• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2000-2006 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 <linux/stddef.h>
20 #include <linux/errno.h>
21 #include <linux/gfp.h>
22 #include <linux/pagemap.h>
23 #include <linux/init.h>
24 #include <linux/vmalloc.h>
25 #include <linux/bio.h>
26 #include <linux/sysctl.h>
27 #include <linux/proc_fs.h>
28 #include <linux/workqueue.h>
29 #include <linux/percpu.h>
30 #include <linux/blkdev.h>
31 #include <linux/hash.h>
32 #include <linux/kthread.h>
33 #include <linux/migrate.h>
34 #include <linux/backing-dev.h>
35 #include <linux/freezer.h>
36 
37 #include "xfs_format.h"
38 #include "xfs_log_format.h"
39 #include "xfs_trans_resv.h"
40 #include "xfs_sb.h"
41 #include "xfs_mount.h"
42 #include "xfs_trace.h"
43 #include "xfs_log.h"
44 
45 static kmem_zone_t *xfs_buf_zone;
46 
47 #ifdef XFS_BUF_LOCK_TRACKING
48 # define XB_SET_OWNER(bp)	((bp)->b_last_holder = current->pid)
49 # define XB_CLEAR_OWNER(bp)	((bp)->b_last_holder = -1)
50 # define XB_GET_OWNER(bp)	((bp)->b_last_holder)
51 #else
52 # define XB_SET_OWNER(bp)	do { } while (0)
53 # define XB_CLEAR_OWNER(bp)	do { } while (0)
54 # define XB_GET_OWNER(bp)	do { } while (0)
55 #endif
56 
57 #define xb_to_gfp(flags) \
58 	((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : GFP_NOFS) | __GFP_NOWARN)
59 
60 
61 static inline int
xfs_buf_is_vmapped(struct xfs_buf * bp)62 xfs_buf_is_vmapped(
63 	struct xfs_buf	*bp)
64 {
65 	/*
66 	 * Return true if the buffer is vmapped.
67 	 *
68 	 * b_addr is null if the buffer is not mapped, but the code is clever
69 	 * enough to know it doesn't have to map a single page, so the check has
70 	 * to be both for b_addr and bp->b_page_count > 1.
71 	 */
72 	return bp->b_addr && bp->b_page_count > 1;
73 }
74 
75 static inline int
xfs_buf_vmap_len(struct xfs_buf * bp)76 xfs_buf_vmap_len(
77 	struct xfs_buf	*bp)
78 {
79 	return (bp->b_page_count * PAGE_SIZE) - bp->b_offset;
80 }
81 
82 /*
83  * When we mark a buffer stale, we remove the buffer from the LRU and clear the
84  * b_lru_ref count so that the buffer is freed immediately when the buffer
85  * reference count falls to zero. If the buffer is already on the LRU, we need
86  * to remove the reference that LRU holds on the buffer.
87  *
88  * This prevents build-up of stale buffers on the LRU.
89  */
90 void
xfs_buf_stale(struct xfs_buf * bp)91 xfs_buf_stale(
92 	struct xfs_buf	*bp)
93 {
94 	ASSERT(xfs_buf_islocked(bp));
95 
96 	bp->b_flags |= XBF_STALE;
97 
98 	/*
99 	 * Clear the delwri status so that a delwri queue walker will not
100 	 * flush this buffer to disk now that it is stale. The delwri queue has
101 	 * a reference to the buffer, so this is safe to do.
102 	 */
103 	bp->b_flags &= ~_XBF_DELWRI_Q;
104 
105 	spin_lock(&bp->b_lock);
106 	atomic_set(&bp->b_lru_ref, 0);
107 	if (!(bp->b_state & XFS_BSTATE_DISPOSE) &&
108 	    (list_lru_del(&bp->b_target->bt_lru, &bp->b_lru)))
109 		atomic_dec(&bp->b_hold);
110 
111 	ASSERT(atomic_read(&bp->b_hold) >= 1);
112 	spin_unlock(&bp->b_lock);
113 }
114 
115 static int
xfs_buf_get_maps(struct xfs_buf * bp,int map_count)116 xfs_buf_get_maps(
117 	struct xfs_buf		*bp,
118 	int			map_count)
119 {
120 	ASSERT(bp->b_maps == NULL);
121 	bp->b_map_count = map_count;
122 
123 	if (map_count == 1) {
124 		bp->b_maps = &bp->__b_map;
125 		return 0;
126 	}
127 
128 	bp->b_maps = kmem_zalloc(map_count * sizeof(struct xfs_buf_map),
129 				KM_NOFS);
130 	if (!bp->b_maps)
131 		return -ENOMEM;
132 	return 0;
133 }
134 
135 /*
136  *	Frees b_pages if it was allocated.
137  */
138 static void
xfs_buf_free_maps(struct xfs_buf * bp)139 xfs_buf_free_maps(
140 	struct xfs_buf	*bp)
141 {
142 	if (bp->b_maps != &bp->__b_map) {
143 		kmem_free(bp->b_maps);
144 		bp->b_maps = NULL;
145 	}
146 }
147 
148 struct xfs_buf *
_xfs_buf_alloc(struct xfs_buftarg * target,struct xfs_buf_map * map,int nmaps,xfs_buf_flags_t flags)149 _xfs_buf_alloc(
150 	struct xfs_buftarg	*target,
151 	struct xfs_buf_map	*map,
152 	int			nmaps,
153 	xfs_buf_flags_t		flags)
154 {
155 	struct xfs_buf		*bp;
156 	int			error;
157 	int			i;
158 
159 	bp = kmem_zone_zalloc(xfs_buf_zone, KM_NOFS);
160 	if (unlikely(!bp))
161 		return NULL;
162 
163 	/*
164 	 * We don't want certain flags to appear in b_flags unless they are
165 	 * specifically set by later operations on the buffer.
166 	 */
167 	flags &= ~(XBF_UNMAPPED | XBF_TRYLOCK | XBF_ASYNC | XBF_READ_AHEAD);
168 
169 	atomic_set(&bp->b_hold, 1);
170 	atomic_set(&bp->b_lru_ref, 1);
171 	init_completion(&bp->b_iowait);
172 	INIT_LIST_HEAD(&bp->b_lru);
173 	INIT_LIST_HEAD(&bp->b_list);
174 	RB_CLEAR_NODE(&bp->b_rbnode);
175 	sema_init(&bp->b_sema, 0); /* held, no waiters */
176 	spin_lock_init(&bp->b_lock);
177 	XB_SET_OWNER(bp);
178 	bp->b_target = target;
179 	bp->b_flags = flags;
180 
181 	/*
182 	 * Set length and io_length to the same value initially.
183 	 * I/O routines should use io_length, which will be the same in
184 	 * most cases but may be reset (e.g. XFS recovery).
185 	 */
186 	error = xfs_buf_get_maps(bp, nmaps);
187 	if (error)  {
188 		kmem_zone_free(xfs_buf_zone, bp);
189 		return NULL;
190 	}
191 
192 	bp->b_bn = map[0].bm_bn;
193 	bp->b_length = 0;
194 	for (i = 0; i < nmaps; i++) {
195 		bp->b_maps[i].bm_bn = map[i].bm_bn;
196 		bp->b_maps[i].bm_len = map[i].bm_len;
197 		bp->b_length += map[i].bm_len;
198 	}
199 	bp->b_io_length = bp->b_length;
200 
201 	atomic_set(&bp->b_pin_count, 0);
202 	init_waitqueue_head(&bp->b_waiters);
203 
204 	XFS_STATS_INC(target->bt_mount, xb_create);
205 	trace_xfs_buf_init(bp, _RET_IP_);
206 
207 	return bp;
208 }
209 
210 /*
211  *	Allocate a page array capable of holding a specified number
212  *	of pages, and point the page buf at it.
213  */
214 STATIC int
_xfs_buf_get_pages(xfs_buf_t * bp,int page_count)215 _xfs_buf_get_pages(
216 	xfs_buf_t		*bp,
217 	int			page_count)
218 {
219 	/* Make sure that we have a page list */
220 	if (bp->b_pages == NULL) {
221 		bp->b_page_count = page_count;
222 		if (page_count <= XB_PAGES) {
223 			bp->b_pages = bp->b_page_array;
224 		} else {
225 			bp->b_pages = kmem_alloc(sizeof(struct page *) *
226 						 page_count, KM_NOFS);
227 			if (bp->b_pages == NULL)
228 				return -ENOMEM;
229 		}
230 		memset(bp->b_pages, 0, sizeof(struct page *) * page_count);
231 	}
232 	return 0;
233 }
234 
235 /*
236  *	Frees b_pages if it was allocated.
237  */
238 STATIC void
_xfs_buf_free_pages(xfs_buf_t * bp)239 _xfs_buf_free_pages(
240 	xfs_buf_t	*bp)
241 {
242 	if (bp->b_pages != bp->b_page_array) {
243 		kmem_free(bp->b_pages);
244 		bp->b_pages = NULL;
245 	}
246 }
247 
248 /*
249  *	Releases the specified buffer.
250  *
251  * 	The modification state of any associated pages is left unchanged.
