blob: 649ade8ef598693e2e57ac66475c1993b57903e3 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scottf07c2252006-09-28 10:52:15 +10002 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
Nathan Scott7b718762005-11-02 14:58:39 +11003 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Vlad Apostolov93c189c2006-11-11 18:03:49 +110018#include "xfs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/stddef.h>
20#include <linux/errno.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090021#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#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>
Christoph Hellwig4df08c52005-09-05 08:34:18 +100032#include <linux/kthread.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080033#include <linux/migrate.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070034#include <linux/backing-dev.h>
Nigel Cunningham7dfb7102006-12-06 20:34:23 -080035#include <linux/freezer.h>
Dave Chinner089716a2010-01-26 15:13:25 +110036#include <linux/list_sort.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Christoph Hellwigb7963132009-03-03 14:48:37 -050038#include "xfs_sb.h"
39#include "xfs_inum.h"
Dave Chinnered3b4d62010-05-21 12:07:08 +100040#include "xfs_log.h"
Christoph Hellwigb7963132009-03-03 14:48:37 -050041#include "xfs_ag.h"
42#include "xfs_dmapi.h"
43#include "xfs_mount.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000044#include "xfs_trace.h"
Christoph Hellwigb7963132009-03-03 14:48:37 -050045
David Chinner7989cb82007-02-10 18:34:56 +110046static kmem_zone_t *xfs_buf_zone;
David Chinnera6867a62006-01-11 15:37:58 +110047STATIC int xfsbufd(void *);
Al Viro27496a82005-10-21 03:20:48 -040048STATIC int xfsbufd_wakeup(int, gfp_t);
Nathan Scottce8e9222006-01-11 15:39:08 +110049STATIC void xfs_buf_delwri_queue(xfs_buf_t *, int);
Rusty Russell8e1f9362007-07-17 04:03:17 -070050static struct shrinker xfs_buf_shake = {
51 .shrink = xfsbufd_wakeup,
52 .seeks = DEFAULT_SEEKS,
53};
Christoph Hellwig23ea4032005-06-21 15:14:01 +100054
David Chinner7989cb82007-02-10 18:34:56 +110055static struct workqueue_struct *xfslogd_workqueue;
Christoph Hellwig0829c362005-09-02 16:58:49 +100056struct workqueue_struct *xfsdatad_workqueue;
Dave Chinnerc626d172009-04-06 18:42:11 +020057struct workqueue_struct *xfsconvertd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Nathan Scottce8e9222006-01-11 15:39:08 +110059#ifdef XFS_BUF_LOCK_TRACKING
60# define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid)
61# define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1)
62# define XB_GET_OWNER(bp) ((bp)->b_last_holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -070063#else
Nathan Scottce8e9222006-01-11 15:39:08 +110064# define XB_SET_OWNER(bp) do { } while (0)
65# define XB_CLEAR_OWNER(bp) do { } while (0)
66# define XB_GET_OWNER(bp) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -070067#endif
68
Nathan Scottce8e9222006-01-11 15:39:08 +110069#define xb_to_gfp(flags) \
70 ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : \
71 ((flags) & XBF_DONT_BLOCK) ? GFP_NOFS : GFP_KERNEL) | __GFP_NOWARN)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
Nathan Scottce8e9222006-01-11 15:39:08 +110073#define xb_to_km(flags) \
74 (((flags) & XBF_DONT_BLOCK) ? KM_NOFS : KM_SLEEP)
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
Nathan Scottce8e9222006-01-11 15:39:08 +110076#define xfs_buf_allocate(flags) \
77 kmem_zone_alloc(xfs_buf_zone, xb_to_km(flags))
78#define xfs_buf_deallocate(bp) \
79 kmem_zone_free(xfs_buf_zone, (bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
James Bottomley73c77e22010-01-25 11:42:24 -060081static inline int
82xfs_buf_is_vmapped(
83 struct xfs_buf *bp)
84{
85 /*
86 * Return true if the buffer is vmapped.
87 *
88 * The XBF_MAPPED flag is set if the buffer should be mapped, but the
89 * code is clever enough to know it doesn't have to map a single page,
90 * so the check has to be both for XBF_MAPPED and bp->b_page_count > 1.
91 */
92 return (bp->b_flags & XBF_MAPPED) && bp->b_page_count > 1;
93}
94
95static inline int
96xfs_buf_vmap_len(
97 struct xfs_buf *bp)
98{
99 return (bp->b_page_count * PAGE_SIZE) - bp->b_offset;
100}
101
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102/*
Nathan Scottce8e9222006-01-11 15:39:08 +1100103 * Page Region interfaces.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 *
Nathan Scottce8e9222006-01-11 15:39:08 +1100105 * For pages in filesystems where the blocksize is smaller than the
106 * pagesize, we use the page->private field (long) to hold a bitmap
107 * of uptodate regions within the page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 *
Nathan Scottce8e9222006-01-11 15:39:08 +1100109 * Each such region is "bytes per page / bits per long" bytes long.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 *
Nathan Scottce8e9222006-01-11 15:39:08 +1100111 * NBPPR == number-of-bytes-per-page-region
112 * BTOPR == bytes-to-page-region (rounded up)
113 * BTOPRT == bytes-to-page-region-truncated (rounded down)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 */
115#if (BITS_PER_LONG == 32)
116#define PRSHIFT (PAGE_CACHE_SHIFT - 5) /* (32 == 1<<5) */
117#elif (BITS_PER_LONG == 64)
118#define PRSHIFT (PAGE_CACHE_SHIFT - 6) /* (64 == 1<<6) */
119#else
120#error BITS_PER_LONG must be 32 or 64
121#endif
122#define NBPPR (PAGE_CACHE_SIZE/BITS_PER_LONG)
123#define BTOPR(b) (((unsigned int)(b) + (NBPPR - 1)) >> PRSHIFT)
124#define BTOPRT(b) (((unsigned int)(b) >> PRSHIFT))
125
126STATIC unsigned long
127page_region_mask(
128 size_t offset,
129 size_t length)
130{
131 unsigned long mask;
132 int first, final;
133
134 first = BTOPR(offset);
135 final = BTOPRT(offset + length - 1);
136 first = min(first, final);
137
138 mask = ~0UL;
139 mask <<= BITS_PER_LONG - (final - first);
140 mask >>= BITS_PER_LONG - (final);
141
142 ASSERT(offset + length <= PAGE_CACHE_SIZE);
143 ASSERT((final - first) < BITS_PER_LONG && (final - first) >= 0);
144
145 return mask;
146}
147
Christoph Hellwigb8f82a42009-11-14 16:17:22 +0000148STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149set_page_region(
150 struct page *page,
151 size_t offset,
152 size_t length)
153{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700154 set_page_private(page,
155 page_private(page) | page_region_mask(offset, length));
156 if (page_private(page) == ~0UL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 SetPageUptodate(page);
158}
159
Christoph Hellwigb8f82a42009-11-14 16:17:22 +0000160STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161test_page_region(
162 struct page *page,
163 size_t offset,
164 size_t length)
165{
166 unsigned long mask = page_region_mask(offset, length);
167
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700168 return (mask && (page_private(page) & mask) == mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169}
170
171/*
Nathan Scottce8e9222006-01-11 15:39:08 +1100172 * Internal xfs_buf_t object manipulation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 */
174
175STATIC void
Nathan Scottce8e9222006-01-11 15:39:08 +1100176_xfs_buf_initialize(
177 xfs_buf_t *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 xfs_buftarg_t *target,
Nathan Scott204ab252006-01-11 20:50:22 +1100179 xfs_off_t range_base,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 size_t range_length,
Nathan Scottce8e9222006-01-11 15:39:08 +1100181 xfs_buf_flags_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182{
183 /*
Nathan Scottce8e9222006-01-11 15:39:08 +1100184 * We don't want certain flags to appear in b_flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 */
Nathan Scottce8e9222006-01-11 15:39:08 +1100186 flags &= ~(XBF_LOCK|XBF_MAPPED|XBF_DONT_BLOCK|XBF_READ_AHEAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187
Nathan Scottce8e9222006-01-11 15:39:08 +1100188 memset(bp, 0, sizeof(xfs_buf_t));
189 atomic_set(&bp->b_hold, 1);
David Chinnerb4dd3302008-08-13 16:36:11 +1000190 init_completion(&bp->b_iowait);
Nathan Scottce8e9222006-01-11 15:39:08 +1100191 INIT_LIST_HEAD(&bp->b_list);
192 INIT_LIST_HEAD(&bp->b_hash_list);
193 init_MUTEX_LOCKED(&bp->b_sema); /* held, no waiters */
194 XB_SET_OWNER(bp);
195 bp->b_target = target;
196 bp->b_file_offset = range_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 /*
198 * Set buffer_length and count_desired to the same value initially.
199 * I/O routines should use count_desired, which will be the same in
200 * most cases but may be reset (e.g. XFS recovery).
201 */
Nathan Scottce8e9222006-01-11 15:39:08 +1100202 bp->b_buffer_length = bp->b_count_desired = range_length;
203 bp->b_flags = flags;
204 bp->b_bn = XFS_BUF_DADDR_NULL;
205 atomic_set(&bp->b_pin_count, 0);
206 init_waitqueue_head(&bp->b_waiters);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207
Nathan Scottce8e9222006-01-11 15:39:08 +1100208 XFS_STATS_INC(xb_create);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000209
210 trace_xfs_buf_init(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211}
212
213/*
Nathan Scottce8e9222006-01-11 15:39:08 +1100214 * Allocate a page array capable of holding a specified number
215 * of pages, and point the page buf at it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 */
217STATIC int
Nathan Scottce8e9222006-01-11 15:39:08 +1100218_xfs_buf_get_pages(
219 xfs_buf_t *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 int page_count,
Nathan Scottce8e9222006-01-11 15:39:08 +1100221 xfs_buf_flags_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222{
223 /* Make sure that we have a page list */
Nathan Scottce8e9222006-01-11 15:39:08 +1100224 if (bp->b_pages == NULL) {
225 bp->b_offset = xfs_buf_poff(bp->b_file_offset);
226 bp->b_page_count = page_count;
227 if (page_count <= XB_PAGES) {
228 bp->b_pages = bp->b_page_array;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 } else {
Nathan Scottce8e9222006-01-11 15:39:08 +1100230 bp->b_pages = kmem_alloc(sizeof(struct page *) *
231 page_count, xb_to_km(flags));
232 if (bp->b_pages == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 return -ENOMEM;
234 }
Nathan Scottce8e9222006-01-11 15:39:08 +1100235 memset(bp->b_pages, 0, sizeof(struct page *) * page_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 }
237 return 0;
238}
239
240/*
Nathan Scottce8e9222006-01-11 15:39:08 +1100241 * Frees b_pages if it was allocated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 */
243STATIC void
Nathan Scottce8e9222006-01-11 15:39:08 +1100244_xfs_buf_free_pages(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 xfs_buf_t *bp)
246{
Nathan Scottce8e9222006-01-11 15:39:08 +1100247 if (bp->b_pages != bp->b_page_array) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +1000248 kmem_free(bp->b_pages);
Dave Chinner3fc98b12009-12-14 23:11:57 +0000249 bp->b_pages = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 }
251}
252
253/*
254 * Releases the specified buffer.
255 *
256 * The modification state of any associated pages is left unchanged.
