blob: 704418a04991704d21382ab0855318a661c3e98c [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
Christoph Hellwigb7963132009-03-03 14:48:37 -050037#include "xfs_sb.h"
38#include "xfs_inum.h"
Dave Chinnered3b4d62010-05-21 12:07:08 +100039#include "xfs_log.h"
Christoph Hellwigb7963132009-03-03 14:48:37 -050040#include "xfs_ag.h"
Christoph Hellwigb7963132009-03-03 14:48:37 -050041#include "xfs_mount.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000042#include "xfs_trace.h"
Christoph Hellwigb7963132009-03-03 14:48:37 -050043
David Chinner7989cb82007-02-10 18:34:56 +110044static kmem_zone_t *xfs_buf_zone;
David Chinnera6867a62006-01-11 15:37:58 +110045STATIC int xfsbufd(void *);
Nathan Scottce8e9222006-01-11 15:39:08 +110046STATIC void xfs_buf_delwri_queue(xfs_buf_t *, int);
Christoph Hellwig23ea4032005-06-21 15:14:01 +100047
David Chinner7989cb82007-02-10 18:34:56 +110048static struct workqueue_struct *xfslogd_workqueue;
Christoph Hellwig0829c362005-09-02 16:58:49 +100049struct workqueue_struct *xfsdatad_workqueue;
Dave Chinnerc626d172009-04-06 18:42:11 +020050struct workqueue_struct *xfsconvertd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
Nathan Scottce8e9222006-01-11 15:39:08 +110052#ifdef XFS_BUF_LOCK_TRACKING
53# define XB_SET_OWNER(bp) ((bp)->b_last_holder = current->pid)
54# define XB_CLEAR_OWNER(bp) ((bp)->b_last_holder = -1)
55# define XB_GET_OWNER(bp) ((bp)->b_last_holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -070056#else
Nathan Scottce8e9222006-01-11 15:39:08 +110057# define XB_SET_OWNER(bp) do { } while (0)
58# define XB_CLEAR_OWNER(bp) do { } while (0)
59# define XB_GET_OWNER(bp) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -070060#endif
61
Nathan Scottce8e9222006-01-11 15:39:08 +110062#define xb_to_gfp(flags) \
63 ((((flags) & XBF_READ_AHEAD) ? __GFP_NORETRY : \
64 ((flags) & XBF_DONT_BLOCK) ? GFP_NOFS : GFP_KERNEL) | __GFP_NOWARN)
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
Nathan Scottce8e9222006-01-11 15:39:08 +110066#define xb_to_km(flags) \
67 (((flags) & XBF_DONT_BLOCK) ? KM_NOFS : KM_SLEEP)
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
Nathan Scottce8e9222006-01-11 15:39:08 +110069#define xfs_buf_allocate(flags) \
70 kmem_zone_alloc(xfs_buf_zone, xb_to_km(flags))
71#define xfs_buf_deallocate(bp) \
72 kmem_zone_free(xfs_buf_zone, (bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -070073
James Bottomley73c77e22010-01-25 11:42:24 -060074static inline int
75xfs_buf_is_vmapped(
76 struct xfs_buf *bp)
77{
78 /*
79 * Return true if the buffer is vmapped.
80 *
81 * The XBF_MAPPED flag is set if the buffer should be mapped, but the
82 * code is clever enough to know it doesn't have to map a single page,
83 * so the check has to be both for XBF_MAPPED and bp->b_page_count > 1.
84 */
85 return (bp->b_flags & XBF_MAPPED) && bp->b_page_count > 1;
86}
87
88static inline int
89xfs_buf_vmap_len(
90 struct xfs_buf *bp)
91{
92 return (bp->b_page_count * PAGE_SIZE) - bp->b_offset;
93}
94
Linus Torvalds1da177e2005-04-16 15:20:36 -070095/*
Dave Chinner430cbeb2010-12-02 16:30:55 +110096 * xfs_buf_lru_add - add a buffer to the LRU.
97 *
98 * The LRU takes a new reference to the buffer so that it will only be freed
99 * once the shrinker takes the buffer off the LRU.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 */
Dave Chinner430cbeb2010-12-02 16:30:55 +1100101STATIC void
102xfs_buf_lru_add(
103 struct xfs_buf *bp)
104{
105 struct xfs_buftarg *btp = bp->b_target;
106
107 spin_lock(&btp->bt_lru_lock);
108 if (list_empty(&bp->b_lru)) {
109 atomic_inc(&bp->b_hold);
110 list_add_tail(&bp->b_lru, &btp->bt_lru);
111 btp->bt_lru_nr++;
112 }
113 spin_unlock(&btp->bt_lru_lock);
114}
115
116/*
117 * xfs_buf_lru_del - remove a buffer from the LRU
118 *
119 * The unlocked check is safe here because it only occurs when there are not
120 * b_lru_ref counts left on the inode under the pag->pag_buf_lock. it is there
121 * to optimise the shrinker removing the buffer from the LRU and calling
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300122 * xfs_buf_free(). i.e. it removes an unnecessary round trip on the
Dave Chinner430cbeb2010-12-02 16:30:55 +1100123 * bt_lru_lock.
124 */
125STATIC void
126xfs_buf_lru_del(
127 struct xfs_buf *bp)
128{
129 struct xfs_buftarg *btp = bp->b_target;
130
131 if (list_empty(&bp->b_lru))
132 return;
133
134 spin_lock(&btp->bt_lru_lock);
135 if (!list_empty(&bp->b_lru)) {
136 list_del_init(&bp->b_lru);
137 btp->bt_lru_nr--;
138 }
139 spin_unlock(&btp->bt_lru_lock);
140}
141
142/*
143 * When we mark a buffer stale, we remove the buffer from the LRU and clear the
144 * b_lru_ref count so that the buffer is freed immediately when the buffer
145 * reference count falls to zero. If the buffer is already on the LRU, we need
146 * to remove the reference that LRU holds on the buffer.
147 *
148 * This prevents build-up of stale buffers on the LRU.
149 */
150void
151xfs_buf_stale(
152 struct xfs_buf *bp)
153{
154 bp->b_flags |= XBF_STALE;
155 atomic_set(&(bp)->b_lru_ref, 0);
156 if (!list_empty(&bp->b_lru)) {
157 struct xfs_buftarg *btp = bp->b_target;
158
159 spin_lock(&btp->bt_lru_lock);
160 if (!list_empty(&bp->b_lru)) {
161 list_del_init(&bp->b_lru);
162 btp->bt_lru_nr--;
163 atomic_dec(&bp->b_hold);
164 }
165 spin_unlock(&btp->bt_lru_lock);
166 }
167 ASSERT(atomic_read(&bp->b_hold) >= 1);
168}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169
170STATIC void
Nathan Scottce8e9222006-01-11 15:39:08 +1100171_xfs_buf_initialize(
172 xfs_buf_t *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 xfs_buftarg_t *target,
Nathan Scott204ab252006-01-11 20:50:22 +1100174 xfs_off_t range_base,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 size_t range_length,
Nathan Scottce8e9222006-01-11 15:39:08 +1100176 xfs_buf_flags_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177{
178 /*
Nathan Scottce8e9222006-01-11 15:39:08 +1100179 * We don't want certain flags to appear in b_flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 */
Nathan Scottce8e9222006-01-11 15:39:08 +1100181 flags &= ~(XBF_LOCK|XBF_MAPPED|XBF_DONT_BLOCK|XBF_READ_AHEAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182
Nathan Scottce8e9222006-01-11 15:39:08 +1100183 memset(bp, 0, sizeof(xfs_buf_t));
184 atomic_set(&bp->b_hold, 1);
Dave Chinner430cbeb2010-12-02 16:30:55 +1100185 atomic_set(&bp->b_lru_ref, 1);
David Chinnerb4dd3302008-08-13 16:36:11 +1000186 init_completion(&bp->b_iowait);
Dave Chinner430cbeb2010-12-02 16:30:55 +1100187 INIT_LIST_HEAD(&bp->b_lru);
Nathan Scottce8e9222006-01-11 15:39:08 +1100188 INIT_LIST_HEAD(&bp->b_list);
Dave Chinner74f75a02010-09-24 19:59:04 +1000189 RB_CLEAR_NODE(&bp->b_rbnode);
Thomas Gleixnera731cd12010-09-07 14:33:15 +0000190 sema_init(&bp->b_sema, 0); /* held, no waiters */
Nathan Scottce8e9222006-01-11 15:39:08 +1100191 XB_SET_OWNER(bp);
192 bp->b_target = target;
193 bp->b_file_offset = range_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 /*
195 * Set buffer_length and count_desired to the same value initially.
196 * I/O routines should use count_desired, which will be the same in
197 * most cases but may be reset (e.g. XFS recovery).
198 */
Nathan Scottce8e9222006-01-11 15:39:08 +1100199 bp->b_buffer_length = bp->b_count_desired = range_length;
200 bp->b_flags = flags;
201 bp->b_bn = XFS_BUF_DADDR_NULL;
202 atomic_set(&bp->b_pin_count, 0);
203 init_waitqueue_head(&bp->b_waiters);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
Nathan Scottce8e9222006-01-11 15:39:08 +1100205 XFS_STATS_INC(xb_create);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000206
207 trace_xfs_buf_init(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208}
209
210/*
Nathan Scottce8e9222006-01-11 15:39:08 +1100211 * Allocate a page array capable of holding a specified number
212 * of pages, and point the page buf at it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 */
214STATIC int
Nathan Scottce8e9222006-01-11 15:39:08 +1100215_xfs_buf_get_pages(
216 xfs_buf_t *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 int page_count,
Nathan Scottce8e9222006-01-11 15:39:08 +1100218 xfs_buf_flags_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219{
220 /* Make sure that we have a page list */
Nathan Scottce8e9222006-01-11 15:39:08 +1100221 if (bp->b_pages == NULL) {
222 bp->b_offset = xfs_buf_poff(bp->b_file_offset);
223 bp->b_page_count = page_count;
224 if (page_count <= XB_PAGES) {
225 bp->b_pages = bp->b_page_array;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 } else {
Nathan Scottce8e9222006-01-11 15:39:08 +1100227 bp->b_pages = kmem_alloc(sizeof(struct page *) *
228 page_count, xb_to_km(flags));
229 if (bp->b_pages == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 return -ENOMEM;
231 }
Nathan Scottce8e9222006-01-11 15:39:08 +1100232 memset(bp->b_pages, 0, sizeof(struct page *) * page_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 }
234 return 0;
235}
236
237/*
Nathan Scottce8e9222006-01-11 15:39:08 +1100238 * Frees b_pages if it was allocated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 */
240STATIC void
Nathan Scottce8e9222006-01-11 15:39:08 +1100241_xfs_buf_free_pages(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 xfs_buf_t *bp)
243{
Nathan Scottce8e9222006-01-11 15:39:08 +1100244 if (bp->b_pages != bp->b_page_array) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +1000245 kmem_free(bp->b_pages);
Dave Chinner3fc98b12009-12-14 23:11:57 +0000246 bp->b_pages = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 }
248}
249
250/*
251 * Releases the specified buffer.
252 *
253 * The modification state of any associated pages is left unchanged.
