1 /*
2  * Copyright (c) 2000-2003 Silicon Graphics, Inc.  All Rights Reserved.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms of version 2 of the GNU General Public License as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it would be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
11  *
12  * Further, this software is distributed without any warranty that it is
13  * free of the rightful claim of any third person regarding infringement
14  * or the like.  Any license provided herein, whether implied or
15  * otherwise, applies only to this software file.  Patent licenses, if
16  * any, provided herein do not apply to combinations of this program with
17  * other software, or any other product whatsoever.
18  *
19  * You should have received a copy of the GNU General Public License along
20  * with this program; if not, write the Free Software Foundation, Inc., 59
21  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
22  *
23  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24  * Mountain View, CA  94043, or:
25  *
26  * http://www.sgi.com
27  *
28  * For further information regarding this notice, see:
29  *
30  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
31  */
32 /*
33  *  fs/xfs/linux/xfs_lrw.c (Linux Read Write stuff)
34  *
35  */
36 
37 #include "xfs.h"
38 
39 #include "xfs_fs.h"
40 #include "xfs_inum.h"
41 #include "xfs_log.h"
42 #include "xfs_trans.h"
43 #include "xfs_sb.h"
44 #include "xfs_ag.h"
45 #include "xfs_dir.h"
46 #include "xfs_dir2.h"
47 #include "xfs_alloc.h"
48 #include "xfs_dmapi.h"
49 #include "xfs_quota.h"
50 #include "xfs_mount.h"
51 #include "xfs_alloc_btree.h"
52 #include "xfs_bmap_btree.h"
53 #include "xfs_ialloc_btree.h"
54 #include "xfs_btree.h"
55 #include "xfs_ialloc.h"
56 #include "xfs_attr_sf.h"
57 #include "xfs_dir_sf.h"
58 #include "xfs_dir2_sf.h"
59 #include "xfs_dinode.h"
60 #include "xfs_inode.h"
61 #include "xfs_bmap.h"
62 #include "xfs_bit.h"
63 #include "xfs_rtalloc.h"
64 #include "xfs_error.h"
65 #include "xfs_itable.h"
66 #include "xfs_rw.h"
67 #include "xfs_refcache.h"
68 #include "xfs_acl.h"
69 #include "xfs_cap.h"
70 #include "xfs_mac.h"
71 #include "xfs_attr.h"
72 #include "xfs_inode_item.h"
73 #include "xfs_buf_item.h"
74 #include "xfs_utils.h"
75 #include "xfs_iomap.h"
76 
77 #include <linux/capability.h>
78 
79 
80 #if defined(XFS_RW_TRACE)
81 void
xfs_rw_enter_trace(int tag,xfs_iocore_t * io,const char * buf,size_t size,loff_t offset,int ioflags)82 xfs_rw_enter_trace(
83 	int		tag,
84 	xfs_iocore_t	*io,
85 	const char	*buf,
86 	size_t		size,
87 	loff_t		offset,
88 	int		ioflags)
89 {
90 	xfs_inode_t	*ip = XFS_IO_INODE(io);
91 
92 	if (ip->i_rwtrace == NULL)
93 		return;
94 	ktrace_enter(ip->i_rwtrace,
95 		(void *)(unsigned long)tag,
96 		(void *)ip,
97 		(void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
98 		(void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
99 		(void *)(__psint_t)buf,
100 		(void *)((unsigned long)size),
101 		(void *)((unsigned long)((offset >> 32) & 0xffffffff)),
102 		(void *)((unsigned long)(offset & 0xffffffff)),
103 		(void *)((unsigned long)ioflags),
104 		(void *)((unsigned long)((io->io_new_size >> 32) & 0xffffffff)),
105 		(void *)((unsigned long)(io->io_new_size & 0xffffffff)),
106 		(void *)NULL,
107 		(void *)NULL,
108 		(void *)NULL,
109 		(void *)NULL,
110 		(void *)NULL);
111 }
112 
113 void
xfs_inval_cached_trace(xfs_iocore_t * io,xfs_off_t offset,xfs_off_t len,xfs_off_t first,xfs_off_t last)114 xfs_inval_cached_trace(
115 	xfs_iocore_t	*io,
116 	xfs_off_t	offset,
117 	xfs_off_t	len,
118 	xfs_off_t	first,
119 	xfs_off_t	last)
120 {
121 	xfs_inode_t	*ip = XFS_IO_INODE(io);
122 
123 	if (ip->i_rwtrace == NULL)
124 		return;
125 	ktrace_enter(ip->i_rwtrace,
126 		(void *)(__psint_t)XFS_INVAL_CACHED,
127 		(void *)ip,
128 		(void *)((unsigned long)((offset >> 32) & 0xffffffff)),
129 		(void *)((unsigned long)(offset & 0xffffffff)),
130 		(void *)((unsigned long)((len >> 32) & 0xffffffff)),
131 		(void *)((unsigned long)(len & 0xffffffff)),
132 		(void *)((unsigned long)((first >> 32) & 0xffffffff)),
133 		(void *)((unsigned long)(first & 0xffffffff)),
134 		(void *)((unsigned long)((last >> 32) & 0xffffffff)),
135 		(void *)((unsigned long)(last & 0xffffffff)),
136 		(void *)NULL,
137 		(void *)NULL,
138 		(void *)NULL,
139 		(void *)NULL,
140 		(void *)NULL,
141 		(void *)NULL);
142 }
143 #endif
144 
145 /*
146  *	xfs_iozero
147  *
148  *	xfs_iozero clears the specified range of buffer supplied,
149  *	and marks all the affected blocks as valid and modified.  If
150  *	an affected block is not allocated, it will be allocated.  If
151  *	an affected block is not completely overwritten, and is not
152  *	valid before the operation, it will be read from disk before
153  *	being partially zeroed.
