  Note because a read(2) of a mmap'ed file
 571  * will always check the page cache first even when the ARC is out
 572  * of date correct data will still be returned.
 573  *
 574  * While this implementation ensures correct behavior it does have
 575  * have some drawbacks.  The most obvious of which is that it
 576  * increases the required memory footprint when access mmap'ed
 577  * files.  It also adds additional complexity to the code keeping
 578  * both caches synchronized.
 579  *
 580  * Longer term it may be possible to cleanly resolve this wart by
 581  * mapping page cache pages directly on to the ARC buffers.  The
 582  * Linux address space operations are flexible enough to allow
 583  * selection of which pages back a particular index.  The trick
 584  * would be working out the details of which subsystem is in
 585  * charge, the ARC, the page cache, or both.  It may also prove
 586  * helpful to move the ARC buffers to a scatter-gather lists
 587  * rather than a vmalloc'ed region.
 588  */
 589 static int
 590 zpl_mmap(struct file *filp, struct vm_area_struct *vma)
 591 {
 592         struct inode *ip = filp->f_mapping->host;
 593         znode_t *zp = ITOZ(ip);
 594         int error;
 595         fstrans_cookie_t cookie;
 596
 597         cookie = spl_fstrans_mark();
 598         error = -zfs_map(ip, vma->vm_pgoff, (caddr_t *)vma->vm_start,
 599             (size_t)(vma->vm_end - vma->vm_start), vma->vm_flags);
 600         spl_fstrans_unmark(cookie);
 601         if (error)
 602                 return (error);
 603
 604         error = generic_file_mmap(filp, vma);
 605         if (error)
 606                 return (error);
 607
 608         mutex_enter(&zp->z_lock);
 609         zp->z_is_mapped = B_TRUE;
 610         mutex_exit(&zp->z_lock);
 611
 612         return (error);
 613 }
 614
 615 /*
 616  * Populate a page with data for the Linux page cache.  This function is
 617  * only used to support mmap(2).  There will be an identical copy of the
 618  * data in the ARC which is kept up to date via .write() and .writepage().
 619  *
 620  * Current this function relies on zpl_read_common() and the O_DIRECT
 621  * flag to read in a page.  This works but the more correct way is to
 622  * update zfs_fillpage() to be Linux friendly and use that interface.
 623  */
 624 static int
 625 zpl_readpage(struct file *filp, struct page *pp)
 626 {
 627         struct inode *ip;
 628         struct page *pl[1];
 629         int error = 0;
 630         fstrans_cookie_t cookie;
 631
 632         ASSERT(PageLocked(pp));
 633         ip = pp->mapping->host;
 634         pl[0] = pp;
 635
 636         cookie = spl_fstrans_mark();
 637         error = -zfs_getpage(ip, pl, 1);
 638         spl_fstrans_unmark(cookie);
 639
 640         if (error) {
 641                 SetPageError(pp);
 642                 ClearPageUptodate(pp);
 643         } else {
 644                 ClearPageError(pp);
 645                 SetPageUptodate(pp);
 646                 flush_dcache_page(pp);
 647         }
 648
 649         unlock_page(pp);
 650         return (error);
 651 }
 652
 653 /*
 654  * Populate a set of pages with data for the Linux page cache.  This
 655  * function will only be called for read ahead and never for demand
 656  * paging.  For simplicity, the code relies on read_cache_pages() to
 657  * correctly lock each page for IO and call zpl_readpage().