Nathan Scottce8e9222006-01-11 15:39:08 +1100257 * The buffer most not be on any hash - use xfs_buf_rele instead for
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 * hashed and refcounted buffers
259 */
260void
Nathan Scottce8e9222006-01-11 15:39:08 +1100261xfs_buf_free(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 xfs_buf_t *bp)
263{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000264 trace_xfs_buf_free(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
Nathan Scottce8e9222006-01-11 15:39:08 +1100266 ASSERT(list_empty(&bp->b_hash_list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
Christoph Hellwig1fa40b02007-05-14 18:23:50 +1000268 if (bp->b_flags & (_XBF_PAGE_CACHE|_XBF_PAGES)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 uint i;
270
James Bottomley73c77e22010-01-25 11:42:24 -0600271 if (xfs_buf_is_vmapped(bp))
Alex Elder8a262e52010-03-16 18:55:56 +0000272 vm_unmap_ram(bp->b_addr - bp->b_offset,
273 bp->b_page_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274
Nathan Scott948ecdb2006-09-28 11:03:13 +1000275 for (i = 0; i < bp->b_page_count; i++) {
276 struct page *page = bp->b_pages[i];
277
Christoph Hellwig1fa40b02007-05-14 18:23:50 +1000278 if (bp->b_flags & _XBF_PAGE_CACHE)
279 ASSERT(!PagePrivate(page));
Nathan Scott948ecdb2006-09-28 11:03:13 +1000280 page_cache_release(page);
281 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 }
Dave Chinner3fc98b12009-12-14 23:11:57 +0000283 _xfs_buf_free_pages(bp);
Nathan Scottce8e9222006-01-11 15:39:08 +1100284 xfs_buf_deallocate(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285}
286
287/*
288 * Finds all pages for buffer in question and builds it's page list.
289 */
290STATIC int
Nathan Scottce8e9222006-01-11 15:39:08 +1100291_xfs_buf_lookup_pages(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 xfs_buf_t *bp,
293 uint flags)
294{
Nathan Scottce8e9222006-01-11 15:39:08 +1100295 struct address_space *mapping = bp->b_target->bt_mapping;
296 size_t blocksize = bp->b_target->bt_bsize;
297 size_t size = bp->b_count_desired;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 size_t nbytes, offset;
Nathan Scottce8e9222006-01-11 15:39:08 +1100299 gfp_t gfp_mask = xb_to_gfp(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 unsigned short page_count, i;
301 pgoff_t first;
Nathan Scott204ab252006-01-11 20:50:22 +1100302 xfs_off_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 int error;
304
Nathan Scottce8e9222006-01-11 15:39:08 +1100305 end = bp->b_file_offset + bp->b_buffer_length;
306 page_count = xfs_buf_btoc(end) - xfs_buf_btoct(bp->b_file_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307
Nathan Scottce8e9222006-01-11 15:39:08 +1100308 error = _xfs_buf_get_pages(bp, page_count, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 if (unlikely(error))
310 return error;
Nathan Scottce8e9222006-01-11 15:39:08 +1100311 bp->b_flags |= _XBF_PAGE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312
Nathan Scottce8e9222006-01-11 15:39:08 +1100313 offset = bp->b_offset;
314 first = bp->b_file_offset >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315
Nathan Scottce8e9222006-01-11 15:39:08 +1100316 for (i = 0; i < bp->b_page_count; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 struct page *page;
318 uint retries = 0;
319
320 retry:
321 page = find_or_create_page(mapping, first + i, gfp_mask);
322 if (unlikely(page == NULL)) {
Nathan Scottce8e9222006-01-11 15:39:08 +1100323 if (flags & XBF_READ_AHEAD) {
324 bp->b_page_count = i;
Christoph Hellwig6ab455e2008-05-19 16:34:42 +1000325 for (i = 0; i < bp->b_page_count; i++)
326 unlock_page(bp->b_pages[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 return -ENOMEM;
328 }
329
330 /*
331 * This could deadlock.
332 *
333 * But until all the XFS lowlevel code is revamped to
334 * handle buffer allocation failures we can't do much.
335 */
336 if (!(++retries % 100))
337 printk(KERN_ERR
338 "XFS: possible memory allocation "
339 "deadlock in %s (mode:0x%x)\n",
Harvey Harrison34a622b2008-04-10 12:19:21 +1000340 __func__, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341
Nathan Scottce8e9222006-01-11 15:39:08 +1100342 XFS_STATS_INC(xb_page_retries);
Christoph Hellwig23ea4032005-06-21 15:14:01 +1000343 xfsbufd_wakeup(0, gfp_mask);
Jens Axboe8aa7e842009-07-09 14:52:32 +0200344 congestion_wait(BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 goto retry;
346 }
347
Nathan Scottce8e9222006-01-11 15:39:08 +1100348 XFS_STATS_INC(xb_page_found);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349
350 nbytes = min_t(size_t, size, PAGE_CACHE_SIZE - offset);
351 size -= nbytes;
352
Nathan Scott948ecdb2006-09-28 11:03:13 +1000353 ASSERT(!PagePrivate(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 if (!PageUptodate(page)) {
355 page_count--;
Christoph Hellwig6ab455e2008-05-19 16:34:42 +1000356 if (blocksize >= PAGE_CACHE_SIZE) {
357 if (flags & XBF_READ)
358 bp->b_flags |= _XBF_PAGE_LOCKED;
359 } else if (!PagePrivate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 if (test_page_region(page, offset, nbytes))
361 page_count++;
362 }
363 }
364
Nathan Scottce8e9222006-01-11 15:39:08 +1100365 bp->b_pages[i] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 offset = 0;
367 }
368
Christoph Hellwig6ab455e2008-05-19 16:34:42 +1000369 if (!(bp->b_flags & _XBF_PAGE_LOCKED)) {
370 for (i = 0; i < bp->b_page_count; i++)
371 unlock_page(bp->b_pages[i]);
372 }
373
Nathan Scottce8e9222006-01-11 15:39:08 +1100374 if (page_count == bp->b_page_count)
375 bp->b_flags |= XBF_DONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 return error;
378}
379
380/*
381 * Map buffer into kernel address-space if nessecary.
382 */
383STATIC int
Nathan Scottce8e9222006-01-11 15:39:08 +1100384_xfs_buf_map_pages(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 xfs_buf_t *bp,
386 uint flags)
387{
388 /* A single page buffer is always mappable */
Nathan Scottce8e9222006-01-11 15:39:08 +1100389 if (bp->b_page_count == 1) {
390 bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset;
391 bp->b_flags |= XBF_MAPPED;
392 } else if (flags & XBF_MAPPED) {
Alex Elder8a262e52010-03-16 18:55:56 +0000393 bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count,
394 -1, PAGE_KERNEL);
Nathan Scottce8e9222006-01-11 15:39:08 +1100395 if (unlikely(bp->b_addr == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 return -ENOMEM;
Nathan Scottce8e9222006-01-11 15:39:08 +1100397 bp->b_addr += bp->b_offset;
398 bp->b_flags |= XBF_MAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 }
400
401 return 0;
402}
403
404/*
405 * Finding and Reading Buffers
406 */
407
408/*
Nathan Scottce8e9222006-01-11 15:39:08 +1100409 * Look up, and creates if absent, a lockable buffer for
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 * a given range of an inode. The buffer is returned
411 * locked. If other overlapping buffers exist, they are
412 * released before the new buffer is created and locked,
413 * which may imply that this call will block until those buffers
414 * are unlocked. No I/O is implied by this call.
415 */
416xfs_buf_t *
Nathan Scottce8e9222006-01-11 15:39:08 +1100417_xfs_buf_find(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 xfs_buftarg_t *btp, /* block device target */
Nathan Scott204ab252006-01-11 20:50:22 +1100419 xfs_off_t ioff, /* starting offset of range */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 size_t isize, /* length of range */
Nathan Scottce8e9222006-01-11 15:39:08 +1100421 xfs_buf_flags_t flags,
422 xfs_buf_t *new_bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423{
Nathan Scott204ab252006-01-11 20:50:22 +1100424 xfs_off_t range_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 size_t range_length;
426 xfs_bufhash_t *hash;
Nathan Scottce8e9222006-01-11 15:39:08 +1100427 xfs_buf_t *bp, *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428
429 range_base = (ioff << BBSHIFT);
430 range_length = (isize << BBSHIFT);
431
432 /* Check for IOs smaller than the sector size / not sector aligned */
Nathan Scottce8e9222006-01-11 15:39:08 +1100433 ASSERT(!(range_length < (1 << btp->bt_sshift)));
Nathan Scott204ab252006-01-11 20:50:22 +1100434 ASSERT(!(range_base & (xfs_off_t)btp->bt_smask));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
436 hash = &btp->bt_hash[hash_long((unsigned long)ioff, btp->bt_hashshift)];
437
438 spin_lock(&hash->bh_lock);
439
Nathan Scottce8e9222006-01-11 15:39:08 +1100440 list_for_each_entry_safe(bp, n, &hash->bh_list, b_hash_list) {
441 ASSERT(btp == bp->b_target);
442 if (bp->b_file_offset == range_base &&
443 bp->b_buffer_length == range_length) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 /*
Nathan Scottce8e9222006-01-11 15:39:08 +1100445 * If we look at something, bring it to the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 * front of the list for next time.
447 */
Nathan Scottce8e9222006-01-11 15:39:08 +1100448 atomic_inc(&bp->b_hold);
449 list_move(&bp->b_hash_list, &hash->bh_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 goto found;
451 }
452 }
453
454 /* No match found */
Nathan Scottce8e9222006-01-11 15:39:08 +1100455 if (new_bp) {
456 _xfs_buf_initialize(new_bp, btp, range_base,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 range_length, flags);
Nathan Scottce8e9222006-01-11 15:39:08 +1100458 new_bp->b_hash = hash;
459 list_add(&new_bp->b_hash_list, &hash->bh_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 } else {
Nathan Scottce8e9222006-01-11 15:39:08 +1100461 XFS_STATS_INC(xb_miss_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 }
463
464 spin_unlock(&hash->bh_lock);
Nathan Scottce8e9222006-01-11 15:39:08 +1100465 return new_bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
467found:
468 spin_unlock(&hash->bh_lock);
469
470 /* Attempt to get the semaphore without sleeping,
471 * if this does not work then we need to drop the
472 * spinlock and do a hard attempt on the semaphore.
473 */
Nathan Scottce8e9222006-01-11 15:39:08 +1100474 if (down_trylock(&bp->b_sema)) {
475 if (!(flags & XBF_TRYLOCK)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 /* wait for buffer ownership */
Nathan Scottce8e9222006-01-11 15:39:08 +1100477 xfs_buf_lock(bp);
478 XFS_STATS_INC(xb_get_locked_waited);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 } else {
480 /* We asked for a trylock and failed, no need
481 * to look at file offset and length here, we
Nathan Scottce8e9222006-01-11 15:39:08 +1100482 * know that this buffer at least overlaps our
483 * buffer and is locked, therefore our buffer
484 * either does not exist, or is this buffer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 */
Nathan Scottce8e9222006-01-11 15:39:08 +1100486 xfs_buf_rele(bp);
487 XFS_STATS_INC(xb_busy_locked);
488 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 }
490 } else {
491 /* trylock worked */
Nathan Scottce8e9222006-01-11 15:39:08 +1100492 XB_SET_OWNER(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 }
494
Nathan Scottce8e9222006-01-11 15:39:08 +1100495 if (bp->b_flags & XBF_STALE) {
496 ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0);
497 bp->b_flags &= XBF_MAPPED;
David Chinner2f926582005-09-05 08:33:35 +1000498 }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000499
500 trace_xfs_buf_find(bp, flags, _RET_IP_);
Nathan Scottce8e9222006-01-11 15:39:08 +1100501 XFS_STATS_INC(xb_get_locked);
502 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503}
504
505/*
Nathan Scottce8e9222006-01-11 15:39:08 +1100506 * Assembles a buffer covering the specified range.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 * Storage in memory for all portions of the buffer will be allocated,
508 * although backing storage may not be.