Nathan Scottce8e9222006-01-11 15:39:08 +1100254 * The buffer most not be on any hash - use xfs_buf_rele instead for
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 * hashed and refcounted buffers
256 */
257void
Nathan Scottce8e9222006-01-11 15:39:08 +1100258xfs_buf_free(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 xfs_buf_t *bp)
260{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000261 trace_xfs_buf_free(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262
Dave Chinner430cbeb2010-12-02 16:30:55 +1100263 ASSERT(list_empty(&bp->b_lru));
264
Dave Chinner0e6e8472011-03-26 09:16:45 +1100265 if (bp->b_flags & _XBF_PAGES) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 uint i;
267
James Bottomley73c77e22010-01-25 11:42:24 -0600268 if (xfs_buf_is_vmapped(bp))
Alex Elder8a262e52010-03-16 18:55:56 +0000269 vm_unmap_ram(bp->b_addr - bp->b_offset,
270 bp->b_page_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271
Nathan Scott948ecdb2006-09-28 11:03:13 +1000272 for (i = 0; i < bp->b_page_count; i++) {
273 struct page *page = bp->b_pages[i];
274
Dave Chinner0e6e8472011-03-26 09:16:45 +1100275 __free_page(page);
Nathan Scott948ecdb2006-09-28 11:03:13 +1000276 }
Dave Chinner0e6e8472011-03-26 09:16:45 +1100277 } else if (bp->b_flags & _XBF_KMEM)
278 kmem_free(bp->b_addr);
Dave Chinner3fc98b12009-12-14 23:11:57 +0000279 _xfs_buf_free_pages(bp);
Nathan Scottce8e9222006-01-11 15:39:08 +1100280 xfs_buf_deallocate(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281}
282
283/*
Dave Chinner0e6e8472011-03-26 09:16:45 +1100284 * Allocates all the pages for buffer in question and builds it's page list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 */
286STATIC int
Dave Chinner0e6e8472011-03-26 09:16:45 +1100287xfs_buf_allocate_memory(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 xfs_buf_t *bp,
289 uint flags)
290{
Nathan Scottce8e9222006-01-11 15:39:08 +1100291 size_t size = bp->b_count_desired;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 size_t nbytes, offset;
Nathan Scottce8e9222006-01-11 15:39:08 +1100293 gfp_t gfp_mask = xb_to_gfp(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 unsigned short page_count, i;
Nathan Scott204ab252006-01-11 20:50:22 +1100295 xfs_off_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 int error;
297
Dave Chinner0e6e8472011-03-26 09:16:45 +1100298 /*
299 * for buffers that are contained within a single page, just allocate
300 * the memory from the heap - there's no need for the complexity of
301 * page arrays to keep allocation down to order 0.
302 */
303 if (bp->b_buffer_length < PAGE_SIZE) {
304 bp->b_addr = kmem_alloc(bp->b_buffer_length, xb_to_km(flags));
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 + bp->b_buffer_length - 1) &
311 PAGE_MASK) !=
312 ((unsigned long)bp->b_addr & PAGE_MASK)) {
313 /* b_addr spans two pages - use alloc_page instead */
314 kmem_free(bp->b_addr);
315 bp->b_addr = NULL;
316 goto use_alloc_page;
317 }
318 bp->b_offset = offset_in_page(bp->b_addr);
319 bp->b_pages = bp->b_page_array;
320 bp->b_pages[0] = virt_to_page(bp->b_addr);
321 bp->b_page_count = 1;
322 bp->b_flags |= XBF_MAPPED | _XBF_KMEM;
323 return 0;
324 }
325
326use_alloc_page:
Nathan Scottce8e9222006-01-11 15:39:08 +1100327 end = bp->b_file_offset + bp->b_buffer_length;
328 page_count = xfs_buf_btoc(end) - xfs_buf_btoct(bp->b_file_offset);
Nathan Scottce8e9222006-01-11 15:39:08 +1100329 error = _xfs_buf_get_pages(bp, page_count, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 if (unlikely(error))
331 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332
Nathan Scottce8e9222006-01-11 15:39:08 +1100333 offset = bp->b_offset;
Dave Chinner0e6e8472011-03-26 09:16:45 +1100334 bp->b_flags |= _XBF_PAGES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335
Nathan Scottce8e9222006-01-11 15:39:08 +1100336 for (i = 0; i < bp->b_page_count; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 struct page *page;
338 uint retries = 0;
Dave Chinner0e6e8472011-03-26 09:16:45 +1100339retry:
340 page = alloc_page(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 if (unlikely(page == NULL)) {
Nathan Scottce8e9222006-01-11 15:39:08 +1100342 if (flags & XBF_READ_AHEAD) {
343 bp->b_page_count = i;
Dave Chinner0e6e8472011-03-26 09:16:45 +1100344 error = ENOMEM;
345 goto out_free_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 }
347
348 /*
349 * This could deadlock.
350 *
351 * But until all the XFS lowlevel code is revamped to
352 * handle buffer allocation failures we can't do much.
353 */
354 if (!(++retries % 100))
Dave Chinner4f107002011-03-07 10:00:35 +1100355 xfs_err(NULL,
356 "possible memory allocation deadlock in %s (mode:0x%x)",
Harvey Harrison34a622b2008-04-10 12:19:21 +1000357 __func__, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358
Nathan Scottce8e9222006-01-11 15:39:08 +1100359 XFS_STATS_INC(xb_page_retries);
Jens Axboe8aa7e842009-07-09 14:52:32 +0200360 congestion_wait(BLK_RW_ASYNC, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 goto retry;
362 }
363
Nathan Scottce8e9222006-01-11 15:39:08 +1100364 XFS_STATS_INC(xb_page_found);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365
Dave Chinner0e6e8472011-03-26 09:16:45 +1100366 nbytes = min_t(size_t, size, PAGE_SIZE - offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 size -= nbytes;
Nathan Scottce8e9222006-01-11 15:39:08 +1100368 bp->b_pages[i] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 offset = 0;
370 }
Dave Chinner0e6e8472011-03-26 09:16:45 +1100371 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372
Dave Chinner0e6e8472011-03-26 09:16:45 +1100373out_free_pages:
374 for (i = 0; i < bp->b_page_count; i++)
375 __free_page(bp->b_pages[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 return error;
377}
378
379/*
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300380 * Map buffer into kernel address-space if necessary.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 */
382STATIC int
Nathan Scottce8e9222006-01-11 15:39:08 +1100383_xfs_buf_map_pages(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 xfs_buf_t *bp,
385 uint flags)
386{
Dave Chinner0e6e8472011-03-26 09:16:45 +1100387 ASSERT(bp->b_flags & _XBF_PAGES);
Nathan Scottce8e9222006-01-11 15:39:08 +1100388 if (bp->b_page_count == 1) {
Dave Chinner0e6e8472011-03-26 09:16:45 +1100389 /* A single page buffer is always mappable */
Nathan Scottce8e9222006-01-11 15:39:08 +1100390 bp->b_addr = page_address(bp->b_pages[0]) + bp->b_offset;
391 bp->b_flags |= XBF_MAPPED;
392 } else if (flags & XBF_MAPPED) {
Dave Chinnera19fb382011-03-26 09:13:42 +1100393 int retried = 0;
394
395 do {
396 bp->b_addr = vm_map_ram(bp->b_pages, bp->b_page_count,
397 -1, PAGE_KERNEL);
398 if (bp->b_addr)
399 break;
400 vm_unmap_aliases();
401 } while (retried++ <= 1);
402
403 if (!bp->b_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 return -ENOMEM;
Nathan Scottce8e9222006-01-11 15:39:08 +1100405 bp->b_addr += bp->b_offset;
406 bp->b_flags |= XBF_MAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 }
408
409 return 0;
410}
411
412/*
413 * Finding and Reading Buffers
414 */
415
416/*
Nathan Scottce8e9222006-01-11 15:39:08 +1100417 * Look up, and creates if absent, a lockable buffer for
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 * a given range of an inode. The buffer is returned
419 * locked. If other overlapping buffers exist, they are
420 * released before the new buffer is created and locked,
421 * which may imply that this call will block until those buffers
422 * are unlocked. No I/O is implied by this call.
423 */
424xfs_buf_t *
Nathan Scottce8e9222006-01-11 15:39:08 +1100425_xfs_buf_find(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 xfs_buftarg_t *btp, /* block device target */
Nathan Scott204ab252006-01-11 20:50:22 +1100427 xfs_off_t ioff, /* starting offset of range */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 size_t isize, /* length of range */
Nathan Scottce8e9222006-01-11 15:39:08 +1100429 xfs_buf_flags_t flags,
430 xfs_buf_t *new_bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431{
Nathan Scott204ab252006-01-11 20:50:22 +1100432 xfs_off_t range_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 size_t range_length;
Dave Chinner74f75a02010-09-24 19:59:04 +1000434 struct xfs_perag *pag;
435 struct rb_node **rbp;
436 struct rb_node *parent;
437 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
439 range_base = (ioff << BBSHIFT);
440 range_length = (isize << BBSHIFT);
441
442 /* Check for IOs smaller than the sector size / not sector aligned */
Nathan Scottce8e9222006-01-11 15:39:08 +1100443 ASSERT(!(range_length < (1 << btp->bt_sshift)));
Nathan Scott204ab252006-01-11 20:50:22 +1100444 ASSERT(!(range_base & (xfs_off_t)btp->bt_smask));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
Dave Chinner74f75a02010-09-24 19:59:04 +1000446 /* get tree root */
447 pag = xfs_perag_get(btp->bt_mount,
448 xfs_daddr_to_agno(btp->bt_mount, ioff));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449
Dave Chinner74f75a02010-09-24 19:59:04 +1000450 /* walk tree */
451 spin_lock(&pag->pag_buf_lock);
452 rbp = &pag->pag_buf_tree.rb_node;
453 parent = NULL;
454 bp = NULL;
455 while (*rbp) {
456 parent = *rbp;
457 bp = rb_entry(parent, struct xfs_buf, b_rbnode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458
Dave Chinner74f75a02010-09-24 19:59:04 +1000459 if (range_base < bp->b_file_offset)
460 rbp = &(*rbp)->rb_left;
461 else if (range_base > bp->b_file_offset)
462 rbp = &(*rbp)->rb_right;
463 else {
464 /*
465 * found a block offset match. If the range doesn't
466 * match, the only way this is allowed is if the buffer
467 * in the cache is stale and the transaction that made
468 * it stale has not yet committed. i.e. we are
469 * reallocating a busy extent. Skip this buffer and
470 * continue searching to the right for an exact match.
471 */
472 if (bp->b_buffer_length != range_length) {
473 ASSERT(bp->b_flags & XBF_STALE);
474 rbp = &(*rbp)->rb_right;
475 continue;
476 }
Nathan Scottce8e9222006-01-11 15:39:08 +1100477 atomic_inc(&bp->b_hold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 goto found;
479 }
480 }
481
482 /* No match found */
Nathan Scottce8e9222006-01-11 15:39:08 +1100483 if (new_bp) {
484 _xfs_buf_initialize(new_bp, btp, range_base,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 range_length, flags);
Dave Chinner74f75a02010-09-24 19:59:04 +1000486 rb_link_node(&new_bp->b_rbnode, parent, rbp);
487 rb_insert_color(&new_bp->b_rbnode, &pag->pag_buf_tree);
488 /* the buffer keeps the perag reference until it is freed */
489 new_bp->b_pag = pag;
490 spin_unlock(&pag->pag_buf_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 } else {
Nathan Scottce8e9222006-01-11 15:39:08 +1100492 XFS_STATS_INC(xb_miss_locked);
Dave Chinner74f75a02010-09-24 19:59:04 +1000493 spin_unlock(&pag->pag_buf_lock);
494 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 }
Nathan Scottce8e9222006-01-11 15:39:08 +1100496 return new_bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
498found:
Dave Chinner74f75a02010-09-24 19:59:04 +1000499 spin_unlock(&pag->pag_buf_lock);
500 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200502 if (!xfs_buf_trylock(bp)) {
503 if (flags & XBF_TRYLOCK) {
Nathan Scottce8e9222006-01-11 15:39:08 +1100504 xfs_buf_rele(bp);
505 XFS_STATS_INC(xb_busy_locked);
506 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 }
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200508 xfs_buf_lock(bp);
509 XFS_STATS_INC(xb_get_locked_waited);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 }
511
Dave Chinner0e6e8472011-03-26 09:16:45 +1100512 /*
513 * if the buffer is stale, clear all the external state associated with
514 * it. We need to keep flags such as how we allocated the buffer memory
515 * intact here.