 658  */
 659 static int
 660 zpl_readpages(struct file *filp, struct address_space *mapping,
 661     struct list_head *pages, unsigned nr_pages)
 662 {
 663         return (read_cache_pages(mapping, pages,
 664             (filler_t *)zpl_readpage, filp));
 665 }
 666
 667 int
 668 zpl_putpage(struct page *pp, struct writeback_control *wbc, void *data)
 669 {
 670         struct address_space *mapping = data;
 671         fstrans_cookie_t cookie;
 672
 673         ASSERT(PageLocked(pp));
 674         ASSERT(!PageWriteback(pp));
 675
 676         cookie = spl_fstrans_mark();
 677         (void) zfs_putpage(mapping->host, pp, wbc);
 678         spl_fstrans_unmark(cookie);
 679
 680         return (0);
 681 }
 682
 683 static int
 684 zpl_writepages(struct address_space *mapping, struct writeback_control *wbc)
 685 {
 686         znode_t         *zp = ITOZ(mapping->host);
 687         zfsvfs_t        *zfsvfs = ITOZSB(mapping->host);
 688         enum writeback_sync_modes sync_mode;
 689         int result;
 690
 691         ZFS_ENTER(zfsvfs);
 692         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
 693                 wbc->sync_mode = WB_SYNC_ALL;
 694         ZFS_EXIT(zfsvfs);
 695         sync_mode = wbc->sync_mode;
 696
 697         /*
 698          * We don't want to run write_cache_pages() in SYNC mode here, because
 699          * that would make putpage() wait for a single page to be committed to
 700          * disk every single time, resulting in atrocious performance. Instead
 701          * we run it once in non-SYNC mode so that the ZIL gets all the data,
 702          * and then we commit it all in one go.
 703          */
 704         wbc->sync_mode = WB_SYNC_NONE;
 705         result = write_cache_pages(mapping, wbc, zpl_putpage, mapping);
 706         if (sync_mode != wbc->sync_mode) {
 707                 ZFS_ENTER(zfsvfs);
 708                 ZFS_VERIFY_ZP(zp);
 709                 if (zfsvfs->z_log != NULL)
 710                         zil_commit(zfsvfs->z_log, zp->z_id);
 711                 ZFS_EXIT(zfsvfs);
 712
 713                 /*
 714                  * We need to call write_cache_pages() again (we can't just
 715                  * return after the commit) because the previous call in
 716                  * non-SYNC mode does not guarantee that we got all the dirty
 717                  * pages (see the implementation of write_cache_pages() for
 718                  * details). That being said, this is a no-op in most cases.
 719                  */
 720                 wbc->sync_mode = sync_mode;
 721                 result = write_cache_pages(mapping, wbc, zpl_putpage, mapping);
 722         }
 723         return (result);
 724 }
 725
 726 /*
 727  * Write out dirty pages to the ARC, this function is only required to
 728  * support mmap(2).  Mapped pages may be dirtied by memory operations
 729  * which never call .write().  These dirty pages are kept in sync with
 730  * the ARC buffers via this hook.
 731  */
 732 static int
 733 zpl_writepage(struct page *pp, struct writeback_control *wbc)
 734 {
 735         if (ITOZSB(pp->mapping->host)->z_os->os_sync == ZFS_SYNC_ALWAYS)
 736                 wbc->sync_mode = WB_SYNC_ALL;
 737
 738         return (zpl_putpage(pp, wbc, pp->mapping));
 739 }
 740
 741 /*
 742  * The only flag combination which matches the behavior of zfs_space()
 743  * is FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE.  The FALLOC_FL_PUNCH_HOLE
 744  * flag was introduced in the 2.6.38 kernel.