509 */
510xfs_buf_t *
Christoph Hellwig6ad112b2009-11-24 18:02:23 +0000511xfs_buf_get(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 xfs_buftarg_t *target,/* target for buffer */
Nathan Scott204ab252006-01-11 20:50:22 +1100513 xfs_off_t ioff, /* starting offset of range */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 size_t isize, /* length of range */
Nathan Scottce8e9222006-01-11 15:39:08 +1100515 xfs_buf_flags_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516{
Nathan Scottce8e9222006-01-11 15:39:08 +1100517 xfs_buf_t *bp, *new_bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 int error = 0, i;
519
Nathan Scottce8e9222006-01-11 15:39:08 +1100520 new_bp = xfs_buf_allocate(flags);
521 if (unlikely(!new_bp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 return NULL;
523
Nathan Scottce8e9222006-01-11 15:39:08 +1100524 bp = _xfs_buf_find(target, ioff, isize, flags, new_bp);
525 if (bp == new_bp) {
526 error = _xfs_buf_lookup_pages(bp, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 if (error)
528 goto no_buffer;
529 } else {
Nathan Scottce8e9222006-01-11 15:39:08 +1100530 xfs_buf_deallocate(new_bp);
531 if (unlikely(bp == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 return NULL;
533 }
534
Nathan Scottce8e9222006-01-11 15:39:08 +1100535 for (i = 0; i < bp->b_page_count; i++)
536 mark_page_accessed(bp->b_pages[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537
Nathan Scottce8e9222006-01-11 15:39:08 +1100538 if (!(bp->b_flags & XBF_MAPPED)) {
539 error = _xfs_buf_map_pages(bp, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 if (unlikely(error)) {
541 printk(KERN_WARNING "%s: failed to map pages\n",
Harvey Harrison34a622b2008-04-10 12:19:21 +1000542 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 goto no_buffer;
544 }
545 }
546
Nathan Scottce8e9222006-01-11 15:39:08 +1100547 XFS_STATS_INC(xb_get);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548
549 /*
550 * Always fill in the block number now, the mapped cases can do
551 * their own overlay of this later.
552 */
Nathan Scottce8e9222006-01-11 15:39:08 +1100553 bp->b_bn = ioff;
554 bp->b_count_desired = bp->b_buffer_length;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000556 trace_xfs_buf_get(bp, flags, _RET_IP_);
Nathan Scottce8e9222006-01-11 15:39:08 +1100557 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558
559 no_buffer:
Nathan Scottce8e9222006-01-11 15:39:08 +1100560 if (flags & (XBF_LOCK | XBF_TRYLOCK))
561 xfs_buf_unlock(bp);
562 xfs_buf_rele(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 return NULL;
564}
565
Christoph Hellwig5d765b92008-12-03 12:20:26 +0100566STATIC int
567_xfs_buf_read(
568 xfs_buf_t *bp,
569 xfs_buf_flags_t flags)
570{
571 int status;
572
Christoph Hellwig5d765b92008-12-03 12:20:26 +0100573 ASSERT(!(flags & (XBF_DELWRI|XBF_WRITE)));
574 ASSERT(bp->b_bn != XFS_BUF_DADDR_NULL);
575
576 bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_DELWRI | \
577 XBF_READ_AHEAD | _XBF_RUN_QUEUES);
578 bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | \
579 XBF_READ_AHEAD | _XBF_RUN_QUEUES);
580
581 status = xfs_buf_iorequest(bp);
582 if (!status && !(flags & XBF_ASYNC))
583 status = xfs_buf_iowait(bp);
584 return status;
585}
586
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587xfs_buf_t *
Christoph Hellwig6ad112b2009-11-24 18:02:23 +0000588xfs_buf_read(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 xfs_buftarg_t *target,
Nathan Scott204ab252006-01-11 20:50:22 +1100590 xfs_off_t ioff,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 size_t isize,
Nathan Scottce8e9222006-01-11 15:39:08 +1100592 xfs_buf_flags_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593{
Nathan Scottce8e9222006-01-11 15:39:08 +1100594 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595
Nathan Scottce8e9222006-01-11 15:39:08 +1100596 flags |= XBF_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597
Christoph Hellwig6ad112b2009-11-24 18:02:23 +0000598 bp = xfs_buf_get(target, ioff, isize, flags);
Nathan Scottce8e9222006-01-11 15:39:08 +1100599 if (bp) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000600 trace_xfs_buf_read(bp, flags, _RET_IP_);
601
Nathan Scottce8e9222006-01-11 15:39:08 +1100602 if (!XFS_BUF_ISDONE(bp)) {
Nathan Scottce8e9222006-01-11 15:39:08 +1100603 XFS_STATS_INC(xb_get_read);
Christoph Hellwig5d765b92008-12-03 12:20:26 +0100604 _xfs_buf_read(bp, flags);
Nathan Scottce8e9222006-01-11 15:39:08 +1100605 } else if (flags & XBF_ASYNC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 /*
607 * Read ahead call which is already satisfied,
608 * drop the buffer
609 */
610 goto no_buffer;
611 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 /* We do not want read in the flags */
Nathan Scottce8e9222006-01-11 15:39:08 +1100613 bp->b_flags &= ~XBF_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 }
615 }
616
Nathan Scottce8e9222006-01-11 15:39:08 +1100617 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
619 no_buffer:
Nathan Scottce8e9222006-01-11 15:39:08 +1100620 if (flags & (XBF_LOCK | XBF_TRYLOCK))
621 xfs_buf_unlock(bp);
622 xfs_buf_rele(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 return NULL;
624}
625
626/*
Nathan Scottce8e9222006-01-11 15:39:08 +1100627 * If we are not low on memory then do the readahead in a deadlock
628 * safe manner.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 */
630void
Nathan Scottce8e9222006-01-11 15:39:08 +1100631xfs_buf_readahead(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 xfs_buftarg_t *target,
Nathan Scott204ab252006-01-11 20:50:22 +1100633 xfs_off_t ioff,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 size_t isize,
Nathan Scottce8e9222006-01-11 15:39:08 +1100635 xfs_buf_flags_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636{
637 struct backing_dev_info *bdi;
638
Nathan Scottce8e9222006-01-11 15:39:08 +1100639 bdi = target->bt_mapping->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 if (bdi_read_congested(bdi))
641 return;
642
Nathan Scottce8e9222006-01-11 15:39:08 +1100643 flags |= (XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD);
Christoph Hellwig6ad112b2009-11-24 18:02:23 +0000644 xfs_buf_read(target, ioff, isize, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645}
646
647xfs_buf_t *
Nathan Scottce8e9222006-01-11 15:39:08 +1100648xfs_buf_get_empty(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 size_t len,
650 xfs_buftarg_t *target)
651{
Nathan Scottce8e9222006-01-11 15:39:08 +1100652 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
Nathan Scottce8e9222006-01-11 15:39:08 +1100654 bp = xfs_buf_allocate(0);
655 if (bp)
656 _xfs_buf_initialize(bp, target, 0, len, 0);
657 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658}
659
660static inline struct page *
661mem_to_page(
662 void *addr)
663{
Christoph Lameter9e2779f2008-02-04 22:28:34 -0800664 if ((!is_vmalloc_addr(addr))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 return virt_to_page(addr);
666 } else {
667 return vmalloc_to_page(addr);
668 }
669}
670
671int
Nathan Scottce8e9222006-01-11 15:39:08 +1100672xfs_buf_associate_memory(
673 xfs_buf_t *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 void *mem,
675 size_t len)
676{
677 int rval;
678 int i = 0;
Lachlan McIlroyd1afb672007-11-27 17:01:24 +1100679 unsigned long pageaddr;
680 unsigned long offset;
681 size_t buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 int page_count;
683
Lachlan McIlroyd1afb672007-11-27 17:01:24 +1100684 pageaddr = (unsigned long)mem & PAGE_CACHE_MASK;
685 offset = (unsigned long)mem - pageaddr;
686 buflen = PAGE_CACHE_ALIGN(len + offset);
687 page_count = buflen >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688
689 /* Free any previous set of page pointers */
Nathan Scottce8e9222006-01-11 15:39:08 +1100690 if (bp->b_pages)
691 _xfs_buf_free_pages(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
Nathan Scottce8e9222006-01-11 15:39:08 +1100693 bp->b_pages = NULL;
694 bp->b_addr = mem;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695
Christoph Hellwig36fae172009-07-18 18:14:58 -0400696 rval = _xfs_buf_get_pages(bp, page_count, XBF_DONT_BLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 if (rval)
698 return rval;
699
Nathan Scottce8e9222006-01-11 15:39:08 +1100700 bp->b_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701
Lachlan McIlroyd1afb672007-11-27 17:01:24 +1100702 for (i = 0; i < bp->b_page_count; i++) {
703 bp->b_pages[i] = mem_to_page((void *)pageaddr);
704 pageaddr += PAGE_CACHE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706
Lachlan McIlroyd1afb672007-11-27 17:01:24 +1100707 bp->b_count_desired = len;
708 bp->b_buffer_length = buflen;
Nathan Scottce8e9222006-01-11 15:39:08 +1100709 bp->b_flags |= XBF_MAPPED;
Christoph Hellwig6ab455e2008-05-19 16:34:42 +1000710 bp->b_flags &= ~_XBF_PAGE_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711
712 return 0;
713}
714
715xfs_buf_t *
Nathan Scottce8e9222006-01-11 15:39:08 +1100716xfs_buf_get_noaddr(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 size_t len,
718 xfs_buftarg_t *target)
719{
Christoph Hellwig1fa40b02007-05-14 18:23:50 +1000720 unsigned long page_count = PAGE_ALIGN(len) >> PAGE_SHIFT;
721 int error, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
Nathan Scottce8e9222006-01-11 15:39:08 +1100724 bp = xfs_buf_allocate(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 if (unlikely(bp == NULL))
726 goto fail;
Nathan Scottce8e9222006-01-11 15:39:08 +1100727 _xfs_buf_initialize(bp, target, 0, len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728
Christoph Hellwig1fa40b02007-05-14 18:23:50 +1000729 error = _xfs_buf_get_pages(bp, page_count, 0);
730 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 goto fail_free_buf;
732
Christoph Hellwig1fa40b02007-05-14 18:23:50 +1000733 for (i = 0; i < page_count; i++) {
734 bp->b_pages[i] = alloc_page(GFP_KERNEL);
735 if (!bp->b_pages[i])
736 goto fail_free_mem;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 }
Christoph Hellwig1fa40b02007-05-14 18:23:50 +1000738 bp->b_flags |= _XBF_PAGES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739
Christoph Hellwig1fa40b02007-05-14 18:23:50 +1000740 error = _xfs_buf_map_pages(bp, XBF_MAPPED);
741 if (unlikely(error)) {
742 printk(KERN_WARNING "%s: failed to map pages\n",
Harvey Harrison34a622b2008-04-10 12:19:21 +1000743 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 goto fail_free_mem;
Christoph Hellwig1fa40b02007-05-14 18:23:50 +1000745 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746
Nathan Scottce8e9222006-01-11 15:39:08 +1100747 xfs_buf_unlock(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000749 trace_xfs_buf_get_noaddr(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 return bp;
Christoph Hellwig1fa40b02007-05-14 18:23:50 +1000751
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 fail_free_mem:
Christoph Hellwig1fa40b02007-05-14 18:23:50 +1000753 while (--i >= 0)
754 __free_page(bp->b_pages[i]);
Christoph Hellwigca165b82007-05-24 15:21:11 +1000755 _xfs_buf_free_pages(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 fail_free_buf:
Christoph Hellwigca165b82007-05-24 15:21:11 +1000757 xfs_buf_deallocate(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 fail:
759 return NULL;
760}
761
762/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 * Increment reference count on buffer, to hold the buffer concurrently
764 * with another thread which may release (free) the buffer asynchronously.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 * Must hold the buffer already to call this function.
