Linux Audio

Check our new training course

Loading...
v4.17
   1/*
   2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3 * All Rights Reserved.
   4 *
   5 * This program is free software; you can redistribute it and/or
   6 * modify it under the terms of the GNU General Public License as
   7 * published by the Free Software Foundation.
   8 *
   9 * This program is distributed in the hope that it would be useful,
  10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 * GNU General Public License for more details.
  13 *
  14 * You should have received a copy of the GNU General Public License
  15 * along with this program; if not, write the Free Software Foundation,
  16 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17 */
  18#include "xfs.h"
  19#include "xfs_fs.h"
  20#include "xfs_shared.h"
  21#include "xfs_format.h"
  22#include "xfs_log_format.h"
  23#include "xfs_trans_resv.h"
 
 
  24#include "xfs_mount.h"
  25#include "xfs_da_format.h"
  26#include "xfs_da_btree.h"
 
  27#include "xfs_inode.h"
  28#include "xfs_trans.h"
  29#include "xfs_inode_item.h"
  30#include "xfs_bmap.h"
  31#include "xfs_bmap_util.h"
  32#include "xfs_error.h"
  33#include "xfs_dir2.h"
  34#include "xfs_dir2_priv.h"
  35#include "xfs_ioctl.h"
  36#include "xfs_trace.h"
  37#include "xfs_log.h"
  38#include "xfs_icache.h"
  39#include "xfs_pnfs.h"
  40#include "xfs_iomap.h"
  41#include "xfs_reflink.h"
  42
  43#include <linux/dcache.h>
  44#include <linux/falloc.h>
  45#include <linux/pagevec.h>
  46#include <linux/backing-dev.h>
  47#include <linux/mman.h>
  48
  49static const struct vm_operations_struct xfs_file_vm_ops;
  50
  51int
  52xfs_update_prealloc_flags(
 
 
 
 
  53	struct xfs_inode	*ip,
  54	enum xfs_prealloc_flags	flags)
  55{
  56	struct xfs_trans	*tp;
  57	int			error;
  58
  59	error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_writeid,
  60			0, 0, 0, &tp);
  61	if (error)
  62		return error;
  63
  64	xfs_ilock(ip, XFS_ILOCK_EXCL);
  65	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
  66
  67	if (!(flags & XFS_PREALLOC_INVISIBLE)) {
  68		VFS_I(ip)->i_mode &= ~S_ISUID;
  69		if (VFS_I(ip)->i_mode & S_IXGRP)
  70			VFS_I(ip)->i_mode &= ~S_ISGID;
  71		xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
  72	}
  73
  74	if (flags & XFS_PREALLOC_SET)
  75		ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
  76	if (flags & XFS_PREALLOC_CLEAR)
  77		ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
 
 
 
 
 
  78
  79	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
  80	if (flags & XFS_PREALLOC_SYNC)
  81		xfs_trans_set_sync(tp);
  82	return xfs_trans_commit(tp);
 
 
 
 
  83}
  84
  85/*
  86 * Fsync operations on directories are much simpler than on regular files,
  87 * as there is no file data to flush, and thus also no need for explicit
  88 * cache flush operations, and there are no non-transaction metadata updates
  89 * on directories either.
 
 
 
 
  90 */
  91STATIC int
  92xfs_dir_fsync(
  93	struct file		*file,
  94	loff_t			start,
  95	loff_t			end,
  96	int			datasync)
  97{
  98	struct xfs_inode	*ip = XFS_I(file->f_mapping->host);
  99	struct xfs_mount	*mp = ip->i_mount;
 100	xfs_lsn_t		lsn = 0;
 101
 102	trace_xfs_dir_fsync(ip);
 103
 104	xfs_ilock(ip, XFS_ILOCK_SHARED);
 105	if (xfs_ipincount(ip))
 106		lsn = ip->i_itemp->ili_last_lsn;
 107	xfs_iunlock(ip, XFS_ILOCK_SHARED);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 108
 109	if (!lsn)
 110		return 0;
 111	return xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
 112}
 113
 114STATIC int
 115xfs_file_fsync(
 116	struct file		*file,
 117	loff_t			start,
 118	loff_t			end,
 119	int			datasync)
 120{
 121	struct inode		*inode = file->f_mapping->host;
 122	struct xfs_inode	*ip = XFS_I(inode);
 123	struct xfs_mount	*mp = ip->i_mount;
 
 124	int			error = 0;
 125	int			log_flushed = 0;
 126	xfs_lsn_t		lsn = 0;
 127
 128	trace_xfs_file_fsync(ip);
 129
 130	error = file_write_and_wait_range(file, start, end);
 131	if (error)
 132		return error;
 133
 134	if (XFS_FORCED_SHUTDOWN(mp))
 135		return -EIO;
 136
 137	xfs_iflags_clear(ip, XFS_ITRUNCATED);
 138
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 139	/*
 140	 * If we have an RT and/or log subvolume we need to make sure to flush
 141	 * the write cache the device used for file data first.  This is to
 142	 * ensure newly written file data make it to disk before logging the new
 143	 * inode size in case of an extending write.
 
 
 
 
 
 
 144	 */
 145	if (XFS_IS_REALTIME_INODE(ip))
 146		xfs_blkdev_issue_flush(mp->m_rtdev_targp);
 147	else if (mp->m_logdev_targp != mp->m_ddev_targp)
 148		xfs_blkdev_issue_flush(mp->m_ddev_targp);
 149
 150	/*
 151	 * All metadata updates are logged, which means that we just have to
 152	 * flush the log up to the latest LSN that touched the inode. If we have
 153	 * concurrent fsync/fdatasync() calls, we need them to all block on the
 154	 * log force before we clear the ili_fsync_fields field. This ensures
 155	 * that we don't get a racing sync operation that does not wait for the
 156	 * metadata to hit the journal before returning. If we race with
 157	 * clearing the ili_fsync_fields, then all that will happen is the log
 158	 * force will do nothing as the lsn will already be on disk. We can't
 159	 * race with setting ili_fsync_fields because that is done under
 160	 * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared
 161	 * until after the ili_fsync_fields is cleared.
 162	 */
 163	xfs_ilock(ip, XFS_ILOCK_SHARED);
 164	if (xfs_ipincount(ip)) {
 165		if (!datasync ||
 166		    (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
 167			lsn = ip->i_itemp->ili_last_lsn;
 168	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 169
 170	if (lsn) {
 171		error = xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
 172		ip->i_itemp->ili_fsync_fields = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 173	}
 174	xfs_iunlock(ip, XFS_ILOCK_SHARED);
 175
 176	/*
 177	 * If we only have a single device, and the log force about was
 178	 * a no-op we might have to flush the data device cache here.
 179	 * This can only happen for fdatasync/O_DSYNC if we were overwriting
 180	 * an already allocated file and thus do not have any metadata to
 181	 * commit.
 182	 */
 183	if (!log_flushed && !XFS_IS_REALTIME_INODE(ip) &&
 184	    mp->m_logdev_targp == mp->m_ddev_targp)
 
 
 185		xfs_blkdev_issue_flush(mp->m_ddev_targp);
 186
 187	return error;
 188}
 189
 190STATIC ssize_t
 191xfs_file_dio_aio_read(
 192	struct kiocb		*iocb,
 193	struct iov_iter		*to)
 
 
 194{
 195	struct xfs_inode	*ip = XFS_I(file_inode(iocb->ki_filp));
 196	size_t			count = iov_iter_count(to);
 197	ssize_t			ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 198
 199	trace_xfs_file_direct_read(ip, count, iocb->ki_pos);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 200
 201	if (!count)
 202		return 0; /* skip atime */
 
 203
 204	file_accessed(iocb->ki_filp);
 
 205
 206	xfs_ilock(ip, XFS_IOLOCK_SHARED);
 207	ret = iomap_dio_rw(iocb, to, &xfs_iomap_ops, NULL);
 208	xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 209
 
 
 
 
 
 
 
 210	return ret;
 211}
 212
 213static noinline ssize_t
 214xfs_file_dax_read(
 215	struct kiocb		*iocb,
 216	struct iov_iter		*to)
 
 
 
 217{
 218	struct xfs_inode	*ip = XFS_I(iocb->ki_filp->f_mapping->host);
 219	size_t			count = iov_iter_count(to);
 220	ssize_t			ret = 0;
 221
 222	trace_xfs_file_dax_read(ip, count, iocb->ki_pos);
 223
 224	if (!count)
 225		return 0; /* skip atime */
 226
 227	if (iocb->ki_flags & IOCB_NOWAIT) {
 228		if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
 229			return -EAGAIN;
 230	} else {
 231		xfs_ilock(ip, XFS_IOLOCK_SHARED);
 232	}
 233
 234	ret = dax_iomap_rw(iocb, to, &xfs_iomap_ops);
 235	xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 
 
 
 
 
 236
 237	file_accessed(iocb->ki_filp);
 238	return ret;
 239}
 240
 241STATIC ssize_t
 242xfs_file_buffered_aio_read(
 243	struct kiocb		*iocb,
 244	struct iov_iter		*to)
 
 245{
 246	struct xfs_inode	*ip = XFS_I(file_inode(iocb->ki_filp));
 247	ssize_t			ret;
 248
 249	trace_xfs_file_buffered_read(ip, iov_iter_count(to), iocb->ki_pos);
 
 250
 251	if (iocb->ki_flags & IOCB_NOWAIT) {
 252		if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
 253			return -EAGAIN;
 254	} else {
 255		xfs_ilock(ip, XFS_IOLOCK_SHARED);
 
 
 
 
 256	}
 257	ret = generic_file_read_iter(iocb, to);
 258	xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 259
 260	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 261}
 262
 
 
 
 
 
 
 
 
 263STATIC ssize_t
 264xfs_file_read_iter(
 265	struct kiocb		*iocb,
 266	struct iov_iter		*to)
 
 
 
 267{
 268	struct inode		*inode = file_inode(iocb->ki_filp);
 269	struct xfs_mount	*mp = XFS_I(inode)->i_mount;
 270	ssize_t			ret = 0;
 
 
 271
 272	XFS_STATS_INC(mp, xs_read_calls);
 273
 274	if (XFS_FORCED_SHUTDOWN(mp))
 
 
 