 745  */
 746 #if defined(HAVE_FILE_FALLOCATE) || defined(HAVE_INODE_FALLOCATE)
 747 long
 748 zpl_fallocate_common(struct inode *ip, int mode, loff_t offset, loff_t len)
 749 {
 750         int error = -EOPNOTSUPP;
 751
 752 #if defined(FALLOC_FL_PUNCH_HOLE) && defined(FALLOC_FL_KEEP_SIZE)
 753         cred_t *cr = CRED();
 754         flock64_t bf;
 755         loff_t olen;
 756         fstrans_cookie_t cookie;
 757
 758         if (mode != (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
 759                 return (error);
 760
 761         if (offset < 0 || len <= 0)
 762                 return (-EINVAL);
 763
 764         spl_inode_lock(ip);
 765         olen = i_size_read(ip);
 766
 767         if (offset > olen) {
 768                 spl_inode_unlock(ip);
 769                 return (0);
 770         }
 771         if (offset + len > olen)
 772                 len = olen - offset;
 773         bf.l_type = F_WRLCK;
 774         bf.l_whence = SEEK_SET;
 775         bf.l_start = offset;
 776         bf.l_len = len;
 777         bf.l_pid = 0;
 778
 779         crhold(cr);
 780         cookie = spl_fstrans_mark();
 781         error = -zfs_space(ip, F_FREESP, &bf, FWRITE, offset, cr);
 782         spl_fstrans_unmark(cookie);
 783         spl_inode_unlock(ip);
 784
 785         crfree(cr);
 786 #endif /* defined(FALLOC_FL_PUNCH_HOLE) && defined(FALLOC_FL_KEEP_SIZE) */
 787
 788         ASSERT3S(error, <=, 0);
 789         return (error);
 790 }
 791 #endif /* defined(HAVE_FILE_FALLOCATE) || defined(HAVE_INODE_FALLOCATE) */
 792
 793 #ifdef HAVE_FILE_FALLOCATE
 794 static long
 795 zpl_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
 796 {
 797         return zpl_fallocate_common(file_inode(filp),
 798             mode, offset, len);
 799 }
 800 #endif /* HAVE_FILE_FALLOCATE */
 801
 802 #define ZFS_FL_USER_VISIBLE     (FS_FL_USER_VISIBLE | ZFS_PROJINHERIT_FL)
 803 #define ZFS_FL_USER_MODIFIABLE  (FS_FL_USER_MODIFIABLE | ZFS_PROJINHERIT_FL)
 804
 805 static uint32_t
 806 __zpl_ioctl_getflags(struct inode *ip)
 807 {
 808         uint64_t zfs_flags = ITOZ(ip)->z_pflags;
 809         uint32_t ioctl_flags = 0;
 810
 811         if (zfs_flags & ZFS_IMMUTABLE)
 812                 ioctl_flags |= FS_IMMUTABLE_FL;
 813
 814         if (zfs_flags & ZFS_APPENDONLY)
 815                 ioctl_flags |= FS_APPEND_FL;
 816
 817         if (zfs_flags & ZFS_NODUMP)
 818                 ioctl_flags |= FS_NODUMP_FL;
 819
 820         if (zfs_flags & ZFS_PROJINHERIT)
 821                 ioctl_flags |= ZFS_PROJINHERIT_FL;
 822
 823         return (ioctl_flags & ZFS_FL_USER_VISIBLE);
 824 }
 825
 826 /*
 827  * Map zfs file z_pflags (xvattr_t) to linux file attributes. Only file
 828  * attributes common to both Linux and Solaris are mapped.
 829  */
 830 static int
 831 zpl_ioctl_getflags(struct file *filp, void __user *arg)
 832 {
 833         uint32_t flags;
 834         int err;
 835
 836         flags = __zpl_ioctl_getflags(file_inode(filp));
 837         err = copy_to_user(arg, &flags, sizeof (flags));
 838
 839         return (err);
 840 }
 841
 842 /*
 843  * fchange() is a helper macro to detect if we have been asked to change a
 844  * flag. This is ugly, but the requirement that we do this is a consequence of
 845  * how the Linux file attribute interface was designed. Another consequence is
 846  * that concurrent modification of files suffers from a TOCTOU race. Neither
 847  * are things we can fix without modifying the kernel-userland interface, which
 848  * is outside of our jurisdiction.