154  */
155 STATIC int
xfs_iozero(struct inode * ip,loff_t pos,size_t count,loff_t end_size)156 xfs_iozero(
157 	struct inode		*ip,	/* inode			*/
158 	loff_t			pos,	/* offset in file		*/
159 	size_t			count,	/* size of data to zero		*/
160 	loff_t			end_size)	/* max file size to set */
161 {
162 	unsigned		bytes;
163 	struct page		*page;
164 	struct address_space	*mapping;
165 	char			*kaddr;
166 	int			status;
167 
168 	mapping = ip->i_mapping;
169 	do {
170 		unsigned long index, offset;
171 
172 		offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
173 		index = pos >> PAGE_CACHE_SHIFT;
174 		bytes = PAGE_CACHE_SIZE - offset;
175 		if (bytes > count)
176 			bytes = count;
177 
178 		status = -ENOMEM;
179 		page = grab_cache_page(mapping, index);
180 		if (!page)
181 			break;
182 
183 		kaddr = kmap(page);
184 		status = mapping->a_ops->prepare_write(NULL, page, offset,
185 							offset + bytes);
186 		if (status) {
187 			goto unlock;
188 		}
189 
190 		memset((void *) (kaddr + offset), 0, bytes);
191 		flush_dcache_page(page);
192 		status = mapping->a_ops->commit_write(NULL, page, offset,
193 							offset + bytes);
194 		if (!status) {
195 			pos += bytes;
196 			count -= bytes;
197 			if (pos > i_size_read(ip))
198 				i_size_write(ip, pos < end_size ? pos : end_size);
199 		}
200 
201 unlock:
202 		kunmap(page);
203 		unlock_page(page);
204 		page_cache_release(page);
205 		if (status)
206 			break;
207 	} while (count);
208 
209 	return (-status);
210 }
211 
212 /*
213  * xfs_inval_cached_pages
214  *
215  * This routine is responsible for keeping direct I/O and buffered I/O
216  * somewhat coherent.  From here we make sure that we're at least
217  * temporarily holding the inode I/O lock exclusively and then call
218  * the page cache to flush and invalidate any cached pages.  If there
219  * are no cached pages this routine will be very quick.
220  */
221 void
xfs_inval_cached_pages(vnode_t * vp,xfs_iocore_t * io,xfs_off_t offset,int write,int relock)222 xfs_inval_cached_pages(
223 	vnode_t		*vp,
224 	xfs_iocore_t	*io,
225 	xfs_off_t	offset,
226 	int		write,
227 	int		relock)
228 {
229 	xfs_mount_t	*mp;
230 
231 	if (!VN_CACHED(vp)) {
232 		return;
233 	}
234 
235 	mp = io->io_mount;
236 
237 	/*
238 	 * We need to get the I/O lock exclusively in order
239 	 * to safely invalidate pages and mappings.