766 */
767void
Nathan Scottce8e9222006-01-11 15:39:08 +1100768xfs_buf_hold(
769 xfs_buf_t *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000771 trace_xfs_buf_hold(bp, _RET_IP_);
Nathan Scottce8e9222006-01-11 15:39:08 +1100772 atomic_inc(&bp->b_hold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773}
774
775/*
Nathan Scottce8e9222006-01-11 15:39:08 +1100776 * Releases a hold on the specified buffer. If the
777 * the hold count is 1, calls xfs_buf_free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 */
779void
Nathan Scottce8e9222006-01-11 15:39:08 +1100780xfs_buf_rele(
781 xfs_buf_t *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782{
Nathan Scottce8e9222006-01-11 15:39:08 +1100783 xfs_bufhash_t *hash = bp->b_hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000785 trace_xfs_buf_rele(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786
Nathan Scottfad3aa12006-02-01 12:14:52 +1100787 if (unlikely(!hash)) {
788 ASSERT(!bp->b_relse);
789 if (atomic_dec_and_test(&bp->b_hold))
790 xfs_buf_free(bp);
791 return;
792 }
793
Lachlan McIlroy37906892008-08-13 15:42:10 +1000794 ASSERT(atomic_read(&bp->b_hold) > 0);
Nathan Scottce8e9222006-01-11 15:39:08 +1100795 if (atomic_dec_and_lock(&bp->b_hold, &hash->bh_lock)) {
796 if (bp->b_relse) {
797 atomic_inc(&bp->b_hold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 spin_unlock(&hash->bh_lock);
Nathan Scottce8e9222006-01-11 15:39:08 +1100799 (*(bp->b_relse)) (bp);
800 } else if (bp->b_flags & XBF_FS_MANAGED) {
Christoph Hellwig7f14d0a2005-11-02 15:09:35 +1100801 spin_unlock(&hash->bh_lock);
802 } else {
Nathan Scottce8e9222006-01-11 15:39:08 +1100803 ASSERT(!(bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q)));
804 list_del_init(&bp->b_hash_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 spin_unlock(&hash->bh_lock);
Nathan Scottce8e9222006-01-11 15:39:08 +1100806 xfs_buf_free(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 }
808 }
809}
810
811
812/*
813 * Mutual exclusion on buffers. Locking model:
814 *
815 * Buffers associated with inodes for which buffer locking
816 * is not enabled are not protected by semaphores, and are
817 * assumed to be exclusively owned by the caller. There is a
818 * spinlock in the buffer, used by the caller when concurrent
819 * access is possible.
820 */
821
822/*
Nathan Scottce8e9222006-01-11 15:39:08 +1100823 * Locks a buffer object, if it is not already locked.
824 * Note that this in no way locks the underlying pages, so it is only
825 * useful for synchronizing concurrent use of buffer objects, not for
826 * synchronizing independent access to the underlying pages.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 */
828int
Nathan Scottce8e9222006-01-11 15:39:08 +1100829xfs_buf_cond_lock(
830 xfs_buf_t *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831{
832 int locked;
833
Nathan Scottce8e9222006-01-11 15:39:08 +1100834 locked = down_trylock(&bp->b_sema) == 0;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000835 if (locked)
Nathan Scottce8e9222006-01-11 15:39:08 +1100836 XB_SET_OWNER(bp);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000837
838 trace_xfs_buf_cond_lock(bp, _RET_IP_);
Nathan Scottce8e9222006-01-11 15:39:08 +1100839 return locked ? 0 : -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840}
841
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842int
Nathan Scottce8e9222006-01-11 15:39:08 +1100843xfs_buf_lock_value(
844 xfs_buf_t *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845{
Stephen Rothwelladaa6932008-04-22 15:26:13 +1000846 return bp->b_sema.count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848
849/*
Nathan Scottce8e9222006-01-11 15:39:08 +1100850 * Locks a buffer object.
851 * Note that this in no way locks the underlying pages, so it is only
852 * useful for synchronizing concurrent use of buffer objects, not for
853 * synchronizing independent access to the underlying pages.
Dave Chinnered3b4d62010-05-21 12:07:08 +1000854 *
855 * If we come across a stale, pinned, locked buffer, we know that we
856 * are being asked to lock a buffer that has been reallocated. Because
857 * it is pinned, we know that the log has not been pushed to disk and
858 * hence it will still be locked. Rather than sleeping until someone
859 * else pushes the log, push it ourselves before trying to get the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 */
Nathan Scottce8e9222006-01-11 15:39:08 +1100861void
862xfs_buf_lock(
863 xfs_buf_t *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000865 trace_xfs_buf_lock(bp, _RET_IP_);
866
Dave Chinnered3b4d62010-05-21 12:07:08 +1000867 if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))
868 xfs_log_force(bp->b_mount, 0);
Nathan Scottce8e9222006-01-11 15:39:08 +1100869 if (atomic_read(&bp->b_io_remaining))
870 blk_run_address_space(bp->b_target->bt_mapping);
871 down(&bp->b_sema);
872 XB_SET_OWNER(bp);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000873
874 trace_xfs_buf_lock_done(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875}
876
877/*
Nathan Scottce8e9222006-01-11 15:39:08 +1100878 * Releases the lock on the buffer object.
David Chinner2f926582005-09-05 08:33:35 +1000879 * If the buffer is marked delwri but is not queued, do so before we
Nathan Scottce8e9222006-01-11 15:39:08 +1100880 * unlock the buffer as we need to set flags correctly. We also need to
David Chinner2f926582005-09-05 08:33:35 +1000881 * take a reference for the delwri queue because the unlocker is going to
882 * drop their's and they don't know we just queued it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 */
884void
Nathan Scottce8e9222006-01-11 15:39:08 +1100885xfs_buf_unlock(
886 xfs_buf_t *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887{
Nathan Scottce8e9222006-01-11 15:39:08 +1100888 if ((bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q)) == XBF_DELWRI) {
889 atomic_inc(&bp->b_hold);
890 bp->b_flags |= XBF_ASYNC;
891 xfs_buf_delwri_queue(bp, 0);
David Chinner2f926582005-09-05 08:33:35 +1000892 }
893
Nathan Scottce8e9222006-01-11 15:39:08 +1100894 XB_CLEAR_OWNER(bp);
895 up(&bp->b_sema);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000896
897 trace_xfs_buf_unlock(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898}
899
900
901/*
902 * Pinning Buffer Storage in Memory
Nathan Scottce8e9222006-01-11 15:39:08 +1100903 * Ensure that no attempt to force a buffer to disk will succeed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 */
905void
Nathan Scottce8e9222006-01-11 15:39:08 +1100906xfs_buf_pin(
907 xfs_buf_t *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000909 trace_xfs_buf_pin(bp, _RET_IP_);
Nathan Scottce8e9222006-01-11 15:39:08 +1100910 atomic_inc(&bp->b_pin_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911}
912
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913void
Nathan Scottce8e9222006-01-11 15:39:08 +1100914xfs_buf_unpin(
915 xfs_buf_t *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000917 trace_xfs_buf_unpin(bp, _RET_IP_);
918
Nathan Scottce8e9222006-01-11 15:39:08 +1100919 if (atomic_dec_and_test(&bp->b_pin_count))
920 wake_up_all(&bp->b_waiters);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921}
922
923int
Nathan Scottce8e9222006-01-11 15:39:08 +1100924xfs_buf_ispin(
925 xfs_buf_t *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926{
Nathan Scottce8e9222006-01-11 15:39:08 +1100927 return atomic_read(&bp->b_pin_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928}
929
Nathan Scottce8e9222006-01-11 15:39:08 +1100930STATIC void
931xfs_buf_wait_unpin(
932 xfs_buf_t *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933{
934 DECLARE_WAITQUEUE (wait, current);
935
Nathan Scottce8e9222006-01-11 15:39:08 +1100936 if (atomic_read(&bp->b_pin_count) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 return;
938
Nathan Scottce8e9222006-01-11 15:39:08 +1100939 add_wait_queue(&bp->b_waiters, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 for (;;) {
941 set_current_state(TASK_UNINTERRUPTIBLE);
Nathan Scottce8e9222006-01-11 15:39:08 +1100942 if (atomic_read(&bp->b_pin_count) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 break;
Nathan Scottce8e9222006-01-11 15:39:08 +1100944 if (atomic_read(&bp->b_io_remaining))
945 blk_run_address_space(bp->b_target->bt_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 schedule();
947 }
Nathan Scottce8e9222006-01-11 15:39:08 +1100948 remove_wait_queue(&bp->b_waiters, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 set_current_state(TASK_RUNNING);
950}
951
952/*
953 * Buffer Utility Routines
954 */
955
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956STATIC void
Nathan Scottce8e9222006-01-11 15:39:08 +1100957xfs_buf_iodone_work(
David Howellsc4028952006-11-22 14:57:56 +0000958 struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959{
David Howellsc4028952006-11-22 14:57:56 +0000960 xfs_buf_t *bp =
961 container_of(work, xfs_buf_t, b_iodone_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962
David Chinner0bfefc42007-05-14 18:24:23 +1000963 /*
964 * We can get an EOPNOTSUPP to ordered writes. Here we clear the
965 * ordered flag and reissue them. Because we can't tell the higher
966 * layers directly that they should not issue ordered I/O anymore, they
Christoph Hellwig73f6aa42008-10-10 17:28:29 +1100967 * need to check if the _XFS_BARRIER_FAILED flag was set during I/O completion.
David Chinner0bfefc42007-05-14 18:24:23 +1000968 */
969 if ((bp->b_error == EOPNOTSUPP) &&
970 (bp->b_flags & (XBF_ORDERED|XBF_ASYNC)) == (XBF_ORDERED|XBF_ASYNC)) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000971 trace_xfs_buf_ordered_retry(bp, _RET_IP_);
David Chinner0bfefc42007-05-14 18:24:23 +1000972 bp->b_flags &= ~XBF_ORDERED;
Christoph Hellwig73f6aa42008-10-10 17:28:29 +1100973 bp->b_flags |= _XFS_BARRIER_FAILED;
David Chinner0bfefc42007-05-14 18:24:23 +1000974 xfs_buf_iorequest(bp);
975 } else if (bp->b_iodone)
Nathan Scottce8e9222006-01-11 15:39:08 +1100976 (*(bp->b_iodone))(bp);
977 else if (bp->b_flags & XBF_ASYNC)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 xfs_buf_relse(bp);
979}
980
981void
Nathan Scottce8e9222006-01-11 15:39:08 +1100982xfs_buf_ioend(
983 xfs_buf_t *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 int schedule)
985{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000986 trace_xfs_buf_iodone(bp, _RET_IP_);
987
Lachlan McIlroy77be55a2007-11-23 16:31:00 +1100988 bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD);
Nathan Scottce8e9222006-01-11 15:39:08 +1100989 if (bp->b_error == 0)
990 bp->b_flags |= XBF_DONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991
Nathan Scottce8e9222006-01-11 15:39:08 +1100992 if ((bp->b_iodone) || (bp->b_flags & XBF_ASYNC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 if (schedule) {
David Howellsc4028952006-11-22 14:57:56 +0000994 INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work);
Nathan Scottce8e9222006-01-11 15:39:08 +1100995 queue_work(xfslogd_workqueue, &bp->b_iodone_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 } else {
David Howellsc4028952006-11-22 14:57:56 +0000997 xfs_buf_iodone_work(&bp->b_iodone_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 }
999 } else {
David Chinnerb4dd3302008-08-13 16:36:11 +10001000 complete(&bp->b_iowait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 }
1002}
1003
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004void
Nathan Scottce8e9222006-01-11 15:39:08 +11001005xfs_buf_ioerror(
1006 xfs_buf_t *bp,
1007 int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008{
1009 ASSERT(error >= 0 && error <= 0xffff);
Nathan Scottce8e9222006-01-11 15:39:08 +11001010 bp->b_error = (unsigned short)error;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001011 trace_xfs_buf_ioerror(bp, error, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012}
1013
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014int
Christoph Hellwig64e0bc72010-01-13 22:17:58 +00001015xfs_bwrite(
1016 struct xfs_mount *mp,
Christoph Hellwig5d765b92008-12-03 12:20:26 +01001017 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018{
Christoph Hellwig8c383662010-03-12 10:59:40 +00001019 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020
Christoph Hellwig64e0bc72010-01-13 22:17:58 +00001021 bp->b_strat = xfs_bdstrat_cb;
1022 bp->b_mount = mp;
1023 bp->b_flags |= XBF_WRITE;
Christoph Hellwig8c383662010-03-12 10:59:40 +00001024 bp->b_flags &= ~(XBF_ASYNC | XBF_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025
Christoph Hellwig5d765b92008-12-03 12:20:26 +01001026 xfs_buf_delwri_dequeue(bp);
Christoph Hellwig64e0bc72010-01-13 22:17:58 +00001027 xfs_buf_iostrategy(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028
Christoph Hellwig8c383662010-03-12 10:59:40 +00001029 error = xfs_buf_iowait(bp);
1030 if (error)
1031 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1032 xfs_buf_relse(bp);
Christoph Hellwig64e0bc72010-01-13 22:17:58 +00001033 return error;
Christoph Hellwig5d765b92008-12-03 12:20:26 +01001034}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035
Christoph Hellwig5d765b92008-12-03 12:20:26 +01001036void
1037xfs_bdwrite(
1038 void *mp,
1039 struct xfs_buf *bp)
1040{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001041 trace_xfs_buf_bdwrite(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042
Christoph Hellwig5d765b92008-12-03 12:20:26 +01001043 bp->b_strat = xfs_bdstrat_cb;
Christoph Hellwig15ac08a2008-12-09 04:47:30 -05001044 bp->b_mount = mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045
Christoph Hellwig5d765b92008-12-03 12:20:26 +01001046 bp->b_flags &= ~XBF_READ;
1047 bp->b_flags |= (XBF_DELWRI | XBF_ASYNC);
1048
1049 xfs_buf_delwri_queue(bp, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050}
1051
Christoph Hellwig4e234712010-01-13 22:17:56 +00001052/*
1053 * Called when we want to stop a buffer from getting written or read.
