xfs_buf.c 44.7 KB
Newer Older
L
Linus Torvalds 已提交
1
/*
2
 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3
 * All Rights Reserved.
L
Linus Torvalds 已提交
4
 *
5 6
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License as
L
Linus Torvalds 已提交
7 8
 * published by the Free Software Foundation.
 *
9 10 11 12
 * This program is distributed in the hope that it would be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
L
Linus Torvalds 已提交
13
 *
14 15 16
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write the Free Software Foundation,
 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
L
Linus Torvalds 已提交
17
 */
18
#include "xfs.h"
L
Linus Torvalds 已提交
19 20
#include <linux/stddef.h>
#include <linux/errno.h>
21
#include <linux/gfp.h>
L
Linus Torvalds 已提交
22 23 24 25 26 27 28 29 30 31
#include <linux/pagemap.h>
#include <linux/init.h>
#include <linux/vmalloc.h>
#include <linux/bio.h>
#include <linux/sysctl.h>
#include <linux/proc_fs.h>
#include <linux/workqueue.h>
#include <linux/percpu.h>
#include <linux/blkdev.h>
#include <linux/hash.h>
32
#include <linux/kthread.h>
C
Christoph Lameter 已提交
33
#include <linux/migrate.h>
34
#include <linux/backing-dev.h>
35
#include <linux/freezer.h>
L
Linus Torvalds 已提交
36

37
#include "xfs_log_format.h"
38
#include "xfs_trans_resv.h"
39
#include "xfs_sb.h"
40 41
#include "xfs_ag.h"
#include "xfs_mount.h"
C
Christoph Hellwig 已提交
42
#include "xfs_trace.h"
43
#include "xfs_log.h"
44

45
static kmem_zone_t *xfs_buf_zone;
46

47
static struct workqueue_struct *xfslogd_workqueue;
L
Linus Torvalds 已提交
48

49 50 51 52
#ifdef XFS_BUF_LOCK_TRACKING
# define XB_SET_OWNER(bp)	((bp)->b_last_holder = current->pid)
# define XB_CLEAR_OWNER(bp)	((bp)->b_last_holder = -1)
# define XB_GET_OWNER(bp)	((bp)->b_last_holder)
L
Linus Torvalds 已提交
53
#else
54 55 56
# define XB_SET_OWNER(bp)	do { } while (0)
# define XB_CLEAR_OWNER(bp)	do { } while (0)
# define XB_GET_OWNER(bp)	do { } while (0)
L
Linus Torvalds 已提交
57 58
#endif

59
#define xb_to_gfp(flags) \
D
Dave Chinner 已提交
60
	((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : GFP_NOFS) | __GFP_NOWARN)
L
Linus Torvalds 已提交
61 62


63 64 65 66 67 68 69
static inline int
xfs_buf_is_vmapped(
	struct xfs_buf	*bp)
{
	/*
	 * Return true if the buffer is vmapped.
	 *
70 71 72
	 * b_addr is null if the buffer is not mapped, but the code is clever
	 * enough to know it doesn't have to map a single page, so the check has
	 * to be both for b_addr and bp->b_page_count > 1.
73
	 */
74
	return bp->b_addr && bp->b_page_count > 1;
75 76 77 78 79 80 81 82 83
}

static inline int
xfs_buf_vmap_len(
	struct xfs_buf	*bp)
{
	return (bp->b_page_count * PAGE_SIZE) - bp->b_offset;
}

84 85 86 87 88 89 90 91 92 93 94 95
/*
 * When we mark a buffer stale, we remove the buffer from the LRU and clear the
 * b_lru_ref count so that the buffer is freed immediately when the buffer
 * reference count falls to zero. If the buffer is already on the LRU, we need
 * to remove the reference that LRU holds on the buffer.
 *
 * This prevents build-up of stale buffers on the LRU.
 */
void
xfs_buf_stale(
	struct xfs_buf	*bp)
{
96 97
	ASSERT(xfs_buf_islocked(bp));

98
	bp->b_flags |= XBF_STALE;
99 100 101 102 103 104 105 106

	/*
	 * Clear the delwri status so that a delwri queue walker will not
	 * flush this buffer to disk now that it is stale. The delwri queue has
	 * a reference to the buffer, so this is safe to do.
	 */
	bp->b_flags &= ~_XBF_DELWRI_Q;

107 108 109
	spin_lock(&bp->b_lock);
	atomic_set(&bp->b_lru_ref, 0);
	if (!(bp->b_state & XFS_BSTATE_DISPOSE) &&
110 111 112
	    (list_lru_del(&bp->b_target->bt_lru, &bp->b_lru)))
		atomic_dec(&bp->b_hold);

113
	ASSERT(atomic_read(&bp->b_hold) >= 1);
114
	spin_unlock(&bp->b_lock);
115
}
L
Linus Torvalds 已提交
116

117 118 119 120 121 122 123 124 125
static int
xfs_buf_get_maps(
	struct xfs_buf		*bp,
	int			map_count)
{
	ASSERT(bp->b_maps == NULL);
	bp->b_map_count = map_count;

	if (map_count == 1) {
126
		bp->b_maps = &bp->__b_map;
127 128 129 130 131 132
		return 0;
	}

	bp->b_maps = kmem_zalloc(map_count * sizeof(struct xfs_buf_map),
				KM_NOFS);
	if (!bp->b_maps)
D
Dave Chinner 已提交
133
		return -ENOMEM;
134 135 136 137 138 139 140 141 142 143
	return 0;
}

/*
 *	Frees b_pages if it was allocated.
 */
static void
xfs_buf_free_maps(
	struct xfs_buf	*bp)
{
144
	if (bp->b_maps != &bp->__b_map) {
145 146 147 148 149
		kmem_free(bp->b_maps);
		bp->b_maps = NULL;
	}
}

150
struct xfs_buf *
151
_xfs_buf_alloc(
152
	struct xfs_buftarg	*target,
153 154
	struct xfs_buf_map	*map,
	int			nmaps,
155
	xfs_buf_flags_t		flags)
L
Linus Torvalds 已提交
156
{
157
	struct xfs_buf		*bp;
158 159
	int			error;
	int			i;
160

D
Dave Chinner 已提交
161
	bp = kmem_zone_zalloc(xfs_buf_zone, KM_NOFS);
162 163 164
	if (unlikely(!bp))
		return NULL;

L
Linus Torvalds 已提交
165
	/*
166 167
	 * We don't want certain flags to appear in b_flags unless they are
	 * specifically set by later operations on the buffer.
L
Linus Torvalds 已提交
168
	 */
169
	flags &= ~(XBF_UNMAPPED | XBF_TRYLOCK | XBF_ASYNC | XBF_READ_AHEAD);
170 171

	atomic_set(&bp->b_hold, 1);
172
	atomic_set(&bp->b_lru_ref, 1);
173
	init_completion(&bp->b_iowait);
174
	INIT_LIST_HEAD(&bp->b_lru);
175
	INIT_LIST_HEAD(&bp->b_list);
176
	RB_CLEAR_NODE(&bp->b_rbnode);
T
Thomas Gleixner 已提交
177
	sema_init(&bp->b_sema, 0); /* held, no waiters */
178
	spin_lock_init(&bp->b_lock);
179 180
	XB_SET_OWNER(bp);
	bp->b_target = target;
181
	bp->b_flags = flags;
D
Dave Chinner 已提交
182

L
Linus Torvalds 已提交
183
	/*
184 185
	 * Set length and io_length to the same value initially.
	 * I/O routines should use io_length, which will be the same in
L
Linus Torvalds 已提交
186 187
	 * most cases but may be reset (e.g. XFS recovery).
	 */
188 189 190 191 192 193 194 195 196 197 198 199 200 201 202
	error = xfs_buf_get_maps(bp, nmaps);
	if (error)  {
		kmem_zone_free(xfs_buf_zone, bp);
		return NULL;
	}

	bp->b_bn = map[0].bm_bn;
	bp->b_length = 0;
	for (i = 0; i < nmaps; i++) {
		bp->b_maps[i].bm_bn = map[i].bm_bn;
		bp->b_maps[i].bm_len = map[i].bm_len;
		bp->b_length += map[i].bm_len;
	}
	bp->b_io_length = bp->b_length;

203 204 205 206
	atomic_set(&bp->b_pin_count, 0);
	init_waitqueue_head(&bp->b_waiters);

	XFS_STATS_INC(xb_create);
C
Christoph Hellwig 已提交
207
	trace_xfs_buf_init(bp, _RET_IP_);
208 209

	return bp;
L
Linus Torvalds 已提交
210 211 212
}

/*
213 214
 *	Allocate a page array capable of holding a specified number
 *	of pages, and point the page buf at it.
L
Linus Torvalds 已提交
215 216
 */
STATIC int
217 218
_xfs_buf_get_pages(
	xfs_buf_t		*bp,
219
	int			page_count)
L
Linus Torvalds 已提交
220 221
{
	/* Make sure that we have a page list */
222 223 224 225
	if (bp->b_pages == NULL) {
		bp->b_page_count = page_count;
		if (page_count <= XB_PAGES) {
			bp->b_pages = bp->b_page_array;
L
Linus Torvalds 已提交
226
		} else {
227
			bp->b_pages = kmem_alloc(sizeof(struct page *) *
D
Dave Chinner 已提交
228
						 page_count, KM_NOFS);
229
			if (bp->b_pages == NULL)
L
Linus Torvalds 已提交
230 231
				return -ENOMEM;
		}
232
		memset(bp->b_pages, 0, sizeof(struct page *) * page_count);
L
Linus Torvalds 已提交
233 234 235 236 237
	}
	return 0;
}

/*
238
 *	Frees b_pages if it was allocated.
L
Linus Torvalds 已提交
239 240
 */
STATIC void
241
_xfs_buf_free_pages(
L
Linus Torvalds 已提交
242 243
	xfs_buf_t	*bp)
{
244
	if (bp->b_pages != bp->b_page_array) {
245
		kmem_free(bp->b_pages);
246
		bp->b_pages = NULL;
L
Linus Torvalds 已提交
247 248 249 250 251 252 253
	}
}