252  * 	The buffer must not be on any hash - use xfs_buf_rele instead for
253  * 	hashed and refcounted buffers
254  */
255 void
xfs_buf_free(xfs_buf_t * bp)256 xfs_buf_free(
257 	xfs_buf_t		*bp)
258 {
259 	trace_xfs_buf_free(bp, _RET_IP_);
260 
261 	ASSERT(list_empty(&bp->b_lru));
262 
263 	if (bp->b_flags & _XBF_PAGES) {
264 		uint		i;
265 
266 		if (xfs_buf_is_vmapped(bp))
267 			vm_unmap_ram(bp->b_addr - bp->b_offset,
268 					bp->b_page_count);
269 
270 		for (i = 0; i < bp->b_page_count; i++) {
271 			struct page	*page = bp->b_pages[i];
272 
273 			__free_page(page);
274 		}
275 	} else if (bp->b_flags & _XBF_KMEM)
276 		kmem_free(bp->b_addr);
277 	_xfs_buf_free_pages(bp);
278 	xfs_buf_free_maps(bp);
279 	kmem_zone_free(xfs_buf_zone, bp);
280 }
281 
282 /*
283  * Allocates all the pages for buffer in question and builds it's page list.
284  */
285 STATIC int
xfs_buf_allocate_memory(xfs_buf_t * bp,uint flags)286 xfs_buf_allocate_memory(
287 	xfs_buf_t		*bp,
288 	uint			flags)
289 {
290 	size_t			size;
291 	size_t			nbytes, offset;
292 	gfp_t			gfp_mask = xb_to_gfp(flags);
293 	unsigned short		page_count, i;
294 	xfs_off_t		start, end;
295 	int			error;
296 
297 	/*
298 	 * for buffers that are contained within a single page, just allocate
299 	 * the memory from the heap - there's no need for the complexity of
300 	 * page arrays to keep allocation down to order 0.
301 	 */
302 	size = BBTOB(bp->b_length);
303 	if (size < PAGE_SIZE) {
304 		bp->b_addr = kmem_alloc(size, KM_NOFS);
305 		if (!bp->b_addr) {
306 			/* low memory - use alloc_page loop instead */
307 			goto use_alloc_page;
308 		}
309 
310 		if (((unsigned long)(bp->b_addr + size - 1) & PAGE_MASK) !=
311 		    ((unsigned long)bp->b_addr & PAGE_MASK)) {
312 			/* b_addr spans two pages - use alloc_page instead */
313 			kmem_free(bp->b_addr);
314 			bp->b_addr = NULL;
315 			goto use_alloc_page;
316 		}
317 		bp->b_offset = offset_in_page(bp->b_addr);
318 		bp->b_pages = bp->b_page_array;
319 		bp->b_pages[0] = virt_to_page(bp->b_addr);
320 		bp->b_page_count = 1;
321 		bp->b_flags |= _XBF_KMEM;
322 		return 0;
323 	}
324 
325 use_alloc_page:
326 	start = BBTOB(bp->b_maps[0].bm_bn) >> PAGE_SHIFT;
327 	end = (BBTOB(bp->b_maps[0].bm_bn + bp->b_length) + PAGE_SIZE - 1)
328 								>> PAGE_SHIFT;
329 	page_count = end - start;
330 	error = _xfs_buf_get_pages(bp, page_count);
331 	if (unlikely(error))
332 		return error;
333 
334 	offset = bp->b_offset;
335 	bp->b_flags |= _XBF_PAGES;
336 
337 	for (i = 0; i < bp->b_page_count; i++) {
338 		struct page	*page;
339 		uint		retries = 0;
340 retry:
341 		page = alloc_page(gfp_mask);
342 		if (unlikely(page == NULL)) {
343 			if (flags & XBF_READ_AHEAD) {
344 				bp->b_page_count = i;
345 				error = -ENOMEM;
346 				goto out_free_pages;
347 			}
348 
349 			/*
350 			 * This could deadlock.
351 			 *
352 			 * But until all the XFS lowlevel code is revamped to
353 			 * handle buffer allocation failures we can't do much.
354 			 */
355 			if (!(++retries % 100))
356 				xfs_err(NULL,
357 		"%s(%u) possible memory allocation deadlock in %s (mode:0x%x)",
358 					current->comm, current->pid,
359 					__func__, gfp_mask);
360 
361 			XFS_STATS_INC(bp->b_target->bt_mount, xb_page_retries);
362 			congestion_wait(BLK_RW_ASYNC, HZ/50);
363 			goto retry;
364 		}
365 
366 		XFS_STATS_INC(bp->b_target->bt_mount, xb_page_found);
367 
368 		nbytes = min_t(size_t, size, PAGE_SIZE - offset);
369 		size -= nbytes;
370 		bp->b_pages[i] = page;
371 		offset = 0;
372 	}
373 	return 0;
374 
375 out_free_pages:
376 	for (i = 0; i < bp->b_page_count; i++)
377 		__free_page(bp->b_pages[i]);
378 	bp->b_flags &= ~_XBF_PAGES;
379 	return error;
380 }
381 
382 /*
383  *	Map buffer into kernel address-space if necessary.
384  */
385 STATIC int
_xfs_buf_map_pages(xfs_buf_t * bp,uint flags)386 _xfs_buf_map_pages(
387 	xfs_buf_t		*bp,
388 	uint			flags)
389 {
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 	} else if (flags & XBF_UNMAPPED) {
395 		bp->b_addr = NULL;
396 	} else {
397 		int retried = 0;
398 		unsigned noio_flag;
399 
400 		/*
401 		 * vm_map_ram() will allocate auxillary structures (e.g.
402 		 * pagetables) with GFP_KERNEL, yet we are likely to be under
403 		 * GFP_NOFS context here. Hence we need to tell memory reclaim
404 		 * that we are in such a context via PF_MEMALLOC_NOIO to prevent
405 		 * memory reclaim re-entering the filesystem here and
406 		 * potentially deadlocking.
407 		 */
408 		noio_flag = memalloc_noio_save();
409 		do {
410 			bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count,
411 						-1, PAGE_KERNEL);
412 			if (bp->b_addr)
413 				break;
414 			vm_unmap_aliases();
415 		} while (retried++ <= 1);
416 		memalloc_noio_restore(noio_flag);
417 
418 		if (!bp->b_addr)
419 			return -ENOMEM;
420 		bp->b_addr += bp->b_offset;
421 	}
422 
423 	return 0;
424 }
425 
426 /*
427  *	Finding and Reading Buffers
428  */
429 
430 /*
431  *	Look up, and creates if absent, a lockable buffer for
432  *	a given range of an inode.  The buffer is returned
433  *	locked.	No I/O is implied by this call.
434  */
435 xfs_buf_t *
_xfs_buf_find(struct xfs_buftarg * btp,struct xfs_buf_map * map,int nmaps,xfs_buf_flags_t flags,xfs_buf_t * new_bp)436 _xfs_buf_find(
437 	struct xfs_buftarg	*btp,
438 	struct xfs_buf_map	*map,
439 	int			nmaps,
440 	xfs_buf_flags_t		flags,
441 	xfs_buf_t		*new_bp)
442 {
443 	struct xfs_perag	*pag;
444 	struct rb_node		**rbp;
445 	struct rb_node		*parent;
446 	xfs_buf_t		*bp;
447 	xfs_daddr_t		blkno = map[0].bm_bn;
448 	xfs_daddr_t		eofs;
449 	int			numblks = 0;
450 	int			i;
451 
452 	for (i = 0; i < nmaps; i++)
453 		numblks += map[i].bm_len;
454 
455 	/* Check for IOs smaller than the sector size / not sector aligned */
456 	ASSERT(!(BBTOB(numblks) < btp->bt_meta_sectorsize));
457 	ASSERT(!(BBTOB(blkno) & (xfs_off_t)btp->bt_meta_sectormask));
458 
459 	/*
460 	 * Corrupted block numbers can get through to here, unfortunately, so we
461 	 * have to check that the buffer falls within the filesystem bounds.
462 	 */
463 	eofs = XFS_FSB_TO_BB(btp->bt_mount, btp->bt_mount->m_sb.sb_dblocks);
464 	if (blkno < 0 || blkno >= eofs) {
465 		/*
466 		 * XXX (dgc): we should really be returning -EFSCORRUPTED here,
467 		 * but none of the higher level infrastructure supports
468 		 * returning a specific error on buffer lookup failures.
469 		 */
470 		xfs_alert(btp->bt_mount,
471 			  "%s: Block out of range: block 0x%llx, EOFS 0x%llx ",
472 			  __func__, blkno, eofs);
473 		WARN_ON(1);
474 		return NULL;
475 	}
476 
477 	/* get tree root */
478 	pag = xfs_perag_get(btp->bt_mount,
479 				xfs_daddr_to_agno(btp->bt_mount, blkno));
480 
481 	/* walk tree */
482 	spin_lock(&pag->pag_buf_lock);
483 	rbp = &pag->pag_buf_tree.rb_node;
484 	parent = NULL;
485 	bp = NULL;
486 	while (*rbp) {
487 		parent = *rbp;
488 		bp = rb_entry(parent, struct xfs_buf, b_rbnode);
489 
490 		if (blkno < bp->b_bn)
491 			rbp = &(*rbp)->rb_left;
492 		else if (blkno > bp->b_bn)
493 			rbp = &(*rbp)->rb_right;
494 		else {
495 			/*
496 			 * found a block number match. If the range doesn't
497 			 * match, the only way this is allowed is if the buffer
498 			 * in the cache is stale and the transaction that made
499 			 * it stale has not yet committed. i.e. we are
500 			 * reallocating a busy extent. Skip this buffer and
501 			 * continue searching to the right for an exact match.