516 */
Nathan Scottce8e9222006-01-11 15:39:08 +1100517 if (bp->b_flags & XBF_STALE) {
518 ASSERT((bp->b_flags & _XBF_DELWRI_Q) == 0);
Dave Chinner0e6e8472011-03-26 09:16:45 +1100519 bp->b_flags &= XBF_MAPPED | _XBF_KMEM | _XBF_PAGES;
David Chinner2f926582005-09-05 08:33:35 +1000520 }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000521
522 trace_xfs_buf_find(bp, flags, _RET_IP_);
Nathan Scottce8e9222006-01-11 15:39:08 +1100523 XFS_STATS_INC(xb_get_locked);
524 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525}
526
527/*
Nathan Scottce8e9222006-01-11 15:39:08 +1100528 * Assembles a buffer covering the specified range.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 * Storage in memory for all portions of the buffer will be allocated,
530 * although backing storage may not be.
531 */
532xfs_buf_t *
Christoph Hellwig6ad112b2009-11-24 18:02:23 +0000533xfs_buf_get(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 xfs_buftarg_t *target,/* target for buffer */
Nathan Scott204ab252006-01-11 20:50:22 +1100535 xfs_off_t ioff, /* starting offset of range */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 size_t isize, /* length of range */
Nathan Scottce8e9222006-01-11 15:39:08 +1100537 xfs_buf_flags_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538{
Nathan Scottce8e9222006-01-11 15:39:08 +1100539 xfs_buf_t *bp, *new_bp;
Dave Chinner0e6e8472011-03-26 09:16:45 +1100540 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
Nathan Scottce8e9222006-01-11 15:39:08 +1100542 new_bp = xfs_buf_allocate(flags);
543 if (unlikely(!new_bp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 return NULL;
545
Nathan Scottce8e9222006-01-11 15:39:08 +1100546 bp = _xfs_buf_find(target, ioff, isize, flags, new_bp);
547 if (bp == new_bp) {
Dave Chinner0e6e8472011-03-26 09:16:45 +1100548 error = xfs_buf_allocate_memory(bp, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 if (error)
550 goto no_buffer;
551 } else {
Nathan Scottce8e9222006-01-11 15:39:08 +1100552 xfs_buf_deallocate(new_bp);
553 if (unlikely(bp == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 return NULL;
555 }
556
Nathan Scottce8e9222006-01-11 15:39:08 +1100557 if (!(bp->b_flags & XBF_MAPPED)) {
558 error = _xfs_buf_map_pages(bp, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 if (unlikely(error)) {
Dave Chinner4f107002011-03-07 10:00:35 +1100560 xfs_warn(target->bt_mount,
561 "%s: failed to map pages\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 goto no_buffer;
563 }
564 }
565
Nathan Scottce8e9222006-01-11 15:39:08 +1100566 XFS_STATS_INC(xb_get);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
568 /*
569 * Always fill in the block number now, the mapped cases can do
570 * their own overlay of this later.
571 */
Nathan Scottce8e9222006-01-11 15:39:08 +1100572 bp->b_bn = ioff;
573 bp->b_count_desired = bp->b_buffer_length;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000575 trace_xfs_buf_get(bp, flags, _RET_IP_);
Nathan Scottce8e9222006-01-11 15:39:08 +1100576 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
578 no_buffer:
Nathan Scottce8e9222006-01-11 15:39:08 +1100579 if (flags & (XBF_LOCK | XBF_TRYLOCK))
580 xfs_buf_unlock(bp);
581 xfs_buf_rele(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 return NULL;
583}
584
Christoph Hellwig5d765b92008-12-03 12:20:26 +0100585STATIC int
586_xfs_buf_read(
587 xfs_buf_t *bp,
588 xfs_buf_flags_t flags)
589{
590 int status;
591
Christoph Hellwig5d765b92008-12-03 12:20:26 +0100592 ASSERT(!(flags & (XBF_DELWRI|XBF_WRITE)));
593 ASSERT(bp->b_bn != XFS_BUF_DADDR_NULL);
594
Christoph Hellwig1d5ae5d2011-07-08 14:36:32 +0200595 bp->b_flags &= ~(XBF_WRITE | XBF_ASYNC | XBF_DELWRI | XBF_READ_AHEAD);
596 bp->b_flags |= flags & (XBF_READ | XBF_ASYNC | XBF_READ_AHEAD);
Christoph Hellwig5d765b92008-12-03 12:20:26 +0100597
598 status = xfs_buf_iorequest(bp);
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +0000599 if (status || bp->b_error || (flags & XBF_ASYNC))
Dave Chinnerec53d1d2010-07-20 17:52:59 +1000600 return status;
601 return xfs_buf_iowait(bp);
Christoph Hellwig5d765b92008-12-03 12:20:26 +0100602}
603
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604xfs_buf_t *
Christoph Hellwig6ad112b2009-11-24 18:02:23 +0000605xfs_buf_read(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 xfs_buftarg_t *target,
Nathan Scott204ab252006-01-11 20:50:22 +1100607 xfs_off_t ioff,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 size_t isize,
Nathan Scottce8e9222006-01-11 15:39:08 +1100609 xfs_buf_flags_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610{
Nathan Scottce8e9222006-01-11 15:39:08 +1100611 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
Nathan Scottce8e9222006-01-11 15:39:08 +1100613 flags |= XBF_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
Christoph Hellwig6ad112b2009-11-24 18:02:23 +0000615 bp = xfs_buf_get(target, ioff, isize, flags);
Nathan Scottce8e9222006-01-11 15:39:08 +1100616 if (bp) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000617 trace_xfs_buf_read(bp, flags, _RET_IP_);
618
Nathan Scottce8e9222006-01-11 15:39:08 +1100619 if (!XFS_BUF_ISDONE(bp)) {
Nathan Scottce8e9222006-01-11 15:39:08 +1100620 XFS_STATS_INC(xb_get_read);
Christoph Hellwig5d765b92008-12-03 12:20:26 +0100621 _xfs_buf_read(bp, flags);
Nathan Scottce8e9222006-01-11 15:39:08 +1100622 } else if (flags & XBF_ASYNC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 /*
624 * Read ahead call which is already satisfied,
625 * drop the buffer
626 */
627 goto no_buffer;
628 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 /* We do not want read in the flags */
Nathan Scottce8e9222006-01-11 15:39:08 +1100630 bp->b_flags &= ~XBF_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 }
632 }
633
Nathan Scottce8e9222006-01-11 15:39:08 +1100634 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635
636 no_buffer:
Nathan Scottce8e9222006-01-11 15:39:08 +1100637 if (flags & (XBF_LOCK | XBF_TRYLOCK))
638 xfs_buf_unlock(bp);
639 xfs_buf_rele(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 return NULL;
641}
642
643/*
Nathan Scottce8e9222006-01-11 15:39:08 +1100644 * If we are not low on memory then do the readahead in a deadlock
645 * safe manner.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 */
647void
Nathan Scottce8e9222006-01-11 15:39:08 +1100648xfs_buf_readahead(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 xfs_buftarg_t *target,
Nathan Scott204ab252006-01-11 20:50:22 +1100650 xfs_off_t ioff,
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +0000651 size_t isize)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652{
Dave Chinner0e6e8472011-03-26 09:16:45 +1100653 if (bdi_read_congested(target->bt_bdi))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 return;
655
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +0000656 xfs_buf_read(target, ioff, isize,
657 XBF_TRYLOCK|XBF_ASYNC|XBF_READ_AHEAD|XBF_DONT_BLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658}
659
Dave Chinner5adc94c2010-09-24 21:58:31 +1000660/*
661 * Read an uncached buffer from disk. Allocates and returns a locked
662 * buffer containing the disk contents or nothing.
663 */
664struct xfs_buf *
665xfs_buf_read_uncached(
666 struct xfs_mount *mp,
667 struct xfs_buftarg *target,
668 xfs_daddr_t daddr,
669 size_t length,
670 int flags)
671{
672 xfs_buf_t *bp;
673 int error;
674
675 bp = xfs_buf_get_uncached(target, length, flags);
676 if (!bp)
677 return NULL;
678
679 /* set up the buffer for a read IO */
Dave Chinner5adc94c2010-09-24 21:58:31 +1000680 XFS_BUF_SET_ADDR(bp, daddr);
681 XFS_BUF_READ(bp);
682 XFS_BUF_BUSY(bp);
683
684 xfsbdstrat(mp, bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +0000685 error = xfs_buf_iowait(bp);
Dave Chinner5adc94c2010-09-24 21:58:31 +1000686 if (error || bp->b_error) {
687 xfs_buf_relse(bp);
688 return NULL;
689 }
690 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691}
692
693xfs_buf_t *
Nathan Scottce8e9222006-01-11 15:39:08 +1100694xfs_buf_get_empty(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 size_t len,
696 xfs_buftarg_t *target)
697{
Nathan Scottce8e9222006-01-11 15:39:08 +1100698 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699
Nathan Scottce8e9222006-01-11 15:39:08 +1100700 bp = xfs_buf_allocate(0);
701 if (bp)
702 _xfs_buf_initialize(bp, target, 0, len, 0);
703 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704}
705
Dave Chinner44396472011-04-21 09:34:27 +0000706/*
707 * Return a buffer allocated as an empty buffer and associated to external
708 * memory via xfs_buf_associate_memory() back to it's empty state.