240 	 */
241 	if (relock) {
242 		XFS_IUNLOCK(mp, io, XFS_IOLOCK_SHARED);
243 		XFS_ILOCK(mp, io, XFS_IOLOCK_EXCL);
244 	}
245 
246 	/* Writing beyond EOF creates a hole that must be zeroed */
247 	if (write && (offset > XFS_SIZE(mp, io))) {
248 		xfs_fsize_t	isize;
249 
250 		XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
251 		isize = XFS_SIZE(mp, io);
252 		if (offset > isize) {
253 			xfs_zero_eof(vp, io, offset, isize, offset);
254 		}
255 		XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
256 	}
257 
258 	xfs_inval_cached_trace(io, offset, -1, ctooff(offtoct(offset)), -1);
259 	VOP_FLUSHINVAL_PAGES(vp, ctooff(offtoct(offset)), -1, FI_REMAPF_LOCKED);
260 	if (relock) {
261 		XFS_ILOCK_DEMOTE(mp, io, XFS_IOLOCK_EXCL);
262 	}
263 }
264 
265 ssize_t			/* bytes read, or (-)  error */
xfs_read(bhv_desc_t * bdp,struct file * file,char * buf,size_t size,loff_t * offset,int ioflags,cred_t * credp)266 xfs_read(
267 	bhv_desc_t      *bdp,
268 	struct file	*file,
269 	char		*buf,
270 	size_t		size,
271 	loff_t		*offset,
272 	int		ioflags,
273 	cred_t          *credp)
274 {
275 	ssize_t		ret;
276 	xfs_fsize_t	n;
277 	xfs_inode_t	*ip;
278 	xfs_mount_t	*mp;
279 
280 	ip = XFS_BHVTOI(bdp);
281 	mp = ip->i_mount;
282 
283 	XFS_STATS_INC(xs_read_calls);
284 
285 	if (unlikely(ioflags & IO_ISDIRECT)) {
286 		if ((ssize_t)size < 0)
287 			return -XFS_ERROR(EINVAL);
288 		if (((__psint_t)buf & BBMASK) ||
289 		    (*offset & mp->m_blockmask) ||
290 		    (size & mp->m_blockmask)) {
291 			if (*offset >= ip->i_d.di_size) {
292 				return (0);
293 			}
294 			return -XFS_ERROR(EINVAL);
295 		}
296 	}
297 
298 	n = XFS_MAXIOFFSET(mp) - *offset;
299 	if ((n <= 0) || (size == 0))
300 		return 0;
301 
302 	if (n < size)
303 		size = n;
304 
305 	if (XFS_FORCED_SHUTDOWN(mp)) {
306 		return -EIO;
307 	}
308 
309 	if (!(ioflags & IO_ISLOCKED))
310 		xfs_ilock(ip, XFS_IOLOCK_SHARED);
311 
312 	if (DM_EVENT_ENABLED(BHV_TO_VNODE(bdp)->v_vfsp, ip, DM_EVENT_READ) &&
313 	    !(ioflags & IO_INVIS)) {
314 		int error;
315 		vrwlock_t locktype = VRWLOCK_READ;
316 		int dmflags = FILP_DELAY_FLAG(file) | DM_SEM_FLAG_RD(ioflags);
317 
318 		error = XFS_SEND_DATA(mp, DM_EVENT_READ, BHV_TO_VNODE(bdp), *offset, size,
319 				      dmflags, &locktype);
320 		if (error) {
321 			if (!(ioflags & IO_ISLOCKED))
322 				xfs_iunlock(ip, XFS_IOLOCK_SHARED);
323 			return -error;
324 		}
325 	}
326 
327 	if (unlikely(ioflags & IO_ISDIRECT)) {
328 		xfs_rw_enter_trace(XFS_DIORD_ENTER, &ip->i_iocore,
329 					buf, size, *offset, ioflags);
330 		ret = (*offset < ip->i_d.di_size) ?
331 			do_generic_direct_read(file, buf, size, offset) : 0;
332 		UPDATE_ATIME(file->f_dentry->d_inode);
333 	} else {
334 		xfs_rw_enter_trace(XFS_READ_ENTER, &ip->i_iocore,
335 					buf, size, *offset, ioflags);
336 		ret = generic_file_read(file, buf, size, offset);
337 	}
338 
339 	if (ret > 0)
340 		XFS_STATS_ADD(xs_read_bytes, ret);
341 
342 	if (!(ioflags & IO_ISLOCKED))
343 		xfs_iunlock(ip, XFS_IOLOCK_SHARED);
344 
345 	if (unlikely(ioflags & IO_INVIS)) {
346 		/* generic_file_read updates the atime but we need to
347 		 * undo that because this I/O was supposed to be invisible.
348 		 */
349 		struct inode *inode = LINVFS_GET_IP(BHV_TO_VNODE(bdp));
350 		inode->i_atime = ip->i_d.di_atime.t_sec;
351 	} else {
352 		xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
353 	}
354 
355 	return ret;
356 }
357 
358 /*
359  * This routine is called to handle zeroing any space in the last
360  * block of the file that is beyond the EOF.  We do this since the
361  * size is being increased without writing anything to that block
362  * and we don't want anyone to read the garbage on the disk.
