Commit 4667a0ec32867865fd4deccf834594b3ea831baf
1 parent
f42ab08529
Exists in
master
and in
7 other branches
GFS2: Make writeback more responsive to system conditions
This patch adds writeback_control to writing back the AIL list. This means that we can then take advantage of the information we get in ->write_inode() in order to set off some pre-emptive writeback. In addition, the AIL code is cleaned up a bit to make it a bit simpler to understand. There is still more which can usefully be done in this area, but this is a good start at least. Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
Showing 8 changed files with 98 additions and 90 deletions Side-by-side Diff
fs/gfs2/export.c
... | ... | @@ -139,7 +139,7 @@ |
139 | 139 | struct gfs2_sbd *sdp = sb->s_fs_info; |
140 | 140 | struct inode *inode; |
141 | 141 | |
142 | - inode = gfs2_ilookup(sb, inum->no_addr); | |
142 | + inode = gfs2_ilookup(sb, inum->no_addr, 0); | |
143 | 143 | if (inode) { |
144 | 144 | if (GFS2_I(inode)->i_no_formal_ino != inum->no_formal_ino) { |
145 | 145 | iput(inode); |
fs/gfs2/glock.c
... | ... | @@ -649,7 +649,7 @@ |
649 | 649 | /* Note: Unsafe to dereference ip as we don't hold right refs/locks */ |
650 | 650 | |
651 | 651 | if (ip) |
652 | - inode = gfs2_ilookup(sdp->sd_vfs, no_addr); | |
652 | + inode = gfs2_ilookup(sdp->sd_vfs, no_addr, 1); | |
653 | 653 | else |
654 | 654 | inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED); |
655 | 655 | if (inode && !IS_ERR(inode)) { |
fs/gfs2/incore.h
... | ... | @@ -20,7 +20,6 @@ |
20 | 20 | |
21 | 21 | #define DIO_WAIT 0x00000010 |
22 | 22 | #define DIO_METADATA 0x00000020 |
23 | -#define DIO_ALL 0x00000100 | |
24 | 23 | |
25 | 24 | struct gfs2_log_operations; |
26 | 25 | struct gfs2_log_element; |
... | ... | @@ -377,8 +376,6 @@ |
377 | 376 | unsigned int ai_first; |
378 | 377 | struct list_head ai_ail1_list; |
379 | 378 | struct list_head ai_ail2_list; |
380 | - | |
381 | - u64 ai_sync_gen; | |
382 | 379 | }; |
383 | 380 | |
384 | 381 | struct gfs2_journal_extent { |
... | ... | @@ -657,7 +654,6 @@ |
657 | 654 | spinlock_t sd_ail_lock; |
658 | 655 | struct list_head sd_ail1_list; |
659 | 656 | struct list_head sd_ail2_list; |
660 | - u64 sd_ail_sync_gen; | |
661 | 657 | |
662 | 658 | /* Replay stuff */ |
663 | 659 |
fs/gfs2/inode.c
... | ... | @@ -74,14 +74,14 @@ |
74 | 74 | return 0; |
75 | 75 | } |
76 | 76 | |
77 | -struct inode *gfs2_ilookup(struct super_block *sb, u64 no_addr) | |
77 | +struct inode *gfs2_ilookup(struct super_block *sb, u64 no_addr, int non_block) | |
78 | 78 | { |
79 | 79 | unsigned long hash = (unsigned long)no_addr; |
80 | 80 | struct gfs2_skip_data data; |
81 | 81 | |
82 | 82 | data.no_addr = no_addr; |
83 | 83 | data.skipped = 0; |
84 | - data.non_block = 0; | |
84 | + data.non_block = non_block; | |
85 | 85 | return ilookup5(sb, hash, iget_test, &data); |
86 | 86 | } |
87 | 87 |
fs/gfs2/inode.h
... | ... | @@ -102,7 +102,7 @@ |