709 */
710void
711xfs_buf_set_empty(
712 struct xfs_buf *bp,
713 size_t len)
714{
715 if (bp->b_pages)
716 _xfs_buf_free_pages(bp);
717
718 bp->b_pages = NULL;
719 bp->b_page_count = 0;
720 bp->b_addr = NULL;
721 bp->b_file_offset = 0;
722 bp->b_buffer_length = bp->b_count_desired = len;
723 bp->b_bn = XFS_BUF_DADDR_NULL;
724 bp->b_flags &= ~XBF_MAPPED;
725}
726
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727static inline struct page *
728mem_to_page(
729 void *addr)
730{
Christoph Lameter9e2779f2008-02-04 22:28:34 -0800731 if ((!is_vmalloc_addr(addr))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 return virt_to_page(addr);
733 } else {
734 return vmalloc_to_page(addr);
735 }
736}
737
738int
Nathan Scottce8e9222006-01-11 15:39:08 +1100739xfs_buf_associate_memory(
740 xfs_buf_t *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 void *mem,
742 size_t len)
743{
744 int rval;
745 int i = 0;
Lachlan McIlroyd1afb672007-11-27 17:01:24 +1100746 unsigned long pageaddr;
747 unsigned long offset;
748 size_t buflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 int page_count;
750
Dave Chinner0e6e8472011-03-26 09:16:45 +1100751 pageaddr = (unsigned long)mem & PAGE_MASK;
Lachlan McIlroyd1afb672007-11-27 17:01:24 +1100752 offset = (unsigned long)mem - pageaddr;
Dave Chinner0e6e8472011-03-26 09:16:45 +1100753 buflen = PAGE_ALIGN(len + offset);
754 page_count = buflen >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755
756 /* Free any previous set of page pointers */
Nathan Scottce8e9222006-01-11 15:39:08 +1100757 if (bp->b_pages)
758 _xfs_buf_free_pages(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759
Nathan Scottce8e9222006-01-11 15:39:08 +1100760 bp->b_pages = NULL;
761 bp->b_addr = mem;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762
Christoph Hellwig36fae172009-07-18 18:14:58 -0400763 rval = _xfs_buf_get_pages(bp, page_count, XBF_DONT_BLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 if (rval)
765 return rval;
766
Nathan Scottce8e9222006-01-11 15:39:08 +1100767 bp->b_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768
Lachlan McIlroyd1afb672007-11-27 17:01:24 +1100769 for (i = 0; i < bp->b_page_count; i++) {
770 bp->b_pages[i] = mem_to_page((void *)pageaddr);
Dave Chinner0e6e8472011-03-26 09:16:45 +1100771 pageaddr += PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773
Lachlan McIlroyd1afb672007-11-27 17:01:24 +1100774 bp->b_count_desired = len;
775 bp->b_buffer_length = buflen;
Nathan Scottce8e9222006-01-11 15:39:08 +1100776 bp->b_flags |= XBF_MAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
778 return 0;
779}
780
781xfs_buf_t *
Dave Chinner686865f2010-09-24 20:07:47 +1000782xfs_buf_get_uncached(
783 struct xfs_buftarg *target,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 size_t len,
Dave Chinner686865f2010-09-24 20:07:47 +1000785 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786{
Christoph Hellwig1fa40b02007-05-14 18:23:50 +1000787 unsigned long page_count = PAGE_ALIGN(len) >> PAGE_SHIFT;
788 int error, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
Nathan Scottce8e9222006-01-11 15:39:08 +1100791 bp = xfs_buf_allocate(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 if (unlikely(bp == NULL))
793 goto fail;
Nathan Scottce8e9222006-01-11 15:39:08 +1100794 _xfs_buf_initialize(bp, target, 0, len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
Christoph Hellwig1fa40b02007-05-14 18:23:50 +1000796 error = _xfs_buf_get_pages(bp, page_count, 0);
797 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 goto fail_free_buf;
799
Christoph Hellwig1fa40b02007-05-14 18:23:50 +1000800 for (i = 0; i < page_count; i++) {
Dave Chinner686865f2010-09-24 20:07:47 +1000801 bp->b_pages[i] = alloc_page(xb_to_gfp(flags));
Christoph Hellwig1fa40b02007-05-14 18:23:50 +1000802 if (!bp->b_pages[i])
803 goto fail_free_mem;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 }
Christoph Hellwig1fa40b02007-05-14 18:23:50 +1000805 bp->b_flags |= _XBF_PAGES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806
Christoph Hellwig1fa40b02007-05-14 18:23:50 +1000807 error = _xfs_buf_map_pages(bp, XBF_MAPPED);
808 if (unlikely(error)) {
Dave Chinner4f107002011-03-07 10:00:35 +1100809 xfs_warn(target->bt_mount,
810 "%s: failed to map pages\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 goto fail_free_mem;
Christoph Hellwig1fa40b02007-05-14 18:23:50 +1000812 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813
Dave Chinner686865f2010-09-24 20:07:47 +1000814 trace_xfs_buf_get_uncached(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 return bp;
Christoph Hellwig1fa40b02007-05-14 18:23:50 +1000816
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 fail_free_mem:
Christoph Hellwig1fa40b02007-05-14 18:23:50 +1000818 while (--i >= 0)
819 __free_page(bp->b_pages[i]);
Christoph Hellwigca165b82007-05-24 15:21:11 +1000820 _xfs_buf_free_pages(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 fail_free_buf:
Christoph Hellwigca165b82007-05-24 15:21:11 +1000822 xfs_buf_deallocate(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 fail:
824 return NULL;
825}
826
827/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 * Increment reference count on buffer, to hold the buffer concurrently
829 * with another thread which may release (free) the buffer asynchronously.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 * Must hold the buffer already to call this function.
831 */
832void
Nathan Scottce8e9222006-01-11 15:39:08 +1100833xfs_buf_hold(
834 xfs_buf_t *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000836 trace_xfs_buf_hold(bp, _RET_IP_);
Nathan Scottce8e9222006-01-11 15:39:08 +1100837 atomic_inc(&bp->b_hold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838}
839
840/*
Nathan Scottce8e9222006-01-11 15:39:08 +1100841 * Releases a hold on the specified buffer. If the
842 * the hold count is 1, calls xfs_buf_free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 */
844void
Nathan Scottce8e9222006-01-11 15:39:08 +1100845xfs_buf_rele(
846 xfs_buf_t *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847{
Dave Chinner74f75a02010-09-24 19:59:04 +1000848 struct xfs_perag *pag = bp->b_pag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000850 trace_xfs_buf_rele(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851
Dave Chinner74f75a02010-09-24 19:59:04 +1000852 if (!pag) {
Dave Chinner430cbeb2010-12-02 16:30:55 +1100853 ASSERT(list_empty(&bp->b_lru));
Dave Chinner74f75a02010-09-24 19:59:04 +1000854 ASSERT(RB_EMPTY_NODE(&bp->b_rbnode));
Nathan Scottfad3aa12006-02-01 12:14:52 +1100855 if (atomic_dec_and_test(&bp->b_hold))
856 xfs_buf_free(bp);
857 return;
858 }
859
Dave Chinner74f75a02010-09-24 19:59:04 +1000860 ASSERT(!RB_EMPTY_NODE(&bp->b_rbnode));
Dave Chinner430cbeb2010-12-02 16:30:55 +1100861
Lachlan McIlroy37906892008-08-13 15:42:10 +1000862 ASSERT(atomic_read(&bp->b_hold) > 0);
Dave Chinner74f75a02010-09-24 19:59:04 +1000863 if (atomic_dec_and_lock(&bp->b_hold, &pag->pag_buf_lock)) {
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000864 if (!(bp->b_flags & XBF_STALE) &&
Dave Chinner430cbeb2010-12-02 16:30:55 +1100865 atomic_read(&bp->b_lru_ref)) {
866 xfs_buf_lru_add(bp);
867 spin_unlock(&pag->pag_buf_lock);
Christoph Hellwig7f14d0a2005-11-02 15:09:35 +1100868 } else {
Dave Chinner430cbeb2010-12-02 16:30:55 +1100869 xfs_buf_lru_del(bp);
Nathan Scottce8e9222006-01-11 15:39:08 +1100870 ASSERT(!(bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q)));
Dave Chinner74f75a02010-09-24 19:59:04 +1000871 rb_erase(&bp->b_rbnode, &pag->pag_buf_tree);
872 spin_unlock(&pag->pag_buf_lock);
873 xfs_perag_put(pag);
Nathan Scottce8e9222006-01-11 15:39:08 +1100874 xfs_buf_free(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 }
876 }
877}
878
879
880/*
Dave Chinner0e6e8472011-03-26 09:16:45 +1100881 * Lock a buffer object, if it is not already locked.
Dave Chinner90810b92010-11-30 15:16:16 +1100882 *
883 * If we come across a stale, pinned, locked buffer, we know that we are
884 * being asked to lock a buffer that has been reallocated. Because it is
885 * pinned, we know that the log has not been pushed to disk and hence it
886 * will still be locked. Rather than continuing to have trylock attempts
887 * fail until someone else pushes the log, push it ourselves before
888 * returning. This means that the xfsaild will not get stuck trying
889 * to push on stale inode buffers.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 */
891int
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200892xfs_buf_trylock(
893 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894{
895 int locked;
896
Nathan Scottce8e9222006-01-11 15:39:08 +1100897 locked = down_trylock(&bp->b_sema) == 0;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000898 if (locked)
Nathan Scottce8e9222006-01-11 15:39:08 +1100899 XB_SET_OWNER(bp);
Dave Chinner90810b92010-11-30 15:16:16 +1100900 else if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))
901 xfs_log_force(bp->b_target->bt_mount, 0);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000902
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200903 trace_xfs_buf_trylock(bp, _RET_IP_);
904 return locked;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906
907/*
Dave Chinner0e6e8472011-03-26 09:16:45 +1100908 * Lock a buffer object.
Dave Chinnered3b4d62010-05-21 12:07:08 +1000909 *
910 * If we come across a stale, pinned, locked buffer, we know that we
911 * are being asked to lock a buffer that has been reallocated. Because
912 * it is pinned, we know that the log has not been pushed to disk and
913 * hence it will still be locked. Rather than sleeping until someone
914 * else pushes the log, push it ourselves before trying to get the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 */
Nathan Scottce8e9222006-01-11 15:39:08 +1100916void
917xfs_buf_lock(
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200918 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000920 trace_xfs_buf_lock(bp, _RET_IP_);
921
Dave Chinnered3b4d62010-05-21 12:07:08 +1000922 if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))
Dave Chinnerebad8612010-09-22 10:47:20 +1000923 xfs_log_force(bp->b_target->bt_mount, 0);
Nathan Scottce8e9222006-01-11 15:39:08 +1100924 down(&bp->b_sema);
925 XB_SET_OWNER(bp);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000926
927 trace_xfs_buf_lock_done(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928}
929
930/*
Nathan Scottce8e9222006-01-11 15:39:08 +1100931 * Releases the lock on the buffer object.
