diff options
Diffstat (limited to 'fs/reiserfs/journal.c')
-rw-r--r-- | fs/reiserfs/journal.c | 6855 |
1 files changed, 3598 insertions, 3257 deletions
diff --git a/fs/reiserfs/journal.c b/fs/reiserfs/journal.c index d1bcf0da6728..c66c27ec4100 100644 --- a/fs/reiserfs/journal.c +++ b/fs/reiserfs/journal.c | |||
@@ -55,7 +55,6 @@ | |||
55 | #include <linux/writeback.h> | 55 | #include <linux/writeback.h> |
56 | #include <linux/blkdev.h> | 56 | #include <linux/blkdev.h> |
57 | 57 | ||
58 | |||
59 | /* gets a struct reiserfs_journal_list * from a list head */ | 58 | /* gets a struct reiserfs_journal_list * from a list head */ |
60 | #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \ | 59 | #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \ |
61 | j_list)) | 60 | j_list)) |
@@ -69,55 +68,61 @@ static int reiserfs_mounted_fs_count; | |||
69 | 68 | ||
70 | static struct workqueue_struct *commit_wq; | 69 | static struct workqueue_struct *commit_wq; |
71 | 70 | ||
72 | #define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit | 71 | #define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit |
73 | structs at 4k */ | 72 | structs at 4k */ |
74 | #define BUFNR 64 /*read ahead */ | 73 | #define BUFNR 64 /*read ahead */ |
75 | 74 | ||
76 | /* cnode stat bits. Move these into reiserfs_fs.h */ | 75 | /* cnode stat bits. Move these into reiserfs_fs.h */ |
77 | 76 | ||
78 | #define BLOCK_FREED 2 /* this block was freed, and can't be written. */ | 77 | #define BLOCK_FREED 2 /* this block was freed, and can't be written. */ |
79 | #define BLOCK_FREED_HOLDER 3 /* this block was freed during this transaction, and can't be written */ | 78 | #define BLOCK_FREED_HOLDER 3 /* this block was freed during this transaction, and can't be written */ |
80 | 79 | ||
81 | #define BLOCK_NEEDS_FLUSH 4 /* used in flush_journal_list */ | 80 | #define BLOCK_NEEDS_FLUSH 4 /* used in flush_journal_list */ |
82 | #define BLOCK_DIRTIED 5 | 81 | #define BLOCK_DIRTIED 5 |
83 | 82 | ||
84 | |||
85 | /* journal list state bits */ | 83 | /* journal list state bits */ |
86 | #define LIST_TOUCHED 1 | 84 | #define LIST_TOUCHED 1 |
87 | #define LIST_DIRTY 2 | 85 | #define LIST_DIRTY 2 |
88 | #define LIST_COMMIT_PENDING 4 /* someone will commit this list */ | 86 | #define LIST_COMMIT_PENDING 4 /* someone will commit this list */ |
89 | 87 | ||
90 | /* flags for do_journal_end */ | 88 | /* flags for do_journal_end */ |
91 | #define FLUSH_ALL 1 /* flush commit and real blocks */ | 89 | #define FLUSH_ALL 1 /* flush commit and real blocks */ |
92 | #define COMMIT_NOW 2 /* end and commit this transaction */ | 90 | #define COMMIT_NOW 2 /* end and commit this transaction */ |
93 | #define WAIT 4 /* wait for the log blocks to hit the disk*/ | 91 | #define WAIT 4 /* wait for the log blocks to hit the disk */ |
94 | 92 | ||
95 | static int do_journal_end(struct reiserfs_transaction_handle *,struct super_block *,unsigned long nblocks,int flags) ; | 93 | static int do_journal_end(struct reiserfs_transaction_handle *, |
96 | static int flush_journal_list(struct super_block *s, struct reiserfs_journal_list *jl, int flushall) ; | 94 | struct super_block *, unsigned long nblocks, |
97 | static int flush_commit_list(struct super_block *s, struct reiserfs_journal_list *jl, int flushall) ; | 95 | int flags); |
98 | static int can_dirty(struct reiserfs_journal_cnode *cn) ; | 96 | static int flush_journal_list(struct super_block *s, |
99 | static int journal_join(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long nblocks); | 97 | struct reiserfs_journal_list *jl, int flushall); |
100 | static int release_journal_dev( struct super_block *super, | 98 | static int flush_commit_list(struct super_block *s, |
101 | struct reiserfs_journal *journal ); | 99 | struct reiserfs_journal_list *jl, int flushall); |
100 | static int can_dirty(struct reiserfs_journal_cnode *cn); | ||
101 | static int journal_join(struct reiserfs_transaction_handle *th, | ||
102 | struct super_block *p_s_sb, unsigned long nblocks); | ||
103 | static int release_journal_dev(struct super_block *super, | ||
104 | struct reiserfs_journal *journal); | ||
102 | static int dirty_one_transaction(struct super_block *s, | 105 | static int dirty_one_transaction(struct super_block *s, |
103 | struct reiserfs_journal_list *jl); | 106 | struct reiserfs_journal_list *jl); |
104 | static void flush_async_commits(void *p); | 107 | static void flush_async_commits(void *p); |
105 | static void queue_log_writer(struct super_block *s); | 108 | static void queue_log_writer(struct super_block *s); |
106 | 109 | ||
107 | /* values for join in do_journal_begin_r */ | 110 | /* values for join in do_journal_begin_r */ |
108 | enum { | 111 | enum { |
109 | JBEGIN_REG = 0, /* regular journal begin */ | 112 | JBEGIN_REG = 0, /* regular journal begin */ |
110 | JBEGIN_JOIN = 1, /* join the running transaction if at all possible */ | 113 | JBEGIN_JOIN = 1, /* join the running transaction if at all possible */ |
111 | JBEGIN_ABORT = 2, /* called from cleanup code, ignores aborted flag */ | 114 | JBEGIN_ABORT = 2, /* called from cleanup code, ignores aborted flag */ |
112 | }; | 115 | }; |
113 | 116 | ||
114 | static int do_journal_begin_r(struct reiserfs_transaction_handle *th, | 117 | static int do_journal_begin_r(struct reiserfs_transaction_handle *th, |
115 | struct super_block * p_s_sb, | 118 | struct super_block *p_s_sb, |
116 | unsigned long nblocks,int join); | 119 | unsigned long nblocks, int join); |
117 | 120 | ||
118 | static void init_journal_hash(struct super_block *p_s_sb) { | 121 | static void init_journal_hash(struct super_block *p_s_sb) |
119 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | 122 | { |
120 | memset(journal->j_hash_table, 0, JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)) ; | 123 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); |
124 | memset(journal->j_hash_table, 0, | ||
125 | JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)); | ||
121 | } | 126 | } |
122 | 127 | ||
123 | /* | 128 | /* |
@@ -125,149 +130,159 @@ static void init_journal_hash(struct super_block *p_s_sb) { | |||
125 | ** make schedule happen after I've freed a block. Look at remove_from_transaction and journal_mark_freed for | 130 | ** make schedule happen after I've freed a block. Look at remove_from_transaction and journal_mark_freed for |
126 | ** more details. | 131 | ** more details. |
127 | */ | 132 | */ |
128 | static int reiserfs_clean_and_file_buffer(struct buffer_head *bh) { | 133 | static int reiserfs_clean_and_file_buffer(struct buffer_head *bh) |
129 | if (bh) { | 134 | { |
130 | clear_buffer_dirty(bh); | 135 | if (bh) { |
131 | clear_buffer_journal_test(bh); | 136 | clear_buffer_dirty(bh); |
132 | } | 137 | clear_buffer_journal_test(bh); |
133 | return 0 ; | 138 | } |
139 | return 0; | ||
134 | } | 140 | } |
135 | 141 | ||
136 | static void disable_barrier(struct super_block *s) | 142 | static void disable_barrier(struct super_block *s) |
137 | { | 143 | { |
138 | REISERFS_SB(s)->s_mount_opt &= ~(1 << REISERFS_BARRIER_FLUSH); | 144 | REISERFS_SB(s)->s_mount_opt &= ~(1 << REISERFS_BARRIER_FLUSH); |
139 | printk("reiserfs: disabling flush barriers on %s\n", reiserfs_bdevname(s)); | 145 | printk("reiserfs: disabling flush barriers on %s\n", |
140 | } | 146 | reiserfs_bdevname(s)); |
141 | 147 | } | |
142 | static struct reiserfs_bitmap_node * | 148 | |
143 | allocate_bitmap_node(struct super_block *p_s_sb) { | 149 | static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block |
144 | struct reiserfs_bitmap_node *bn ; | 150 | *p_s_sb) |
145 | static int id; | 151 | { |
146 | 152 | struct reiserfs_bitmap_node *bn; | |
147 | bn = reiserfs_kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS, p_s_sb) ; | 153 | static int id; |
148 | if (!bn) { | 154 | |
149 | return NULL ; | 155 | bn = reiserfs_kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS, |
150 | } | 156 | p_s_sb); |
151 | bn->data = reiserfs_kmalloc(p_s_sb->s_blocksize, GFP_NOFS, p_s_sb) ; | 157 | if (!bn) { |
152 | if (!bn->data) { | 158 | return NULL; |
153 | reiserfs_kfree(bn, sizeof(struct reiserfs_bitmap_node), p_s_sb) ; | 159 | } |
154 | return NULL ; | 160 | bn->data = reiserfs_kmalloc(p_s_sb->s_blocksize, GFP_NOFS, p_s_sb); |
155 | } | 161 | if (!bn->data) { |
156 | bn->id = id++ ; | 162 | reiserfs_kfree(bn, sizeof(struct reiserfs_bitmap_node), p_s_sb); |
157 | memset(bn->data, 0, p_s_sb->s_blocksize) ; | 163 | return NULL; |
158 | INIT_LIST_HEAD(&bn->list) ; | 164 | } |
159 | return bn ; | 165 | bn->id = id++; |
160 | } | 166 | memset(bn->data, 0, p_s_sb->s_blocksize); |
161 | 167 | INIT_LIST_HEAD(&bn->list); | |
162 | static struct reiserfs_bitmap_node * | 168 | return bn; |
163 | get_bitmap_node(struct super_block *p_s_sb) { | 169 | } |
164 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | 170 | |
165 | struct reiserfs_bitmap_node *bn = NULL; | 171 | static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *p_s_sb) |
166 | struct list_head *entry = journal->j_bitmap_nodes.next ; | 172 | { |
167 | 173 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); | |
168 | journal->j_used_bitmap_nodes++ ; | 174 | struct reiserfs_bitmap_node *bn = NULL; |
169 | repeat: | 175 | struct list_head *entry = journal->j_bitmap_nodes.next; |
170 | 176 | ||
171 | if(entry != &journal->j_bitmap_nodes) { | 177 | journal->j_used_bitmap_nodes++; |
172 | bn = list_entry(entry, struct reiserfs_bitmap_node, list) ; | 178 | repeat: |
173 | list_del(entry) ; | 179 | |
174 | memset(bn->data, 0, p_s_sb->s_blocksize) ; | 180 | if (entry != &journal->j_bitmap_nodes) { |
175 | journal->j_free_bitmap_nodes-- ; | 181 | bn = list_entry(entry, struct reiserfs_bitmap_node, list); |
176 | return bn ; | 182 | list_del(entry); |
177 | } | 183 | memset(bn->data, 0, p_s_sb->s_blocksize); |
178 | bn = allocate_bitmap_node(p_s_sb) ; | 184 | journal->j_free_bitmap_nodes--; |
179 | if (!bn) { | 185 | return bn; |
180 | yield(); | 186 | } |
181 | goto repeat ; | 187 | bn = allocate_bitmap_node(p_s_sb); |
182 | } | 188 | if (!bn) { |
183 | return bn ; | 189 | yield(); |
190 | goto repeat; | ||
191 | } | ||
192 | return bn; | ||
184 | } | 193 | } |
185 | static inline void free_bitmap_node(struct super_block *p_s_sb, | 194 | static inline void free_bitmap_node(struct super_block *p_s_sb, |
186 | struct reiserfs_bitmap_node *bn) { | 195 | struct reiserfs_bitmap_node *bn) |
187 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | 196 | { |
188 | journal->j_used_bitmap_nodes-- ; | 197 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); |
189 | if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) { | 198 | journal->j_used_bitmap_nodes--; |
190 | reiserfs_kfree(bn->data, p_s_sb->s_blocksize, p_s_sb) ; | 199 | if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) { |
191 | reiserfs_kfree(bn, sizeof(struct reiserfs_bitmap_node), p_s_sb) ; | 200 | reiserfs_kfree(bn->data, p_s_sb->s_blocksize, p_s_sb); |
192 | } else { | 201 | reiserfs_kfree(bn, sizeof(struct reiserfs_bitmap_node), p_s_sb); |
193 | list_add(&bn->list, &journal->j_bitmap_nodes) ; | 202 | } else { |
194 | journal->j_free_bitmap_nodes++ ; | 203 | list_add(&bn->list, &journal->j_bitmap_nodes); |
195 | } | 204 | journal->j_free_bitmap_nodes++; |
196 | } | 205 | } |
197 | 206 | } | |
198 | static void allocate_bitmap_nodes(struct super_block *p_s_sb) { | 207 | |
199 | int i ; | 208 | static void allocate_bitmap_nodes(struct super_block *p_s_sb) |
200 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | 209 | { |
201 | struct reiserfs_bitmap_node *bn = NULL ; | 210 | int i; |
202 | for (i = 0 ; i < REISERFS_MIN_BITMAP_NODES ; i++) { | 211 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); |
203 | bn = allocate_bitmap_node(p_s_sb) ; | 212 | struct reiserfs_bitmap_node *bn = NULL; |
204 | if (bn) { | 213 | for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) { |
205 | list_add(&bn->list, &journal->j_bitmap_nodes) ; | 214 | bn = allocate_bitmap_node(p_s_sb); |
206 | journal->j_free_bitmap_nodes++ ; | 215 | if (bn) { |
207 | } else { | 216 | list_add(&bn->list, &journal->j_bitmap_nodes); |
208 | break ; // this is ok, we'll try again when more are needed | 217 | journal->j_free_bitmap_nodes++; |
209 | } | 218 | } else { |
210 | } | 219 | break; // this is ok, we'll try again when more are needed |
220 | } | ||
221 | } | ||
211 | } | 222 | } |
212 | 223 | ||
213 | static int set_bit_in_list_bitmap(struct super_block *p_s_sb, int block, | 224 | static int set_bit_in_list_bitmap(struct super_block *p_s_sb, int block, |
214 | struct reiserfs_list_bitmap *jb) { | 225 | struct reiserfs_list_bitmap *jb) |
215 | int bmap_nr = block / (p_s_sb->s_blocksize << 3) ; | 226 | { |
216 | int bit_nr = block % (p_s_sb->s_blocksize << 3) ; | 227 | int bmap_nr = block / (p_s_sb->s_blocksize << 3); |
228 | int bit_nr = block % (p_s_sb->s_blocksize << 3); | ||
217 | 229 | ||
218 | if (!jb->bitmaps[bmap_nr]) { | 230 | if (!jb->bitmaps[bmap_nr]) { |
219 | jb->bitmaps[bmap_nr] = get_bitmap_node(p_s_sb) ; | 231 | jb->bitmaps[bmap_nr] = get_bitmap_node(p_s_sb); |
220 | } | 232 | } |
221 | set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data) ; | 233 | set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data); |
222 | return 0 ; | 234 | return 0; |
223 | } | 235 | } |
224 | 236 | ||
225 | static void cleanup_bitmap_list(struct super_block *p_s_sb, | 237 | static void cleanup_bitmap_list(struct super_block *p_s_sb, |
226 | struct reiserfs_list_bitmap *jb) { | 238 | struct reiserfs_list_bitmap *jb) |
227 | int i; | 239 | { |
228 | if (jb->bitmaps == NULL) | 240 | int i; |
229 | return; | 241 | if (jb->bitmaps == NULL) |
230 | 242 | return; | |
231 | for (i = 0 ; i < SB_BMAP_NR(p_s_sb) ; i++) { | 243 | |
232 | if (jb->bitmaps[i]) { | 244 | for (i = 0; i < SB_BMAP_NR(p_s_sb); i++) { |
233 | free_bitmap_node(p_s_sb, jb->bitmaps[i]) ; | 245 | if (jb->bitmaps[i]) { |
234 | jb->bitmaps[i] = NULL ; | 246 | free_bitmap_node(p_s_sb, jb->bitmaps[i]); |
235 | } | 247 | jb->bitmaps[i] = NULL; |
236 | } | 248 | } |
249 | } | ||
237 | } | 250 | } |
238 | 251 | ||
239 | /* | 252 | /* |
240 | ** only call this on FS unmount. | 253 | ** only call this on FS unmount. |
241 | */ | 254 | */ |
242 | static int free_list_bitmaps(struct super_block *p_s_sb, | 255 | static int free_list_bitmaps(struct super_block *p_s_sb, |
243 | struct reiserfs_list_bitmap *jb_array) { | 256 | struct reiserfs_list_bitmap *jb_array) |
244 | int i ; | 257 | { |
245 | struct reiserfs_list_bitmap *jb ; | 258 | int i; |
246 | for (i = 0 ; i < JOURNAL_NUM_BITMAPS ; i++) { | 259 | struct reiserfs_list_bitmap *jb; |
247 | jb = jb_array + i ; | 260 | for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) { |
248 | jb->journal_list = NULL ; | 261 | jb = jb_array + i; |
249 | cleanup_bitmap_list(p_s_sb, jb) ; | 262 | jb->journal_list = NULL; |
250 | vfree(jb->bitmaps) ; | 263 | cleanup_bitmap_list(p_s_sb, jb); |
251 | jb->bitmaps = NULL ; | 264 | vfree(jb->bitmaps); |
252 | } | 265 | jb->bitmaps = NULL; |
253 | return 0; | 266 | } |
254 | } | 267 | return 0; |
255 | 268 | } | |
256 | static int free_bitmap_nodes(struct super_block *p_s_sb) { | 269 | |
257 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | 270 | static int free_bitmap_nodes(struct super_block *p_s_sb) |
258 | struct list_head *next = journal->j_bitmap_nodes.next ; | 271 | { |
259 | struct reiserfs_bitmap_node *bn ; | 272 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); |
260 | 273 | struct list_head *next = journal->j_bitmap_nodes.next; | |
261 | while(next != &journal->j_bitmap_nodes) { | 274 | struct reiserfs_bitmap_node *bn; |
262 | bn = list_entry(next, struct reiserfs_bitmap_node, list) ; | 275 | |
263 | list_del(next) ; | 276 | while (next != &journal->j_bitmap_nodes) { |
264 | reiserfs_kfree(bn->data, p_s_sb->s_blocksize, p_s_sb) ; | 277 | bn = list_entry(next, struct reiserfs_bitmap_node, list); |
265 | reiserfs_kfree(bn, sizeof(struct reiserfs_bitmap_node), p_s_sb) ; | 278 | list_del(next); |
266 | next = journal->j_bitmap_nodes.next ; | 279 | reiserfs_kfree(bn->data, p_s_sb->s_blocksize, p_s_sb); |
267 | journal->j_free_bitmap_nodes-- ; | 280 | reiserfs_kfree(bn, sizeof(struct reiserfs_bitmap_node), p_s_sb); |
268 | } | 281 | next = journal->j_bitmap_nodes.next; |
269 | 282 | journal->j_free_bitmap_nodes--; | |
270 | return 0 ; | 283 | } |
284 | |||
285 | return 0; | ||
271 | } | 286 | } |
272 | 287 | ||
273 | /* | 288 | /* |
@@ -275,59 +290,65 @@ static int free_bitmap_nodes(struct super_block *p_s_sb) { | |||
275 | ** jb_array is the array to be filled in. | 290 | ** jb_array is the array to be filled in. |
276 | */ | 291 | */ |
277 | int reiserfs_allocate_list_bitmaps(struct super_block *p_s_sb, | 292 | int reiserfs_allocate_list_bitmaps(struct super_block *p_s_sb, |
278 | struct reiserfs_list_bitmap *jb_array, | 293 | struct reiserfs_list_bitmap *jb_array, |
279 | int bmap_nr) { | 294 | int bmap_nr) |
280 | int i ; | 295 | { |
281 | int failed = 0 ; | 296 | int i; |
282 | struct reiserfs_list_bitmap *jb ; | 297 | int failed = 0; |
283 | int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *) ; | 298 | struct reiserfs_list_bitmap *jb; |
284 | 299 | int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *); | |
285 | for (i = 0 ; i < JOURNAL_NUM_BITMAPS ; i++) { | 300 | |
286 | jb = jb_array + i ; | 301 | for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) { |
287 | jb->journal_list = NULL ; | 302 | jb = jb_array + i; |
288 | jb->bitmaps = vmalloc( mem ) ; | 303 | jb->journal_list = NULL; |
289 | if (!jb->bitmaps) { | 304 | jb->bitmaps = vmalloc(mem); |
290 | reiserfs_warning(p_s_sb, "clm-2000, unable to allocate bitmaps for journal lists") ; | 305 | if (!jb->bitmaps) { |
291 | failed = 1; | 306 | reiserfs_warning(p_s_sb, |
292 | break ; | 307 | "clm-2000, unable to allocate bitmaps for journal lists"); |
293 | } | 308 | failed = 1; |
294 | memset(jb->bitmaps, 0, mem) ; | 309 | break; |
295 | } | 310 | } |
296 | if (failed) { | 311 | memset(jb->bitmaps, 0, mem); |
297 | free_list_bitmaps(p_s_sb, jb_array) ; | 312 | } |
298 | return -1 ; | 313 | if (failed) { |
299 | } | 314 | free_list_bitmaps(p_s_sb, jb_array); |
300 | return 0 ; | 315 | return -1; |
316 | } | ||
317 | return 0; | ||
301 | } | 318 | } |
302 | 319 | ||
303 | /* | 320 | /* |
304 | ** find an available list bitmap. If you can't find one, flush a commit list | 321 | ** find an available list bitmap. If you can't find one, flush a commit list |
305 | ** and try again | 322 | ** and try again |
306 | */ | 323 | */ |
307 | static struct reiserfs_list_bitmap * | 324 | static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *p_s_sb, |
308 | get_list_bitmap(struct super_block *p_s_sb, struct reiserfs_journal_list *jl) { | 325 | struct reiserfs_journal_list |
309 | int i,j ; | 326 | *jl) |
310 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | 327 | { |
311 | struct reiserfs_list_bitmap *jb = NULL ; | 328 | int i, j; |
312 | 329 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); | |
313 | for (j = 0 ; j < (JOURNAL_NUM_BITMAPS * 3) ; j++) { | 330 | struct reiserfs_list_bitmap *jb = NULL; |
314 | i = journal->j_list_bitmap_index ; | 331 | |
315 | journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS ; | 332 | for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) { |
316 | jb = journal->j_list_bitmap + i ; | 333 | i = journal->j_list_bitmap_index; |
317 | if (journal->j_list_bitmap[i].journal_list) { | 334 | journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS; |
318 | flush_commit_list(p_s_sb, journal->j_list_bitmap[i].journal_list, 1) ; | 335 | jb = journal->j_list_bitmap + i; |
319 | if (!journal->j_list_bitmap[i].journal_list) { | 336 | if (journal->j_list_bitmap[i].journal_list) { |
320 | break ; | 337 | flush_commit_list(p_s_sb, |
321 | } | 338 | journal->j_list_bitmap[i]. |
322 | } else { | 339 | journal_list, 1); |
323 | break ; | 340 | if (!journal->j_list_bitmap[i].journal_list) { |
324 | } | 341 | break; |
325 | } | 342 | } |
326 | if (jb->journal_list) { /* double check to make sure if flushed correctly */ | 343 | } else { |
327 | return NULL ; | 344 | break; |
328 | } | 345 | } |
329 | jb->journal_list = jl ; | 346 | } |
330 | return jb ; | 347 | if (jb->journal_list) { /* double check to make sure if flushed correctly */ |
348 | return NULL; | ||
349 | } | ||
350 | jb->journal_list = jl; | ||
351 | return jb; | ||
331 | } | 352 | } |
332 | 353 | ||
333 | /* | 354 | /* |
@@ -335,104 +356,114 @@ get_list_bitmap(struct super_block *p_s_sb, struct reiserfs_journal_list *jl) { | |||
335 | ** Uses the cnode->next and cnode->prev pointers | 356 | ** Uses the cnode->next and cnode->prev pointers |
336 | ** returns NULL on failure | 357 | ** returns NULL on failure |
337 | */ | 358 | */ |
338 | static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes) { | 359 | static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes) |
339 | struct reiserfs_journal_cnode *head ; | 360 | { |
340 | int i ; | 361 | struct reiserfs_journal_cnode *head; |
341 | if (num_cnodes <= 0) { | 362 | int i; |
342 | return NULL ; | 363 | if (num_cnodes <= 0) { |
343 | } | 364 | return NULL; |
344 | head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode)) ; | 365 | } |
345 | if (!head) { | 366 | head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode)); |
346 | return NULL ; | 367 | if (!head) { |
347 | } | 368 | return NULL; |
348 | memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode)) ; | 369 | } |
349 | head[0].prev = NULL ; | 370 | memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode)); |
350 | head[0].next = head + 1 ; | 371 | head[0].prev = NULL; |
351 | for (i = 1 ; i < num_cnodes; i++) { | 372 | head[0].next = head + 1; |
352 | head[i].prev = head + (i - 1) ; | 373 | for (i = 1; i < num_cnodes; i++) { |
353 | head[i].next = head + (i + 1) ; /* if last one, overwrite it after the if */ | 374 | head[i].prev = head + (i - 1); |
354 | } | 375 | head[i].next = head + (i + 1); /* if last one, overwrite it after the if */ |
355 | head[num_cnodes -1].next = NULL ; | 376 | } |
356 | return head ; | 377 | head[num_cnodes - 1].next = NULL; |
378 | return head; | ||
357 | } | 379 | } |
358 | 380 | ||
359 | /* | 381 | /* |
360 | ** pulls a cnode off the free list, or returns NULL on failure | 382 | ** pulls a cnode off the free list, or returns NULL on failure |
361 | */ | 383 | */ |
362 | static struct reiserfs_journal_cnode *get_cnode(struct super_block *p_s_sb) { | 384 | static struct reiserfs_journal_cnode *get_cnode(struct super_block *p_s_sb) |
363 | struct reiserfs_journal_cnode *cn ; | 385 | { |
364 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | 386 | struct reiserfs_journal_cnode *cn; |
365 | 387 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); | |
366 | reiserfs_check_lock_depth(p_s_sb, "get_cnode") ; | 388 | |
367 | 389 | reiserfs_check_lock_depth(p_s_sb, "get_cnode"); | |
368 | if (journal->j_cnode_free <= 0) { | 390 | |
369 | return NULL ; | 391 | if (journal->j_cnode_free <= 0) { |
370 | } | 392 | return NULL; |
371 | journal->j_cnode_used++ ; | 393 | } |
372 | journal->j_cnode_free-- ; | 394 | journal->j_cnode_used++; |
373 | cn = journal->j_cnode_free_list ; | 395 | journal->j_cnode_free--; |
374 | if (!cn) { | 396 | cn = journal->j_cnode_free_list; |
375 | return cn ; | 397 | if (!cn) { |
376 | } | 398 | return cn; |
377 | if (cn->next) { | 399 | } |
378 | cn->next->prev = NULL ; | 400 | if (cn->next) { |
379 | } | 401 | cn->next->prev = NULL; |
380 | journal->j_cnode_free_list = cn->next ; | 402 | } |
381 | memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; | 403 | journal->j_cnode_free_list = cn->next; |
382 | return cn ; | 404 | memset(cn, 0, sizeof(struct reiserfs_journal_cnode)); |
405 | return cn; | ||
383 | } | 406 | } |
384 | 407 | ||
385 | /* | 408 | /* |
386 | ** returns a cnode to the free list | 409 | ** returns a cnode to the free list |
387 | */ | 410 | */ |
388 | static void free_cnode(struct super_block *p_s_sb, struct reiserfs_journal_cnode *cn) { | 411 | static void free_cnode(struct super_block *p_s_sb, |
389 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | 412 | struct reiserfs_journal_cnode *cn) |
413 | { | ||
414 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); | ||
390 | 415 | ||
391 | reiserfs_check_lock_depth(p_s_sb, "free_cnode") ; | 416 | reiserfs_check_lock_depth(p_s_sb, "free_cnode"); |
392 | 417 | ||
393 | journal->j_cnode_used-- ; | 418 | journal->j_cnode_used--; |
394 | journal->j_cnode_free++ ; | 419 | journal->j_cnode_free++; |
395 | /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */ | 420 | /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */ |
396 | cn->next = journal->j_cnode_free_list ; | 421 | cn->next = journal->j_cnode_free_list; |
397 | if (journal->j_cnode_free_list) { | 422 | if (journal->j_cnode_free_list) { |
398 | journal->j_cnode_free_list->prev = cn ; | 423 | journal->j_cnode_free_list->prev = cn; |
399 | } | 424 | } |
400 | cn->prev = NULL ; /* not needed with the memset, but I might kill the memset, and forget to do this */ | 425 | cn->prev = NULL; /* not needed with the memset, but I might kill the memset, and forget to do this */ |
401 | journal->j_cnode_free_list = cn ; | 426 | journal->j_cnode_free_list = cn; |
402 | } | 427 | } |
403 | 428 | ||
404 | static void clear_prepared_bits(struct buffer_head *bh) { | 429 | static void clear_prepared_bits(struct buffer_head *bh) |
405 | clear_buffer_journal_prepared (bh); | 430 | { |
406 | clear_buffer_journal_restore_dirty (bh); | 431 | clear_buffer_journal_prepared(bh); |
432 | clear_buffer_journal_restore_dirty(bh); | ||
407 | } | 433 | } |
408 | 434 | ||
409 | /* utility function to force a BUG if it is called without the big | 435 | /* utility function to force a BUG if it is called without the big |
410 | ** kernel lock held. caller is the string printed just before calling BUG() | 436 | ** kernel lock held. caller is the string printed just before calling BUG() |
411 | */ | 437 | */ |
412 | void reiserfs_check_lock_depth(struct super_block *sb, char *caller) { | 438 | void reiserfs_check_lock_depth(struct super_block *sb, char *caller) |
439 | { | ||
413 | #ifdef CONFIG_SMP | 440 | #ifdef CONFIG_SMP |
414 | if (current->lock_depth < 0) { | 441 | if (current->lock_depth < 0) { |
415 | reiserfs_panic (sb, "%s called without kernel lock held", caller) ; | 442 | reiserfs_panic(sb, "%s called without kernel lock held", |
416 | } | 443 | caller); |
444 | } | ||
417 | #else | 445 | #else |
418 | ; | 446 | ; |
419 | #endif | 447 | #endif |