363  */
364 STATIC int				/* error (positive) */
xfs_zero_last_block(struct inode * ip,xfs_iocore_t * io,xfs_off_t offset,xfs_fsize_t isize,xfs_fsize_t end_size)365 xfs_zero_last_block(
366 	struct inode	*ip,
367 	xfs_iocore_t	*io,
368 	xfs_off_t	offset,
369 	xfs_fsize_t	isize,
370 	xfs_fsize_t	end_size)
371 {
372 	xfs_fileoff_t	last_fsb;
373 	xfs_mount_t	*mp;
374 	int		nimaps;
375 	int		zero_offset;
376 	int		zero_len;
377 	int		isize_fsb_offset;
378 	int		error = 0;
379 	xfs_bmbt_irec_t	imap;
380 	loff_t		loff;
381 	size_t		lsize;
382 
383 	ASSERT(ismrlocked(io->io_lock, MR_UPDATE) != 0);
384 	ASSERT(offset > isize);
385 
386 	mp = io->io_mount;
387 
388 	isize_fsb_offset = XFS_B_FSB_OFFSET(mp, isize);
389 	if (isize_fsb_offset == 0) {
390 		/*
391 		 * There are no extra bytes in the last block on disk to
392 		 * zero, so return.
393 		 */
394 		return 0;
395 	}
396 
397 	last_fsb = XFS_B_TO_FSBT(mp, isize);
398 	nimaps = 1;
399 	error = XFS_BMAPI(mp, NULL, io, last_fsb, 1, 0, NULL, 0, &imap,
400 			  &nimaps, NULL);
401 	if (error) {
402 		return error;
403 	}
404 	ASSERT(nimaps > 0);
405 	/*
406 	 * If the block underlying isize is just a hole, then there
407 	 * is nothing to zero.
408 	 */
409 	if (imap.br_startblock == HOLESTARTBLOCK) {
410 		return 0;
411 	}
412 	/*
413 	 * Zero the part of the last block beyond the EOF, and write it
414 	 * out sync.  We need to drop the ilock while we do this so we
415 	 * don't deadlock when the buffer cache calls back to us.
416 	 */
417 	XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL| XFS_EXTSIZE_RD);
418 	loff = XFS_FSB_TO_B(mp, last_fsb);
419 	lsize = XFS_FSB_TO_B(mp, 1);
420 
421 	zero_offset = isize_fsb_offset;
422 	zero_len = mp->m_sb.sb_blocksize - isize_fsb_offset;
423 
424 	error = xfs_iozero(ip, loff + zero_offset, zero_len, end_size);
425 
426 	XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
427 	ASSERT(error >= 0);
428 	return error;
429 }
430 
431 /*
432  * Zero any on disk space between the current EOF and the new,
433  * larger EOF.  This handles the normal case of zeroing the remainder
434  * of the last block in the file and the unusual case of zeroing blocks
435  * out beyond the size of the file.  This second case only happens
436  * with fixed size extents and when the system crashes before the inode
437  * size was updated but after blocks were allocated.  If fill is set,
438  * then any holes in the range are filled and zeroed.  If not, the holes
439  * are left alone as holes.
440  */
441 
442 int					/* error (positive) */
xfs_zero_eof(vnode_t * vp,xfs_iocore_t * io,xfs_off_t offset,xfs_fsize_t isize,xfs_fsize_t end_size)443 xfs_zero_eof(
444 	vnode_t		*vp,
445 	xfs_iocore_t	*io,
446 	xfs_off_t	offset,		/* starting I/O offset */
447 	xfs_fsize_t	isize,		/* current inode size */
448 	xfs_fsize_t	end_size)	/* terminal inode size */
449 {
450 	struct inode	*ip = LINVFS_GET_IP(vp);
451 	xfs_fileoff_t	start_zero_fsb;
452 	xfs_fileoff_t	end_zero_fsb;
453 	xfs_fileoff_t	prev_zero_fsb;
454 	xfs_fileoff_t	zero_count_fsb;
455 	xfs_fileoff_t	last_fsb;
456 	xfs_extlen_t	buf_len_fsb;
457 	xfs_extlen_t	prev_zero_count;
458 	xfs_mount_t	*mp;
459 	int		nimaps;
460 	int		error = 0;
461 	xfs_bmbt_irec_t	imap;
462 	loff_t		loff;
463 	size_t		lsize;
464 
465 	ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
466 	ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
467 
468 	mp = io->io_mount;
469 
470 	/*
471 	 * First handle zeroing the block on which isize resides.
472 	 * We only zero a part of that block so it is handled specially.
473 	 */
474 	error = xfs_zero_last_block(ip, io, offset, isize, end_size);
475 	if (error) {
476 		ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
477 		ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
478 		return error;
479 	}
480 
481 	/*
482 	 * Calculate the range between the new size and the old
483 	 * where blocks needing to be zeroed may exist.  To get the
484 	 * block where the last byte in the file currently resides,
485 	 * we need to subtract one from the size and truncate back
486 	 * to a block boundary.  We subtract 1 in case the size is
487 	 * exactly on a block boundary.