102 | 102 | extern struct inode *gfs2_lookup_by_inum(struct gfs2_sbd *sdp, u64 no_addr, |
103 | 103 | u64 *no_formal_ino, |
104 | 104 | unsigned int blktype); |
105 | -extern struct inode *gfs2_ilookup(struct super_block *sb, u64 no_addr); | |
105 | +extern struct inode *gfs2_ilookup(struct super_block *sb, u64 no_addr, int nonblock); | |
106 | 106 | |
107 | 107 | extern int gfs2_inode_refresh(struct gfs2_inode *ip); |
108 | 108 |
fs/gfs2/log.c
... | ... | @@ -18,6 +18,7 @@ |
18 | 18 | #include <linux/kthread.h> |
19 | 19 | #include <linux/freezer.h> |
20 | 20 | #include <linux/bio.h> |
21 | +#include <linux/writeback.h> | |
21 | 22 | |
22 | 23 | #include "gfs2.h" |
23 | 24 | #include "incore.h" |
24 | 25 | |
25 | 26 | |
26 | 27 | |
27 | 28 | |
28 | 29 | |
29 | 30 | |
30 | 31 | |
31 | 32 | |
32 | 33 | |
33 | 34 | |
34 | 35 | |
35 | 36 | |
36 | 37 | |
... | ... | @@ -83,60 +84,100 @@ |
83 | 84 | /** |
84 | 85 | * gfs2_ail1_start_one - Start I/O on a part of the AIL |
85 | 86 | * @sdp: the filesystem |
86 | - * @tr: the part of the AIL | |
87 | + * @wbc: The writeback control structure | |
88 | + * @ai: The ail structure | |
87 | 89 | * |
88 | 90 | */ |
89 | 91 | |
90 | -static void gfs2_ail1_start_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai) | |
92 | +static void gfs2_ail1_start_one(struct gfs2_sbd *sdp, | |
93 | + struct writeback_control *wbc, | |
94 | + struct gfs2_ail *ai) | |
91 | 95 | __releases(&sdp->sd_ail_lock) |
92 | 96 | __acquires(&sdp->sd_ail_lock) |
93 | 97 | { |
94 | 98 | struct gfs2_glock *gl = NULL; |
99 | + struct address_space *mapping; | |
95 | 100 | struct gfs2_bufdata *bd, *s; |
96 | 101 | struct buffer_head *bh; |
97 | - int retry; | |
98 | 102 | |
99 | - do { | |
100 | - retry = 0; | |
103 | +restart: | |
104 | + list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list, bd_ail_st_list) { | |
105 | + bh = bd->bd_bh; | |
101 | 106 | |
102 | - list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list, | |
103 | - bd_ail_st_list) { | |
104 | - bh = bd->bd_bh; | |
107 | + gfs2_assert(sdp, bd->bd_ail == ai); | |
105 | 108 | |
106 | - gfs2_assert(sdp, bd->bd_ail == ai); | |
109 | + if (!buffer_busy(bh)) { | |
110 | + if (!buffer_uptodate(bh)) | |
111 | + gfs2_io_error_bh(sdp, bh); | |
112 | + list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list); | |
113 | + continue; | |
114 | + } | |
107 | 115 | |
108 | - if (!buffer_busy(bh)) { | |
109 | - if (!buffer_uptodate(bh)) | |
110 | - gfs2_io_error_bh(sdp, bh); | |
111 | - list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list); | |
112 | - continue; | |
113 | - } | |
116 | + if (!buffer_dirty(bh)) | |
117 | + continue; | |
118 | + if (gl == bd->bd_gl) | |
119 | + continue; | |
120 | + gl = bd->bd_gl; | |
121 | + list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list); | |
122 | + mapping = bh->b_page->mapping; | |
123 | + spin_unlock(&sdp->sd_ail_lock); | |