/*
 *	Releases the specified buffer.
 *
 * 	The modification state of any associated pages is left unchanged.
254
 * 	The buffer must not be on any hash - use xfs_buf_rele instead for
L
Linus Torvalds 已提交
255 256 257
 * 	hashed and refcounted buffers
 */
void
258
xfs_buf_free(
L
Linus Torvalds 已提交
259 260
	xfs_buf_t		*bp)
{
C
Christoph Hellwig 已提交
261
	trace_xfs_buf_free(bp, _RET_IP_);
L
Linus Torvalds 已提交
262

263 264
	ASSERT(list_empty(&bp->b_lru));

265
	if (bp->b_flags & _XBF_PAGES) {
L
Linus Torvalds 已提交
266 267
		uint		i;

268
		if (xfs_buf_is_vmapped(bp))
A
Alex Elder 已提交
269 270
			vm_unmap_ram(bp->b_addr - bp->b_offset,
					bp->b_page_count);
L
Linus Torvalds 已提交
271

272 273 274
		for (i = 0; i < bp->b_page_count; i++) {
			struct page	*page = bp->b_pages[i];

275
			__free_page(page);
276
		}
277 278
	} else if (bp->b_flags & _XBF_KMEM)
		kmem_free(bp->b_addr);
279
	_xfs_buf_free_pages(bp);
280
	xfs_buf_free_maps(bp);
281
	kmem_zone_free(xfs_buf_zone, bp);
L
Linus Torvalds 已提交
282 283 284
}

/*
285
 * Allocates all the pages for buffer in question and builds it's page list.
L
Linus Torvalds 已提交
286 287
 */
STATIC int
288
xfs_buf_allocate_memory(
L
Linus Torvalds 已提交
289 290 291
	xfs_buf_t		*bp,
	uint			flags)
{
292
	size_t			size;
L
Linus Torvalds 已提交
293
	size_t			nbytes, offset;
294
	gfp_t			gfp_mask = xb_to_gfp(flags);
L
Linus Torvalds 已提交
295
	unsigned short		page_count, i;
D
Dave Chinner 已提交
296
	xfs_off_t		start, end;
L
Linus Torvalds 已提交
297 298
	int			error;

299 300 301 302 303
	/*
	 * for buffers that are contained within a single page, just allocate
	 * the memory from the heap - there's no need for the complexity of
	 * page arrays to keep allocation down to order 0.
	 */
D
Dave Chinner 已提交
304 305
	size = BBTOB(bp->b_length);
	if (size < PAGE_SIZE) {
D
Dave Chinner 已提交
306
		bp->b_addr = kmem_alloc(size, KM_NOFS);
307 308 309 310 311
		if (!bp->b_addr) {
			/* low memory - use alloc_page loop instead */
			goto use_alloc_page;
		}

D
Dave Chinner 已提交
312
		if (((unsigned long)(bp->b_addr + size - 1) & PAGE_MASK) !=
313 314 315 316 317 318 319 320 321 322
		    ((unsigned long)bp->b_addr & PAGE_MASK)) {
			/* b_addr spans two pages - use alloc_page instead */
			kmem_free(bp->b_addr);
			bp->b_addr = NULL;
			goto use_alloc_page;
		}
		bp->b_offset = offset_in_page(bp->b_addr);
		bp->b_pages = bp->b_page_array;
		bp->b_pages[0] = virt_to_page(bp->b_addr);
		bp->b_page_count = 1;
323
		bp->b_flags |= _XBF_KMEM;
324 325 326 327
		return 0;
	}

use_alloc_page:
328 329
	start = BBTOB(bp->b_maps[0].bm_bn) >> PAGE_SHIFT;
	end = (BBTOB(bp->b_maps[0].bm_bn + bp->b_length) + PAGE_SIZE - 1)
330
								>> PAGE_SHIFT;
D
Dave Chinner 已提交
331
	page_count = end - start;
332
	error = _xfs_buf_get_pages(bp, page_count);
L
Linus Torvalds 已提交
333 334 335
	if (unlikely(error))
		return error;

336
	offset = bp->b_offset;
337
	bp->b_flags |= _XBF_PAGES;
L
Linus Torvalds 已提交
338

339
	for (i = 0; i < bp->b_page_count; i++) {
L
Linus Torvalds 已提交
340 341
		struct page	*page;
		uint		retries = 0;
342 343
retry:
		page = alloc_page(gfp_mask);
L
Linus Torvalds 已提交
344
		if (unlikely(page == NULL)) {
345 346
			if (flags & XBF_READ_AHEAD) {
				bp->b_page_count = i;
D
Dave Chinner 已提交
347
				error = -ENOMEM;
348
				goto out_free_pages;
L
Linus Torvalds 已提交
349 350 351 352 353 354 355 356 357
			}

			/*
			 * This could deadlock.
			 *
			 * But until all the XFS lowlevel code is revamped to
			 * handle buffer allocation failures we can't do much.
			 */
			if (!(++retries % 100))
358 359
				xfs_err(NULL,
		"possible memory allocation deadlock in %s (mode:0x%x)",
360
					__func__, gfp_mask);
L
Linus Torvalds 已提交
361

362
			XFS_STATS_INC(xb_page_retries);
363
			congestion_wait(BLK_RW_ASYNC, HZ/50);
L
Linus Torvalds 已提交
364 365 366
			goto retry;
		}

367
		XFS_STATS_INC(xb_page_found);
L
Linus Torvalds 已提交
368

369
		nbytes = min_t(size_t, size, PAGE_SIZE - offset);
L
Linus Torvalds 已提交
370
		size -= nbytes;
371
		bp->b_pages[i] = page;
L
Linus Torvalds 已提交
372 373
		offset = 0;
	}
374
	return 0;
L
Linus Torvalds 已提交
375

376 377 378
out_free_pages:
	for (i = 0; i < bp->b_page_count; i++)
		__free_page(bp->b_pages[i]);
L
Linus Torvalds 已提交
379 380 381 382
	return error;
}

/*
L
Lucas De Marchi 已提交
383
 *	Map buffer into kernel address-space if necessary.
L
Linus Torvalds 已提交
384 385
 */
STATIC int
386
_xfs_buf_map_pages(
L
Linus Torvalds 已提交
387 388 389
	xfs_buf_t		*bp,
	uint			flags)
{
390
	ASSERT(bp->b_flags & _XBF_PAGES);
391
	if (bp->b_page_count == 1) {
392
		/* A single page buffer is always mappable */
393
		bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset;
394 395 396
	} else if (flags & XBF_UNMAPPED) {
		bp->b_addr = NULL;
	} else {
397
		int retried = 0;
398 399 400 401 402 403 404 405 406 407 408
		unsigned noio_flag;

		/*
		 * vm_map_ram() will allocate auxillary structures (e.g.
		 * pagetables) with GFP_KERNEL, yet we are likely to be under
		 * GFP_NOFS context here. Hence we need to tell memory reclaim
		 * that we are in such a context via PF_MEMALLOC_NOIO to prevent
		 * memory reclaim re-entering the filesystem here and
		 * potentially deadlocking.
		 */
		noio_flag = memalloc_noio_save();
409 410 411 412 413 414 415
		do {
			bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count,
						-1, PAGE_KERNEL);
			if (bp->b_addr)
				break;
			vm_unmap_aliases();
		} while (retried++ <= 1);
416
		memalloc_noio_restore(noio_flag);
417 418

		if (!bp->b_addr)
L
Linus Torvalds 已提交
419
			return -ENOMEM;
420
		bp->b_addr += bp->b_offset;
L
Linus Torvalds 已提交
421 422 423 424 425 426 427 428 429 430
	}

	return 0;
}

/*
 *	Finding and Reading Buffers
 */

/*
431
 *	Look up, and creates if absent, a lockable buffer for
L
Linus Torvalds 已提交
432
 *	a given range of an inode.  The buffer is returned
433
 *	locked.	No I/O is implied by this call.
L
Linus Torvalds 已提交
434 435
 */
xfs_buf_t *
436
_xfs_buf_find(
437
	struct xfs_buftarg	*btp,
438 439
	struct xfs_buf_map	*map,
	int			nmaps,
440 441
	xfs_buf_flags_t		flags,
	xfs_buf_t		*new_bp)
L
Linus Torvalds 已提交
442
{
443
	size_t			numbytes;
444 445 446 447
	struct xfs_perag	*pag;
	struct rb_node		**rbp;
	struct rb_node		*parent;
	xfs_buf_t		*bp;
448
	xfs_daddr_t		blkno = map[0].bm_bn;
449
	xfs_daddr_t		eofs;
450 451
	int			numblks = 0;
	int			i;
L
Linus Torvalds 已提交
452

453 454
	for (i = 0; i < nmaps; i++)
		numblks += map[i].bm_len;
455
	numbytes = BBTOB(numblks);
L
Linus Torvalds 已提交
456 457

	/* Check for IOs smaller than the sector size / not sector aligned */
458 459
	ASSERT(!(numbytes < btp->bt_meta_sectorsize));
	ASSERT(!(BBTOB(blkno) & (xfs_off_t)btp->bt_meta_sectormask));
L
Linus Torvalds 已提交
460

461 462 463 464 465 466 467
	/*
	 * Corrupted block numbers can get through to here, unfortunately, so we
	 * have to check that the buffer falls within the filesystem bounds.
	 */
	eofs = XFS_FSB_TO_BB(btp->bt_mount, btp->bt_mount->m_sb.sb_dblocks);
	if (blkno >= eofs) {
		/*
D
Dave Chinner 已提交
468
		 * XXX (dgc): we should really be returning -EFSCORRUPTED here,
469 470 471 472 473 474
		 * but none of the higher level infrastructure supports
		 * returning a specific error on buffer lookup failures.
		 */
		xfs_alert(btp->bt_mount,
			  "%s: Block out of range: block 0x%llx, EOFS 0x%llx ",
			  __func__, blkno, eofs);
D
Dave Chinner 已提交
475
		WARN_ON(1);
476 477 478
		return NULL;
	}