David Chinner2f926582005-09-05 08:33:35 +1000932 * If the buffer is marked delwri but is not queued, do so before we
Nathan Scottce8e9222006-01-11 15:39:08 +1100933 * unlock the buffer as we need to set flags correctly. We also need to
David Chinner2f926582005-09-05 08:33:35 +1000934 * take a reference for the delwri queue because the unlocker is going to
935 * drop their's and they don't know we just queued it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 */
937void
Nathan Scottce8e9222006-01-11 15:39:08 +1100938xfs_buf_unlock(
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200939 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940{
Nathan Scottce8e9222006-01-11 15:39:08 +1100941 if ((bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q)) == XBF_DELWRI) {
942 atomic_inc(&bp->b_hold);
943 bp->b_flags |= XBF_ASYNC;
944 xfs_buf_delwri_queue(bp, 0);
David Chinner2f926582005-09-05 08:33:35 +1000945 }
946
Nathan Scottce8e9222006-01-11 15:39:08 +1100947 XB_CLEAR_OWNER(bp);
948 up(&bp->b_sema);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000949
950 trace_xfs_buf_unlock(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951}
952
Nathan Scottce8e9222006-01-11 15:39:08 +1100953STATIC void
954xfs_buf_wait_unpin(
955 xfs_buf_t *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956{
957 DECLARE_WAITQUEUE (wait, current);
958
Nathan Scottce8e9222006-01-11 15:39:08 +1100959 if (atomic_read(&bp->b_pin_count) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 return;
961
Nathan Scottce8e9222006-01-11 15:39:08 +1100962 add_wait_queue(&bp->b_waiters, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 for (;;) {
964 set_current_state(TASK_UNINTERRUPTIBLE);
Nathan Scottce8e9222006-01-11 15:39:08 +1100965 if (atomic_read(&bp->b_pin_count) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 break;
Jens Axboe7eaceac2011-03-10 08:52:07 +0100967 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 }
Nathan Scottce8e9222006-01-11 15:39:08 +1100969 remove_wait_queue(&bp->b_waiters, &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 set_current_state(TASK_RUNNING);
971}
972
973/*
974 * Buffer Utility Routines
975 */
976
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977STATIC void
Nathan Scottce8e9222006-01-11 15:39:08 +1100978xfs_buf_iodone_work(
David Howellsc4028952006-11-22 14:57:56 +0000979 struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980{
David Howellsc4028952006-11-22 14:57:56 +0000981 xfs_buf_t *bp =
982 container_of(work, xfs_buf_t, b_iodone_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983
Christoph Hellwig80f6c292010-08-18 05:29:11 -0400984 if (bp->b_iodone)
Nathan Scottce8e9222006-01-11 15:39:08 +1100985 (*(bp->b_iodone))(bp);
986 else if (bp->b_flags & XBF_ASYNC)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 xfs_buf_relse(bp);
988}
989
990void
Nathan Scottce8e9222006-01-11 15:39:08 +1100991xfs_buf_ioend(
992 xfs_buf_t *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 int schedule)
994{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000995 trace_xfs_buf_iodone(bp, _RET_IP_);
996
Lachlan McIlroy77be55a2007-11-23 16:31:00 +1100997 bp->b_flags &= ~(XBF_READ | XBF_WRITE | XBF_READ_AHEAD);
Nathan Scottce8e9222006-01-11 15:39:08 +1100998 if (bp->b_error == 0)
999 bp->b_flags |= XBF_DONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
Nathan Scottce8e9222006-01-11 15:39:08 +11001001 if ((bp->b_iodone) || (bp->b_flags & XBF_ASYNC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 if (schedule) {
David Howellsc4028952006-11-22 14:57:56 +00001003 INIT_WORK(&bp->b_iodone_work, xfs_buf_iodone_work);
Nathan Scottce8e9222006-01-11 15:39:08 +11001004 queue_work(xfslogd_workqueue, &bp->b_iodone_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 } else {
David Howellsc4028952006-11-22 14:57:56 +00001006 xfs_buf_iodone_work(&bp->b_iodone_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 }
1008 } else {
David Chinnerb4dd3302008-08-13 16:36:11 +10001009 complete(&bp->b_iowait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 }
1011}
1012
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013void
Nathan Scottce8e9222006-01-11 15:39:08 +11001014xfs_buf_ioerror(
1015 xfs_buf_t *bp,
1016 int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017{
1018 ASSERT(error >= 0 && error <= 0xffff);
Nathan Scottce8e9222006-01-11 15:39:08 +11001019 bp->b_error = (unsigned short)error;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001020 trace_xfs_buf_ioerror(bp, error, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021}
1022
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023int
Christoph Hellwig64e0bc72010-01-13 22:17:58 +00001024xfs_bwrite(
1025 struct xfs_mount *mp,
Christoph Hellwig5d765b92008-12-03 12:20:26 +01001026 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027{
Christoph Hellwig8c383662010-03-12 10:59:40 +00001028 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029
Christoph Hellwig64e0bc72010-01-13 22:17:58 +00001030 bp->b_flags |= XBF_WRITE;
Christoph Hellwig8c383662010-03-12 10:59:40 +00001031 bp->b_flags &= ~(XBF_ASYNC | XBF_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032
Christoph Hellwig5d765b92008-12-03 12:20:26 +01001033 xfs_buf_delwri_dequeue(bp);
Christoph Hellwig939d7232010-07-20 17:51:16 +10001034 xfs_bdstrat_cb(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035
Christoph Hellwig8c383662010-03-12 10:59:40 +00001036 error = xfs_buf_iowait(bp);
1037 if (error)
1038 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1039 xfs_buf_relse(bp);
Christoph Hellwig64e0bc72010-01-13 22:17:58 +00001040 return error;
Christoph Hellwig5d765b92008-12-03 12:20:26 +01001041}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042
Christoph Hellwig5d765b92008-12-03 12:20:26 +01001043void
1044xfs_bdwrite(
1045 void *mp,
1046 struct xfs_buf *bp)
1047{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001048 trace_xfs_buf_bdwrite(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049
Christoph Hellwig5d765b92008-12-03 12:20:26 +01001050 bp->b_flags &= ~XBF_READ;
1051 bp->b_flags |= (XBF_DELWRI | XBF_ASYNC);
1052
1053 xfs_buf_delwri_queue(bp, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054}
1055
Christoph Hellwig4e234712010-01-13 22:17:56 +00001056/*
1057 * Called when we want to stop a buffer from getting written or read.
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00001058 * We attach the EIO error, muck with its flags, and call xfs_buf_ioend
Christoph Hellwig4e234712010-01-13 22:17:56 +00001059 * so that the proper iodone callbacks get called.
1060 */
1061STATIC int
1062xfs_bioerror(
1063 xfs_buf_t *bp)
1064{
1065#ifdef XFSERRORDEBUG
1066 ASSERT(XFS_BUF_ISREAD(bp) || bp->b_iodone);
1067#endif
1068
1069 /*
1070 * No need to wait until the buffer is unpinned, we aren't flushing it.
1071 */
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00001072 xfs_buf_ioerror(bp, EIO);
Christoph Hellwig4e234712010-01-13 22:17:56 +00001073
1074 /*
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00001075 * We're calling xfs_buf_ioend, so delete XBF_DONE flag.
Christoph Hellwig4e234712010-01-13 22:17:56 +00001076 */
1077 XFS_BUF_UNREAD(bp);
1078 XFS_BUF_UNDELAYWRITE(bp);
1079 XFS_BUF_UNDONE(bp);
1080 XFS_BUF_STALE(bp);
1081
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00001082 xfs_buf_ioend(bp, 0);
Christoph Hellwig4e234712010-01-13 22:17:56 +00001083
1084 return EIO;
1085}
1086
1087/*
1088 * Same as xfs_bioerror, except that we are releasing the buffer
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00001089 * here ourselves, and avoiding the xfs_buf_ioend call.
Christoph Hellwig4e234712010-01-13 22:17:56 +00001090 * This is meant for userdata errors; metadata bufs come with
1091 * iodone functions attached, so that we can track down errors.
1092 */
1093STATIC int
1094xfs_bioerror_relse(
1095 struct xfs_buf *bp)
1096{
Chandra Seetharamaned432332011-07-22 23:39:39 +00001097 int64_t fl = bp->b_flags;
Christoph Hellwig4e234712010-01-13 22:17:56 +00001098 /*
1099 * No need to wait until the buffer is unpinned.
1100 * We aren't flushing it.
1101 *
1102 * chunkhold expects B_DONE to be set, whether
1103 * we actually finish the I/O or not. We don't want to
1104 * change that interface.
1105 */
1106 XFS_BUF_UNREAD(bp);
1107 XFS_BUF_UNDELAYWRITE(bp);
1108 XFS_BUF_DONE(bp);
1109 XFS_BUF_STALE(bp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02001110 bp->b_iodone = NULL;
Christoph Hellwig0cadda12010-01-19 09:56:44 +00001111 if (!(fl & XBF_ASYNC)) {
Christoph Hellwig4e234712010-01-13 22:17:56 +00001112 /*
1113 * Mark b_error and B_ERROR _both_.
1114 * Lot's of chunkcache code assumes that.
1115 * There's no reason to mark error for
1116 * ASYNC buffers.
1117 */
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00001118 xfs_buf_ioerror(bp, EIO);
Christoph Hellwig4e234712010-01-13 22:17:56 +00001119 XFS_BUF_FINISH_IOWAIT(bp);
1120 } else {
1121 xfs_buf_relse(bp);
1122 }
1123
1124 return EIO;
1125}
1126
1127
1128/*
1129 * All xfs metadata buffers except log state machine buffers
1130 * get this attached as their b_bdstrat callback function.
1131 * This is so that we can catch a buffer
1132 * after prematurely unpinning it to forcibly shutdown the filesystem.
1133 */
1134int
1135xfs_bdstrat_cb(
1136 struct xfs_buf *bp)
1137{
Dave Chinnerebad8612010-09-22 10:47:20 +10001138 if (XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) {
Christoph Hellwig4e234712010-01-13 22:17:56 +00001139 trace_xfs_bdstrat_shut(bp, _RET_IP_);
1140 /*
1141 * Metadata write that didn't get logged but
1142 * written delayed anyway. These aren't associated
1143 * with a transaction, and can be ignored.
1144 */
1145 if (!bp->b_iodone && !XFS_BUF_ISREAD(bp))
1146 return xfs_bioerror_relse(bp);
1147 else
1148 return xfs_bioerror(bp);
1149 }
1150
1151 xfs_buf_iorequest(bp);
1152 return 0;
1153}
1154
1155/*
1156 * Wrapper around bdstrat so that we can stop data from going to disk in case
1157 * we are shutting down the filesystem. Typically user data goes thru this
1158 * path; one of the exceptions is the superblock.