1054 * We attach the EIO error, muck with its flags, and call biodone
1055 * so that the proper iodone callbacks get called.
1056 */
1057STATIC int
1058xfs_bioerror(
1059 xfs_buf_t *bp)
1060{
1061#ifdef XFSERRORDEBUG
1062 ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone);
1063#endif
1064
1065 /*
1066 * No need to wait until the buffer is unpinned, we aren't flushing it.
1067 */
1068 XFS_BUF_ERROR(bp, EIO);
1069
1070 /*
1071 * We're calling biodone, so delete XBF_DONE flag.
1072 */
1073 XFS_BUF_UNREAD(bp);
1074 XFS_BUF_UNDELAYWRITE(bp);
1075 XFS_BUF_UNDONE(bp);
1076 XFS_BUF_STALE(bp);
1077
1078 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
1079 xfs_biodone(bp);
1080
1081 return EIO;
1082}
1083
1084/*
1085 * Same as xfs_bioerror, except that we are releasing the buffer
1086 * here ourselves, and avoiding the biodone call.
1087 * This is meant for userdata errors; metadata bufs come with
1088 * iodone functions attached, so that we can track down errors.
1089 */
1090STATIC int
1091xfs_bioerror_relse(
1092 struct xfs_buf *bp)
1093{
1094 int64_t fl = XFS_BUF_BFLAGS(bp);
1095 /*
1096 * No need to wait until the buffer is unpinned.
1097 * We aren't flushing it.
1098 *
1099 * chunkhold expects B_DONE to be set, whether
1100 * we actually finish the I/O or not. We don't want to
1101 * change that interface.
1102 */
1103 XFS_BUF_UNREAD(bp);
1104 XFS_BUF_UNDELAYWRITE(bp);
1105 XFS_BUF_DONE(bp);
1106 XFS_BUF_STALE(bp);
1107 XFS_BUF_CLR_IODONE_FUNC(bp);
1108 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
Christoph Hellwig0cadda12010-01-19 09:56:44 +00001109 if (!(fl & XBF_ASYNC)) {
Christoph Hellwig4e234712010-01-13 22:17:56 +00001110 /*
1111 * Mark b_error and B_ERROR _both_.
1112 * Lot's of chunkcache code assumes that.
1113 * There's no reason to mark error for
1114 * ASYNC buffers.
1115 */
1116 XFS_BUF_ERROR(bp, EIO);
1117 XFS_BUF_FINISH_IOWAIT(bp);
1118 } else {
1119 xfs_buf_relse(bp);
1120 }
1121
1122 return EIO;
1123}
1124
1125
1126/*
1127 * All xfs metadata buffers except log state machine buffers
1128 * get this attached as their b_bdstrat callback function.
1129 * This is so that we can catch a buffer
1130 * after prematurely unpinning it to forcibly shutdown the filesystem.
1131 */
1132int
1133xfs_bdstrat_cb(
1134 struct xfs_buf *bp)
1135{
1136 if (XFS_FORCED_SHUTDOWN(bp->b_mount)) {
1137 trace_xfs_bdstrat_shut(bp, _RET_IP_);
1138 /*
1139 * Metadata write that didn't get logged but
1140 * written delayed anyway. These aren't associated
1141 * with a transaction, and can be ignored.
1142 */
1143 if (!bp->b_iodone && !XFS_BUF_ISREAD(bp))
1144 return xfs_bioerror_relse(bp);
1145 else
1146 return xfs_bioerror(bp);
1147 }
1148
1149 xfs_buf_iorequest(bp);
1150 return 0;
1151}
1152
1153/*
1154 * Wrapper around bdstrat so that we can stop data from going to disk in case
1155 * we are shutting down the filesystem. Typically user data goes thru this
1156 * path; one of the exceptions is the superblock.
1157 */
1158void
1159xfsbdstrat(
1160 struct xfs_mount *mp,
1161 struct xfs_buf *bp)
1162{
1163 if (XFS_FORCED_SHUTDOWN(mp)) {
1164 trace_xfs_bdstrat_shut(bp, _RET_IP_);
1165 xfs_bioerror_relse(bp);
1166 return;
1167 }
1168
1169 xfs_buf_iorequest(bp);
1170}
1171
Christoph Hellwigb8f82a42009-11-14 16:17:22 +00001172STATIC void
Nathan Scottce8e9222006-01-11 15:39:08 +11001173_xfs_buf_ioend(
1174 xfs_buf_t *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 int schedule)
1176{
Christoph Hellwig6ab455e2008-05-19 16:34:42 +10001177 if (atomic_dec_and_test(&bp->b_io_remaining) == 1) {
1178 bp->b_flags &= ~_XBF_PAGE_LOCKED;
Nathan Scottce8e9222006-01-11 15:39:08 +11001179 xfs_buf_ioend(bp, schedule);
Christoph Hellwig6ab455e2008-05-19 16:34:42 +10001180 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181}
1182
Al Viro782e3b32007-10-12 07:17:47 +01001183STATIC void
Nathan Scottce8e9222006-01-11 15:39:08 +11001184xfs_buf_bio_end_io(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 int error)
1187{
Nathan Scottce8e9222006-01-11 15:39:08 +11001188 xfs_buf_t *bp = (xfs_buf_t *)bio->bi_private;
1189 unsigned int blocksize = bp->b_target->bt_bsize;
Nathan Scotteedb5532005-09-02 16:39:56 +10001190 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
Lachlan McIlroycfbe5262008-12-12 15:27:25 +11001192 xfs_buf_ioerror(bp, -error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
James Bottomley73c77e22010-01-25 11:42:24 -06001194 if (!error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ))
1195 invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp));
1196
Nathan Scotteedb5532005-09-02 16:39:56 +10001197 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 struct page *page = bvec->bv_page;
1199
Nathan Scott948ecdb2006-09-28 11:03:13 +10001200 ASSERT(!PagePrivate(page));
Nathan Scottce8e9222006-01-11 15:39:08 +11001201 if (unlikely(bp->b_error)) {
1202 if (bp->b_flags & XBF_READ)
Nathan Scotteedb5532005-09-02 16:39:56 +10001203 ClearPageUptodate(page);
Nathan Scottce8e9222006-01-11 15:39:08 +11001204 } else if (blocksize >= PAGE_CACHE_SIZE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 SetPageUptodate(page);
1206 } else if (!PagePrivate(page) &&
Nathan Scottce8e9222006-01-11 15:39:08 +11001207 (bp->b_flags & _XBF_PAGE_CACHE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 set_page_region(page, bvec->bv_offset, bvec->bv_len);
1209 }
1210
Nathan Scotteedb5532005-09-02 16:39:56 +10001211 if (--bvec >= bio->bi_io_vec)
1212 prefetchw(&bvec->bv_page->flags);
Christoph Hellwig6ab455e2008-05-19 16:34:42 +10001213
1214 if (bp->b_flags & _XBF_PAGE_LOCKED)
1215 unlock_page(page);
Nathan Scotteedb5532005-09-02 16:39:56 +10001216 } while (bvec >= bio->bi_io_vec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217
Nathan Scottce8e9222006-01-11 15:39:08 +11001218 _xfs_buf_ioend(bp, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220}
1221
1222STATIC void
Nathan Scottce8e9222006-01-11 15:39:08 +11001223_xfs_buf_ioapply(
1224 xfs_buf_t *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225{
Christoph Hellwiga9759f22007-12-07 14:07:08 +11001226 int rw, map_i, total_nr_pages, nr_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 struct bio *bio;
Nathan Scottce8e9222006-01-11 15:39:08 +11001228 int offset = bp->b_offset;
1229 int size = bp->b_count_desired;
1230 sector_t sector = bp->b_bn;
1231 unsigned int blocksize = bp->b_target->bt_bsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232
Nathan Scottce8e9222006-01-11 15:39:08 +11001233 total_nr_pages = bp->b_page_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 map_i = 0;
1235
Nathan Scottce8e9222006-01-11 15:39:08 +11001236 if (bp->b_flags & XBF_ORDERED) {
1237 ASSERT(!(bp->b_flags & XBF_READ));
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001238 rw = WRITE_BARRIER;
Dave Chinner2ee1aba2009-11-24 18:03:15 +00001239 } else if (bp->b_flags & XBF_LOG_BUFFER) {
Nathan Scott51bdd702006-09-28 11:01:57 +10001240 ASSERT(!(bp->b_flags & XBF_READ_AHEAD));
1241 bp->b_flags &= ~_XBF_RUN_QUEUES;
1242 rw = (bp->b_flags & XBF_WRITE) ? WRITE_SYNC : READ_SYNC;
Dave Chinner2ee1aba2009-11-24 18:03:15 +00001243 } else if (bp->b_flags & _XBF_RUN_QUEUES) {
1244 ASSERT(!(bp->b_flags & XBF_READ_AHEAD));
1245 bp->b_flags &= ~_XBF_RUN_QUEUES;
1246 rw = (bp->b_flags & XBF_WRITE) ? WRITE_META : READ_META;
Nathan Scott51bdd702006-09-28 11:01:57 +10001247 } else {
1248 rw = (bp->b_flags & XBF_WRITE) ? WRITE :
1249 (bp->b_flags & XBF_READ_AHEAD) ? READA : READ;
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001250 }
1251
Nathan Scottce8e9222006-01-11 15:39:08 +11001252 /* Special code path for reading a sub page size buffer in --
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 * we populate up the whole page, and hence the other metadata
1254 * in the same page. This optimization is only valid when the
Nathan Scottce8e9222006-01-11 15:39:08 +11001255 * filesystem block size is not smaller than the page size.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 */
Nathan Scottce8e9222006-01-11 15:39:08 +11001257 if ((bp->b_buffer_length < PAGE_CACHE_SIZE) &&
Christoph Hellwig6ab455e2008-05-19 16:34:42 +10001258 ((bp->b_flags & (XBF_READ|_XBF_PAGE_LOCKED)) ==
1259 (XBF_READ|_XBF_PAGE_LOCKED)) &&
Nathan Scottce8e9222006-01-11 15:39:08 +11001260 (blocksize >= PAGE_CACHE_SIZE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 bio = bio_alloc(GFP_NOIO, 1);
1262
Nathan Scottce8e9222006-01-11 15:39:08 +11001263 bio->bi_bdev = bp->b_target->bt_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 bio->bi_sector = sector - (offset >> BBSHIFT);
Nathan Scottce8e9222006-01-11 15:39:08 +11001265 bio->bi_end_io = xfs_buf_bio_end_io;
1266 bio->bi_private = bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267
Nathan Scottce8e9222006-01-11 15:39:08 +11001268 bio_add_page(bio, bp->b_pages[0], PAGE_CACHE_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 size = 0;
1270
Nathan Scottce8e9222006-01-11 15:39:08 +11001271 atomic_inc(&bp->b_io_remaining);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272
1273 goto submit_io;
1274 }
1275
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276next_chunk:
Nathan Scottce8e9222006-01-11 15:39:08 +11001277 atomic_inc(&bp->b_io_remaining);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT);
1279 if (nr_pages > total_nr_pages)
1280 nr_pages = total_nr_pages;
1281
1282 bio = bio_alloc(GFP_NOIO, nr_pages);
Nathan Scottce8e9222006-01-11 15:39:08 +11001283 bio->bi_bdev = bp->b_target->bt_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 bio->bi_sector = sector;
Nathan Scottce8e9222006-01-11 15:39:08 +11001285 bio->bi_end_io = xfs_buf_bio_end_io;
1286 bio->bi_private = bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287
1288 for (; size && nr_pages; nr_pages--, map_i++) {
Nathan Scottce8e9222006-01-11 15:39:08 +11001289 int rbytes, nbytes = PAGE_CACHE_SIZE - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290
1291 if (nbytes > size)
1292 nbytes = size;
1293
Nathan Scottce8e9222006-01-11 15:39:08 +11001294 rbytes = bio_add_page(bio, bp->b_pages[map_i], nbytes, offset);
1295 if (rbytes < nbytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 break;
1297
1298 offset = 0;
1299 sector += nbytes >> BBSHIFT;
1300 size -= nbytes;
1301 total_nr_pages--;
1302 }
1303
1304submit_io:
1305 if (likely(bio->bi_size)) {
James Bottomley73c77e22010-01-25 11:42:24 -06001306 if (xfs_buf_is_vmapped(bp)) {
1307 flush_kernel_vmap_range(bp->b_addr,
1308 xfs_buf_vmap_len(bp));
1309 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 submit_bio(rw, bio);
1311 if (size)
1312 goto next_chunk;
1313 } else {
1314 bio_put(bio);
Nathan Scottce8e9222006-01-11 15:39:08 +11001315 xfs_buf_ioerror(bp, EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 }
1317}
1318
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319int
Nathan Scottce8e9222006-01-11 15:39:08 +11001320xfs_buf_iorequest(
1321 xfs_buf_t *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001323 trace_xfs_buf_iorequest(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324
Nathan Scottce8e9222006-01-11 15:39:08 +11001325 if (bp->b_flags & XBF_DELWRI) {
1326 xfs_buf_delwri_queue(bp, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 return 0;
1328 }
1329
Nathan Scottce8e9222006-01-11 15:39:08 +11001330 if (bp->b_flags & XBF_WRITE) {
1331 xfs_buf_wait_unpin(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 }
1333
Nathan Scottce8e9222006-01-11 15:39:08 +11001334 xfs_buf_hold(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335
1336 /* Set the count to 1 initially, this will stop an I/O
1337 * completion callout which happens before we have started
Nathan Scottce8e9222006-01-11 15:39:08 +11001338 * all the I/O from calling xfs_buf_ioend too early.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 */
Nathan Scottce8e9222006-01-11 15:39:08 +11001340 atomic_set(&bp->b_io_remaining, 1);
1341 _xfs_buf_ioapply(bp);
1342 _xfs_buf_ioend(bp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343
Nathan Scottce8e9222006-01-11 15:39:08 +11001344 xfs_buf_rele(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 return 0;
1346}
1347
1348/*
Nathan Scottce8e9222006-01-11 15:39:08 +11001349 * Waits for I/O to complete on the buffer supplied.