488 	 */
489 	last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
490 	start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
491 	end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
492 	ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
493 	if (last_fsb == end_zero_fsb) {
494 		/*
495 		 * The size was only incremented on its last block.
496 		 * We took care of that above, so just return.
497 		 */
498 		return 0;
499 	}
500 
501 	ASSERT(start_zero_fsb <= end_zero_fsb);
502 	prev_zero_fsb = NULLFILEOFF;
503 	prev_zero_count = 0;
504 	while (start_zero_fsb <= end_zero_fsb) {
505 		nimaps = 1;
506 		zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
507 		error = XFS_BMAPI(mp, NULL, io, start_zero_fsb, zero_count_fsb,
508 				  0, NULL, 0, &imap, &nimaps, NULL);
509 		if (error) {
510 			ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
511 			ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
512 			return error;
513 		}
514 		ASSERT(nimaps > 0);
515 
516 		if (imap.br_state == XFS_EXT_UNWRITTEN ||
517 		    imap.br_startblock == HOLESTARTBLOCK) {
518 			/*
519 			 * This loop handles initializing pages that were
520 			 * partially initialized by the code below this
521 			 * loop. It basically zeroes the part of the page
522 			 * that sits on a hole and sets the page as P_HOLE
523 			 * and calls remapf if it is a mapped file.
524 			 */
525 			prev_zero_fsb = NULLFILEOFF;
526 			prev_zero_count = 0;
527 			start_zero_fsb = imap.br_startoff +
528 					 imap.br_blockcount;
529 			ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
530 			continue;
531 		}
532 
533 		/*
534 		 * There are blocks in the range requested.
535 		 * Zero them a single write at a time.  We actually
536 		 * don't zero the entire range returned if it is
537 		 * too big and simply loop around to get the rest.
538 		 * That is not the most efficient thing to do, but it
539 		 * is simple and this path should not be exercised often.
540 		 */
541 		buf_len_fsb = XFS_FILBLKS_MIN(imap.br_blockcount,
542 					      mp->m_writeio_blocks << 8);
543 		/*
544 		 * Drop the inode lock while we're doing the I/O.
545 		 * We'll still have the iolock to protect us.
546 		 */
547 		XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
548 
549 		loff = XFS_FSB_TO_B(mp, start_zero_fsb);
550 		lsize = XFS_FSB_TO_B(mp, buf_len_fsb);
551 
552 		error = xfs_iozero(ip, loff, lsize, end_size);
553 
554 		if (error) {
555 			goto out_lock;
556 		}
557 
558 		prev_zero_fsb = start_zero_fsb;
559 		prev_zero_count = buf_len_fsb;
560 		start_zero_fsb = imap.br_startoff + buf_len_fsb;
561 		ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
562 
563 		XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
564 	}
565 
566 	return 0;
567 
568 out_lock:
569 
570 	XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
571 	ASSERT(error >= 0);
572 	return error;
573 }
574 
575 ssize_t				/* bytes written, or (-) error */
xfs_write(bhv_desc_t * bdp,struct file * file,const char * buf,size_t size,loff_t * offset,int ioflags,cred_t * credp)576 xfs_write(
577 	bhv_desc_t      *bdp,
578 	struct file	*file,
579 	const char	*buf,
580 	size_t		size,
581 	loff_t		*offset,
582 	int		ioflags,
583 	cred_t          *credp)
584 {
585 	xfs_inode_t	*xip;
586 	xfs_mount_t	*mp;
587 	ssize_t		ret;
588 	int		error = 0;
589 	xfs_fsize_t     isize, new_size;
590 	xfs_fsize_t	n, limit;
591 	xfs_iocore_t    *io;
592 	vnode_t		*vp;
593 	int		iolock;
594 	int		eventsent = 0;
595 	vrwlock_t	locktype;
596 
597 	XFS_STATS_INC(xs_write_calls);
598 
599 	vp = BHV_TO_VNODE(bdp);
600 	xip = XFS_BHVTOI(bdp);
601 
602 	if (size == 0)
603 		return 0;
604 
605 	io = &xip->i_iocore;
606 	mp = io->io_mount;
607 
608 	fs_check_frozen(vp->v_vfsp, SB_FREEZE_WRITE);
609 
610 	if (XFS_FORCED_SHUTDOWN(xip->i_mount)) {
611 		return -EIO;
612 	}
613 
614 	if (unlikely(ioflags & IO_ISDIRECT)) {