420 | } | 448 | } |
421 | 449 | ||
422 | /* return a cnode with same dev, block number and size in table, or null if not found */ | 450 | /* return a cnode with same dev, block number and size in table, or null if not found */ |
423 | static inline struct reiserfs_journal_cnode * | 451 | static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct |
424 | get_journal_hash_dev(struct super_block *sb, | 452 | super_block |
425 | struct reiserfs_journal_cnode **table, | 453 | *sb, |
426 | long bl) | 454 | struct |
455 | reiserfs_journal_cnode | ||
456 | **table, | ||
457 | long bl) | ||
427 | { | 458 | { |
428 | struct reiserfs_journal_cnode *cn ; | 459 | struct reiserfs_journal_cnode *cn; |
429 | cn = journal_hash(table, sb, bl) ; | 460 | cn = journal_hash(table, sb, bl); |
430 | while(cn) { | 461 | while (cn) { |
431 | if (cn->blocknr == bl && cn->sb == sb) | 462 | if (cn->blocknr == bl && cn->sb == sb) |
432 | return cn ; | 463 | return cn; |
433 | cn = cn->hnext ; | 464 | cn = cn->hnext; |
434 | } | 465 | } |
435 | return (struct reiserfs_journal_cnode *)0 ; | 466 | return (struct reiserfs_journal_cnode *)0; |
436 | } | 467 | } |
437 | 468 | ||
438 | /* | 469 | /* |
@@ -454,91 +485,103 @@ get_journal_hash_dev(struct super_block *sb, | |||
454 | ** | 485 | ** |
455 | */ | 486 | */ |
456 | int reiserfs_in_journal(struct super_block *p_s_sb, | 487 | int reiserfs_in_journal(struct super_block *p_s_sb, |
457 | int bmap_nr, int bit_nr, int search_all, | 488 | int bmap_nr, int bit_nr, int search_all, |
458 | b_blocknr_t *next_zero_bit) { | 489 | b_blocknr_t * next_zero_bit) |
459 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | 490 | { |
460 | struct reiserfs_journal_cnode *cn ; | 491 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); |
461 | struct reiserfs_list_bitmap *jb ; | 492 | struct reiserfs_journal_cnode *cn; |
462 | int i ; | 493 | struct reiserfs_list_bitmap *jb; |
463 | unsigned long bl; | 494 | int i; |
464 | 495 | unsigned long bl; | |
465 | *next_zero_bit = 0 ; /* always start this at zero. */ | 496 | |
466 | 497 | *next_zero_bit = 0; /* always start this at zero. */ | |
467 | PROC_INFO_INC( p_s_sb, journal.in_journal ); | 498 | |
468 | /* If we aren't doing a search_all, this is a metablock, and it will be logged before use. | 499 | PROC_INFO_INC(p_s_sb, journal.in_journal); |
469 | ** if we crash before the transaction that freed it commits, this transaction won't | 500 | /* If we aren't doing a search_all, this is a metablock, and it will be logged before use. |
470 | ** have committed either, and the block will never be written | 501 | ** if we crash before the transaction that freed it commits, this transaction won't |
471 | */ | 502 | ** have committed either, and the block will never be written |
472 | if (search_all) { | 503 | */ |
473 | for (i = 0 ; i < JOURNAL_NUM_BITMAPS ; i++) { | 504 | if (search_all) { |
474 | PROC_INFO_INC( p_s_sb, journal.in_journal_bitmap ); | 505 | for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) { |
475 | jb = journal->j_list_bitmap + i ; | 506 | PROC_INFO_INC(p_s_sb, journal.in_journal_bitmap); |
476 | if (jb->journal_list && jb->bitmaps[bmap_nr] && | 507 | jb = journal->j_list_bitmap + i; |
477 | test_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data)) { | 508 | if (jb->journal_list && jb->bitmaps[bmap_nr] && |
478 | *next_zero_bit = find_next_zero_bit((unsigned long *) | 509 | test_bit(bit_nr, |
479 | (jb->bitmaps[bmap_nr]->data), | 510 | (unsigned long *)jb->bitmaps[bmap_nr]-> |
480 | p_s_sb->s_blocksize << 3, bit_nr+1) ; | 511 | data)) { |
481 | return 1 ; | 512 | *next_zero_bit = |
482 | } | 513 | find_next_zero_bit((unsigned long *) |
483 | } | 514 | (jb->bitmaps[bmap_nr]-> |
484 | } | 515 | data), |
485 | 516 | p_s_sb->s_blocksize << 3, | |
486 | bl = bmap_nr * (p_s_sb->s_blocksize << 3) + bit_nr; | 517 | bit_nr + 1); |
487 | /* is it in any old transactions? */ | 518 | return 1; |
488 | if (search_all && (cn = get_journal_hash_dev(p_s_sb, journal->j_list_hash_table, bl))) { | 519 | } |
489 | return 1; | 520 | } |
490 | } | 521 | } |
491 | 522 | ||
492 | /* is it in the current transaction. This should never happen */ | 523 | bl = bmap_nr * (p_s_sb->s_blocksize << 3) + bit_nr; |
493 | if ((cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, bl))) { | 524 | /* is it in any old transactions? */ |
494 | BUG(); | 525 | if (search_all |
495 | return 1; | 526 | && (cn = |
496 | } | 527 | get_journal_hash_dev(p_s_sb, journal->j_list_hash_table, bl))) { |
497 | 528 | return 1; | |
498 | PROC_INFO_INC( p_s_sb, journal.in_journal_reusable ); | 529 | } |
499 | /* safe for reuse */ | 530 | |
500 | return 0 ; | 531 | /* is it in the current transaction. This should never happen */ |
532 | if ((cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, bl))) { | ||
533 | BUG(); | ||
534 | return 1; | ||
535 | } | ||
536 | |||
537 | PROC_INFO_INC(p_s_sb, journal.in_journal_reusable); | ||
538 | /* safe for reuse */ | ||
539 | return 0; | ||
501 | } | 540 | } |
502 | 541 | ||
503 | /* insert cn into table | 542 | /* insert cn into table |
504 | */ | 543 | */ |
505 | static inline void insert_journal_hash(struct reiserfs_journal_cnode **table, struct reiserfs_journal_cnode *cn) { | 544 | static inline void insert_journal_hash(struct reiserfs_journal_cnode **table, |
506 | struct reiserfs_journal_cnode *cn_orig ; | 545 | struct reiserfs_journal_cnode *cn) |
546 | { | ||
547 | struct reiserfs_journal_cnode *cn_orig; | ||
507 | 548 | ||
508 | cn_orig = journal_hash(table, cn->sb, cn->blocknr) ; | 549 | cn_orig = journal_hash(table, cn->sb, cn->blocknr); |
509 | cn->hnext = cn_orig ; | 550 | cn->hnext = cn_orig; |
510 | cn->hprev = NULL ; | 551 | cn->hprev = NULL; |
511 | if (cn_orig) { | 552 | if (cn_orig) { |
512 | cn_orig->hprev = cn ; | 553 | cn_orig->hprev = cn; |
513 | } | 554 | } |
514 | journal_hash(table, cn->sb, cn->blocknr) = cn ; | 555 | journal_hash(table, cn->sb, cn->blocknr) = cn; |
515 | } | 556 | } |
516 | 557 | ||
517 | /* lock the current transaction */ | 558 | /* lock the current transaction */ |
518 | inline static void lock_journal(struct super_block *p_s_sb) { | 559 | inline static void lock_journal(struct super_block *p_s_sb) |
519 | PROC_INFO_INC( p_s_sb, journal.lock_journal ); | 560 | { |
520 | down(&SB_JOURNAL(p_s_sb)->j_lock); | 561 | PROC_INFO_INC(p_s_sb, journal.lock_journal); |
562 | down(&SB_JOURNAL(p_s_sb)->j_lock); | ||
521 | } | 563 | } |
522 | 564 | ||
523 | /* unlock the current transaction */ | 565 | /* unlock the current transaction */ |
524 | inline static void unlock_journal(struct super_block *p_s_sb) { | 566 | inline static void unlock_journal(struct super_block *p_s_sb) |
525 | up(&SB_JOURNAL(p_s_sb)->j_lock); | 567 | { |
568 | up(&SB_JOURNAL(p_s_sb)->j_lock); | ||
526 | } | 569 | } |
527 | 570 | ||
528 | static inline void get_journal_list(struct reiserfs_journal_list *jl) | 571 | static inline void get_journal_list(struct reiserfs_journal_list *jl) |
529 | { | 572 | { |
530 | jl->j_refcount++; | 573 | jl->j_refcount++; |
531 | } | 574 | } |
532 | 575 | ||
533 | static inline void put_journal_list(struct super_block *s, | 576 | static inline void put_journal_list(struct super_block *s, |
534 | struct reiserfs_journal_list *jl) | 577 | struct reiserfs_journal_list *jl) |
535 | { | 578 | { |
536 | if (jl->j_refcount < 1) { | 579 | if (jl->j_refcount < 1) { |
537 | reiserfs_panic (s, "trans id %lu, refcount at %d", jl->j_trans_id, | 580 | reiserfs_panic(s, "trans id %lu, refcount at %d", |
538 | jl->j_refcount); | 581 | jl->j_trans_id, jl->j_refcount); |
539 | } | 582 | } |
540 | if (--jl->j_refcount == 0) | 583 | if (--jl->j_refcount == 0) |
541 | reiserfs_kfree(jl, sizeof(struct reiserfs_journal_list), s); | 584 | reiserfs_kfree(jl, sizeof(struct reiserfs_journal_list), s); |
542 | } | 585 | } |
543 | 586 | ||
544 | /* | 587 | /* |
@@ -546,358 +589,375 @@ static inline void put_journal_list(struct super_block *s, | |||
546 | ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a | 589 | ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a |
547 | ** transaction. | 590 | ** transaction. |
548 | */ | 591 | */ |
549 | static void cleanup_freed_for_journal_list(struct super_block *p_s_sb, struct reiserfs_journal_list *jl) { | 592 | static void cleanup_freed_for_journal_list(struct super_block *p_s_sb, |
593 | struct reiserfs_journal_list *jl) | ||
594 | { | ||
550 | 595 | ||
551 | struct reiserfs_list_bitmap *jb = jl->j_list_bitmap ; | 596 | struct reiserfs_list_bitmap *jb = jl->j_list_bitmap; |
552 | if (jb) { | 597 | if (jb) { |
553 | cleanup_bitmap_list(p_s_sb, jb) ; | 598 | cleanup_bitmap_list(p_s_sb, jb); |
554 | } | 599 | } |
555 | jl->j_list_bitmap->journal_list = NULL ; | 600 | jl->j_list_bitmap->journal_list = NULL; |
556 | jl->j_list_bitmap = NULL ; | 601 | jl->j_list_bitmap = NULL; |
557 | } | 602 | } |
558 | 603 | ||
559 | static int journal_list_still_alive(struct super_block *s, | 604 | static int journal_list_still_alive(struct super_block *s, |
560 | unsigned long trans_id) | 605 | unsigned long trans_id) |
561 | { | 606 | { |
562 | struct reiserfs_journal *journal = SB_JOURNAL (s); | 607 | struct reiserfs_journal *journal = SB_JOURNAL(s); |
563 | struct list_head *entry = &journal->j_journal_list; | 608 | struct list_head *entry = &journal->j_journal_list; |
564 | struct reiserfs_journal_list *jl; | 609 | struct reiserfs_journal_list *jl; |
565 | 610 | ||
566 | if (!list_empty(entry)) { | 611 | if (!list_empty(entry)) { |
567 | jl = JOURNAL_LIST_ENTRY(entry->next); | 612 | jl = JOURNAL_LIST_ENTRY(entry->next); |
568 | if (jl->j_trans_id <= trans_id) { | 613 | if (jl->j_trans_id <= trans_id) { |
569 | return 1; | 614 | return 1; |
570 | } | 615 | } |
571 | } | 616 | } |
572 | return 0; | 617 | return 0; |
573 | } | 618 | } |
574 | 619 | ||
575 | static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate) { | 620 | static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate) |
576 | char b[BDEVNAME_SIZE]; | 621 | { |
577 | 622 | char b[BDEVNAME_SIZE]; | |
578 | if (buffer_journaled(bh)) { | 623 | |
579 | reiserfs_warning(NULL, "clm-2084: pinned buffer %lu:%s sent to disk", | 624 | if (buffer_journaled(bh)) { |
580 | bh->b_blocknr, bdevname(bh->b_bdev, b)) ; | 625 | reiserfs_warning(NULL, |
581 | } | 626 | "clm-2084: pinned buffer %lu:%s sent to disk", |
582 | if (uptodate) | 627 | bh->b_blocknr, bdevname(bh->b_bdev, b)); |
583 | set_buffer_uptodate(bh) ; | 628 | } |
584 | else | 629 | if (uptodate) |
585 | clear_buffer_uptodate(bh) ; | 630 | set_buffer_uptodate(bh); |
586 | unlock_buffer(bh) ; | 631 | else |
587 | put_bh(bh) ; | 632 | clear_buffer_uptodate(bh); |
588 | } | 633 | unlock_buffer(bh); |
589 | 634 | put_bh(bh); | |
590 | static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate) { | 635 | } |
591 | if (uptodate) | 636 | |
592 | set_buffer_uptodate(bh) ; | 637 | static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate) |
593 | else | 638 | { |
594 | clear_buffer_uptodate(bh) ; | 639 | if (uptodate) |
595 | unlock_buffer(bh) ; | 640 | set_buffer_uptodate(bh); |
596 | put_bh(bh) ; | 641 | else |
597 | } | 642 | clear_buffer_uptodate(bh); |
598 | 643 | unlock_buffer(bh); | |
599 | static void submit_logged_buffer(struct buffer_head *bh) { | 644 | put_bh(bh); |
600 | get_bh(bh) ; | 645 | } |
601 | bh->b_end_io = reiserfs_end_buffer_io_sync ; | 646 | |
602 | clear_buffer_journal_new (bh); | 647 | static void submit_logged_buffer(struct buffer_head *bh) |
603 | clear_buffer_dirty(bh) ; | 648 | { |
604 | if (!test_clear_buffer_journal_test (bh)) | 649 | get_bh(bh); |
605 | BUG(); | 650 | bh->b_end_io = reiserfs_end_buffer_io_sync; |
606 | if (!buffer_uptodate(bh)) | 651 | clear_buffer_journal_new(bh); |
607 | BUG(); | 652 | clear_buffer_dirty(bh); |
608 | submit_bh(WRITE, bh) ; | 653 | if (!test_clear_buffer_journal_test(bh)) |
609 | } | 654 | BUG(); |
610 | 655 | if (!buffer_uptodate(bh)) | |
611 | static void submit_ordered_buffer(struct buffer_head *bh) { | 656 | BUG(); |
612 | get_bh(bh) ; | 657 | submit_bh(WRITE, bh); |
613 | bh->b_end_io = reiserfs_end_ordered_io; | 658 | } |
614 | clear_buffer_dirty(bh) ; | 659 | |
615 | if (!buffer_uptodate(bh)) | 660 | static void submit_ordered_buffer(struct buffer_head *bh) |
616 | BUG(); | 661 | { |
617 | submit_bh(WRITE, bh) ; | 662 | get_bh(bh); |
618 | } | 663 | bh->b_end_io = reiserfs_end_ordered_io; |
619 | 664 | clear_buffer_dirty(bh); | |
620 | static int submit_barrier_buffer(struct buffer_head *bh) { | 665 | if (!buffer_uptodate(bh)) |
621 | get_bh(bh) ; | 666 | BUG(); |
622 | bh->b_end_io = reiserfs_end_ordered_io; | 667 | submit_bh(WRITE, bh); |
623 | clear_buffer_dirty(bh) ; | 668 | } |
624 | if (!buffer_uptodate(bh)) | 669 | |
625 | BUG(); | 670 | static int submit_barrier_buffer(struct buffer_head *bh) |
626 | return submit_bh(WRITE_BARRIER, bh) ; | 671 | { |
672 | get_bh(bh); | ||
673 | bh->b_end_io = reiserfs_end_ordered_io; | ||
674 | clear_buffer_dirty(bh); | ||
675 | if (!buffer_uptodate(bh)) | ||
676 | BUG(); | ||
677 | return submit_bh(WRITE_BARRIER, bh); | ||
627 | } | 678 | } |
628 | 679 | ||
629 | static void check_barrier_completion(struct super_block *s, | 680 | static void check_barrier_completion(struct super_block *s, |
630 | struct buffer_head *bh) { | 681 | struct buffer_head *bh) |
631 | if (buffer_eopnotsupp(bh)) { | 682 | { |
632 | clear_buffer_eopnotsupp(bh); | 683 | if (buffer_eopnotsupp(bh)) { |
633 | disable_barrier(s); | 684 | clear_buffer_eopnotsupp(bh); |
634 | set_buffer_uptodate(bh); | 685 | disable_barrier(s); |
635 | set_buffer_dirty(bh); | 686 | set_buffer_uptodate(bh); |
636 | sync_dirty_buffer(bh); | 687 | set_buffer_dirty(bh); |
637 | } | 688 | sync_dirty_buffer(bh); |
689 | } | ||
638 | } | 690 | } |
639 | 691 | ||
640 | #define CHUNK_SIZE 32 | 692 | #define CHUNK_SIZE 32 |
641 | struct buffer_chunk { | 693 | struct buffer_chunk { |
642 | struct buffer_head *bh[CHUNK_SIZE]; | 694 | struct buffer_head *bh[CHUNK_SIZE]; |
643 | int nr; | 695 | int nr; |
644 | }; | 696 | }; |
645 | 697 | ||
646 | static void write_chunk(struct buffer_chunk *chunk) { | 698 | static void write_chunk(struct buffer_chunk *chunk) |
647 | int i; | 699 | { |
648 | get_fs_excl(); | 700 | int i; |
649 | for (i = 0; i < chunk->nr ; i++) { | 701 | get_fs_excl(); |
650 | submit_logged_buffer(chunk->bh[i]) ; | 702 | for (i = 0; i < chunk->nr; i++) { |
651 | } | 703 | submit_logged_buffer(chunk->bh[i]); |
652 | chunk->nr = 0; | 704 | } |
653 | put_fs_excl(); | 705 | chunk->nr = 0; |
706 | put_fs_excl(); | ||
654 | } | 707 | } |
655 | 708 | ||
656 | static void write_ordered_chunk(struct buffer_chunk *chunk) { | 709 | static void write_ordered_chunk(struct buffer_chunk *chunk) |
657 | int i; | 710 | { |
658 | get_fs_excl(); | 711 | int i; |
659 | for (i = 0; i < chunk->nr ; i++) { | 712 | get_fs_excl(); |
660 | submit_ordered_buffer(chunk->bh[i]) ; | 713 | for (i = 0; i < chunk->nr; i++) { |
661 | } | 714 | submit_ordered_buffer(chunk->bh[i]); |
662 | chunk->nr = 0; | 715 | } |
663 | put_fs_excl(); | 716 | chunk->nr = 0; |
717 | put_fs_excl(); | ||
664 | } | 718 | } |
665 | 719 | ||
666 | static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh, | 720 | static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh, |
667 | spinlock_t *lock, | 721 | spinlock_t * lock, void (fn) (struct buffer_chunk *)) |
668 | void (fn)(struct buffer_chunk *)) | ||
669 | { | 722 | { |
670 | int ret = 0; | 723 | int ret = 0; |
671 | if (chunk->nr >= CHUNK_SIZE) | 724 | if (chunk->nr >= CHUNK_SIZE) |
672 | BUG(); | 725 | BUG(); |
673 | chunk->bh[chunk->nr++] = bh; | 726 | chunk->bh[chunk->nr++] = bh; |
674 | if (chunk->nr >= CHUNK_SIZE) { | 727 | if (chunk->nr >= CHUNK_SIZE) { |
675 | ret = 1; | 728 | ret = 1; |
676 | if (lock) | 729 | if (lock) |
677 | spin_unlock(lock); | 730 | spin_unlock(lock); |
678 | fn(chunk); | 731 | fn(chunk); |
679 | if (lock) | 732 | if (lock) |
680 | spin_lock(lock); | 733 | spin_lock(lock); |
681 | } | 734 | } |
682 | return ret; | 735 | return ret; |
683 | } | 736 | } |
684 | 737 | ||
685 | |||
686 | static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0); | 738 | static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0); |
687 | static struct reiserfs_jh *alloc_jh(void) { | 739 | static struct reiserfs_jh *alloc_jh(void) |
688 | struct reiserfs_jh *jh; | 740 | { |
689 | while(1) { | 741 | struct reiserfs_jh *jh; |
690 | jh = kmalloc(sizeof(*jh), GFP_NOFS); | 742 | while (1) { |
691 | if (jh) { | 743 | jh = kmalloc(sizeof(*jh), GFP_NOFS); |
692 | atomic_inc(&nr_reiserfs_jh); | 744 | if (jh) { |
693 | return jh; | 745 | atomic_inc(&nr_reiserfs_jh); |
746 | return jh; | ||
747 | } | ||
748 | yield(); | ||
694 | } | 749 | } |
695 | yield(); | ||
696 | } | ||
697 | } | 750 | } |
698 | 751 | ||
699 | /* | 752 | /* |
700 | * we want to free the jh when the buffer has been written | 753 | * we want to free the jh when the buffer has been written |
701 | * and waited on | 754 | * and waited on |
702 | */ | 755 | */ |
703 | void reiserfs_free_jh(struct buffer_head *bh) { | 756 | void reiserfs_free_jh(struct buffer_head *bh) |
704 | struct reiserfs_jh *jh; | 757 | { |
705 | 758 | struct reiserfs_jh *jh; | |
706 | jh = bh->b_private; | 759 | |
707 | if (jh) { | 760 | jh = bh->b_private; |
708 | bh->b_private = NULL; | 761 | if (jh) { |
709 | jh->bh = NULL; | 762 | bh->b_private = NULL; |
710 | list_del_init(&jh->list); | 763 | jh->bh = NULL; |
711 | kfree(jh); | 764 | list_del_init(&jh->list); |
712 | if (atomic_read(&nr_reiserfs_jh) <= 0) | 765 | kfree(jh); |
713 | BUG(); | 766 | if (atomic_read(&nr_reiserfs_jh) <= 0) |
714 | atomic_dec(&nr_reiserfs_jh); | 767 | BUG(); |
715 | put_bh(bh); | 768 | atomic_dec(&nr_reiserfs_jh); |
716 | } | 769 | put_bh(bh); |
770 | } | ||
717 | } | 771 | } |
718 | 772 | ||
719 | static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh, | 773 | static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh, |
720 | int tail) | 774 | int tail) |
721 | { | 775 | { |
722 | struct reiserfs_jh *jh; | 776 | struct reiserfs_jh *jh; |
723 | 777 | ||
724 | if (bh->b_private) { | 778 | if (bh->b_private) { |
725 | spin_lock(&j->j_dirty_buffers_lock); | 779 | spin_lock(&j->j_dirty_buffers_lock); |
726 | if (!bh->b_private) { | 780 | if (!bh->b_private) { |
727 | spin_unlock(&j->j_dirty_buffers_lock); | 781 | spin_unlock(&j->j_dirty_buffers_lock); |
728 | goto no_jh; | 782 | goto no_jh; |
783 | } | ||
784 | jh = bh->b_private; | ||
785 | list_del_init(&jh->list); | ||
786 | } else { | ||
787 | no_jh: | ||
788 | get_bh(bh); | ||
789 | jh = alloc_jh(); | ||
790 | spin_lock(&j->j_dirty_buffers_lock); | ||
791 | /* buffer must be locked for __add_jh, should be able to have | ||
792 | * two adds at the same time | ||
793 | */ | ||
794 | if (bh->b_private) | ||
795 | BUG(); | ||
796 | jh->bh = bh; | ||
797 | bh->b_private = jh; | ||
729 | } | 798 | } |
730 | jh = bh->b_private; | 799 | jh->jl = j->j_current_jl; |
731 | list_del_init(&jh->list); | 800 | if (tail) |
732 | } else { | 801 | list_add_tail(&jh->list, &jh->jl->j_tail_bh_list); |
733 | no_jh: | 802 | else { |
734 | get_bh(bh); | 803 | list_add_tail(&jh->list, &jh->jl->j_bh_list); |
735 | jh = alloc_jh(); | 804 | } |
736 | spin_lock(&j->j_dirty_buffers_lock); | 805 | spin_unlock(&j->j_dirty_buffers_lock); |
737 | /* buffer must be locked for __add_jh, should be able to have | 806 | return 0; |
738 | * two adds at the same time | ||
739 | */ | ||
740 | if (bh->b_private) | ||
741 | BUG(); | ||
742 | jh->bh = bh; | ||
743 | bh->b_private = jh; | ||
744 | } | ||
745 | jh->jl = j->j_current_jl; | ||
746 | if (tail) | ||
747 | list_add_tail(&jh->list, &jh->jl->j_tail_bh_list); | ||
748 | else { | ||
749 | list_add_tail(&jh->list, &jh->jl->j_bh_list); | ||
750 | } | ||
751 | spin_unlock(&j->j_dirty_buffers_lock); | ||
752 | return 0; | ||
753 | } | 807 | } |
754 | 808 | ||
755 | int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh) { | 809 | int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh) |
756 | return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1); | 810 | { |
811 | return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1); | ||
757 | } | 812 | } |
758 | int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh) { | 813 | int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh) |
759 | return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0); | 814 | { |
815 | return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0); | ||
760 | } | 816 | } |
761 | 817 | ||
762 | #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list) | 818 | #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list) |
763 | static int write_ordered_buffers(spinlock_t *lock, | 819 | static int write_ordered_buffers(spinlock_t * lock, |
764 | struct reiserfs_journal *j, | 820 | struct reiserfs_journal *j, |
765 | struct reiserfs_journal_list *jl, | 821 | struct reiserfs_journal_list *jl, |
766 | struct list_head *list) | 822 | struct list_head *list) |
767 | { | 823 | { |
768 | struct buffer_head *bh; | 824 | struct buffer_head *bh; |
769 | struct reiserfs_jh *jh; | 825 | struct reiserfs_jh *jh; |
770 | int ret = j->j_errno; | 826 | int ret = j->j_errno; |
771 | struct buffer_chunk chunk; | 827 | struct buffer_chunk chunk; |
772 | struct list_head tmp; | 828 | struct list_head tmp; |
773 | INIT_LIST_HEAD(&tmp); | 829 | INIT_LIST_HEAD(&tmp); |
774 | 830 | ||
775 | chunk.nr = 0; | 831 | chunk.nr = 0; |
776 | spin_lock(lock); | 832 | spin_lock(lock); |
777 | while(!list_empty(list)) { | 833 | while (!list_empty(list)) { |
778 | jh = JH_ENTRY(list->next); | 834 | jh = JH_ENTRY(list->next); |
779 | bh = jh->bh; | 835 | bh = jh->bh; |
780 | get_bh(bh); | 836 | get_bh(bh); |
781 | if (test_set_buffer_locked(bh)) { | 837 | if (test_set_buffer_locked(bh)) { |
782 | if (!buffer_dirty(bh)) { | 838 | if (!buffer_dirty(bh)) { |
783 | list_del_init(&jh->list); | 839 | list_del_init(&jh->list); |
784 | list_add(&jh->list, &tmp); | 840 | list_add(&jh->list, &tmp); |
785 | goto loop_next; | 841 | goto loop_next; |
786 | } | 842 | } |
787 | spin_unlock(lock); | 843 | spin_unlock(lock); |
788 | if (chunk.nr) | 844 | if (chunk.nr) |
845 | write_ordered_chunk(&chunk); | ||
846 | wait_on_buffer(bh); | ||
847 | cond_resched(); | ||
848 | spin_lock(lock); | ||
849 | goto loop_next; | ||
850 | } | ||
851 | if (buffer_dirty(bh)) { | ||
852 | list_del_init(&jh->list); | ||
853 | list_add(&jh->list, &tmp); | ||
854 | add_to_chunk(&chunk, bh, lock, write_ordered_chunk); | ||
855 | } else { | ||
856 | reiserfs_free_jh(bh); | ||
857 | unlock_buffer(bh); | ||
858 | } | ||
859 | loop_next: | ||
860 | put_bh(bh); | ||
861 | cond_resched_lock(lock); | ||
862 | } | ||
863 | if (chunk.nr) { | ||
864 | spin_unlock(lock); | ||
789 | write_ordered_chunk(&chunk); | 865 | write_ordered_chunk(&chunk); |
790 | wait_on_buffer(bh); | 866 | spin_lock(lock); |
791 | cond_resched(); | ||
792 | spin_lock(lock); | ||
793 | goto loop_next; | ||
794 | } | ||
795 | if (buffer_dirty(bh)) { | ||
796 | list_del_init(&jh->list); | ||
797 | list_add(&jh->list, &tmp); | ||
798 | add_to_chunk(&chunk, bh, lock, write_ordered_chunk); | ||
799 | } else { | ||
800 | reiserfs_free_jh(bh); | ||
801 | unlock_buffer(bh); | ||
802 | } | 867 | } |
803 | loop_next: | 868 | while (!list_empty(&tmp)) { |
804 | put_bh(bh); | 869 | jh = JH_ENTRY(tmp.prev); |
805 | cond_resched_lock(lock); | 870 | bh = jh->bh; |
806 | } | 871 | get_bh(bh); |
807 | if (chunk.nr) { | 872 | reiserfs_free_jh(bh); |
808 | spin_unlock(lock); | 873 | |
809 | write_ordered_chunk(&chunk); | 874 | if (buffer_locked(bh)) { |
810 | spin_lock(lock); | 875 | spin_unlock(lock); |
811 | } | 876 | wait_on_buffer(bh); |
812 | while(!list_empty(&tmp)) { | 877 | spin_lock(lock); |
813 | jh = JH_ENTRY(tmp.prev); | 878 | } |
814 | bh = jh->bh; | 879 | if (!buffer_uptodate(bh)) { |
815 | get_bh(bh); | 880 | ret = -EIO; |
816 | reiserfs_free_jh(bh); | 881 | } |
817 | 882 | put_bh(bh); | |
818 | if (buffer_locked(bh)) { | 883 | cond_resched_lock(lock); |
819 | spin_unlock(lock); | ||
820 | wait_on_buffer(bh); | ||
821 | spin_lock(lock); | ||
822 | } | 884 | } |
823 | if (!buffer_uptodate(bh)) { | 885 | spin_unlock(lock); |
824 | ret = -EIO; | 886 | return ret; |
825 | } | 887 | } |
826 | put_bh(bh); | ||
827 | cond_resched_lock(lock); | ||
828 | } | ||
829 | spin_unlock(lock); | ||
830 | return ret; | ||
831 | } | ||
832 | |||
833 | static int flush_older_commits(struct super_block *s, struct reiserfs_journal_list *jl) { | ||
834 | struct reiserfs_journal *journal = SB_JOURNAL (s); | ||
835 | struct reiserfs_journal_list *other_jl; | ||
836 | struct reiserfs_journal_list *first_jl; | ||
837 | struct list_head *entry; | ||
838 | unsigned long trans_id = jl->j_trans_id; | ||
839 | unsigned long other_trans_id; | ||
840 | unsigned long first_trans_id; | ||
841 | |||
842 | find_first: | ||
843 | /* | ||
844 | * first we walk backwards to find the oldest uncommitted transation | ||
845 | */ | ||
846 | first_jl = jl; | ||
847 | entry = jl->j_list.prev; | ||
848 | while(1) { | ||
849 | other_jl = JOURNAL_LIST_ENTRY(entry); | ||
850 | if (entry == &journal->j_journal_list || | ||
851 | atomic_read(&other_jl->j_older_commits_done)) | ||
852 | break; | ||
853 | |||
854 | first_jl = other_jl; | ||
855 | entry = other_jl->j_list.prev; | ||
856 | } | ||
857 | |||
858 | /* if we didn't find any older uncommitted transactions, return now */ | ||
859 | if (first_jl == jl) { | ||
860 | return 0; | ||
861 | } | ||
862 | |||
863 | first_trans_id = first_jl->j_trans_id; | ||
864 | 888 | ||
865 | entry = &first_jl->j_list; | 889 | static int flush_older_commits(struct super_block *s, |
866 | while(1) { | 890 | struct reiserfs_journal_list *jl) |
867 | other_jl = JOURNAL_LIST_ENTRY(entry); | 891 | { |
868 | other_trans_id = other_jl->j_trans_id; | 892 | struct reiserfs_journal *journal = SB_JOURNAL(s); |
893 | struct reiserfs_journal_list *other_jl; | ||
894 | struct reiserfs_journal_list *first_jl; | ||
895 | struct list_head *entry; | ||
896 | unsigned long trans_id = jl->j_trans_id; | ||
897 | unsigned long other_trans_id; | ||
898 | unsigned long first_trans_id; | ||
899 | |||
900 | find_first: | ||
901 | /* | ||
902 | * first we walk backwards to find the oldest uncommitted transation | ||
903 | */ | ||
904 | first_jl = jl; | ||
905 | entry = jl->j_list.prev; | ||
906 | while (1) { | ||
907 | other_jl = JOURNAL_LIST_ENTRY(entry); | ||
908 | if (entry == &journal->j_journal_list || | ||
909 | atomic_read(&other_jl->j_older_commits_done)) | ||
910 | break; | ||
869 | 911 | ||