502 			 */
503 			if (bp->b_length != numblks) {
504 				ASSERT(bp->b_flags & XBF_STALE);
505 				rbp = &(*rbp)->rb_right;
506 				continue;
507 			}
508 			atomic_inc(&bp->b_hold);
509 			goto found;
510 		}
511 	}
512 
513 	/* No match found */
514 	if (new_bp) {
515 		rb_link_node(&new_bp->b_rbnode, parent, rbp);
516 		rb_insert_color(&new_bp->b_rbnode, &pag->pag_buf_tree);
517 		/* the buffer keeps the perag reference until it is freed */
518 		new_bp->b_pag = pag;
519 		spin_unlock(&pag->pag_buf_lock);
520 	} else {
521 		XFS_STATS_INC(btp->bt_mount, xb_miss_locked);
522 		spin_unlock(&pag->pag_buf_lock);
523 		xfs_perag_put(pag);
524 	}
525 	return new_bp;
526 
527 found:
528 	spin_unlock(&pag->pag_buf_lock);
529 	xfs_perag_put(pag);
530 
531 	if (!xfs_buf_trylock(bp)) {
532 		if (flags & XBF_TRYLOCK) {
533 			xfs_buf_rele(bp);
534 			XFS_STATS_INC(btp->bt_mount, xb_busy_locked);
535 			return NULL;
536 		}
537 		xfs_buf_lock(bp);
538 		XFS_STATS_INC(btp->bt_mount, xb_get_locked_waited);
539 	}
540 
541 	/*
542 	 * if the buffer is stale, clear all the external state associated with
543 	 * it. We need to keep flags such as how we allocated the buffer memory
544 	 * intact here.
545 	 */
546 	if (bp->b_flags & XBF_STALE) {
547 		ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0);
548 		ASSERT(bp->b_iodone == NULL);
549 		bp->b_flags &= _XBF_KMEM | _XBF_PAGES;
550 		bp->b_ops = NULL;
551 	}
552 
553 	trace_xfs_buf_find(bp, flags, _RET_IP_);
554 	XFS_STATS_INC(btp->bt_mount, xb_get_locked);
555 	return bp;
556 }
557 
558 /*
559  * Assembles a buffer covering the specified range. The code is optimised for
560  * cache hits, as metadata intensive workloads will see 3 orders of magnitude
561  * more hits than misses.
562  */
563 struct xfs_buf *
xfs_buf_get_map(struct xfs_buftarg * target,struct xfs_buf_map * map,int nmaps,xfs_buf_flags_t flags)564 xfs_buf_get_map(
565 	struct xfs_buftarg	*target,
566 	struct xfs_buf_map	*map,
567 	int			nmaps,
568 	xfs_buf_flags_t		flags)
569 {
570 	struct xfs_buf		*bp;
571 	struct xfs_buf		*new_bp;
572 	int			error = 0;
573 
574 	bp = _xfs_buf_find(target, map, nmaps, flags, NULL);
575 	if (likely(bp))
576 		goto found;
577 
578 	new_bp = _xfs_buf_alloc(target, map, nmaps, flags);
579 	if (unlikely(!new_bp))
580 		return NULL;
581 
582 	error = xfs_buf_allocate_memory(new_bp, flags);
583 	if (error) {
584 		xfs_buf_free(new_bp);
585 		return NULL;
586 	}
587 
588 	bp = _xfs_buf_find(target, map, nmaps, flags, new_bp);
589 	if (!bp) {
590 		xfs_buf_free(new_bp);
591 		return NULL;
592 	}
593 
594 	if (bp != new_bp)
595 		xfs_buf_free(new_bp);
596 
597 found:
598 	if (!bp->b_addr) {
599 		error = _xfs_buf_map_pages(bp, flags);
600 		if (unlikely(error)) {
601 			xfs_warn(target->bt_mount,
602 				"%s: failed to map pagesn", __func__);
603 			xfs_buf_relse(bp);
604 			return NULL;
605 		}
606 	}
607 
608 	/*
609 	 * Clear b_error if this is a lookup from a caller that doesn't expect
610 	 * valid data to be found in the buffer.
611 	 */
612 	if (!(flags & XBF_READ))
613 		xfs_buf_ioerror(bp, 0);
614 
615 	XFS_STATS_INC(target->bt_mount, xb_get);
616 	trace_xfs_buf_get(bp, flags, _RET_IP_);
617 	return bp;
618 }
619 
620 STATIC int
_xfs_buf_read(xfs_buf_t * bp,xfs_buf_flags_t flags)621 _xfs_buf_read(
622 	xfs_buf_t		*bp,
623 	xfs_buf_flags_t		flags)
624 {
625 	ASSERT(!(flags & XBF_WRITE));
626 	ASSERT(bp->b_maps[0].bm_bn != XFS_BUF_DADDR_NULL);
627 
628 	bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_READ_AHEAD);
629 	bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD);
630 
631 	if (flags & XBF_ASYNC) {
632 		xfs_buf_submit(bp);
633 		return 0;
634 	}
635 	return xfs_buf_submit_wait(bp);
636 }
637 
638 xfs_buf_t *
xfs_buf_read_map(struct xfs_buftarg * target,struct xfs_buf_map * map,int nmaps,xfs_buf_flags_t flags,const struct xfs_buf_ops * ops)639 xfs_buf_read_map(
640 	struct xfs_buftarg	*target,
641 	struct xfs_buf_map	*map,
642 	int			nmaps,
643 	xfs_buf_flags_t		flags,
644 	const struct xfs_buf_ops *ops)
645 {
646 	struct xfs_buf		*bp;
647 
648 	flags |= XBF_READ;
649 
650 	bp = xfs_buf_get_map(target, map, nmaps, flags);
651 	if (bp) {
652 		trace_xfs_buf_read(bp, flags, _RET_IP_);
653 
654 		if (!XFS_BUF_ISDONE(bp)) {
655 			XFS_STATS_INC(target->bt_mount, xb_get_read);
656 			bp->b_ops = ops;
657 			_xfs_buf_read(bp, flags);
658 		} else if (flags & XBF_ASYNC) {
659 			/*
660 			 * Read ahead call which is already satisfied,
661 			 * drop the buffer
662 			 */
663 			xfs_buf_relse(bp);
664 			return NULL;
665 		} else {
666 			/* We do not want read in the flags */
667 			bp->b_flags &= ~XBF_READ;
668 		}
669 	}
670 
671 	return bp;
672 }
673 
674 /*
675  *	If we are not low on memory then do the readahead in a deadlock
676  *	safe manner.
677  */
678 void
xfs_buf_readahead_map(struct xfs_buftarg * target,struct xfs_buf_map * map,int nmaps,const struct xfs_buf_ops * ops)679 xfs_buf_readahead_map(
680 	struct xfs_buftarg	*target,
681 	struct xfs_buf_map	*map,
682 	int			nmaps,
683 	const struct xfs_buf_ops *ops)
684 {
685 	if (bdi_read_congested(target->bt_bdi))
686 		return;
687 
688 	xfs_buf_read_map(target, map, nmaps,
689 		     XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD, ops);
690 }
691 
692 /*
693  * Read an uncached buffer from disk. Allocates and returns a locked
694  * buffer containing the disk contents or nothing.
695  */
696 int
xfs_buf_read_uncached(struct xfs_buftarg * target,xfs_daddr_t daddr,size_t numblks,int flags,struct xfs_buf ** bpp,const struct xfs_buf_ops * ops)697 xfs_buf_read_uncached(
698 	struct xfs_buftarg	*target,
699 	xfs_daddr_t		daddr,
700 	size_t			numblks,
701 	int			flags,
702 	struct xfs_buf		**bpp,
703 	const struct xfs_buf_ops *ops)
704 {
705 	struct xfs_buf		*bp;
706 
707 	*bpp = NULL;
708 
709 	bp = xfs_buf_get_uncached(target, numblks, flags);
710 	if (!bp)
711 		return -ENOMEM;
712 
713 	/* set up the buffer for a read IO */
714 	ASSERT(bp->b_map_count == 1);
715 	bp->b_bn = XFS_BUF_DADDR_NULL;  /* always null for uncached buffers */
716 	bp->b_maps[0].bm_bn = daddr;
717 	bp->b_flags |= XBF_READ;
718 	bp->b_ops = ops;
719 
720 	xfs_buf_submit_wait(bp);
721 	if (bp->b_error) {
722 		int	error = bp->b_error;
723 		xfs_buf_relse(bp);
724 		return error;
725 	}
726 
727 	*bpp = bp;
728 	return 0;
729 }
730 
731 /*
732  * Return a buffer allocated as an empty buffer and associated to external
733  * memory via xfs_buf_associate_memory() back to it's empty state.