615 		if (((__psint_t)buf & BBMASK) ||
616 		    (*offset & mp->m_blockmask) ||
617 		    (size  & mp->m_blockmask)) {
618 			return XFS_ERROR(-EINVAL);
619 		}
620 		iolock = XFS_IOLOCK_SHARED;
621 		locktype = VRWLOCK_WRITE_DIRECT;
622 	} else {
623 		iolock = XFS_IOLOCK_EXCL;
624 		locktype = VRWLOCK_WRITE;
625 	}
626 
627 	if (ioflags & IO_ISLOCKED)
628 		iolock = 0;
629 
630 	xfs_ilock(xip, XFS_ILOCK_EXCL|iolock);
631 
632 	isize = xip->i_d.di_size;
633 	limit = XFS_MAXIOFFSET(mp);
634 
635 	if (file->f_flags & O_APPEND)
636 		*offset = isize;
637 
638 start:
639 	n = limit - *offset;
640 	if (n <= 0) {
641 		xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
642 		return -EFBIG;
643 	}
644 	if (n < size)
645 		size = n;
646 
647 	new_size = *offset + size;
648 	if (new_size > isize) {
649 		io->io_new_size = new_size;
650 	}
651 
652 	if ((DM_EVENT_ENABLED(vp->v_vfsp, xip, DM_EVENT_WRITE) &&
653 	    !(ioflags & IO_INVIS) && !eventsent)) {
654 		loff_t		savedsize = *offset;
655 		int dmflags = FILP_DELAY_FLAG(file) | DM_SEM_FLAG_RD(ioflags);
656 
657 		xfs_iunlock(xip, XFS_ILOCK_EXCL);
658 		error = XFS_SEND_DATA(xip->i_mount, DM_EVENT_WRITE, vp,
659 				      *offset, size,
660 				      dmflags, &locktype);
661 		if (error) {
662 			if (iolock) xfs_iunlock(xip, iolock);
663 			return -error;
664 		}
665 		xfs_ilock(xip, XFS_ILOCK_EXCL);
666 		eventsent = 1;
667 
668 		/*
669 		 * The iolock was dropped and reaquired in XFS_SEND_DATA
670 		 * so we have to recheck the size when appending.
671 		 * We will only "goto start;" once, since having sent the
672 		 * event prevents another call to XFS_SEND_DATA, which is
673 		 * what allows the size to change in the first place.
674 		 */
675 		if ((file->f_flags & O_APPEND) &&
676 		    savedsize != xip->i_d.di_size) {
677 			*offset = isize = xip->i_d.di_size;
678 			goto start;
679 		}
680 	}
681 
682 	/*
683 	 * If the offset is beyond the size of the file, we have a couple
684 	 * of things to do. First, if there is already space allocated
685 	 * we need to either create holes or zero the disk or ...
686 	 *
687 	 * If there is a page where the previous size lands, we need
688 	 * to zero it out up to the new size.
689 	 */
690 
691 	if (!(ioflags & IO_ISDIRECT) && (*offset > isize && isize)) {
692 		error = xfs_zero_eof(BHV_TO_VNODE(bdp), io, *offset,
693 			isize, *offset + size);
694 		if (error) {
695 			xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
696 			return(-error);
697 		}
698 	}
699 	xfs_iunlock(xip, XFS_ILOCK_EXCL);
700 
701 	/*
702 	 * If we're writing the file then make sure to clear the
703 	 * setuid and setgid bits if the process is not being run
704 	 * by root.  This keeps people from modifying setuid and
705 	 * setgid binaries.
706 	 */
707 
708 	if (((xip->i_d.di_mode & S_ISUID) ||
709 	    ((xip->i_d.di_mode & (S_ISGID | S_IXGRP)) ==
710 		(S_ISGID | S_IXGRP))) &&
711 	     !capable(CAP_FSETID)) {
712 		error = xfs_write_clear_setuid(xip);
713 		if (error) {
714 			xfs_iunlock(xip, iolock);
715 			return -error;
716 		}
717 	}
718 
719 
720 	if ((ssize_t) size < 0) {
721 		ret = -EINVAL;
722 		goto error;
723 	}
724 
725 	if (!access_ok(VERIFY_READ, buf, size)) {
726 		ret = -EINVAL;
727 		goto error;
728 	}
729 
730 retry:
731 	if (unlikely(ioflags & IO_ISDIRECT)) {
732 		xfs_inval_cached_pages(vp, io, *offset, 1, 1);
733 		xfs_rw_enter_trace(XFS_DIOWR_ENTER,
734 					io, buf, size, *offset, ioflags);
735 		ret = do_generic_direct_write(file, buf, size, offset);
736 	} else {
737 		xfs_rw_enter_trace(XFS_WRITE_ENTER,
738 					io, buf, size, *offset, ioflags);
739 		ret = do_generic_file_write(file, buf, size, offset);
740 	}
741 
742 	if (unlikely(ioflags & IO_INVIS)) {
743 		/* generic_file_write updates the mtime/ctime but we need
744 		 * to undo that because this I/O was supposed to be
745 		 * invisible.