124 | + generic_writepages(mapping, wbc); | |
125 | + spin_lock(&sdp->sd_ail_lock); | |
126 | + if (wbc->nr_to_write <= 0) | |
127 | + break; | |
128 | + goto restart; | |
129 | + } | |
130 | +} | |
114 | 131 | |
115 | - if (!buffer_dirty(bh)) | |
116 | - continue; | |
117 | - if (gl == bd->bd_gl) | |
118 | - continue; | |
119 | - gl = bd->bd_gl; | |
120 | - list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list); | |
121 | 132 | |
122 | - spin_unlock(&sdp->sd_ail_lock); | |
123 | - filemap_fdatawrite(gfs2_glock2aspace(gl)); | |
124 | - spin_lock(&sdp->sd_ail_lock); | |
133 | +/** | |
134 | + * gfs2_ail1_flush - start writeback of some ail1 entries | |
135 | + * @sdp: The super block | |
136 | + * @wbc: The writeback control structure | |
137 | + * | |
138 | + * Writes back some ail1 entries, according to the limits in the | |
139 | + * writeback control structure | |
140 | + */ | |
125 | 141 | |
126 | - retry = 1; | |
142 | +void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc) | |
143 | +{ | |
144 | + struct list_head *head = &sdp->sd_ail1_list; | |
145 | + struct gfs2_ail *ai; | |
146 | + | |
147 | + spin_lock(&sdp->sd_ail_lock); | |
148 | + list_for_each_entry_reverse(ai, head, ai_list) { | |
149 | + if (wbc->nr_to_write <= 0) | |
127 | 150 | break; |
128 | - } | |
129 | - } while (retry); | |
151 | + gfs2_ail1_start_one(sdp, wbc, ai); /* This may drop ail lock */ | |
152 | + } | |
153 | + spin_unlock(&sdp->sd_ail_lock); | |
130 | 154 | } |
131 | 155 | |
132 | 156 | /** |
157 | + * gfs2_ail1_start - start writeback of all ail1 entries | |
158 | + * @sdp: The superblock | |
159 | + */ | |
160 | + | |
161 | +static void gfs2_ail1_start(struct gfs2_sbd *sdp) | |
162 | +{ | |
163 | + struct writeback_control wbc = { | |
164 | + .sync_mode = WB_SYNC_NONE, | |
165 | + .nr_to_write = LONG_MAX, | |
166 | + .range_start = 0, | |
167 | + .range_end = LLONG_MAX, | |
168 | + }; | |
169 | + | |
170 | + return gfs2_ail1_flush(sdp, &wbc); | |
171 | +} | |
172 | + | |
173 | +/** | |
133 | 174 | * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced |
134 | 175 | * @sdp: the filesystem |
135 | 176 | * @ai: the AIL entry |
136 | 177 | * |
137 | 178 | */ |
138 | 179 | |
139 | -static int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai, int flags) | |
180 | +static void gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai) | |
140 | 181 | { |
141 | 182 | struct gfs2_bufdata *bd, *s; |
142 | 183 | struct buffer_head *bh; |
143 | 184 | |
144 | 185 | |
145 | 186 | |
146 | 187 | |
147 | 188 | |
148 | 189 | |
149 | 190 | |
150 | 191 | |
151 | 192 | |
152 | 193 | |
... | ... | @@ -144,71 +185,37 @@ |
144 | 185 | list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list, |
145 | 186 | bd_ail_st_list) { |
146 | 187 | bh = bd->bd_bh; |
147 | - | |
148 | 188 | gfs2_assert(sdp, bd->bd_ail == ai); |
149 | - | |
150 | - if (buffer_busy(bh)) { | |
151 | - if (flags & DIO_ALL) | |