479 480
	/* get tree root */
	pag = xfs_perag_get(btp->bt_mount,
481
				xfs_daddr_to_agno(btp->bt_mount, blkno));
482 483 484 485 486 487 488 489 490 491

	/* walk tree */
	spin_lock(&pag->pag_buf_lock);
	rbp = &pag->pag_buf_tree.rb_node;
	parent = NULL;
	bp = NULL;
	while (*rbp) {
		parent = *rbp;
		bp = rb_entry(parent, struct xfs_buf, b_rbnode);

D
Dave Chinner 已提交
492
		if (blkno < bp->b_bn)
493
			rbp = &(*rbp)->rb_left;
D
Dave Chinner 已提交
494
		else if (blkno > bp->b_bn)
495 496 497
			rbp = &(*rbp)->rb_right;
		else {
			/*
D
Dave Chinner 已提交
498
			 * found a block number match. If the range doesn't
499 500 501 502 503 504
			 * match, the only way this is allowed is if the buffer
			 * in the cache is stale and the transaction that made
			 * it stale has not yet committed. i.e. we are
			 * reallocating a busy extent. Skip this buffer and
			 * continue searching to the right for an exact match.
			 */
505
			if (bp->b_length != numblks) {
506 507 508 509
				ASSERT(bp->b_flags & XBF_STALE);
				rbp = &(*rbp)->rb_right;
				continue;
			}
510
			atomic_inc(&bp->b_hold);
L
Linus Torvalds 已提交
511 512 513 514 515
			goto found;
		}
	}

	/* No match found */
516
	if (new_bp) {
517 518 519 520 521
		rb_link_node(&new_bp->b_rbnode, parent, rbp);
		rb_insert_color(&new_bp->b_rbnode, &pag->pag_buf_tree);
		/* the buffer keeps the perag reference until it is freed */
		new_bp->b_pag = pag;
		spin_unlock(&pag->pag_buf_lock);
L
Linus Torvalds 已提交
522
	} else {
523
		XFS_STATS_INC(xb_miss_locked);
524 525
		spin_unlock(&pag->pag_buf_lock);
		xfs_perag_put(pag);
L
Linus Torvalds 已提交
526
	}
527
	return new_bp;
L
Linus Torvalds 已提交
528 529

found:
530 531
	spin_unlock(&pag->pag_buf_lock);
	xfs_perag_put(pag);
L
Linus Torvalds 已提交
532

533 534
	if (!xfs_buf_trylock(bp)) {
		if (flags & XBF_TRYLOCK) {
535 536 537
			xfs_buf_rele(bp);
			XFS_STATS_INC(xb_busy_locked);
			return NULL;
L
Linus Torvalds 已提交
538
		}
539 540
		xfs_buf_lock(bp);
		XFS_STATS_INC(xb_get_locked_waited);
L
Linus Torvalds 已提交
541 542
	}

543 544 545 546 547
	/*
	 * if the buffer is stale, clear all the external state associated with
	 * it. We need to keep flags such as how we allocated the buffer memory
	 * intact here.
	 */
548 549
	if (bp->b_flags & XBF_STALE) {
		ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0);
D
Dave Chinner 已提交
550
		ASSERT(bp->b_iodone == NULL);
551
		bp->b_flags &= _XBF_KMEM | _XBF_PAGES;
552
		bp->b_ops = NULL;
553
	}
C
Christoph Hellwig 已提交
554 555

	trace_xfs_buf_find(bp, flags, _RET_IP_);
556 557
	XFS_STATS_INC(xb_get_locked);
	return bp;
L
Linus Torvalds 已提交
558 559 560
}

/*
561 562 563
 * Assembles a buffer covering the specified range. The code is optimised for
 * cache hits, as metadata intensive workloads will see 3 orders of magnitude
 * more hits than misses.
L
Linus Torvalds 已提交
564
 */
565
struct xfs_buf *
566 567 568 569
xfs_buf_get_map(
	struct xfs_buftarg	*target,
	struct xfs_buf_map	*map,
	int			nmaps,
570
	xfs_buf_flags_t		flags)
L
Linus Torvalds 已提交
571
{
572 573
	struct xfs_buf		*bp;
	struct xfs_buf		*new_bp;
574
	int			error = 0;
L
Linus Torvalds 已提交
575

576
	bp = _xfs_buf_find(target, map, nmaps, flags, NULL);
577 578 579
	if (likely(bp))
		goto found;

580
	new_bp = _xfs_buf_alloc(target, map, nmaps, flags);
581
	if (unlikely(!new_bp))
L
Linus Torvalds 已提交
582 583
		return NULL;

584 585
	error = xfs_buf_allocate_memory(new_bp, flags);
	if (error) {
586
		xfs_buf_free(new_bp);
587 588 589
		return NULL;
	}

590
	bp = _xfs_buf_find(target, map, nmaps, flags, new_bp);
591
	if (!bp) {
592
		xfs_buf_free(new_bp);
593 594 595
		return NULL;
	}

596 597
	if (bp != new_bp)
		xfs_buf_free(new_bp);
L
Linus Torvalds 已提交
598

599
found:
600
	if (!bp->b_addr) {
601
		error = _xfs_buf_map_pages(bp, flags);
L
Linus Torvalds 已提交
602
		if (unlikely(error)) {
603
			xfs_warn(target->bt_mount,
604
				"%s: failed to map pagesn", __func__);
D
Dave Chinner 已提交
605 606
			xfs_buf_relse(bp);
			return NULL;
L
Linus Torvalds 已提交
607 608 609
		}
	}

610
	XFS_STATS_INC(xb_get);
C
Christoph Hellwig 已提交
611
	trace_xfs_buf_get(bp, flags, _RET_IP_);
612
	return bp;
L
Linus Torvalds 已提交
613 614
}

C
Christoph Hellwig 已提交
615 616 617 618 619
STATIC int
_xfs_buf_read(
	xfs_buf_t		*bp,
	xfs_buf_flags_t		flags)
{
620
	ASSERT(!(flags & XBF_WRITE));
621
	ASSERT(bp->b_maps[0].bm_bn != XFS_BUF_DADDR_NULL);
C
Christoph Hellwig 已提交
622

623
	bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_READ_AHEAD);
624
	bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD);
C
Christoph Hellwig 已提交
625

626 627 628
	xfs_buf_iorequest(bp);
	if (flags & XBF_ASYNC)
		return 0;
629
	return xfs_buf_iowait(bp);
C
Christoph Hellwig 已提交
630 631
}

L
Linus Torvalds 已提交
632
xfs_buf_t *
633 634 635 636
xfs_buf_read_map(
	struct xfs_buftarg	*target,
	struct xfs_buf_map	*map,
	int			nmaps,
637
	xfs_buf_flags_t		flags,
638
	const struct xfs_buf_ops *ops)
L
Linus Torvalds 已提交
639
{
640
	struct xfs_buf		*bp;
641 642 643

	flags |= XBF_READ;

644
	bp = xfs_buf_get_map(target, map, nmaps, flags);
645
	if (bp) {
C
Christoph Hellwig 已提交
646 647
		trace_xfs_buf_read(bp, flags, _RET_IP_);

648 649
		if (!XFS_BUF_ISDONE(bp)) {
			XFS_STATS_INC(xb_get_read);
650
			bp->b_ops = ops;
C
Christoph Hellwig 已提交
651
			_xfs_buf_read(bp, flags);
652
		} else if (flags & XBF_ASYNC) {
L
Linus Torvalds 已提交
653 654 655 656
			/*
			 * Read ahead call which is already satisfied,
			 * drop the buffer
			 */
D
Dave Chinner 已提交
657 658
			xfs_buf_relse(bp);
			return NULL;
L
Linus Torvalds 已提交
659 660
		} else {
			/* We do not want read in the flags */
661
			bp->b_flags &= ~XBF_READ;
L
Linus Torvalds 已提交
662 663 664
		}
	}

665
	return bp;
L
Linus Torvalds 已提交
666 667 668
}

/*
669 670
 *	If we are not low on memory then do the readahead in a deadlock
 *	safe manner.
L
Linus Torvalds 已提交
671 672
 */
void
673 674 675
xfs_buf_readahead_map(
	struct xfs_buftarg	*target,
	struct xfs_buf_map	*map,
676
	int			nmaps,
677
	const struct xfs_buf_ops *ops)
L
Linus Torvalds 已提交
678
{
679
	if (bdi_read_congested(target->bt_bdi))
L
Linus Torvalds 已提交
680 681
		return;

682
	xfs_buf_read_map(target, map, nmaps,
683
		     XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD, ops);
L
Linus Torvalds 已提交
684 685
}

686 687 688 689 690 691 692 693
/*
 * Read an uncached buffer from disk. Allocates and returns a locked
 * buffer containing the disk contents or nothing.
 */
struct xfs_buf *
xfs_buf_read_uncached(
	struct xfs_buftarg	*target,
	xfs_daddr_t		daddr,
694
	size_t			numblks,
695
	int			flags,
696
	const struct xfs_buf_ops *ops)
697
{
698
	struct xfs_buf		*bp;
699

700
	bp = xfs_buf_get_uncached(target, numblks, flags);
701 702 703 704
	if (!bp)
		return NULL;

	/* set up the buffer for a read IO */
705 706 707
	ASSERT(bp->b_map_count == 1);
	bp->b_bn = daddr;
	bp->b_maps[0].bm_bn = daddr;
708
	bp->b_flags |= XBF_READ;
709
	bp->b_ops = ops;
710

C
Christoph Hellwig 已提交
711 712 713 714 715
	if (XFS_FORCED_SHUTDOWN(target->bt_mount)) {
		xfs_buf_relse(bp);
		return NULL;
	}
	xfs_buf_iorequest(bp);
716
	xfs_buf_iowait(bp);
717
	return bp;
L
Linus Torvalds 已提交
718 719
}

720 721 722 723 724 725 726
/*
 * Return a buffer allocated as an empty buffer and associated to external
 * memory via xfs_buf_associate_memory() back to it's empty state.
 */
void
xfs_buf_set_empty(
	struct xfs_buf		*bp,
727
	size_t			numblks)
728 729 730 731 732 733 734
{
	if (bp->b_pages)
		_xfs_buf_free_pages(bp);

	bp->b_pages = NULL;
	bp->b_page_count = 0;
	bp->b_addr = NULL;
735
	bp->b_length = numblks;
736
	bp->b_io_length = numblks;
737 738