746 		 */
747 		struct inode	*inode = LINVFS_GET_IP(vp);
748 		inode->i_mtime = xip->i_d.di_mtime.t_sec;
749 		inode->i_ctime = xip->i_d.di_ctime.t_sec;
750 	} else {
751 		xfs_ichgtime(xip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
752 	}
753 
754 	if ((ret == -ENOSPC) &&
755 	    DM_EVENT_ENABLED(vp->v_vfsp, xip, DM_EVENT_NOSPACE) &&
756 	    !(ioflags & IO_INVIS)) {
757 
758 		xfs_rwunlock(bdp, locktype);
759 		error = XFS_SEND_NAMESP(xip->i_mount, DM_EVENT_NOSPACE, vp,
760 				DM_RIGHT_NULL, vp, DM_RIGHT_NULL, NULL, NULL,
761 				0, 0, 0); /* Delay flag intentionally  unused */
762 		if (error)
763 			return -error;
764 		xfs_rwlock(bdp, locktype);
765 		*offset = xip->i_d.di_size;
766 		goto retry;
767 	}
768 
769 error:
770 	if (ret <= 0) {
771 		if (iolock)
772 			xfs_rwunlock(bdp, locktype);
773 		return ret;
774 	}
775 
776 	XFS_STATS_ADD(xs_write_bytes, ret);
777 
778 	if (*offset > xip->i_d.di_size) {
779 		xfs_ilock(xip, XFS_ILOCK_EXCL);
780 		if (*offset > xip->i_d.di_size) {
781 			struct inode	*inode = LINVFS_GET_IP(vp);
782 
783 			xip->i_d.di_size = *offset;
784 			i_size_write(inode, *offset);
785 			xip->i_update_core = 1;
786 			xip->i_update_size = 1;
787 			mark_inode_dirty_sync(inode);
788 		}
789 		xfs_iunlock(xip, XFS_ILOCK_EXCL);
790 	}
791 
792 	/* Handle various SYNC-type writes */
793 	if ((file->f_flags & O_SYNC) || IS_SYNC(file->f_dentry->d_inode)) {
794 
795 		/*
796 		 * If we're treating this as O_DSYNC and we have not updated the
797 		 * size, force the log.
798 		 */
799 
800 		if (!(mp->m_flags & XFS_MOUNT_OSYNCISOSYNC)
801 			&& !(xip->i_update_size)) {
802 			/*
803 			 * If an allocation transaction occurred
804 			 * without extending the size, then we have to force
805 			 * the log up the proper point to ensure that the
806 			 * allocation is permanent.  We can't count on
807 			 * the fact that buffered writes lock out direct I/O
808 			 * writes - the direct I/O write could have extended
809 			 * the size nontransactionally, then finished before
810 			 * we started.  xfs_write_file will think that the file
811 			 * didn't grow but the update isn't safe unless the
812 			 * size change is logged.
813 			 *
814 			 * Force the log if we've committed a transaction
815 			 * against the inode or if someone else has and
816 			 * the commit record hasn't gone to disk (e.g.
817 			 * the inode is pinned).  This guarantees that
818 			 * all changes affecting the inode are permanent
819 			 * when we return.
820 			 */
821 
822 			xfs_inode_log_item_t *iip;
823 			xfs_lsn_t lsn;
824 
825 			iip = xip->i_itemp;
826 			if (iip && iip->ili_last_lsn) {
827 				lsn = iip->ili_last_lsn;
828 				xfs_log_force(mp, lsn,
829 						XFS_LOG_FORCE | XFS_LOG_SYNC);
830 			} else if (xfs_ipincount(xip) > 0) {
831 				xfs_log_force(mp, (xfs_lsn_t)0,
832 						XFS_LOG_FORCE | XFS_LOG_SYNC);
833 			}
834 
835 		} else {
836 			xfs_trans_t	*tp;
837 
838 			/*
839 			 * O_SYNC or O_DSYNC _with_ a size update are handled
840 			 * the same way.
841 			 *
842 			 * If the write was synchronous then we need to make
843 			 * sure that the inode modification time is permanent.
844 			 * We'll have updated the timestamp above, so here
845 			 * we use a synchronous transaction to log the inode.
846 			 * It's not fast, but it's necessary.
847 			 *
848 			 * If this a dsync write and the size got changed
849 			 * non-transactionally, then we need to ensure that
850 			 * the size change gets logged in a synchronous
851 			 * transaction.