 849  */
 850
 851 #define fchange(f0, f1, b0, b1) (!((f0) & (b0)) != !((f1) & (b1)))
 852
 853 static int
 854 __zpl_ioctl_setflags(struct inode *ip, uint32_t ioctl_flags, xvattr_t *xva)
 855 {
 856         uint64_t zfs_flags = ITOZ(ip)->z_pflags;
 857         xoptattr_t *xoap;
 858
 859         if (ioctl_flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NODUMP_FL |
 860             ZFS_PROJINHERIT_FL))
 861                 return (-EOPNOTSUPP);
 862
 863         if (ioctl_flags & ~ZFS_FL_USER_MODIFIABLE)
 864                 return (-EACCES);
 865
 866         if ((fchange(ioctl_flags, zfs_flags, FS_IMMUTABLE_FL, ZFS_IMMUTABLE) ||
 867             fchange(ioctl_flags, zfs_flags, FS_APPEND_FL, ZFS_APPENDONLY)) &&
 868             !capable(CAP_LINUX_IMMUTABLE))
 869                 return (-EACCES);
 870
 871         if (!zpl_inode_owner_or_capable(ip))
 872                 return (-EACCES);
 873
 874         xva_init(xva);
 875         xoap = xva_getxoptattr(xva);
 876
 877         XVA_SET_REQ(xva, XAT_IMMUTABLE);
 878         if (ioctl_flags & FS_IMMUTABLE_FL)
 879                 xoap->xoa_immutable = B_TRUE;
 880
 881         XVA_SET_REQ(xva, XAT_APPENDONLY);
 882         if (ioctl_flags & FS_APPEND_FL)
 883                 xoap->xoa_appendonly = B_TRUE;
 884
 885         XVA_SET_REQ(xva, XAT_NODUMP);
 886         if (ioctl_flags & FS_NODUMP_FL)
 887                 xoap->xoa_nodump = B_TRUE;
 888
 889         XVA_SET_REQ(xva, XAT_PROJINHERIT);
 890         if (ioctl_flags & ZFS_PROJINHERIT_FL)
 891                 xoap->xoa_projinherit = B_TRUE;
 892
 893         return (0);
 894 }
 895
 896 static int
 897 zpl_ioctl_setflags(struct file *filp, void __user *arg)
 898 {
 899         struct inode *ip = file_inode(filp);
 900         uint32_t flags;
 901         cred_t *cr = CRED();
 902         xvattr_t xva;
 903         int err;
 904         fstrans_cookie_t cookie;
 905
 906         if (copy_from_user(&flags, arg, sizeof (flags)))
 907                 return (-EFAULT);
 908
 909         err = __zpl_ioctl_setflags(ip, flags, &xva);
 910         if (err)
 911                 return (err);
 912
 913         crhold(cr);
 914         cookie = spl_fstrans_mark();
 915         err = -zfs_setattr(ip, (vattr_t *)&xva, 0, cr);
 916         spl_fstrans_unmark(cookie);
 917         crfree(cr);
 918
 919         return (err);
 920 }
 921
 922 static int
 923 zpl_ioctl_getxattr(struct file *filp, void __user *arg)
 924 {
 925         zfsxattr_t fsx = { 0 };
 926         struct inode *ip = file_inode(filp);
 927         int err;
 928
 929         fsx.fsx_xflags = __zpl_ioctl_getflags(ip);
 930         fsx.fsx_projid = ITOZ(ip)->z_projid;
 931         err = copy_to_user(arg, &fsx, sizeof (fsx));
 932
 933         return (err);
 934 }
 935
 936 static int
 937 zpl_ioctl_setxattr(struct file *filp, void __user *arg)
 938 {
 939         struct inode *ip = file_inode(filp);
 940         zfsxattr_t fsx;
 941         cred_t *cr = CRED();
 942         xvattr_t xva;
 943         xoptattr_t *xoap;
 944         int err;
 945         fstrans_cookie_t cookie;
 946
 947         if (copy_from_user(&fsx, arg, sizeof (fsx)))
 948                 return (-EFAULT);
 949
 950         if (!zpl_is_valid_projid(fsx.fsx_projid))
 951                 return (-EINVAL);
 952
 953         err = __zpl_ioctl_setflags(ip, fsx.fsx_xflags, &xva);
 954         if (err)
 955                 return (err);
 956
 957         xoap = xva_getxoptattr(&xva);