1350 * It returns immediately if no I/O is pending.
1351 * It returns the I/O error code, if any, or 0 if there was no error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 */
1353int
Nathan Scottce8e9222006-01-11 15:39:08 +11001354xfs_buf_iowait(
1355 xfs_buf_t *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001357 trace_xfs_buf_iowait(bp, _RET_IP_);
1358
Nathan Scottce8e9222006-01-11 15:39:08 +11001359 if (atomic_read(&bp->b_io_remaining))
1360 blk_run_address_space(bp->b_target->bt_mapping);
David Chinnerb4dd3302008-08-13 16:36:11 +10001361 wait_for_completion(&bp->b_iowait);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001362
1363 trace_xfs_buf_iowait_done(bp, _RET_IP_);
Nathan Scottce8e9222006-01-11 15:39:08 +11001364 return bp->b_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365}
1366
Nathan Scottce8e9222006-01-11 15:39:08 +11001367xfs_caddr_t
1368xfs_buf_offset(
1369 xfs_buf_t *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 size_t offset)
1371{
1372 struct page *page;
1373
Nathan Scottce8e9222006-01-11 15:39:08 +11001374 if (bp->b_flags & XBF_MAPPED)
1375 return XFS_BUF_PTR(bp) + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376
Nathan Scottce8e9222006-01-11 15:39:08 +11001377 offset += bp->b_offset;
1378 page = bp->b_pages[offset >> PAGE_CACHE_SHIFT];
1379 return (xfs_caddr_t)page_address(page) + (offset & (PAGE_CACHE_SIZE-1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380}
1381
1382/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 * Move data into or out of a buffer.
1384 */
1385void
Nathan Scottce8e9222006-01-11 15:39:08 +11001386xfs_buf_iomove(
1387 xfs_buf_t *bp, /* buffer to process */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 size_t boff, /* starting buffer offset */
1389 size_t bsize, /* length to copy */
Dave Chinnerb9c48642010-01-20 10:47:39 +11001390 void *data, /* data address */
Nathan Scottce8e9222006-01-11 15:39:08 +11001391 xfs_buf_rw_t mode) /* read/write/zero flag */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392{
1393 size_t bend, cpoff, csize;
1394 struct page *page;
1395
1396 bend = boff + bsize;
1397 while (boff < bend) {
Nathan Scottce8e9222006-01-11 15:39:08 +11001398 page = bp->b_pages[xfs_buf_btoct(boff + bp->b_offset)];
1399 cpoff = xfs_buf_poff(boff + bp->b_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 csize = min_t(size_t,
Nathan Scottce8e9222006-01-11 15:39:08 +11001401 PAGE_CACHE_SIZE-cpoff, bp->b_count_desired-boff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402
1403 ASSERT(((csize + cpoff) <= PAGE_CACHE_SIZE));
1404
1405 switch (mode) {
Nathan Scottce8e9222006-01-11 15:39:08 +11001406 case XBRW_ZERO:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 memset(page_address(page) + cpoff, 0, csize);
1408 break;
Nathan Scottce8e9222006-01-11 15:39:08 +11001409 case XBRW_READ:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 memcpy(data, page_address(page) + cpoff, csize);
1411 break;
Nathan Scottce8e9222006-01-11 15:39:08 +11001412 case XBRW_WRITE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 memcpy(page_address(page) + cpoff, data, csize);
1414 }
1415
1416 boff += csize;
1417 data += csize;
1418 }
1419}
1420
1421/*
Nathan Scottce8e9222006-01-11 15:39:08 +11001422 * Handling of buffer targets (buftargs).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 */
1424
1425/*
Nathan Scottce8e9222006-01-11 15:39:08 +11001426 * Wait for any bufs with callbacks that have been submitted but
1427 * have not yet returned... walk the hash list for the target.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 */
1429void
1430xfs_wait_buftarg(
1431 xfs_buftarg_t *btp)
1432{
1433 xfs_buf_t *bp, *n;
1434 xfs_bufhash_t *hash;
1435 uint i;
1436
1437 for (i = 0; i < (1 << btp->bt_hashshift); i++) {
1438 hash = &btp->bt_hash[i];
1439again:
1440 spin_lock(&hash->bh_lock);
Nathan Scottce8e9222006-01-11 15:39:08 +11001441 list_for_each_entry_safe(bp, n, &hash->bh_list, b_hash_list) {
1442 ASSERT(btp == bp->b_target);
1443 if (!(bp->b_flags & XBF_FS_MANAGED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 spin_unlock(&hash->bh_lock);
David Chinner2f926582005-09-05 08:33:35 +10001445 /*
1446 * Catch superblock reference count leaks
1447 * immediately
1448 */
Nathan Scottce8e9222006-01-11 15:39:08 +11001449 BUG_ON(bp->b_bn == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 delay(100);
1451 goto again;
1452 }
1453 }
1454 spin_unlock(&hash->bh_lock);
1455 }
1456}
1457
1458/*
Nathan Scottce8e9222006-01-11 15:39:08 +11001459 * Allocate buffer hash table for a given target.
1460 * For devices containing metadata (i.e. not the log/realtime devices)
1461 * we need to allocate a much larger hash table.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 */
1463STATIC void
1464xfs_alloc_bufhash(
1465 xfs_buftarg_t *btp,
1466 int external)
1467{
1468 unsigned int i;
1469
1470 btp->bt_hashshift = external ? 3 : 8; /* 8 or 256 buckets */
1471 btp->bt_hashmask = (1 << btp->bt_hashshift) - 1;
Christoph Hellwigbdfb0432010-01-20 21:55:30 +00001472 btp->bt_hash = kmem_zalloc_large((1 << btp->bt_hashshift) *
1473 sizeof(xfs_bufhash_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 for (i = 0; i < (1 << btp->bt_hashshift); i++) {
1475 spin_lock_init(&btp->bt_hash[i].bh_lock);
1476 INIT_LIST_HEAD(&btp->bt_hash[i].bh_list);
1477 }
1478}
1479
1480STATIC void
1481xfs_free_bufhash(
1482 xfs_buftarg_t *btp)
1483{
Christoph Hellwigbdfb0432010-01-20 21:55:30 +00001484 kmem_free_large(btp->bt_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 btp->bt_hash = NULL;
1486}
1487
David Chinnera6867a62006-01-11 15:37:58 +11001488/*
Nathan Scottce8e9222006-01-11 15:39:08 +11001489 * buftarg list for delwrite queue processing
David Chinnera6867a62006-01-11 15:37:58 +11001490 */
Tim Shimmine6a0e9c2007-05-08 13:49:59 +10001491static LIST_HEAD(xfs_buftarg_list);
David Chinner7989cb82007-02-10 18:34:56 +11001492static DEFINE_SPINLOCK(xfs_buftarg_lock);
David Chinnera6867a62006-01-11 15:37:58 +11001493
1494STATIC void
1495xfs_register_buftarg(
1496 xfs_buftarg_t *btp)
1497{
1498 spin_lock(&xfs_buftarg_lock);
1499 list_add(&btp->bt_list, &xfs_buftarg_list);
1500 spin_unlock(&xfs_buftarg_lock);
1501}
1502
1503STATIC void
1504xfs_unregister_buftarg(
1505 xfs_buftarg_t *btp)
1506{
1507 spin_lock(&xfs_buftarg_lock);
1508 list_del(&btp->bt_list);
1509 spin_unlock(&xfs_buftarg_lock);
1510}
1511
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512void
1513xfs_free_buftarg(
Christoph Hellwigb7963132009-03-03 14:48:37 -05001514 struct xfs_mount *mp,
1515 struct xfs_buftarg *btp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516{
1517 xfs_flush_buftarg(btp, 1);
Christoph Hellwigb7963132009-03-03 14:48:37 -05001518 if (mp->m_flags & XFS_MOUNT_BARRIER)
1519 xfs_blkdev_issue_flush(btp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 xfs_free_bufhash(btp);
Nathan Scottce8e9222006-01-11 15:39:08 +11001521 iput(btp->bt_mapping->host);
David Chinnera6867a62006-01-11 15:37:58 +11001522
Nathan Scottce8e9222006-01-11 15:39:08 +11001523 /* Unregister the buftarg first so that we don't get a
1524 * wakeup finding a non-existent task
1525 */
David Chinnera6867a62006-01-11 15:37:58 +11001526 xfs_unregister_buftarg(btp);
1527 kthread_stop(btp->bt_task);
1528
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001529 kmem_free(btp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530}
1531
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532STATIC int
1533xfs_setsize_buftarg_flags(
1534 xfs_buftarg_t *btp,
1535 unsigned int blocksize,
1536 unsigned int sectorsize,
1537 int verbose)
1538{
Nathan Scottce8e9222006-01-11 15:39:08 +11001539 btp->bt_bsize = blocksize;
1540 btp->bt_sshift = ffs(sectorsize) - 1;
1541 btp->bt_smask = sectorsize - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542
Nathan Scottce8e9222006-01-11 15:39:08 +11001543 if (set_blocksize(btp->bt_bdev, sectorsize)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 printk(KERN_WARNING
1545 "XFS: Cannot set_blocksize to %u on device %s\n",
1546 sectorsize, XFS_BUFTARG_NAME(btp));
1547 return EINVAL;
1548 }
1549
1550 if (verbose &&
1551 (PAGE_CACHE_SIZE / BITS_PER_LONG) > sectorsize) {
1552 printk(KERN_WARNING
1553 "XFS: %u byte sectors in use on device %s. "
1554 "This is suboptimal; %u or greater is ideal.\n",
1555 sectorsize, XFS_BUFTARG_NAME(btp),
1556 (unsigned int)PAGE_CACHE_SIZE / BITS_PER_LONG);
1557 }
1558
1559 return 0;
1560}
1561
1562/*
Nathan Scottce8e9222006-01-11 15:39:08 +11001563 * When allocating the initial buffer target we have not yet
1564 * read in the superblock, so don't know what sized sectors
1565 * are being used is at this early stage. Play safe.