152 | - continue; | |
153 | - else | |
154 | - break; | |
155 | - } | |
156 | - | |
189 | + if (buffer_busy(bh)) | |
190 | + continue; | |
157 | 191 | if (!buffer_uptodate(bh)) |
158 | 192 | gfs2_io_error_bh(sdp, bh); |
159 | - | |
160 | 193 | list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list); |
161 | 194 | } |
162 | 195 | |
163 | - return list_empty(&ai->ai_ail1_list); | |
164 | 196 | } |
165 | 197 | |
166 | -static void gfs2_ail1_start(struct gfs2_sbd *sdp) | |
167 | -{ | |
168 | - struct list_head *head; | |
169 | - u64 sync_gen; | |
170 | - struct gfs2_ail *ai; | |
171 | - int done = 0; | |
198 | +/** | |
199 | + * gfs2_ail1_empty - Try to empty the ail1 lists | |
200 | + * @sdp: The superblock | |
201 | + * | |
202 | + * Tries to empty the ail1 lists, starting with the oldest first | |
203 | + */ | |
172 | 204 | |
173 | - spin_lock(&sdp->sd_ail_lock); | |
174 | - head = &sdp->sd_ail1_list; | |
175 | - if (list_empty(head)) { | |
176 | - spin_unlock(&sdp->sd_ail_lock); | |
177 | - return; | |
178 | - } | |
179 | - sync_gen = sdp->sd_ail_sync_gen++; | |
180 | - | |
181 | - while(!done) { | |
182 | - done = 1; | |
183 | - list_for_each_entry_reverse(ai, head, ai_list) { | |
184 | - if (ai->ai_sync_gen >= sync_gen) | |
185 | - continue; | |
186 | - ai->ai_sync_gen = sync_gen; | |
187 | - gfs2_ail1_start_one(sdp, ai); /* This may drop ail lock */ | |
188 | - done = 0; | |
189 | - break; | |
190 | - } | |
191 | - } | |
192 | - | |
193 | - spin_unlock(&sdp->sd_ail_lock); | |
194 | -} | |
195 | - | |
196 | -static int gfs2_ail1_empty(struct gfs2_sbd *sdp, int flags) | |
205 | +static int gfs2_ail1_empty(struct gfs2_sbd *sdp) | |
197 | 206 | { |
198 | 207 | struct gfs2_ail *ai, *s; |
199 | 208 | int ret; |
200 | 209 | |
201 | 210 | spin_lock(&sdp->sd_ail_lock); |
202 | - | |
203 | 211 | list_for_each_entry_safe_reverse(ai, s, &sdp->sd_ail1_list, ai_list) { |
204 | - if (gfs2_ail1_empty_one(sdp, ai, flags)) | |
212 | + gfs2_ail1_empty_one(sdp, ai); | |
213 | + if (list_empty(&ai->ai_ail1_list)) | |
205 | 214 | list_move(&ai->ai_list, &sdp->sd_ail2_list); |
206 | - else if (!(flags & DIO_ALL)) | |
215 | + else | |
207 | 216 | break; |
208 | 217 | } |
209 | - | |
210 | 218 | ret = list_empty(&sdp->sd_ail1_list); |
211 | - | |
212 | 219 | spin_unlock(&sdp->sd_ail_lock); |
213 | 220 | |
214 | 221 | return ret; |
... | ... | @@ -569,7 +576,7 @@ |
569 | 576 | set_buffer_uptodate(bh); |
570 | 577 | clear_buffer_dirty(bh); |
571 | 578 | |
572 | - gfs2_ail1_empty(sdp, 0); | |
579 | + gfs2_ail1_empty(sdp); | |
573 | 580 | tail = current_tail(sdp); |
574 | 581 | |
575 | 582 | lh = (struct gfs2_log_header *)bh->b_data; |
... | ... | @@ -864,7 +871,7 @@ |
864 | 871 | gfs2_log_flush(sdp, NULL); |
865 | 872 | for (;;) { |
866 | 873 | gfs2_ail1_start(sdp); |
867 | - if (gfs2_ail1_empty(sdp, DIO_ALL)) | |
874 | + if (gfs2_ail1_empty(sdp)) | |
868 | 875 | break; |
869 | 876 | msleep(10); |
870 | 877 | } |