870 | if (other_trans_id < trans_id) { | 912 | first_jl = other_jl; |
871 | if (atomic_read(&other_jl->j_commit_left) != 0) { | 913 | entry = other_jl->j_list.prev; |
872 | flush_commit_list(s, other_jl, 0); | 914 | } |
873 | 915 | ||
874 | /* list we were called with is gone, return */ | 916 | /* if we didn't find any older uncommitted transactions, return now */ |
875 | if (!journal_list_still_alive(s, trans_id)) | 917 | if (first_jl == jl) { |
876 | return 1; | 918 | return 0; |
919 | } | ||
877 | 920 | ||
878 | /* the one we just flushed is gone, this means all | 921 | first_trans_id = first_jl->j_trans_id; |
879 | * older lists are also gone, so first_jl is no longer | 922 | |
880 | * valid either. Go back to the beginning. | 923 | entry = &first_jl->j_list; |
881 | */ | 924 | while (1) { |
882 | if (!journal_list_still_alive(s, other_trans_id)) { | 925 | other_jl = JOURNAL_LIST_ENTRY(entry); |
883 | goto find_first; | 926 | other_trans_id = other_jl->j_trans_id; |
927 | |||
928 | if (other_trans_id < trans_id) { | ||
929 | if (atomic_read(&other_jl->j_commit_left) != 0) { | ||
930 | flush_commit_list(s, other_jl, 0); | ||
931 | |||
932 | /* list we were called with is gone, return */ | ||
933 | if (!journal_list_still_alive(s, trans_id)) | ||
934 | return 1; | ||
935 | |||
936 | /* the one we just flushed is gone, this means all | ||
937 | * older lists are also gone, so first_jl is no longer | ||
938 | * valid either. Go back to the beginning. | ||
939 | */ | ||
940 | if (!journal_list_still_alive | ||
941 | (s, other_trans_id)) { | ||
942 | goto find_first; | ||
943 | } | ||
944 | } | ||
945 | entry = entry->next; | ||
946 | if (entry == &journal->j_journal_list) | ||
947 | return 0; | ||
948 | } else { | ||
949 | return 0; | ||
884 | } | 950 | } |
885 | } | ||
886 | entry = entry->next; | ||
887 | if (entry == &journal->j_journal_list) | ||
888 | return 0; | ||
889 | } else { | ||
890 | return 0; | ||
891 | } | 951 | } |
892 | } | 952 | return 0; |
893 | return 0; | ||
894 | } | 953 | } |
895 | int reiserfs_async_progress_wait(struct super_block *s) { | 954 | int reiserfs_async_progress_wait(struct super_block *s) |
896 | DEFINE_WAIT(wait); | 955 | { |
897 | struct reiserfs_journal *j = SB_JOURNAL(s); | 956 | DEFINE_WAIT(wait); |
898 | if (atomic_read(&j->j_async_throttle)) | 957 | struct reiserfs_journal *j = SB_JOURNAL(s); |
899 | blk_congestion_wait(WRITE, HZ/10); | 958 | if (atomic_read(&j->j_async_throttle)) |
900 | return 0; | 959 | blk_congestion_wait(WRITE, HZ / 10); |
960 | return 0; | ||
901 | } | 961 | } |
902 | 962 | ||
903 | /* | 963 | /* |
@@ -907,212 +967,225 @@ int reiserfs_async_progress_wait(struct super_block *s) { | |||
907 | ** Before the commit block can by written, every other log block must be safely on disk | 967 | ** Before the commit block can by written, every other log block must be safely on disk |
908 | ** | 968 | ** |
909 | */ | 969 | */ |
910 | static int flush_commit_list(struct super_block *s, struct reiserfs_journal_list *jl, int flushall) { | 970 | static int flush_commit_list(struct super_block *s, |
911 | int i; | 971 | struct reiserfs_journal_list *jl, int flushall) |
912 | int bn ; | 972 | { |
913 | struct buffer_head *tbh = NULL ; | 973 | int i; |
914 | unsigned long trans_id = jl->j_trans_id; | 974 | int bn; |
915 | struct reiserfs_journal *journal = SB_JOURNAL (s); | 975 | struct buffer_head *tbh = NULL; |
916 | int barrier = 0; | 976 | unsigned long trans_id = jl->j_trans_id; |
917 | int retval = 0; | 977 | struct reiserfs_journal *journal = SB_JOURNAL(s); |
918 | 978 | int barrier = 0; | |
919 | reiserfs_check_lock_depth(s, "flush_commit_list") ; | 979 | int retval = 0; |
920 | 980 | ||
921 | if (atomic_read(&jl->j_older_commits_done)) { | 981 | reiserfs_check_lock_depth(s, "flush_commit_list"); |
922 | return 0 ; | 982 | |
923 | } | 983 | if (atomic_read(&jl->j_older_commits_done)) { |
924 | 984 | return 0; | |
925 | get_fs_excl(); | 985 | } |
926 | 986 | ||
927 | /* before we can put our commit blocks on disk, we have to make sure everyone older than | 987 | get_fs_excl(); |
928 | ** us is on disk too | 988 | |
929 | */ | 989 | /* before we can put our commit blocks on disk, we have to make sure everyone older than |
930 | BUG_ON (jl->j_len <= 0); | 990 | ** us is on disk too |
931 | BUG_ON (trans_id == journal->j_trans_id); | 991 | */ |
932 | 992 | BUG_ON(jl->j_len <= 0); | |
933 | get_journal_list(jl); | 993 | BUG_ON(trans_id == journal->j_trans_id); |
934 | if (flushall) { | 994 | |
935 | if (flush_older_commits(s, jl) == 1) { | 995 | get_journal_list(jl); |
936 | /* list disappeared during flush_older_commits. return */ | 996 | if (flushall) { |
937 | goto put_jl; | 997 | if (flush_older_commits(s, jl) == 1) { |
938 | } | 998 | /* list disappeared during flush_older_commits. return */ |
939 | } | 999 | goto put_jl; |
940 | 1000 | } | |
941 | /* make sure nobody is trying to flush this one at the same time */ | 1001 | } |
942 | down(&jl->j_commit_lock); | 1002 | |
943 | if (!journal_list_still_alive(s, trans_id)) { | 1003 | /* make sure nobody is trying to flush this one at the same time */ |
944 | up(&jl->j_commit_lock); | 1004 | down(&jl->j_commit_lock); |
945 | goto put_jl; | 1005 | if (!journal_list_still_alive(s, trans_id)) { |
946 | } | 1006 | up(&jl->j_commit_lock); |
947 | BUG_ON (jl->j_trans_id == 0); | 1007 | goto put_jl; |
948 | 1008 | } | |
949 | /* this commit is done, exit */ | 1009 | BUG_ON(jl->j_trans_id == 0); |
950 | if (atomic_read(&(jl->j_commit_left)) <= 0) { | 1010 | |
951 | if (flushall) { | 1011 | /* this commit is done, exit */ |
952 | atomic_set(&(jl->j_older_commits_done), 1) ; | 1012 | if (atomic_read(&(jl->j_commit_left)) <= 0) { |
953 | } | 1013 | if (flushall) { |
954 | up(&jl->j_commit_lock); | 1014 | atomic_set(&(jl->j_older_commits_done), 1); |
955 | goto put_jl; | 1015 | } |
956 | } | 1016 | up(&jl->j_commit_lock); |
957 | 1017 | goto put_jl; | |
958 | if (!list_empty(&jl->j_bh_list)) { | 1018 | } |
959 | unlock_kernel(); | 1019 | |
960 | write_ordered_buffers(&journal->j_dirty_buffers_lock, | 1020 | if (!list_empty(&jl->j_bh_list)) { |
961 | journal, jl, &jl->j_bh_list); | 1021 | unlock_kernel(); |
962 | lock_kernel(); | 1022 | write_ordered_buffers(&journal->j_dirty_buffers_lock, |
963 | } | 1023 | journal, jl, &jl->j_bh_list); |
964 | BUG_ON (!list_empty(&jl->j_bh_list)); | 1024 | lock_kernel(); |
965 | /* | 1025 | } |
966 | * for the description block and all the log blocks, submit any buffers | 1026 | BUG_ON(!list_empty(&jl->j_bh_list)); |
967 | * that haven't already reached the disk | 1027 | /* |
968 | */ | 1028 | * for the description block and all the log blocks, submit any buffers |
969 | atomic_inc(&journal->j_async_throttle); | 1029 | * that haven't already reached the disk |
970 | for (i = 0 ; i < (jl->j_len + 1) ; i++) { | 1030 | */ |
971 | bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start+i) % | 1031 | atomic_inc(&journal->j_async_throttle); |
972 | SB_ONDISK_JOURNAL_SIZE(s); | 1032 | for (i = 0; i < (jl->j_len + 1); i++) { |
973 | tbh = journal_find_get_block(s, bn) ; | 1033 | bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) % |
974 | if (buffer_dirty(tbh)) /* redundant, ll_rw_block() checks */ | 1034 | SB_ONDISK_JOURNAL_SIZE(s); |
975 | ll_rw_block(WRITE, 1, &tbh) ; | 1035 | tbh = journal_find_get_block(s, bn); |
976 | put_bh(tbh) ; | 1036 | if (buffer_dirty(tbh)) /* redundant, ll_rw_block() checks */ |
977 | } | 1037 | ll_rw_block(WRITE, 1, &tbh); |
978 | atomic_dec(&journal->j_async_throttle); | 1038 | put_bh(tbh); |
979 | 1039 | } | |
980 | /* wait on everything written so far before writing the commit | 1040 | atomic_dec(&journal->j_async_throttle); |
981 | * if we are in barrier mode, send the commit down now | 1041 | |
982 | */ | 1042 | /* wait on everything written so far before writing the commit |
983 | barrier = reiserfs_barrier_flush(s); | 1043 | * if we are in barrier mode, send the commit down now |
984 | if (barrier) { | 1044 | */ |
985 | int ret; | 1045 | barrier = reiserfs_barrier_flush(s); |
986 | lock_buffer(jl->j_commit_bh); | 1046 | if (barrier) { |
987 | ret = submit_barrier_buffer(jl->j_commit_bh); | 1047 | int ret; |
988 | if (ret == -EOPNOTSUPP) { | 1048 | lock_buffer(jl->j_commit_bh); |
989 | set_buffer_uptodate(jl->j_commit_bh); | 1049 | ret = submit_barrier_buffer(jl->j_commit_bh); |
990 | disable_barrier(s); | 1050 | if (ret == -EOPNOTSUPP) { |
991 | barrier = 0; | 1051 | set_buffer_uptodate(jl->j_commit_bh); |
992 | } | 1052 | disable_barrier(s); |
993 | } | 1053 | barrier = 0; |
994 | for (i = 0 ; i < (jl->j_len + 1) ; i++) { | 1054 | } |
995 | bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + | 1055 | } |
996 | (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s) ; | 1056 | for (i = 0; i < (jl->j_len + 1); i++) { |
997 | tbh = journal_find_get_block(s, bn) ; | 1057 | bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + |
998 | wait_on_buffer(tbh) ; | 1058 | (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s); |
999 | // since we're using ll_rw_blk above, it might have skipped over | 1059 | tbh = journal_find_get_block(s, bn); |
1000 | // a locked buffer. Double check here | 1060 | wait_on_buffer(tbh); |
1001 | // | 1061 | // since we're using ll_rw_blk above, it might have skipped over |
1002 | if (buffer_dirty(tbh)) /* redundant, sync_dirty_buffer() checks */ | 1062 | // a locked buffer. Double check here |
1003 | sync_dirty_buffer(tbh); | 1063 | // |
1004 | if (unlikely (!buffer_uptodate(tbh))) { | 1064 | if (buffer_dirty(tbh)) /* redundant, sync_dirty_buffer() checks */ |
1065 | sync_dirty_buffer(tbh); | ||
1066 | if (unlikely(!buffer_uptodate(tbh))) { | ||
1005 | #ifdef CONFIG_REISERFS_CHECK | 1067 | #ifdef CONFIG_REISERFS_CHECK |
1006 | reiserfs_warning(s, "journal-601, buffer write failed") ; | 1068 | reiserfs_warning(s, "journal-601, buffer write failed"); |
1007 | #endif | 1069 | #endif |
1008 | retval = -EIO; | 1070 | retval = -EIO; |
1009 | } | 1071 | } |
1010 | put_bh(tbh) ; /* once for journal_find_get_block */ | 1072 | put_bh(tbh); /* once for journal_find_get_block */ |
1011 | put_bh(tbh) ; /* once due to original getblk in do_journal_end */ | 1073 | put_bh(tbh); /* once due to original getblk in do_journal_end */ |
1012 | atomic_dec(&(jl->j_commit_left)) ; | 1074 | atomic_dec(&(jl->j_commit_left)); |
1013 | } | 1075 | } |
1014 | 1076 | ||
1015 | BUG_ON (atomic_read(&(jl->j_commit_left)) != 1); | 1077 | BUG_ON(atomic_read(&(jl->j_commit_left)) != 1); |
1016 | 1078 | ||
1017 | if (!barrier) { | 1079 | if (!barrier) { |
1018 | if (buffer_dirty(jl->j_commit_bh)) | 1080 | if (buffer_dirty(jl->j_commit_bh)) |
1019 | BUG(); | 1081 | BUG(); |
1020 | mark_buffer_dirty(jl->j_commit_bh) ; | 1082 | mark_buffer_dirty(jl->j_commit_bh); |
1021 | sync_dirty_buffer(jl->j_commit_bh) ; | 1083 | sync_dirty_buffer(jl->j_commit_bh); |
1022 | } else | 1084 | } else |
1023 | wait_on_buffer(jl->j_commit_bh); | 1085 | wait_on_buffer(jl->j_commit_bh); |
1024 | 1086 | ||
1025 | check_barrier_completion(s, jl->j_commit_bh); | 1087 | check_barrier_completion(s, jl->j_commit_bh); |
1026 | 1088 | ||
1027 | /* If there was a write error in the journal - we can't commit this | 1089 | /* If there was a write error in the journal - we can't commit this |
1028 | * transaction - it will be invalid and, if successful, will just end | 1090 | * transaction - it will be invalid and, if successful, will just end |
1029 | * up propogating the write error out to the filesystem. */ | 1091 | * up propogating the write error out to the filesystem. */ |
1030 | if (unlikely (!buffer_uptodate(jl->j_commit_bh))) { | 1092 | if (unlikely(!buffer_uptodate(jl->j_commit_bh))) { |
1031 | #ifdef CONFIG_REISERFS_CHECK | 1093 | #ifdef CONFIG_REISERFS_CHECK |
1032 | reiserfs_warning(s, "journal-615: buffer write failed") ; | 1094 | reiserfs_warning(s, "journal-615: buffer write failed"); |
1033 | #endif | 1095 | #endif |
1034 | retval = -EIO; | 1096 | retval = -EIO; |
1035 | } | 1097 | } |
1036 | bforget(jl->j_commit_bh) ; | 1098 | bforget(jl->j_commit_bh); |
1037 | if (journal->j_last_commit_id != 0 && | 1099 | if (journal->j_last_commit_id != 0 && |
1038 | (jl->j_trans_id - journal->j_last_commit_id) != 1) { | 1100 | (jl->j_trans_id - journal->j_last_commit_id) != 1) { |
1039 | reiserfs_warning(s, "clm-2200: last commit %lu, current %lu", | 1101 | reiserfs_warning(s, "clm-2200: last commit %lu, current %lu", |
1040 | journal->j_last_commit_id, | 1102 | journal->j_last_commit_id, jl->j_trans_id); |
1041 | jl->j_trans_id); | 1103 | } |
1042 | } | 1104 | journal->j_last_commit_id = jl->j_trans_id; |
1043 | journal->j_last_commit_id = jl->j_trans_id; | 1105 | |
1044 | 1106 | /* now, every commit block is on the disk. It is safe to allow blocks freed during this transaction to be reallocated */ | |
1045 | /* now, every commit block is on the disk. It is safe to allow blocks freed during this transaction to be reallocated */ | 1107 | cleanup_freed_for_journal_list(s, jl); |
1046 | cleanup_freed_for_journal_list(s, jl) ; | 1108 | |
1047 | 1109 | retval = retval ? retval : journal->j_errno; | |
1048 | retval = retval ? retval : journal->j_errno; | 1110 | |
1049 | 1111 | /* mark the metadata dirty */ | |
1050 | /* mark the metadata dirty */ | 1112 | if (!retval) |
1051 | if (!retval) | 1113 | dirty_one_transaction(s, jl); |
1052 | dirty_one_transaction(s, jl); | 1114 | atomic_dec(&(jl->j_commit_left)); |
1053 | atomic_dec(&(jl->j_commit_left)) ; | 1115 | |
1054 | 1116 | if (flushall) { | |
1055 | if (flushall) { | 1117 | atomic_set(&(jl->j_older_commits_done), 1); |
1056 | atomic_set(&(jl->j_older_commits_done), 1) ; | 1118 | } |
1057 | } | 1119 | up(&jl->j_commit_lock); |
1058 | up(&jl->j_commit_lock); | 1120 | put_jl: |
1059 | put_jl: | 1121 | put_journal_list(s, jl); |
1060 | put_journal_list(s, jl); | 1122 | |
1061 | 1123 | if (retval) | |
1062 | if (retval) | 1124 | reiserfs_abort(s, retval, "Journal write error in %s", |
1063 | reiserfs_abort (s, retval, "Journal write error in %s", __FUNCTION__); | 1125 | __FUNCTION__); |
1064 | put_fs_excl(); | 1126 | put_fs_excl(); |
1065 | return retval; | 1127 | return retval; |
1066 | } | 1128 | } |
1067 | 1129 | ||
1068 | /* | 1130 | /* |
1069 | ** flush_journal_list frequently needs to find a newer transaction for a given block. This does that, or | 1131 | ** flush_journal_list frequently needs to find a newer transaction for a given block. This does that, or |
1070 | ** returns NULL if it can't find anything | 1132 | ** returns NULL if it can't find anything |
1071 | */ | 1133 | */ |
1072 | static struct reiserfs_journal_list *find_newer_jl_for_cn(struct reiserfs_journal_cnode *cn) { | 1134 | static struct reiserfs_journal_list *find_newer_jl_for_cn(struct |
1073 | struct super_block *sb = cn->sb; | 1135 | reiserfs_journal_cnode |
1074 | b_blocknr_t blocknr = cn->blocknr ; | 1136 | *cn) |
1137 | { | ||
1138 | struct super_block *sb = cn->sb; | ||
1139 | b_blocknr_t blocknr = cn->blocknr; | ||
1075 | 1140 | ||
1076 | cn = cn->hprev ; | 1141 | cn = cn->hprev; |
1077 | while(cn) { | 1142 | while (cn) { |
1078 | if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) { | 1143 | if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) { |
1079 | return cn->jlist ; | 1144 | return cn->jlist; |
1080 | } | 1145 | } |
1081 | cn = cn->hprev ; | 1146 | cn = cn->hprev; |
1082 | } | 1147 | } |
1083 | return NULL ; | 1148 | return NULL; |
1084 | } | 1149 | } |
1085 | 1150 | ||
1086 | static void remove_journal_hash(struct super_block *, struct reiserfs_journal_cnode **, | 1151 | static void remove_journal_hash(struct super_block *, |
1087 | struct reiserfs_journal_list *, unsigned long, int); | 1152 | struct reiserfs_journal_cnode **, |
1153 | struct reiserfs_journal_list *, unsigned long, | ||
1154 | int); | ||
1088 | 1155 | ||
1089 | /* | 1156 | /* |
1090 | ** once all the real blocks have been flushed, it is safe to remove them from the | 1157 | ** once all the real blocks have been flushed, it is safe to remove them from the |
1091 | ** journal list for this transaction. Aside from freeing the cnode, this also allows the | 1158 | ** journal list for this transaction. Aside from freeing the cnode, this also allows the |
1092 | ** block to be reallocated for data blocks if it had been deleted. | 1159 | ** block to be reallocated for data blocks if it had been deleted. |
1093 | */ | 1160 | */ |
1094 | static void remove_all_from_journal_list(struct super_block *p_s_sb, struct reiserfs_journal_list *jl, int debug) { | 1161 | static void remove_all_from_journal_list(struct super_block *p_s_sb, |
1095 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | 1162 | struct reiserfs_journal_list *jl, |
1096 | struct reiserfs_journal_cnode *cn, *last ; | 1163 | int debug) |
1097 | cn = jl->j_realblock ; | 1164 | { |
1098 | 1165 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); | |
1099 | /* which is better, to lock once around the whole loop, or | 1166 | struct reiserfs_journal_cnode *cn, *last; |
1100 | ** to lock for each call to remove_journal_hash? | 1167 | cn = jl->j_realblock; |
1101 | */ | 1168 | |
1102 | while(cn) { | 1169 | /* which is better, to lock once around the whole loop, or |
1103 | if (cn->blocknr != 0) { | 1170 | ** to lock for each call to remove_journal_hash? |
1104 | if (debug) { | 1171 | */ |
1105 | reiserfs_warning (p_s_sb, "block %u, bh is %d, state %ld", cn->blocknr, | 1172 | while (cn) { |
1106 | cn->bh ? 1: 0, cn->state) ; | 1173 | if (cn->blocknr != 0) { |
1107 | } | 1174 | if (debug) { |
1108 | cn->state = 0 ; | 1175 | reiserfs_warning(p_s_sb, |
1109 | remove_journal_hash(p_s_sb, journal->j_list_hash_table, jl, cn->blocknr, 1) ; | 1176 | "block %u, bh is %d, state %ld", |
1110 | } | 1177 | cn->blocknr, cn->bh ? 1 : 0, |
1111 | last = cn ; | 1178 | cn->state); |
1112 | cn = cn->next ; | 1179 | } |
1113 | free_cnode(p_s_sb, last) ; | 1180 | cn->state = 0; |
1114 | } | 1181 | remove_journal_hash(p_s_sb, journal->j_list_hash_table, |
1115 | jl->j_realblock = NULL ; | 1182 | jl, cn->blocknr, 1); |
1183 | } | ||
1184 | last = cn; | ||
1185 | cn = cn->next; | ||
1186 | free_cnode(p_s_sb, last); | ||
1187 | } | ||
1188 | jl->j_realblock = NULL; | ||
1116 | } | 1189 | } |
1117 | 1190 | ||
1118 | /* | 1191 | /* |
@@ -1122,98 +1195,107 @@ static void remove_all_from_journal_list(struct super_block *p_s_sb, struct reis | |||
1122 | ** called by flush_journal_list, before it calls remove_all_from_journal_list | 1195 | ** called by flush_journal_list, before it calls remove_all_from_journal_list |
1123 | ** | 1196 | ** |
1124 | */ | 1197 | */ |
1125 | static int _update_journal_header_block(struct super_block *p_s_sb, unsigned long offset, unsigned long trans_id) { | 1198 | static int _update_journal_header_block(struct super_block *p_s_sb, |
1126 | struct reiserfs_journal_header *jh ; | 1199 | unsigned long offset, |
1127 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | 1200 | unsigned long trans_id) |
1201 | { | ||
1202 | struct reiserfs_journal_header *jh; | ||
1203 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); | ||
1128 | 1204 | ||
1129 | if (reiserfs_is_journal_aborted (journal)) | 1205 | if (reiserfs_is_journal_aborted(journal)) |
1130 | return -EIO; | 1206 | return -EIO; |
1131 | 1207 | ||
1132 | if (trans_id >= journal->j_last_flush_trans_id) { | 1208 | if (trans_id >= journal->j_last_flush_trans_id) { |
1133 | if (buffer_locked((journal->j_header_bh))) { | 1209 | if (buffer_locked((journal->j_header_bh))) { |
1134 | wait_on_buffer((journal->j_header_bh)) ; | 1210 | wait_on_buffer((journal->j_header_bh)); |
1135 | if (unlikely (!buffer_uptodate(journal->j_header_bh))) { | 1211 | if (unlikely(!buffer_uptodate(journal->j_header_bh))) { |
1136 | #ifdef CONFIG_REISERFS_CHECK | 1212 | #ifdef CONFIG_REISERFS_CHECK |
1137 | reiserfs_warning (p_s_sb, "journal-699: buffer write failed") ; | 1213 | reiserfs_warning(p_s_sb, |
1214 | "journal-699: buffer write failed"); | ||
1138 | #endif | 1215 | #endif |
1139 | return -EIO; | 1216 | return -EIO; |
1140 | } | 1217 | } |
1141 | } | 1218 | } |
1142 | journal->j_last_flush_trans_id = trans_id ; | 1219 | journal->j_last_flush_trans_id = trans_id; |
1143 | journal->j_first_unflushed_offset = offset ; | 1220 | journal->j_first_unflushed_offset = offset; |
1144 | jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data) ; | 1221 | jh = (struct reiserfs_journal_header *)(journal->j_header_bh-> |
1145 | jh->j_last_flush_trans_id = cpu_to_le32(trans_id) ; | 1222 | b_data); |
1146 | jh->j_first_unflushed_offset = cpu_to_le32(offset) ; | 1223 | jh->j_last_flush_trans_id = cpu_to_le32(trans_id); |
1147 | jh->j_mount_id = cpu_to_le32(journal->j_mount_id) ; | 1224 | jh->j_first_unflushed_offset = cpu_to_le32(offset); |
1148 | 1225 | jh->j_mount_id = cpu_to_le32(journal->j_mount_id); | |
1149 | if (reiserfs_barrier_flush(p_s_sb)) { | 1226 | |
1150 | int ret; | 1227 | if (reiserfs_barrier_flush(p_s_sb)) { |
1151 | lock_buffer(journal->j_header_bh); | 1228 | int ret; |
1152 | ret = submit_barrier_buffer(journal->j_header_bh); | 1229 | lock_buffer(journal->j_header_bh); |
1153 | if (ret == -EOPNOTSUPP) { | 1230 | ret = submit_barrier_buffer(journal->j_header_bh); |
1154 | set_buffer_uptodate(journal->j_header_bh); | 1231 | if (ret == -EOPNOTSUPP) { |
1155 | disable_barrier(p_s_sb); | 1232 | set_buffer_uptodate(journal->j_header_bh); |
1156 | goto sync; | 1233 | disable_barrier(p_s_sb); |
1157 | } | 1234 | goto sync; |
1158 | wait_on_buffer(journal->j_header_bh); | 1235 | } |
1159 | check_barrier_completion(p_s_sb, journal->j_header_bh); | 1236 | wait_on_buffer(journal->j_header_bh); |
1160 | } else { | 1237 | check_barrier_completion(p_s_sb, journal->j_header_bh); |
1161 | sync: | 1238 | } else { |
1162 | set_buffer_dirty(journal->j_header_bh) ; | 1239 | sync: |
1163 | sync_dirty_buffer(journal->j_header_bh) ; | 1240 | set_buffer_dirty(journal->j_header_bh); |
1164 | } | 1241 | sync_dirty_buffer(journal->j_header_bh); |
1165 | if (!buffer_uptodate(journal->j_header_bh)) { | 1242 | } |
1166 | reiserfs_warning (p_s_sb, "journal-837: IO error during journal replay"); | 1243 | if (!buffer_uptodate(journal->j_header_bh)) { |
1167 | return -EIO ; | 1244 | reiserfs_warning(p_s_sb, |
1168 | } | 1245 | "journal-837: IO error during journal replay"); |
1169 | } | 1246 | return -EIO; |
1170 | return 0 ; | 1247 | } |
1171 | } | 1248 | } |
1172 | 1249 | return 0; | |
1173 | static int update_journal_header_block(struct super_block *p_s_sb, | ||
1174 | unsigned long offset, | ||
1175 | unsigned long trans_id) { | ||
1176 | return _update_journal_header_block(p_s_sb, offset, trans_id); | ||
1177 | } | 1250 | } |
1251 | |||
1252 | static int update_journal_header_block(struct super_block *p_s_sb, | ||
1253 | unsigned long offset, | ||
1254 | unsigned long trans_id) | ||
1255 | { | ||
1256 | return _update_journal_header_block(p_s_sb, offset, trans_id); | ||
1257 | } | ||
1258 | |||
1178 | /* | 1259 | /* |
1179 | ** flush any and all journal lists older than you are | 1260 | ** flush any and all journal lists older than you are |
1180 | ** can only be called from flush_journal_list | 1261 | ** can only be called from flush_journal_list |
1181 | */ | 1262 | */ |
1182 | static int flush_older_journal_lists(struct super_block *p_s_sb, | 1263 | static int flush_older_journal_lists(struct super_block *p_s_sb, |
1183 | struct reiserfs_journal_list *jl) | 1264 | struct reiserfs_journal_list *jl) |
1184 | { | 1265 | { |
1185 | struct list_head *entry; | 1266 | struct list_head *entry; |
1186 | struct reiserfs_journal_list *other_jl ; | 1267 | struct reiserfs_journal_list *other_jl; |
1187 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | 1268 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); |
1188 | unsigned long trans_id = jl->j_trans_id; | 1269 | unsigned long trans_id = jl->j_trans_id; |
1189 | 1270 | ||
1190 | /* we know we are the only ones flushing things, no extra race | 1271 | /* we know we are the only ones flushing things, no extra race |
1191 | * protection is required. | 1272 | * protection is required. |
1192 | */ | 1273 | */ |
1193 | restart: | 1274 | restart: |
1194 | entry = journal->j_journal_list.next; | 1275 | entry = journal->j_journal_list.next; |
1195 | /* Did we wrap? */ | 1276 | /* Did we wrap? */ |
1196 | if (entry == &journal->j_journal_list) | 1277 | if (entry == &journal->j_journal_list) |
1197 | return 0; | 1278 | return 0; |
1198 | other_jl = JOURNAL_LIST_ENTRY(entry); | 1279 | other_jl = JOURNAL_LIST_ENTRY(entry); |
1199 | if (other_jl->j_trans_id < trans_id) { | 1280 | if (other_jl->j_trans_id < trans_id) { |
1200 | BUG_ON (other_jl->j_refcount <= 0); | 1281 | BUG_ON(other_jl->j_refcount <= 0); |
1201 | /* do not flush all */ | 1282 | /* do not flush all */ |
1202 | flush_journal_list(p_s_sb, other_jl, 0) ; | 1283 | flush_journal_list(p_s_sb, other_jl, 0); |
1203 | 1284 | ||
1204 | /* other_jl is now deleted from the list */ | 1285 | /* other_jl is now deleted from the list */ |
1205 | goto restart; | 1286 | goto restart; |
1206 | } | 1287 | } |
1207 | return 0 ; | 1288 | return 0; |
1208 | } | 1289 | } |
1209 | 1290 | ||
1210 | static void del_from_work_list(struct super_block *s, | 1291 | static void del_from_work_list(struct super_block *s, |
1211 | struct reiserfs_journal_list *jl) { | 1292 | struct reiserfs_journal_list *jl) |
1212 | struct reiserfs_journal *journal = SB_JOURNAL (s); | 1293 | { |
1213 | if (!list_empty(&jl->j_working_list)) { | 1294 | struct reiserfs_journal *journal = SB_JOURNAL(s); |
1214 | list_del_init(&jl->j_working_list); | 1295 | if (!list_empty(&jl->j_working_list)) { |
1215 | journal->j_num_work_lists--; | 1296 | list_del_init(&jl->j_working_list); |
1216 | } | 1297 | journal->j_num_work_lists--; |
1298 | } | ||
1217 | } | 1299 | } |
1218 | 1300 | ||
1219 | /* flush a journal list, both commit and real blocks | 1301 | /* flush a journal list, both commit and real blocks |
@@ -1225,386 +1307,407 @@ static void del_from_work_list(struct super_block *s, | |||
1225 | ** and the journal is locked. That means it can only be called from | 1307 | ** and the journal is locked. That means it can only be called from |
1226 | ** do_journal_end, or by journal_release | 1308 | ** do_journal_end, or by journal_release |
1227 | */ | 1309 | */ |
1228 | static int flush_journal_list(struct super_block *s, | 1310 | static int flush_journal_list(struct super_block *s, |
1229 | struct reiserfs_journal_list *jl, int flushall) { | 1311 | struct reiserfs_journal_list *jl, int flushall) |
1230 | struct reiserfs_journal_list *pjl ; | 1312 | { |