 275		return -EIO;
 276
 277	if (IS_DAX(inode))
 278		ret = xfs_file_dax_read(iocb, to);
 279	else if (iocb->ki_flags & IOCB_DIRECT)
 280		ret = xfs_file_dio_aio_read(iocb, to);
 281	else
 282		ret = xfs_file_buffered_aio_read(iocb, to);
 283
 284	if (ret > 0)
 285		XFS_STATS_ADD(mp, xs_read_bytes, ret);
 
 
 
 
 
 
 
 
 
 
 
 
 286	return ret;
 287}
 288
 289/*
 290 * Common pre-write limit and setup checks.
 291 *
 292 * Called with the iolocked held either shared and exclusive according to
 293 * @iolock, and returns with it held.  Might upgrade the iolock to exclusive
 294 * if called for a direct write beyond i_size.
 295 */
 296STATIC ssize_t
 297xfs_file_aio_write_checks(
 298	struct kiocb		*iocb,
 299	struct iov_iter		*from,
 300	int			*iolock)
 301{
 302	struct file		*file = iocb->ki_filp;
 303	struct inode		*inode = file->f_mapping->host;
 304	struct xfs_inode	*ip = XFS_I(inode);
 305	ssize_t			error = 0;
 306	size_t			count = iov_iter_count(from);
 307	bool			drained_dio = false;
 308	loff_t			isize;
 309
 310restart:
 311	error = generic_write_checks(iocb, from);
 312	if (error <= 0)
 313		return error;
 314
 315	error = xfs_break_layouts(inode, iolock);
 316	if (error)
 317		return error;
 318
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 319	/*
 320	 * For changing security info in file_remove_privs() we need i_rwsem
 321	 * exclusively.
 322	 */
 323	if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
 324		xfs_iunlock(ip, *iolock);
 325		*iolock = XFS_IOLOCK_EXCL;
 326		xfs_ilock(ip, *iolock);
 327		goto restart;
 328	}
 329	/*
 330	 * If the offset is beyond the size of the file, we need to zero any
 331	 * blocks that fall between the existing EOF and the start of this
 332	 * write.  If zeroing is needed and we are currently holding the
 333	 * iolock shared, we need to update it to exclusive which implies
 334	 * having to redo all checks before.
 335	 *
 336	 * We need to serialise against EOF updates that occur in IO
 337	 * completions here. We want to make sure that nobody is changing the
 338	 * size while we do this check until we have placed an IO barrier (i.e.
 339	 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
 340	 * The spinlock effectively forms a memory barrier once we have the
 341	 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
 342	 * and hence be able to correctly determine if we need to run zeroing.
 343	 */
 344	spin_lock(&ip->i_flags_lock);
 345	isize = i_size_read(inode);
 346	if (iocb->ki_pos > isize) {
 347		spin_unlock(&ip->i_flags_lock);
 348		if (!drained_dio) {
 349			if (*iolock == XFS_IOLOCK_SHARED) {
 350				xfs_iunlock(ip, *iolock);
 351				*iolock = XFS_IOLOCK_EXCL;
 352				xfs_ilock(ip, *iolock);
 353				iov_iter_reexpand(from, count);
 354			}
 355			/*
 356			 * We now have an IO submission barrier in place, but
 357			 * AIO can do EOF updates during IO completion and hence
 358			 * we now need to wait for all of them to drain. Non-AIO
 359			 * DIO will have drained before we are given the
 360			 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
 361			 * no-op.
 362			 */
 363			inode_dio_wait(inode);
 364			drained_dio = true;
 365			goto restart;
 366		}
 367	
 368		trace_xfs_zero_eof(ip, isize, iocb->ki_pos - isize);
 369		error = iomap_zero_range(inode, isize, iocb->ki_pos - isize,
 370				NULL, &xfs_iomap_ops);
 371		if (error)
 372			return error;
 373	} else
 374		spin_unlock(&ip->i_flags_lock);
 
 
 
 
 
 
 
 
 
 
 
 375
 376	/*
 377	 * Updating the timestamps will grab the ilock again from
 378	 * xfs_fs_dirty_inode, so we have to call it after dropping the
 379	 * lock above.  Eventually we should look into a way to avoid
 380	 * the pointless lock roundtrip.
 381	 */
 382	if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
 383		error = file_update_time(file);
 384		if (error)
 385			return error;
 386	}
 387
 388	/*
 389	 * If we're writing the file then make sure to clear the setuid and
 390	 * setgid bits if the process is not being run by root.  This keeps
 391	 * people from modifying setuid and setgid binaries.
 
 
 
 392	 */
 393	if (!IS_NOSEC(inode))
 394		return file_remove_privs(file);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 395	return 0;
 
 
 
 
 
 396}
 397
 398static int
 399xfs_dio_write_end_io(
 400	struct kiocb		*iocb,
 401	ssize_t			size,
 402	unsigned		flags)
 
 
 
 
 
 
 403{
 404	struct inode		*inode = file_inode(iocb->ki_filp);
 405	struct xfs_inode	*ip = XFS_I(inode);
 406	loff_t			offset = iocb->ki_pos;
 407	int			error = 0;
 408
 409	trace_xfs_end_io_direct_write(ip, offset, size);
 410
 411	if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 412		return -EIO;
 
 
 413
 414	if (size <= 0)
 415		return size;
 
 416
 417	if (flags & IOMAP_DIO_COW) {
 418		error = xfs_reflink_end_cow(ip, offset, size);
 419		if (error)
 420			return error;
 421	}
 422
 423	/*
 424	 * Unwritten conversion updates the in-core isize after extent
 425	 * conversion but before updating the on-disk size. Updating isize any
 426	 * earlier allows a racing dio read to find unwritten extents before
 427	 * they are converted.
 428	 */
 429	if (flags & IOMAP_DIO_UNWRITTEN)
 430		return xfs_iomap_write_unwritten(ip, offset, size, true);
 
 
 
 
 431
 432	/*
 433	 * We need to update the in-core inode size here so that we don't end up
 434	 * with the on-disk inode size being outside the in-core inode size. We
 435	 * have no other method of updating EOF for AIO, so always do it here
 436	 * if necessary.
 437	 *
 438	 * We need to lock the test/set EOF update as we can be racing with
 439	 * other IO completions here to update the EOF. Failing to serialise
 440	 * here can result in EOF moving backwards and Bad Things Happen when
 441	 * that occurs.
 442	 */
 443	spin_lock(&ip->i_flags_lock);
 444	if (offset + size > i_size_read(inode)) {
 445		i_size_write(inode, offset + size);
 446		spin_unlock(&ip->i_flags_lock);
 447		error = xfs_setfilesize(ip, offset, size);
 448	} else {
 449		spin_unlock(&ip->i_flags_lock);
 450	}
 451
 452	return error;
 453}
 454
 455/*
 456 * xfs_file_dio_aio_write - handle direct IO writes
 457 *
 458 * Lock the inode appropriately to prepare for and issue a direct IO write.
 459 * By separating it from the buffered write path we remove all the tricky to
 460 * follow locking changes and looping.
 461 *
 462 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
 463 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
 464 * pages are flushed out.
 465 *
 466 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
 467 * allowing them to be done in parallel with reads and other direct IO writes.
 468 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
 469 * needs to do sub-block zeroing and that requires serialisation against other
 470 * direct IOs to the same block. In this case we need to serialise the
 471 * submission of the unaligned IOs so that we don't get racing block zeroing in
 472 * the dio layer.  To avoid the problem with aio, we also need to wait for
 473 * outstanding IOs to complete so that unwritten extent conversion is completed
 474 * before we try to map the overlapping block. This is currently implemented by
 475 * hitting it with a big hammer (i.e. inode_dio_wait()).
 476 *
 477 * Returns with locks held indicated by @iolock and errors indicated by
 478 * negative return values.
 479 */
 480STATIC ssize_t
 481xfs_file_dio_aio_write(
 482	struct kiocb		*iocb,
 483	struct iov_iter		*from)
 
 
 
 
 484{
 485	struct file		*file = iocb->ki_filp;
 486	struct address_space	*mapping = file->f_mapping;
 487	struct inode		*inode = mapping->host;
 488	struct xfs_inode	*ip = XFS_I(inode);
 489	struct xfs_mount	*mp = ip->i_mount;
 490	ssize_t			ret = 0;
 
 491	int			unaligned_io = 0;
 492	int			iolock;
 493	size_t			count = iov_iter_count(from);
 494	struct xfs_buftarg      *target = XFS_IS_REALTIME_INODE(ip) ?
 495					mp->m_rtdev_targp : mp->m_ddev_targp;
 496
 497	/* DIO must be aligned to device logical sector size */
 498	if ((iocb->ki_pos | count) & target->bt_logical_sectormask)
 499		return -EINVAL;
 500
 501	/*
 502	 * Don't take the exclusive iolock here unless the I/O is unaligned to
 503	 * the file system block size.  We don't need to consider the EOF
 504	 * extension case here because xfs_file_aio_write_checks() will relock
 505	 * the inode as necessary for EOF zeroing cases and fill out the new
 506	 * inode size as appropriate.
 507	 */
 508	if ((iocb->ki_pos & mp->m_blockmask) ||
 509	    ((iocb->ki_pos + count) & mp->m_blockmask)) {
 510		unaligned_io = 1;
 511
 512		/*
 513		 * We can't properly handle unaligned direct I/O to reflink
 514		 * files yet, as we can't unshare a partial block.
 515		 */
 516		if (xfs_is_reflink_inode(ip)) {
 517			trace_xfs_reflink_bounce_dio_write(ip, iocb->ki_pos, count);
 518			return -EREMCHG;
 519		}
 520		iolock = XFS_IOLOCK_EXCL;
 521	} else {
 522		iolock = XFS_IOLOCK_SHARED;
 523	}
 524
 525	if (iocb->ki_flags & IOCB_NOWAIT) {
 526		if (!xfs_ilock_nowait(ip, iolock))
 527			return -EAGAIN;
 528	} else {
 529		xfs_ilock(ip, iolock);
 530	}
 531
 532	ret = xfs_file_aio_write_checks(iocb, from, &iolock);
 533	if (ret)
 534		goto out;
 535	count = iov_iter_count(from);
 
 
 
 
 
 
 