 958         XVA_SET_REQ(&xva, XAT_PROJID);
 959         xoap->xoa_projid = fsx.fsx_projid;
 960
 961         crhold(cr);
 962         cookie = spl_fstrans_mark();
 963         err = -zfs_setattr(ip, (vattr_t *)&xva, 0, cr);
 964         spl_fstrans_unmark(cookie);
 965         crfree(cr);
 966
 967         return (err);
 968 }
 969
 970 static long
 971 zpl_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 972 {
 973         switch (cmd) {
 974         case FS_IOC_GETFLAGS:
 975                 return (zpl_ioctl_getflags(filp, (void *)arg));
 976         case FS_IOC_SETFLAGS:
 977                 return (zpl_ioctl_setflags(filp, (void *)arg));
 978         case ZFS_IOC_FSGETXATTR:
 979                 return (zpl_ioctl_getxattr(filp, (void *)arg));
 980         case ZFS_IOC_FSSETXATTR:
 981                 return (zpl_ioctl_setxattr(filp, (void *)arg));
 982         default:
 983                 return (-ENOTTY);
 984         }
 985 }
 986
 987 #ifdef CONFIG_COMPAT
 988 static long
 989 zpl_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 990 {
 991         switch (cmd) {
 992         case FS_IOC32_GETFLAGS:
 993                 cmd = FS_IOC_GETFLAGS;
 994                 break;
 995         case FS_IOC32_SETFLAGS:
 996                 cmd = FS_IOC_SETFLAGS;
 997                 break;
 998         default:
 999                 return (-ENOTTY);
1000         }
1001         return (zpl_ioctl(filp, cmd, (unsigned long)compat_ptr(arg)));
1002 }
1003 #endif /* CONFIG_COMPAT */
1004
1005
1006 const struct address_space_operations zpl_address_space_operations = {
1007         .readpages      = zpl_readpages,
1008         .readpage       = zpl_readpage,
1009         .writepage      = zpl_writepage,
1010         .writepages     = zpl_writepages,
1011         .direct_IO      = zpl_direct_IO,
1012 };
1013
1014 const struct file_operations zpl_file_operations = {
1015         .open           = zpl_open,
1016         .release        = zpl_release,
1017         .llseek         = zpl_llseek,
1018 #ifdef HAVE_VFS_RW_ITERATE
1019 #ifdef HAVE_NEW_SYNC_READ
1020         .read           = new_sync_read,
1021         .write          = new_sync_write,
1022 #endif
1023         .read_iter      = zpl_iter_read,
1024         .write_iter     = zpl_iter_write,
1025 #else
1026         .read           = do_sync_read,
1027         .write          = do_sync_write,
1028         .aio_read       = zpl_aio_read,
1029         .aio_write      = zpl_aio_write,
1030 #endif
1031         .mmap           = zpl_mmap,
1032         .fsync          = zpl_fsync,
1033 #ifdef HAVE_FILE_AIO_FSYNC
1034         .aio_fsync      = zpl_aio_fsync,
1035 #endif
1036 #ifdef HAVE_FILE_FALLOCATE
1037         .fallocate      = zpl_fallocate,
1038 #endif /* HAVE_FILE_FALLOCATE */
1039         .unlocked_ioctl = zpl_ioctl,
1040 #ifdef CONFIG_COMPAT
1041         .compat_ioctl   = zpl_compat_ioctl,
1042 #endif
1043 };
1044
1045 const struct file_operations zpl_dir_file_operations = {
1046         .llseek         = generic_file_llseek,
1047         .read           = generic_read_dir,
1048 #if defined(HAVE_VFS_ITERATE_SHARED)
1049         .iterate_shared = zpl_iterate,
1050 #elif defined(HAVE_VFS_ITERATE)
1051         .iterate        = zpl_iterate,
1052 #else
1053         .readdir        = zpl_readdir,
1054 #endif
1055         .fsync          = zpl_fsync,
1056         .unlocked_ioctl = zpl_ioctl,
1057 #ifdef CONFIG_COMPAT
1058         .compat_ioctl   = zpl_compat_ioctl,
1059 #endif
1060 };