1231 | struct reiserfs_journal_cnode *cn, *last ; | 1313 | struct reiserfs_journal_list *pjl; |
1232 | int count ; | 1314 | struct reiserfs_journal_cnode *cn, *last; |
1233 | int was_jwait = 0 ; | 1315 | int count; |
1234 | int was_dirty = 0 ; | 1316 | int was_jwait = 0; |
1235 | struct buffer_head *saved_bh ; | 1317 | int was_dirty = 0; |
1236 | unsigned long j_len_saved = jl->j_len ; | 1318 | struct buffer_head *saved_bh; |
1237 | struct reiserfs_journal *journal = SB_JOURNAL (s); | 1319 | unsigned long j_len_saved = jl->j_len; |
1238 | int err = 0; | 1320 | struct reiserfs_journal *journal = SB_JOURNAL(s); |
1239 | 1321 | int err = 0; | |
1240 | BUG_ON (j_len_saved <= 0); | 1322 | |
1241 | 1323 | BUG_ON(j_len_saved <= 0); | |
1242 | if (atomic_read(&journal->j_wcount) != 0) { | 1324 | |
1243 | reiserfs_warning(s, "clm-2048: flush_journal_list called with wcount %d", | 1325 | if (atomic_read(&journal->j_wcount) != 0) { |
1244 | atomic_read(&journal->j_wcount)) ; | 1326 | reiserfs_warning(s, |
1245 | } | 1327 | "clm-2048: flush_journal_list called with wcount %d", |
1246 | BUG_ON (jl->j_trans_id == 0); | 1328 | atomic_read(&journal->j_wcount)); |
1247 | 1329 | } | |
1248 | /* if flushall == 0, the lock is already held */ | 1330 | BUG_ON(jl->j_trans_id == 0); |
1249 | if (flushall) { | 1331 | |
1250 | down(&journal->j_flush_sem); | 1332 | /* if flushall == 0, the lock is already held */ |
1251 | } else if (!down_trylock(&journal->j_flush_sem)) { | 1333 | if (flushall) { |
1252 | BUG(); | 1334 | down(&journal->j_flush_sem); |
1253 | } | 1335 | } else if (!down_trylock(&journal->j_flush_sem)) { |
1254 | 1336 | BUG(); | |
1255 | count = 0 ; | 1337 | } |
1256 | if (j_len_saved > journal->j_trans_max) { | 1338 | |
1257 | reiserfs_panic(s, "journal-715: flush_journal_list, length is %lu, trans id %lu\n", j_len_saved, jl->j_trans_id); | 1339 | count = 0; |
1258 | return 0 ; | 1340 | if (j_len_saved > journal->j_trans_max) { |
1259 | } | 1341 | reiserfs_panic(s, |
1260 | 1342 | "journal-715: flush_journal_list, length is %lu, trans id %lu\n", | |
1261 | get_fs_excl(); | 1343 | j_len_saved, jl->j_trans_id); |
1262 | 1344 | return 0; | |
1263 | /* if all the work is already done, get out of here */ | 1345 | } |
1264 | if (atomic_read(&(jl->j_nonzerolen)) <= 0 && | 1346 | |
1265 | atomic_read(&(jl->j_commit_left)) <= 0) { | 1347 | get_fs_excl(); |
1266 | goto flush_older_and_return ; | 1348 | |
1267 | } | 1349 | /* if all the work is already done, get out of here */ |
1268 | 1350 | if (atomic_read(&(jl->j_nonzerolen)) <= 0 && | |
1269 | /* start by putting the commit list on disk. This will also flush | 1351 | atomic_read(&(jl->j_commit_left)) <= 0) { |
1270 | ** the commit lists of any olders transactions | 1352 | goto flush_older_and_return; |
1271 | */ | 1353 | } |
1272 | flush_commit_list(s, jl, 1) ; | 1354 | |
1273 | 1355 | /* start by putting the commit list on disk. This will also flush | |
1274 | if (!(jl->j_state & LIST_DIRTY) && !reiserfs_is_journal_aborted (journal)) | 1356 | ** the commit lists of any olders transactions |
1275 | BUG(); | 1357 | */ |
1276 | 1358 | flush_commit_list(s, jl, 1); | |
1277 | /* are we done now? */ | 1359 | |
1278 | if (atomic_read(&(jl->j_nonzerolen)) <= 0 && | 1360 | if (!(jl->j_state & LIST_DIRTY) |
1279 | atomic_read(&(jl->j_commit_left)) <= 0) { | 1361 | && !reiserfs_is_journal_aborted(journal)) |
1280 | goto flush_older_and_return ; | 1362 | BUG(); |
1281 | } | 1363 | |
1282 | 1364 | /* are we done now? */ | |
1283 | /* loop through each cnode, see if we need to write it, | 1365 | if (atomic_read(&(jl->j_nonzerolen)) <= 0 && |
1284 | ** or wait on a more recent transaction, or just ignore it | 1366 | atomic_read(&(jl->j_commit_left)) <= 0) { |
1285 | */ | 1367 | goto flush_older_and_return; |
1286 | if (atomic_read(&(journal->j_wcount)) != 0) { | 1368 | } |
1287 | reiserfs_panic(s, "journal-844: panic journal list is flushing, wcount is not 0\n") ; | 1369 | |
1288 | } | 1370 | /* loop through each cnode, see if we need to write it, |
1289 | cn = jl->j_realblock ; | 1371 | ** or wait on a more recent transaction, or just ignore it |
1290 | while(cn) { | 1372 | */ |
1291 | was_jwait = 0 ; | 1373 | if (atomic_read(&(journal->j_wcount)) != 0) { |
1292 | was_dirty = 0 ; | 1374 | reiserfs_panic(s, |
1293 | saved_bh = NULL ; | 1375 | "journal-844: panic journal list is flushing, wcount is not 0\n"); |
1294 | /* blocknr of 0 is no longer in the hash, ignore it */ | 1376 | } |
1295 | if (cn->blocknr == 0) { | 1377 | cn = jl->j_realblock; |
1296 | goto free_cnode ; | 1378 | while (cn) { |
1297 | } | 1379 | was_jwait = 0; |
1298 | 1380 | was_dirty = 0; | |
1299 | /* This transaction failed commit. Don't write out to the disk */ | 1381 | saved_bh = NULL; |
1300 | if (!(jl->j_state & LIST_DIRTY)) | 1382 | /* blocknr of 0 is no longer in the hash, ignore it */ |
1301 | goto free_cnode; | 1383 | if (cn->blocknr == 0) { |
1302 | 1384 | goto free_cnode; | |
1303 | pjl = find_newer_jl_for_cn(cn) ; | 1385 | } |
1304 | /* the order is important here. We check pjl to make sure we | 1386 | |
1305 | ** don't clear BH_JDirty_wait if we aren't the one writing this | 1387 | /* This transaction failed commit. Don't write out to the disk */ |
1306 | ** block to disk | 1388 | if (!(jl->j_state & LIST_DIRTY)) |
1307 | */ | 1389 | goto free_cnode; |
1308 | if (!pjl && cn->bh) { | 1390 | |
1309 | saved_bh = cn->bh ; | 1391 | pjl = find_newer_jl_for_cn(cn); |
1310 | 1392 | /* the order is important here. We check pjl to make sure we | |
1311 | /* we do this to make sure nobody releases the buffer while | 1393 | ** don't clear BH_JDirty_wait if we aren't the one writing this |
1312 | ** we are working with it | 1394 | ** block to disk |
1313 | */ | 1395 | */ |
1314 | get_bh(saved_bh) ; | 1396 | if (!pjl && cn->bh) { |
1315 | 1397 | saved_bh = cn->bh; | |
1316 | if (buffer_journal_dirty(saved_bh)) { | 1398 | |
1317 | BUG_ON (!can_dirty (cn)); | 1399 | /* we do this to make sure nobody releases the buffer while |
1318 | was_jwait = 1 ; | 1400 | ** we are working with it |
1319 | was_dirty = 1 ; | 1401 | */ |
1320 | } else if (can_dirty(cn)) { | 1402 | get_bh(saved_bh); |
1321 | /* everything with !pjl && jwait should be writable */ | 1403 | |
1322 | BUG(); | 1404 | if (buffer_journal_dirty(saved_bh)) { |
1323 | } | 1405 | BUG_ON(!can_dirty(cn)); |
1324 | } | 1406 | was_jwait = 1; |
1325 | 1407 | was_dirty = 1; | |
1326 | /* if someone has this block in a newer transaction, just make | 1408 | } else if (can_dirty(cn)) { |
1327 | ** sure they are commited, and don't try writing it to disk | 1409 | /* everything with !pjl && jwait should be writable */ |
1328 | */ | 1410 | BUG(); |
1329 | if (pjl) { | 1411 | } |
1330 | if (atomic_read(&pjl->j_commit_left)) | 1412 | } |
1331 | flush_commit_list(s, pjl, 1) ; | 1413 | |
1332 | goto free_cnode ; | 1414 | /* if someone has this block in a newer transaction, just make |
1333 | } | 1415 | ** sure they are commited, and don't try writing it to disk |
1334 | 1416 | */ | |
1335 | /* bh == NULL when the block got to disk on its own, OR, | 1417 | if (pjl) { |
1336 | ** the block got freed in a future transaction | 1418 | if (atomic_read(&pjl->j_commit_left)) |
1337 | */ | 1419 | flush_commit_list(s, pjl, 1); |
1338 | if (saved_bh == NULL) { | 1420 | goto free_cnode; |
1339 | goto free_cnode ; | 1421 | } |
1340 | } | 1422 | |
1341 | 1423 | /* bh == NULL when the block got to disk on its own, OR, | |
1342 | /* this should never happen. kupdate_one_transaction has this list | 1424 | ** the block got freed in a future transaction |
1343 | ** locked while it works, so we should never see a buffer here that | 1425 | */ |
1344 | ** is not marked JDirty_wait | 1426 | if (saved_bh == NULL) { |
1345 | */ | 1427 | goto free_cnode; |
1346 | if ((!was_jwait) && !buffer_locked(saved_bh)) { | 1428 | } |
1347 | reiserfs_warning (s, "journal-813: BAD! buffer %llu %cdirty %cjwait, " | 1429 | |
1348 | "not in a newer tranasction", | 1430 | /* this should never happen. kupdate_one_transaction has this list |
1349 | (unsigned long long)saved_bh->b_blocknr, | 1431 | ** locked while it works, so we should never see a buffer here that |
1350 | was_dirty ? ' ' : '!', was_jwait ? ' ' : '!') ; | 1432 | ** is not marked JDirty_wait |
1351 | } | 1433 | */ |
1352 | if (was_dirty) { | 1434 | if ((!was_jwait) && !buffer_locked(saved_bh)) { |
1353 | /* we inc again because saved_bh gets decremented at free_cnode */ | 1435 | reiserfs_warning(s, |
1354 | get_bh(saved_bh) ; | 1436 | "journal-813: BAD! buffer %llu %cdirty %cjwait, " |
1355 | set_bit(BLOCK_NEEDS_FLUSH, &cn->state) ; | 1437 | "not in a newer tranasction", |
1356 | lock_buffer(saved_bh); | 1438 | (unsigned long long)saved_bh-> |
1357 | BUG_ON (cn->blocknr != saved_bh->b_blocknr); | 1439 | b_blocknr, was_dirty ? ' ' : '!', |
1358 | if (buffer_dirty(saved_bh)) | 1440 | was_jwait ? ' ' : '!'); |
1359 | submit_logged_buffer(saved_bh) ; | 1441 | } |
1360 | else | 1442 | if (was_dirty) { |
1361 | unlock_buffer(saved_bh); | 1443 | /* we inc again because saved_bh gets decremented at free_cnode */ |
1362 | count++ ; | 1444 | get_bh(saved_bh); |
1363 | } else { | 1445 | set_bit(BLOCK_NEEDS_FLUSH, &cn->state); |
1364 | reiserfs_warning (s, "clm-2082: Unable to flush buffer %llu in %s", | 1446 | lock_buffer(saved_bh); |
1365 | (unsigned long long)saved_bh->b_blocknr, __FUNCTION__); | 1447 | BUG_ON(cn->blocknr != saved_bh->b_blocknr); |
1366 | } | 1448 | if (buffer_dirty(saved_bh)) |
1367 | free_cnode: | 1449 | submit_logged_buffer(saved_bh); |
1368 | last = cn ; | 1450 | else |
1369 | cn = cn->next ; | 1451 | unlock_buffer(saved_bh); |
1370 | if (saved_bh) { | 1452 | count++; |
1371 | /* we incremented this to keep others from taking the buffer head away */ | 1453 | } else { |
1372 | put_bh(saved_bh) ; | 1454 | reiserfs_warning(s, |
1373 | if (atomic_read(&(saved_bh->b_count)) < 0) { | 1455 | "clm-2082: Unable to flush buffer %llu in %s", |
1374 | reiserfs_warning (s, "journal-945: saved_bh->b_count < 0"); | 1456 | (unsigned long long)saved_bh-> |
1375 | } | 1457 | b_blocknr, __FUNCTION__); |
1376 | } | 1458 | } |
1377 | } | 1459 | free_cnode: |
1378 | if (count > 0) { | 1460 | last = cn; |
1379 | cn = jl->j_realblock ; | 1461 | cn = cn->next; |
1380 | while(cn) { | 1462 | if (saved_bh) { |
1381 | if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) { | 1463 | /* we incremented this to keep others from taking the buffer head away */ |
1382 | if (!cn->bh) { | 1464 | put_bh(saved_bh); |
1383 | reiserfs_panic(s, "journal-1011: cn->bh is NULL\n") ; | 1465 | if (atomic_read(&(saved_bh->b_count)) < 0) { |
1384 | } | 1466 | reiserfs_warning(s, |
1385 | wait_on_buffer(cn->bh) ; | 1467 | "journal-945: saved_bh->b_count < 0"); |
1386 | if (!cn->bh) { | 1468 | } |
1387 | reiserfs_panic(s, "journal-1012: cn->bh is NULL\n") ; | 1469 | } |
1388 | } | 1470 | } |
1389 | if (unlikely (!buffer_uptodate(cn->bh))) { | 1471 | if (count > 0) { |
1472 | cn = jl->j_realblock; | ||
1473 | while (cn) { | ||
1474 | if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) { | ||
1475 | if (!cn->bh) { | ||
1476 | reiserfs_panic(s, | ||
1477 | "journal-1011: cn->bh is NULL\n"); | ||
1478 | } | ||
1479 | wait_on_buffer(cn->bh); | ||
1480 | if (!cn->bh) { | ||
1481 | reiserfs_panic(s, | ||
1482 | "journal-1012: cn->bh is NULL\n"); | ||
1483 | } | ||
1484 | if (unlikely(!buffer_uptodate(cn->bh))) { | ||
1390 | #ifdef CONFIG_REISERFS_CHECK | 1485 | #ifdef CONFIG_REISERFS_CHECK |
1391 | reiserfs_warning(s, "journal-949: buffer write failed\n") ; | 1486 | reiserfs_warning(s, |
1487 | "journal-949: buffer write failed\n"); | ||
1392 | #endif | 1488 | #endif |
1393 | err = -EIO; | 1489 | err = -EIO; |
1394 | } | 1490 | } |
1395 | /* note, we must clear the JDirty_wait bit after the up to date | 1491 | /* note, we must clear the JDirty_wait bit after the up to date |
1396 | ** check, otherwise we race against our flushpage routine | 1492 | ** check, otherwise we race against our flushpage routine |
1397 | */ | 1493 | */ |
1398 | BUG_ON (!test_clear_buffer_journal_dirty (cn->bh)); | 1494 | BUG_ON(!test_clear_buffer_journal_dirty |
1399 | 1495 | (cn->bh)); | |
1400 | /* undo the inc from journal_mark_dirty */ | 1496 | |
1401 | put_bh(cn->bh) ; | 1497 | /* undo the inc from journal_mark_dirty */ |
1402 | brelse(cn->bh) ; | 1498 | put_bh(cn->bh); |
1403 | } | 1499 | brelse(cn->bh); |
1404 | cn = cn->next ; | 1500 | } |
1405 | } | 1501 | cn = cn->next; |
1406 | } | 1502 | } |
1407 | 1503 | } | |
1408 | if (err) | 1504 | |
1409 | reiserfs_abort (s, -EIO, "Write error while pushing transaction to disk in %s", __FUNCTION__); | 1505 | if (err) |
1410 | flush_older_and_return: | 1506 | reiserfs_abort(s, -EIO, |
1411 | 1507 | "Write error while pushing transaction to disk in %s", | |
1412 | 1508 | __FUNCTION__); | |
1413 | /* before we can update the journal header block, we _must_ flush all | 1509 | flush_older_and_return: |
1414 | ** real blocks from all older transactions to disk. This is because | 1510 | |
1415 | ** once the header block is updated, this transaction will not be | 1511 | /* before we can update the journal header block, we _must_ flush all |
1416 | ** replayed after a crash | 1512 | ** real blocks from all older transactions to disk. This is because |
1417 | */ | 1513 | ** once the header block is updated, this transaction will not be |
1418 | if (flushall) { | 1514 | ** replayed after a crash |
1419 | flush_older_journal_lists(s, jl); | 1515 | */ |
1420 | } | 1516 | if (flushall) { |
1421 | 1517 | flush_older_journal_lists(s, jl); | |
1422 | err = journal->j_errno; | 1518 | } |
1423 | /* before we can remove everything from the hash tables for this | 1519 | |
1424 | ** transaction, we must make sure it can never be replayed | 1520 | err = journal->j_errno; |
1425 | ** | 1521 | /* before we can remove everything from the hash tables for this |
1426 | ** since we are only called from do_journal_end, we know for sure there | 1522 | ** transaction, we must make sure it can never be replayed |
1427 | ** are no allocations going on while we are flushing journal lists. So, | 1523 | ** |
1428 | ** we only need to update the journal header block for the last list | 1524 | ** since we are only called from do_journal_end, we know for sure there |
1429 | ** being flushed | 1525 | ** are no allocations going on while we are flushing journal lists. So, |
1430 | */ | 1526 | ** we only need to update the journal header block for the last list |
1431 | if (!err && flushall) { | 1527 | ** being flushed |
1432 | err = update_journal_header_block(s, (jl->j_start + jl->j_len + 2) % SB_ONDISK_JOURNAL_SIZE(s), jl->j_trans_id) ; | 1528 | */ |
1433 | if (err) | 1529 | if (!err && flushall) { |
1434 | reiserfs_abort (s, -EIO, "Write error while updating journal header in %s", __FUNCTION__); | 1530 | err = |
1435 | } | 1531 | update_journal_header_block(s, |
1436 | remove_all_from_journal_list(s, jl, 0) ; | 1532 | (jl->j_start + jl->j_len + |
1437 | list_del_init(&jl->j_list); | 1533 | 2) % SB_ONDISK_JOURNAL_SIZE(s), |
1438 | journal->j_num_lists--; | 1534 | jl->j_trans_id); |
1439 | del_from_work_list(s, jl); | 1535 | if (err) |
1440 | 1536 | reiserfs_abort(s, -EIO, | |
1441 | if (journal->j_last_flush_id != 0 && | 1537 | "Write error while updating journal header in %s", |
1442 | (jl->j_trans_id - journal->j_last_flush_id) != 1) { | 1538 | __FUNCTION__); |
1443 | reiserfs_warning(s, "clm-2201: last flush %lu, current %lu", | 1539 | } |
1444 | journal->j_last_flush_id, | 1540 | remove_all_from_journal_list(s, jl, 0); |
1445 | jl->j_trans_id); | 1541 | list_del_init(&jl->j_list); |
1446 | } | 1542 | journal->j_num_lists--; |
1447 | journal->j_last_flush_id = jl->j_trans_id; | 1543 | del_from_work_list(s, jl); |
1448 | 1544 | ||
1449 | /* not strictly required since we are freeing the list, but it should | 1545 | if (journal->j_last_flush_id != 0 && |
1450 | * help find code using dead lists later on | 1546 | (jl->j_trans_id - journal->j_last_flush_id) != 1) { |
1451 | */ | 1547 | reiserfs_warning(s, "clm-2201: last flush %lu, current %lu", |
1452 | jl->j_len = 0 ; | 1548 | journal->j_last_flush_id, jl->j_trans_id); |
1453 | atomic_set(&(jl->j_nonzerolen), 0) ; | 1549 | } |
1454 | jl->j_start = 0 ; | 1550 | journal->j_last_flush_id = jl->j_trans_id; |
1455 | jl->j_realblock = NULL ; | 1551 | |
1456 | jl->j_commit_bh = NULL ; | 1552 | /* not strictly required since we are freeing the list, but it should |
1457 | jl->j_trans_id = 0 ; | 1553 | * help find code using dead lists later on |
1458 | jl->j_state = 0; | 1554 | */ |
1459 | put_journal_list(s, jl); | 1555 | jl->j_len = 0; |
1460 | if (flushall) | 1556 | atomic_set(&(jl->j_nonzerolen), 0); |
1461 | up(&journal->j_flush_sem); | 1557 | jl->j_start = 0; |
1462 | put_fs_excl(); | 1558 | jl->j_realblock = NULL; |
1463 | return err ; | 1559 | jl->j_commit_bh = NULL; |
1464 | } | 1560 | jl->j_trans_id = 0; |
1561 | jl->j_state = 0; | ||
1562 | put_journal_list(s, jl); | ||
1563 | if (flushall) | ||
1564 | up(&journal->j_flush_sem); | ||
1565 | put_fs_excl(); | ||
1566 | return err; | ||
1567 | } | ||
1465 | 1568 | ||
1466 | static int write_one_transaction(struct super_block *s, | 1569 | static int write_one_transaction(struct super_block *s, |
1467 | struct reiserfs_journal_list *jl, | 1570 | struct reiserfs_journal_list *jl, |
1468 | struct buffer_chunk *chunk) | 1571 | struct buffer_chunk *chunk) |
1469 | { | 1572 | { |
1470 | struct reiserfs_journal_cnode *cn; | 1573 | struct reiserfs_journal_cnode *cn; |
1471 | int ret = 0 ; | 1574 | int ret = 0; |
1472 | 1575 | ||
1473 | jl->j_state |= LIST_TOUCHED; | 1576 | jl->j_state |= LIST_TOUCHED; |
1474 | del_from_work_list(s, jl); | 1577 | del_from_work_list(s, jl); |
1475 | if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) { | 1578 | if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) { |
1476 | return 0; | 1579 | return 0; |
1477 | } | 1580 | } |
1478 | 1581 | ||
1479 | cn = jl->j_realblock ; | 1582 | cn = jl->j_realblock; |
1480 | while(cn) { | 1583 | while (cn) { |
1481 | /* if the blocknr == 0, this has been cleared from the hash, | 1584 | /* if the blocknr == 0, this has been cleared from the hash, |
1482 | ** skip it | 1585 | ** skip it |
1483 | */ | 1586 | */ |
1484 | if (cn->blocknr == 0) { | 1587 | if (cn->blocknr == 0) { |
1485 | goto next ; | 1588 | goto next; |
1486 | } | 1589 | } |
1487 | if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) { | 1590 | if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) { |
1488 | struct buffer_head *tmp_bh; | 1591 | struct buffer_head *tmp_bh; |
1489 | /* we can race against journal_mark_freed when we try | 1592 | /* we can race against journal_mark_freed when we try |
1490 | * to lock_buffer(cn->bh), so we have to inc the buffer | 1593 | * to lock_buffer(cn->bh), so we have to inc the buffer |
1491 | * count, and recheck things after locking | 1594 | * count, and recheck things after locking |
1492 | */ | 1595 | */ |
1493 | tmp_bh = cn->bh; | 1596 | tmp_bh = cn->bh; |
1494 | get_bh(tmp_bh); | 1597 | get_bh(tmp_bh); |
1495 | lock_buffer(tmp_bh); | 1598 | lock_buffer(tmp_bh); |
1496 | if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) { | 1599 | if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) { |
1497 | if (!buffer_journal_dirty(tmp_bh) || | 1600 | if (!buffer_journal_dirty(tmp_bh) || |
1498 | buffer_journal_prepared(tmp_bh)) | 1601 | buffer_journal_prepared(tmp_bh)) |
1499 | BUG(); | 1602 | BUG(); |
1500 | add_to_chunk(chunk, tmp_bh, NULL, write_chunk); | 1603 | add_to_chunk(chunk, tmp_bh, NULL, write_chunk); |
1501 | ret++; | 1604 | ret++; |
1502 | } else { | 1605 | } else { |
1503 | /* note, cn->bh might be null now */ | 1606 | /* note, cn->bh might be null now */ |
1504 | unlock_buffer(tmp_bh); | 1607 | unlock_buffer(tmp_bh); |
1505 | } | 1608 | } |
1506 | put_bh(tmp_bh); | 1609 | put_bh(tmp_bh); |
1507 | } | 1610 | } |
1508 | next: | 1611 | next: |
1509 | cn = cn->next ; | 1612 | cn = cn->next; |
1510 | cond_resched(); | 1613 | cond_resched(); |
1511 | } | 1614 | } |
1512 | return ret ; | 1615 | return ret; |
1513 | } | 1616 | } |
1514 | 1617 | ||
1515 | /* used by flush_commit_list */ | 1618 | /* used by flush_commit_list */ |
1516 | static int dirty_one_transaction(struct super_block *s, | 1619 | static int dirty_one_transaction(struct super_block *s, |
1517 | struct reiserfs_journal_list *jl) | 1620 | struct reiserfs_journal_list *jl) |
1518 | { | 1621 | { |
1519 | struct reiserfs_journal_cnode *cn; | 1622 | struct reiserfs_journal_cnode *cn; |
1520 | struct reiserfs_journal_list *pjl; | 1623 | struct reiserfs_journal_list *pjl; |
1521 | int ret = 0 ; | 1624 | int ret = 0; |
1522 | 1625 | ||
1523 | jl->j_state |= LIST_DIRTY; | 1626 | jl->j_state |= LIST_DIRTY; |
1524 | cn = jl->j_realblock ; | 1627 | cn = jl->j_realblock; |
1525 | while(cn) { | 1628 | while (cn) { |
1526 | /* look for a more recent transaction that logged this | 1629 | /* look for a more recent transaction that logged this |
1527 | ** buffer. Only the most recent transaction with a buffer in | 1630 | ** buffer. Only the most recent transaction with a buffer in |
1528 | ** it is allowed to send that buffer to disk | 1631 | ** it is allowed to send that buffer to disk |
1529 | */ | 1632 | */ |
1530 | pjl = find_newer_jl_for_cn(cn) ; | 1633 | pjl = find_newer_jl_for_cn(cn); |
1531 | if (!pjl && cn->blocknr && cn->bh && buffer_journal_dirty(cn->bh)) | 1634 | if (!pjl && cn->blocknr && cn->bh |
1532 | { | 1635 | && buffer_journal_dirty(cn->bh)) { |
1533 | BUG_ON (!can_dirty(cn)); | 1636 | BUG_ON(!can_dirty(cn)); |
1534 | /* if the buffer is prepared, it will either be logged | 1637 | /* if the buffer is prepared, it will either be logged |
1535 | * or restored. If restored, we need to make sure | 1638 | * or restored. If restored, we need to make sure |
1536 | * it actually gets marked dirty | 1639 | * it actually gets marked dirty |
1537 | */ | 1640 | */ |
1538 | clear_buffer_journal_new (cn->bh); | 1641 | clear_buffer_journal_new(cn->bh); |
1539 | if (buffer_journal_prepared (cn->bh)) { | 1642 | if (buffer_journal_prepared(cn->bh)) { |
1540 | set_buffer_journal_restore_dirty (cn->bh); | 1643 | set_buffer_journal_restore_dirty(cn->bh); |
1541 | } else { | 1644 | } else { |
1542 | set_buffer_journal_test (cn->bh); | 1645 | set_buffer_journal_test(cn->bh); |
1543 | mark_buffer_dirty(cn->bh); | 1646 | mark_buffer_dirty(cn->bh); |
1544 | } | 1647 | } |
1545 | } | 1648 | } |
1546 | cn = cn->next ; | 1649 | cn = cn->next; |
1547 | } | 1650 | } |
1548 | return ret ; | 1651 | return ret; |
1549 | } | 1652 | } |
1550 | 1653 | ||
1551 | static int kupdate_transactions(struct super_block *s, | 1654 | static int kupdate_transactions(struct super_block *s, |
1552 | struct reiserfs_journal_list *jl, | 1655 | struct reiserfs_journal_list *jl, |
1553 | struct reiserfs_journal_list **next_jl, | 1656 | struct reiserfs_journal_list **next_jl, |
1554 | unsigned long *next_trans_id, | 1657 | unsigned long *next_trans_id, |
1555 | int num_blocks, | 1658 | int num_blocks, int num_trans) |
1556 | int num_trans) { | 1659 | { |
1557 | int ret = 0; | 1660 | int ret = 0; |
1558 | int written = 0 ; | 1661 | int written = 0; |
1559 | int transactions_flushed = 0; | 1662 | int transactions_flushed = 0; |
1560 | unsigned long orig_trans_id = jl->j_trans_id; | 1663 | unsigned long orig_trans_id = jl->j_trans_id; |
1561 | struct buffer_chunk chunk; | 1664 | struct buffer_chunk chunk; |
1562 | struct list_head *entry; | 1665 | struct list_head *entry; |
1563 | struct reiserfs_journal *journal = SB_JOURNAL (s); | 1666 | struct reiserfs_journal *journal = SB_JOURNAL(s); |
1564 | chunk.nr = 0; | 1667 | chunk.nr = 0; |
1565 | 1668 | ||
1566 | down(&journal->j_flush_sem); | 1669 | down(&journal->j_flush_sem); |
1567 | if (!journal_list_still_alive(s, orig_trans_id)) { | 1670 | if (!journal_list_still_alive(s, orig_trans_id)) { |
1568 | goto done; | 1671 | goto done; |
1569 | } | 1672 | } |
1570 | 1673 | ||
1571 | /* we've got j_flush_sem held, nobody is going to delete any | 1674 | /* we've got j_flush_sem held, nobody is going to delete any |
1572 | * of these lists out from underneath us | 1675 | * of these lists out from underneath us |
1573 | */ | 1676 | */ |
1574 | while((num_trans && transactions_flushed < num_trans) || | 1677 | while ((num_trans && transactions_flushed < num_trans) || |
1575 | (!num_trans && written < num_blocks)) { | 1678 | (!num_trans && written < num_blocks)) { |
1576 | 1679 | ||
1577 | if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) || | 1680 | if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) || |
1578 | atomic_read(&jl->j_commit_left) || !(jl->j_state & LIST_DIRTY)) | 1681 | atomic_read(&jl->j_commit_left) |
1579 | { | 1682 | || !(jl->j_state & LIST_DIRTY)) { |
1580 | del_from_work_list(s, jl); | 1683 | del_from_work_list(s, jl); |
1581 | break; | 1684 | break; |
1582 | } | 1685 | } |
1583 | ret = write_one_transaction(s, jl, &chunk); | 1686 | ret = write_one_transaction(s, jl, &chunk); |
1584 | 1687 | ||
1585 | if (ret < 0) | 1688 | if (ret < 0) |
1586 | goto done; | 1689 | goto done; |
1587 | transactions_flushed++; | 1690 | transactions_flushed++; |
1588 | written += ret; | 1691 | written += ret; |
1589 | entry = jl->j_list.next; | 1692 | entry = jl->j_list.next; |
1590 | 1693 | ||
1591 | /* did we wrap? */ | 1694 | /* did we wrap? */ |
1592 | if (entry == &journal->j_journal_list) { | 1695 | if (entry == &journal->j_journal_list) { |
1593 | break; | 1696 | break; |
1594 | } | 1697 | } |
1595 | jl = JOURNAL_LIST_ENTRY(entry); | 1698 | jl = JOURNAL_LIST_ENTRY(entry); |
1596 | 1699 | ||
1597 | /* don't bother with older transactions */ | 1700 | /* don't bother with older transactions */ |
1598 | if (jl->j_trans_id <= orig_trans_id) | 1701 | if (jl->j_trans_id <= orig_trans_id) |
1599 | break; | 1702 | break; |
1600 | } | 1703 | } |
1601 | if (chunk.nr) { | 1704 | if (chunk.nr) { |
1602 | write_chunk(&chunk); | 1705 | write_chunk(&chunk); |
1603 | } | 1706 | } |
1604 | 1707 | ||
1605 | done: | 1708 | done: |
1606 | up(&journal->j_flush_sem); | 1709 | up(&journal->j_flush_sem); |
1607 | return ret; | 1710 | return ret; |
1608 | } | 1711 | } |
1609 | 1712 | ||
1610 | /* for o_sync and fsync heavy applications, they tend to use | 1713 | /* for o_sync and fsync heavy applications, they tend to use |
@@ -1617,47 +1720,48 @@ done: | |||
1617 | ** list updates the header block | 1720 | ** list updates the header block |
1618 | */ | 1721 | */ |