	ASSERT(bp->b_map_count == 1);
739
	bp->b_bn = XFS_BUF_DADDR_NULL;
740 741
	bp->b_maps[0].bm_bn = XFS_BUF_DADDR_NULL;
	bp->b_maps[0].bm_len = bp->b_length;
742 743
}

L
Linus Torvalds 已提交
744 745 746 747
static inline struct page *
mem_to_page(
	void			*addr)
{
748
	if ((!is_vmalloc_addr(addr))) {
L
Linus Torvalds 已提交
749 750 751 752 753 754 755
		return virt_to_page(addr);
	} else {
		return vmalloc_to_page(addr);
	}
}

int
756 757
xfs_buf_associate_memory(
	xfs_buf_t		*bp,
L
Linus Torvalds 已提交
758 759 760 761 762
	void			*mem,
	size_t			len)
{
	int			rval;
	int			i = 0;
763 764 765
	unsigned long		pageaddr;
	unsigned long		offset;
	size_t			buflen;
L
Linus Torvalds 已提交
766 767
	int			page_count;

768
	pageaddr = (unsigned long)mem & PAGE_MASK;
769
	offset = (unsigned long)mem - pageaddr;
770 771
	buflen = PAGE_ALIGN(len + offset);
	page_count = buflen >> PAGE_SHIFT;
L
Linus Torvalds 已提交
772 773

	/* Free any previous set of page pointers */
774 775
	if (bp->b_pages)
		_xfs_buf_free_pages(bp);
L
Linus Torvalds 已提交
776

777 778
	bp->b_pages = NULL;
	bp->b_addr = mem;
L
Linus Torvalds 已提交
779

780
	rval = _xfs_buf_get_pages(bp, page_count);
L
Linus Torvalds 已提交
781 782 783
	if (rval)
		return rval;

784
	bp->b_offset = offset;
785 786 787

	for (i = 0; i < bp->b_page_count; i++) {
		bp->b_pages[i] = mem_to_page((void *)pageaddr);
788
		pageaddr += PAGE_SIZE;
L
Linus Torvalds 已提交
789 790
	}

791
	bp->b_io_length = BTOBB(len);
792
	bp->b_length = BTOBB(buflen);
L
Linus Torvalds 已提交
793 794 795 796 797

	return 0;
}

xfs_buf_t *
798 799
xfs_buf_get_uncached(
	struct xfs_buftarg	*target,
800
	size_t			numblks,
801
	int			flags)
L
Linus Torvalds 已提交
802
{
803
	unsigned long		page_count;
804
	int			error, i;
805 806
	struct xfs_buf		*bp;
	DEFINE_SINGLE_BUF_MAP(map, XFS_BUF_DADDR_NULL, numblks);
L
Linus Torvalds 已提交
807

808
	bp = _xfs_buf_alloc(target, &map, 1, 0);
L
Linus Torvalds 已提交
809 810 811
	if (unlikely(bp == NULL))
		goto fail;

812
	page_count = PAGE_ALIGN(numblks << BBSHIFT) >> PAGE_SHIFT;
813
	error = _xfs_buf_get_pages(bp, page_count);
814
	if (error)
L
Linus Torvalds 已提交
815 816
		goto fail_free_buf;

817
	for (i = 0; i < page_count; i++) {
818
		bp->b_pages[i] = alloc_page(xb_to_gfp(flags));
819 820
		if (!bp->b_pages[i])
			goto fail_free_mem;
L
Linus Torvalds 已提交
821
	}
822
	bp->b_flags |= _XBF_PAGES;
L
Linus Torvalds 已提交
823

824
	error = _xfs_buf_map_pages(bp, 0);
825
	if (unlikely(error)) {
826
		xfs_warn(target->bt_mount,
827
			"%s: failed to map pages", __func__);
L
Linus Torvalds 已提交
828
		goto fail_free_mem;
829
	}
L
Linus Torvalds 已提交
830

831
	trace_xfs_buf_get_uncached(bp, _RET_IP_);
L
Linus Torvalds 已提交
832
	return bp;
833

L
Linus Torvalds 已提交
834
 fail_free_mem:
835 836
	while (--i >= 0)
		__free_page(bp->b_pages[i]);
837
	_xfs_buf_free_pages(bp);
L
Linus Torvalds 已提交
838
 fail_free_buf:
839
	xfs_buf_free_maps(bp);
840
	kmem_zone_free(xfs_buf_zone, bp);
L
Linus Torvalds 已提交
841 842 843 844 845 846 847 848 849 850
 fail:
	return NULL;
}

/*
 *	Increment reference count on buffer, to hold the buffer concurrently
 *	with another thread which may release (free) the buffer asynchronously.
 *	Must hold the buffer already to call this function.
 */
void
851 852
xfs_buf_hold(
	xfs_buf_t		*bp)
L
Linus Torvalds 已提交
853
{
C
Christoph Hellwig 已提交
854
	trace_xfs_buf_hold(bp, _RET_IP_);
855
	atomic_inc(&bp->b_hold);
L
Linus Torvalds 已提交
856 857 858
}

/*
859 860
 *	Releases a hold on the specified buffer.  If the
 *	the hold count is 1, calls xfs_buf_free.
L
Linus Torvalds 已提交
861 862
 */
void
863 864
xfs_buf_rele(
	xfs_buf_t		*bp)
L
Linus Torvalds 已提交
865
{
866
	struct xfs_perag	*pag = bp->b_pag;
L
Linus Torvalds 已提交
867

C
Christoph Hellwig 已提交
868
	trace_xfs_buf_rele(bp, _RET_IP_);
L
Linus Torvalds 已提交
869

870
	if (!pag) {
871
		ASSERT(list_empty(&bp->b_lru));
872
		ASSERT(RB_EMPTY_NODE(&bp->b_rbnode));
873 874 875 876 877
		if (atomic_dec_and_test(&bp->b_hold))
			xfs_buf_free(bp);
		return;
	}

878
	ASSERT(!RB_EMPTY_NODE(&bp->b_rbnode));
879

880
	ASSERT(atomic_read(&bp->b_hold) > 0);
881
	if (atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock)) {
882 883 884 885 886 887 888 889 890 891 892 893
		spin_lock(&bp->b_lock);
		if (!(bp->b_flags & XBF_STALE) && atomic_read(&bp->b_lru_ref)) {
			/*
			 * If the buffer is added to the LRU take a new
			 * reference to the buffer for the LRU and clear the
			 * (now stale) dispose list state flag
			 */
			if (list_lru_add(&bp->b_target->bt_lru, &bp->b_lru)) {
				bp->b_state &= ~XFS_BSTATE_DISPOSE;
				atomic_inc(&bp->b_hold);
			}
			spin_unlock(&bp->b_lock);
894
			spin_unlock(&pag->pag_buf_lock);
L
Linus Torvalds 已提交
895
		} else {
896 897 898 899 900 901 902 903 904 905 906 907 908
			/*
			 * most of the time buffers will already be removed from
			 * the LRU, so optimise that case by checking for the
			 * XFS_BSTATE_DISPOSE flag indicating the last list the
			 * buffer was on was the disposal list
			 */
			if (!(bp->b_state & XFS_BSTATE_DISPOSE)) {
				list_lru_del(&bp->b_target->bt_lru, &bp->b_lru);
			} else {
				ASSERT(list_empty(&bp->b_lru));
			}
			spin_unlock(&bp->b_lock);

909
			ASSERT(!(bp->b_flags & _XBF_DELWRI_Q));
910 911 912
			rb_erase(&bp->b_rbnode, &pag->pag_buf_tree);
			spin_unlock(&pag->pag_buf_lock);
			xfs_perag_put(pag);
913
			xfs_buf_free(bp);
L
Linus Torvalds 已提交
914 915 916 917 918 919
		}
	}
}


/*
920
 *	Lock a buffer object, if it is not already locked.
921 922 923 924 925 926 927 928
 *
 *	If we come across a stale, pinned, locked buffer, we know that we are
 *	being asked to lock a buffer that has been reallocated. Because it is
 *	pinned, we know that the log has not been pushed to disk and hence it
 *	will still be locked.  Rather than continuing to have trylock attempts
 *	fail until someone else pushes the log, push it ourselves before
 *	returning.  This means that the xfsaild will not get stuck trying
 *	to push on stale inode buffers.
L
Linus Torvalds 已提交
929 930
 */
int
931 932
xfs_buf_trylock(
	struct xfs_buf		*bp)
L
Linus Torvalds 已提交
933 934 935
{
	int			locked;

936
	locked = down_trylock(&bp->b_sema) == 0;
C
Christoph Hellwig 已提交
937
	if (locked)
938
		XB_SET_OWNER(bp);
C
Christoph Hellwig 已提交
939

940 941
	trace_xfs_buf_trylock(bp, _RET_IP_);
	return locked;
L
Linus Torvalds 已提交
942 943 944
}

/*
945
 *	Lock a buffer object.
946 947 948 949 950 951
 *
 *	If we come across a stale, pinned, locked buffer, we know that we
 *	are being asked to lock a buffer that has been reallocated. Because
 *	it is pinned, we know that the log has not been pushed to disk and
 *	hence it will still be locked. Rather than sleeping until someone
 *	else pushes the log, push it ourselves before trying to get the lock.
L
Linus Torvalds 已提交
952
 */
953 954
void
xfs_buf_lock(
955
	struct xfs_buf		*bp)
L
Linus Torvalds 已提交
956
{
C
Christoph Hellwig 已提交
957 958
	trace_xfs_buf_lock(bp, _RET_IP_);