734  */
735 void
xfs_buf_set_empty(struct xfs_buf * bp,size_t numblks)736 xfs_buf_set_empty(
737 	struct xfs_buf		*bp,
738 	size_t			numblks)
739 {
740 	if (bp->b_pages)
741 		_xfs_buf_free_pages(bp);
742 
743 	bp->b_pages = NULL;
744 	bp->b_page_count = 0;
745 	bp->b_addr = NULL;
746 	bp->b_length = numblks;
747 	bp->b_io_length = numblks;
748 
749 	ASSERT(bp->b_map_count == 1);
750 	bp->b_bn = XFS_BUF_DADDR_NULL;
751 	bp->b_maps[0].bm_bn = XFS_BUF_DADDR_NULL;
752 	bp->b_maps[0].bm_len = bp->b_length;
753 }
754 
755 static inline struct page *
mem_to_page(void * addr)756 mem_to_page(
757 	void			*addr)
758 {
759 	if ((!is_vmalloc_addr(addr))) {
760 		return virt_to_page(addr);
761 	} else {
762 		return vmalloc_to_page(addr);
763 	}
764 }
765 
766 int
xfs_buf_associate_memory(xfs_buf_t * bp,void * mem,size_t len)767 xfs_buf_associate_memory(
768 	xfs_buf_t		*bp,
769 	void			*mem,
770 	size_t			len)
771 {
772 	int			rval;
773 	int			i = 0;
774 	unsigned long		pageaddr;
775 	unsigned long		offset;
776 	size_t			buflen;
777 	int			page_count;
778 
779 	pageaddr = (unsigned long)mem & PAGE_MASK;
780 	offset = (unsigned long)mem - pageaddr;
781 	buflen = PAGE_ALIGN(len + offset);
782 	page_count = buflen >> PAGE_SHIFT;
783 
784 	/* Free any previous set of page pointers */
785 	if (bp->b_pages)
786 		_xfs_buf_free_pages(bp);
787 
788 	bp->b_pages = NULL;
789 	bp->b_addr = mem;
790 
791 	rval = _xfs_buf_get_pages(bp, page_count);
792 	if (rval)
793 		return rval;
794 
795 	bp->b_offset = offset;
796 
797 	for (i = 0; i < bp->b_page_count; i++) {
798 		bp->b_pages[i] = mem_to_page((void *)pageaddr);
799 		pageaddr += PAGE_SIZE;
800 	}
801 
802 	bp->b_io_length = BTOBB(len);
803 	bp->b_length = BTOBB(buflen);
804 
805 	return 0;
806 }
807 
808 xfs_buf_t *
xfs_buf_get_uncached(struct xfs_buftarg * target,size_t numblks,int flags)809 xfs_buf_get_uncached(
810 	struct xfs_buftarg	*target,
811 	size_t			numblks,
812 	int			flags)
813 {
814 	unsigned long		page_count;
815 	int			error, i;
816 	struct xfs_buf		*bp;
817 	DEFINE_SINGLE_BUF_MAP(map, XFS_BUF_DADDR_NULL, numblks);
818 
819 	bp = _xfs_buf_alloc(target, &map, 1, 0);
820 	if (unlikely(bp == NULL))
821 		goto fail;
822 
823 	page_count = PAGE_ALIGN(numblks << BBSHIFT) >> PAGE_SHIFT;
824 	error = _xfs_buf_get_pages(bp, page_count);
825 	if (error)
826 		goto fail_free_buf;
827 
828 	for (i = 0; i < page_count; i++) {
829 		bp->b_pages[i] = alloc_page(xb_to_gfp(flags));
830 		if (!bp->b_pages[i])
831 			goto fail_free_mem;
832 	}
833 	bp->b_flags |= _XBF_PAGES;
834 
835 	error = _xfs_buf_map_pages(bp, 0);
836 	if (unlikely(error)) {
837 		xfs_warn(target->bt_mount,
838 			"%s: failed to map pages", __func__);
839 		goto fail_free_mem;
840 	}
841 
842 	trace_xfs_buf_get_uncached(bp, _RET_IP_);
843 	return bp;
844 
845  fail_free_mem:
846 	while (--i >= 0)
847 		__free_page(bp->b_pages[i]);
848 	_xfs_buf_free_pages(bp);
849  fail_free_buf:
850 	xfs_buf_free_maps(bp);
851 	kmem_zone_free(xfs_buf_zone, bp);
852  fail:
853 	return NULL;
854 }
855 
856 /*
857  *	Increment reference count on buffer, to hold the buffer concurrently
858  *	with another thread which may release (free) the buffer asynchronously.
859  *	Must hold the buffer already to call this function.
860  */
861 void
xfs_buf_hold(xfs_buf_t * bp)862 xfs_buf_hold(
863 	xfs_buf_t		*bp)
864 {
865 	trace_xfs_buf_hold(bp, _RET_IP_);
866 	atomic_inc(&bp->b_hold);
867 }
868 
869 /*
870  *	Releases a hold on the specified buffer.  If the
871  *	the hold count is 1, calls xfs_buf_free.
872  */
873 void
xfs_buf_rele(xfs_buf_t * bp)874 xfs_buf_rele(
875 	xfs_buf_t		*bp)
876 {
877 	struct xfs_perag	*pag = bp->b_pag;
878 
879 	trace_xfs_buf_rele(bp, _RET_IP_);
880 
881 	if (!pag) {
882 		ASSERT(list_empty(&bp->b_lru));
883 		ASSERT(RB_EMPTY_NODE(&bp->b_rbnode));
884 		if (atomic_dec_and_test(&bp->b_hold))
885 			xfs_buf_free(bp);
886 		return;
887 	}
888 
889 	ASSERT(!RB_EMPTY_NODE(&bp->b_rbnode));
890 
891 	ASSERT(atomic_read(&bp->b_hold) > 0);
892 	if (atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock)) {
893 		spin_lock(&bp->b_lock);
894 		if (!(bp->b_flags & XBF_STALE) && atomic_read(&bp->b_lru_ref)) {
895 			/*
896 			 * If the buffer is added to the LRU take a new
897 			 * reference to the buffer for the LRU and clear the
898 			 * (now stale) dispose list state flag
899 			 */
900 			if (list_lru_add(&bp->b_target->bt_lru, &bp->b_lru)) {
901 				bp->b_state &= ~XFS_BSTATE_DISPOSE;
902 				atomic_inc(&bp->b_hold);
903 			}
904 			spin_unlock(&bp->b_lock);
905 			spin_unlock(&pag->pag_buf_lock);
906 		} else {
907 			/*
908 			 * most of the time buffers will already be removed from
909 			 * the LRU, so optimise that case by checking for the
910 			 * XFS_BSTATE_DISPOSE flag indicating the last list the
911 			 * buffer was on was the disposal list
912 			 */
913 			if (!(bp->b_state & XFS_BSTATE_DISPOSE)) {
914 				list_lru_del(&bp->b_target->bt_lru, &bp->b_lru);
915 			} else {
916 				ASSERT(list_empty(&bp->b_lru));
917 			}
918 			spin_unlock(&bp->b_lock);
919 
920 			ASSERT(!(bp->b_flags & _XBF_DELWRI_Q));
921 			rb_erase(&bp->b_rbnode, &pag->pag_buf_tree);
922 			spin_unlock(&pag->pag_buf_lock);
923 			xfs_perag_put(pag);
924 			xfs_buf_free(bp);
925 		}
926 	}
927 }
928 
929 
930 /*
931  *	Lock a buffer object, if it is not already locked.
932  *
933  *	If we come across a stale, pinned, locked buffer, we know that we are
934  *	being asked to lock a buffer that has been reallocated. Because it is
935  *	pinned, we know that the log has not been pushed to disk and hence it
936  *	will still be locked.  Rather than continuing to have trylock attempts
937  *	fail until someone else pushes the log, push it ourselves before
938  *	returning.  This means that the xfsaild will not get stuck trying
939  *	to push on stale inode buffers.
940  */
941 int
xfs_buf_trylock(struct xfs_buf * bp)942 xfs_buf_trylock(
943 	struct xfs_buf		*bp)
944 {
945 	int			locked;
946 
947 	locked = down_trylock(&bp->b_sema) == 0;
948 	if (locked)
949 		XB_SET_OWNER(bp);
950 
951 	trace_xfs_buf_trylock(bp, _RET_IP_);
952 	return locked;
953 }
954 
955 /*
956  *	Lock a buffer object.
957  *
958  *	If we come across a stale, pinned, locked buffer, we know that we
959  *	are being asked to lock a buffer that has been reallocated. Because
960  *	it is pinned, we know that the log has not been pushed to disk and
961  *	hence it will still be locked. Rather than sleeping until someone
962  *	else pushes the log, push it ourselves before trying to get the lock.
963  */
964 void
xfs_buf_lock(struct xfs_buf * bp)965 xfs_buf_lock(
966 	struct xfs_buf		*bp)
967 {
968 	trace_xfs_buf_lock(bp, _RET_IP_);
969 
970 	if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))
971 		xfs_log_force(bp->b_target->bt_mount, 0);
972 	down(&bp->b_sema);
973 	XB_SET_OWNER(bp);
974 
975 	trace_xfs_buf_lock_done(bp, _RET_IP_);
976 }
977 
978 void
xfs_buf_unlock(struct xfs_buf * bp)979 xfs_buf_unlock(
980 	struct xfs_buf		*bp)
981 {
982 	ASSERT(xfs_buf_islocked(bp));
983 
984 	XB_CLEAR_OWNER(bp);
985 	up(&bp->b_sema);
986 
987 	trace_xfs_buf_unlock(bp, _RET_IP_);
988 }
989 
990 STATIC void
xfs_buf_wait_unpin(xfs_buf_t * bp)991 xfs_buf_wait_unpin(
992 	xfs_buf_t		*bp)
993 {
994 	DECLARE_WAITQUEUE	(wait, current);
995 
996 	if (atomic_read(&bp->b_pin_count) == 0)
997 		return;
998 
999 	add_wait_queue(&bp->b_waiters, &wait);
1000 	for (;;) {
1001 		set_current_state(TASK_UNINTERRUPTIBLE);
1002 		if (atomic_read(&bp->b_pin_count) == 0)
1003 			break;
1004 		io_schedule();
1005 	}
1006 	remove_wait_queue(&bp->b_waiters, &wait);
1007 	set_current_state(TASK_RUNNING);
1008 }
1009 
1010 /*
1011  *	Buffer Utility Routines
1012  */
1013 
1014 void
xfs_buf_ioend(struct xfs_buf * bp)1015 xfs_buf_ioend(
1016 	struct xfs_buf	*bp)
1017 {
1018 	bool		read = bp->b_flags & XBF_READ;
1019 
1020 	trace_xfs_buf_iodone(bp, _RET_IP_);
1021 
1022 	bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD);
1023 
1024 	/*
1025 	 * Pull in IO completion errors now. We are guaranteed to be running
1026 	 * single threaded, so we don't need the lock to read b_io_error.