1566 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567STATIC int
1568xfs_setsize_buftarg_early(
1569 xfs_buftarg_t *btp,
1570 struct block_device *bdev)
1571{
1572 return xfs_setsize_buftarg_flags(btp,
Martin K. Petersene1defc42009-05-22 17:17:49 -04001573 PAGE_CACHE_SIZE, bdev_logical_block_size(bdev), 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574}
1575
1576int
1577xfs_setsize_buftarg(
1578 xfs_buftarg_t *btp,
1579 unsigned int blocksize,
1580 unsigned int sectorsize)
1581{
1582 return xfs_setsize_buftarg_flags(btp, blocksize, sectorsize, 1);
1583}
1584
1585STATIC int
1586xfs_mapping_buftarg(
1587 xfs_buftarg_t *btp,
1588 struct block_device *bdev)
1589{
1590 struct backing_dev_info *bdi;
1591 struct inode *inode;
1592 struct address_space *mapping;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001593 static const struct address_space_operations mapping_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 .sync_page = block_sync_page,
Christoph Lametere965f962006-02-01 03:05:41 -08001595 .migratepage = fail_migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 };
1597
1598 inode = new_inode(bdev->bd_inode->i_sb);
1599 if (!inode) {
1600 printk(KERN_WARNING
1601 "XFS: Cannot allocate mapping inode for device %s\n",
1602 XFS_BUFTARG_NAME(btp));
1603 return ENOMEM;
1604 }
1605 inode->i_mode = S_IFBLK;
1606 inode->i_bdev = bdev;
1607 inode->i_rdev = bdev->bd_dev;
1608 bdi = blk_get_backing_dev_info(bdev);
1609 if (!bdi)
1610 bdi = &default_backing_dev_info;
1611 mapping = &inode->i_data;
1612 mapping->a_ops = &mapping_aops;
1613 mapping->backing_dev_info = bdi;
1614 mapping_set_gfp_mask(mapping, GFP_NOFS);
Nathan Scottce8e9222006-01-11 15:39:08 +11001615 btp->bt_mapping = mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 return 0;
1617}
1618
David Chinnera6867a62006-01-11 15:37:58 +11001619STATIC int
1620xfs_alloc_delwrite_queue(
Jan Engelhardte2a07812010-03-23 09:52:55 +11001621 xfs_buftarg_t *btp,
1622 const char *fsname)
David Chinnera6867a62006-01-11 15:37:58 +11001623{
1624 int error = 0;
1625
1626 INIT_LIST_HEAD(&btp->bt_list);
1627 INIT_LIST_HEAD(&btp->bt_delwrite_queue);
Eric Sandeen007c61c2007-10-11 17:43:56 +10001628 spin_lock_init(&btp->bt_delwrite_lock);
David Chinnera6867a62006-01-11 15:37:58 +11001629 btp->bt_flags = 0;
Jan Engelhardte2a07812010-03-23 09:52:55 +11001630 btp->bt_task = kthread_run(xfsbufd, btp, "xfsbufd/%s", fsname);
David Chinnera6867a62006-01-11 15:37:58 +11001631 if (IS_ERR(btp->bt_task)) {
1632 error = PTR_ERR(btp->bt_task);
1633 goto out_error;
1634 }
1635 xfs_register_buftarg(btp);
1636out_error:
1637 return error;
1638}
1639
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640xfs_buftarg_t *
1641xfs_alloc_buftarg(
1642 struct block_device *bdev,
Jan Engelhardte2a07812010-03-23 09:52:55 +11001643 int external,
1644 const char *fsname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645{
1646 xfs_buftarg_t *btp;
1647
1648 btp = kmem_zalloc(sizeof(*btp), KM_SLEEP);
1649
Nathan Scottce8e9222006-01-11 15:39:08 +11001650 btp->bt_dev = bdev->bd_dev;
1651 btp->bt_bdev = bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 if (xfs_setsize_buftarg_early(btp, bdev))
1653 goto error;
1654 if (xfs_mapping_buftarg(btp, bdev))
1655 goto error;
Jan Engelhardte2a07812010-03-23 09:52:55 +11001656 if (xfs_alloc_delwrite_queue(btp, fsname))
David Chinnera6867a62006-01-11 15:37:58 +11001657 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 xfs_alloc_bufhash(btp, external);
1659 return btp;
1660
1661error:
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001662 kmem_free(btp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 return NULL;
1664}
1665
1666
1667/*
Nathan Scottce8e9222006-01-11 15:39:08 +11001668 * Delayed write buffer handling
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670STATIC void
Nathan Scottce8e9222006-01-11 15:39:08 +11001671xfs_buf_delwri_queue(
1672 xfs_buf_t *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 int unlock)
1674{
Nathan Scottce8e9222006-01-11 15:39:08 +11001675 struct list_head *dwq = &bp->b_target->bt_delwrite_queue;
1676 spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock;
David Chinnera6867a62006-01-11 15:37:58 +11001677
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001678 trace_xfs_buf_delwri_queue(bp, _RET_IP_);
1679
Nathan Scottce8e9222006-01-11 15:39:08 +11001680 ASSERT((bp->b_flags&(XBF_DELWRI|XBF_ASYNC)) == (XBF_DELWRI|XBF_ASYNC));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681
David Chinnera6867a62006-01-11 15:37:58 +11001682 spin_lock(dwlk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 /* If already in the queue, dequeue and place at tail */
Nathan Scottce8e9222006-01-11 15:39:08 +11001684 if (!list_empty(&bp->b_list)) {
1685 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
1686 if (unlock)
1687 atomic_dec(&bp->b_hold);
1688 list_del(&bp->b_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 }
1690
Dave Chinnerc9c12972010-01-11 11:49:59 +00001691 if (list_empty(dwq)) {
1692 /* start xfsbufd as it is about to have something to do */
1693 wake_up_process(bp->b_target->bt_task);
1694 }
1695
Nathan Scottce8e9222006-01-11 15:39:08 +11001696 bp->b_flags |= _XBF_DELWRI_Q;
1697 list_add_tail(&bp->b_list, dwq);
1698 bp->b_queuetime = jiffies;
David Chinnera6867a62006-01-11 15:37:58 +11001699 spin_unlock(dwlk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700
1701 if (unlock)
Nathan Scottce8e9222006-01-11 15:39:08 +11001702 xfs_buf_unlock(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703}
1704
1705void
Nathan Scottce8e9222006-01-11 15:39:08 +11001706xfs_buf_delwri_dequeue(
1707 xfs_buf_t *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708{
Nathan Scottce8e9222006-01-11 15:39:08 +11001709 spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 int dequeued = 0;
1711
David Chinnera6867a62006-01-11 15:37:58 +11001712 spin_lock(dwlk);
Nathan Scottce8e9222006-01-11 15:39:08 +11001713 if ((bp->b_flags & XBF_DELWRI) && !list_empty(&bp->b_list)) {
1714 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
1715 list_del_init(&bp->b_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 dequeued = 1;
1717 }
Nathan Scottce8e9222006-01-11 15:39:08 +11001718 bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q);
David Chinnera6867a62006-01-11 15:37:58 +11001719 spin_unlock(dwlk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720
1721 if (dequeued)
Nathan Scottce8e9222006-01-11 15:39:08 +11001722 xfs_buf_rele(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001724 trace_xfs_buf_delwri_dequeue(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725}
1726
Dave Chinnerd808f612010-02-02 10:13:42 +11001727/*
1728 * If a delwri buffer needs to be pushed before it has aged out, then promote
1729 * it to the head of the delwri queue so that it will be flushed on the next
1730 * xfsbufd run. We do this by resetting the queuetime of the buffer to be older
1731 * than the age currently needed to flush the buffer. Hence the next time the
1732 * xfsbufd sees it is guaranteed to be considered old enough to flush.
1733 */
1734void
1735xfs_buf_delwri_promote(
1736 struct xfs_buf *bp)
1737{
1738 struct xfs_buftarg *btp = bp->b_target;
1739 long age = xfs_buf_age_centisecs * msecs_to_jiffies(10) + 1;
1740
1741 ASSERT(bp->b_flags & XBF_DELWRI);
1742 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
1743
1744 /*
1745 * Check the buffer age before locking the delayed write queue as we
1746 * don't need to promote buffers that are already past the flush age.
1747 */
1748 if (bp->b_queuetime < jiffies - age)
1749 return;
1750 bp->b_queuetime = jiffies - age;
1751 spin_lock(&btp->bt_delwrite_lock);
1752 list_move(&bp->b_list, &btp->bt_delwrite_queue);
1753 spin_unlock(&btp->bt_delwrite_lock);
1754}
1755
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756STATIC void
Nathan Scottce8e9222006-01-11 15:39:08 +11001757xfs_buf_runall_queues(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 struct workqueue_struct *queue)
1759{
1760 flush_workqueue(queue);
1761}
1762
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763STATIC int
Christoph Hellwig23ea4032005-06-21 15:14:01 +10001764xfsbufd_wakeup(
Nathan Scott15c84a42005-11-04 10:51:01 +11001765 int priority,
1766 gfp_t mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767{
Christoph Hellwigda7f93e2006-01-11 20:49:57 +11001768 xfs_buftarg_t *btp;
David Chinnera6867a62006-01-11 15:37:58 +11001769
1770 spin_lock(&xfs_buftarg_lock);
Christoph Hellwigda7f93e2006-01-11 20:49:57 +11001771 list_for_each_entry(btp, &xfs_buftarg_list, bt_list) {
Nathan Scottce8e9222006-01-11 15:39:08 +11001772 if (test_bit(XBT_FORCE_SLEEP, &btp->bt_flags))
David Chinnera6867a62006-01-11 15:37:58 +11001773 continue;
Dave Chinnerc9c12972010-01-11 11:49:59 +00001774 if (list_empty(&btp->bt_delwrite_queue))
1775 continue;
Nathan Scottce8e9222006-01-11 15:39:08 +11001776 set_bit(XBT_FORCE_FLUSH, &btp->bt_flags);
David Chinnera6867a62006-01-11 15:37:58 +11001777 wake_up_process(btp->bt_task);
1778 }
1779 spin_unlock(&xfs_buftarg_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 return 0;
1781}
1782
David Chinner585e6d82007-02-10 18:32:29 +11001783/*
1784 * Move as many buffers as specified to the supplied list
1785 * idicating if we skipped any buffers to prevent deadlocks.