959
	if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))
960
		xfs_log_force(bp->b_target->bt_mount, 0);
961 962
	down(&bp->b_sema);
	XB_SET_OWNER(bp);
C
Christoph Hellwig 已提交
963 964

	trace_xfs_buf_lock_done(bp, _RET_IP_);
L
Linus Torvalds 已提交
965 966 967
}

void
968
xfs_buf_unlock(
969
	struct xfs_buf		*bp)
L
Linus Torvalds 已提交
970
{
971 972
	XB_CLEAR_OWNER(bp);
	up(&bp->b_sema);
C
Christoph Hellwig 已提交
973 974

	trace_xfs_buf_unlock(bp, _RET_IP_);
L
Linus Torvalds 已提交
975 976
}

977 978 979
STATIC void
xfs_buf_wait_unpin(
	xfs_buf_t		*bp)
L
Linus Torvalds 已提交
980 981 982
{
	DECLARE_WAITQUEUE	(wait, current);

983
	if (atomic_read(&bp->b_pin_count) == 0)
L
Linus Torvalds 已提交
984 985
		return;

986
	add_wait_queue(&bp->b_waiters, &wait);
L
Linus Torvalds 已提交
987 988
	for (;;) {
		set_current_state(TASK_UNINTERRUPTIBLE);
989
		if (atomic_read(&bp->b_pin_count) == 0)
L
Linus Torvalds 已提交
990
			break;
J
Jens Axboe 已提交
991
		io_schedule();
L
Linus Torvalds 已提交
992
	}
993
	remove_wait_queue(&bp->b_waiters, &wait);
L
Linus Torvalds 已提交
994 995 996 997 998 999 1000
	set_current_state(TASK_RUNNING);
}

/*
 *	Buffer Utility Routines
 */

1001 1002 1003
void
xfs_buf_ioend(
	struct xfs_buf	*bp)
L
Linus Torvalds 已提交
1004
{
1005 1006 1007
	bool		read = bp->b_flags & XBF_READ;

	trace_xfs_buf_iodone(bp, _RET_IP_);
1008 1009

	bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD);
1010

1011 1012 1013 1014 1015 1016 1017
	/*
	 * Pull in IO completion errors now. We are guaranteed to be running
	 * single threaded, so we don't need the lock to read b_io_error.
	 */
	if (!bp->b_error && bp->b_io_error)
		xfs_buf_ioerror(bp, bp->b_io_error);

1018 1019 1020
	/* Only validate buffers that were read without errors */
	if (read && !bp->b_error && bp->b_ops) {
		ASSERT(!bp->b_iodone);
1021
		bp->b_ops->verify_read(bp);
1022 1023 1024 1025
	}

	if (!bp->b_error)
		bp->b_flags |= XBF_DONE;
L
Linus Torvalds 已提交
1026

1027
	if (bp->b_iodone)
1028 1029
		(*(bp->b_iodone))(bp);
	else if (bp->b_flags & XBF_ASYNC)
L
Linus Torvalds 已提交
1030
		xfs_buf_relse(bp);
1031 1032
	else {
		complete(&bp->b_iowait);
1033 1034 1035

		/* release the !XBF_ASYNC ref now we are done. */
		xfs_buf_rele(bp);
1036
	}
L
Linus Torvalds 已提交
1037 1038
}

1039 1040 1041
static void
xfs_buf_ioend_work(
	struct work_struct	*work)
L
Linus Torvalds 已提交
1042
{
1043 1044
	struct xfs_buf		*bp =
		container_of(work, xfs_buf_t, b_iodone_work);
C
Christoph Hellwig 已提交
1045

1046 1047
	xfs_buf_ioend(bp);
}
L
Linus Torvalds 已提交
1048

1049 1050 1051 1052 1053 1054
void
xfs_buf_ioend_async(
	struct xfs_buf	*bp)
{
	INIT_WORK(&bp->b_iodone_work, xfs_buf_ioend_work);
	queue_work(xfslogd_workqueue, &bp->b_iodone_work);
L
Linus Torvalds 已提交
1055 1056 1057
}

void
1058 1059 1060
xfs_buf_ioerror(
	xfs_buf_t		*bp,
	int			error)
L
Linus Torvalds 已提交
1061
{
D
Dave Chinner 已提交
1062 1063
	ASSERT(error <= 0 && error >= -1000);
	bp->b_error = error;
C
Christoph Hellwig 已提交
1064
	trace_xfs_buf_ioerror(bp, error, _RET_IP_);
L
Linus Torvalds 已提交
1065 1066
}

1067 1068 1069 1070 1071 1072
void
xfs_buf_ioerror_alert(
	struct xfs_buf		*bp,
	const char		*func)
{
	xfs_alert(bp->b_target->bt_mount,
1073
"metadata I/O error: block 0x%llx (\"%s\") error %d numblks %d",
D
Dave Chinner 已提交
1074
		(__uint64_t)XFS_BUF_ADDR(bp), func, -bp->b_error, bp->b_length);
1075 1076
}

1077 1078
/*
 * Called when we want to stop a buffer from getting written or read.
C
Christoph Hellwig 已提交
1079
 * We attach the EIO error, muck with its flags, and call xfs_buf_ioend
1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092
 * so that the proper iodone callbacks get called.
 */
STATIC int
xfs_bioerror(
	xfs_buf_t *bp)
{
#ifdef XFSERRORDEBUG
	ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone);
#endif

	/*
	 * No need to wait until the buffer is unpinned, we aren't flushing it.
	 */
D
Dave Chinner 已提交
1093
	xfs_buf_ioerror(bp, -EIO);
1094 1095

	/*
1096 1097
	 * We're calling xfs_buf_ioend, so delete XBF_DONE flag. For
	 * sync IO, xfs_buf_ioend is going to remove a ref here.
1098
	 */
1099 1100
	if (!(bp->b_flags & XBF_ASYNC))
		xfs_buf_hold(bp);
1101 1102
	XFS_BUF_UNREAD(bp);
	XFS_BUF_UNDONE(bp);
1103
	xfs_buf_stale(bp);
1104

1105
	xfs_buf_ioend(bp);
1106

D
Dave Chinner 已提交
1107
	return -EIO;
1108 1109 1110 1111
}

/*
 * Same as xfs_bioerror, except that we are releasing the buffer
C
Christoph Hellwig 已提交
1112
 * here ourselves, and avoiding the xfs_buf_ioend call.
1113 1114 1115
 * This is meant for userdata errors; metadata bufs come with
 * iodone functions attached, so that we can track down errors.
 */
C
Christoph Hellwig 已提交
1116
int
1117 1118 1119
xfs_bioerror_relse(
	struct xfs_buf	*bp)
{
1120
	int64_t		fl = bp->b_flags;
1121 1122 1123 1124 1125 1126 1127 1128 1129 1130
	/*
	 * No need to wait until the buffer is unpinned.
	 * We aren't flushing it.
	 *
	 * chunkhold expects B_DONE to be set, whether
	 * we actually finish the I/O or not. We don't want to
	 * change that interface.
	 */
	XFS_BUF_UNREAD(bp);
	XFS_BUF_DONE(bp);
1131
	xfs_buf_stale(bp);
1132
	bp->b_iodone = NULL;
1133
	if (!(fl & XBF_ASYNC)) {
1134 1135 1136 1137 1138 1139
		/*
		 * Mark b_error and B_ERROR _both_.
		 * Lot's of chunkcache code assumes that.
		 * There's no reason to mark error for
		 * ASYNC buffers.
		 */
D
Dave Chinner 已提交
1140
		xfs_buf_ioerror(bp, -EIO);
1141
		complete(&bp->b_iowait);
1142 1143 1144 1145
	} else {
		xfs_buf_relse(bp);
	}

D
Dave Chinner 已提交
1146
	return -EIO;
1147 1148
}

1149 1150 1151 1152 1153 1154 1155 1156 1157
int
xfs_bwrite(
	struct xfs_buf		*bp)
{
	int			error;

	ASSERT(xfs_buf_islocked(bp));

	bp->b_flags |= XBF_WRITE;
1158
	bp->b_flags &= ~(XBF_ASYNC | XBF_READ | _XBF_DELWRI_Q | XBF_WRITE_FAIL);
1159

D
Dave Chinner 已提交
1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173
	if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) {
		trace_xfs_bdstrat_shut(bp, _RET_IP_);

		/*
		 * Metadata write that didn't get logged but written anyway.
		 * These aren't associated with a transaction, and can be
		 * ignored.
		 */
		if (!bp->b_iodone)
			return xfs_bioerror_relse(bp);
		return xfs_bioerror(bp);
	}

	xfs_buf_iorequest(bp);
1174 1175 1176 1177 1178 1179 1180 1181 1182

	error = xfs_buf_iowait(bp);
	if (error) {
		xfs_force_shutdown(bp->b_target->bt_mount,
				   SHUTDOWN_META_IO_ERROR);
	}
	return error;
}

A
Al Viro 已提交
1183
STATIC void
1184
xfs_buf_bio_end_io(
L
Linus Torvalds 已提交
1185 1186 1187
	struct bio		*bio,
	int			error)
{
1188
	xfs_buf_t		*bp = (xfs_buf_t *)bio->bi_private;
L
Linus Torvalds 已提交
1189

1190 1191 1192 1193
	/*
	 * don't overwrite existing errors - otherwise we can lose errors on
	 * buffers that require multiple bios to complete.
	 */
1194 1195 1196 1197 1198 1199
	if (error) {
		spin_lock(&bp->b_lock);
		if (!bp->b_io_error)
			bp->b_io_error = error;
		spin_unlock(&bp->b_lock);
	}
L
Linus Torvalds 已提交
1200

1201
	if (!bp->b_error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ))
1202 1203
		invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp));

1204 1205
	if (atomic_dec_and_test(&bp->b_io_remaining) == 1)
		xfs_buf_ioend_async(bp);
L
Linus Torvalds 已提交
1206 1207 1208
	bio_put(bio);
}

1209 1210 1211 1212 1213 1214 1215
static void
xfs_buf_ioapply_map(
	struct xfs_buf	*bp,
	int		map,
	int		*buf_offset,
	int		*count,
	int		rw)
L
Linus Torvalds 已提交
1216
{
1217 1218 1219 1220 1221 1222 1223
	int		page_index;
	int		total_nr_pages = bp->b_page_count;
	int		nr_pages;
	struct bio	*bio;
	sector_t	sector =  bp->b_maps[map].bm_bn;
	int		size;
	int		offset;
L
Linus Torvalds 已提交
1224