1619 | static int flush_used_journal_lists(struct super_block *s, | 1722 | static int flush_used_journal_lists(struct super_block *s, |
1620 | struct reiserfs_journal_list *jl) { | 1723 | struct reiserfs_journal_list *jl) |
1621 | unsigned long len = 0; | 1724 | { |
1622 | unsigned long cur_len; | 1725 | unsigned long len = 0; |
1623 | int ret; | 1726 | unsigned long cur_len; |
1624 | int i; | 1727 | int ret; |
1625 | int limit = 256; | 1728 | int i; |
1626 | struct reiserfs_journal_list *tjl; | 1729 | int limit = 256; |
1627 | struct reiserfs_journal_list *flush_jl; | 1730 | struct reiserfs_journal_list *tjl; |
1628 | unsigned long trans_id; | 1731 | struct reiserfs_journal_list *flush_jl; |
1629 | struct reiserfs_journal *journal = SB_JOURNAL (s); | 1732 | unsigned long trans_id; |
1630 | 1733 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
1631 | flush_jl = tjl = jl; | 1734 | |
1632 | 1735 | flush_jl = tjl = jl; | |
1633 | /* in data logging mode, try harder to flush a lot of blocks */ | 1736 | |
1634 | if (reiserfs_data_log(s)) | 1737 | /* in data logging mode, try harder to flush a lot of blocks */ |
1635 | limit = 1024; | 1738 | if (reiserfs_data_log(s)) |
1636 | /* flush for 256 transactions or limit blocks, whichever comes first */ | 1739 | limit = 1024; |
1637 | for(i = 0 ; i < 256 && len < limit ; i++) { | 1740 | /* flush for 256 transactions or limit blocks, whichever comes first */ |
1638 | if (atomic_read(&tjl->j_commit_left) || | 1741 | for (i = 0; i < 256 && len < limit; i++) { |
1639 | tjl->j_trans_id < jl->j_trans_id) { | 1742 | if (atomic_read(&tjl->j_commit_left) || |
1640 | break; | 1743 | tjl->j_trans_id < jl->j_trans_id) { |
1641 | } | 1744 | break; |
1642 | cur_len = atomic_read(&tjl->j_nonzerolen); | 1745 | } |
1643 | if (cur_len > 0) { | 1746 | cur_len = atomic_read(&tjl->j_nonzerolen); |
1644 | tjl->j_state &= ~LIST_TOUCHED; | 1747 | if (cur_len > 0) { |
1645 | } | 1748 | tjl->j_state &= ~LIST_TOUCHED; |
1646 | len += cur_len; | 1749 | } |
1647 | flush_jl = tjl; | 1750 | len += cur_len; |
1648 | if (tjl->j_list.next == &journal->j_journal_list) | 1751 | flush_jl = tjl; |
1649 | break; | 1752 | if (tjl->j_list.next == &journal->j_journal_list) |
1650 | tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next); | 1753 | break; |
1651 | } | 1754 | tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next); |
1652 | /* try to find a group of blocks we can flush across all the | 1755 | } |
1653 | ** transactions, but only bother if we've actually spanned | 1756 | /* try to find a group of blocks we can flush across all the |
1654 | ** across multiple lists | 1757 | ** transactions, but only bother if we've actually spanned |
1655 | */ | 1758 | ** across multiple lists |
1656 | if (flush_jl != jl) { | 1759 | */ |
1657 | ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i); | 1760 | if (flush_jl != jl) { |
1658 | } | 1761 | ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i); |
1659 | flush_journal_list(s, flush_jl, 1); | 1762 | } |
1660 | return 0; | 1763 | flush_journal_list(s, flush_jl, 1); |
1764 | return 0; | ||
1661 | } | 1765 | } |
1662 | 1766 | ||
1663 | /* | 1767 | /* |
@@ -1665,207 +1769,248 @@ static int flush_used_journal_lists(struct super_block *s, | |||
1665 | ** only touchs the hnext and hprev pointers. | 1769 | ** only touchs the hnext and hprev pointers. |
1666 | */ | 1770 | */ |
1667 | void remove_journal_hash(struct super_block *sb, | 1771 | void remove_journal_hash(struct super_block *sb, |
1668 | struct reiserfs_journal_cnode **table, | 1772 | struct reiserfs_journal_cnode **table, |
1669 | struct reiserfs_journal_list *jl, | 1773 | struct reiserfs_journal_list *jl, |
1670 | unsigned long block, int remove_freed) | 1774 | unsigned long block, int remove_freed) |
1671 | { | 1775 | { |
1672 | struct reiserfs_journal_cnode *cur ; | 1776 | struct reiserfs_journal_cnode *cur; |
1673 | struct reiserfs_journal_cnode **head ; | 1777 | struct reiserfs_journal_cnode **head; |
1674 | 1778 | ||
1675 | head= &(journal_hash(table, sb, block)) ; | 1779 | head = &(journal_hash(table, sb, block)); |
1676 | if (!head) { | 1780 | if (!head) { |
1677 | return ; | 1781 | return; |
1678 | } | 1782 | } |
1679 | cur = *head ; | 1783 | cur = *head; |
1680 | while(cur) { | 1784 | while (cur) { |
1681 | if (cur->blocknr == block && cur->sb == sb && (jl == NULL || jl == cur->jlist) && | 1785 | if (cur->blocknr == block && cur->sb == sb |
1682 | (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) { | 1786 | && (jl == NULL || jl == cur->jlist) |
1683 | if (cur->hnext) { | 1787 | && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) { |
1684 | cur->hnext->hprev = cur->hprev ; | 1788 | if (cur->hnext) { |
1685 | } | 1789 | cur->hnext->hprev = cur->hprev; |
1686 | if (cur->hprev) { | 1790 | } |
1687 | cur->hprev->hnext = cur->hnext ; | 1791 | if (cur->hprev) { |
1688 | } else { | 1792 | cur->hprev->hnext = cur->hnext; |
1689 | *head = cur->hnext ; | 1793 | } else { |
1690 | } | 1794 | *head = cur->hnext; |
1691 | cur->blocknr = 0 ; | 1795 | } |
1692 | cur->sb = NULL ; | 1796 | cur->blocknr = 0; |
1693 | cur->state = 0 ; | 1797 | cur->sb = NULL; |
1694 | if (cur->bh && cur->jlist) /* anybody who clears the cur->bh will also dec the nonzerolen */ | 1798 | cur->state = 0; |
1695 | atomic_dec(&(cur->jlist->j_nonzerolen)) ; | 1799 | if (cur->bh && cur->jlist) /* anybody who clears the cur->bh will also dec the nonzerolen */ |
1696 | cur->bh = NULL ; | 1800 | atomic_dec(&(cur->jlist->j_nonzerolen)); |
1697 | cur->jlist = NULL ; | 1801 | cur->bh = NULL; |
1698 | } | 1802 | cur->jlist = NULL; |
1699 | cur = cur->hnext ; | 1803 | } |
1700 | } | 1804 | cur = cur->hnext; |
1701 | } | 1805 | } |
1702 | 1806 | } | |
1703 | static void free_journal_ram(struct super_block *p_s_sb) { | 1807 | |
1704 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); | 1808 | static void free_journal_ram(struct super_block *p_s_sb) |
1705 | reiserfs_kfree(journal->j_current_jl, | 1809 | { |
1706 | sizeof(struct reiserfs_journal_list), p_s_sb); | 1810 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); |
1707 | journal->j_num_lists--; | 1811 | reiserfs_kfree(journal->j_current_jl, |
1708 | 1812 | sizeof(struct reiserfs_journal_list), p_s_sb); | |
1709 | vfree(journal->j_cnode_free_orig) ; | 1813 | journal->j_num_lists--; |
1710 | free_list_bitmaps(p_s_sb, journal->j_list_bitmap) ; | 1814 | |
1711 | free_bitmap_nodes(p_s_sb) ; /* must be after free_list_bitmaps */ | 1815 | vfree(journal->j_cnode_free_orig); |
1712 | if (journal->j_header_bh) { | 1816 | free_list_bitmaps(p_s_sb, journal->j_list_bitmap); |
1713 | brelse(journal->j_header_bh) ; | 1817 | free_bitmap_nodes(p_s_sb); /* must be after free_list_bitmaps */ |
1714 | } | 1818 | if (journal->j_header_bh) { |
1715 | /* j_header_bh is on the journal dev, make sure not to release the journal | 1819 | brelse(journal->j_header_bh); |
1716 | * dev until we brelse j_header_bh | 1820 | } |
1717 | */ | 1821 | /* j_header_bh is on the journal dev, make sure not to release the journal |
1718 | release_journal_dev(p_s_sb, journal); | 1822 | * dev until we brelse j_header_bh |
1719 | vfree(journal) ; | 1823 | */ |
1824 | release_journal_dev(p_s_sb, journal); | ||
1825 | vfree(journal); | ||
1720 | } | 1826 | } |
1721 | 1827 | ||
1722 | /* | 1828 | /* |
1723 | ** call on unmount. Only set error to 1 if you haven't made your way out | 1829 | ** call on unmount. Only set error to 1 if you haven't made your way out |
1724 | ** of read_super() yet. Any other caller must keep error at 0. | 1830 | ** of read_super() yet. Any other caller must keep error at 0. |
1725 | */ | 1831 | */ |
1726 | static int do_journal_release(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, int error) { | 1832 | static int do_journal_release(struct reiserfs_transaction_handle *th, |
1727 | struct reiserfs_transaction_handle myth ; | 1833 | struct super_block *p_s_sb, int error) |
1728 | int flushed = 0; | 1834 | { |
1729 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); | 1835 | struct reiserfs_transaction_handle myth; |
1730 | 1836 | int flushed = 0; | |
1731 | /* we only want to flush out transactions if we were called with error == 0 | 1837 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); |
1732 | */ | 1838 | |
1733 | if (!error && !(p_s_sb->s_flags & MS_RDONLY)) { | 1839 | /* we only want to flush out transactions if we were called with error == 0 |
1734 | /* end the current trans */ | 1840 | */ |
1735 | BUG_ON (!th->t_trans_id); | 1841 | if (!error && !(p_s_sb->s_flags & MS_RDONLY)) { |
1736 | do_journal_end(th, p_s_sb,10, FLUSH_ALL) ; | 1842 | /* end the current trans */ |
1737 | 1843 | BUG_ON(!th->t_trans_id); | |
1738 | /* make sure something gets logged to force our way into the flush code */ | 1844 | do_journal_end(th, p_s_sb, 10, FLUSH_ALL); |
1739 | if (!journal_join(&myth, p_s_sb, 1)) { | 1845 | |
1740 | reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ; | 1846 | /* make sure something gets logged to force our way into the flush code */ |
1741 | journal_mark_dirty(&myth, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ; | 1847 | if (!journal_join(&myth, p_s_sb, 1)) { |
1742 | do_journal_end(&myth, p_s_sb,1, FLUSH_ALL) ; | 1848 | reiserfs_prepare_for_journal(p_s_sb, |
1743 | flushed = 1; | 1849 | SB_BUFFER_WITH_SB(p_s_sb), |
1744 | } | 1850 | 1); |
1745 | } | 1851 | journal_mark_dirty(&myth, p_s_sb, |
1746 | 1852 | SB_BUFFER_WITH_SB(p_s_sb)); | |
1747 | /* this also catches errors during the do_journal_end above */ | 1853 | do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL); |
1748 | if (!error && reiserfs_is_journal_aborted(journal)) { | 1854 | flushed = 1; |
1749 | memset(&myth, 0, sizeof(myth)); | 1855 | } |
1750 | if (!journal_join_abort(&myth, p_s_sb, 1)) { | 1856 | } |
1751 | reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ; | 1857 | |
1752 | journal_mark_dirty(&myth, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ; | 1858 | /* this also catches errors during the do_journal_end above */ |
1753 | do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL) ; | 1859 | if (!error && reiserfs_is_journal_aborted(journal)) { |
1754 | } | 1860 | memset(&myth, 0, sizeof(myth)); |
1755 | } | 1861 | if (!journal_join_abort(&myth, p_s_sb, 1)) { |
1756 | 1862 | reiserfs_prepare_for_journal(p_s_sb, | |
1757 | reiserfs_mounted_fs_count-- ; | 1863 | SB_BUFFER_WITH_SB(p_s_sb), |
1758 | /* wait for all commits to finish */ | 1864 | 1); |
1759 | cancel_delayed_work(&SB_JOURNAL(p_s_sb)->j_work); | 1865 | journal_mark_dirty(&myth, p_s_sb, |
1760 | flush_workqueue(commit_wq); | 1866 | SB_BUFFER_WITH_SB(p_s_sb)); |
1761 | if (!reiserfs_mounted_fs_count) { | 1867 | do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL); |
1762 | destroy_workqueue(commit_wq); | 1868 | } |
1763 | commit_wq = NULL; | 1869 | } |
1764 | } | 1870 | |
1765 | 1871 | reiserfs_mounted_fs_count--; | |
1766 | free_journal_ram(p_s_sb) ; | 1872 | /* wait for all commits to finish */ |
1767 | 1873 | cancel_delayed_work(&SB_JOURNAL(p_s_sb)->j_work); | |
1768 | return 0 ; | 1874 | flush_workqueue(commit_wq); |
1875 | if (!reiserfs_mounted_fs_count) { | ||
1876 | destroy_workqueue(commit_wq); | ||
1877 | commit_wq = NULL; | ||
1878 | } | ||
1879 | |||
1880 | free_journal_ram(p_s_sb); | ||
1881 | |||
1882 | return 0; | ||
1769 | } | 1883 | } |
1770 | 1884 | ||
1771 | /* | 1885 | /* |
1772 | ** call on unmount. flush all journal trans, release all alloc'd ram | 1886 | ** call on unmount. flush all journal trans, release all alloc'd ram |
1773 | */ | 1887 | */ |
1774 | int journal_release(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb) { | 1888 | int journal_release(struct reiserfs_transaction_handle *th, |
1775 | return do_journal_release(th, p_s_sb, 0) ; | 1889 | struct super_block *p_s_sb) |
1890 | { | ||
1891 | return do_journal_release(th, p_s_sb, 0); | ||
1776 | } | 1892 | } |
1893 | |||
1777 | /* | 1894 | /* |
1778 | ** only call from an error condition inside reiserfs_read_super! | 1895 | ** only call from an error condition inside reiserfs_read_super! |
1779 | */ | 1896 | */ |
1780 | int journal_release_error(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb) { | 1897 | int journal_release_error(struct reiserfs_transaction_handle *th, |
1781 | return do_journal_release(th, p_s_sb, 1) ; | 1898 | struct super_block *p_s_sb) |
1899 | { | ||
1900 | return do_journal_release(th, p_s_sb, 1); | ||
1782 | } | 1901 | } |
1783 | 1902 | ||
1784 | /* compares description block with commit block. returns 1 if they differ, 0 if they are the same */ | 1903 | /* compares description block with commit block. returns 1 if they differ, 0 if they are the same */ |
1785 | static int journal_compare_desc_commit(struct super_block *p_s_sb, struct reiserfs_journal_desc *desc, | 1904 | static int journal_compare_desc_commit(struct super_block *p_s_sb, |
1786 | struct reiserfs_journal_commit *commit) { | 1905 | struct reiserfs_journal_desc *desc, |
1787 | if (get_commit_trans_id (commit) != get_desc_trans_id (desc) || | 1906 | struct reiserfs_journal_commit *commit) |
1788 | get_commit_trans_len (commit) != get_desc_trans_len (desc) || | 1907 | { |
1789 | get_commit_trans_len (commit) > SB_JOURNAL(p_s_sb)->j_trans_max || | 1908 | if (get_commit_trans_id(commit) != get_desc_trans_id(desc) || |
1790 | get_commit_trans_len (commit) <= 0 | 1909 | get_commit_trans_len(commit) != get_desc_trans_len(desc) || |
1791 | ) { | 1910 | get_commit_trans_len(commit) > SB_JOURNAL(p_s_sb)->j_trans_max || |
1792 | return 1 ; | 1911 | get_commit_trans_len(commit) <= 0) { |
1793 | } | 1912 | return 1; |
1794 | return 0 ; | 1913 | } |
1914 | return 0; | ||
1795 | } | 1915 | } |
1916 | |||
1796 | /* returns 0 if it did not find a description block | 1917 | /* returns 0 if it did not find a description block |
1797 | ** returns -1 if it found a corrupt commit block | 1918 | ** returns -1 if it found a corrupt commit block |
1798 | ** returns 1 if both desc and commit were valid | 1919 | ** returns 1 if both desc and commit were valid |
1799 | */ | 1920 | */ |
1800 | static int journal_transaction_is_valid(struct super_block *p_s_sb, struct buffer_head *d_bh, unsigned long *oldest_invalid_trans_id, unsigned long *newest_mount_id) { | 1921 | static int journal_transaction_is_valid(struct super_block *p_s_sb, |
1801 | struct reiserfs_journal_desc *desc ; | 1922 | struct buffer_head *d_bh, |
1802 | struct reiserfs_journal_commit *commit ; | 1923 | unsigned long *oldest_invalid_trans_id, |
1803 | struct buffer_head *c_bh ; | 1924 | unsigned long *newest_mount_id) |
1804 | unsigned long offset ; | 1925 | { |
1805 | 1926 | struct reiserfs_journal_desc *desc; | |
1806 | if (!d_bh) | 1927 | struct reiserfs_journal_commit *commit; |
1807 | return 0 ; | 1928 | struct buffer_head *c_bh; |
1808 | 1929 | unsigned long offset; | |
1809 | desc = (struct reiserfs_journal_desc *)d_bh->b_data ; | 1930 | |
1810 | if (get_desc_trans_len(desc) > 0 && !memcmp(get_journal_desc_magic (d_bh), JOURNAL_DESC_MAGIC, 8)) { | 1931 | if (!d_bh) |
1811 | if (oldest_invalid_trans_id && *oldest_invalid_trans_id && get_desc_trans_id(desc) > *oldest_invalid_trans_id) { | 1932 | return 0; |
1812 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-986: transaction " | 1933 | |
1813 | "is valid returning because trans_id %d is greater than " | 1934 | desc = (struct reiserfs_journal_desc *)d_bh->b_data; |
1814 | "oldest_invalid %lu", get_desc_trans_id(desc), | 1935 | if (get_desc_trans_len(desc) > 0 |
1815 | *oldest_invalid_trans_id); | 1936 | && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) { |
1816 | return 0 ; | 1937 | if (oldest_invalid_trans_id && *oldest_invalid_trans_id |
1817 | } | 1938 | && get_desc_trans_id(desc) > *oldest_invalid_trans_id) { |
1818 | if (newest_mount_id && *newest_mount_id > get_desc_mount_id (desc)) { | 1939 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, |
1819 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1087: transaction " | 1940 | "journal-986: transaction " |
1820 | "is valid returning because mount_id %d is less than " | 1941 | "is valid returning because trans_id %d is greater than " |
1821 | "newest_mount_id %lu", get_desc_mount_id (desc), | 1942 | "oldest_invalid %lu", |
1822 | *newest_mount_id) ; | 1943 | get_desc_trans_id(desc), |
1823 | return -1 ; | 1944 | *oldest_invalid_trans_id); |
1824 | } | 1945 | return 0; |
1825 | if ( get_desc_trans_len(desc) > SB_JOURNAL(p_s_sb)->j_trans_max ) { | 1946 | } |
1826 | reiserfs_warning(p_s_sb, "journal-2018: Bad transaction length %d encountered, ignoring transaction", get_desc_trans_len(desc)); | 1947 | if (newest_mount_id |
1827 | return -1 ; | 1948 | && *newest_mount_id > get_desc_mount_id(desc)) { |
1828 | } | 1949 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, |
1829 | offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) ; | 1950 | "journal-1087: transaction " |
1830 | 1951 | "is valid returning because mount_id %d is less than " | |
1831 | /* ok, we have a journal description block, lets see if the transaction was valid */ | 1952 | "newest_mount_id %lu", |
1832 | c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + | 1953 | get_desc_mount_id(desc), |
1833 | ((offset + get_desc_trans_len(desc) + 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb))) ; | 1954 | *newest_mount_id); |
1834 | if (!c_bh) | 1955 | return -1; |
1835 | return 0 ; | 1956 | } |
1836 | commit = (struct reiserfs_journal_commit *)c_bh->b_data ; | 1957 | if (get_desc_trans_len(desc) > SB_JOURNAL(p_s_sb)->j_trans_max) { |
1837 | if (journal_compare_desc_commit(p_s_sb, desc, commit)) { | 1958 | reiserfs_warning(p_s_sb, |
1838 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, | 1959 | "journal-2018: Bad transaction length %d encountered, ignoring transaction", |
1839 | "journal_transaction_is_valid, commit offset %ld had bad " | 1960 | get_desc_trans_len(desc)); |
1840 | "time %d or length %d", | 1961 | return -1; |
1841 | c_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), | 1962 | } |
1842 | get_commit_trans_id (commit), | 1963 | offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb); |
1843 | get_commit_trans_len(commit)); | 1964 | |
1844 | brelse(c_bh) ; | 1965 | /* ok, we have a journal description block, lets see if the transaction was valid */ |
1845 | if (oldest_invalid_trans_id) { | 1966 | c_bh = |
1846 | *oldest_invalid_trans_id = get_desc_trans_id(desc) ; | 1967 | journal_bread(p_s_sb, |
1847 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1004: " | 1968 | SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + |
1848 | "transaction_is_valid setting oldest invalid trans_id " | 1969 | ((offset + get_desc_trans_len(desc) + |
1849 | "to %d", get_desc_trans_id(desc)) ; | 1970 | 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb))); |
1850 | } | 1971 | if (!c_bh) |
1851 | return -1; | 1972 | return 0; |
1852 | } | 1973 | commit = (struct reiserfs_journal_commit *)c_bh->b_data; |
1853 | brelse(c_bh) ; | 1974 | if (journal_compare_desc_commit(p_s_sb, desc, commit)) { |
1854 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1006: found valid " | 1975 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, |
1855 | "transaction start offset %llu, len %d id %d", | 1976 | "journal_transaction_is_valid, commit offset %ld had bad " |
1856 | d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), | 1977 | "time %d or length %d", |
1857 | get_desc_trans_len(desc), get_desc_trans_id(desc)) ; | 1978 | c_bh->b_blocknr - |
1858 | return 1 ; | 1979 | SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), |
1859 | } else { | 1980 | get_commit_trans_id(commit), |
1860 | return 0 ; | 1981 | get_commit_trans_len(commit)); |
1861 | } | 1982 | brelse(c_bh); |
1862 | } | 1983 | if (oldest_invalid_trans_id) { |
1863 | 1984 | *oldest_invalid_trans_id = | |
1864 | static void brelse_array(struct buffer_head **heads, int num) { | 1985 | get_desc_trans_id(desc); |
1865 | int i ; | 1986 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, |
1866 | for (i = 0 ; i < num ; i++) { | 1987 | "journal-1004: " |
1867 | brelse(heads[i]) ; | 1988 | "transaction_is_valid setting oldest invalid trans_id " |
1868 | } | 1989 | "to %d", |
1990 | get_desc_trans_id(desc)); | ||
1991 | } | ||
1992 | return -1; | ||
1993 | } | ||
1994 | brelse(c_bh); | ||
1995 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, | ||
1996 | "journal-1006: found valid " | ||
1997 | "transaction start offset %llu, len %d id %d", | ||
1998 | d_bh->b_blocknr - | ||
1999 | SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), | ||
2000 | get_desc_trans_len(desc), | ||
2001 | get_desc_trans_id(desc)); | ||
2002 | return 1; | ||
2003 | } else { | ||
2004 | return 0; | ||
2005 | } | ||
2006 | } | ||
2007 | |||
2008 | static void brelse_array(struct buffer_head **heads, int num) | ||
2009 | { | ||
2010 | int i; | ||
2011 | for (i = 0; i < num; i++) { | ||
2012 | brelse(heads[i]); | ||
2013 | } | ||
1869 | } | 2014 | } |
1870 | 2015 | ||
1871 | /* | 2016 | /* |
@@ -1873,149 +2018,202 @@ static void brelse_array(struct buffer_head **heads, int num) { | |||
1873 | ** this either reads in a replays a transaction, or returns because the transaction | 2018 | ** this either reads in a replays a transaction, or returns because the transaction |
1874 | ** is invalid, or too old. | 2019 | ** is invalid, or too old. |
1875 | */ | 2020 | */ |
1876 | static int journal_read_transaction(struct super_block *p_s_sb, unsigned long cur_dblock, unsigned long oldest_start, | 2021 | static int journal_read_transaction(struct super_block *p_s_sb, |
1877 | unsigned long oldest_trans_id, unsigned long newest_mount_id) { | 2022 | unsigned long cur_dblock, |
1878 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | 2023 | unsigned long oldest_start, |
1879 | struct reiserfs_journal_desc *desc ; | 2024 | unsigned long oldest_trans_id, |
1880 | struct reiserfs_journal_commit *commit ; | 2025 | unsigned long newest_mount_id) |
1881 | unsigned long trans_id = 0 ; | 2026 | { |
1882 | struct buffer_head *c_bh ; | 2027 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); |
1883 | struct buffer_head *d_bh ; | 2028 | struct reiserfs_journal_desc *desc; |
1884 | struct buffer_head **log_blocks = NULL ; | 2029 | struct reiserfs_journal_commit *commit; |
1885 | struct buffer_head **real_blocks = NULL ; | 2030 | unsigned long trans_id = 0; |
1886 | unsigned long trans_offset ; | 2031 | struct buffer_head *c_bh; |
1887 | int i; | 2032 | struct buffer_head *d_bh; |
1888 | int trans_half; | 2033 | struct buffer_head **log_blocks = NULL; |
1889 | 2034 | struct buffer_head **real_blocks = NULL; | |
1890 | d_bh = journal_bread(p_s_sb, cur_dblock) ; | 2035 | unsigned long trans_offset; |
1891 | if (!d_bh) | 2036 | int i; |
1892 | return 1 ; | 2037 | int trans_half; |
1893 | desc = (struct reiserfs_journal_desc *)d_bh->b_data ; | 2038 | |
1894 | trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) ; | 2039 | d_bh = journal_bread(p_s_sb, cur_dblock); |
1895 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: " | 2040 | if (!d_bh) |
1896 | "journal_read_transaction, offset %llu, len %d mount_id %d", | 2041 | return 1; |
1897 | d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), | 2042 | desc = (struct reiserfs_journal_desc *)d_bh->b_data; |
1898 | get_desc_trans_len(desc), get_desc_mount_id(desc)) ; | 2043 | trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb); |
1899 | if (get_desc_trans_id(desc) < oldest_trans_id) { | 2044 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: " |
1900 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: " | 2045 | "journal_read_transaction, offset %llu, len %d mount_id %d", |
1901 | "journal_read_trans skipping because %lu is too old", | 2046 | d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), |
1902 | cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)) ; | 2047 | get_desc_trans_len(desc), get_desc_mount_id(desc)); |
1903 | brelse(d_bh) ; | 2048 | if (get_desc_trans_id(desc) < oldest_trans_id) { |
1904 | return 1 ; | 2049 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: " |
1905 | } | 2050 | "journal_read_trans skipping because %lu is too old", |
1906 | if (get_desc_mount_id(desc) != newest_mount_id) { | 2051 | cur_dblock - |
1907 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: " | 2052 | SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)); |
1908 | "journal_read_trans skipping because %d is != " | 2053 | brelse(d_bh); |
1909 | "newest_mount_id %lu", get_desc_mount_id(desc), | 2054 | return 1; |
1910 | newest_mount_id) ; | 2055 | } |
1911 | brelse(d_bh) ; | 2056 | if (get_desc_mount_id(desc) != newest_mount_id) { |
1912 | return 1 ; | 2057 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: " |
1913 | } | 2058 | "journal_read_trans skipping because %d is != " |
1914 | c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + | 2059 | "newest_mount_id %lu", get_desc_mount_id(desc), |
1915 | ((trans_offset + get_desc_trans_len(desc) + 1) % | 2060 | newest_mount_id); |
1916 | SB_ONDISK_JOURNAL_SIZE(p_s_sb))) ; | 2061 | brelse(d_bh); |
1917 | if (!c_bh) { | 2062 | return 1; |
1918 | brelse(d_bh) ; | 2063 | } |
1919 | return 1 ; | 2064 | c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + |
1920 | } | 2065 | ((trans_offset + get_desc_trans_len(desc) + 1) % |
1921 | commit = (struct reiserfs_journal_commit *)c_bh->b_data ; | 2066 | SB_ONDISK_JOURNAL_SIZE(p_s_sb))); |
1922 | if (journal_compare_desc_commit(p_s_sb, desc, commit)) { | 2067 | if (!c_bh) { |
1923 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal_read_transaction, " | 2068 | brelse(d_bh); |
1924 | "commit offset %llu had bad time %d or length %d", | 2069 | return 1; |
1925 | c_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), | 2070 | } |
1926 | get_commit_trans_id(commit), get_commit_trans_len(commit)); | 2071 | commit = (struct reiserfs_journal_commit *)c_bh->b_data; |
1927 | brelse(c_bh) ; | 2072 | if (journal_compare_desc_commit(p_s_sb, desc, commit)) { |
1928 | brelse(d_bh) ; | 2073 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, |
1929 | return 1; | 2074 | "journal_read_transaction, " |
1930 | } | 2075 | "commit offset %llu had bad time %d or length %d", |
1931 | trans_id = get_desc_trans_id(desc) ; | 2076 | c_bh->b_blocknr - |
1932 | /* now we know we've got a good transaction, and it was inside the valid time ranges */ | 2077 | SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), |
1933 | log_blocks = reiserfs_kmalloc(get_desc_trans_len(desc) * sizeof(struct buffer_head *), GFP_NOFS, p_s_sb) ; | 2078 | get_commit_trans_id(commit), |