 536
 537	/*
 538	 * If we are doing unaligned IO, wait for all other IO to drain,
 539	 * otherwise demote the lock if we had to take the exclusive lock
 540	 * for other reasons in xfs_file_aio_write_checks.
 541	 */
 542	if (unaligned_io) {
 543		/* If we are going to wait for other DIO to finish, bail */
 544		if (iocb->ki_flags & IOCB_NOWAIT) {
 545			if (atomic_read(&inode->i_dio_count))
 546				return -EAGAIN;
 547		} else {
 548			inode_dio_wait(inode);
 549		}
 550	} else if (iolock == XFS_IOLOCK_EXCL) {
 551		xfs_ilock_demote(ip, XFS_IOLOCK_EXCL);
 552		iolock = XFS_IOLOCK_SHARED;
 553	}
 554
 555	trace_xfs_file_direct_write(ip, count, iocb->ki_pos);
 556	ret = iomap_dio_rw(iocb, from, &xfs_iomap_ops, xfs_dio_write_end_io);
 557out:
 558	xfs_iunlock(ip, iolock);
 559
 560	/*
 561	 * No fallback to buffered IO on errors for XFS, direct IO will either
 562	 * complete fully or fail.
 563	 */
 564	ASSERT(ret < 0 || ret == count);
 565	return ret;
 566}
 567
 568static noinline ssize_t
 569xfs_file_dax_write(
 570	struct kiocb		*iocb,
 571	struct iov_iter		*from)
 572{
 573	struct inode		*inode = iocb->ki_filp->f_mapping->host;
 574	struct xfs_inode	*ip = XFS_I(inode);
 575	int			iolock = XFS_IOLOCK_EXCL;
 576	ssize_t			ret, error = 0;
 577	size_t			count;
 578	loff_t			pos;
 579
 580	if (iocb->ki_flags & IOCB_NOWAIT) {
 581		if (!xfs_ilock_nowait(ip, iolock))
 582			return -EAGAIN;
 583	} else {
 584		xfs_ilock(ip, iolock);
 585	}
 586
 587	ret = xfs_file_aio_write_checks(iocb, from, &iolock);
 588	if (ret)
 589		goto out;
 590
 591	pos = iocb->ki_pos;
 592	count = iov_iter_count(from);
 593
 594	trace_xfs_file_dax_write(ip, count, pos);
 595	ret = dax_iomap_rw(iocb, from, &xfs_iomap_ops);
 596	if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
 597		i_size_write(inode, iocb->ki_pos);
 598		error = xfs_setfilesize(ip, pos, ret);
 599	}
 600out:
 601	xfs_iunlock(ip, iolock);
 602	return error ? error : ret;
 603}
 604
 605STATIC ssize_t
 606xfs_file_buffered_aio_write(
 607	struct kiocb		*iocb,
 608	struct iov_iter		*from)
 
 
 
 
 609{
 610	struct file		*file = iocb->ki_filp;
 611	struct address_space	*mapping = file->f_mapping;
 612	struct inode		*inode = mapping->host;
 613	struct xfs_inode	*ip = XFS_I(inode);
 614	ssize_t			ret;
 615	int			enospc = 0;
 616	int			iolock;
 617
 618	if (iocb->ki_flags & IOCB_NOWAIT)
 619		return -EOPNOTSUPP;
 620
 621write_retry:
 622	iolock = XFS_IOLOCK_EXCL;
 623	xfs_ilock(ip, iolock);
 624
 625	ret = xfs_file_aio_write_checks(iocb, from, &iolock);
 626	if (ret)
 627		goto out;
 628
 629	/* We can write back this queue in page reclaim */
 630	current->backing_dev_info = inode_to_bdi(inode);
 631
 632	trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos);
 633	ret = iomap_file_buffered_write(iocb, from, &xfs_iomap_ops);
 634	if (likely(ret >= 0))
 635		iocb->ki_pos += ret;
 636
 
 
 
 
 637	/*
 638	 * If we hit a space limit, try to free up some lingering preallocated
 639	 * space before returning an error. In the case of ENOSPC, first try to
 640	 * write back all dirty inodes to free up some of the excess reserved
 641	 * metadata space. This reduces the chances that the eofblocks scan
 642	 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
 643	 * also behaves as a filter to prevent too many eofblocks scans from
 644	 * running at the same time.
 645	 */
 646	if (ret == -EDQUOT && !enospc) {
 647		xfs_iunlock(ip, iolock);
 648		enospc = xfs_inode_free_quota_eofblocks(ip);
 649		if (enospc)
 650			goto write_retry;
 651		enospc = xfs_inode_free_quota_cowblocks(ip);
 652		if (enospc)
 653			goto write_retry;
 654		iolock = 0;
 655	} else if (ret == -ENOSPC && !enospc) {
 656		struct xfs_eofblocks eofb = {0};
 657
 658		enospc = 1;
 659		xfs_flush_inodes(ip->i_mount);
 660
 661		xfs_iunlock(ip, iolock);
 662		eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
 663		xfs_icache_free_eofblocks(ip->i_mount, &eofb);
 664		xfs_icache_free_cowblocks(ip->i_mount, &eofb);
 665		goto write_retry;
 666	}
 667
 668	current->backing_dev_info = NULL;
 669out:
 670	if (iolock)
 671		xfs_iunlock(ip, iolock);
 672	return ret;
 673}
 674
 675STATIC ssize_t
 676xfs_file_write_iter(
 677	struct kiocb		*iocb,
 678	struct iov_iter		*from)
 
 
 679{
 680	struct file		*file = iocb->ki_filp;
 681	struct address_space	*mapping = file->f_mapping;
 682	struct inode		*inode = mapping->host;
 683	struct xfs_inode	*ip = XFS_I(inode);
 684	ssize_t			ret;
 685	size_t			ocount = iov_iter_count(from);
 
 
 
 686
 687	XFS_STATS_INC(ip->i_mount, xs_write_calls);
 
 
 
 
 688
 689	if (ocount == 0)
 690		return 0;
 691
 
 
 692	if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 693		return -EIO;
 694
 695	if (IS_DAX(inode))
 696		ret = xfs_file_dax_write(iocb, from);
 697	else if (iocb->ki_flags & IOCB_DIRECT) {
 698		/*
 699		 * Allow a directio write to fall back to a buffered
 700		 * write *only* in the case that we're doing a reflink
 701		 * CoW.  In all other directio scenarios we do not
 702		 * allow an operation to fall back to buffered mode.
 703		 */
 704		ret = xfs_file_dio_aio_write(iocb, from);
 705		if (ret == -EREMCHG)
 706			goto buffered;
 707	} else {
 708buffered:
 709		ret = xfs_file_buffered_aio_write(iocb, from);
 710	}
 711
 712	if (ret > 0) {
 713		XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
 714
 715		/* Handle various SYNC-type writes */
 716		ret = generic_write_sync(iocb, ret);
 
 
 
 
 
 
 
 
 
 717	}
 
 
 
 
 718	return ret;
 719}
 720
 721#define	XFS_FALLOC_FL_SUPPORTED						\
 722		(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |		\
 723		 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |	\
 724		 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
 725
 726STATIC long
 727xfs_file_fallocate(
 728	struct file		*file,
 729	int			mode,
 730	loff_t			offset,
 731	loff_t			len)
 732{
 733	struct inode		*inode = file_inode(file);
 734	struct xfs_inode	*ip = XFS_I(inode);
 735	long			error;
 736	enum xfs_prealloc_flags	flags = 0;
 737	uint			iolock = XFS_IOLOCK_EXCL;
 738	loff_t			new_size = 0;
 739	bool			do_file_insert = false;
 740
 741	if (!S_ISREG(inode->i_mode))
 742		return -EINVAL;
 743	if (mode & ~XFS_FALLOC_FL_SUPPORTED)
 744		return -EOPNOTSUPP;
 745
 746	xfs_ilock(ip, iolock);
 747	error = xfs_break_layouts(inode, &iolock);
 748	if (error)
 749		goto out_unlock;
 750
 751	xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
 752	iolock |= XFS_MMAPLOCK_EXCL;
 753
 754	if (mode & FALLOC_FL_PUNCH_HOLE) {
 755		error = xfs_free_file_space(ip, offset, len);
 756		if (error)
 757			goto out_unlock;
 758	} else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
 759		unsigned int blksize_mask = i_blocksize(inode) - 1;
 760
 761		if (offset & blksize_mask || len & blksize_mask) {
 762			error = -EINVAL;
 763			goto out_unlock;
 764		}
 765
 766		/*
 767		 * There is no need to overlap collapse range with EOF,
 768		 * in which case it is effectively a truncate operation
 769		 */
 770		if (offset + len >= i_size_read(inode)) {
 771			error = -EINVAL;
 772			goto out_unlock;
 773		}
 774
 775		new_size = i_size_read(inode) - len;
 776
 777		error = xfs_collapse_file_space(ip, offset, len);
 778		if (error)
 779			goto out_unlock;
 780	} else if (mode & FALLOC_FL_INSERT_RANGE) {
 781		unsigned int	blksize_mask = i_blocksize(inode) - 1;
 782		loff_t		isize = i_size_read(inode);
 783
 784		if (offset & blksize_mask || len & blksize_mask) {
 785			error = -EINVAL;
 786			goto out_unlock;
 787		}
 788
 789		/*
 790		 * New inode size must not exceed ->s_maxbytes, accounting for
 791		 * possible signed overflow.
 792		 */
 793		if (inode->i_sb->s_maxbytes - isize < len) {
 794			error = -EFBIG;
 795			goto out_unlock;
 796		}
 797		new_size = isize + len;
 798
 799		/* Offset should be less than i_size */
 800		if (offset >= isize) {
 801			error = -EINVAL;
 802			goto out_unlock;
 803		}
 804		do_file_insert = true;
 805	} else {
 806		flags |= XFS_PREALLOC_SET;
 807
 808		if (!(mode & FALLOC_FL_KEEP_SIZE) &&
 809		    offset + len > i_size_read(inode)) {
 810			new_size = offset + len;
 811			error = inode_newsize_ok(inode, new_size);
 812			if (error)
 813				goto out_unlock;
 814		}
 815
 816		if (mode & FALLOC_FL_ZERO_RANGE)
 817			error = xfs_zero_file_space(ip, offset, len);
 818		else {
 819			if (mode & FALLOC_FL_UNSHARE_RANGE) {
 820				error = xfs_reflink_unshare(ip, offset, len);
 821				if (error)
 822					goto out_unlock;
 823			}
 824			error = xfs_alloc_file_space(ip, offset, len,
 825						     XFS_BMAPI_PREALLOC);
 826		}
 