1225
	total_nr_pages = bp->b_page_count;
L
Linus Torvalds 已提交
1226

1227 1228 1229 1230 1231 1232
	/* skip the pages in the buffer before the start offset */
	page_index = 0;
	offset = *buf_offset;
	while (offset >= PAGE_SIZE) {
		page_index++;
		offset -= PAGE_SIZE;
1233 1234
	}

1235 1236 1237 1238 1239 1240 1241
	/*
	 * Limit the IO size to the length of the current vector, and update the
	 * remaining IO count for the next time around.
	 */
	size = min_t(int, BBTOB(bp->b_maps[map].bm_len), *count);
	*count -= size;
	*buf_offset += size;
1242

L
Linus Torvalds 已提交
1243
next_chunk:
1244
	atomic_inc(&bp->b_io_remaining);
L
Linus Torvalds 已提交
1245 1246 1247 1248 1249
	nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT);
	if (nr_pages > total_nr_pages)
		nr_pages = total_nr_pages;

	bio = bio_alloc(GFP_NOIO, nr_pages);
1250
	bio->bi_bdev = bp->b_target->bt_bdev;
1251
	bio->bi_iter.bi_sector = sector;
1252 1253
	bio->bi_end_io = xfs_buf_bio_end_io;
	bio->bi_private = bp;
L
Linus Torvalds 已提交
1254

1255

1256
	for (; size && nr_pages; nr_pages--, page_index++) {
1257
		int	rbytes, nbytes = PAGE_SIZE - offset;
L
Linus Torvalds 已提交
1258 1259 1260 1261

		if (nbytes > size)
			nbytes = size;

1262 1263
		rbytes = bio_add_page(bio, bp->b_pages[page_index], nbytes,
				      offset);
1264
		if (rbytes < nbytes)
L
Linus Torvalds 已提交
1265 1266 1267
			break;

		offset = 0;
1268
		sector += BTOBB(nbytes);
L
Linus Torvalds 已提交
1269 1270 1271 1272
		size -= nbytes;
		total_nr_pages--;
	}

1273
	if (likely(bio->bi_iter.bi_size)) {
1274 1275 1276 1277
		if (xfs_buf_is_vmapped(bp)) {
			flush_kernel_vmap_range(bp->b_addr,
						xfs_buf_vmap_len(bp));
		}
L
Linus Torvalds 已提交
1278 1279 1280 1281
		submit_bio(rw, bio);
		if (size)
			goto next_chunk;
	} else {
1282 1283 1284 1285 1286
		/*
		 * This is guaranteed not to be the last io reference count
		 * because the caller (xfs_buf_iorequest) holds a count itself.
		 */
		atomic_dec(&bp->b_io_remaining);
D
Dave Chinner 已提交
1287
		xfs_buf_ioerror(bp, -EIO);
1288
		bio_put(bio);
L
Linus Torvalds 已提交
1289
	}
1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302

}

STATIC void
_xfs_buf_ioapply(
	struct xfs_buf	*bp)
{
	struct blk_plug	plug;
	int		rw;
	int		offset;
	int		size;
	int		i;

1303 1304 1305 1306 1307 1308
	/*
	 * Make sure we capture only current IO errors rather than stale errors
	 * left over from previous use of the buffer (e.g. failed readahead).
	 */
	bp->b_error = 0;

1309 1310 1311 1312 1313 1314 1315 1316 1317
	if (bp->b_flags & XBF_WRITE) {
		if (bp->b_flags & XBF_SYNCIO)
			rw = WRITE_SYNC;
		else
			rw = WRITE;
		if (bp->b_flags & XBF_FUA)
			rw |= REQ_FUA;
		if (bp->b_flags & XBF_FLUSH)
			rw |= REQ_FLUSH;
1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330

		/*
		 * Run the write verifier callback function if it exists. If
		 * this function fails it will mark the buffer with an error and
		 * the IO should not be dispatched.
		 */
		if (bp->b_ops) {
			bp->b_ops->verify_write(bp);
			if (bp->b_error) {
				xfs_force_shutdown(bp->b_target->bt_mount,
						   SHUTDOWN_CORRUPT_INCORE);
				return;
			}
1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344
		} else if (bp->b_bn != XFS_BUF_DADDR_NULL) {
			struct xfs_mount *mp = bp->b_target->bt_mount;

			/*
			 * non-crc filesystems don't attach verifiers during
			 * log recovery, so don't warn for such filesystems.
			 */
			if (xfs_sb_version_hascrc(&mp->m_sb)) {
				xfs_warn(mp,
					"%s: no ops on block 0x%llx/0x%x",
					__func__, bp->b_bn, bp->b_length);
				xfs_hex_dump(bp->b_addr, 64);
				dump_stack();
			}
1345
		}
1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371
	} else if (bp->b_flags & XBF_READ_AHEAD) {
		rw = READA;
	} else {
		rw = READ;
	}

	/* we only use the buffer cache for meta-data */
	rw |= REQ_META;

	/*
	 * Walk all the vectors issuing IO on them. Set up the initial offset
	 * into the buffer and the desired IO size before we start -
	 * _xfs_buf_ioapply_vec() will modify them appropriately for each
	 * subsequent call.
	 */
	offset = bp->b_offset;
	size = BBTOB(bp->b_io_length);
	blk_start_plug(&plug);
	for (i = 0; i < bp->b_map_count; i++) {
		xfs_buf_ioapply_map(bp, i, &offset, &size, rw);
		if (bp->b_error)
			break;
		if (size <= 0)
			break;	/* all done */
	}
	blk_finish_plug(&plug);
L
Linus Torvalds 已提交
1372 1373
}

1374
void
1375 1376
xfs_buf_iorequest(
	xfs_buf_t		*bp)
L
Linus Torvalds 已提交
1377
{
C
Christoph Hellwig 已提交
1378
	trace_xfs_buf_iorequest(bp, _RET_IP_);
L
Linus Torvalds 已提交
1379

1380
	ASSERT(!(bp->b_flags & _XBF_DELWRI_Q));
L
Linus Torvalds 已提交
1381

1382
	if (bp->b_flags & XBF_WRITE)
1383
		xfs_buf_wait_unpin(bp);
1384

1385 1386 1387
	/* clear the internal error state to avoid spurious errors */
	bp->b_io_error = 0;

1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403
	/*
	 * Take references to the buffer. For XBF_ASYNC buffers, holding a
	 * reference for as long as submission takes is all that is necessary
	 * here. The IO inherits the lock and hold count from the submitter,
	 * and these are release during IO completion processing. Taking a hold
	 * over submission ensures that the buffer is not freed until we have
	 * completed all processing, regardless of when IO errors occur or are
	 * reported.
	 *
	 * However, for synchronous IO, the IO does not inherit the submitters
	 * reference count, nor the buffer lock. Hence we need to take an extra
	 * reference to the buffer for the for the IO context so that we can
	 * guarantee the buffer is not freed until all IO completion processing
	 * is done. Otherwise the caller can drop their reference while the IO
	 * is still in progress and hence trigger a use-after-free situation.
	 */
1404
	xfs_buf_hold(bp);
1405 1406 1407
	if (!(bp->b_flags & XBF_ASYNC))
		xfs_buf_hold(bp);

L
Linus Torvalds 已提交
1408

1409
	/*
1410 1411 1412
	 * Set the count to 1 initially, this will stop an I/O completion
	 * callout which happens before we have started all the I/O from calling
	 * xfs_buf_ioend too early.
L
Linus Torvalds 已提交
1413
	 */
1414 1415
	atomic_set(&bp->b_io_remaining, 1);
	_xfs_buf_ioapply(bp);
1416

1417
	/*
1418 1419
	 * If _xfs_buf_ioapply failed or we are doing synchronous IO that
	 * completes extremely quickly, we can get back here with only the IO
1420 1421 1422 1423
	 * reference we took above. If we drop it to zero, run completion
	 * processing synchronously so that we don't return to the caller with
	 * completion still pending. This avoids unnecessary context switches
	 * associated with the end_io workqueue.
1424
	 */
1425 1426 1427 1428 1429 1430
	if (atomic_dec_and_test(&bp->b_io_remaining) == 1) {
		if (bp->b_error || !(bp->b_flags & XBF_ASYNC))
			xfs_buf_ioend(bp);
		else
			xfs_buf_ioend_async(bp);
	}
L
Linus Torvalds 已提交
1431

1432
	xfs_buf_rele(bp);
L
Linus Torvalds 已提交
1433 1434 1435
}

/*
1436
 * Waits for I/O to complete on the buffer supplied.  It returns immediately if
1437 1438 1439
 * no I/O is pending or there is already a pending error on the buffer, in which
 * case nothing will ever complete.  It returns the I/O error code, if any, or
 * 0 if there was no error.
L
Linus Torvalds 已提交
1440 1441
 */
int
1442 1443
xfs_buf_iowait(
	xfs_buf_t		*bp)
L
Linus Torvalds 已提交
1444
{
C
Christoph Hellwig 已提交
1445 1446
	trace_xfs_buf_iowait(bp, _RET_IP_);

1447 1448
	if (!bp->b_error)
		wait_for_completion(&bp->b_iowait);
C
Christoph Hellwig 已提交
1449 1450

	trace_xfs_buf_iowait_done(bp, _RET_IP_);
1451
	return bp->b_error;
L
Linus Torvalds 已提交
1452 1453
}

1454 1455 1456
xfs_caddr_t
xfs_buf_offset(
	xfs_buf_t		*bp,
L
Linus Torvalds 已提交
1457 1458 1459 1460
	size_t			offset)
{
	struct page		*page;

1461
	if (bp->b_addr)
1462
		return bp->b_addr + offset;
L
Linus Torvalds 已提交
1463

1464
	offset += bp->b_offset;
1465 1466
	page = bp->b_pages[offset >> PAGE_SHIFT];
	return (xfs_caddr_t)page_address(page) + (offset & (PAGE_SIZE-1));
L
Linus Torvalds 已提交
1467 1468 1469 1470 1471 1472
}