1934 | real_blocks = reiserfs_kmalloc(get_desc_trans_len(desc) * sizeof(struct buffer_head *), GFP_NOFS, p_s_sb) ; | 2079 | get_commit_trans_len(commit)); |
1935 | if (!log_blocks || !real_blocks) { | 2080 | brelse(c_bh); |
1936 | brelse(c_bh) ; | 2081 | brelse(d_bh); |
1937 | brelse(d_bh) ; | 2082 | return 1; |
1938 | reiserfs_kfree(log_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ; | 2083 | } |
1939 | reiserfs_kfree(real_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ; | 2084 | trans_id = get_desc_trans_id(desc); |
1940 | reiserfs_warning(p_s_sb, "journal-1169: kmalloc failed, unable to mount FS") ; | 2085 | /* now we know we've got a good transaction, and it was inside the valid time ranges */ |
1941 | return -1 ; | 2086 | log_blocks = |
1942 | } | 2087 | reiserfs_kmalloc(get_desc_trans_len(desc) * |
1943 | /* get all the buffer heads */ | 2088 | sizeof(struct buffer_head *), GFP_NOFS, p_s_sb); |
1944 | trans_half = journal_trans_half (p_s_sb->s_blocksize) ; | 2089 | real_blocks = |
1945 | for(i = 0 ; i < get_desc_trans_len(desc) ; i++) { | 2090 | reiserfs_kmalloc(get_desc_trans_len(desc) * |
1946 | log_blocks[i] = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + (trans_offset + 1 + i) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)); | 2091 | sizeof(struct buffer_head *), GFP_NOFS, p_s_sb); |
1947 | if (i < trans_half) { | 2092 | if (!log_blocks || !real_blocks) { |
1948 | real_blocks[i] = sb_getblk(p_s_sb, le32_to_cpu(desc->j_realblock[i])) ; | 2093 | brelse(c_bh); |
1949 | } else { | 2094 | brelse(d_bh); |
1950 | real_blocks[i] = sb_getblk(p_s_sb, le32_to_cpu(commit->j_realblock[i - trans_half])) ; | 2095 | reiserfs_kfree(log_blocks, |
1951 | } | 2096 | get_desc_trans_len(desc) * |
1952 | if ( real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(p_s_sb) ) { | 2097 | sizeof(struct buffer_head *), p_s_sb); |
1953 | reiserfs_warning(p_s_sb, "journal-1207: REPLAY FAILURE fsck required! Block to replay is outside of filesystem"); | 2098 | reiserfs_kfree(real_blocks, |
1954 | goto abort_replay; | 2099 | get_desc_trans_len(desc) * |
1955 | } | 2100 | sizeof(struct buffer_head *), p_s_sb); |
1956 | /* make sure we don't try to replay onto log or reserved area */ | 2101 | reiserfs_warning(p_s_sb, |
1957 | if (is_block_in_log_or_reserved_area(p_s_sb, real_blocks[i]->b_blocknr)) { | 2102 | "journal-1169: kmalloc failed, unable to mount FS"); |
1958 | reiserfs_warning(p_s_sb, "journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block") ; | 2103 | return -1; |
1959 | abort_replay: | 2104 | } |
1960 | brelse_array(log_blocks, i) ; | 2105 | /* get all the buffer heads */ |
1961 | brelse_array(real_blocks, i) ; | 2106 | trans_half = journal_trans_half(p_s_sb->s_blocksize); |
1962 | brelse(c_bh) ; | 2107 | for (i = 0; i < get_desc_trans_len(desc); i++) { |
1963 | brelse(d_bh) ; | 2108 | log_blocks[i] = |
1964 | reiserfs_kfree(log_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ; | 2109 | journal_getblk(p_s_sb, |
1965 | reiserfs_kfree(real_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ; | 2110 | SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + |
1966 | return -1 ; | 2111 | (trans_offset + 1 + |
1967 | } | 2112 | i) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)); |
1968 | } | 2113 | if (i < trans_half) { |
1969 | /* read in the log blocks, memcpy to the corresponding real block */ | 2114 | real_blocks[i] = |
1970 | ll_rw_block(READ, get_desc_trans_len(desc), log_blocks) ; | 2115 | sb_getblk(p_s_sb, |
1971 | for (i = 0 ; i < get_desc_trans_len(desc) ; i++) { | 2116 | le32_to_cpu(desc->j_realblock[i])); |
1972 | wait_on_buffer(log_blocks[i]) ; | 2117 | } else { |
1973 | if (!buffer_uptodate(log_blocks[i])) { | 2118 | real_blocks[i] = |
1974 | reiserfs_warning(p_s_sb, "journal-1212: REPLAY FAILURE fsck required! buffer write failed") ; | 2119 | sb_getblk(p_s_sb, |
1975 | brelse_array(log_blocks + i, get_desc_trans_len(desc) - i) ; | 2120 | le32_to_cpu(commit-> |
1976 | brelse_array(real_blocks, get_desc_trans_len(desc)) ; | 2121 | j_realblock[i - trans_half])); |
1977 | brelse(c_bh) ; | 2122 | } |
1978 | brelse(d_bh) ; | 2123 | if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(p_s_sb)) { |
1979 | reiserfs_kfree(log_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ; | 2124 | reiserfs_warning(p_s_sb, |
1980 | reiserfs_kfree(real_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ; | 2125 | "journal-1207: REPLAY FAILURE fsck required! Block to replay is outside of filesystem"); |
1981 | return -1 ; | 2126 | goto abort_replay; |
1982 | } | 2127 | } |
1983 | memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data, real_blocks[i]->b_size) ; | 2128 | /* make sure we don't try to replay onto log or reserved area */ |
1984 | set_buffer_uptodate(real_blocks[i]) ; | 2129 | if (is_block_in_log_or_reserved_area |
1985 | brelse(log_blocks[i]) ; | 2130 | (p_s_sb, real_blocks[i]->b_blocknr)) { |
1986 | } | 2131 | reiserfs_warning(p_s_sb, |
1987 | /* flush out the real blocks */ | 2132 | "journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block"); |
1988 | for (i = 0 ; i < get_desc_trans_len(desc) ; i++) { | 2133 | abort_replay: |
1989 | set_buffer_dirty(real_blocks[i]) ; | 2134 | brelse_array(log_blocks, i); |
1990 | ll_rw_block(WRITE, 1, real_blocks + i) ; | 2135 | brelse_array(real_blocks, i); |
1991 | } | 2136 | brelse(c_bh); |
1992 | for (i = 0 ; i < get_desc_trans_len(desc) ; i++) { | 2137 | brelse(d_bh); |
1993 | wait_on_buffer(real_blocks[i]) ; | 2138 | reiserfs_kfree(log_blocks, |
1994 | if (!buffer_uptodate(real_blocks[i])) { | 2139 | get_desc_trans_len(desc) * |
1995 | reiserfs_warning(p_s_sb, "journal-1226: REPLAY FAILURE, fsck required! buffer write failed") ; | 2140 | sizeof(struct buffer_head *), p_s_sb); |
1996 | brelse_array(real_blocks + i, get_desc_trans_len(desc) - i) ; | 2141 | reiserfs_kfree(real_blocks, |
1997 | brelse(c_bh) ; | 2142 | get_desc_trans_len(desc) * |
1998 | brelse(d_bh) ; | 2143 | sizeof(struct buffer_head *), p_s_sb); |
1999 | reiserfs_kfree(log_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ; | 2144 | return -1; |
2000 | reiserfs_kfree(real_blocks, get_desc_trans_len(desc) * sizeof(struct buffer_head *), p_s_sb) ; | 2145 | } |
2001 | return -1 ; | 2146 | } |
2002 | } | 2147 | /* read in the log blocks, memcpy to the corresponding real block */ |
2003 | brelse(real_blocks[i]) ; | 2148 | ll_rw_block(READ, get_desc_trans_len(desc), log_blocks); |
2004 | } | 2149 | for (i = 0; i < get_desc_trans_len(desc); i++) { |
2005 | cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + ((trans_offset + get_desc_trans_len(desc) + 2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)) ; | 2150 | wait_on_buffer(log_blocks[i]); |
2006 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1095: setting journal " | 2151 | if (!buffer_uptodate(log_blocks[i])) { |
2007 | "start to offset %ld", | 2152 | reiserfs_warning(p_s_sb, |
2008 | cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)) ; | 2153 | "journal-1212: REPLAY FAILURE fsck required! buffer write failed"); |
2009 | 2154 | brelse_array(log_blocks + i, | |
2010 | /* init starting values for the first transaction, in case this is the last transaction to be replayed. */ | 2155 | get_desc_trans_len(desc) - i); |
2011 | journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) ; | 2156 | brelse_array(real_blocks, get_desc_trans_len(desc)); |
2012 | journal->j_last_flush_trans_id = trans_id ; | 2157 | brelse(c_bh); |
2013 | journal->j_trans_id = trans_id + 1; | 2158 | brelse(d_bh); |
2014 | brelse(c_bh) ; | 2159 | reiserfs_kfree(log_blocks, |
2015 | brelse(d_bh) ; | 2160 | get_desc_trans_len(desc) * |
2016 | reiserfs_kfree(log_blocks, le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), p_s_sb) ; | 2161 | sizeof(struct buffer_head *), p_s_sb); |
2017 | reiserfs_kfree(real_blocks, le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), p_s_sb) ; | 2162 | reiserfs_kfree(real_blocks, |
2018 | return 0 ; | 2163 | get_desc_trans_len(desc) * |
2164 | sizeof(struct buffer_head *), p_s_sb); | ||
2165 | return -1; | ||
2166 | } | ||
2167 | memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data, | ||
2168 | real_blocks[i]->b_size); | ||
2169 | set_buffer_uptodate(real_blocks[i]); | ||
2170 | brelse(log_blocks[i]); | ||
2171 | } | ||
2172 | /* flush out the real blocks */ | ||
2173 | for (i = 0; i < get_desc_trans_len(desc); i++) { | ||
2174 | set_buffer_dirty(real_blocks[i]); | ||
2175 | ll_rw_block(WRITE, 1, real_blocks + i); | ||
2176 | } | ||
2177 | for (i = 0; i < get_desc_trans_len(desc); i++) { | ||
2178 | wait_on_buffer(real_blocks[i]); | ||
2179 | if (!buffer_uptodate(real_blocks[i])) { | ||
2180 | reiserfs_warning(p_s_sb, | ||
2181 | "journal-1226: REPLAY FAILURE, fsck required! buffer write failed"); | ||
2182 | brelse_array(real_blocks + i, | ||
2183 | get_desc_trans_len(desc) - i); | ||
2184 | brelse(c_bh); | ||
2185 | brelse(d_bh); | ||
2186 | reiserfs_kfree(log_blocks, | ||
2187 | get_desc_trans_len(desc) * | ||
2188 | sizeof(struct buffer_head *), p_s_sb); | ||
2189 | reiserfs_kfree(real_blocks, | ||
2190 | get_desc_trans_len(desc) * | ||
2191 | sizeof(struct buffer_head *), p_s_sb); | ||
2192 | return -1; | ||
2193 | } | ||
2194 | brelse(real_blocks[i]); | ||
2195 | } | ||
2196 | cur_dblock = | ||
2197 | SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + | ||
2198 | ((trans_offset + get_desc_trans_len(desc) + | ||
2199 | 2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)); | ||
2200 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, | ||
2201 | "journal-1095: setting journal " "start to offset %ld", | ||
2202 | cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)); | ||
2203 | |||
2204 | /* init starting values for the first transaction, in case this is the last transaction to be replayed. */ | ||
2205 | journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb); | ||
2206 | journal->j_last_flush_trans_id = trans_id; | ||
2207 | journal->j_trans_id = trans_id + 1; | ||
2208 | brelse(c_bh); | ||
2209 | brelse(d_bh); | ||
2210 | reiserfs_kfree(log_blocks, | ||
2211 | le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), | ||
2212 | p_s_sb); | ||
2213 | reiserfs_kfree(real_blocks, | ||
2214 | le32_to_cpu(desc->j_len) * sizeof(struct buffer_head *), | ||
2215 | p_s_sb); | ||
2216 | return 0; | ||
2019 | } | 2217 | } |
2020 | 2218 | ||
2021 | /* This function reads blocks starting from block and to max_block of bufsize | 2219 | /* This function reads blocks starting from block and to max_block of bufsize |
@@ -2024,39 +2222,39 @@ abort_replay: | |||
2024 | Right now it is only used from journal code. But later we might use it | 2222 | Right now it is only used from journal code. But later we might use it |
2025 | from other places. | 2223 | from other places. |
2026 | Note: Do not use journal_getblk/sb_getblk functions here! */ | 2224 | Note: Do not use journal_getblk/sb_getblk functions here! */ |
2027 | static struct buffer_head * reiserfs_breada (struct block_device *dev, int block, int bufsize, | 2225 | static struct buffer_head *reiserfs_breada(struct block_device *dev, int block, |
2028 | unsigned int max_block) | 2226 | int bufsize, unsigned int max_block) |
2029 | { | 2227 | { |
2030 | struct buffer_head * bhlist[BUFNR]; | 2228 | struct buffer_head *bhlist[BUFNR]; |
2031 | unsigned int blocks = BUFNR; | 2229 | unsigned int blocks = BUFNR; |
2032 | struct buffer_head * bh; | 2230 | struct buffer_head *bh; |
2033 | int i, j; | 2231 | int i, j; |
2034 | 2232 | ||
2035 | bh = __getblk (dev, block, bufsize ); | 2233 | bh = __getblk(dev, block, bufsize); |
2036 | if (buffer_uptodate (bh)) | 2234 | if (buffer_uptodate(bh)) |
2037 | return (bh); | 2235 | return (bh); |
2038 | 2236 | ||
2039 | if (block + BUFNR > max_block) { | 2237 | if (block + BUFNR > max_block) { |
2040 | blocks = max_block - block; | 2238 | blocks = max_block - block; |
2041 | } | 2239 | } |
2042 | bhlist[0] = bh; | 2240 | bhlist[0] = bh; |
2043 | j = 1; | 2241 | j = 1; |
2044 | for (i = 1; i < blocks; i++) { | 2242 | for (i = 1; i < blocks; i++) { |
2045 | bh = __getblk (dev, block + i, bufsize); | 2243 | bh = __getblk(dev, block + i, bufsize); |
2046 | if (buffer_uptodate (bh)) { | 2244 | if (buffer_uptodate(bh)) { |
2047 | brelse (bh); | 2245 | brelse(bh); |
2048 | break; | 2246 | break; |
2049 | } | 2247 | } else |
2050 | else bhlist[j++] = bh; | 2248 | bhlist[j++] = bh; |
2051 | } | 2249 | } |
2052 | ll_rw_block (READ, j, bhlist); | 2250 | ll_rw_block(READ, j, bhlist); |
2053 | for(i = 1; i < j; i++) | 2251 | for (i = 1; i < j; i++) |
2054 | brelse (bhlist[i]); | 2252 | brelse(bhlist[i]); |
2055 | bh = bhlist[0]; | 2253 | bh = bhlist[0]; |
2056 | wait_on_buffer (bh); | 2254 | wait_on_buffer(bh); |
2057 | if (buffer_uptodate (bh)) | 2255 | if (buffer_uptodate(bh)) |
2058 | return bh; | 2256 | return bh; |
2059 | brelse (bh); | 2257 | brelse(bh); |
2060 | return NULL; | 2258 | return NULL; |
2061 | } | 2259 | } |
2062 | 2260 | ||
@@ -2069,218 +2267,250 @@ static struct buffer_head * reiserfs_breada (struct block_device *dev, int block | |||
2069 | ** | 2267 | ** |
2070 | ** On exit, it sets things up so the first transaction will work correctly. | 2268 | ** On exit, it sets things up so the first transaction will work correctly. |
2071 | */ | 2269 | */ |
2072 | static int journal_read(struct super_block *p_s_sb) { | 2270 | static int journal_read(struct super_block *p_s_sb) |
2073 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | 2271 | { |
2074 | struct reiserfs_journal_desc *desc ; | 2272 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); |
2075 | unsigned long oldest_trans_id = 0; | 2273 | struct reiserfs_journal_desc *desc; |
2076 | unsigned long oldest_invalid_trans_id = 0 ; | 2274 | unsigned long oldest_trans_id = 0; |
2077 | time_t start ; | 2275 | unsigned long oldest_invalid_trans_id = 0; |
2078 | unsigned long oldest_start = 0; | 2276 | time_t start; |
2079 | unsigned long cur_dblock = 0 ; | 2277 | unsigned long oldest_start = 0; |
2080 | unsigned long newest_mount_id = 9 ; | 2278 | unsigned long cur_dblock = 0; |
2081 | struct buffer_head *d_bh ; | 2279 | unsigned long newest_mount_id = 9; |
2082 | struct reiserfs_journal_header *jh ; | 2280 | struct buffer_head *d_bh; |
2083 | int valid_journal_header = 0 ; | 2281 | struct reiserfs_journal_header *jh; |
2084 | int replay_count = 0 ; | 2282 | int valid_journal_header = 0; |
2085 | int continue_replay = 1 ; | 2283 | int replay_count = 0; |
2086 | int ret ; | 2284 | int continue_replay = 1; |
2087 | char b[BDEVNAME_SIZE]; | 2285 | int ret; |
2088 | 2286 | char b[BDEVNAME_SIZE]; | |
2089 | cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) ; | 2287 | |
2090 | reiserfs_info (p_s_sb, "checking transaction log (%s)\n", | 2288 | cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb); |
2091 | bdevname(journal->j_dev_bd, b)); | 2289 | reiserfs_info(p_s_sb, "checking transaction log (%s)\n", |
2092 | start = get_seconds(); | 2290 | bdevname(journal->j_dev_bd, b)); |
2093 | 2291 | start = get_seconds(); | |
2094 | /* step 1, read in the journal header block. Check the transaction it says | 2292 | |
2095 | ** is the first unflushed, and if that transaction is not valid, | 2293 | /* step 1, read in the journal header block. Check the transaction it says |
2096 | ** replay is done | 2294 | ** is the first unflushed, and if that transaction is not valid, |
2097 | */ | 2295 | ** replay is done |
2098 | journal->j_header_bh = journal_bread(p_s_sb, | 2296 | */ |
2099 | SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + | 2297 | journal->j_header_bh = journal_bread(p_s_sb, |
2100 | SB_ONDISK_JOURNAL_SIZE(p_s_sb)); | 2298 | SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) |
2101 | if (!journal->j_header_bh) { | 2299 | + SB_ONDISK_JOURNAL_SIZE(p_s_sb)); |
2102 | return 1 ; | 2300 | if (!journal->j_header_bh) { |
2103 | } | 2301 | return 1; |
2104 | jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data) ; | 2302 | } |
2105 | if (le32_to_cpu(jh->j_first_unflushed_offset) >= 0 && | 2303 | jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data); |
2106 | le32_to_cpu(jh->j_first_unflushed_offset) < SB_ONDISK_JOURNAL_SIZE(p_s_sb) && | 2304 | if (le32_to_cpu(jh->j_first_unflushed_offset) >= 0 && |
2107 | le32_to_cpu(jh->j_last_flush_trans_id) > 0) { | 2305 | le32_to_cpu(jh->j_first_unflushed_offset) < |
2108 | oldest_start = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + | 2306 | SB_ONDISK_JOURNAL_SIZE(p_s_sb) |
2109 | le32_to_cpu(jh->j_first_unflushed_offset) ; | 2307 | && le32_to_cpu(jh->j_last_flush_trans_id) > 0) { |
2110 | oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1; | 2308 | oldest_start = |
2111 | newest_mount_id = le32_to_cpu(jh->j_mount_id); | 2309 | SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + |
2112 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1153: found in " | 2310 | le32_to_cpu(jh->j_first_unflushed_offset); |
2113 | "header: first_unflushed_offset %d, last_flushed_trans_id " | 2311 | oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1; |
2114 | "%lu", le32_to_cpu(jh->j_first_unflushed_offset), | 2312 | newest_mount_id = le32_to_cpu(jh->j_mount_id); |
2115 | le32_to_cpu(jh->j_last_flush_trans_id)) ; | 2313 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, |
2116 | valid_journal_header = 1 ; | 2314 | "journal-1153: found in " |
2117 | 2315 | "header: first_unflushed_offset %d, last_flushed_trans_id " | |
2118 | /* now, we try to read the first unflushed offset. If it is not valid, | 2316 | "%lu", le32_to_cpu(jh->j_first_unflushed_offset), |
2119 | ** there is nothing more we can do, and it makes no sense to read | 2317 | le32_to_cpu(jh->j_last_flush_trans_id)); |
2120 | ** through the whole log. | 2318 | valid_journal_header = 1; |
2121 | */ | 2319 | |
2122 | d_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + le32_to_cpu(jh->j_first_unflushed_offset)) ; | 2320 | /* now, we try to read the first unflushed offset. If it is not valid, |
2123 | ret = journal_transaction_is_valid(p_s_sb, d_bh, NULL, NULL) ; | 2321 | ** there is nothing more we can do, and it makes no sense to read |
2124 | if (!ret) { | 2322 | ** through the whole log. |
2125 | continue_replay = 0 ; | 2323 | */ |
2126 | } | 2324 | d_bh = |
2127 | brelse(d_bh) ; | 2325 | journal_bread(p_s_sb, |
2128 | goto start_log_replay; | 2326 | SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + |
2129 | } | 2327 | le32_to_cpu(jh->j_first_unflushed_offset)); |
2130 | 2328 | ret = journal_transaction_is_valid(p_s_sb, d_bh, NULL, NULL); | |
2131 | if (continue_replay && bdev_read_only(p_s_sb->s_bdev)) { | 2329 | if (!ret) { |
2132 | reiserfs_warning (p_s_sb, | 2330 | continue_replay = 0; |
2133 | "clm-2076: device is readonly, unable to replay log") ; | 2331 | } |
2134 | return -1 ; | 2332 | brelse(d_bh); |
2135 | } | 2333 | goto start_log_replay; |
2136 | 2334 | } | |
2137 | /* ok, there are transactions that need to be replayed. start with the first log block, find | 2335 | |
2138 | ** all the valid transactions, and pick out the oldest. | 2336 | if (continue_replay && bdev_read_only(p_s_sb->s_bdev)) { |
2139 | */ | 2337 | reiserfs_warning(p_s_sb, |
2140 | while(continue_replay && cur_dblock < (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + SB_ONDISK_JOURNAL_SIZE(p_s_sb))) { | 2338 | "clm-2076: device is readonly, unable to replay log"); |
2141 | /* Note that it is required for blocksize of primary fs device and journal | 2339 | return -1; |
2142 | device to be the same */ | 2340 | } |
2143 | d_bh = reiserfs_breada(journal->j_dev_bd, cur_dblock, p_s_sb->s_blocksize, | 2341 | |
2144 | SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + SB_ONDISK_JOURNAL_SIZE(p_s_sb)) ; | 2342 | /* ok, there are transactions that need to be replayed. start with the first log block, find |
2145 | ret = journal_transaction_is_valid(p_s_sb, d_bh, &oldest_invalid_trans_id, &newest_mount_id) ; | 2343 | ** all the valid transactions, and pick out the oldest. |
2146 | if (ret == 1) { | 2344 | */ |
2147 | desc = (struct reiserfs_journal_desc *)d_bh->b_data ; | 2345 | while (continue_replay |
2148 | if (oldest_start == 0) { /* init all oldest_ values */ | 2346 | && cur_dblock < |
2149 | oldest_trans_id = get_desc_trans_id(desc) ; | 2347 | (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + |
2150 | oldest_start = d_bh->b_blocknr ; | 2348 | SB_ONDISK_JOURNAL_SIZE(p_s_sb))) { |
2151 | newest_mount_id = get_desc_mount_id(desc) ; | 2349 | /* Note that it is required for blocksize of primary fs device and journal |
2152 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1179: Setting " | 2350 | device to be the same */ |
2153 | "oldest_start to offset %llu, trans_id %lu", | 2351 | d_bh = |
2154 | oldest_start - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), | 2352 | reiserfs_breada(journal->j_dev_bd, cur_dblock, |
2155 | oldest_trans_id) ; | 2353 | p_s_sb->s_blocksize, |
2156 | } else if (oldest_trans_id > get_desc_trans_id(desc)) { | 2354 | SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + |
2157 | /* one we just read was older */ | 2355 | SB_ONDISK_JOURNAL_SIZE(p_s_sb)); |
2158 | oldest_trans_id = get_desc_trans_id(desc) ; | 2356 | ret = |
2159 | oldest_start = d_bh->b_blocknr ; | 2357 | journal_transaction_is_valid(p_s_sb, d_bh, |
2160 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1180: Resetting " | 2358 | &oldest_invalid_trans_id, |
2161 | "oldest_start to offset %lu, trans_id %lu", | 2359 | &newest_mount_id); |
2162 | oldest_start - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), | 2360 | if (ret == 1) { |
2163 | oldest_trans_id) ; | 2361 | desc = (struct reiserfs_journal_desc *)d_bh->b_data; |
2164 | } | 2362 | if (oldest_start == 0) { /* init all oldest_ values */ |
2165 | if (newest_mount_id < get_desc_mount_id(desc)) { | 2363 | oldest_trans_id = get_desc_trans_id(desc); |
2166 | newest_mount_id = get_desc_mount_id(desc) ; | 2364 | oldest_start = d_bh->b_blocknr; |
2365 | newest_mount_id = get_desc_mount_id(desc); | ||
2366 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, | ||
2367 | "journal-1179: Setting " | ||
2368 | "oldest_start to offset %llu, trans_id %lu", | ||
2369 | oldest_start - | ||
2370 | SB_ONDISK_JOURNAL_1st_BLOCK | ||
2371 | (p_s_sb), oldest_trans_id); | ||
2372 | } else if (oldest_trans_id > get_desc_trans_id(desc)) { | ||
2373 | /* one we just read was older */ | ||
2374 | oldest_trans_id = get_desc_trans_id(desc); | ||
2375 | oldest_start = d_bh->b_blocknr; | ||
2376 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, | ||
2377 | "journal-1180: Resetting " | ||
2378 | "oldest_start to offset %lu, trans_id %lu", | ||
2379 | oldest_start - | ||
2380 | SB_ONDISK_JOURNAL_1st_BLOCK | ||
2381 | (p_s_sb), oldest_trans_id); | ||
2382 | } | ||
2383 | if (newest_mount_id < get_desc_mount_id(desc)) { | ||
2384 | newest_mount_id = get_desc_mount_id(desc); | ||
2385 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, | ||
2386 | "journal-1299: Setting " | ||
2387 | "newest_mount_id to %d", | ||
2388 | get_desc_mount_id(desc)); | ||
2389 | } | ||
2390 | cur_dblock += get_desc_trans_len(desc) + 2; | ||
2391 | } else { | ||
2392 | cur_dblock++; | ||
2393 | } | ||
2394 | brelse(d_bh); | ||
2395 | } | ||
2396 | |||
2397 | start_log_replay: | ||
2398 | cur_dblock = oldest_start; | ||
2399 | if (oldest_trans_id) { | ||
2400 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, | ||
2401 | "journal-1206: Starting replay " | ||
2402 | "from offset %llu, trans_id %lu", | ||
2403 | cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), | ||
2404 | oldest_trans_id); | ||
2405 | |||
2406 | } | ||
2407 | replay_count = 0; | ||
2408 | while (continue_replay && oldest_trans_id > 0) { | ||
2409 | ret = | ||
2410 | journal_read_transaction(p_s_sb, cur_dblock, oldest_start, | ||
2411 | oldest_trans_id, newest_mount_id); | ||
2412 | if (ret < 0) { | ||
2413 | return ret; | ||
2414 | } else if (ret != 0) { | ||
2415 | break; | ||
2416 | } | ||
2417 | cur_dblock = | ||
2418 | SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + journal->j_start; | ||
2419 | replay_count++; | ||
2420 | if (cur_dblock == oldest_start) | ||
2421 | break; | ||
2422 | } | ||
2423 | |||
2424 | if (oldest_trans_id == 0) { | ||
2425 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, | ||
2426 | "journal-1225: No valid " "transactions found"); | ||
2427 | } | ||
2428 | /* j_start does not get set correctly if we don't replay any transactions. | ||
2429 | ** if we had a valid journal_header, set j_start to the first unflushed transaction value, | ||
2430 | ** copy the trans_id from the header | ||
2431 | */ | ||
2432 | if (valid_journal_header && replay_count == 0) { | ||
2433 | journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset); | ||
2434 | journal->j_trans_id = | ||
2435 | le32_to_cpu(jh->j_last_flush_trans_id) + 1; | ||
2436 | journal->j_last_flush_trans_id = | ||
2437 | le32_to_cpu(jh->j_last_flush_trans_id); | ||
2438 | journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1; | ||
2439 | } else { | ||
2440 | journal->j_mount_id = newest_mount_id + 1; | ||
2441 | } | ||
2167 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting " | 2442 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting " |
2168 | "newest_mount_id to %d", get_desc_mount_id(desc)); | 2443 | "newest_mount_id to %lu", journal->j_mount_id); |
2169 | } | 2444 | journal->j_first_unflushed_offset = journal->j_start; |
2170 | cur_dblock += get_desc_trans_len(desc) + 2 ; | 2445 | if (replay_count > 0) { |
2171 | } else { | 2446 | reiserfs_info(p_s_sb, |
2172 | cur_dblock++ ; | 2447 | "replayed %d transactions in %lu seconds\n", |
2173 | } | 2448 | replay_count, get_seconds() - start); |
2174 | brelse(d_bh) ; | 2449 | } |
2175 | } | 2450 | if (!bdev_read_only(p_s_sb->s_bdev) && |
2176 | 2451 | _update_journal_header_block(p_s_sb, journal->j_start, | |
2177 | start_log_replay: | 2452 | journal->j_last_flush_trans_id)) { |
2178 | cur_dblock = oldest_start ; | 2453 | /* replay failed, caller must call free_journal_ram and abort |
2179 | if (oldest_trans_id) { | 2454 | ** the mount |
2180 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1206: Starting replay " | 2455 | */ |
2181 | "from offset %llu, trans_id %lu", | 2456 | return -1; |
2182 | cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), | 2457 | } |
2183 | oldest_trans_id) ; | 2458 | return 0; |
2184 | |||
2185 | } | ||
2186 | replay_count = 0 ; | ||
2187 | while(continue_replay && oldest_trans_id > 0) { | ||
2188 | ret = journal_read_transaction(p_s_sb, cur_dblock, oldest_start, oldest_trans_id, newest_mount_id) ; | ||