1027 	 */
1028 	if (!bp->b_error && bp->b_io_error)
1029 		xfs_buf_ioerror(bp, bp->b_io_error);
1030 
1031 	/* Only validate buffers that were read without errors */
1032 	if (read && !bp->b_error && bp->b_ops) {
1033 		ASSERT(!bp->b_iodone);
1034 		bp->b_ops->verify_read(bp);
1035 	}
1036 
1037 	if (!bp->b_error)
1038 		bp->b_flags |= XBF_DONE;
1039 
1040 	if (bp->b_iodone)
1041 		(*(bp->b_iodone))(bp);
1042 	else if (bp->b_flags & XBF_ASYNC)
1043 		xfs_buf_relse(bp);
1044 	else
1045 		complete(&bp->b_iowait);
1046 }
1047 
1048 static void
xfs_buf_ioend_work(struct work_struct * work)1049 xfs_buf_ioend_work(
1050 	struct work_struct	*work)
1051 {
1052 	struct xfs_buf		*bp =
1053 		container_of(work, xfs_buf_t, b_ioend_work);
1054 
1055 	xfs_buf_ioend(bp);
1056 }
1057 
1058 void
xfs_buf_ioend_async(struct xfs_buf * bp)1059 xfs_buf_ioend_async(
1060 	struct xfs_buf	*bp)
1061 {
1062 	INIT_WORK(&bp->b_ioend_work, xfs_buf_ioend_work);
1063 	queue_work(bp->b_ioend_wq, &bp->b_ioend_work);
1064 }
1065 
1066 void
xfs_buf_ioerror(xfs_buf_t * bp,int error)1067 xfs_buf_ioerror(
1068 	xfs_buf_t		*bp,
1069 	int			error)
1070 {
1071 	ASSERT(error <= 0 && error >= -1000);
1072 	bp->b_error = error;
1073 	trace_xfs_buf_ioerror(bp, error, _RET_IP_);
1074 }
1075 
1076 void
xfs_buf_ioerror_alert(struct xfs_buf * bp,const char * func)1077 xfs_buf_ioerror_alert(
1078 	struct xfs_buf		*bp,
1079 	const char		*func)
1080 {
1081 	xfs_alert(bp->b_target->bt_mount,
1082 "metadata I/O error: block 0x%llx (\"%s\") error %d numblks %d",
1083 		(__uint64_t)XFS_BUF_ADDR(bp), func, -bp->b_error, bp->b_length);
1084 }
1085 
1086 int
xfs_bwrite(struct xfs_buf * bp)1087 xfs_bwrite(
1088 	struct xfs_buf		*bp)
1089 {
1090 	int			error;
1091 
1092 	ASSERT(xfs_buf_islocked(bp));
1093 
1094 	bp->b_flags |= XBF_WRITE;
1095 	bp->b_flags &= ~(XBF_ASYNC | XBF_READ | _XBF_DELWRI_Q |
1096 			 XBF_WRITE_FAIL | XBF_DONE);
1097 
1098 	error = xfs_buf_submit_wait(bp);
1099 	if (error) {
1100 		xfs_force_shutdown(bp->b_target->bt_mount,
1101 				   SHUTDOWN_META_IO_ERROR);
1102 	}
1103 	return error;
1104 }
1105 
1106 STATIC void
xfs_buf_bio_end_io(struct bio * bio)1107 xfs_buf_bio_end_io(
1108 	struct bio		*bio)
1109 {
1110 	xfs_buf_t		*bp = (xfs_buf_t *)bio->bi_private;
1111 
1112 	/*
1113 	 * don't overwrite existing errors - otherwise we can lose errors on
1114 	 * buffers that require multiple bios to complete.
1115 	 */
1116 	if (bio->bi_error) {
1117 		spin_lock(&bp->b_lock);
1118 		if (!bp->b_io_error)
1119 			bp->b_io_error = bio->bi_error;
1120 		spin_unlock(&bp->b_lock);
1121 	}
1122 
1123 	if (!bp->b_error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ))
1124 		invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp));
1125 
1126 	if (atomic_dec_and_test(&bp->b_io_remaining) == 1)
1127 		xfs_buf_ioend_async(bp);
1128 	bio_put(bio);
1129 }
1130 
1131 static void
xfs_buf_ioapply_map(struct xfs_buf * bp,int map,int * buf_offset,int * count,int rw)1132 xfs_buf_ioapply_map(
1133 	struct xfs_buf	*bp,
1134 	int		map,
1135 	int		*buf_offset,
1136 	int		*count,
1137 	int		rw)
1138 {
1139 	int		page_index;
1140 	int		total_nr_pages = bp->b_page_count;
1141 	int		nr_pages;
1142 	struct bio	*bio;
1143 	sector_t	sector =  bp->b_maps[map].bm_bn;
1144 	int		size;
1145 	int		offset;
1146 
1147 	total_nr_pages = bp->b_page_count;
1148 
1149 	/* skip the pages in the buffer before the start offset */
1150 	page_index = 0;
1151 	offset = *buf_offset;
1152 	while (offset >= PAGE_SIZE) {
1153 		page_index++;
1154 		offset -= PAGE_SIZE;
1155 	}
1156 
1157 	/*
1158 	 * Limit the IO size to the length of the current vector, and update the
1159 	 * remaining IO count for the next time around.
1160 	 */
1161 	size = min_t(int, BBTOB(bp->b_maps[map].bm_len), *count);
1162 	*count -= size;
1163 	*buf_offset += size;
1164 
1165 next_chunk:
1166 	atomic_inc(&bp->b_io_remaining);
1167 	nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT);
1168 	if (nr_pages > total_nr_pages)
1169 		nr_pages = total_nr_pages;
1170 
1171 	bio = bio_alloc(GFP_NOIO, nr_pages);
1172 	bio->bi_bdev = bp->b_target->bt_bdev;
1173 	bio->bi_iter.bi_sector = sector;
1174 	bio->bi_end_io = xfs_buf_bio_end_io;
1175 	bio->bi_private = bp;
1176 
1177 
1178 	for (; size && nr_pages; nr_pages--, page_index++) {
1179 		int	rbytes, nbytes = PAGE_SIZE - offset;
1180 
1181 		if (nbytes > size)
1182 			nbytes = size;
1183 
1184 		rbytes = bio_add_page(bio, bp->b_pages[page_index], nbytes,
1185 				      offset);
1186 		if (rbytes < nbytes)
1187 			break;
1188 
1189 		offset = 0;
1190 		sector += BTOBB(nbytes);
1191 		size -= nbytes;
1192 		total_nr_pages--;
1193 	}
1194 
1195 	if (likely(bio->bi_iter.bi_size)) {
1196 		if (xfs_buf_is_vmapped(bp)) {
1197 			flush_kernel_vmap_range(bp->b_addr,
1198 						xfs_buf_vmap_len(bp));
1199 		}
1200 		submit_bio(rw, bio);
1201 		if (size)
1202 			goto next_chunk;
1203 	} else {
1204 		/*
1205 		 * This is guaranteed not to be the last io reference count
1206 		 * because the caller (xfs_buf_submit) holds a count itself.
1207 		 */
1208 		atomic_dec(&bp->b_io_remaining);
1209 		xfs_buf_ioerror(bp, -EIO);
1210 		bio_put(bio);
1211 	}
1212 
1213 }
1214 
1215 STATIC void
_xfs_buf_ioapply(struct xfs_buf * bp)1216 _xfs_buf_ioapply(
1217 	struct xfs_buf	*bp)
1218 {
1219 	struct blk_plug	plug;
1220 	int		rw;
1221 	int		offset;
1222 	int		size;
1223 	int		i;
1224 
1225 	/*
1226 	 * Make sure we capture only current IO errors rather than stale errors
1227 	 * left over from previous use of the buffer (e.g. failed readahead).
1228 	 */
1229 	bp->b_error = 0;
1230 
1231 	/*
1232 	 * Initialize the I/O completion workqueue if we haven't yet or the
1233 	 * submitter has not opted to specify a custom one.
1234 	 */
1235 	if (!bp->b_ioend_wq)
1236 		bp->b_ioend_wq = bp->b_target->bt_mount->m_buf_workqueue;
1237 
1238 	if (bp->b_flags & XBF_WRITE) {
1239 		if (bp->b_flags & XBF_SYNCIO)
1240 			rw = WRITE_SYNC;
1241 		else
1242 			rw = WRITE;
1243 		if (bp->b_flags & XBF_FUA)
1244 			rw |= REQ_FUA;
1245 		if (bp->b_flags & XBF_FLUSH)
1246 			rw |= REQ_FLUSH;
1247 
1248 		/*
1249 		 * Run the write verifier callback function if it exists. If
1250 		 * this function fails it will mark the buffer with an error and
1251 		 * the IO should not be dispatched.