1159 */
1160void
1161xfsbdstrat(
1162 struct xfs_mount *mp,
1163 struct xfs_buf *bp)
1164{
1165 if (XFS_FORCED_SHUTDOWN(mp)) {
1166 trace_xfs_bdstrat_shut(bp, _RET_IP_);
1167 xfs_bioerror_relse(bp);
1168 return;
1169 }
1170
1171 xfs_buf_iorequest(bp);
1172}
1173
Christoph Hellwigb8f82a42009-11-14 16:17:22 +00001174STATIC void
Nathan Scottce8e9222006-01-11 15:39:08 +11001175_xfs_buf_ioend(
1176 xfs_buf_t *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 int schedule)
1178{
Dave Chinner0e6e8472011-03-26 09:16:45 +11001179 if (atomic_dec_and_test(&bp->b_io_remaining) == 1)
Nathan Scottce8e9222006-01-11 15:39:08 +11001180 xfs_buf_ioend(bp, schedule);
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;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189
Lachlan McIlroycfbe5262008-12-12 15:27:25 +11001190 xfs_buf_ioerror(bp, -error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
James Bottomley73c77e22010-01-25 11:42:24 -06001192 if (!error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ))
1193 invalidate_kernel_vmap_range(bp->b_addr, xfs_buf_vmap_len(bp));
1194
Nathan Scottce8e9222006-01-11 15:39:08 +11001195 _xfs_buf_ioend(bp, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197}
1198
1199STATIC void
Nathan Scottce8e9222006-01-11 15:39:08 +11001200_xfs_buf_ioapply(
1201 xfs_buf_t *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202{
Christoph Hellwiga9759f22007-12-07 14:07:08 +11001203 int rw, map_i, total_nr_pages, nr_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 struct bio *bio;
Nathan Scottce8e9222006-01-11 15:39:08 +11001205 int offset = bp->b_offset;
1206 int size = bp->b_count_desired;
1207 sector_t sector = bp->b_bn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208
Nathan Scottce8e9222006-01-11 15:39:08 +11001209 total_nr_pages = bp->b_page_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 map_i = 0;
1211
Christoph Hellwig1d5ae5d2011-07-08 14:36:32 +02001212 if (bp->b_flags & XBF_WRITE) {
1213 if (bp->b_flags & XBF_SYNCIO)
1214 rw = WRITE_SYNC;
1215 else
1216 rw = WRITE;
1217 if (bp->b_flags & XBF_FUA)
1218 rw |= REQ_FUA;
1219 if (bp->b_flags & XBF_FLUSH)
1220 rw |= REQ_FLUSH;
1221 } else if (bp->b_flags & XBF_READ_AHEAD) {
1222 rw = READA;
Nathan Scott51bdd702006-09-28 11:01:57 +10001223 } else {
Christoph Hellwig1d5ae5d2011-07-08 14:36:32 +02001224 rw = READ;
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001225 }
1226
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227next_chunk:
Nathan Scottce8e9222006-01-11 15:39:08 +11001228 atomic_inc(&bp->b_io_remaining);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 nr_pages = BIO_MAX_SECTORS >> (PAGE_SHIFT - BBSHIFT);
1230 if (nr_pages > total_nr_pages)
1231 nr_pages = total_nr_pages;
1232
1233 bio = bio_alloc(GFP_NOIO, nr_pages);
Nathan Scottce8e9222006-01-11 15:39:08 +11001234 bio->bi_bdev = bp->b_target->bt_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 bio->bi_sector = sector;
Nathan Scottce8e9222006-01-11 15:39:08 +11001236 bio->bi_end_io = xfs_buf_bio_end_io;
1237 bio->bi_private = bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238
Dave Chinner0e6e8472011-03-26 09:16:45 +11001239
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 for (; size && nr_pages; nr_pages--, map_i++) {
Dave Chinner0e6e8472011-03-26 09:16:45 +11001241 int rbytes, nbytes = PAGE_SIZE - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242
1243 if (nbytes > size)
1244 nbytes = size;
1245
Nathan Scottce8e9222006-01-11 15:39:08 +11001246 rbytes = bio_add_page(bio, bp->b_pages[map_i], nbytes, offset);
1247 if (rbytes < nbytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 break;
1249
1250 offset = 0;
1251 sector += nbytes >> BBSHIFT;
1252 size -= nbytes;
1253 total_nr_pages--;
1254 }
1255
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 if (likely(bio->bi_size)) {
James Bottomley73c77e22010-01-25 11:42:24 -06001257 if (xfs_buf_is_vmapped(bp)) {
1258 flush_kernel_vmap_range(bp->b_addr,
1259 xfs_buf_vmap_len(bp));
1260 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 submit_bio(rw, bio);
1262 if (size)
1263 goto next_chunk;
1264 } else {
Nathan Scottce8e9222006-01-11 15:39:08 +11001265 xfs_buf_ioerror(bp, EIO);
Dave Chinnerec53d1d2010-07-20 17:52:59 +10001266 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 }
1268}
1269
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270int
Nathan Scottce8e9222006-01-11 15:39:08 +11001271xfs_buf_iorequest(
1272 xfs_buf_t *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001274 trace_xfs_buf_iorequest(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
Nathan Scottce8e9222006-01-11 15:39:08 +11001276 if (bp->b_flags & XBF_DELWRI) {
1277 xfs_buf_delwri_queue(bp, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 return 0;
1279 }
1280
Nathan Scottce8e9222006-01-11 15:39:08 +11001281 if (bp->b_flags & XBF_WRITE) {
1282 xfs_buf_wait_unpin(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 }
1284
Nathan Scottce8e9222006-01-11 15:39:08 +11001285 xfs_buf_hold(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286
1287 /* Set the count to 1 initially, this will stop an I/O
1288 * completion callout which happens before we have started
Nathan Scottce8e9222006-01-11 15:39:08 +11001289 * all the I/O from calling xfs_buf_ioend too early.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 */
Nathan Scottce8e9222006-01-11 15:39:08 +11001291 atomic_set(&bp->b_io_remaining, 1);
1292 _xfs_buf_ioapply(bp);
1293 _xfs_buf_ioend(bp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294
Nathan Scottce8e9222006-01-11 15:39:08 +11001295 xfs_buf_rele(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 return 0;
1297}
1298
1299/*
Nathan Scottce8e9222006-01-11 15:39:08 +11001300 * Waits for I/O to complete on the buffer supplied.
1301 * It returns immediately if no I/O is pending.
1302 * It returns the I/O error code, if any, or 0 if there was no error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 */
1304int
Nathan Scottce8e9222006-01-11 15:39:08 +11001305xfs_buf_iowait(
1306 xfs_buf_t *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001308 trace_xfs_buf_iowait(bp, _RET_IP_);
1309
David Chinnerb4dd3302008-08-13 16:36:11 +10001310 wait_for_completion(&bp->b_iowait);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001311
1312 trace_xfs_buf_iowait_done(bp, _RET_IP_);
Nathan Scottce8e9222006-01-11 15:39:08 +11001313 return bp->b_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314}
1315
Nathan Scottce8e9222006-01-11 15:39:08 +11001316xfs_caddr_t
1317xfs_buf_offset(
1318 xfs_buf_t *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 size_t offset)
1320{
1321 struct page *page;
1322
Nathan Scottce8e9222006-01-11 15:39:08 +11001323 if (bp->b_flags & XBF_MAPPED)
1324 return XFS_BUF_PTR(bp) + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325
Nathan Scottce8e9222006-01-11 15:39:08 +11001326 offset += bp->b_offset;
Dave Chinner0e6e8472011-03-26 09:16:45 +11001327 page = bp->b_pages[offset >> PAGE_SHIFT];
1328 return (xfs_caddr_t)page_address(page) + (offset & (PAGE_SIZE-1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329}
1330
1331/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 * Move data into or out of a buffer.
1333 */
1334void
Nathan Scottce8e9222006-01-11 15:39:08 +11001335xfs_buf_iomove(
1336 xfs_buf_t *bp, /* buffer to process */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 size_t boff, /* starting buffer offset */
1338 size_t bsize, /* length to copy */
Dave Chinnerb9c48642010-01-20 10:47:39 +11001339 void *data, /* data address */
Nathan Scottce8e9222006-01-11 15:39:08 +11001340 xfs_buf_rw_t mode) /* read/write/zero flag */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341{
1342 size_t bend, cpoff, csize;
1343 struct page *page;
1344
1345 bend = boff + bsize;
1346 while (boff < bend) {
Nathan Scottce8e9222006-01-11 15:39:08 +11001347 page = bp->b_pages[xfs_buf_btoct(boff + bp->b_offset)];
1348 cpoff = xfs_buf_poff(boff + bp->b_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 csize = min_t(size_t,
Dave Chinner0e6e8472011-03-26 09:16:45 +11001350 PAGE_SIZE-cpoff, bp->b_count_desired-boff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351
Dave Chinner0e6e8472011-03-26 09:16:45 +11001352 ASSERT(((csize + cpoff) <= PAGE_SIZE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353
1354 switch (mode) {
Nathan Scottce8e9222006-01-11 15:39:08 +11001355 case XBRW_ZERO:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 memset(page_address(page) + cpoff, 0, csize);
1357 break;
Nathan Scottce8e9222006-01-11 15:39:08 +11001358 case XBRW_READ:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 memcpy(data, page_address(page) + cpoff, csize);
1360 break;
Nathan Scottce8e9222006-01-11 15:39:08 +11001361 case XBRW_WRITE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 memcpy(page_address(page) + cpoff, data, csize);
1363 }
1364
1365 boff += csize;
1366 data += csize;
1367 }
1368}
1369
1370/*
Nathan Scottce8e9222006-01-11 15:39:08 +11001371 * Handling of buffer targets (buftargs).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 */
1373
1374/*
Dave Chinner430cbeb2010-12-02 16:30:55 +11001375 * Wait for any bufs with callbacks that have been submitted but have not yet
1376 * returned. These buffers will have an elevated hold count, so wait on those
1377 * while freeing all the buffers only held by the LRU.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 */
1379void
1380xfs_wait_buftarg(
Dave Chinner74f75a02010-09-24 19:59:04 +10001381 struct xfs_buftarg *btp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382{
Dave Chinner430cbeb2010-12-02 16:30:55 +11001383 struct xfs_buf *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384
Dave Chinner430cbeb2010-12-02 16:30:55 +11001385restart:
1386 spin_lock(&btp->bt_lru_lock);
1387 while (!list_empty(&btp->bt_lru)) {
1388 bp = list_first_entry(&btp->bt_lru, struct xfs_buf, b_lru);
1389 if (atomic_read(&bp->b_hold) > 1) {
1390 spin_unlock(&btp->bt_lru_lock);
Dave Chinner26af6552010-09-22 10:47:20 +10001391 delay(100);
Dave Chinner430cbeb2010-12-02 16:30:55 +11001392 goto restart;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 }
Dave Chinner430cbeb2010-12-02 16:30:55 +11001394 /*
1395 * clear the LRU reference count so the bufer doesn't get
1396 * ignored in xfs_buf_rele().
1397 */
1398 atomic_set(&bp->b_lru_ref, 0);
1399 spin_unlock(&btp->bt_lru_lock);
1400 xfs_buf_rele(bp);
1401 spin_lock(&btp->bt_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 }
Dave Chinner430cbeb2010-12-02 16:30:55 +11001403 spin_unlock(&btp->bt_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404}
1405
Dave Chinnerff57ab22010-11-30 17:27:57 +11001406int
1407xfs_buftarg_shrink(
1408 struct shrinker *shrink,
Ying Han1495f232011-05-24 17:12:27 -07001409 struct shrink_control *sc)
David Chinnera6867a62006-01-11 15:37:58 +11001410{
Dave Chinnerff57ab22010-11-30 17:27:57 +11001411 struct xfs_buftarg *btp = container_of(shrink,
1412 struct xfs_buftarg, bt_shrinker);
Dave Chinner430cbeb2010-12-02 16:30:55 +11001413 struct xfs_buf *bp;
Ying Han1495f232011-05-24 17:12:27 -07001414 int nr_to_scan = sc->nr_to_scan;
Dave Chinner430cbeb2010-12-02 16:30:55 +11001415 LIST_HEAD(dispose);
1416
1417 if (!nr_to_scan)
1418 return btp->bt_lru_nr;
1419
1420 spin_lock(&btp->bt_lru_lock);
1421 while (!list_empty(&btp->bt_lru)) {
1422 if (nr_to_scan-- <= 0)
1423 break;
1424
1425 bp = list_first_entry(&btp->bt_lru, struct xfs_buf, b_lru);
1426
1427 /*
1428 * Decrement the b_lru_ref count unless the value is already
1429 * zero. If the value is already zero, we need to reclaim the
1430 * buffer, otherwise it gets another trip through the LRU.
1431 */
1432 if (!atomic_add_unless(&bp->b_lru_ref, -1, 0)) {
1433 list_move_tail(&bp->b_lru, &btp->bt_lru);
1434 continue;
1435 }
1436
1437 /*
1438 * remove the buffer from the LRU now to avoid needing another
1439 * lock round trip inside xfs_buf_rele().