2189 | if (ret < 0) { | ||
2190 | return ret ; | ||
2191 | } else if (ret != 0) { | ||
2192 | break ; | ||
2193 | } | ||
2194 | cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + journal->j_start ; | ||
2195 | replay_count++ ; | ||
2196 | if (cur_dblock == oldest_start) | ||
2197 | break; | ||
2198 | } | ||
2199 | |||
2200 | if (oldest_trans_id == 0) { | ||
2201 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1225: No valid " | ||
2202 | "transactions found") ; | ||
2203 | } | ||
2204 | /* j_start does not get set correctly if we don't replay any transactions. | ||
2205 | ** if we had a valid journal_header, set j_start to the first unflushed transaction value, | ||
2206 | ** copy the trans_id from the header | ||
2207 | */ | ||
2208 | if (valid_journal_header && replay_count == 0) { | ||
2209 | journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset) ; | ||
2210 | journal->j_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1; | ||
2211 | journal->j_last_flush_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) ; | ||
2212 | journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1; | ||
2213 | } else { | ||
2214 | journal->j_mount_id = newest_mount_id + 1 ; | ||
2215 | } | ||
2216 | reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting " | ||
2217 | "newest_mount_id to %lu", journal->j_mount_id) ; | ||
2218 | journal->j_first_unflushed_offset = journal->j_start ; | ||
2219 | if (replay_count > 0) { | ||
2220 | reiserfs_info (p_s_sb, "replayed %d transactions in %lu seconds\n", | ||
2221 | replay_count, get_seconds() - start) ; | ||
2222 | } | ||
2223 | if (!bdev_read_only(p_s_sb->s_bdev) && | ||
2224 | _update_journal_header_block(p_s_sb, journal->j_start, | ||
2225 | journal->j_last_flush_trans_id)) | ||
2226 | { | ||
2227 | /* replay failed, caller must call free_journal_ram and abort | ||
2228 | ** the mount | ||
2229 | */ | ||
2230 | return -1 ; | ||
2231 | } | ||
2232 | return 0 ; | ||
2233 | } | 2459 | } |
2234 | 2460 | ||
2235 | static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s) | 2461 | static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s) |
2236 | { | 2462 | { |
2237 | struct reiserfs_journal_list *jl; | 2463 | struct reiserfs_journal_list *jl; |
2238 | retry: | 2464 | retry: |
2239 | jl = reiserfs_kmalloc(sizeof(struct reiserfs_journal_list), GFP_NOFS, s); | 2465 | jl = reiserfs_kmalloc(sizeof(struct reiserfs_journal_list), GFP_NOFS, |
2240 | if (!jl) { | 2466 | s); |
2241 | yield(); | 2467 | if (!jl) { |
2242 | goto retry; | 2468 | yield(); |
2243 | } | 2469 | goto retry; |
2244 | memset(jl, 0, sizeof(*jl)); | 2470 | } |
2245 | INIT_LIST_HEAD(&jl->j_list); | 2471 | memset(jl, 0, sizeof(*jl)); |
2246 | INIT_LIST_HEAD(&jl->j_working_list); | 2472 | INIT_LIST_HEAD(&jl->j_list); |
2247 | INIT_LIST_HEAD(&jl->j_tail_bh_list); | 2473 | INIT_LIST_HEAD(&jl->j_working_list); |
2248 | INIT_LIST_HEAD(&jl->j_bh_list); | 2474 | INIT_LIST_HEAD(&jl->j_tail_bh_list); |
2249 | sema_init(&jl->j_commit_lock, 1); | 2475 | INIT_LIST_HEAD(&jl->j_bh_list); |
2250 | SB_JOURNAL(s)->j_num_lists++; | 2476 | sema_init(&jl->j_commit_lock, 1); |
2251 | get_journal_list(jl); | 2477 | SB_JOURNAL(s)->j_num_lists++; |
2252 | return jl; | 2478 | get_journal_list(jl); |
2253 | } | 2479 | return jl; |
2254 | 2480 | } | |
2255 | static void journal_list_init(struct super_block *p_s_sb) { | 2481 | |
2256 | SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb); | 2482 | static void journal_list_init(struct super_block *p_s_sb) |
2257 | } | 2483 | { |
2258 | 2484 | SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb); | |
2259 | static int release_journal_dev( struct super_block *super, | 2485 | } |
2260 | struct reiserfs_journal *journal ) | 2486 | |
2261 | { | 2487 | static int release_journal_dev(struct super_block *super, |
2262 | int result; | 2488 | struct reiserfs_journal *journal) |
2263 | 2489 | { | |
2264 | result = 0; | 2490 | int result; |
2265 | 2491 | ||
2266 | if( journal -> j_dev_file != NULL ) { | 2492 | result = 0; |
2267 | result = filp_close( journal -> j_dev_file, NULL ); | 2493 | |
2268 | journal -> j_dev_file = NULL; | 2494 | if (journal->j_dev_file != NULL) { |
2269 | journal -> j_dev_bd = NULL; | 2495 | result = filp_close(journal->j_dev_file, NULL); |
2270 | } else if( journal -> j_dev_bd != NULL ) { | 2496 | journal->j_dev_file = NULL; |
2271 | result = blkdev_put( journal -> j_dev_bd ); | 2497 | journal->j_dev_bd = NULL; |
2272 | journal -> j_dev_bd = NULL; | 2498 | } else if (journal->j_dev_bd != NULL) { |
2273 | } | 2499 | result = blkdev_put(journal->j_dev_bd); |
2274 | 2500 | journal->j_dev_bd = NULL; | |
2275 | if( result != 0 ) { | 2501 | } |
2276 | reiserfs_warning(super, "sh-457: release_journal_dev: Cannot release journal device: %i", result ); | 2502 | |
2277 | } | 2503 | if (result != 0) { |
2278 | return result; | 2504 | reiserfs_warning(super, |
2279 | } | 2505 | "sh-457: release_journal_dev: Cannot release journal device: %i", |
2280 | 2506 | result); | |
2281 | static int journal_init_dev( struct super_block *super, | 2507 | } |
2282 | struct reiserfs_journal *journal, | 2508 | return result; |
2283 | const char *jdev_name ) | 2509 | } |
2510 | |||
2511 | static int journal_init_dev(struct super_block *super, | ||
2512 | struct reiserfs_journal *journal, | ||
2513 | const char *jdev_name) | ||
2284 | { | 2514 | { |
2285 | int result; | 2515 | int result; |
2286 | dev_t jdev; | 2516 | dev_t jdev; |
@@ -2289,50 +2519,51 @@ static int journal_init_dev( struct super_block *super, | |||
2289 | 2519 | ||
2290 | result = 0; | 2520 | result = 0; |
2291 | 2521 | ||
2292 | journal -> j_dev_bd = NULL; | 2522 | journal->j_dev_bd = NULL; |
2293 | journal -> j_dev_file = NULL; | 2523 | journal->j_dev_file = NULL; |
2294 | jdev = SB_ONDISK_JOURNAL_DEVICE( super ) ? | 2524 | jdev = SB_ONDISK_JOURNAL_DEVICE(super) ? |
2295 | new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev; | 2525 | new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev; |
2296 | 2526 | ||
2297 | if (bdev_read_only(super->s_bdev)) | 2527 | if (bdev_read_only(super->s_bdev)) |
2298 | blkdev_mode = FMODE_READ; | 2528 | blkdev_mode = FMODE_READ; |
2299 | 2529 | ||
2300 | /* there is no "jdev" option and journal is on separate device */ | 2530 | /* there is no "jdev" option and journal is on separate device */ |
2301 | if( ( !jdev_name || !jdev_name[ 0 ] ) ) { | 2531 | if ((!jdev_name || !jdev_name[0])) { |
2302 | journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode); | 2532 | journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode); |
2303 | if (IS_ERR(journal->j_dev_bd)) { | 2533 | if (IS_ERR(journal->j_dev_bd)) { |
2304 | result = PTR_ERR(journal->j_dev_bd); | 2534 | result = PTR_ERR(journal->j_dev_bd); |
2305 | journal->j_dev_bd = NULL; | 2535 | journal->j_dev_bd = NULL; |
2306 | reiserfs_warning (super, "sh-458: journal_init_dev: " | 2536 | reiserfs_warning(super, "sh-458: journal_init_dev: " |
2307 | "cannot init journal device '%s': %i", | 2537 | "cannot init journal device '%s': %i", |
2308 | __bdevname(jdev, b), result ); | 2538 | __bdevname(jdev, b), result); |
2309 | return result; | 2539 | return result; |
2310 | } else if (jdev != super->s_dev) | 2540 | } else if (jdev != super->s_dev) |
2311 | set_blocksize(journal->j_dev_bd, super->s_blocksize); | 2541 | set_blocksize(journal->j_dev_bd, super->s_blocksize); |
2312 | return 0; | 2542 | return 0; |
2313 | } | 2543 | } |
2314 | 2544 | ||
2315 | journal -> j_dev_file = filp_open( jdev_name, 0, 0 ); | 2545 | journal->j_dev_file = filp_open(jdev_name, 0, 0); |
2316 | if( !IS_ERR( journal -> j_dev_file ) ) { | 2546 | if (!IS_ERR(journal->j_dev_file)) { |
2317 | struct inode *jdev_inode = journal->j_dev_file->f_mapping->host; | 2547 | struct inode *jdev_inode = journal->j_dev_file->f_mapping->host; |
2318 | if( !S_ISBLK( jdev_inode -> i_mode ) ) { | 2548 | if (!S_ISBLK(jdev_inode->i_mode)) { |
2319 | reiserfs_warning(super, "journal_init_dev: '%s' is " | 2549 | reiserfs_warning(super, "journal_init_dev: '%s' is " |
2320 | "not a block device", jdev_name ); | 2550 | "not a block device", jdev_name); |
2321 | result = -ENOTBLK; | 2551 | result = -ENOTBLK; |
2322 | release_journal_dev( super, journal ); | 2552 | release_journal_dev(super, journal); |
2323 | } else { | 2553 | } else { |
2324 | /* ok */ | 2554 | /* ok */ |
2325 | journal->j_dev_bd = I_BDEV(jdev_inode); | 2555 | journal->j_dev_bd = I_BDEV(jdev_inode); |
2326 | set_blocksize(journal->j_dev_bd, super->s_blocksize); | 2556 | set_blocksize(journal->j_dev_bd, super->s_blocksize); |
2327 | reiserfs_info(super, "journal_init_dev: journal device: %s\n", | 2557 | reiserfs_info(super, |
2558 | "journal_init_dev: journal device: %s\n", | ||
2328 | bdevname(journal->j_dev_bd, b)); | 2559 | bdevname(journal->j_dev_bd, b)); |
2329 | } | 2560 | } |
2330 | } else { | 2561 | } else { |
2331 | result = PTR_ERR( journal -> j_dev_file ); | 2562 | result = PTR_ERR(journal->j_dev_file); |
2332 | journal -> j_dev_file = NULL; | 2563 | journal->j_dev_file = NULL; |
2333 | reiserfs_warning (super, | 2564 | reiserfs_warning(super, |
2334 | "journal_init_dev: Cannot open '%s': %i", | 2565 | "journal_init_dev: Cannot open '%s': %i", |
2335 | jdev_name, result ); | 2566 | jdev_name, result); |
2336 | } | 2567 | } |
2337 | return result; | 2568 | return result; |
2338 | } | 2569 | } |
@@ -2340,193 +2571,214 @@ static int journal_init_dev( struct super_block *super, | |||
2340 | /* | 2571 | /* |
2341 | ** must be called once on fs mount. calls journal_read for you | 2572 | ** must be called once on fs mount. calls journal_read for you |
2342 | */ | 2573 | */ |
2343 | int journal_init(struct super_block *p_s_sb, const char * j_dev_name, int old_format, unsigned int commit_max_age) { | 2574 | int journal_init(struct super_block *p_s_sb, const char *j_dev_name, |
2344 | int num_cnodes = SB_ONDISK_JOURNAL_SIZE(p_s_sb) * 2 ; | 2575 | int old_format, unsigned int commit_max_age) |
2345 | struct buffer_head *bhjh; | 2576 | { |
2346 | struct reiserfs_super_block * rs; | 2577 | int num_cnodes = SB_ONDISK_JOURNAL_SIZE(p_s_sb) * 2; |
2347 | struct reiserfs_journal_header *jh; | 2578 | struct buffer_head *bhjh; |
2348 | struct reiserfs_journal *journal; | 2579 | struct reiserfs_super_block *rs; |
2349 | struct reiserfs_journal_list *jl; | 2580 | struct reiserfs_journal_header *jh; |
2350 | char b[BDEVNAME_SIZE]; | 2581 | struct reiserfs_journal *journal; |
2351 | 2582 | struct reiserfs_journal_list *jl; | |
2352 | journal = SB_JOURNAL(p_s_sb) = vmalloc(sizeof (struct reiserfs_journal)) ; | 2583 | char b[BDEVNAME_SIZE]; |
2353 | if (!journal) { | 2584 | |
2354 | reiserfs_warning (p_s_sb, "journal-1256: unable to get memory for journal structure") ; | 2585 | journal = SB_JOURNAL(p_s_sb) = vmalloc(sizeof(struct reiserfs_journal)); |
2355 | return 1 ; | 2586 | if (!journal) { |
2356 | } | 2587 | reiserfs_warning(p_s_sb, |
2357 | memset(journal, 0, sizeof(struct reiserfs_journal)) ; | 2588 | "journal-1256: unable to get memory for journal structure"); |
2358 | INIT_LIST_HEAD(&journal->j_bitmap_nodes) ; | 2589 | return 1; |
2359 | INIT_LIST_HEAD (&journal->j_prealloc_list); | 2590 | } |
2360 | INIT_LIST_HEAD(&journal->j_working_list); | 2591 | memset(journal, 0, sizeof(struct reiserfs_journal)); |
2361 | INIT_LIST_HEAD(&journal->j_journal_list); | 2592 | INIT_LIST_HEAD(&journal->j_bitmap_nodes); |
2362 | journal->j_persistent_trans = 0; | 2593 | INIT_LIST_HEAD(&journal->j_prealloc_list); |
2363 | if (reiserfs_allocate_list_bitmaps(p_s_sb, | 2594 | INIT_LIST_HEAD(&journal->j_working_list); |
2364 | journal->j_list_bitmap, | 2595 | INIT_LIST_HEAD(&journal->j_journal_list); |
2365 | SB_BMAP_NR(p_s_sb))) | 2596 | journal->j_persistent_trans = 0; |
2366 | goto free_and_return ; | 2597 | if (reiserfs_allocate_list_bitmaps(p_s_sb, |
2367 | allocate_bitmap_nodes(p_s_sb) ; | 2598 | journal->j_list_bitmap, |
2368 | 2599 | SB_BMAP_NR(p_s_sb))) | |
2369 | /* reserved for journal area support */ | 2600 | goto free_and_return; |
2370 | SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) = (old_format ? | 2601 | allocate_bitmap_nodes(p_s_sb); |
2371 | REISERFS_OLD_DISK_OFFSET_IN_BYTES / p_s_sb->s_blocksize + | 2602 | |
2372 | SB_BMAP_NR(p_s_sb) + 1 : | 2603 | /* reserved for journal area support */ |
2373 | REISERFS_DISK_OFFSET_IN_BYTES / p_s_sb->s_blocksize + 2); | 2604 | SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) = (old_format ? |
2374 | 2605 | REISERFS_OLD_DISK_OFFSET_IN_BYTES | |
2375 | /* Sanity check to see is the standard journal fitting withing first bitmap | 2606 | / p_s_sb->s_blocksize + |
2376 | (actual for small blocksizes) */ | 2607 | SB_BMAP_NR(p_s_sb) + |
2377 | if ( !SB_ONDISK_JOURNAL_DEVICE( p_s_sb ) && | 2608 | 1 : |
2378 | (SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) + SB_ONDISK_JOURNAL_SIZE(p_s_sb) > p_s_sb->s_blocksize * 8) ) { | 2609 | REISERFS_DISK_OFFSET_IN_BYTES / |
2379 | reiserfs_warning (p_s_sb, "journal-1393: journal does not fit for area " | 2610 | p_s_sb->s_blocksize + 2); |
2380 | "addressed by first of bitmap blocks. It starts at " | 2611 | |
2381 | "%u and its size is %u. Block size %ld", | 2612 | /* Sanity check to see is the standard journal fitting withing first bitmap |
2382 | SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb), | 2613 | (actual for small blocksizes) */ |
2383 | SB_ONDISK_JOURNAL_SIZE(p_s_sb), p_s_sb->s_blocksize); | 2614 | if (!SB_ONDISK_JOURNAL_DEVICE(p_s_sb) && |
2384 | goto free_and_return; | 2615 | (SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) + |
2385 | } | 2616 | SB_ONDISK_JOURNAL_SIZE(p_s_sb) > p_s_sb->s_blocksize * 8)) { |
2386 | 2617 | reiserfs_warning(p_s_sb, | |
2387 | if( journal_init_dev( p_s_sb, journal, j_dev_name ) != 0 ) { | 2618 | "journal-1393: journal does not fit for area " |
2388 | reiserfs_warning (p_s_sb, "sh-462: unable to initialize jornal device"); | 2619 | "addressed by first of bitmap blocks. It starts at " |
2389 | goto free_and_return; | 2620 | "%u and its size is %u. Block size %ld", |
2390 | } | 2621 | SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb), |
2391 | 2622 | SB_ONDISK_JOURNAL_SIZE(p_s_sb), | |
2392 | rs = SB_DISK_SUPER_BLOCK(p_s_sb); | 2623 | p_s_sb->s_blocksize); |
2393 | 2624 | goto free_and_return; | |
2394 | /* read journal header */ | 2625 | } |
2395 | bhjh = journal_bread(p_s_sb, | 2626 | |
2396 | SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + SB_ONDISK_JOURNAL_SIZE(p_s_sb)); | 2627 | if (journal_init_dev(p_s_sb, journal, j_dev_name) != 0) { |
2397 | if (!bhjh) { | 2628 | reiserfs_warning(p_s_sb, |
2398 | reiserfs_warning (p_s_sb, "sh-459: unable to read journal header"); | 2629 | "sh-462: unable to initialize jornal device"); |
2399 | goto free_and_return; | 2630 | goto free_and_return; |
2400 | } | 2631 | } |
2401 | jh = (struct reiserfs_journal_header *)(bhjh->b_data); | 2632 | |
2402 | 2633 | rs = SB_DISK_SUPER_BLOCK(p_s_sb); | |
2403 | /* make sure that journal matches to the super block */ | 2634 | |
2404 | if (is_reiserfs_jr(rs) && (le32_to_cpu(jh->jh_journal.jp_journal_magic) != sb_jp_journal_magic(rs))) { | 2635 | /* read journal header */ |
2405 | reiserfs_warning (p_s_sb, "sh-460: journal header magic %x " | 2636 | bhjh = journal_bread(p_s_sb, |
2406 | "(device %s) does not match to magic found in super " | 2637 | SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + |
2407 | "block %x", | 2638 | SB_ONDISK_JOURNAL_SIZE(p_s_sb)); |
2408 | jh->jh_journal.jp_journal_magic, | 2639 | if (!bhjh) { |
2409 | bdevname( journal->j_dev_bd, b), | 2640 | reiserfs_warning(p_s_sb, |
2410 | sb_jp_journal_magic(rs)); | 2641 | "sh-459: unable to read journal header"); |
2411 | brelse (bhjh); | 2642 | goto free_and_return; |
2412 | goto free_and_return; | 2643 | } |
2413 | } | 2644 | jh = (struct reiserfs_journal_header *)(bhjh->b_data); |
2414 | 2645 | ||
2415 | journal->j_trans_max = le32_to_cpu (jh->jh_journal.jp_journal_trans_max); | 2646 | /* make sure that journal matches to the super block */ |
2416 | journal->j_max_batch = le32_to_cpu (jh->jh_journal.jp_journal_max_batch); | 2647 | if (is_reiserfs_jr(rs) |
2417 | journal->j_max_commit_age = le32_to_cpu (jh->jh_journal.jp_journal_max_commit_age); | 2648 | && (le32_to_cpu(jh->jh_journal.jp_journal_magic) != |
2418 | journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE; | 2649 | sb_jp_journal_magic(rs))) { |
2419 | 2650 | reiserfs_warning(p_s_sb, | |
2420 | if (journal->j_trans_max) { | 2651 | "sh-460: journal header magic %x " |
2421 | /* make sure these parameters are available, assign it if they are not */ | 2652 | "(device %s) does not match to magic found in super " |
2422 | __u32 initial = journal->j_trans_max; | 2653 | "block %x", jh->jh_journal.jp_journal_magic, |
2423 | __u32 ratio = 1; | 2654 | bdevname(journal->j_dev_bd, b), |
2424 | 2655 | sb_jp_journal_magic(rs)); | |
2425 | if (p_s_sb->s_blocksize < 4096) | 2656 | brelse(bhjh); |
2426 | ratio = 4096 / p_s_sb->s_blocksize; | 2657 | goto free_and_return; |
2427 | 2658 | } | |
2428 | if (SB_ONDISK_JOURNAL_SIZE(p_s_sb)/journal->j_trans_max < JOURNAL_MIN_RATIO) | 2659 | |
2429 | journal->j_trans_max = SB_ONDISK_JOURNAL_SIZE(p_s_sb) / JOURNAL_MIN_RATIO; | 2660 | journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max); |
2430 | if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio) | 2661 | journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch); |
2431 | journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT / ratio; | 2662 | journal->j_max_commit_age = |
2432 | if (journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio) | 2663 | le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age); |
2433 | journal->j_trans_max = JOURNAL_TRANS_MIN_DEFAULT / ratio; | 2664 | journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE; |
2434 | 2665 | ||
2435 | if (journal->j_trans_max != initial) | 2666 | if (journal->j_trans_max) { |
2436 | reiserfs_warning (p_s_sb, "sh-461: journal_init: wrong transaction max size (%u). Changed to %u", | 2667 | /* make sure these parameters are available, assign it if they are not */ |
2437 | initial, journal->j_trans_max); | 2668 | __u32 initial = journal->j_trans_max; |
2438 | 2669 | __u32 ratio = 1; | |
2439 | journal->j_max_batch = journal->j_trans_max* | 2670 | |
2440 | JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT; | 2671 | if (p_s_sb->s_blocksize < 4096) |
2441 | } | 2672 | ratio = 4096 / p_s_sb->s_blocksize; |
2442 | 2673 | ||
2443 | if (!journal->j_trans_max) { | 2674 | if (SB_ONDISK_JOURNAL_SIZE(p_s_sb) / journal->j_trans_max < |
2444 | /*we have the file system was created by old version of mkreiserfs | 2675 | JOURNAL_MIN_RATIO) |
2445 | so this field contains zero value */ | 2676 | journal->j_trans_max = |
2446 | journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT ; | 2677 | SB_ONDISK_JOURNAL_SIZE(p_s_sb) / JOURNAL_MIN_RATIO; |
2447 | journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT ; | 2678 | if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio) |
2448 | journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE ; | 2679 | journal->j_trans_max = |
2449 | 2680 | JOURNAL_TRANS_MAX_DEFAULT / ratio; | |
2450 | /* for blocksize >= 4096 - max transaction size is 1024. For block size < 4096 | 2681 | if (journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio) |
2451 | trans max size is decreased proportionally */ | 2682 | journal->j_trans_max = |
2452 | if (p_s_sb->s_blocksize < 4096) { | 2683 | JOURNAL_TRANS_MIN_DEFAULT / ratio; |
2453 | journal->j_trans_max /= (4096 / p_s_sb->s_blocksize) ; | 2684 | |
2454 | journal->j_max_batch = (journal->j_trans_max) * 9 / 10 ; | 2685 | if (journal->j_trans_max != initial) |
2455 | } | 2686 | reiserfs_warning(p_s_sb, |
2456 | } | 2687 | "sh-461: journal_init: wrong transaction max size (%u). Changed to %u", |
2457 | 2688 | initial, journal->j_trans_max); | |
2458 | journal->j_default_max_commit_age = journal->j_max_commit_age; | 2689 | |
2459 | 2690 | journal->j_max_batch = journal->j_trans_max * | |
2460 | if (commit_max_age != 0) { | 2691 | JOURNAL_MAX_BATCH_DEFAULT / JOURNAL_TRANS_MAX_DEFAULT; |
2461 | journal->j_max_commit_age = commit_max_age; | 2692 | } |
2462 | journal->j_max_trans_age = commit_max_age; | 2693 | |
2463 | } | 2694 | if (!journal->j_trans_max) { |
2464 | 2695 | /*we have the file system was created by old version of mkreiserfs | |
2465 | reiserfs_info (p_s_sb, "journal params: device %s, size %u, " | 2696 | so this field contains zero value */ |
2466 | "journal first block %u, max trans len %u, max batch %u, " | 2697 | journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT; |
2467 | "max commit age %u, max trans age %u\n", | 2698 | journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT; |
2468 | bdevname( journal->j_dev_bd, b), | 2699 | journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE; |
2469 | SB_ONDISK_JOURNAL_SIZE(p_s_sb), | 2700 | |
2470 | SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), | 2701 | /* for blocksize >= 4096 - max transaction size is 1024. For block size < 4096 |
2471 | journal->j_trans_max, | 2702 | trans max size is decreased proportionally */ |
2472 | journal->j_max_batch, | 2703 | if (p_s_sb->s_blocksize < 4096) { |
2473 | journal->j_max_commit_age, | 2704 | journal->j_trans_max /= (4096 / p_s_sb->s_blocksize); |
2474 | journal->j_max_trans_age); | 2705 | journal->j_max_batch = (journal->j_trans_max) * 9 / 10; |
2475 | 2706 | } | |
2476 | brelse (bhjh); | 2707 | } |
2477 | 2708 | ||
2478 | journal->j_list_bitmap_index = 0 ; | 2709 | journal->j_default_max_commit_age = journal->j_max_commit_age; |
2479 | journal_list_init(p_s_sb) ; | 2710 | |
2480 | 2711 | if (commit_max_age != 0) { | |
2481 | memset(journal->j_list_hash_table, 0, JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)) ; | 2712 | journal->j_max_commit_age = commit_max_age; |
2482 | 2713 | journal->j_max_trans_age = commit_max_age; | |
2483 | INIT_LIST_HEAD(&journal->j_dirty_buffers) ; | 2714 | } |
2484 | spin_lock_init(&journal->j_dirty_buffers_lock) ; | 2715 | |
2485 | 2716 | reiserfs_info(p_s_sb, "journal params: device %s, size %u, " | |
2486 | journal->j_start = 0 ; | 2717 | "journal first block %u, max trans len %u, max batch %u, " |
2487 | journal->j_len = 0 ; | 2718 | "max commit age %u, max trans age %u\n", |
2488 | journal->j_len_alloc = 0 ; | 2719 | bdevname(journal->j_dev_bd, b), |
2489 | atomic_set(&(journal->j_wcount), 0) ; | 2720 | SB_ONDISK_JOURNAL_SIZE(p_s_sb), |
2490 | atomic_set(&(journal->j_async_throttle), 0) ; | 2721 | SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb), |
2491 | journal->j_bcount = 0 ; | 2722 | journal->j_trans_max, |
2492 | journal->j_trans_start_time = 0 ; | 2723 | journal->j_max_batch, |
2493 | journal->j_last = NULL ; | 2724 | journal->j_max_commit_age, journal->j_max_trans_age); |
2494 | journal->j_first = NULL ; | 2725 | |
2495 | init_waitqueue_head(&(journal->j_join_wait)) ; | 2726 | brelse(bhjh); |
2496 | sema_init(&journal->j_lock, 1); | 2727 | |
2497 | sema_init(&journal->j_flush_sem, 1); | 2728 | journal->j_list_bitmap_index = 0; |
2498 | 2729 | journal_list_init(p_s_sb); | |
2499 | journal->j_trans_id = 10 ; | 2730 | |
2500 | journal->j_mount_id = 10 ; | 2731 | memset(journal->j_list_hash_table, 0, |
2501 | journal->j_state = 0 ; | 2732 | JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *)); |
2502 | atomic_set(&(journal->j_jlock), 0) ; | 2733 | |
2503 | journal->j_cnode_free_list = allocate_cnodes(num_cnodes) ; | 2734 | INIT_LIST_HEAD(&journal->j_dirty_buffers); |
2504 | journal->j_cnode_free_orig = journal->j_cnode_free_list ; | 2735 | spin_lock_init(&journal->j_dirty_buffers_lock); |
2505 | journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0 ; | 2736 | |
2506 | journal->j_cnode_used = 0 ; | 2737 | journal->j_start = 0; |
2507 | journal->j_must_wait = 0 ; | 2738 | journal->j_len = 0; |
2508 | 2739 | journal->j_len_alloc = 0; | |
2509 | init_journal_hash(p_s_sb) ; | 2740 | atomic_set(&(journal->j_wcount), 0); |
2510 | jl = journal->j_current_jl; | 2741 | atomic_set(&(journal->j_async_throttle), 0); |
2511 | jl->j_list_bitmap = get_list_bitmap(p_s_sb, jl); | 2742 | journal->j_bcount = 0; |
2512 | if (!jl->j_list_bitmap) { | 2743 | journal->j_trans_start_time = 0; |
2513 | reiserfs_warning(p_s_sb, "journal-2005, get_list_bitmap failed for journal list 0") ; | 2744 | journal->j_last = NULL; |
2514 | goto free_and_return; | 2745 | journal->j_first = NULL; |
2515 | } | 2746 | init_waitqueue_head(&(journal->j_join_wait)); |
2516 | if (journal_read(p_s_sb) < 0) { | 2747 | sema_init(&journal->j_lock, 1); |
2517 | reiserfs_warning(p_s_sb, "Replay Failure, unable to mount") ; | 2748 | sema_init(&journal->j_flush_sem, 1); |
2518 | goto free_and_return; | 2749 | |
2519 | } | 2750 | journal->j_trans_id = 10; |
2520 | 2751 | journal->j_mount_id = 10; | |
2521 | reiserfs_mounted_fs_count++ ; | 2752 | journal->j_state = 0; |
2522 | if (reiserfs_mounted_fs_count <= 1) | 2753 | atomic_set(&(journal->j_jlock), 0); |
2523 | commit_wq = create_workqueue("reiserfs"); | 2754 | journal->j_cnode_free_list = allocate_cnodes(num_cnodes); |
2524 | 2755 | journal->j_cnode_free_orig = journal->j_cnode_free_list; | |
2525 | INIT_WORK(&journal->j_work, flush_async_commits, p_s_sb); | 2756 | journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0; |
2526 | return 0 ; | 2757 | journal->j_cnode_used = 0; |
2527 | free_and_return: | 2758 | journal->j_must_wait = 0; |
2528 | free_journal_ram(p_s_sb); | 2759 | |
2529 | return 1; | 2760 | init_journal_hash(p_s_sb); |
2761 | jl = journal->j_current_jl; | ||
2762 | jl->j_list_bitmap = get_list_bitmap(p_s_sb, jl); | ||
2763 | if (!jl->j_list_bitmap) { | ||
2764 | reiserfs_warning(p_s_sb, | ||
2765 | "journal-2005, get_list_bitmap failed for journal list 0"); | ||
2766 | goto free_and_return; | ||
2767 | } | ||
2768 | if (journal_read(p_s_sb) < 0) { | ||
2769 | reiserfs_warning(p_s_sb, "Replay Failure, unable to mount"); | ||
2770 | goto free_and_return; | ||
2771 | } | ||
2772 | |||
2773 | reiserfs_mounted_fs_count++; | ||
2774 | if (reiserfs_mounted_fs_count <= 1) | ||
2775 | commit_wq = create_workqueue("reiserfs"); | ||
2776 | |||
2777 | INIT_WORK(&journal->j_work, flush_async_commits, p_s_sb); | ||
2778 | return 0; | ||
2779 | free_and_return: | ||
2780 | free_journal_ram(p_s_sb); | ||
2781 | return 1; | ||
2530 | } | 2782 | } |
2531 | 2783 | ||
2532 | /* | 2784 | /* |
@@ -2534,96 +2786,102 @@ free_and_return: | |||