1252 		 */
1253 		if (bp->b_ops) {
1254 			bp->b_ops->verify_write(bp);
1255 			if (bp->b_error) {
1256 				xfs_force_shutdown(bp->b_target->bt_mount,
1257 						   SHUTDOWN_CORRUPT_INCORE);
1258 				return;
1259 			}
1260 		} else if (bp->b_bn != XFS_BUF_DADDR_NULL) {
1261 			struct xfs_mount *mp = bp->b_target->bt_mount;
1262 
1263 			/*
1264 			 * non-crc filesystems don't attach verifiers during
1265 			 * log recovery, so don't warn for such filesystems.
1266 			 */
1267 			if (xfs_sb_version_hascrc(&mp->m_sb)) {
1268 				xfs_warn(mp,
1269 					"%s: no ops on block 0x%llx/0x%x",
1270 					__func__, bp->b_bn, bp->b_length);
1271 				xfs_hex_dump(bp->b_addr, 64);
1272 				dump_stack();
1273 			}
1274 		}
1275 	} else if (bp->b_flags & XBF_READ_AHEAD) {
1276 		rw = READA;
1277 	} else {
1278 		rw = READ;
1279 	}
1280 
1281 	/* we only use the buffer cache for meta-data */
1282 	rw |= REQ_META;
1283 
1284 	/*
1285 	 * Walk all the vectors issuing IO on them. Set up the initial offset
1286 	 * into the buffer and the desired IO size before we start -
1287 	 * _xfs_buf_ioapply_vec() will modify them appropriately for each
1288 	 * subsequent call.
1289 	 */
1290 	offset = bp->b_offset;
1291 	size = BBTOB(bp->b_io_length);
1292 	blk_start_plug(&plug);
1293 	for (i = 0; i < bp->b_map_count; i++) {
1294 		xfs_buf_ioapply_map(bp, i, &offset, &size, rw);
1295 		if (bp->b_error)
1296 			break;
1297 		if (size <= 0)
1298 			break;	/* all done */
1299 	}
1300 	blk_finish_plug(&plug);
1301 }
1302 
1303 /*
1304  * Asynchronous IO submission path. This transfers the buffer lock ownership and
1305  * the current reference to the IO. It is not safe to reference the buffer after
1306  * a call to this function unless the caller holds an additional reference
1307  * itself.
1308  */
1309 void
xfs_buf_submit(struct xfs_buf * bp)1310 xfs_buf_submit(
1311 	struct xfs_buf	*bp)
1312 {
1313 	trace_xfs_buf_submit(bp, _RET_IP_);
1314 
1315 	ASSERT(!(bp->b_flags & _XBF_DELWRI_Q));
1316 	ASSERT(bp->b_flags & XBF_ASYNC);
1317 
1318 	/* on shutdown we stale and complete the buffer immediately */
1319 	if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) {
1320 		xfs_buf_ioerror(bp, -EIO);
1321 		bp->b_flags &= ~XBF_DONE;
1322 		xfs_buf_stale(bp);
1323 		xfs_buf_ioend(bp);
1324 		return;
1325 	}
1326 
1327 	if (bp->b_flags & XBF_WRITE)
1328 		xfs_buf_wait_unpin(bp);
1329 
1330 	/* clear the internal error state to avoid spurious errors */
1331 	bp->b_io_error = 0;
1332 
1333 	/*
1334 	 * The caller's reference is released during I/O completion.
1335 	 * This occurs some time after the last b_io_remaining reference is
1336 	 * released, so after we drop our Io reference we have to have some
1337 	 * other reference to ensure the buffer doesn't go away from underneath
1338 	 * us. Take a direct reference to ensure we have safe access to the
1339 	 * buffer until we are finished with it.
1340 	 */
1341 	xfs_buf_hold(bp);
1342 
1343 	/*
1344 	 * Set the count to 1 initially, this will stop an I/O completion
1345 	 * callout which happens before we have started all the I/O from calling
1346 	 * xfs_buf_ioend too early.
1347 	 */
1348 	atomic_set(&bp->b_io_remaining, 1);
1349 	_xfs_buf_ioapply(bp);
1350 
1351 	/*
1352 	 * If _xfs_buf_ioapply failed, we can get back here with only the IO
1353 	 * reference we took above. If we drop it to zero, run completion so
1354 	 * that we don't return to the caller with completion still pending.
1355 	 */
1356 	if (atomic_dec_and_test(&bp->b_io_remaining) == 1) {
1357 		if (bp->b_error)
1358 			xfs_buf_ioend(bp);
1359 		else
1360 			xfs_buf_ioend_async(bp);
1361 	}
1362 
1363 	xfs_buf_rele(bp);
1364 	/* Note: it is not safe to reference bp now we've dropped our ref */
1365 }
1366 
1367 /*
1368  * Synchronous buffer IO submission path, read or write.
1369  */
1370 int
xfs_buf_submit_wait(struct xfs_buf * bp)1371 xfs_buf_submit_wait(
1372 	struct xfs_buf	*bp)
1373 {
1374 	int		error;
1375 
1376 	trace_xfs_buf_submit_wait(bp, _RET_IP_);
1377 
1378 	ASSERT(!(bp->b_flags & (_XBF_DELWRI_Q | XBF_ASYNC)));
1379 
1380 	if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) {
1381 		xfs_buf_ioerror(bp, -EIO);
1382 		xfs_buf_stale(bp);
1383 		bp->b_flags &= ~XBF_DONE;
1384 		return -EIO;
1385 	}
1386 
1387 	if (bp->b_flags & XBF_WRITE)
1388 		xfs_buf_wait_unpin(bp);
1389 
1390 	/* clear the internal error state to avoid spurious errors */
1391 	bp->b_io_error = 0;
1392 
1393 	/*
1394 	 * For synchronous IO, the IO does not inherit the submitters reference
1395 	 * count, nor the buffer lock. Hence we cannot release the reference we
1396 	 * are about to take until we've waited for all IO completion to occur,
1397 	 * including any xfs_buf_ioend_async() work that may be pending.
1398 	 */
1399 	xfs_buf_hold(bp);
1400 
1401 	/*
1402 	 * Set the count to 1 initially, this will stop an I/O completion
1403 	 * callout which happens before we have started all the I/O from calling
1404 	 * xfs_buf_ioend too early.
1405 	 */
1406 	atomic_set(&bp->b_io_remaining, 1);
1407 	_xfs_buf_ioapply(bp);
1408 
1409 	/*
1410 	 * make sure we run completion synchronously if it raced with us and is
1411 	 * already complete.
1412 	 */
1413 	if (atomic_dec_and_test(&bp->b_io_remaining) == 1)
1414 		xfs_buf_ioend(bp);
1415 
1416 	/* wait for completion before gathering the error from the buffer */
1417 	trace_xfs_buf_iowait(bp, _RET_IP_);
1418 	wait_for_completion(&bp->b_iowait);
1419 	trace_xfs_buf_iowait_done(bp, _RET_IP_);
1420 	error = bp->b_error;
1421 
1422 	/*
1423 	 * all done now, we can release the hold that keeps the buffer
1424 	 * referenced for the entire IO.
1425 	 */
1426 	xfs_buf_rele(bp);
1427 	return error;
1428 }
1429 
1430 void *
xfs_buf_offset(struct xfs_buf * bp,size_t offset)1431 xfs_buf_offset(
1432 	struct xfs_buf		*bp,
1433 	size_t			offset)
1434 {
1435 	struct page		*page;
1436 
1437 	if (bp->b_addr)
1438 		return bp->b_addr + offset;
1439 
1440 	offset += bp->b_offset;
1441 	page = bp->b_pages[offset >> PAGE_SHIFT];
1442 	return page_address(page) + (offset & (PAGE_SIZE-1));
1443 }
1444 
1445 /*
1446  *	Move data into or out of a buffer.
1447  */
1448 void
xfs_buf_iomove(xfs_buf_t * bp,size_t boff,size_t bsize,void * data,xfs_buf_rw_t mode)1449 xfs_buf_iomove(
1450 	xfs_buf_t		*bp,	/* buffer to process		*/
1451 	size_t			boff,	/* starting buffer offset	*/
1452 	size_t			bsize,	/* length to copy		*/
1453 	void			*data,	/* data address			*/
1454 	xfs_buf_rw_t		mode)	/* read/write/zero flag		*/
1455 {
1456 	size_t			bend;
1457 
1458 	bend = boff + bsize;
1459 	while (boff < bend) {
1460 		struct page	*page;
1461 		int		page_index, page_offset, csize;
1462 
1463 		page_index = (boff + bp->b_offset) >> PAGE_SHIFT;
1464 		page_offset = (boff + bp->b_offset) & ~PAGE_MASK;
1465 		page = bp->b_pages[page_index];
1466 		csize = min_t(size_t, PAGE_SIZE - page_offset,
1467 				      BBTOB(bp->b_io_length) - boff);
1468 
1469 		ASSERT((csize + page_offset) <= PAGE_SIZE);
1470 
1471 		switch (mode) {
1472 		case XBRW_ZERO:
1473 			memset(page_address(page) + page_offset, 0, csize);
1474 			break;
1475 		case XBRW_READ:
1476 			memcpy(data, page_address(page) + page_offset, csize);
1477 			break;
1478 		case XBRW_WRITE:
1479 			memcpy(page_address(page) + page_offset, data, csize);
1480 		}
1481 
1482 		boff += csize;
1483 		data += csize;
1484 	}
1485 }
1486 
1487 /*
1488  *	Handling of buffer targets (buftargs).