/*
 *	Move data into or out of a buffer.
 */
void
1473 1474
xfs_buf_iomove(
	xfs_buf_t		*bp,	/* buffer to process		*/
L
Linus Torvalds 已提交
1475 1476
	size_t			boff,	/* starting buffer offset	*/
	size_t			bsize,	/* length to copy		*/
1477
	void			*data,	/* data address			*/
1478
	xfs_buf_rw_t		mode)	/* read/write/zero flag		*/
L
Linus Torvalds 已提交
1479
{
D
Dave Chinner 已提交
1480
	size_t			bend;
L
Linus Torvalds 已提交
1481 1482 1483

	bend = boff + bsize;
	while (boff < bend) {
D
Dave Chinner 已提交
1484 1485 1486 1487 1488 1489 1490 1491
		struct page	*page;
		int		page_index, page_offset, csize;

		page_index = (boff + bp->b_offset) >> PAGE_SHIFT;
		page_offset = (boff + bp->b_offset) & ~PAGE_MASK;
		page = bp->b_pages[page_index];
		csize = min_t(size_t, PAGE_SIZE - page_offset,
				      BBTOB(bp->b_io_length) - boff);
L
Linus Torvalds 已提交
1492

D
Dave Chinner 已提交
1493
		ASSERT((csize + page_offset) <= PAGE_SIZE);
L
Linus Torvalds 已提交
1494 1495

		switch (mode) {
1496
		case XBRW_ZERO:
D
Dave Chinner 已提交
1497
			memset(page_address(page) + page_offset, 0, csize);
L
Linus Torvalds 已提交
1498
			break;
1499
		case XBRW_READ:
D
Dave Chinner 已提交
1500
			memcpy(data, page_address(page) + page_offset, csize);
L
Linus Torvalds 已提交
1501
			break;
1502
		case XBRW_WRITE:
D
Dave Chinner 已提交
1503
			memcpy(page_address(page) + page_offset, data, csize);
L
Linus Torvalds 已提交
1504 1505 1506 1507 1508 1509 1510 1511
		}

		boff += csize;
		data += csize;
	}
}

/*
1512
 *	Handling of buffer targets (buftargs).
L
Linus Torvalds 已提交
1513 1514 1515
 */

/*
1516 1517 1518
 * Wait for any bufs with callbacks that have been submitted but have not yet
 * returned. These buffers will have an elevated hold count, so wait on those
 * while freeing all the buffers only held by the LRU.
L
Linus Torvalds 已提交
1519
 */
1520 1521 1522 1523 1524 1525
static enum lru_status
xfs_buftarg_wait_rele(
	struct list_head	*item,
	spinlock_t		*lru_lock,
	void			*arg)

L
Linus Torvalds 已提交
1526
{
1527
	struct xfs_buf		*bp = container_of(item, struct xfs_buf, b_lru);
1528
	struct list_head	*dispose = arg;
1529

1530
	if (atomic_read(&bp->b_hold) > 1) {
1531
		/* need to wait, so skip it this pass */
1532
		trace_xfs_buf_wait_buftarg(bp, _RET_IP_);
1533
		return LRU_SKIP;
L
Linus Torvalds 已提交
1534
	}
1535 1536
	if (!spin_trylock(&bp->b_lock))
		return LRU_SKIP;
1537

1538 1539 1540 1541 1542 1543 1544 1545 1546
	/*
	 * clear the LRU reference count so the buffer doesn't get
	 * ignored in xfs_buf_rele().
	 */
	atomic_set(&bp->b_lru_ref, 0);
	bp->b_state |= XFS_BSTATE_DISPOSE;
	list_move(item, dispose);
	spin_unlock(&bp->b_lock);
	return LRU_REMOVED;
L
Linus Torvalds 已提交
1547 1548
}

1549 1550 1551 1552
void
xfs_wait_buftarg(
	struct xfs_buftarg	*btp)
{
1553 1554 1555 1556 1557
	LIST_HEAD(dispose);
	int loop = 0;

	/* loop until there is nothing left on the lru list. */
	while (list_lru_count(&btp->bt_lru)) {
1558
		list_lru_walk(&btp->bt_lru, xfs_buftarg_wait_rele,
1559 1560 1561 1562 1563 1564
			      &dispose, LONG_MAX);

		while (!list_empty(&dispose)) {
			struct xfs_buf *bp;
			bp = list_first_entry(&dispose, struct xfs_buf, b_lru);
			list_del_init(&bp->b_lru);
1565 1566 1567 1568 1569 1570
			if (bp->b_flags & XBF_WRITE_FAIL) {
				xfs_alert(btp->bt_mount,
"Corruption Alert: Buffer at block 0x%llx had permanent write failures!\n"
"Please run xfs_repair to determine the extent of the problem.",
					(long long)bp->b_bn);
			}
1571 1572 1573 1574 1575
			xfs_buf_rele(bp);
		}
		if (loop++ != 0)
			delay(100);
	}
1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586
}

static enum lru_status
xfs_buftarg_isolate(
	struct list_head	*item,
	spinlock_t		*lru_lock,
	void			*arg)
{
	struct xfs_buf		*bp = container_of(item, struct xfs_buf, b_lru);
	struct list_head	*dispose = arg;

1587 1588 1589 1590 1591 1592
	/*
	 * we are inverting the lru lock/bp->b_lock here, so use a trylock.
	 * If we fail to get the lock, just skip it.
	 */
	if (!spin_trylock(&bp->b_lock))
		return LRU_SKIP;
1593 1594 1595 1596 1597
	/*
	 * Decrement the b_lru_ref count unless the value is already
	 * zero. If the value is already zero, we need to reclaim the
	 * buffer, otherwise it gets another trip through the LRU.
	 */
1598 1599
	if (!atomic_add_unless(&bp->b_lru_ref, -1, 0)) {
		spin_unlock(&bp->b_lock);
1600
		return LRU_ROTATE;
1601
	}
1602

1603
	bp->b_state |= XFS_BSTATE_DISPOSE;
1604
	list_move(item, dispose);
1605
	spin_unlock(&bp->b_lock);
1606 1607 1608
	return LRU_REMOVED;
}

1609
static unsigned long
1610
xfs_buftarg_shrink_scan(
1611
	struct shrinker		*shrink,
1612
	struct shrink_control	*sc)
1613
{
1614 1615
	struct xfs_buftarg	*btp = container_of(shrink,
					struct xfs_buftarg, bt_shrinker);
1616
	LIST_HEAD(dispose);
1617
	unsigned long		freed;
1618
	unsigned long		nr_to_scan = sc->nr_to_scan;
1619

1620 1621
	freed = list_lru_walk_node(&btp->bt_lru, sc->nid, xfs_buftarg_isolate,
				       &dispose, &nr_to_scan);
1622 1623

	while (!list_empty(&dispose)) {
1624
		struct xfs_buf *bp;
1625 1626 1627 1628 1629
		bp = list_first_entry(&dispose, struct xfs_buf, b_lru);
		list_del_init(&bp->b_lru);
		xfs_buf_rele(bp);
	}

1630 1631 1632
	return freed;
}

1633
static unsigned long
1634 1635 1636 1637 1638 1639 1640
xfs_buftarg_shrink_count(
	struct shrinker		*shrink,
	struct shrink_control	*sc)
{
	struct xfs_buftarg	*btp = container_of(shrink,
					struct xfs_buftarg, bt_shrinker);
	return list_lru_count_node(&btp->bt_lru, sc->nid);
1641 1642
}

L
Linus Torvalds 已提交
1643 1644
void
xfs_free_buftarg(
1645 1646
	struct xfs_mount	*mp,
	struct xfs_buftarg	*btp)
L
Linus Torvalds 已提交
1647
{
1648
	unregister_shrinker(&btp->bt_shrinker);
G
Glauber Costa 已提交
1649
	list_lru_destroy(&btp->bt_lru);
1650

1651 1652
	if (mp->m_flags & XFS_MOUNT_BARRIER)
		xfs_blkdev_issue_flush(btp);
1653

1654
	kmem_free(btp);
L
Linus Torvalds 已提交
1655 1656
}

1657 1658
int
xfs_setsize_buftarg(
L
Linus Torvalds 已提交
1659
	xfs_buftarg_t		*btp,
1660
	unsigned int		sectorsize)
L
Linus Torvalds 已提交
1661
{
1662
	/* Set up metadata sector size info */
1663 1664
	btp->bt_meta_sectorsize = sectorsize;
	btp->bt_meta_sectormask = sectorsize - 1;
L
Linus Torvalds 已提交
1665

1666
	if (set_blocksize(btp->bt_bdev, sectorsize)) {
1667 1668 1669 1670
		char name[BDEVNAME_SIZE];

		bdevname(btp->bt_bdev, name);

1671
		xfs_warn(btp->bt_mount,
1672
			"Cannot set_blocksize to %u on device %s",
1673
			sectorsize, name);
D
Dave Chinner 已提交
1674
		return -EINVAL;
L
Linus Torvalds 已提交
1675 1676
	}

1677 1678 1679 1680
	/* Set up device logical sector size mask */
	btp->bt_logical_sectorsize = bdev_logical_block_size(btp->bt_bdev);
	btp->bt_logical_sectormask = bdev_logical_block_size(btp->bt_bdev) - 1;

L
Linus Torvalds 已提交
1681 1682 1683 1684
	return 0;
}

/*
1685 1686 1687
 * When allocating the initial buffer target we have not yet
 * read in the superblock, so don't know what sized sectors
 * are being used at this early stage.  Play safe.
1688
 */
L
Linus Torvalds 已提交
1689 1690 1691 1692 1693
STATIC int
xfs_setsize_buftarg_early(
	xfs_buftarg_t		*btp,
	struct block_device	*bdev)
{
1694
	return xfs_setsize_buftarg(btp, bdev_logical_block_size(bdev));
L
Linus Torvalds 已提交
1695 1696 1697 1698
}

xfs_buftarg_t *
xfs_alloc_buftarg(
1699
	struct xfs_mount	*mp,
1700
	struct block_device	*bdev)
L
Linus Torvalds 已提交
1701 1702 1703
{
	xfs_buftarg_t		*btp;