1440 */
1441 list_move(&bp->b_lru, &dispose);
1442 btp->bt_lru_nr--;
Dave Chinnerff57ab22010-11-30 17:27:57 +11001443 }
Dave Chinner430cbeb2010-12-02 16:30:55 +11001444 spin_unlock(&btp->bt_lru_lock);
1445
1446 while (!list_empty(&dispose)) {
1447 bp = list_first_entry(&dispose, struct xfs_buf, b_lru);
1448 list_del_init(&bp->b_lru);
1449 xfs_buf_rele(bp);
1450 }
1451
1452 return btp->bt_lru_nr;
David Chinnera6867a62006-01-11 15:37:58 +11001453}
1454
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455void
1456xfs_free_buftarg(
Christoph Hellwigb7963132009-03-03 14:48:37 -05001457 struct xfs_mount *mp,
1458 struct xfs_buftarg *btp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459{
Dave Chinnerff57ab22010-11-30 17:27:57 +11001460 unregister_shrinker(&btp->bt_shrinker);
1461
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 xfs_flush_buftarg(btp, 1);
Christoph Hellwigb7963132009-03-03 14:48:37 -05001463 if (mp->m_flags & XFS_MOUNT_BARRIER)
1464 xfs_blkdev_issue_flush(btp);
David Chinnera6867a62006-01-11 15:37:58 +11001465
David Chinnera6867a62006-01-11 15:37:58 +11001466 kthread_stop(btp->bt_task);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001467 kmem_free(btp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468}
1469
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470STATIC int
1471xfs_setsize_buftarg_flags(
1472 xfs_buftarg_t *btp,
1473 unsigned int blocksize,
1474 unsigned int sectorsize,
1475 int verbose)
1476{
Nathan Scottce8e9222006-01-11 15:39:08 +11001477 btp->bt_bsize = blocksize;
1478 btp->bt_sshift = ffs(sectorsize) - 1;
1479 btp->bt_smask = sectorsize - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480
Nathan Scottce8e9222006-01-11 15:39:08 +11001481 if (set_blocksize(btp->bt_bdev, sectorsize)) {
Dave Chinner4f107002011-03-07 10:00:35 +11001482 xfs_warn(btp->bt_mount,
1483 "Cannot set_blocksize to %u on device %s\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 sectorsize, XFS_BUFTARG_NAME(btp));
1485 return EINVAL;
1486 }
1487
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 return 0;
1489}
1490
1491/*
Nathan Scottce8e9222006-01-11 15:39:08 +11001492 * When allocating the initial buffer target we have not yet
1493 * read in the superblock, so don't know what sized sectors
1494 * are being used is at this early stage. Play safe.
1495 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496STATIC int
1497xfs_setsize_buftarg_early(
1498 xfs_buftarg_t *btp,
1499 struct block_device *bdev)
1500{
1501 return xfs_setsize_buftarg_flags(btp,
Dave Chinner0e6e8472011-03-26 09:16:45 +11001502 PAGE_SIZE, bdev_logical_block_size(bdev), 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503}
1504
1505int
1506xfs_setsize_buftarg(
1507 xfs_buftarg_t *btp,
1508 unsigned int blocksize,
1509 unsigned int sectorsize)
1510{
1511 return xfs_setsize_buftarg_flags(btp, blocksize, sectorsize, 1);
1512}
1513
1514STATIC int
David Chinnera6867a62006-01-11 15:37:58 +11001515xfs_alloc_delwrite_queue(
Jan Engelhardte2a07812010-03-23 09:52:55 +11001516 xfs_buftarg_t *btp,
1517 const char *fsname)
David Chinnera6867a62006-01-11 15:37:58 +11001518{
David Chinnera6867a62006-01-11 15:37:58 +11001519 INIT_LIST_HEAD(&btp->bt_delwrite_queue);
Eric Sandeen007c61c2007-10-11 17:43:56 +10001520 spin_lock_init(&btp->bt_delwrite_lock);
David Chinnera6867a62006-01-11 15:37:58 +11001521 btp->bt_flags = 0;
Jan Engelhardte2a07812010-03-23 09:52:55 +11001522 btp->bt_task = kthread_run(xfsbufd, btp, "xfsbufd/%s", fsname);
Dave Chinnerff57ab22010-11-30 17:27:57 +11001523 if (IS_ERR(btp->bt_task))
1524 return PTR_ERR(btp->bt_task);
1525 return 0;
David Chinnera6867a62006-01-11 15:37:58 +11001526}
1527
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528xfs_buftarg_t *
1529xfs_alloc_buftarg(
Dave Chinnerebad8612010-09-22 10:47:20 +10001530 struct xfs_mount *mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 struct block_device *bdev,
Jan Engelhardte2a07812010-03-23 09:52:55 +11001532 int external,
1533 const char *fsname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534{
1535 xfs_buftarg_t *btp;
1536
1537 btp = kmem_zalloc(sizeof(*btp), KM_SLEEP);
1538
Dave Chinnerebad8612010-09-22 10:47:20 +10001539 btp->bt_mount = mp;
Nathan Scottce8e9222006-01-11 15:39:08 +11001540 btp->bt_dev = bdev->bd_dev;
1541 btp->bt_bdev = bdev;
Dave Chinner0e6e8472011-03-26 09:16:45 +11001542 btp->bt_bdi = blk_get_backing_dev_info(bdev);
1543 if (!btp->bt_bdi)
1544 goto error;
1545
Dave Chinner430cbeb2010-12-02 16:30:55 +11001546 INIT_LIST_HEAD(&btp->bt_lru);
1547 spin_lock_init(&btp->bt_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 if (xfs_setsize_buftarg_early(btp, bdev))
1549 goto error;
Jan Engelhardte2a07812010-03-23 09:52:55 +11001550 if (xfs_alloc_delwrite_queue(btp, fsname))
David Chinnera6867a62006-01-11 15:37:58 +11001551 goto error;
Dave Chinnerff57ab22010-11-30 17:27:57 +11001552 btp->bt_shrinker.shrink = xfs_buftarg_shrink;
1553 btp->bt_shrinker.seeks = DEFAULT_SEEKS;
1554 register_shrinker(&btp->bt_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 return btp;
1556
1557error:
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001558 kmem_free(btp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 return NULL;
1560}
1561
1562
1563/*
Nathan Scottce8e9222006-01-11 15:39:08 +11001564 * Delayed write buffer handling
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566STATIC void
Nathan Scottce8e9222006-01-11 15:39:08 +11001567xfs_buf_delwri_queue(
1568 xfs_buf_t *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 int unlock)
1570{
Nathan Scottce8e9222006-01-11 15:39:08 +11001571 struct list_head *dwq = &bp->b_target->bt_delwrite_queue;
1572 spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock;
David Chinnera6867a62006-01-11 15:37:58 +11001573
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001574 trace_xfs_buf_delwri_queue(bp, _RET_IP_);
1575
Nathan Scottce8e9222006-01-11 15:39:08 +11001576 ASSERT((bp->b_flags&(XBF_DELWRI|XBF_ASYNC)) == (XBF_DELWRI|XBF_ASYNC));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577
David Chinnera6867a62006-01-11 15:37:58 +11001578 spin_lock(dwlk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 /* If already in the queue, dequeue and place at tail */
Nathan Scottce8e9222006-01-11 15:39:08 +11001580 if (!list_empty(&bp->b_list)) {
1581 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
1582 if (unlock)
1583 atomic_dec(&bp->b_hold);
1584 list_del(&bp->b_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 }
1586
Dave Chinnerc9c12972010-01-11 11:49:59 +00001587 if (list_empty(dwq)) {
1588 /* start xfsbufd as it is about to have something to do */
1589 wake_up_process(bp->b_target->bt_task);
1590 }
1591
Nathan Scottce8e9222006-01-11 15:39:08 +11001592 bp->b_flags |= _XBF_DELWRI_Q;
1593 list_add_tail(&bp->b_list, dwq);
1594 bp->b_queuetime = jiffies;
David Chinnera6867a62006-01-11 15:37:58 +11001595 spin_unlock(dwlk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596
1597 if (unlock)
Nathan Scottce8e9222006-01-11 15:39:08 +11001598 xfs_buf_unlock(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599}
1600
1601void
Nathan Scottce8e9222006-01-11 15:39:08 +11001602xfs_buf_delwri_dequeue(
1603 xfs_buf_t *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604{
Nathan Scottce8e9222006-01-11 15:39:08 +11001605 spinlock_t *dwlk = &bp->b_target->bt_delwrite_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 int dequeued = 0;
1607
David Chinnera6867a62006-01-11 15:37:58 +11001608 spin_lock(dwlk);
Nathan Scottce8e9222006-01-11 15:39:08 +11001609 if ((bp->b_flags & XBF_DELWRI) && !list_empty(&bp->b_list)) {
1610 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
1611 list_del_init(&bp->b_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 dequeued = 1;
1613 }
Nathan Scottce8e9222006-01-11 15:39:08 +11001614 bp->b_flags &= ~(XBF_DELWRI|_XBF_DELWRI_Q);
David Chinnera6867a62006-01-11 15:37:58 +11001615 spin_unlock(dwlk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616
1617 if (dequeued)
Nathan Scottce8e9222006-01-11 15:39:08 +11001618 xfs_buf_rele(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001620 trace_xfs_buf_delwri_dequeue(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621}
1622
Dave Chinnerd808f612010-02-02 10:13:42 +11001623/*
1624 * If a delwri buffer needs to be pushed before it has aged out, then promote
1625 * it to the head of the delwri queue so that it will be flushed on the next
1626 * xfsbufd run. We do this by resetting the queuetime of the buffer to be older
1627 * than the age currently needed to flush the buffer. Hence the next time the
1628 * xfsbufd sees it is guaranteed to be considered old enough to flush.
1629 */
1630void
1631xfs_buf_delwri_promote(
1632 struct xfs_buf *bp)
1633{
1634 struct xfs_buftarg *btp = bp->b_target;
1635 long age = xfs_buf_age_centisecs * msecs_to_jiffies(10) + 1;
1636
1637 ASSERT(bp->b_flags & XBF_DELWRI);
1638 ASSERT(bp->b_flags & _XBF_DELWRI_Q);
1639
1640 /*
1641 * Check the buffer age before locking the delayed write queue as we
1642 * don't need to promote buffers that are already past the flush age.
1643 */
1644 if (bp->b_queuetime < jiffies - age)
1645 return;
1646 bp->b_queuetime = jiffies - age;
1647 spin_lock(&btp->bt_delwrite_lock);
1648 list_move(&bp->b_list, &btp->bt_delwrite_queue);
1649 spin_unlock(&btp->bt_delwrite_lock);
1650}
1651
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652STATIC void
Nathan Scottce8e9222006-01-11 15:39:08 +11001653xfs_buf_runall_queues(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 struct workqueue_struct *queue)
1655{
1656 flush_workqueue(queue);
1657}
1658
David Chinner585e6d82007-02-10 18:32:29 +11001659/*
1660 * Move as many buffers as specified to the supplied list
1661 * idicating if we skipped any buffers to prevent deadlocks.