1489  */
1490 
1491 /*
1492  * Wait for any bufs with callbacks that have been submitted but have not yet
1493  * returned. These buffers will have an elevated hold count, so wait on those
1494  * while freeing all the buffers only held by the LRU.
1495  */
1496 static enum lru_status
xfs_buftarg_wait_rele(struct list_head * item,struct list_lru_one * lru,spinlock_t * lru_lock,void * arg)1497 xfs_buftarg_wait_rele(
1498 	struct list_head	*item,
1499 	struct list_lru_one	*lru,
1500 	spinlock_t		*lru_lock,
1501 	void			*arg)
1502 
1503 {
1504 	struct xfs_buf		*bp = container_of(item, struct xfs_buf, b_lru);
1505 	struct list_head	*dispose = arg;
1506 
1507 	if (atomic_read(&bp->b_hold) > 1) {
1508 		/* need to wait, so skip it this pass */
1509 		trace_xfs_buf_wait_buftarg(bp, _RET_IP_);
1510 		return LRU_SKIP;
1511 	}
1512 	if (!spin_trylock(&bp->b_lock))
1513 		return LRU_SKIP;
1514 
1515 	/*
1516 	 * clear the LRU reference count so the buffer doesn't get
1517 	 * ignored in xfs_buf_rele().
1518 	 */
1519 	atomic_set(&bp->b_lru_ref, 0);
1520 	bp->b_state |= XFS_BSTATE_DISPOSE;
1521 	list_lru_isolate_move(lru, item, dispose);
1522 	spin_unlock(&bp->b_lock);
1523 	return LRU_REMOVED;
1524 }
1525 
1526 void
xfs_wait_buftarg(struct xfs_buftarg * btp)1527 xfs_wait_buftarg(
1528 	struct xfs_buftarg	*btp)
1529 {
1530 	LIST_HEAD(dispose);
1531 	int loop = 0;
1532 
1533 	/*
1534 	 * We need to flush the buffer workqueue to ensure that all IO
1535 	 * completion processing is 100% done. Just waiting on buffer locks is
1536 	 * not sufficient for async IO as the reference count held over IO is
1537 	 * not released until after the buffer lock is dropped. Hence we need to
1538 	 * ensure here that all reference counts have been dropped before we
1539 	 * start walking the LRU list.
1540 	 */
1541 	flush_workqueue(btp->bt_mount->m_buf_workqueue);
1542 
1543 	/* loop until there is nothing left on the lru list. */
1544 	while (list_lru_count(&btp->bt_lru)) {
1545 		list_lru_walk(&btp->bt_lru, xfs_buftarg_wait_rele,
1546 			      &dispose, LONG_MAX);
1547 
1548 		while (!list_empty(&dispose)) {
1549 			struct xfs_buf *bp;
1550 			bp = list_first_entry(&dispose, struct xfs_buf, b_lru);
1551 			list_del_init(&bp->b_lru);
1552 			if (bp->b_flags & XBF_WRITE_FAIL) {
1553 				xfs_alert(btp->bt_mount,
1554 "Corruption Alert: Buffer at block 0x%llx had permanent write failures!",
1555 					(long long)bp->b_bn);
1556 				xfs_alert(btp->bt_mount,
1557 "Please run xfs_repair to determine the extent of the problem.");
1558 			}
1559 			xfs_buf_rele(bp);
1560 		}
1561 		if (loop++ != 0)
1562 			delay(100);
1563 	}
1564 }
1565 
1566 static enum lru_status
xfs_buftarg_isolate(struct list_head * item,struct list_lru_one * lru,spinlock_t * lru_lock,void * arg)1567 xfs_buftarg_isolate(
1568 	struct list_head	*item,
1569 	struct list_lru_one	*lru,
1570 	spinlock_t		*lru_lock,
1571 	void			*arg)
1572 {
1573 	struct xfs_buf		*bp = container_of(item, struct xfs_buf, b_lru);
1574 	struct list_head	*dispose = arg;
1575 
1576 	/*
1577 	 * we are inverting the lru lock/bp->b_lock here, so use a trylock.
1578 	 * If we fail to get the lock, just skip it.
1579 	 */
1580 	if (!spin_trylock(&bp->b_lock))
1581 		return LRU_SKIP;
1582 	/*
1583 	 * Decrement the b_lru_ref count unless the value is already
1584 	 * zero. If the value is already zero, we need to reclaim the
1585 	 * buffer, otherwise it gets another trip through the LRU.
1586 	 */
1587 	if (atomic_add_unless(&bp->b_lru_ref, -1, 0)) {
1588 		spin_unlock(&bp->b_lock);
1589 		return LRU_ROTATE;
1590 	}
1591 
1592 	bp->b_state |= XFS_BSTATE_DISPOSE;
1593 	list_lru_isolate_move(lru, item, dispose);
1594 	spin_unlock(&bp->b_lock);
1595 	return LRU_REMOVED;
1596 }
1597 
1598 static unsigned long
xfs_buftarg_shrink_scan(struct shrinker * shrink,struct shrink_control * sc)1599 xfs_buftarg_shrink_scan(
1600 	struct shrinker		*shrink,
1601 	struct shrink_control	*sc)
1602 {
1603 	struct xfs_buftarg	*btp = container_of(shrink,
1604 					struct xfs_buftarg, bt_shrinker);
1605 	LIST_HEAD(dispose);
1606 	unsigned long		freed;
1607 
1608 	freed = list_lru_shrink_walk(&btp->bt_lru, sc,
1609 				     xfs_buftarg_isolate, &dispose);
1610 
1611 	while (!list_empty(&dispose)) {
1612 		struct xfs_buf *bp;
1613 		bp = list_first_entry(&dispose, struct xfs_buf, b_lru);
1614 		list_del_init(&bp->b_lru);
1615 		xfs_buf_rele(bp);
1616 	}
1617 
1618 	return freed;
1619 }
1620 
1621 static unsigned long
xfs_buftarg_shrink_count(struct shrinker * shrink,struct shrink_control * sc)1622 xfs_buftarg_shrink_count(
1623 	struct shrinker		*shrink,
1624 	struct shrink_control	*sc)
1625 {
1626 	struct xfs_buftarg	*btp = container_of(shrink,
1627 					struct xfs_buftarg, bt_shrinker);
1628 	return list_lru_shrink_count(&btp->bt_lru, sc);
1629 }
1630 
1631 void
xfs_free_buftarg(struct xfs_mount * mp,struct xfs_buftarg * btp)1632 xfs_free_buftarg(
1633 	struct xfs_mount	*mp,
1634 	struct xfs_buftarg	*btp)
1635 {
1636 	unregister_shrinker(&btp->bt_shrinker);
1637 	list_lru_destroy(&btp->bt_lru);
1638 
1639 	if (mp->m_flags & XFS_MOUNT_BARRIER)
1640 		xfs_blkdev_issue_flush(btp);
1641 
1642 	kmem_free(btp);
1643 }
1644 
1645 int
xfs_setsize_buftarg(xfs_buftarg_t * btp,unsigned int sectorsize)1646 xfs_setsize_buftarg(
1647 	xfs_buftarg_t		*btp,
1648 	unsigned int		sectorsize)
1649 {
1650 	/* Set up metadata sector size info */
1651 	btp->bt_meta_sectorsize = sectorsize;
1652 	btp->bt_meta_sectormask = sectorsize - 1;
1653 
1654 	if (set_blocksize(btp->bt_bdev, sectorsize)) {
1655 		char name[BDEVNAME_SIZE];
1656 
1657 		bdevname(btp->bt_bdev, name);
1658 
1659 		xfs_warn(btp->bt_mount,
1660 			"Cannot set_blocksize to %u on device %s",
1661 			sectorsize, name);
1662 		return -EINVAL;
1663 	}
1664 
1665 	/* Set up device logical sector size mask */
1666 	btp->bt_logical_sectorsize = bdev_logical_block_size(btp->bt_bdev);
1667 	btp->bt_logical_sectormask = bdev_logical_block_size(btp->bt_bdev) - 1;
1668 
1669 	return 0;
1670 }
1671 
1672 /*
1673  * When allocating the initial buffer target we have not yet
1674  * read in the superblock, so don't know what sized sectors
1675  * are being used at this early stage.  Play safe.
1676  */
1677 STATIC int
xfs_setsize_buftarg_early(xfs_buftarg_t * btp,struct block_device * bdev)1678 xfs_setsize_buftarg_early(
1679 	xfs_buftarg_t		*btp,
1680 	struct block_device	*bdev)
1681 {
1682 	return xfs_setsize_buftarg(btp, bdev_logical_block_size(bdev));
1683 }
1684 
1685 xfs_buftarg_t *
xfs_alloc_buftarg(struct xfs_mount * mp,struct block_device * bdev)1686 xfs_alloc_buftarg(
1687 	struct xfs_mount	*mp,
1688 	struct block_device	*bdev)
1689 {
1690 	xfs_buftarg_t		*btp;
1691 
1692 	btp = kmem_zalloc(sizeof(*btp), KM_SLEEP | KM_NOFS);
1693 
1694 	btp->bt_mount = mp;
1695 	btp->bt_dev =  bdev->bd_dev;
1696 	btp->bt_bdev = bdev;
1697 	btp->bt_bdi = blk_get_backing_dev_info(bdev);
1698 
1699 	if (xfs_setsize_buftarg_early(btp, bdev))
1700 		goto error;
1701 
1702 	if (list_lru_init(&btp->bt_lru))
1703 		goto error;
1704 
1705 	btp->bt_shrinker.count_objects = xfs_buftarg_shrink_count;
1706 	btp->bt_shrinker.scan_objects = xfs_buftarg_shrink_scan;
1707 	btp->bt_shrinker.seeks = DEFAULT_SEEKS;
1708 	btp->bt_shrinker.flags = SHRINKER_NUMA_AWARE;
1709 	register_shrinker(&btp->bt_shrinker);
1710 	return btp;
1711 
1712 error:
1713 	kmem_free(btp);
1714 	return NULL;
1715 }
1716 
1717 /*
1718  * Cancel a delayed write list.