1704
	btp = kmem_zalloc(sizeof(*btp), KM_SLEEP | KM_NOFS);
L
Linus Torvalds 已提交
1705

1706
	btp->bt_mount = mp;
1707 1708
	btp->bt_dev =  bdev->bd_dev;
	btp->bt_bdev = bdev;
1709 1710 1711 1712
	btp->bt_bdi = blk_get_backing_dev_info(bdev);
	if (!btp->bt_bdi)
		goto error;

L
Linus Torvalds 已提交
1713 1714
	if (xfs_setsize_buftarg_early(btp, bdev))
		goto error;
1715 1716 1717 1718

	if (list_lru_init(&btp->bt_lru))
		goto error;

1719 1720
	btp->bt_shrinker.count_objects = xfs_buftarg_shrink_count;
	btp->bt_shrinker.scan_objects = xfs_buftarg_shrink_scan;
1721
	btp->bt_shrinker.seeks = DEFAULT_SEEKS;
1722
	btp->bt_shrinker.flags = SHRINKER_NUMA_AWARE;
1723
	register_shrinker(&btp->bt_shrinker);
L
Linus Torvalds 已提交
1724 1725 1726
	return btp;

error:
1727
	kmem_free(btp);
L
Linus Torvalds 已提交
1728 1729 1730 1731
	return NULL;
}

/*
1732 1733 1734 1735 1736 1737 1738 1739 1740
 * Add a buffer to the delayed write list.
 *
 * This queues a buffer for writeout if it hasn't already been.  Note that
 * neither this routine nor the buffer list submission functions perform
 * any internal synchronization.  It is expected that the lists are thread-local
 * to the callers.
 *
 * Returns true if we queued up the buffer, or false if it already had
 * been on the buffer list.
L
Linus Torvalds 已提交
1741
 */
1742
bool
1743
xfs_buf_delwri_queue(
1744 1745
	struct xfs_buf		*bp,
	struct list_head	*list)
L
Linus Torvalds 已提交
1746
{
1747
	ASSERT(xfs_buf_islocked(bp));
1748
	ASSERT(!(bp->b_flags & XBF_READ));
L
Linus Torvalds 已提交
1749

1750 1751 1752 1753 1754 1755 1756 1757
	/*
	 * If the buffer is already marked delwri it already is queued up
	 * by someone else for imediate writeout.  Just ignore it in that
	 * case.
	 */
	if (bp->b_flags & _XBF_DELWRI_Q) {
		trace_xfs_buf_delwri_queued(bp, _RET_IP_);
		return false;
L
Linus Torvalds 已提交
1758 1759
	}

1760
	trace_xfs_buf_delwri_queue(bp, _RET_IP_);
1761 1762

	/*
1763 1764 1765 1766 1767 1768
	 * If a buffer gets written out synchronously or marked stale while it
	 * is on a delwri list we lazily remove it. To do this, the other party
	 * clears the  _XBF_DELWRI_Q flag but otherwise leaves the buffer alone.
	 * It remains referenced and on the list.  In a rare corner case it
	 * might get readded to a delwri list after the synchronous writeout, in
	 * which case we need just need to re-add the flag here.
1769
	 */
1770 1771 1772 1773
	bp->b_flags |= _XBF_DELWRI_Q;
	if (list_empty(&bp->b_list)) {
		atomic_inc(&bp->b_hold);
		list_add_tail(&bp->b_list, list);
1774 1775
	}

1776
	return true;
1777 1778
}

1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793
/*
 * Compare function is more complex than it needs to be because
 * the return value is only 32 bits and we are doing comparisons
 * on 64 bit values
 */
static int
xfs_buf_cmp(
	void		*priv,
	struct list_head *a,
	struct list_head *b)
{
	struct xfs_buf	*ap = container_of(a, struct xfs_buf, b_list);
	struct xfs_buf	*bp = container_of(b, struct xfs_buf, b_list);
	xfs_daddr_t		diff;

1794
	diff = ap->b_maps[0].bm_bn - bp->b_maps[0].bm_bn;
1795 1796 1797 1798 1799 1800 1801
	if (diff < 0)
		return -1;
	if (diff > 0)
		return 1;
	return 0;
}

1802 1803 1804 1805 1806
static int
__xfs_buf_delwri_submit(
	struct list_head	*buffer_list,
	struct list_head	*io_list,
	bool			wait)
L
Linus Torvalds 已提交
1807
{
1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822
	struct blk_plug		plug;
	struct xfs_buf		*bp, *n;
	int			pinned = 0;

	list_for_each_entry_safe(bp, n, buffer_list, b_list) {
		if (!wait) {
			if (xfs_buf_ispinned(bp)) {
				pinned++;
				continue;
			}
			if (!xfs_buf_trylock(bp))
				continue;
		} else {
			xfs_buf_lock(bp);
		}
1823

1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834
		/*
		 * Someone else might have written the buffer synchronously or
		 * marked it stale in the meantime.  In that case only the
		 * _XBF_DELWRI_Q flag got cleared, and we have to drop the
		 * reference and remove it from the list here.
		 */
		if (!(bp->b_flags & _XBF_DELWRI_Q)) {
			list_del_init(&bp->b_list);
			xfs_buf_relse(bp);
			continue;
		}
D
Dave Chinner 已提交
1835

1836 1837 1838
		list_move_tail(&bp->b_list, io_list);
		trace_xfs_buf_delwri_split(bp, _RET_IP_);
	}
L
Linus Torvalds 已提交
1839

1840
	list_sort(NULL, io_list, xfs_buf_cmp);
L
Linus Torvalds 已提交
1841

1842 1843
	blk_start_plug(&plug);
	list_for_each_entry_safe(bp, n, io_list, b_list) {
1844
		bp->b_flags &= ~(_XBF_DELWRI_Q | XBF_ASYNC | XBF_WRITE_FAIL);
1845
		bp->b_flags |= XBF_WRITE | XBF_ASYNC;
1846

1847 1848 1849 1850 1851 1852 1853 1854
		/*
		 * we do all Io submission async. This means if we need to wait
		 * for IO completion we need to take an extra reference so the
		 * buffer is still valid on the other side.
		 */
		if (wait)
			xfs_buf_hold(bp);
		else
1855
			list_del_init(&bp->b_list);
D
Dave Chinner 已提交
1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866

		if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) {
			trace_xfs_bdstrat_shut(bp, _RET_IP_);

			if (!bp->b_iodone)
				xfs_bioerror_relse(bp);
			else
				xfs_bioerror(bp);
			continue;
		}
		xfs_buf_iorequest(bp);
1867 1868
	}
	blk_finish_plug(&plug);
L
Linus Torvalds 已提交
1869

1870
	return pinned;
L
Linus Torvalds 已提交
1871 1872 1873
}

/*
1874 1875 1876 1877 1878 1879 1880
 * Write out a buffer list asynchronously.
 *
 * This will take the @buffer_list, write all non-locked and non-pinned buffers
 * out and not wait for I/O completion on any of the buffers.  This interface
 * is only safely useable for callers that can track I/O completion by higher
 * level means, e.g. AIL pushing as the @buffer_list is consumed in this
 * function.
L
Linus Torvalds 已提交
1881 1882
 */
int
1883 1884
xfs_buf_delwri_submit_nowait(
	struct list_head	*buffer_list)
L
Linus Torvalds 已提交
1885
{
1886 1887 1888
	LIST_HEAD		(io_list);
	return __xfs_buf_delwri_submit(buffer_list, &io_list, false);
}
L
Linus Torvalds 已提交
1889

1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904
/*
 * Write out a buffer list synchronously.
 *
 * This will take the @buffer_list, write all buffers out and wait for I/O
 * completion on all of the buffers. @buffer_list is consumed by the function,
 * so callers must have some other way of tracking buffers if they require such
 * functionality.
 */
int
xfs_buf_delwri_submit(
	struct list_head	*buffer_list)
{
	LIST_HEAD		(io_list);
	int			error = 0, error2;
	struct xfs_buf		*bp;
L
Linus Torvalds 已提交
1905

1906
	__xfs_buf_delwri_submit(buffer_list, &io_list, true);
L
Linus Torvalds 已提交
1907

1908 1909 1910
	/* Wait for IO to complete. */
	while (!list_empty(&io_list)) {
		bp = list_first_entry(&io_list, struct xfs_buf, b_list);
1911

1912
		list_del_init(&bp->b_list);
1913 1914 1915 1916

		/* locking the buffer will wait for async IO completion. */
		xfs_buf_lock(bp);
		error2 = bp->b_error;
1917 1918 1919
		xfs_buf_relse(bp);
		if (!error)
			error = error2;
L
Linus Torvalds 已提交
1920 1921
	}

1922
	return error;
L
Linus Torvalds 已提交
1923 1924
}

1925
int __init
1926
xfs_buf_init(void)
L
Linus Torvalds 已提交
1927
{
1928 1929
	xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf",
						KM_ZONE_HWALIGN, NULL);
1930
	if (!xfs_buf_zone)
C
Christoph Hellwig 已提交
1931
		goto out;
1932

1933
	xfslogd_workqueue = alloc_workqueue("xfslogd",
1934
					WQ_MEM_RECLAIM | WQ_HIGHPRI, 1);
1935
	if (!xfslogd_workqueue)
1936
		goto out_free_buf_zone;
L
Linus Torvalds 已提交
1937

1938
	return 0;
L
Linus Torvalds 已提交
1939

1940
 out_free_buf_zone:
1941
	kmem_zone_destroy(xfs_buf_zone);
C
Christoph Hellwig 已提交
1942
 out:
1943
	return -ENOMEM;
L
Linus Torvalds 已提交
1944 1945 1946
}

void
1947
xfs_buf_terminate(void)
L
Linus Torvalds 已提交
1948
{
1949
	destroy_workqueue(xfslogd_workqueue);
1950
	kmem_zone_destroy(xfs_buf_zone);
L
Linus Torvalds 已提交
1951
}