 
 
 827		if (error)
 828			goto out_unlock;
 829	}
 830
 831	if (file->f_flags & O_DSYNC)
 832		flags |= XFS_PREALLOC_SYNC;
 833
 834	error = xfs_update_prealloc_flags(ip, flags);
 835	if (error)
 836		goto out_unlock;
 837
 838	/* Change file size if needed */
 839	if (new_size) {
 840		struct iattr iattr;
 841
 842		iattr.ia_valid = ATTR_SIZE;
 843		iattr.ia_size = new_size;
 844		error = xfs_vn_setattr_size(file_dentry(file), &iattr);
 845		if (error)
 846			goto out_unlock;
 847	}
 848
 849	/*
 850	 * Perform hole insertion now that the file size has been
 851	 * updated so that if we crash during the operation we don't
 852	 * leave shifted extents past EOF and hence losing access to
 853	 * the data that is contained within them.
 854	 */
 855	if (do_file_insert)
 856		error = xfs_insert_file_space(ip, offset, len);
 857
 858out_unlock:
 859	xfs_iunlock(ip, iolock);
 860	return error;
 861}
 862
 863STATIC int
 864xfs_file_clone_range(
 865	struct file	*file_in,
 866	loff_t		pos_in,
 867	struct file	*file_out,
 868	loff_t		pos_out,
 869	u64		len)
 870{
 871	return xfs_reflink_remap_range(file_in, pos_in, file_out, pos_out,
 872				     len, false);
 873}
 874
 875STATIC ssize_t
 876xfs_file_dedupe_range(
 877	struct file	*src_file,
 878	u64		loff,
 879	u64		len,
 880	struct file	*dst_file,
 881	u64		dst_loff)
 882{
 883	struct inode	*srci = file_inode(src_file);
 884	u64		max_dedupe;
 885	int		error;
 886
 887	/*
 888	 * Since we have to read all these pages in to compare them, cut
 889	 * it off at MAX_RW_COUNT/2 rounded down to the nearest block.
 890	 * That means we won't do more than MAX_RW_COUNT IO per request.
 891	 */
 892	max_dedupe = (MAX_RW_COUNT >> 1) & ~(i_blocksize(srci) - 1);
 893	if (len > max_dedupe)
 894		len = max_dedupe;
 895	error = xfs_reflink_remap_range(src_file, loff, dst_file, dst_loff,
 896				     len, true);
 897	if (error)
 898		return error;
 899	return len;
 900}
 901
 902STATIC int
 903xfs_file_open(
 904	struct inode	*inode,
 905	struct file	*file)
 906{
 907	if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
 908		return -EFBIG;
 909	if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
 910		return -EIO;
 911	file->f_mode |= FMODE_NOWAIT;
 912	return 0;
 913}
 914
 915STATIC int
 916xfs_dir_open(
 917	struct inode	*inode,
 918	struct file	*file)
 919{
 920	struct xfs_inode *ip = XFS_I(inode);
 921	int		mode;
 922	int		error;
 923
 924	error = xfs_file_open(inode, file);
 925	if (error)
 926		return error;
 927
 928	/*
 929	 * If there are any blocks, read-ahead block 0 as we're almost
 930	 * certain to have the next operation be a read there.
 931	 */
 932	mode = xfs_ilock_data_map_shared(ip);
 933	if (ip->i_d.di_nextents > 0)
 934		error = xfs_dir3_data_readahead(ip, 0, -1);
 935	xfs_iunlock(ip, mode);
 936	return error;
 937}
 938
 939STATIC int
 940xfs_file_release(
 941	struct inode	*inode,
 942	struct file	*filp)
 943{
 944	return xfs_release(XFS_I(inode));
 945}
 946
 947STATIC int
 948xfs_file_readdir(
 949	struct file	*file,
 950	struct dir_context *ctx)
 
 951{
 952	struct inode	*inode = file_inode(file);
 953	xfs_inode_t	*ip = XFS_I(inode);
 
 954	size_t		bufsize;
 955
 956	/*
 957	 * The Linux API doesn't pass down the total size of the buffer
 958	 * we read into down to the filesystem.  With the filldir concept
 959	 * it's not needed for correct information, but the XFS dir2 leaf
 960	 * code wants an estimate of the buffer size to calculate it's
 961	 * readahead window and size the buffers used for mapping to
 962	 * physical blocks.
 963	 *
 964	 * Try to give it an estimate that's good enough, maybe at some
 965	 * point we can change the ->readdir prototype to include the
 966	 * buffer size.  For now we use the current glibc buffer size.
 967	 */
 968	bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE, ip->i_d.di_size);
 969
 970	return xfs_readdir(NULL, ip, ctx, bufsize);
 971}
 972
 973STATIC loff_t
 974xfs_file_llseek(
 975	struct file	*file,
 976	loff_t		offset,
 977	int		whence)
 978{
 979	struct inode		*inode = file->f_mapping->host;
 980
 981	if (XFS_FORCED_SHUTDOWN(XFS_I(inode)->i_mount))
 982		return -EIO;
 983
 984	switch (whence) {
 985	default:
 986		return generic_file_llseek(file, offset, whence);
 987	case SEEK_HOLE:
 988		offset = iomap_seek_hole(inode, offset, &xfs_iomap_ops);
 989		break;
 990	case SEEK_DATA:
 991		offset = iomap_seek_data(inode, offset, &xfs_iomap_ops);
 992		break;
 993	}
 994
 995	if (offset < 0)
 996		return offset;
 997	return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
 998}
 999
1000/*
1001 * Locking for serialisation of IO during page faults. This results in a lock
1002 * ordering of:
1003 *
1004 * mmap_sem (MM)
1005 *   sb_start_pagefault(vfs, freeze)
1006 *     i_mmaplock (XFS - truncate serialisation)
1007 *       page_lock (MM)
1008 *         i_lock (XFS - extent map serialisation)
1009 */
1010static int
1011__xfs_filemap_fault(
1012	struct vm_fault		*vmf,
1013	enum page_entry_size	pe_size,
1014	bool			write_fault)
1015{
1016	struct inode		*inode = file_inode(vmf->vma->vm_file);
1017	struct xfs_inode	*ip = XFS_I(inode);
1018	int			ret;
1019
1020	trace_xfs_filemap_fault(ip, pe_size, write_fault);
1021
1022	if (write_fault) {
1023		sb_start_pagefault(inode->i_sb);
1024		file_update_time(vmf->vma->vm_file);
1025	}
1026
1027	xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1028	if (IS_DAX(inode)) {
1029		pfn_t pfn;
1030
1031		ret = dax_iomap_fault(vmf, pe_size, &pfn, NULL, &xfs_iomap_ops);
1032		if (ret & VM_FAULT_NEEDDSYNC)
1033			ret = dax_finish_sync_fault(vmf, pe_size, pfn);
1034	} else {
1035		if (write_fault)
1036			ret = iomap_page_mkwrite(vmf, &xfs_iomap_ops);
1037		else
1038			ret = filemap_fault(vmf);
1039	}
1040	xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1041
1042	if (write_fault)
1043		sb_end_pagefault(inode->i_sb);
1044	return ret;
1045}
1046
1047static int
1048xfs_filemap_fault(
1049	struct vm_fault		*vmf)
1050{
1051	/* DAX can shortcut the normal fault path on write faults! */
1052	return __xfs_filemap_fault(vmf, PE_SIZE_PTE,
1053			IS_DAX(file_inode(vmf->vma->vm_file)) &&
1054			(vmf->flags & FAULT_FLAG_WRITE));
1055}
1056
1057static int
1058xfs_filemap_huge_fault(
1059	struct vm_fault		*vmf,
1060	enum page_entry_size	pe_size)
1061{
1062	if (!IS_DAX(file_inode(vmf->vma->vm_file)))
1063		return VM_FAULT_FALLBACK;
1064
1065	/* DAX can shortcut the normal fault path on write faults! */
1066	return __xfs_filemap_fault(vmf, pe_size,
1067			(vmf->flags & FAULT_FLAG_WRITE));
1068}
1069
1070static int
1071xfs_filemap_page_mkwrite(
1072	struct vm_fault		*vmf)
1073{
1074	return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1075}
1076
1077/*
1078 * pfn_mkwrite was originally intended to ensure we capture time stamp updates
1079 * on write faults. In reality, it needs to serialise against truncate and
1080 * prepare memory for writing so handle is as standard write fault.
1081 */
1082static int
1083xfs_filemap_pfn_mkwrite(
1084	struct vm_fault		*vmf)
1085{
1086
1087	return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1088}
1089
1090static const struct vm_operations_struct xfs_file_vm_ops = {
1091	.fault		= xfs_filemap_fault,
1092	.huge_fault	= xfs_filemap_huge_fault,
1093	.map_pages	= filemap_map_pages,
1094	.page_mkwrite	= xfs_filemap_page_mkwrite,
1095	.pfn_mkwrite	= xfs_filemap_pfn_mkwrite,
1096};
1097
1098STATIC int
1099xfs_file_mmap(
1100	struct file	*filp,
1101	struct vm_area_struct *vma)
1102{
1103	/*
1104	 * We don't support synchronous mappings for non-DAX files. At least
1105	 * until someone comes with a sensible use case.
1106	 */
1107	if (!IS_DAX(file_inode(filp)) && (vma->vm_flags & VM_SYNC))
1108		return -EOPNOTSUPP;
1109
1110	file_accessed(filp);
1111	vma->vm_ops = &xfs_file_vm_ops;
1112	if (IS_DAX(file_inode(filp)))
1113		vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
1114	return 0;
1115}
1116
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1117const struct file_operations xfs_file_operations = {
1118	.llseek		= xfs_file_llseek,
1119	.read_iter	= xfs_file_read_iter,
1120	.write_iter	= xfs_file_write_iter,
1121	.splice_read	= generic_file_splice_read,
1122	.splice_write	= iter_file_splice_write,
 