1719  *
1720  * Remove each buffer from the list, clear the delwri queue flag and drop the
1721  * associated buffer reference.
1722  */
1723 void
xfs_buf_delwri_cancel(struct list_head * list)1724 xfs_buf_delwri_cancel(
1725 	struct list_head	*list)
1726 {
1727 	struct xfs_buf		*bp;
1728 
1729 	while (!list_empty(list)) {
1730 		bp = list_first_entry(list, struct xfs_buf, b_list);
1731 
1732 		xfs_buf_lock(bp);
1733 		bp->b_flags &= ~_XBF_DELWRI_Q;
1734 		list_del_init(&bp->b_list);
1735 		xfs_buf_relse(bp);
1736 	}
1737 }
1738 
1739 /*
1740  * Add a buffer to the delayed write list.
1741  *
1742  * This queues a buffer for writeout if it hasn't already been.  Note that
1743  * neither this routine nor the buffer list submission functions perform
1744  * any internal synchronization.  It is expected that the lists are thread-local
1745  * to the callers.
1746  *
1747  * Returns true if we queued up the buffer, or false if it already had
1748  * been on the buffer list.
1749  */
1750 bool
xfs_buf_delwri_queue(struct xfs_buf * bp,struct list_head * list)1751 xfs_buf_delwri_queue(
1752 	struct xfs_buf		*bp,
1753 	struct list_head	*list)
1754 {
1755 	ASSERT(xfs_buf_islocked(bp));
1756 	ASSERT(!(bp->b_flags & XBF_READ));
1757 
1758 	/*
1759 	 * If the buffer is already marked delwri it already is queued up
1760 	 * by someone else for imediate writeout.  Just ignore it in that
1761 	 * case.
1762 	 */
1763 	if (bp->b_flags & _XBF_DELWRI_Q) {
1764 		trace_xfs_buf_delwri_queued(bp, _RET_IP_);
1765 		return false;
1766 	}
1767 
1768 	trace_xfs_buf_delwri_queue(bp, _RET_IP_);
1769 
1770 	/*
1771 	 * If a buffer gets written out synchronously or marked stale while it
1772 	 * is on a delwri list we lazily remove it. To do this, the other party
1773 	 * clears the  _XBF_DELWRI_Q flag but otherwise leaves the buffer alone.
1774 	 * It remains referenced and on the list.  In a rare corner case it
1775 	 * might get readded to a delwri list after the synchronous writeout, in
1776 	 * which case we need just need to re-add the flag here.
1777 	 */
1778 	bp->b_flags |= _XBF_DELWRI_Q;
1779 	if (list_empty(&bp->b_list)) {
1780 		atomic_inc(&bp->b_hold);
1781 		list_add_tail(&bp->b_list, list);
1782 	}
1783 
1784 	return true;
1785 }
1786 
1787 /*
1788  * Compare function is more complex than it needs to be because
1789  * the return value is only 32 bits and we are doing comparisons
1790  * on 64 bit values
1791  */
1792 static int
xfs_buf_cmp(void * priv,struct list_head * a,struct list_head * b)1793 xfs_buf_cmp(
1794 	void		*priv,
1795 	struct list_head *a,
1796 	struct list_head *b)
1797 {
1798 	struct xfs_buf	*ap = container_of(a, struct xfs_buf, b_list);
1799 	struct xfs_buf	*bp = container_of(b, struct xfs_buf, b_list);
1800 	xfs_daddr_t		diff;
1801 
1802 	diff = ap->b_maps[0].bm_bn - bp->b_maps[0].bm_bn;
1803 	if (diff < 0)
1804 		return -1;
1805 	if (diff > 0)
1806 		return 1;
1807 	return 0;
1808 }
1809 
1810 static int
__xfs_buf_delwri_submit(struct list_head * buffer_list,struct list_head * io_list,bool wait)1811 __xfs_buf_delwri_submit(
1812 	struct list_head	*buffer_list,
1813 	struct list_head	*io_list,
1814 	bool			wait)
1815 {
1816 	struct blk_plug		plug;
1817 	struct xfs_buf		*bp, *n;
1818 	int			pinned = 0;
1819 
1820 	list_for_each_entry_safe(bp, n, buffer_list, b_list) {
1821 		if (!wait) {
1822 			if (xfs_buf_ispinned(bp)) {
1823 				pinned++;
1824 				continue;
1825 			}
1826 			if (!xfs_buf_trylock(bp))
1827 				continue;
1828 		} else {
1829 			xfs_buf_lock(bp);
1830 		}
1831 
1832 		/*
1833 		 * Someone else might have written the buffer synchronously or
1834 		 * marked it stale in the meantime.  In that case only the
1835 		 * _XBF_DELWRI_Q flag got cleared, and we have to drop the
1836 		 * reference and remove it from the list here.
1837 		 */
1838 		if (!(bp->b_flags & _XBF_DELWRI_Q)) {
1839 			list_del_init(&bp->b_list);
1840 			xfs_buf_relse(bp);
1841 			continue;
1842 		}
1843 
1844 		list_move_tail(&bp->b_list, io_list);
1845 		trace_xfs_buf_delwri_split(bp, _RET_IP_);
1846 	}
1847 
1848 	list_sort(NULL, io_list, xfs_buf_cmp);
1849 
1850 	blk_start_plug(&plug);
1851 	list_for_each_entry_safe(bp, n, io_list, b_list) {
1852 		bp->b_flags &= ~(_XBF_DELWRI_Q | XBF_ASYNC | XBF_WRITE_FAIL);
1853 		bp->b_flags |= XBF_WRITE | XBF_ASYNC;
1854 
1855 		/*
1856 		 * we do all Io submission async. This means if we need to wait
1857 		 * for IO completion we need to take an extra reference so the
1858 		 * buffer is still valid on the other side.
1859 		 */
1860 		if (wait)
1861 			xfs_buf_hold(bp);
1862 		else
1863 			list_del_init(&bp->b_list);
1864 
1865 		xfs_buf_submit(bp);
1866 	}
1867 	blk_finish_plug(&plug);
1868 
1869 	return pinned;
1870 }
1871 
1872 /*
1873  * Write out a buffer list asynchronously.
1874  *
1875  * This will take the @buffer_list, write all non-locked and non-pinned buffers
1876  * out and not wait for I/O completion on any of the buffers.  This interface
1877  * is only safely useable for callers that can track I/O completion by higher
1878  * level means, e.g. AIL pushing as the @buffer_list is consumed in this
1879  * function.
1880  */
1881 int
xfs_buf_delwri_submit_nowait(struct list_head * buffer_list)1882 xfs_buf_delwri_submit_nowait(
1883 	struct list_head	*buffer_list)
1884 {
1885 	LIST_HEAD		(io_list);
1886 	return __xfs_buf_delwri_submit(buffer_list, &io_list, false);
1887 }
1888 
1889 /*
1890  * Write out a buffer list synchronously.
1891  *
1892  * This will take the @buffer_list, write all buffers out and wait for I/O
1893  * completion on all of the buffers. @buffer_list is consumed by the function,
1894  * so callers must have some other way of tracking buffers if they require such
1895  * functionality.
1896  */
1897 int
xfs_buf_delwri_submit(struct list_head * buffer_list)1898 xfs_buf_delwri_submit(
1899 	struct list_head	*buffer_list)
1900 {
1901 	LIST_HEAD		(io_list);
1902 	int			error = 0, error2;
1903 	struct xfs_buf		*bp;
1904 
1905 	__xfs_buf_delwri_submit(buffer_list, &io_list, true);
1906 
1907 	/* Wait for IO to complete. */
1908 	while (!list_empty(&io_list)) {
1909 		bp = list_first_entry(&io_list, struct xfs_buf, b_list);
1910 
1911 		list_del_init(&bp->b_list);
1912 
1913 		/* locking the buffer will wait for async IO completion. */
1914 		xfs_buf_lock(bp);
1915 		error2 = bp->b_error;
1916 		xfs_buf_relse(bp);
1917 		if (!error)
1918 			error = error2;
1919 	}
1920 
1921 	return error;
1922 }
1923 
1924 int __init
xfs_buf_init(void)1925 xfs_buf_init(void)
1926 {
1927 	xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf",
1928 						KM_ZONE_HWALIGN, NULL);
1929 	if (!xfs_buf_zone)
1930 		goto out;
1931 
1932 	return 0;
1933 
1934  out:
1935 	return -ENOMEM;
1936 }
1937 
1938 void
xfs_buf_terminate(void)1939 xfs_buf_terminate(void)
1940 {
1941 	kmem_zone_destroy(xfs_buf_zone);
1942 }
1943