852 			 */
853 
854 			tp = xfs_trans_alloc(mp, XFS_TRANS_WRITE_SYNC);
855 			if ((error = xfs_trans_reserve(tp, 0,
856 						      XFS_SWRITE_LOG_RES(mp),
857 						      0, 0, 0))) {
858 				/* Transaction reserve failed */
859 				xfs_trans_cancel(tp, 0);
860 			} else {
861 				/* Transaction reserve successful */
862 				xfs_ilock(xip, XFS_ILOCK_EXCL);
863 				xfs_trans_ijoin(tp, xip, XFS_ILOCK_EXCL);
864 				xfs_trans_ihold(tp, xip);
865 				xfs_trans_log_inode(tp, xip, XFS_ILOG_CORE);
866 				xfs_trans_set_sync(tp);
867 				error = xfs_trans_commit(tp, 0, NULL);
868 				xfs_iunlock(xip, XFS_ILOCK_EXCL);
869 			}
870 		}
871 	} /* (ioflags & O_SYNC) */
872 
873 	/*
874 	 * If we are coming from an nfsd thread then insert into the
875 	 * reference cache.
876 	 */
877 
878 	if (!strcmp(current->comm, "nfsd"))
879 		xfs_refcache_insert(xip);
880 
881 	/* Drop lock this way - the old refcache release is in here */
882 	if (iolock)
883 		xfs_rwunlock(bdp, locktype);
884 
885 	return(ret);
886 }
887 
888 /*
889  * All xfs metadata buffers except log state machine buffers
890  * get this attached as their b_bdstrat callback function.
891  * This is so that we can catch a buffer
892  * after prematurely unpinning it to forcibly shutdown the filesystem.
893  */
894 int
xfs_bdstrat_cb(struct xfs_buf * bp)895 xfs_bdstrat_cb(struct xfs_buf *bp)
896 {
897 	xfs_mount_t	*mp;
898 
899 	mp = XFS_BUF_FSPRIVATE3(bp, xfs_mount_t *);
900 	if (!XFS_FORCED_SHUTDOWN(mp)) {
901 		pagebuf_iorequest(bp);
902 		return 0;
903 	} else {
904 		xfs_buftrace("XFS__BDSTRAT IOERROR", bp);
905 		/*
906 		 * Metadata write that didn't get logged but
907 		 * written delayed anyway. These aren't associated
908 		 * with a transaction, and can be ignored.
909 		 */
910 		if (XFS_BUF_IODONE_FUNC(bp) == NULL &&
911 		    (XFS_BUF_ISREAD(bp)) == 0)
912 			return (xfs_bioerror_relse(bp));
913 		else
914 			return (xfs_bioerror(bp));
915 	}
916 }
917 
918 
919 int
xfs_bmap(bhv_desc_t * bdp,xfs_off_t offset,ssize_t count,int flags,xfs_iomap_t * iomapp,int * niomaps)920 xfs_bmap(bhv_desc_t	*bdp,
921 	xfs_off_t	offset,
922 	ssize_t		count,
923 	int		flags,
924 	xfs_iomap_t	*iomapp,
925 	int		*niomaps)
926 {
927 	xfs_inode_t	*ip = XFS_BHVTOI(bdp);
928 	xfs_iocore_t	*io = &ip->i_iocore;
929 
930 	ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
931 	ASSERT(((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != 0) ==
932 	       ((ip->i_iocore.io_flags & XFS_IOCORE_RT) != 0));
933 
934 	return xfs_iomap(io, offset, count, flags, iomapp, niomaps);
935 }
936 
937 /*
938  * Wrapper around bdstrat so that we can stop data
939  * from going to disk in case we are shutting down the filesystem.
940  * Typically user data goes thru this path; one of the exceptions
941  * is the superblock.
942  */
943 int
xfsbdstrat(struct xfs_mount * mp,struct xfs_buf * bp)944 xfsbdstrat(
945 	struct xfs_mount	*mp,
946 	struct xfs_buf		*bp)
947 {
948 	ASSERT(mp);
949 	if (!XFS_FORCED_SHUTDOWN(mp)) {
950 		/* Grio redirection would go here
951 		 * if (XFS_BUF_IS_GRIO(bp)) {
952 		 */
953 
954 		pagebuf_iorequest(bp);
955 		return 0;
956 	}
957 
958 	xfs_buftrace("XFSBDSTRAT IOERROR", bp);
959 	return (xfs_bioerror_relse(bp));
960 }
961 
962 /*
963  * If the underlying (data/log/rt) device is readonly, there are some
964  * operations that cannot proceed.
965  */
966 int
xfs_dev_is_read_only(xfs_mount_t * mp,char * message)967 xfs_dev_is_read_only(
968 	xfs_mount_t		*mp,
969 	char			*message)
970 {
971 	if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
972 	    xfs_readonly_buftarg(mp->m_logdev_targp) ||
973 	    (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
974 		cmn_err(CE_NOTE,
975 			"XFS: %s required on read-only device.", message);
976 		cmn_err(CE_NOTE,
977 			"XFS: write access unavailable, cannot proceed.");
978 		return EROFS;
979 	}
980 	return 0;
981 }
982