 
1123	.unlocked_ioctl	= xfs_file_ioctl,
1124#ifdef CONFIG_COMPAT
1125	.compat_ioctl	= xfs_file_compat_ioctl,
1126#endif
1127	.mmap		= xfs_file_mmap,
1128	.mmap_supported_flags = MAP_SYNC,
1129	.open		= xfs_file_open,
1130	.release	= xfs_file_release,
1131	.fsync		= xfs_file_fsync,
1132	.get_unmapped_area = thp_get_unmapped_area,
1133	.fallocate	= xfs_file_fallocate,
1134	.clone_file_range = xfs_file_clone_range,
1135	.dedupe_file_range = xfs_file_dedupe_range,
1136};
1137
1138const struct file_operations xfs_dir_file_operations = {
1139	.open		= xfs_dir_open,
1140	.read		= generic_read_dir,
1141	.iterate_shared	= xfs_file_readdir,
1142	.llseek		= generic_file_llseek,
1143	.unlocked_ioctl	= xfs_file_ioctl,
1144#ifdef CONFIG_COMPAT
1145	.compat_ioctl	= xfs_file_compat_ioctl,
1146#endif
1147	.fsync		= xfs_dir_fsync,
 
 
 
 
 
1148};
v3.1
   1/*
   2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3 * All Rights Reserved.
   4 *
   5 * This program is free software; you can redistribute it and/or
   6 * modify it under the terms of the GNU General Public License as
   7 * published by the Free Software Foundation.
   8 *
   9 * This program is distributed in the hope that it would be useful,
  10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 * GNU General Public License for more details.
  13 *
  14 * You should have received a copy of the GNU General Public License
  15 * along with this program; if not, write the Free Software Foundation,
  16 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17 */
  18#include "xfs.h"
  19#include "xfs_fs.h"
  20#include "xfs_bit.h"
  21#include "xfs_log.h"
  22#include "xfs_inum.h"
  23#include "xfs_sb.h"
  24#include "xfs_ag.h"
  25#include "xfs_trans.h"
  26#include "xfs_mount.h"
  27#include "xfs_bmap_btree.h"
  28#include "xfs_alloc.h"
  29#include "xfs_dinode.h"
  30#include "xfs_inode.h"
 
  31#include "xfs_inode_item.h"
  32#include "xfs_bmap.h"
 
  33#include "xfs_error.h"
  34#include "xfs_vnodeops.h"
  35#include "xfs_da_btree.h"
  36#include "xfs_ioctl.h"
  37#include "xfs_trace.h"
 
 
 
 
 
  38
  39#include <linux/dcache.h>
  40#include <linux/falloc.h>
 
 
 
  41
  42static const struct vm_operations_struct xfs_file_vm_ops;
  43
  44/*
  45 * Locking primitives for read and write IO paths to ensure we consistently use
  46 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
  47 */
  48static inline void
  49xfs_rw_ilock(
  50	struct xfs_inode	*ip,
  51	int			type)
  52{
  53	if (type & XFS_IOLOCK_EXCL)
  54		mutex_lock(&VFS_I(ip)->i_mutex);
  55	xfs_ilock(ip, type);
  56}
 
 
 
 
 
 
 
 
 
 
 
 
 
  57
  58static inline void
  59xfs_rw_iunlock(
  60	struct xfs_inode	*ip,
  61	int			type)
  62{
  63	xfs_iunlock(ip, type);
  64	if (type & XFS_IOLOCK_EXCL)
  65		mutex_unlock(&VFS_I(ip)->i_mutex);
  66}
  67
  68static inline void
  69xfs_rw_ilock_demote(
  70	struct xfs_inode	*ip,
  71	int			type)
  72{
  73	xfs_ilock_demote(ip, type);
  74	if (type & XFS_IOLOCK_EXCL)
  75		mutex_unlock(&VFS_I(ip)->i_mutex);
  76}
  77
  78/*
  79 *	xfs_iozero
  80 *
  81 *	xfs_iozero clears the specified range of buffer supplied,
  82 *	and marks all the affected blocks as valid and modified.  If
  83 *	an affected block is not allocated, it will be allocated.  If
  84 *	an affected block is not completely overwritten, and is not
  85 *	valid before the operation, it will be read from disk before
  86 *	being partially zeroed.
  87 */
  88STATIC int
  89xfs_iozero(
  90	struct xfs_inode	*ip,	/* inode			*/
  91	loff_t			pos,	/* offset in file		*/
  92	size_t			count)	/* size of data to zero		*/
  93{
  94	struct page		*page;
  95	struct address_space	*mapping;
  96	int			status;
  97
  98	mapping = VFS_I(ip)->i_mapping;
  99	do {
 100		unsigned offset, bytes;
 101		void *fsdata;
 102
 103		offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
 104		bytes = PAGE_CACHE_SIZE - offset;
 105		if (bytes > count)
 106			bytes = count;
 107
 108		status = pagecache_write_begin(NULL, mapping, pos, bytes,
 109					AOP_FLAG_UNINTERRUPTIBLE,
 110					&page, &fsdata);
 111		if (status)
 112			break;
 113
 114		zero_user(page, offset, bytes);
 115
 116		status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
 117					page, fsdata);
 118		WARN_ON(status <= 0); /* can't return less than zero! */
 119		pos += bytes;
 120		count -= bytes;
 121		status = 0;
 122	} while (count);
 123
 124	return (-status);
 
 
 125}
 126
 127STATIC int
 128xfs_file_fsync(
 129	struct file		*file,
 130	loff_t			start,
 131	loff_t			end,
 132	int			datasync)
 133{
 134	struct inode		*inode = file->f_mapping->host;
 135	struct xfs_inode	*ip = XFS_I(inode);
 136	struct xfs_mount	*mp = ip->i_mount;
 137	struct xfs_trans	*tp;
 138	int			error = 0;
 139	int			log_flushed = 0;
 
 140
 141	trace_xfs_file_fsync(ip);
 142
 143	error = filemap_write_and_wait_range(inode->i_mapping, start, end);
 144	if (error)
 145		return error;
 146
 147	if (XFS_FORCED_SHUTDOWN(mp))
 148		return -XFS_ERROR(EIO);
 149
 150	xfs_iflags_clear(ip, XFS_ITRUNCATED);
 151
 152	xfs_ilock(ip, XFS_IOLOCK_SHARED);
 153	xfs_ioend_wait(ip);
 154	xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 155
 156	if (mp->m_flags & XFS_MOUNT_BARRIER) {
 157		/*
 158		 * If we have an RT and/or log subvolume we need to make sure
 159		 * to flush the write cache the device used for file data
 160		 * first.  This is to ensure newly written file data make
 161		 * it to disk before logging the new inode size in case of
 162		 * an extending write.
 163		 */
 164		if (XFS_IS_REALTIME_INODE(ip))
 165			xfs_blkdev_issue_flush(mp->m_rtdev_targp);
 166		else if (mp->m_logdev_targp != mp->m_ddev_targp)
 167			xfs_blkdev_issue_flush(mp->m_ddev_targp);
 168	}
 169
 170	/*
 171	 * We always need to make sure that the required inode state is safe on
 172	 * disk.  The inode might be clean but we still might need to force the
 173	 * log because of committed transactions that haven't hit the disk yet.
 174	 * Likewise, there could be unflushed non-transactional changes to the
 175	 * inode core that have to go to disk and this requires us to issue
 176	 * a synchronous transaction to capture these changes correctly.
 177	 *
 178	 * This code relies on the assumption that if the i_update_core field
 179	 * of the inode is clear and the inode is unpinned then it is clean
 180	 * and no action is required.
 181	 */
 182	xfs_ilock(ip, XFS_ILOCK_SHARED);
 
 
 
 183
 184	/*
 185	 * First check if the VFS inode is marked dirty.  All the dirtying
 186	 * of non-transactional updates no goes through mark_inode_dirty*,
 187	 * which allows us to distinguish beteeen pure timestamp updates
 188	 * and i_size updates which need to be caught for fdatasync.
 189	 * After that also theck for the dirty state in the XFS inode, which
 190	 * might gets cleared when the inode gets written out via the AIL
 191	 * or xfs_iflush_cluster.
 
 
 
 
 192	 */
 193	if (((inode->i_state & I_DIRTY_DATASYNC) ||
 194	    ((inode->i_state & I_DIRTY_SYNC) && !datasync)) &&
 195	    ip->i_update_core) {
 196		/*
 197		 * Kick off a transaction to log the inode core to get the
 198		 * updates.  The sync transaction will also force the log.
 199		 */
 200		xfs_iunlock(ip, XFS_ILOCK_SHARED);
 201		tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
 202		error = xfs_trans_reserve(tp, 0,
 203				XFS_FSYNC_TS_LOG_RES(mp), 0, 0, 0);
 204		if (error) {
 205			xfs_trans_cancel(tp, 0);
 206			return -error;
 207		}
 208		xfs_ilock(ip, XFS_ILOCK_EXCL);
 209
 210		/*
 211		 * Note - it's possible that we might have pushed ourselves out
 212		 * of the way during trans_reserve which would flush the inode.
 213		 * But there's no guarantee that the inode buffer has actually
 214		 * gone out yet (it's delwri).	Plus the buffer could be pinned
 215		 * anyway if it's part of an inode in another recent
 216		 * transaction.	 So we play it safe and fire off the
 217		 * transaction anyway.
 218		 */
 219		xfs_trans_ijoin(tp, ip);
 220		xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 221		xfs_trans_set_sync(tp);
 222		error = _xfs_trans_commit(tp, 0, &log_flushed);
 223
 224		xfs_iunlock(ip, XFS_ILOCK_EXCL);
 225	} else {
 226		/*
 227		 * Timestamps/size haven't changed since last inode flush or
 228		 * inode transaction commit.  That means either nothing got
 229		 * written or a transaction committed which caught the updates.
 230		 * If the latter happened and the transaction hasn't hit the
 231		 * disk yet, the inode will be still be pinned.  If it is,
 232		 * force the log.
 233		 */
 234		if (xfs_ipincount(ip)) {
 235			error = _xfs_log_force_lsn(mp,
 236					ip->i_itemp->ili_last_lsn,
 237					XFS_LOG_SYNC, &log_flushed);
 238		}
 239		xfs_iunlock(ip, XFS_ILOCK_SHARED);
 240	}
 