1786 */
1787STATIC int
1788xfs_buf_delwri_split(
1789 xfs_buftarg_t *target,
1790 struct list_head *list,
David Chinner5e6a07d2007-02-10 18:34:49 +11001791 unsigned long age)
David Chinner585e6d82007-02-10 18:32:29 +11001792{
1793 xfs_buf_t *bp, *n;
1794 struct list_head *dwq = &target->bt_delwrite_queue;
1795 spinlock_t *dwlk = &target->bt_delwrite_lock;
1796 int skipped = 0;
David Chinner5e6a07d2007-02-10 18:34:49 +11001797 int force;
David Chinner585e6d82007-02-10 18:32:29 +11001798
David Chinner5e6a07d2007-02-10 18:34:49 +11001799 force = test_and_clear_bit(XBT_FORCE_FLUSH, &target->bt_flags);
David Chinner585e6d82007-02-10 18:32:29 +11001800 INIT_LIST_HEAD(list);
1801 spin_lock(dwlk);
1802 list_for_each_entry_safe(bp, n, dwq, b_list) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001803 trace_xfs_buf_delwri_split(bp, _RET_IP_);
David Chinner585e6d82007-02-10 18:32:29 +11001804 ASSERT(bp->b_flags & XBF_DELWRI);
1805
1806 if (!xfs_buf_ispin(bp) && !xfs_buf_cond_lock(bp)) {
David Chinner5e6a07d2007-02-10 18:34:49 +11001807 if (!force &&
David Chinner585e6d82007-02-10 18:32:29 +11001808 time_before(jiffies, bp->b_queuetime + age)) {
1809 xfs_buf_unlock(bp);
1810 break;
1811 }
1812
1813 bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q|
1814 _XBF_RUN_QUEUES);
1815 bp->b_flags |= XBF_WRITE;
1816 list_move_tail(&bp->b_list, list);
1817 } else
1818 skipped++;
1819 }
1820 spin_unlock(dwlk);
1821
1822 return skipped;
1823
1824}
1825
Dave Chinner089716a2010-01-26 15:13:25 +11001826/*
1827 * Compare function is more complex than it needs to be because
1828 * the return value is only 32 bits and we are doing comparisons
1829 * on 64 bit values
1830 */
1831static int
1832xfs_buf_cmp(
1833 void *priv,
1834 struct list_head *a,
1835 struct list_head *b)
1836{
1837 struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list);
1838 struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list);
1839 xfs_daddr_t diff;
1840
1841 diff = ap->b_bn - bp->b_bn;
1842 if (diff < 0)
1843 return -1;
1844 if (diff > 0)
1845 return 1;
1846 return 0;
1847}
1848
1849void
1850xfs_buf_delwri_sort(
1851 xfs_buftarg_t *target,
1852 struct list_head *list)
1853{
1854 list_sort(NULL, list, xfs_buf_cmp);
1855}
1856
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857STATIC int
Christoph Hellwig23ea4032005-06-21 15:14:01 +10001858xfsbufd(
David Chinner585e6d82007-02-10 18:32:29 +11001859 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860{
Dave Chinner089716a2010-01-26 15:13:25 +11001861 xfs_buftarg_t *target = (xfs_buftarg_t *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 current->flags |= PF_MEMALLOC;
1864
Rafael J. Wysocki978c7b22007-12-07 14:09:02 +11001865 set_freezable();
1866
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 do {
Dave Chinnerc9c12972010-01-11 11:49:59 +00001868 long age = xfs_buf_age_centisecs * msecs_to_jiffies(10);
1869 long tout = xfs_buf_timer_centisecs * msecs_to_jiffies(10);
Dave Chinner089716a2010-01-26 15:13:25 +11001870 int count = 0;
1871 struct list_head tmp;
Dave Chinnerc9c12972010-01-11 11:49:59 +00001872
Christoph Lameter3e1d1d22005-06-24 23:13:50 -07001873 if (unlikely(freezing(current))) {
Nathan Scottce8e9222006-01-11 15:39:08 +11001874 set_bit(XBT_FORCE_SLEEP, &target->bt_flags);
Christoph Lameter3e1d1d22005-06-24 23:13:50 -07001875 refrigerator();
Nathan Scottabd0cf72005-05-05 13:30:13 -07001876 } else {
Nathan Scottce8e9222006-01-11 15:39:08 +11001877 clear_bit(XBT_FORCE_SLEEP, &target->bt_flags);
Nathan Scottabd0cf72005-05-05 13:30:13 -07001878 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879
Dave Chinnerc9c12972010-01-11 11:49:59 +00001880 /* sleep for a long time if there is nothing to do. */
1881 if (list_empty(&target->bt_delwrite_queue))
1882 tout = MAX_SCHEDULE_TIMEOUT;
1883 schedule_timeout_interruptible(tout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884
Dave Chinnerc9c12972010-01-11 11:49:59 +00001885 xfs_buf_delwri_split(target, &tmp, age);
Dave Chinner089716a2010-01-26 15:13:25 +11001886 list_sort(NULL, &tmp, xfs_buf_cmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 while (!list_empty(&tmp)) {
Dave Chinner089716a2010-01-26 15:13:25 +11001888 struct xfs_buf *bp;
1889 bp = list_first_entry(&tmp, struct xfs_buf, b_list);
Nathan Scottce8e9222006-01-11 15:39:08 +11001890 list_del_init(&bp->b_list);
1891 xfs_buf_iostrategy(bp);
David Chinner585e6d82007-02-10 18:32:29 +11001892 count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 }
Nathan Scottf07c2252006-09-28 10:52:15 +10001894 if (count)
1895 blk_run_address_space(target->bt_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896
Christoph Hellwig4df08c52005-09-05 08:34:18 +10001897 } while (!kthread_should_stop());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898
Christoph Hellwig4df08c52005-09-05 08:34:18 +10001899 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900}
1901
1902/*
Nathan Scottce8e9222006-01-11 15:39:08 +11001903 * Go through all incore buffers, and release buffers if they belong to
1904 * the given device. This is used in filesystem error handling to
1905 * preserve the consistency of its metadata.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 */
1907int
1908xfs_flush_buftarg(
David Chinner585e6d82007-02-10 18:32:29 +11001909 xfs_buftarg_t *target,
1910 int wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911{
Dave Chinner089716a2010-01-26 15:13:25 +11001912 xfs_buf_t *bp;
David Chinner585e6d82007-02-10 18:32:29 +11001913 int pincount = 0;
Dave Chinner089716a2010-01-26 15:13:25 +11001914 LIST_HEAD(tmp_list);
1915 LIST_HEAD(wait_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916
Dave Chinnerc626d172009-04-06 18:42:11 +02001917 xfs_buf_runall_queues(xfsconvertd_workqueue);
Nathan Scottce8e9222006-01-11 15:39:08 +11001918 xfs_buf_runall_queues(xfsdatad_workqueue);
1919 xfs_buf_runall_queues(xfslogd_workqueue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920
David Chinner5e6a07d2007-02-10 18:34:49 +11001921 set_bit(XBT_FORCE_FLUSH, &target->bt_flags);
Dave Chinner089716a2010-01-26 15:13:25 +11001922 pincount = xfs_buf_delwri_split(target, &tmp_list, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923
1924 /*
Dave Chinner089716a2010-01-26 15:13:25 +11001925 * Dropped the delayed write list lock, now walk the temporary list.
1926 * All I/O is issued async and then if we need to wait for completion
1927 * we do that after issuing all the IO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928 */
Dave Chinner089716a2010-01-26 15:13:25 +11001929 list_sort(NULL, &tmp_list, xfs_buf_cmp);
1930 while (!list_empty(&tmp_list)) {
1931 bp = list_first_entry(&tmp_list, struct xfs_buf, b_list);
David Chinner585e6d82007-02-10 18:32:29 +11001932 ASSERT(target == bp->b_target);
Dave Chinner089716a2010-01-26 15:13:25 +11001933 list_del_init(&bp->b_list);
1934 if (wait) {
Nathan Scottce8e9222006-01-11 15:39:08 +11001935 bp->b_flags &= ~XBF_ASYNC;
Dave Chinner089716a2010-01-26 15:13:25 +11001936 list_add(&bp->b_list, &wait_list);
1937 }
Nathan Scottce8e9222006-01-11 15:39:08 +11001938 xfs_buf_iostrategy(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 }
1940
Dave Chinner089716a2010-01-26 15:13:25 +11001941 if (wait) {
1942 /* Expedite and wait for IO to complete. */
Nathan Scottf07c2252006-09-28 10:52:15 +10001943 blk_run_address_space(target->bt_mapping);
Dave Chinner089716a2010-01-26 15:13:25 +11001944 while (!list_empty(&wait_list)) {
1945 bp = list_first_entry(&wait_list, struct xfs_buf, b_list);
Nathan Scottf07c2252006-09-28 10:52:15 +10001946
Dave Chinner089716a2010-01-26 15:13:25 +11001947 list_del_init(&bp->b_list);
1948 xfs_iowait(bp);
1949 xfs_buf_relse(bp);
1950 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 }
1952
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 return pincount;
1954}
1955
Christoph Hellwig04d8b282005-11-02 10:15:05 +11001956int __init
Nathan Scottce8e9222006-01-11 15:39:08 +11001957xfs_buf_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958{
Nathan Scott87582802006-03-14 13:18:19 +11001959 xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf",
1960 KM_ZONE_HWALIGN, NULL);
Nathan Scottce8e9222006-01-11 15:39:08 +11001961 if (!xfs_buf_zone)
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001962 goto out;
Christoph Hellwig04d8b282005-11-02 10:15:05 +11001963
Rafael J. Wysockib4337692007-03-22 00:11:27 -08001964 xfslogd_workqueue = create_workqueue("xfslogd");
Christoph Hellwig23ea4032005-06-21 15:14:01 +10001965 if (!xfslogd_workqueue)
Christoph Hellwig04d8b282005-11-02 10:15:05 +11001966 goto out_free_buf_zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967
Rafael J. Wysockib4337692007-03-22 00:11:27 -08001968 xfsdatad_workqueue = create_workqueue("xfsdatad");
Christoph Hellwig23ea4032005-06-21 15:14:01 +10001969 if (!xfsdatad_workqueue)
1970 goto out_destroy_xfslogd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971
Dave Chinnerc626d172009-04-06 18:42:11 +02001972 xfsconvertd_workqueue = create_workqueue("xfsconvertd");
1973 if (!xfsconvertd_workqueue)
1974 goto out_destroy_xfsdatad_workqueue;
1975
Rusty Russell8e1f9362007-07-17 04:03:17 -07001976 register_shrinker(&xfs_buf_shake);
Christoph Hellwig23ea4032005-06-21 15:14:01 +10001977 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978
Dave Chinnerc626d172009-04-06 18:42:11 +02001979 out_destroy_xfsdatad_workqueue:
1980 destroy_workqueue(xfsdatad_workqueue);
Christoph Hellwig23ea4032005-06-21 15:14:01 +10001981 out_destroy_xfslogd_workqueue:
1982 destroy_workqueue(xfslogd_workqueue);
Christoph Hellwig23ea4032005-06-21 15:14:01 +10001983 out_free_buf_zone:
Nathan Scottce8e9222006-01-11 15:39:08 +11001984 kmem_zone_destroy(xfs_buf_zone);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001985 out:
Nathan Scott87582802006-03-14 13:18:19 +11001986 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987}
1988
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989void
Nathan Scottce8e9222006-01-11 15:39:08 +11001990xfs_buf_terminate(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991{
Rusty Russell8e1f9362007-07-17 04:03:17 -07001992 unregister_shrinker(&xfs_buf_shake);
Dave Chinnerc626d172009-04-06 18:42:11 +02001993 destroy_workqueue(xfsconvertd_workqueue);
Christoph Hellwig04d8b282005-11-02 10:15:05 +11001994 destroy_workqueue(xfsdatad_workqueue);
1995 destroy_workqueue(xfslogd_workqueue);
Nathan Scottce8e9222006-01-11 15:39:08 +11001996 kmem_zone_destroy(xfs_buf_zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997}
Tim Shimmine6a0e9c2007-05-08 13:49:59 +10001998
1999#ifdef CONFIG_KDB_MODULES
2000struct list_head *
2001xfs_get_buftarg_list(void)
2002{
2003 return &xfs_buftarg_list;
2004}
2005#endif