1662 */
1663STATIC int
1664xfs_buf_delwri_split(
1665 xfs_buftarg_t *target,
1666 struct list_head *list,
David Chinner5e6a07d2007-02-10 18:34:49 +11001667 unsigned long age)
David Chinner585e6d82007-02-10 18:32:29 +11001668{
1669 xfs_buf_t *bp, *n;
1670 struct list_head *dwq = &target->bt_delwrite_queue;
1671 spinlock_t *dwlk = &target->bt_delwrite_lock;
1672 int skipped = 0;
David Chinner5e6a07d2007-02-10 18:34:49 +11001673 int force;
David Chinner585e6d82007-02-10 18:32:29 +11001674
David Chinner5e6a07d2007-02-10 18:34:49 +11001675 force = test_and_clear_bit(XBT_FORCE_FLUSH, &target->bt_flags);
David Chinner585e6d82007-02-10 18:32:29 +11001676 INIT_LIST_HEAD(list);
1677 spin_lock(dwlk);
1678 list_for_each_entry_safe(bp, n, dwq, b_list) {
David Chinner585e6d82007-02-10 18:32:29 +11001679 ASSERT(bp->b_flags & XBF_DELWRI);
1680
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001681 if (!XFS_BUF_ISPINNED(bp) && xfs_buf_trylock(bp)) {
David Chinner5e6a07d2007-02-10 18:34:49 +11001682 if (!force &&
David Chinner585e6d82007-02-10 18:32:29 +11001683 time_before(jiffies, bp->b_queuetime + age)) {
1684 xfs_buf_unlock(bp);
1685 break;
1686 }
1687
Christoph Hellwig1d5ae5d2011-07-08 14:36:32 +02001688 bp->b_flags &= ~(XBF_DELWRI | _XBF_DELWRI_Q);
David Chinner585e6d82007-02-10 18:32:29 +11001689 bp->b_flags |= XBF_WRITE;
1690 list_move_tail(&bp->b_list, list);
Dave Chinnerbfe27412010-11-08 08:55:05 +00001691 trace_xfs_buf_delwri_split(bp, _RET_IP_);
David Chinner585e6d82007-02-10 18:32:29 +11001692 } else
1693 skipped++;
1694 }
1695 spin_unlock(dwlk);
1696
1697 return skipped;
1698
1699}
1700
Dave Chinner089716a2010-01-26 15:13:25 +11001701/*
1702 * Compare function is more complex than it needs to be because
1703 * the return value is only 32 bits and we are doing comparisons
1704 * on 64 bit values
1705 */
1706static int
1707xfs_buf_cmp(
1708 void *priv,
1709 struct list_head *a,
1710 struct list_head *b)
1711{
1712 struct xfs_buf *ap = container_of(a, struct xfs_buf, b_list);
1713 struct xfs_buf *bp = container_of(b, struct xfs_buf, b_list);
1714 xfs_daddr_t diff;
1715
1716 diff = ap->b_bn - bp->b_bn;
1717 if (diff < 0)
1718 return -1;
1719 if (diff > 0)
1720 return 1;
1721 return 0;
1722}
1723
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724STATIC int
Christoph Hellwig23ea4032005-06-21 15:14:01 +10001725xfsbufd(
David Chinner585e6d82007-02-10 18:32:29 +11001726 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727{
Dave Chinner089716a2010-01-26 15:13:25 +11001728 xfs_buftarg_t *target = (xfs_buftarg_t *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 current->flags |= PF_MEMALLOC;
1731
Rafael J. Wysocki978c7b22007-12-07 14:09:02 +11001732 set_freezable();
1733
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 do {
Dave Chinnerc9c12972010-01-11 11:49:59 +00001735 long age = xfs_buf_age_centisecs * msecs_to_jiffies(10);
1736 long tout = xfs_buf_timer_centisecs * msecs_to_jiffies(10);
Dave Chinner089716a2010-01-26 15:13:25 +11001737 struct list_head tmp;
Christoph Hellwiga1b7ea52011-03-30 11:05:09 +00001738 struct blk_plug plug;
Dave Chinnerc9c12972010-01-11 11:49:59 +00001739
Christoph Lameter3e1d1d22005-06-24 23:13:50 -07001740 if (unlikely(freezing(current))) {
Nathan Scottce8e9222006-01-11 15:39:08 +11001741 set_bit(XBT_FORCE_SLEEP, &target->bt_flags);
Christoph Lameter3e1d1d22005-06-24 23:13:50 -07001742 refrigerator();
Nathan Scottabd0cf72005-05-05 13:30:13 -07001743 } else {
Nathan Scottce8e9222006-01-11 15:39:08 +11001744 clear_bit(XBT_FORCE_SLEEP, &target->bt_flags);
Nathan Scottabd0cf72005-05-05 13:30:13 -07001745 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746
Dave Chinnerc9c12972010-01-11 11:49:59 +00001747 /* sleep for a long time if there is nothing to do. */
1748 if (list_empty(&target->bt_delwrite_queue))
1749 tout = MAX_SCHEDULE_TIMEOUT;
1750 schedule_timeout_interruptible(tout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751
Dave Chinnerc9c12972010-01-11 11:49:59 +00001752 xfs_buf_delwri_split(target, &tmp, age);
Dave Chinner089716a2010-01-26 15:13:25 +11001753 list_sort(NULL, &tmp, xfs_buf_cmp);
Christoph Hellwiga1b7ea52011-03-30 11:05:09 +00001754
1755 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 while (!list_empty(&tmp)) {
Dave Chinner089716a2010-01-26 15:13:25 +11001757 struct xfs_buf *bp;
1758 bp = list_first_entry(&tmp, struct xfs_buf, b_list);
Nathan Scottce8e9222006-01-11 15:39:08 +11001759 list_del_init(&bp->b_list);
Christoph Hellwig939d7232010-07-20 17:51:16 +10001760 xfs_bdstrat_cb(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 }
Christoph Hellwiga1b7ea52011-03-30 11:05:09 +00001762 blk_finish_plug(&plug);
Christoph Hellwig4df08c52005-09-05 08:34:18 +10001763 } while (!kthread_should_stop());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764
Christoph Hellwig4df08c52005-09-05 08:34:18 +10001765 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766}
1767
1768/*
Nathan Scottce8e9222006-01-11 15:39:08 +11001769 * Go through all incore buffers, and release buffers if they belong to
1770 * the given device. This is used in filesystem error handling to
1771 * preserve the consistency of its metadata.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 */
1773int
1774xfs_flush_buftarg(
David Chinner585e6d82007-02-10 18:32:29 +11001775 xfs_buftarg_t *target,
1776 int wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777{
Dave Chinner089716a2010-01-26 15:13:25 +11001778 xfs_buf_t *bp;
David Chinner585e6d82007-02-10 18:32:29 +11001779 int pincount = 0;
Dave Chinner089716a2010-01-26 15:13:25 +11001780 LIST_HEAD(tmp_list);
1781 LIST_HEAD(wait_list);
Christoph Hellwiga1b7ea52011-03-30 11:05:09 +00001782 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783
Dave Chinnerc626d172009-04-06 18:42:11 +02001784 xfs_buf_runall_queues(xfsconvertd_workqueue);
Nathan Scottce8e9222006-01-11 15:39:08 +11001785 xfs_buf_runall_queues(xfsdatad_workqueue);
1786 xfs_buf_runall_queues(xfslogd_workqueue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787
David Chinner5e6a07d2007-02-10 18:34:49 +11001788 set_bit(XBT_FORCE_FLUSH, &target->bt_flags);
Dave Chinner089716a2010-01-26 15:13:25 +11001789 pincount = xfs_buf_delwri_split(target, &tmp_list, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790
1791 /*
Dave Chinner089716a2010-01-26 15:13:25 +11001792 * Dropped the delayed write list lock, now walk the temporary list.
1793 * All I/O is issued async and then if we need to wait for completion
1794 * we do that after issuing all the IO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 */
Dave Chinner089716a2010-01-26 15:13:25 +11001796 list_sort(NULL, &tmp_list, xfs_buf_cmp);
Christoph Hellwiga1b7ea52011-03-30 11:05:09 +00001797
1798 blk_start_plug(&plug);
Dave Chinner089716a2010-01-26 15:13:25 +11001799 while (!list_empty(&tmp_list)) {
1800 bp = list_first_entry(&tmp_list, struct xfs_buf, b_list);
David Chinner585e6d82007-02-10 18:32:29 +11001801 ASSERT(target == bp->b_target);
Dave Chinner089716a2010-01-26 15:13:25 +11001802 list_del_init(&bp->b_list);
1803 if (wait) {
Nathan Scottce8e9222006-01-11 15:39:08 +11001804 bp->b_flags &= ~XBF_ASYNC;
Dave Chinner089716a2010-01-26 15:13:25 +11001805 list_add(&bp->b_list, &wait_list);
1806 }
Christoph Hellwig939d7232010-07-20 17:51:16 +10001807 xfs_bdstrat_cb(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 }
Christoph Hellwiga1b7ea52011-03-30 11:05:09 +00001809 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810
Dave Chinner089716a2010-01-26 15:13:25 +11001811 if (wait) {
Christoph Hellwiga1b7ea52011-03-30 11:05:09 +00001812 /* Wait for IO to complete. */
Dave Chinner089716a2010-01-26 15:13:25 +11001813 while (!list_empty(&wait_list)) {
1814 bp = list_first_entry(&wait_list, struct xfs_buf, b_list);
Nathan Scottf07c2252006-09-28 10:52:15 +10001815
Dave Chinner089716a2010-01-26 15:13:25 +11001816 list_del_init(&bp->b_list);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00001817 xfs_buf_iowait(bp);
Dave Chinner089716a2010-01-26 15:13:25 +11001818 xfs_buf_relse(bp);
1819 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 }
1821
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 return pincount;
1823}
1824
Christoph Hellwig04d8b282005-11-02 10:15:05 +11001825int __init
Nathan Scottce8e9222006-01-11 15:39:08 +11001826xfs_buf_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827{
Nathan Scott87582802006-03-14 13:18:19 +11001828 xfs_buf_zone = kmem_zone_init_flags(sizeof(xfs_buf_t), "xfs_buf",
1829 KM_ZONE_HWALIGN, NULL);
Nathan Scottce8e9222006-01-11 15:39:08 +11001830 if (!xfs_buf_zone)
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001831 goto out;
Christoph Hellwig04d8b282005-11-02 10:15:05 +11001832
Dave Chinner51749e42010-09-08 09:00:22 +00001833 xfslogd_workqueue = alloc_workqueue("xfslogd",
Tejun Heo6370a6a2010-10-11 15:12:27 +02001834 WQ_MEM_RECLAIM | WQ_HIGHPRI, 1);
Christoph Hellwig23ea4032005-06-21 15:14:01 +10001835 if (!xfslogd_workqueue)
Christoph Hellwig04d8b282005-11-02 10:15:05 +11001836 goto out_free_buf_zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837
Tejun Heo83e75902011-02-01 11:42:43 +01001838 xfsdatad_workqueue = alloc_workqueue("xfsdatad", WQ_MEM_RECLAIM, 1);
Christoph Hellwig23ea4032005-06-21 15:14:01 +10001839 if (!xfsdatad_workqueue)
1840 goto out_destroy_xfslogd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841
Tejun Heo83e75902011-02-01 11:42:43 +01001842 xfsconvertd_workqueue = alloc_workqueue("xfsconvertd",
1843 WQ_MEM_RECLAIM, 1);
Dave Chinnerc626d172009-04-06 18:42:11 +02001844 if (!xfsconvertd_workqueue)
1845 goto out_destroy_xfsdatad_workqueue;
1846
Christoph Hellwig23ea4032005-06-21 15:14:01 +10001847 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848
Dave Chinnerc626d172009-04-06 18:42:11 +02001849 out_destroy_xfsdatad_workqueue:
1850 destroy_workqueue(xfsdatad_workqueue);
Christoph Hellwig23ea4032005-06-21 15:14:01 +10001851 out_destroy_xfslogd_workqueue:
1852 destroy_workqueue(xfslogd_workqueue);
Christoph Hellwig23ea4032005-06-21 15:14:01 +10001853 out_free_buf_zone:
Nathan Scottce8e9222006-01-11 15:39:08 +11001854 kmem_zone_destroy(xfs_buf_zone);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001855 out:
Nathan Scott87582802006-03-14 13:18:19 +11001856 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857}
1858
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859void
Nathan Scottce8e9222006-01-11 15:39:08 +11001860xfs_buf_terminate(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861{
Dave Chinnerc626d172009-04-06 18:42:11 +02001862 destroy_workqueue(xfsconvertd_workqueue);
Christoph Hellwig04d8b282005-11-02 10:15:05 +11001863 destroy_workqueue(xfsdatad_workqueue);
1864 destroy_workqueue(xfslogd_workqueue);
Nathan Scottce8e9222006-01-11 15:39:08 +11001865 kmem_zone_destroy(xfs_buf_zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866}
Tim Shimmine6a0e9c2007-05-08 13:49:59 +10001867
1868#ifdef CONFIG_KDB_MODULES
1869struct list_head *
1870xfs_get_buftarg_list(void)
1871{
1872 return &xfs_buftarg_list;
1873}
1874#endif