 241
 242	/*
 243	 * If we only have a single device, and the log force about was
 244	 * a no-op we might have to flush the data device cache here.
 245	 * This can only happen for fdatasync/O_DSYNC if we were overwriting
 246	 * an already allocated file and thus do not have any metadata to
 247	 * commit.
 248	 */
 249	if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
 250	    mp->m_logdev_targp == mp->m_ddev_targp &&
 251	    !XFS_IS_REALTIME_INODE(ip) &&
 252	    !log_flushed)
 253		xfs_blkdev_issue_flush(mp->m_ddev_targp);
 254
 255	return -error;
 256}
 257
 258STATIC ssize_t
 259xfs_file_aio_read(
 260	struct kiocb		*iocb,
 261	const struct iovec	*iovp,
 262	unsigned long		nr_segs,
 263	loff_t			pos)
 264{
 265	struct file		*file = iocb->ki_filp;
 266	struct inode		*inode = file->f_mapping->host;
 267	struct xfs_inode	*ip = XFS_I(inode);
 268	struct xfs_mount	*mp = ip->i_mount;
 269	size_t			size = 0;
 270	ssize_t			ret = 0;
 271	int			ioflags = 0;
 272	xfs_fsize_t		n;
 273	unsigned long		seg;
 274
 275	XFS_STATS_INC(xs_read_calls);
 276
 277	BUG_ON(iocb->ki_pos != pos);
 278
 279	if (unlikely(file->f_flags & O_DIRECT))
 280		ioflags |= IO_ISDIRECT;
 281	if (file->f_mode & FMODE_NOCMTIME)
 282		ioflags |= IO_INVIS;
 283
 284	/* START copy & waste from filemap.c */
 285	for (seg = 0; seg < nr_segs; seg++) {
 286		const struct iovec *iv = &iovp[seg];
 287
 288		/*
 289		 * If any segment has a negative length, or the cumulative
 290		 * length ever wraps negative then return -EINVAL.
 291		 */
 292		size += iv->iov_len;
 293		if (unlikely((ssize_t)(size|iv->iov_len) < 0))
 294			return XFS_ERROR(-EINVAL);
 295	}
 296	/* END copy & waste from filemap.c */
 297
 298	if (unlikely(ioflags & IO_ISDIRECT)) {
 299		xfs_buftarg_t	*target =
 300			XFS_IS_REALTIME_INODE(ip) ?
 301				mp->m_rtdev_targp : mp->m_ddev_targp;
 302		if ((iocb->ki_pos & target->bt_smask) ||
 303		    (size & target->bt_smask)) {
 304			if (iocb->ki_pos == ip->i_size)
 305				return 0;
 306			return -XFS_ERROR(EINVAL);
 307		}
 308	}
 309
 310	n = XFS_MAXIOFFSET(mp) - iocb->ki_pos;
 311	if (n <= 0 || size == 0)
 312		return 0;
 313
 314	if (n < size)
 315		size = n;
 316
 317	if (XFS_FORCED_SHUTDOWN(mp))
 318		return -EIO;
 319
 320	if (unlikely(ioflags & IO_ISDIRECT)) {
 321		xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
 322
 323		if (inode->i_mapping->nrpages) {
 324			ret = -xfs_flushinval_pages(ip,
 325					(iocb->ki_pos & PAGE_CACHE_MASK),
 326					-1, FI_REMAPF_LOCKED);
 327			if (ret) {
 328				xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
 329				return ret;
 330			}
 331		}
 332		xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
 333	} else
 334		xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
 335
 336	trace_xfs_file_read(ip, size, iocb->ki_pos, ioflags);
 337
 338	ret = generic_file_aio_read(iocb, iovp, nr_segs, iocb->ki_pos);
 339	if (ret > 0)
 340		XFS_STATS_ADD(xs_read_bytes, ret);
 341
 342	xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
 343	return ret;
 344}
 345
 346STATIC ssize_t
 347xfs_file_splice_read(
 348	struct file		*infilp,
 349	loff_t			*ppos,
 350	struct pipe_inode_info	*pipe,
 351	size_t			count,
 352	unsigned int		flags)
 353{
 354	struct xfs_inode	*ip = XFS_I(infilp->f_mapping->host);
 355	int			ioflags = 0;
 356	ssize_t			ret;
 357
 358	XFS_STATS_INC(xs_read_calls);
 359
 360	if (infilp->f_mode & FMODE_NOCMTIME)
 361		ioflags |= IO_INVIS;
 362
 363	if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 364		return -EIO;
 
 
 
 
 365
 366	xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
 367
 368	trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
 369
 370	ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
 371	if (ret > 0)
 372		XFS_STATS_ADD(xs_read_bytes, ret);
 373
 374	xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
 375	return ret;
 376}
 377
 378STATIC void
 379xfs_aio_write_isize_update(
 380	struct inode	*inode,
 381	loff_t		*ppos,
 382	ssize_t		bytes_written)
 383{
 384	struct xfs_inode	*ip = XFS_I(inode);
 385	xfs_fsize_t		isize = i_size_read(inode);
 386
 387	if (bytes_written > 0)
 388		XFS_STATS_ADD(xs_write_bytes, bytes_written);
 389
 390	if (unlikely(bytes_written < 0 && bytes_written != -EFAULT &&
 391					*ppos > isize))
 392		*ppos = isize;
 393
 394	if (*ppos > ip->i_size) {
 395		xfs_rw_ilock(ip, XFS_ILOCK_EXCL);
 396		if (*ppos > ip->i_size)
 397			ip->i_size = *ppos;
 398		xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
 399	}
 400}
 
 401
 402/*
 403 * If this was a direct or synchronous I/O that failed (such as ENOSPC) then
 404 * part of the I/O may have been written to disk before the error occurred.  In
 405 * this case the on-disk file size may have been adjusted beyond the in-memory
 406 * file size and now needs to be truncated back.
 407 */
 408STATIC void
 409xfs_aio_write_newsize_update(
 410	struct xfs_inode	*ip)
 411{
 412	if (ip->i_new_size) {
 413		xfs_rw_ilock(ip, XFS_ILOCK_EXCL);
 414		ip->i_new_size = 0;
 415		if (ip->i_d.di_size > ip->i_size)
 416			ip->i_d.di_size = ip->i_size;
 417		xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
 418	}
 419}
 420
 421/*
 422 * xfs_file_splice_write() does not use xfs_rw_ilock() because
 423 * generic_file_splice_write() takes the i_mutex itself. This, in theory,
 424 * couuld cause lock inversions between the aio_write path and the splice path
 425 * if someone is doing concurrent splice(2) based writes and write(2) based
 426 * writes to the same inode. The only real way to fix this is to re-implement
 427 * the generic code here with correct locking orders.
 428 */
 429STATIC ssize_t
 430xfs_file_splice_write(
 431	struct pipe_inode_info	*pipe,
 432	struct file		*outfilp,
 433	loff_t			*ppos,
 434	size_t			count,
 435	unsigned int		flags)
 436{
 437	struct inode		*inode = outfilp->f_mapping->host;
 438	struct xfs_inode	*ip = XFS_I(inode);
 439	xfs_fsize_t		new_size;
 440	int			ioflags = 0;
 441	ssize_t			ret;
 442
 443	XFS_STATS_INC(xs_write_calls);
 444
 445	if (outfilp->f_mode & FMODE_NOCMTIME)
 446		ioflags |= IO_INVIS;
 447
 448	if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 449		return -EIO;
 450
 451	xfs_ilock(ip, XFS_IOLOCK_EXCL);
 
 
 
 
 
 452
 453	new_size = *ppos + count;
 454
 455	xfs_ilock(ip, XFS_ILOCK_EXCL);
 456	if (new_size > ip->i_size)
 457		ip->i_new_size = new_size;
 458	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 459
 460	trace_xfs_file_splice_write(ip, count, *ppos, ioflags);
 461
 462	ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
 463
 464	xfs_aio_write_isize_update(inode, ppos, ret);
 465	xfs_aio_write_newsize_update(ip);
 466	xfs_iunlock(ip, XFS_IOLOCK_EXCL);
 467	return ret;
 468}
 469
 470/*
 471 * This routine is called to handle zeroing any space in the last
 472 * block of the file that is beyond the EOF.  We do this since the
 473 * size is being increased without writing anything to that block
 474 * and we don't want anyone to read the garbage on the disk.
 
 475 */
 476STATIC int				/* error (positive) */
 477xfs_zero_last_block(
 478	xfs_inode_t	*ip,
 479	xfs_fsize_t	offset,
 480	xfs_fsize_t	isize)
 481{
 482	xfs_fileoff_t	last_fsb;
 483	xfs_mount_t	*mp = ip->i_mount;
 484	int		nimaps;
 485	int		zero_offset;
 486	int		zero_len;
 487	int		error = 0;
 488	xfs_bmbt_irec_t	imap;
 
 
 
 
 
 489
 490	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 
 
 491
 492	zero_offset = XFS_B_FSB_OFFSET(mp, isize);
 493	if (zero_offset == 0) {
 494		/*
 495		 * There are no extra bytes in the last block on disk to
 496		 * zero, so return.
 497		 */
 498		return 0;
 499	}
 500
 501	last_fsb = XFS_B_TO_FSBT(mp, isize);
 502	nimaps = 1;
 503	error = xfs_bmapi(NULL, ip, last_fsb, 1, 0, NULL, 0, &imap,
 504			  &nimaps, NULL);
 505	if (error) {
 506		return error;
 507	}
 508	ASSERT(nimaps > 0);
 509	/*
 510	 * If the block underlying isize is just a hole, then there
 511	 * is nothing to zero.
 512	 */
 513	if (imap.br_startblock == HOLESTARTBLOCK) {
 514		return 0;
 
 
 
 515	}
 516	/*
 517	 * Zero the part of the last block beyond the EOF, and write it
 518	 * out sync.  We need to drop the ilock while we do this so we
 519	 * don't deadlock when the buffer cache calls back to us.
 