871 | 878 | |
872 | 879 | |
873 | 880 | |
... | ... | @@ -900,17 +907,15 @@ |
900 | 907 | |
901 | 908 | preflush = atomic_read(&sdp->sd_log_pinned); |
902 | 909 | if (gfs2_jrnl_flush_reqd(sdp) || t == 0) { |
903 | - gfs2_ail1_empty(sdp, DIO_ALL); | |
910 | + gfs2_ail1_empty(sdp); | |
904 | 911 | gfs2_log_flush(sdp, NULL); |
905 | - gfs2_ail1_empty(sdp, DIO_ALL); | |
906 | 912 | } |
907 | 913 | |
908 | 914 | if (gfs2_ail_flush_reqd(sdp)) { |
909 | 915 | gfs2_ail1_start(sdp); |
910 | 916 | io_schedule(); |
911 | - gfs2_ail1_empty(sdp, 0); | |
917 | + gfs2_ail1_empty(sdp); | |
912 | 918 | gfs2_log_flush(sdp, NULL); |
913 | - gfs2_ail1_empty(sdp, DIO_ALL); | |
914 | 919 | } |
915 | 920 | |
916 | 921 | wake_up(&sdp->sd_log_waitq); |
fs/gfs2/log.h
... | ... | @@ -12,6 +12,7 @@ |
12 | 12 | |
13 | 13 | #include <linux/list.h> |
14 | 14 | #include <linux/spinlock.h> |
15 | +#include <linux/writeback.h> | |
15 | 16 | #include "incore.h" |
16 | 17 | |
17 | 18 | /** |
... | ... | @@ -59,6 +60,7 @@ |
59 | 60 | extern void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl); |
60 | 61 | extern void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *trans); |
61 | 62 | extern void gfs2_remove_from_ail(struct gfs2_bufdata *bd); |
63 | +extern void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc); | |
62 | 64 | |
63 | 65 | extern void gfs2_log_shutdown(struct gfs2_sbd *sdp); |
64 | 66 | extern void gfs2_meta_syncfs(struct gfs2_sbd *sdp); |
fs/gfs2/super.c
... | ... | @@ -23,6 +23,7 @@ |
23 | 23 | #include <linux/time.h> |
24 | 24 | #include <linux/wait.h> |
25 | 25 | #include <linux/writeback.h> |
26 | +#include <linux/backing-dev.h> | |
26 | 27 | |
27 | 28 | #include "gfs2.h" |
28 | 29 | #include "incore.h" |
... | ... | @@ -714,6 +715,7 @@ |
714 | 715 | struct gfs2_inode *ip = GFS2_I(inode); |
715 | 716 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
716 | 717 | struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl); |
718 | + struct backing_dev_info *bdi = metamapping->backing_dev_info; | |
717 | 719 | struct gfs2_holder gh; |
718 | 720 | struct buffer_head *bh; |
719 | 721 | struct timespec atime; |
... | ... | @@ -747,6 +749,8 @@ |
747 | 749 | if (wbc->sync_mode == WB_SYNC_ALL) |
748 | 750 | gfs2_log_flush(GFS2_SB(inode), ip->i_gl); |
749 | 751 | filemap_fdatawrite(metamapping); |
752 | + if (bdi->dirty_exceeded) | |
753 | + gfs2_ail1_flush(sdp, wbc); | |
750 | 754 | if (!ret && (wbc->sync_mode == WB_SYNC_ALL)) |
751 | 755 | ret = filemap_fdatawait(metamapping); |
752 | 756 | if (ret) |
... | ... | @@ -1366,7 +1370,8 @@ |
1366 | 1370 | if (error) |
1367 | 1371 | goto out_rindex_relse; |
1368 | 1372 | |
1369 | - error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA, 1); | |
1373 | + error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA, | |
1374 | + sdp->sd_jdesc->jd_blocks); | |
1370 | 1375 | if (error) |
1371 | 1376 | goto out_rg_gunlock; |
1372 | 1377 |