 
 
 
 
 
 
 
 
 
 520	 */
 521	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 522
 523	zero_len = mp->m_sb.sb_blocksize - zero_offset;
 524	if (isize + zero_len > offset)
 525		zero_len = offset - isize;
 526	error = xfs_iozero(ip, isize, zero_len);
 527
 528	xfs_ilock(ip, XFS_ILOCK_EXCL);
 529	ASSERT(error >= 0);
 530	return error;
 531}
 532
 533/*
 534 * Zero any on disk space between the current EOF and the new,
 535 * larger EOF.  This handles the normal case of zeroing the remainder
 536 * of the last block in the file and the unusual case of zeroing blocks
 537 * out beyond the size of the file.  This second case only happens
 538 * with fixed size extents and when the system crashes before the inode
 539 * size was updated but after blocks were allocated.  If fill is set,
 540 * then any holes in the range are filled and zeroed.  If not, the holes
 541 * are left alone as holes.
 542 */
 543
 544int					/* error (positive) */
 545xfs_zero_eof(
 546	xfs_inode_t	*ip,
 547	xfs_off_t	offset,		/* starting I/O offset */
 548	xfs_fsize_t	isize)		/* current inode size */
 549{
 550	xfs_mount_t	*mp = ip->i_mount;
 551	xfs_fileoff_t	start_zero_fsb;
 552	xfs_fileoff_t	end_zero_fsb;
 553	xfs_fileoff_t	zero_count_fsb;
 554	xfs_fileoff_t	last_fsb;
 555	xfs_fileoff_t	zero_off;
 556	xfs_fsize_t	zero_len;
 557	int		nimaps;
 558	int		error = 0;
 559	xfs_bmbt_irec_t	imap;
 560
 561	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
 562	ASSERT(offset > isize);
 563
 564	/*
 565	 * First handle zeroing the block on which isize resides.
 566	 * We only zero a part of that block so it is handled specially.
 
 
 567	 */
 568	error = xfs_zero_last_block(ip, offset, isize);
 569	if (error) {
 570		ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
 571		return error;
 572	}
 573
 574	/*
 575	 * Calculate the range between the new size and the old
 576	 * where blocks needing to be zeroed may exist.  To get the
 577	 * block where the last byte in the file currently resides,
 578	 * we need to subtract one from the size and truncate back
 579	 * to a block boundary.  We subtract 1 in case the size is
 580	 * exactly on a block boundary.
 581	 */
 582	last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
 583	start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
 584	end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
 585	ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
 586	if (last_fsb == end_zero_fsb) {
 587		/*
 588		 * The size was only incremented on its last block.
 589		 * We took care of that above, so just return.
 590		 */
 591		return 0;
 592	}
 593
 594	ASSERT(start_zero_fsb <= end_zero_fsb);
 595	while (start_zero_fsb <= end_zero_fsb) {
 596		nimaps = 1;
 597		zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
 598		error = xfs_bmapi(NULL, ip, start_zero_fsb, zero_count_fsb,
 599				  0, NULL, 0, &imap, &nimaps, NULL);
 600		if (error) {
 601			ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
 602			return error;
 603		}
 604		ASSERT(nimaps > 0);
 605
 606		if (imap.br_state == XFS_EXT_UNWRITTEN ||
 607		    imap.br_startblock == HOLESTARTBLOCK) {
 608			/*
 609			 * This loop handles initializing pages that were
 610			 * partially initialized by the code below this
 611			 * loop. It basically zeroes the part of the page
 612			 * that sits on a hole and sets the page as P_HOLE
 613			 * and calls remapf if it is a mapped file.
 614			 */
 615			start_zero_fsb = imap.br_startoff + imap.br_blockcount;
 616			ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
 617			continue;
 618		}
 619
 620		/*
 621		 * There are blocks we need to zero.
 622		 * Drop the inode lock while we're doing the I/O.
 623		 * We'll still have the iolock to protect us.
 624		 */
 625		xfs_iunlock(ip, XFS_ILOCK_EXCL);
 626
 627		zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
 628		zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
 629
 630		if ((zero_off + zero_len) > offset)
 631			zero_len = offset - zero_off;
 632
 633		error = xfs_iozero(ip, zero_off, zero_len);
 634		if (error) {
 635			goto out_lock;
 636		}
 637
 638		start_zero_fsb = imap.br_startoff + imap.br_blockcount;
 639		ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
 640
 641		xfs_ilock(ip, XFS_ILOCK_EXCL);
 642	}
 643
 644	return 0;
 645
 646out_lock:
 647	xfs_ilock(ip, XFS_ILOCK_EXCL);
 648	ASSERT(error >= 0);
 649	return error;
 650}
 651
 652/*
 653 * Common pre-write limit and setup checks.
 654 *
 655 * Returns with iolock held according to @iolock.
 656 */
 657STATIC ssize_t
 658xfs_file_aio_write_checks(
 659	struct file		*file,
 660	loff_t			*pos,
 661	size_t			*count,
 662	int			*iolock)
 663{
 664	struct inode		*inode = file->f_mapping->host;
 665	struct xfs_inode	*ip = XFS_I(inode);
 666	xfs_fsize_t		new_size;
 667	int			error = 0;
 668
 669	error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
 670	if (error) {
 671		xfs_rw_iunlock(ip, XFS_ILOCK_EXCL | *iolock);
 672		*iolock = 0;
 673		return error;
 674	}
 675
 676	new_size = *pos + *count;
 677	if (new_size > ip->i_size)
 678		ip->i_new_size = new_size;
 679
 680	if (likely(!(file->f_mode & FMODE_NOCMTIME)))
 681		file_update_time(file);
 
 
 
 682
 683	/*
 684	 * If the offset is beyond the size of the file, we need to zero any
 685	 * blocks that fall between the existing EOF and the start of this
 686	 * write.
 
 687	 */
 688	if (*pos > ip->i_size)
 689		error = -xfs_zero_eof(ip, *pos, ip->i_size);
 690
 691	xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
 692	if (error)
 693		return error;
 694
 695	/*
 696	 * If we're writing the file then make sure to clear the setuid and
 697	 * setgid bits if the process is not being run by root.  This keeps
 698	 * people from modifying setuid and setgid binaries.
 
 
 
 
 
 
 699	 */
 700	return file_remove_suid(file);
 
 
 
 
 
 
 
 701
 
 702}
 703
 704/*
 705 * xfs_file_dio_aio_write - handle direct IO writes
 706 *
 707 * Lock the inode appropriately to prepare for and issue a direct IO write.
 708 * By separating it from the buffered write path we remove all the tricky to
 709 * follow locking changes and looping.
 710 *
 711 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
 712 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
 713 * pages are flushed out.
 714 *
 715 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
 716 * allowing them to be done in parallel with reads and other direct IO writes.
 717 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
 718 * needs to do sub-block zeroing and that requires serialisation against other
 719 * direct IOs to the same block. In this case we need to serialise the
 720 * submission of the unaligned IOs so that we don't get racing block zeroing in
 721 * the dio layer.  To avoid the problem with aio, we also need to wait for
 722 * outstanding IOs to complete so that unwritten extent conversion is completed
 723 * before we try to map the overlapping block. This is currently implemented by
 724 * hitting it with a big hammer (i.e. xfs_ioend_wait()).
 725 *
 726 * Returns with locks held indicated by @iolock and errors indicated by
 727 * negative return values.
 728 */
 729STATIC ssize_t
 730xfs_file_dio_aio_write(
 731	struct kiocb		*iocb,
 732	const struct iovec	*iovp,
 733	unsigned long		nr_segs,
 734	loff_t			pos,
 735	size_t			ocount,
 736	int			*iolock)
 737{
 738	struct file		*file = iocb->ki_filp;
 739	struct address_space	*mapping = file->f_mapping;
 740	struct inode		*inode = mapping->host;
 741	struct xfs_inode	*ip = XFS_I(inode);
 742	struct xfs_mount	*mp = ip->i_mount;
 743	ssize_t			ret = 0;
 744	size_t			count = ocount;
 745	int			unaligned_io = 0;
 746	struct xfs_buftarg	*target = XFS_IS_REALTIME_INODE(ip) ?
 
 
 747					mp->m_rtdev_targp : mp->m_ddev_targp;
 748
 749	*iolock = 0;
 750	if ((pos & target->bt_smask) || (count & target->bt_smask))
 751		return -XFS_ERROR(EINVAL);
 752
 753	if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
 
 
 
 
 
 
 
 
 754		unaligned_io = 1;
 755
 756	if (unaligned_io || mapping->nrpages || pos > ip->i_size)
 757		*iolock = XFS_IOLOCK_EXCL;
 758	else
 759		*iolock = XFS_IOLOCK_SHARED;
 760	xfs_rw_ilock(ip, XFS_ILOCK_EXCL | *iolock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 761
 762	ret = xfs_file_aio_write_checks(file, &pos, &count, iolock);
 763	if (ret)
 764		return ret;
 765
 766	if (mapping->nrpages) {
 767		WARN_ON(*iolock != XFS_IOLOCK_EXCL);
 768		ret = -xfs_flushinval_pages(ip, (pos & PAGE_CACHE_MASK), -1,
 769							FI_REMAPF_LOCKED);
 770		if (ret)
 771			return ret;
 772	}
 773
 774	/*
 775	 * If we are doing unaligned IO, wait for all other IO to drain,
 776	 * otherwise demote the lock if we had to flush cached pages
 
 777	 */
 778	if (unaligned_io)
 779		xfs_ioend_wait(ip);
 780	else if (*iolock == XFS_IOLOCK_EXCL) {
 781		xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
 782		*iolock = XFS_IOLOCK_SHARED;
 
 
 
 
 
 
 783	}
 784
 785	trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
 786	ret = generic_file_direct_write(iocb, iovp,
 787			&nr_segs, pos, &iocb->ki_pos, count, ocount);
 
 788
 789	/* No fallback to buffered IO on errors for XFS. */
 
 
 
 790	ASSERT(ret < 0 || ret == count);
 791	return ret;
 792}
 793
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 794STATIC ssize_t
 795xfs_file_buffered_aio_write(
 796	struct kiocb		*iocb,
 797	const struct iovec	*iovp,
 798	unsigned long		nr_segs,
 799	loff_t			pos,
 800	size_t			ocount,
 801	int			*iolock)
 802{
 803	struct file		*file = iocb->ki_filp;
 804	struct address_space	*mapping = file->f_mapping;
 805	struct inode		*inode = mapping->host;
 806	struct xfs_inode	*ip = XFS_I(inode);
 807	ssize_t			ret;
 808	int			enospc = 0;
 809	size_t			count = ocount;
 810
 811	*iolock = XFS_IOLOCK_EXCL;
 812	xfs_rw_ilock(ip, XFS_ILOCK_EXCL | *iolock);
 
 
 
 
 813
 814	ret = xfs_file_aio_write_checks(file, &pos, &count, iolock);
 815	if (ret)
 816		return ret;
 817
 818	/* We can write back this queue in page reclaim */
 819	current->backing_dev_info = mapping->backing_dev_info;
 
 
 
 
 
 820
 821write_retry:
 822	trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
 823	ret = generic_file_buffered_write(iocb, iovp, nr_segs,
 824			pos, &iocb->ki_pos, count, ret);
 825	/*
 826	 * if we just got an ENOSPC, flush the inode now we aren't holding any
 827	 * page locks and retry *once*
 
 
 
 
 
 828	 */
 829	if (ret == -ENOSPC && !enospc) {
 830		ret = -xfs_flush_pages(ip, 0, -1, 0, FI_NONE);
 831		if (ret)
 832			return ret;
 
 
 
 
 
 
 
 
 833		enospc = 1;
 
 
 
 
 
 
 834		goto write_retry;
 835	}
 
 836	current->backing_dev_info = NULL;
 
 
 
 837	return ret;
 838}
 839
 840STATIC ssize_t
 841xfs_file_aio_write(
 842	struct kiocb		*iocb,
 843	const struct iovec	*iovp,
 844	unsigned long		nr_segs,
 845	loff_t			pos)
 846{
 847	struct file		*file = iocb->ki_filp;
 848	struct address_space	*mapping = file->f_mapping;
 849	struct inode		*inode = mapping->host;
 850	struct xfs_inode	*ip = XFS_I(inode);
 851	ssize_t			ret;
 852	int			iolock;
 853	size_t			ocount = 0;
 854
 855	XFS_STATS_INC(xs_write_calls);
 856
 857	BUG_ON(iocb->ki_pos != pos);
 858
 859	ret = generic_segment_checks(iovp, &nr_segs, &ocount, VERIFY_READ);
 860	if (ret)
 861		return ret;
 862
 863	if (ocount == 0)
 864		return 0;
 865
 866	xfs_wait_for_freeze(ip->i_mount, SB_FREEZE_WRITE);
 867
 868	if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 869		return -EIO;
 870
 871	if (unlikely(file->f_flags & O_DIRECT))
 872		ret = xfs_file_dio_aio_write(iocb, iovp, nr_segs, pos,
 873						ocount, &iolock);
 874	else
 875		ret = xfs_file_buffered_aio_write(iocb, iovp, nr_segs, pos,
 876						ocount, &iolock);
 877
 878	xfs_aio_write_isize_update(inode, &iocb->ki_pos, ret);
 
 
 
 
 
 
 
 
 879
 880	if (ret <= 0)
 881		goto out_unlock;
 882
 883	/* Handle various SYNC-type writes */
 884	if ((file->f_flags & O_DSYNC) || IS_SYNC(inode)) {
 885		loff_t end = pos + ret - 1;
 886		int error;
 887
 888		xfs_rw_iunlock(ip, iolock);
 889		error = xfs_file_fsync(file, pos, end,
 890				      (file->f_flags & __O_SYNC) ? 0 : 1);
 891		xfs_rw_ilock(ip, iolock);
 892		if (error)
 893			ret = error;
 894	}
 895
 896out_unlock:
 897	xfs_aio_write_newsize_update(ip);
 898	xfs_rw_iunlock(ip, iolock);
 899	return ret;
 900}
 901
 
 
 
 
 
 902STATIC long
 903xfs_file_fallocate(
 904	struct file	*file,
 905	int		mode,
 906	loff_t		offset,
 907	loff_t		len)
 908{
 909	struct inode	*inode = file->f_path.dentry->d_inode;
 910	long		error;
 911	loff_t		new_size = 0;
 912	xfs_flock64_t	bf;
 913	xfs_inode_t	*ip = XFS_I(inode);
 914	int		cmd = XFS_IOC_RESVSP;
 915	int		attr_flags = XFS_ATTR_NOLOCK;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 916
 917	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
 918		return -EOPNOTSUPP;
 
 
 
 
 
 919
 920	bf.l_whence = 0;
 921	bf.l_start = offset;
 922	bf.l_len = len;
 923
 924	xfs_ilock(ip, XFS_IOLOCK_EXCL);
 925
 926	if (mode & FALLOC_FL_PUNCH_HOLE)
 927		cmd = XFS_IOC_UNRESVSP;
 928
 929	/* check the new inode size is valid before allocating */
 930	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
 931	    offset + len > i_size_read(inode)) {
 932		new_size = offset + len;
 933		error = inode_newsize_ok(inode, new_size);
 934		if (error)
 935			goto out_unlock;
 936	}
 937
 938	if (file->f_flags & O_DSYNC)
 939		attr_flags |= XFS_ATTR_SYNC;
 940
 941	error = -xfs_change_file_space(ip, cmd, &bf, 0, attr_flags);
 942	if (error)
 943		goto out_unlock;
 944
 945	/* Change file size if needed */
 946	if (new_size) {
 947		struct iattr iattr;
 948
 949		iattr.ia_valid = ATTR_SIZE;
 950		iattr.ia_size = new_size;
 951		error = -xfs_setattr_size(ip, &iattr, XFS_ATTR_NOLOCK);
 
 
 952	}
 953
 
 
 
 
 
 
 
 
 
 954out_unlock:
 955	xfs_iunlock(ip, XFS_IOLOCK_EXCL);
 956	return error;
 957}
 958
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 959
 960STATIC int
 961xfs_file_open(
 962	struct inode	*inode,
 963	struct file	*file)
 964{
 965	if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
 966		return -EFBIG;
 967	if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
 968		return -EIO;
 
 969	return 0;
 970}
 971
 972STATIC int
 973xfs_dir_open(
 974	struct inode	*inode,
 975	struct file	*file)
 976{
 977	struct xfs_inode *ip = XFS_I(inode);
 978	int		mode;
 979	int		error;
 980
 981	error = xfs_file_open(inode, file);
 982	if (error)
 983		return error;
 984
 985	/*
 986	 * If there are any blocks, read-ahead block 0 as we're almost
 987	 * certain to have the next operation be a read there.
 988	 */
 989	mode = xfs_ilock_map_shared(ip);
 990	if (ip->i_d.di_nextents > 0)
 991		xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
 992	xfs_iunlock(ip, mode);
 993	return 0;
 994}
 995
 996STATIC int
 997xfs_file_release(
 998	struct inode	*inode,
 999	struct file	*filp)
1000{
1001	return -xfs_release(XFS_I(inode));
1002}
1003
1004STATIC int
1005xfs_file_readdir(
1006	struct file	*filp,
1007	void		*dirent,
1008	filldir_t	filldir)
1009{
1010	struct inode	*inode = filp->f_path.dentry->d_inode;
1011	xfs_inode_t	*ip = XFS_I(inode);
1012	int		error;
1013	size_t		bufsize;
1014
1015	/*
1016	 * The Linux API doesn't pass down the total size of the buffer
1017	 * we read into down to the filesystem.  With the filldir concept
1018	 * it's not needed for correct information, but the XFS dir2 leaf
1019	 * code wants an estimate of the buffer size to calculate it's
1020	 * readahead window and size the buffers used for mapping to
1021	 * physical blocks.
1022	 *
1023	 * Try to give it an estimate that's good enough, maybe at some
1024	 * point we can change the ->readdir prototype to include the
1025	 * buffer size.  For now we use the current glibc buffer size.
1026	 */
1027	bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
 
 
 
 
 
 
 
 
 
 
 
1028
1029	error = xfs_readdir(ip, dirent, bufsize,
1030				(xfs_off_t *)&filp->f_pos, filldir);
1031	if (error)
1032		return -error;
1033	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1034}
1035
 
 
 
 
 
 
 
 
1036STATIC int
1037xfs_file_mmap(
1038	struct file	*filp,
1039	struct vm_area_struct *vma)
1040{
1041	vma->vm_ops = &xfs_file_vm_ops;
1042	vma->vm_flags |= VM_CAN_NONLINEAR;
 
 
 
 
1043
1044	file_accessed(filp);
 
 
 
1045	return 0;
1046}
1047
1048/*
1049 * mmap()d file has taken write protection fault and is being made
1050 * writable. We can set the page state up correctly for a writable
1051 * page, which means we can do correct delalloc accounting (ENOSPC
1052 * checking!) and unwritten extent mapping.
1053 */
1054STATIC int
1055xfs_vm_page_mkwrite(
1056	struct vm_area_struct	*vma,
1057	struct vm_fault		*vmf)
1058{
1059	return block_page_mkwrite(vma, vmf, xfs_get_blocks);
1060}
1061
1062const struct file_operations xfs_file_operations = {
1063	.llseek		= generic_file_llseek,
1064	.read		= do_sync_read,
1065	.write		= do_sync_write,
1066	.aio_read	= xfs_file_aio_read,
1067	.aio_write	= xfs_file_aio_write,
1068	.splice_read	= xfs_file_splice_read,
1069	.splice_write	= xfs_file_splice_write,
1070	.unlocked_ioctl	= xfs_file_ioctl,
1071#ifdef CONFIG_COMPAT
1072	.compat_ioctl	= xfs_file_compat_ioctl,
1073#endif
1074	.mmap		= xfs_file_mmap,
 
1075	.open		= xfs_file_open,
1076	.release	= xfs_file_release,
1077	.fsync		= xfs_file_fsync,
 
1078	.fallocate	= xfs_file_fallocate,
 
 
1079};
1080
1081const struct file_operations xfs_dir_file_operations = {
1082	.open		= xfs_dir_open,
1083	.read		= generic_read_dir,
1084	.readdir	= xfs_file_readdir,
1085	.llseek		= generic_file_llseek,
1086	.unlocked_ioctl	= xfs_file_ioctl,
1087#ifdef CONFIG_COMPAT
1088	.compat_ioctl	= xfs_file_compat_ioctl,
1089#endif
1090	.fsync		= xfs_file_fsync,
1091};
1092
1093static const struct vm_operations_struct xfs_file_vm_ops = {
1094	.fault		= filemap_fault,
1095	.page_mkwrite	= xfs_vm_page_mkwrite,
1096};