2534 | ** be used by delete to make sure they don't write more than can fit inside a single | 2786 | ** be used by delete to make sure they don't write more than can fit inside a single |
2535 | ** transaction | 2787 | ** transaction |
2536 | */ | 2788 | */ |
2537 | int journal_transaction_should_end(struct reiserfs_transaction_handle *th, int new_alloc) { | 2789 | int journal_transaction_should_end(struct reiserfs_transaction_handle *th, |
2538 | struct reiserfs_journal *journal = SB_JOURNAL (th->t_super); | 2790 | int new_alloc) |
2539 | time_t now = get_seconds() ; | 2791 | { |
2540 | /* cannot restart while nested */ | 2792 | struct reiserfs_journal *journal = SB_JOURNAL(th->t_super); |
2541 | BUG_ON (!th->t_trans_id); | 2793 | time_t now = get_seconds(); |
2542 | if (th->t_refcount > 1) | 2794 | /* cannot restart while nested */ |
2543 | return 0 ; | 2795 | BUG_ON(!th->t_trans_id); |
2544 | if ( journal->j_must_wait > 0 || | 2796 | if (th->t_refcount > 1) |
2545 | (journal->j_len_alloc + new_alloc) >= journal->j_max_batch || | 2797 | return 0; |
2546 | atomic_read(&(journal->j_jlock)) || | 2798 | if (journal->j_must_wait > 0 || |
2547 | (now - journal->j_trans_start_time) > journal->j_max_trans_age || | 2799 | (journal->j_len_alloc + new_alloc) >= journal->j_max_batch || |
2548 | journal->j_cnode_free < (journal->j_trans_max * 3)) { | 2800 | atomic_read(&(journal->j_jlock)) || |
2549 | return 1 ; | 2801 | (now - journal->j_trans_start_time) > journal->j_max_trans_age || |
2550 | } | 2802 | journal->j_cnode_free < (journal->j_trans_max * 3)) { |
2551 | return 0 ; | 2803 | return 1; |
2804 | } | ||
2805 | return 0; | ||
2552 | } | 2806 | } |
2553 | 2807 | ||
2554 | /* this must be called inside a transaction, and requires the | 2808 | /* this must be called inside a transaction, and requires the |
2555 | ** kernel_lock to be held | 2809 | ** kernel_lock to be held |
2556 | */ | 2810 | */ |
2557 | void reiserfs_block_writes(struct reiserfs_transaction_handle *th) { | 2811 | void reiserfs_block_writes(struct reiserfs_transaction_handle *th) |
2558 | struct reiserfs_journal *journal = SB_JOURNAL (th->t_super); | 2812 | { |
2559 | BUG_ON (!th->t_trans_id); | 2813 | struct reiserfs_journal *journal = SB_JOURNAL(th->t_super); |
2560 | journal->j_must_wait = 1 ; | 2814 | BUG_ON(!th->t_trans_id); |
2561 | set_bit(J_WRITERS_BLOCKED, &journal->j_state) ; | 2815 | journal->j_must_wait = 1; |
2562 | return ; | 2816 | set_bit(J_WRITERS_BLOCKED, &journal->j_state); |
2817 | return; | ||
2563 | } | 2818 | } |
2564 | 2819 | ||
2565 | /* this must be called without a transaction started, and does not | 2820 | /* this must be called without a transaction started, and does not |
2566 | ** require BKL | 2821 | ** require BKL |
2567 | */ | 2822 | */ |
2568 | void reiserfs_allow_writes(struct super_block *s) { | 2823 | void reiserfs_allow_writes(struct super_block *s) |
2569 | struct reiserfs_journal *journal = SB_JOURNAL (s); | 2824 | { |
2570 | clear_bit(J_WRITERS_BLOCKED, &journal->j_state) ; | 2825 | struct reiserfs_journal *journal = SB_JOURNAL(s); |
2571 | wake_up(&journal->j_join_wait) ; | 2826 | clear_bit(J_WRITERS_BLOCKED, &journal->j_state); |
2827 | wake_up(&journal->j_join_wait); | ||
2572 | } | 2828 | } |
2573 | 2829 | ||
2574 | /* this must be called without a transaction started, and does not | 2830 | /* this must be called without a transaction started, and does not |
2575 | ** require BKL | 2831 | ** require BKL |
2576 | */ | 2832 | */ |
2577 | void reiserfs_wait_on_write_block(struct super_block *s) { | 2833 | void reiserfs_wait_on_write_block(struct super_block *s) |
2578 | struct reiserfs_journal *journal = SB_JOURNAL (s); | 2834 | { |
2579 | wait_event(journal->j_join_wait, | 2835 | struct reiserfs_journal *journal = SB_JOURNAL(s); |
2580 | !test_bit(J_WRITERS_BLOCKED, &journal->j_state)) ; | 2836 | wait_event(journal->j_join_wait, |
2581 | } | 2837 | !test_bit(J_WRITERS_BLOCKED, &journal->j_state)); |
2582 | 2838 | } | |
2583 | static void queue_log_writer(struct super_block *s) { | 2839 | |
2584 | wait_queue_t wait; | 2840 | static void queue_log_writer(struct super_block *s) |
2585 | struct reiserfs_journal *journal = SB_JOURNAL (s); | 2841 | { |
2586 | set_bit(J_WRITERS_QUEUED, &journal->j_state); | 2842 | wait_queue_t wait; |
2587 | 2843 | struct reiserfs_journal *journal = SB_JOURNAL(s); | |
2588 | /* | 2844 | set_bit(J_WRITERS_QUEUED, &journal->j_state); |
2589 | * we don't want to use wait_event here because | 2845 | |
2590 | * we only want to wait once. | 2846 | /* |
2591 | */ | 2847 | * we don't want to use wait_event here because |
2592 | init_waitqueue_entry(&wait, current); | 2848 | * we only want to wait once. |
2593 | add_wait_queue(&journal->j_join_wait, &wait); | 2849 | */ |
2594 | set_current_state(TASK_UNINTERRUPTIBLE); | 2850 | init_waitqueue_entry(&wait, current); |
2595 | if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) | 2851 | add_wait_queue(&journal->j_join_wait, &wait); |
2596 | schedule(); | ||
2597 | current->state = TASK_RUNNING; | ||
2598 | remove_wait_queue(&journal->j_join_wait, &wait); | ||
2599 | } | ||
2600 | |||
2601 | static void wake_queued_writers(struct super_block *s) { | ||
2602 | struct reiserfs_journal *journal = SB_JOURNAL (s); | ||
2603 | if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state)) | ||
2604 | wake_up(&journal->j_join_wait); | ||
2605 | } | ||
2606 | |||
2607 | static void let_transaction_grow(struct super_block *sb, | ||
2608 | unsigned long trans_id) | ||
2609 | { | ||
2610 | struct reiserfs_journal *journal = SB_JOURNAL (sb); | ||
2611 | unsigned long bcount = journal->j_bcount; | ||
2612 | while(1) { | ||
2613 | set_current_state(TASK_UNINTERRUPTIBLE); | 2852 | set_current_state(TASK_UNINTERRUPTIBLE); |
2614 | schedule_timeout(1); | 2853 | if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) |
2615 | journal->j_current_jl->j_state |= LIST_COMMIT_PENDING; | 2854 | schedule(); |
2616 | while ((atomic_read(&journal->j_wcount) > 0 || | 2855 | current->state = TASK_RUNNING; |
2617 | atomic_read(&journal->j_jlock)) && | 2856 | remove_wait_queue(&journal->j_join_wait, &wait); |
2618 | journal->j_trans_id == trans_id) { | 2857 | } |
2619 | queue_log_writer(sb); | 2858 | |
2859 | static void wake_queued_writers(struct super_block *s) | ||
2860 | { | ||
2861 | struct reiserfs_journal *journal = SB_JOURNAL(s); | ||
2862 | if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state)) | ||
2863 | wake_up(&journal->j_join_wait); | ||
2864 | } | ||
2865 | |||
2866 | static void let_transaction_grow(struct super_block *sb, unsigned long trans_id) | ||
2867 | { | ||
2868 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | ||
2869 | unsigned long bcount = journal->j_bcount; | ||
2870 | while (1) { | ||
2871 | set_current_state(TASK_UNINTERRUPTIBLE); | ||
2872 | schedule_timeout(1); | ||
2873 | journal->j_current_jl->j_state |= LIST_COMMIT_PENDING; | ||
2874 | while ((atomic_read(&journal->j_wcount) > 0 || | ||
2875 | atomic_read(&journal->j_jlock)) && | ||
2876 | journal->j_trans_id == trans_id) { | ||
2877 | queue_log_writer(sb); | ||
2878 | } | ||
2879 | if (journal->j_trans_id != trans_id) | ||
2880 | break; | ||
2881 | if (bcount == journal->j_bcount) | ||
2882 | break; | ||
2883 | bcount = journal->j_bcount; | ||
2620 | } | 2884 | } |
2621 | if (journal->j_trans_id != trans_id) | ||
2622 | break; | ||
2623 | if (bcount == journal->j_bcount) | ||
2624 | break; | ||
2625 | bcount = journal->j_bcount; | ||
2626 | } | ||
2627 | } | 2885 | } |
2628 | 2886 | ||
2629 | /* join == true if you must join an existing transaction. | 2887 | /* join == true if you must join an existing transaction. |
@@ -2632,224 +2890,244 @@ static void let_transaction_grow(struct super_block *sb, | |||
2632 | ** this will block until the transaction is joinable. send the number of blocks you | 2890 | ** this will block until the transaction is joinable. send the number of blocks you |
2633 | ** expect to use in nblocks. | 2891 | ** expect to use in nblocks. |
2634 | */ | 2892 | */ |
2635 | static int do_journal_begin_r(struct reiserfs_transaction_handle *th, struct super_block * p_s_sb,unsigned long nblocks,int join) { | 2893 | static int do_journal_begin_r(struct reiserfs_transaction_handle *th, |
2636 | time_t now = get_seconds() ; | 2894 | struct super_block *p_s_sb, unsigned long nblocks, |
2637 | int old_trans_id ; | 2895 | int join) |
2638 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); | 2896 | { |
2639 | struct reiserfs_transaction_handle myth; | 2897 | time_t now = get_seconds(); |
2640 | int sched_count = 0; | 2898 | int old_trans_id; |
2641 | int retval; | 2899 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); |
2642 | 2900 | struct reiserfs_transaction_handle myth; | |
2643 | reiserfs_check_lock_depth(p_s_sb, "journal_begin") ; | 2901 | int sched_count = 0; |
2644 | if (nblocks > journal->j_trans_max) | 2902 | int retval; |
2645 | BUG(); | 2903 | |
2646 | 2904 | reiserfs_check_lock_depth(p_s_sb, "journal_begin"); | |
2647 | PROC_INFO_INC( p_s_sb, journal.journal_being ); | 2905 | if (nblocks > journal->j_trans_max) |
2648 | /* set here for journal_join */ | 2906 | BUG(); |
2649 | th->t_refcount = 1; | 2907 | |
2650 | th->t_super = p_s_sb ; | 2908 | PROC_INFO_INC(p_s_sb, journal.journal_being); |
2651 | 2909 | /* set here for journal_join */ | |
2652 | relock: | 2910 | th->t_refcount = 1; |
2653 | lock_journal(p_s_sb) ; | 2911 | th->t_super = p_s_sb; |
2654 | if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted (journal)) { | 2912 | |
2655 | unlock_journal (p_s_sb); | 2913 | relock: |
2656 | retval = journal->j_errno; | 2914 | lock_journal(p_s_sb); |
2657 | goto out_fail; | 2915 | if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) { |
2658 | } | 2916 | unlock_journal(p_s_sb); |
2659 | journal->j_bcount++; | 2917 | retval = journal->j_errno; |
2660 | 2918 | goto out_fail; | |
2661 | if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) { | 2919 | } |
2662 | unlock_journal(p_s_sb) ; | 2920 | journal->j_bcount++; |
2663 | reiserfs_wait_on_write_block(p_s_sb) ; | 2921 | |
2664 | PROC_INFO_INC( p_s_sb, journal.journal_relock_writers ); | 2922 | if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) { |
2665 | goto relock ; | 2923 | unlock_journal(p_s_sb); |
2666 | } | 2924 | reiserfs_wait_on_write_block(p_s_sb); |
2667 | now = get_seconds(); | 2925 | PROC_INFO_INC(p_s_sb, journal.journal_relock_writers); |
2668 | 2926 | goto relock; | |
2669 | /* if there is no room in the journal OR | 2927 | } |
2670 | ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning | 2928 | now = get_seconds(); |
2671 | ** we don't sleep if there aren't other writers | 2929 | |
2672 | */ | 2930 | /* if there is no room in the journal OR |
2673 | 2931 | ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning | |
2674 | if ( (!join && journal->j_must_wait > 0) || | 2932 | ** we don't sleep if there aren't other writers |
2675 | ( !join && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch) || | 2933 | */ |
2676 | (!join && atomic_read(&journal->j_wcount) > 0 && journal->j_trans_start_time > 0 && | 2934 | |
2677 | (now - journal->j_trans_start_time) > journal->j_max_trans_age) || | 2935 | if ((!join && journal->j_must_wait > 0) || |
2678 | (!join && atomic_read(&journal->j_jlock)) || | 2936 | (!join |
2679 | (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) { | 2937 | && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch) |
2680 | 2938 | || (!join && atomic_read(&journal->j_wcount) > 0 | |
2681 | old_trans_id = journal->j_trans_id; | 2939 | && journal->j_trans_start_time > 0 |
2682 | unlock_journal(p_s_sb) ; /* allow others to finish this transaction */ | 2940 | && (now - journal->j_trans_start_time) > |
2683 | 2941 | journal->j_max_trans_age) || (!join | |
2684 | if (!join && (journal->j_len_alloc + nblocks + 2) >= | 2942 | && atomic_read(&journal->j_jlock)) |
2685 | journal->j_max_batch && | 2943 | || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) { |
2686 | ((journal->j_len + nblocks + 2) * 100) < (journal->j_len_alloc * 75)) | 2944 | |
2687 | { | 2945 | old_trans_id = journal->j_trans_id; |
2688 | if (atomic_read(&journal->j_wcount) > 10) { | 2946 | unlock_journal(p_s_sb); /* allow others to finish this transaction */ |
2689 | sched_count++; | 2947 | |
2690 | queue_log_writer(p_s_sb); | 2948 | if (!join && (journal->j_len_alloc + nblocks + 2) >= |
2691 | goto relock; | 2949 | journal->j_max_batch && |
2692 | } | 2950 | ((journal->j_len + nblocks + 2) * 100) < |
2693 | } | 2951 | (journal->j_len_alloc * 75)) { |
2694 | /* don't mess with joining the transaction if all we have to do is | 2952 | if (atomic_read(&journal->j_wcount) > 10) { |
2695 | * wait for someone else to do a commit | 2953 | sched_count++; |
2696 | */ | 2954 | queue_log_writer(p_s_sb); |
2697 | if (atomic_read(&journal->j_jlock)) { | 2955 | goto relock; |
2698 | while (journal->j_trans_id == old_trans_id && | 2956 | } |
2699 | atomic_read(&journal->j_jlock)) { | 2957 | } |
2700 | queue_log_writer(p_s_sb); | 2958 | /* don't mess with joining the transaction if all we have to do is |
2701 | } | 2959 | * wait for someone else to do a commit |
2702 | goto relock; | 2960 | */ |
2703 | } | 2961 | if (atomic_read(&journal->j_jlock)) { |
2704 | retval = journal_join(&myth, p_s_sb, 1) ; | 2962 | while (journal->j_trans_id == old_trans_id && |
2705 | if (retval) | 2963 | atomic_read(&journal->j_jlock)) { |
2706 | goto out_fail; | 2964 | queue_log_writer(p_s_sb); |
2707 | 2965 | } | |
2708 | /* someone might have ended the transaction while we joined */ | 2966 | goto relock; |
2709 | if (old_trans_id != journal->j_trans_id) { | 2967 | } |
2710 | retval = do_journal_end(&myth, p_s_sb, 1, 0) ; | 2968 | retval = journal_join(&myth, p_s_sb, 1); |
2711 | } else { | 2969 | if (retval) |
2712 | retval = do_journal_end(&myth, p_s_sb, 1, COMMIT_NOW) ; | 2970 | goto out_fail; |
2713 | } | 2971 | |
2714 | 2972 | /* someone might have ended the transaction while we joined */ | |
2715 | if (retval) | 2973 | if (old_trans_id != journal->j_trans_id) { |
2716 | goto out_fail; | 2974 | retval = do_journal_end(&myth, p_s_sb, 1, 0); |
2717 | 2975 | } else { | |
2718 | PROC_INFO_INC( p_s_sb, journal.journal_relock_wcount ); | 2976 | retval = do_journal_end(&myth, p_s_sb, 1, COMMIT_NOW); |
2719 | goto relock ; | 2977 | } |
2720 | } | 2978 | |
2721 | /* we are the first writer, set trans_id */ | 2979 | if (retval) |
2722 | if (journal->j_trans_start_time == 0) { | 2980 | goto out_fail; |
2723 | journal->j_trans_start_time = get_seconds(); | 2981 | |
2724 | } | 2982 | PROC_INFO_INC(p_s_sb, journal.journal_relock_wcount); |
2725 | atomic_inc(&(journal->j_wcount)) ; | 2983 | goto relock; |
2726 | journal->j_len_alloc += nblocks ; | 2984 | } |
2727 | th->t_blocks_logged = 0 ; | 2985 | /* we are the first writer, set trans_id */ |
2728 | th->t_blocks_allocated = nblocks ; | 2986 | if (journal->j_trans_start_time == 0) { |
2729 | th->t_trans_id = journal->j_trans_id ; | 2987 | journal->j_trans_start_time = get_seconds(); |
2730 | unlock_journal(p_s_sb) ; | 2988 | } |
2731 | INIT_LIST_HEAD (&th->t_list); | 2989 | atomic_inc(&(journal->j_wcount)); |
2732 | get_fs_excl(); | 2990 | journal->j_len_alloc += nblocks; |
2733 | return 0 ; | 2991 | th->t_blocks_logged = 0; |
2734 | 2992 | th->t_blocks_allocated = nblocks; | |
2735 | out_fail: | 2993 | th->t_trans_id = journal->j_trans_id; |
2736 | memset (th, 0, sizeof (*th)); | 2994 | unlock_journal(p_s_sb); |
2737 | /* Re-set th->t_super, so we can properly keep track of how many | 2995 | INIT_LIST_HEAD(&th->t_list); |
2738 | * persistent transactions there are. We need to do this so if this | 2996 | get_fs_excl(); |
2739 | * call is part of a failed restart_transaction, we can free it later */ | 2997 | return 0; |
2740 | th->t_super = p_s_sb; | 2998 | |
2741 | return retval; | 2999 | out_fail: |
2742 | } | 3000 | memset(th, 0, sizeof(*th)); |
2743 | 3001 | /* Re-set th->t_super, so we can properly keep track of how many | |
2744 | struct reiserfs_transaction_handle * | 3002 | * persistent transactions there are. We need to do this so if this |
2745 | reiserfs_persistent_transaction(struct super_block *s, int nblocks) { | 3003 | * call is part of a failed restart_transaction, we can free it later */ |
2746 | int ret ; | 3004 | th->t_super = p_s_sb; |
2747 | struct reiserfs_transaction_handle *th ; | 3005 | return retval; |
2748 | 3006 | } | |
2749 | /* if we're nesting into an existing transaction. It will be | 3007 | |
2750 | ** persistent on its own | 3008 | struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct |
2751 | */ | 3009 | super_block |
2752 | if (reiserfs_transaction_running(s)) { | 3010 | *s, |
2753 | th = current->journal_info ; | 3011 | int nblocks) |
2754 | th->t_refcount++ ; | 3012 | { |
2755 | if (th->t_refcount < 2) { | 3013 | int ret; |
2756 | BUG() ; | 3014 | struct reiserfs_transaction_handle *th; |
2757 | } | 3015 | |
2758 | return th ; | 3016 | /* if we're nesting into an existing transaction. It will be |
2759 | } | 3017 | ** persistent on its own |
2760 | th = reiserfs_kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS, s) ; | 3018 | */ |
2761 | if (!th) | 3019 | if (reiserfs_transaction_running(s)) { |
2762 | return NULL; | 3020 | th = current->journal_info; |
2763 | ret = journal_begin(th, s, nblocks) ; | 3021 | th->t_refcount++; |
2764 | if (ret) { | 3022 | if (th->t_refcount < 2) { |
2765 | reiserfs_kfree(th, sizeof(struct reiserfs_transaction_handle), s) ; | 3023 | BUG(); |
2766 | return NULL; | 3024 | } |
2767 | } | 3025 | return th; |
2768 | 3026 | } | |
2769 | SB_JOURNAL(s)->j_persistent_trans++; | 3027 | th = reiserfs_kmalloc(sizeof(struct reiserfs_transaction_handle), |
2770 | return th ; | 3028 | GFP_NOFS, s); |
2771 | } | 3029 | if (!th) |
2772 | 3030 | return NULL; | |
2773 | int | 3031 | ret = journal_begin(th, s, nblocks); |
2774 | reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th) { | 3032 | if (ret) { |
2775 | struct super_block *s = th->t_super; | 3033 | reiserfs_kfree(th, sizeof(struct reiserfs_transaction_handle), |
2776 | int ret = 0; | 3034 | s); |
2777 | if (th->t_trans_id) | 3035 | return NULL; |
2778 | ret = journal_end(th, th->t_super, th->t_blocks_allocated); | 3036 | } |
2779 | else | 3037 | |
2780 | ret = -EIO; | 3038 | SB_JOURNAL(s)->j_persistent_trans++; |
2781 | if (th->t_refcount == 0) { | 3039 | return th; |
2782 | SB_JOURNAL(s)->j_persistent_trans--; | 3040 | } |
2783 | reiserfs_kfree(th, sizeof(struct reiserfs_transaction_handle), s) ; | 3041 | |
2784 | } | 3042 | int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th) |
2785 | return ret; | 3043 | { |
2786 | } | 3044 | struct super_block *s = th->t_super; |
2787 | 3045 | int ret = 0; | |
2788 | static int journal_join(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long nblocks) { | 3046 | if (th->t_trans_id) |
2789 | struct reiserfs_transaction_handle *cur_th = current->journal_info; | 3047 | ret = journal_end(th, th->t_super, th->t_blocks_allocated); |
2790 | 3048 | else | |
2791 | /* this keeps do_journal_end from NULLing out the current->journal_info | 3049 | ret = -EIO; |
2792 | ** pointer | 3050 | if (th->t_refcount == 0) { |
2793 | */ | 3051 | SB_JOURNAL(s)->j_persistent_trans--; |
2794 | th->t_handle_save = cur_th ; | 3052 | reiserfs_kfree(th, sizeof(struct reiserfs_transaction_handle), |
2795 | if (cur_th && cur_th->t_refcount > 1) { | 3053 | s); |
2796 | BUG() ; | 3054 | } |
2797 | } | 3055 | return ret; |
2798 | return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_JOIN) ; | 3056 | } |
2799 | } | 3057 | |
2800 | 3058 | static int journal_join(struct reiserfs_transaction_handle *th, | |
2801 | int journal_join_abort(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long nblocks) { | 3059 | struct super_block *p_s_sb, unsigned long nblocks) |
2802 | struct reiserfs_transaction_handle *cur_th = current->journal_info; | 3060 | { |
2803 | 3061 | struct reiserfs_transaction_handle *cur_th = current->journal_info; | |
2804 | /* this keeps do_journal_end from NULLing out the current->journal_info | 3062 | |
2805 | ** pointer | 3063 | /* this keeps do_journal_end from NULLing out the current->journal_info |
2806 | */ | 3064 | ** pointer |
2807 | th->t_handle_save = cur_th ; | 3065 | */ |
2808 | if (cur_th && cur_th->t_refcount > 1) { | 3066 | th->t_handle_save = cur_th; |
2809 | BUG() ; | 3067 | if (cur_th && cur_th->t_refcount > 1) { |
2810 | } | 3068 | BUG(); |
2811 | return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_ABORT) ; | 3069 | } |
2812 | } | 3070 | return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_JOIN); |
2813 | 3071 | } | |
2814 | int journal_begin(struct reiserfs_transaction_handle *th, struct super_block * p_s_sb, unsigned long nblocks) { | 3072 | |
2815 | struct reiserfs_transaction_handle *cur_th = current->journal_info ; | 3073 | int journal_join_abort(struct reiserfs_transaction_handle *th, |
2816 | int ret ; | 3074 | struct super_block *p_s_sb, unsigned long nblocks) |
2817 | 3075 | { | |
2818 | th->t_handle_save = NULL ; | 3076 | struct reiserfs_transaction_handle *cur_th = current->journal_info; |
2819 | if (cur_th) { | 3077 | |
2820 | /* we are nesting into the current transaction */ | 3078 | /* this keeps do_journal_end from NULLing out the current->journal_info |
2821 | if (cur_th->t_super == p_s_sb) { | 3079 | ** pointer |
2822 | BUG_ON (!cur_th->t_refcount); | 3080 | */ |
2823 | cur_th->t_refcount++ ; | 3081 | th->t_handle_save = cur_th; |
2824 | memcpy(th, cur_th, sizeof(*th)); | 3082 | if (cur_th && cur_th->t_refcount > 1) { |
2825 | if (th->t_refcount <= 1) | 3083 | BUG(); |
2826 | reiserfs_warning (p_s_sb, "BAD: refcount <= 1, but journal_info != 0"); | 3084 | } |
2827 | return 0; | 3085 | return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_ABORT); |
3086 | } | ||
3087 | |||
3088 | int journal_begin(struct reiserfs_transaction_handle *th, | ||
3089 | struct super_block *p_s_sb, unsigned long nblocks) | ||
3090 | { | ||
3091 | struct reiserfs_transaction_handle *cur_th = current->journal_info; | ||
3092 | int ret; | ||
3093 | |||
3094 | th->t_handle_save = NULL; | ||
3095 | if (cur_th) { | ||
3096 | /* we are nesting into the current transaction */ | ||
3097 | if (cur_th->t_super == p_s_sb) { | ||
3098 | BUG_ON(!cur_th->t_refcount); | ||
3099 | cur_th->t_refcount++; | ||
3100 | memcpy(th, cur_th, sizeof(*th)); | ||
3101 | if (th->t_refcount <= 1) | ||
3102 | reiserfs_warning(p_s_sb, | ||
3103 | "BAD: refcount <= 1, but journal_info != 0"); | ||
3104 | return 0; | ||
3105 | } else { | ||
3106 | /* we've ended up with a handle from a different filesystem. | ||
3107 | ** save it and restore on journal_end. This should never | ||
3108 | ** really happen... | ||
3109 | */ | ||
3110 | reiserfs_warning(p_s_sb, | ||
3111 | "clm-2100: nesting info a different FS"); | ||
3112 | th->t_handle_save = current->journal_info; | ||
3113 | current->journal_info = th; | ||
3114 | } | ||
2828 | } else { | 3115 | } else { |
2829 | /* we've ended up with a handle from a different filesystem. | 3116 | current->journal_info = th; |
2830 | ** save it and restore on journal_end. This should never | 3117 | } |
2831 | ** really happen... | 3118 | ret = do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_REG); |
2832 | */ | 3119 | if (current->journal_info != th) |
2833 | reiserfs_warning(p_s_sb, "clm-2100: nesting info a different FS") ; | 3120 | BUG(); |
2834 | th->t_handle_save = current->journal_info ; | ||
2835 | current->journal_info = th; | ||
2836 | } | ||
2837 | } else { | ||
2838 | current->journal_info = th; | ||
2839 | } | ||
2840 | ret = do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_REG) ; | ||
2841 | if (current->journal_info != th) | ||
2842 | BUG() ; | ||
2843 | 3121 | ||
2844 | /* I guess this boils down to being the reciprocal of clm-2100 above. | 3122 | /* I guess this boils down to being the reciprocal of clm-2100 above. |
2845 | * If do_journal_begin_r fails, we need to put it back, since journal_end | 3123 | * If do_journal_begin_r fails, we need to put it back, since journal_end |
2846 | * won't be called to do it. */ | 3124 | * won't be called to do it. */ |
2847 | if (ret) | 3125 | if (ret) |
2848 | current->journal_info = th->t_handle_save; | 3126 | current->journal_info = th->t_handle_save; |
2849 | else | 3127 | else |
2850 | BUG_ON (!th->t_refcount); | 3128 | BUG_ON(!th->t_refcount); |
2851 | 3129 | ||
2852 | return ret ; | 3130 | return ret; |
2853 | } | 3131 | } |
2854 | 3132 | ||
2855 | /* | 3133 | /* |
@@ -2861,129 +3139,140 @@ int journal_begin(struct reiserfs_transaction_handle *th, struct super_block * | |||
2861 | ** | 3139 | ** |
2862 | ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len. | 3140 | ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len. |
2863 | */ | 3141 | */ |
2864 | int journal_mark_dirty(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, struct buffer_head *bh) { | 3142 | int journal_mark_dirty(struct reiserfs_transaction_handle *th, |
2865 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | 3143 | struct super_block *p_s_sb, struct buffer_head *bh) |
2866 | struct reiserfs_journal_cnode *cn = NULL; | 3144 | { |
2867 | int count_already_incd = 0 ; | 3145 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); |
2868 | int prepared = 0 ; | 3146 | struct reiserfs_journal_cnode *cn = NULL; |
2869 | BUG_ON (!th->t_trans_id); | 3147 | int count_already_incd = 0; |
2870 | 3148 | int prepared = 0; | |
2871 | PROC_INFO_INC( p_s_sb, journal.mark_dirty ); | 3149 | BUG_ON(!th->t_trans_id); |
2872 | if (th->t_trans_id != journal->j_trans_id) { | 3150 | |
2873 | reiserfs_panic(th->t_super, "journal-1577: handle trans id %ld != current trans id %ld\n", | 3151 | PROC_INFO_INC(p_s_sb, journal.mark_dirty); |
2874 | th->t_trans_id, journal->j_trans_id); | 3152 | if (th->t_trans_id != journal->j_trans_id) { |
2875 | } | 3153 | reiserfs_panic(th->t_super, |
2876 | 3154 | "journal-1577: handle trans id %ld != current trans id %ld\n", | |
2877 | p_s_sb->s_dirt = 1; | 3155 | th->t_trans_id, journal->j_trans_id); |
2878 | 3156 | } | |
2879 | prepared = test_clear_buffer_journal_prepared (bh); | 3157 | |
2880 | clear_buffer_journal_restore_dirty (bh); | 3158 | p_s_sb->s_dirt = 1; |
2881 | /* already in this transaction, we are done */ | 3159 | |
2882 | if (buffer_journaled(bh)) { | 3160 | prepared = test_clear_buffer_journal_prepared(bh); |
2883 | PROC_INFO_INC( p_s_sb, journal.mark_dirty_already ); | 3161 | clear_buffer_journal_restore_dirty(bh); |
2884 | return 0 ; | 3162 | /* already in this transaction, we are done */ |
2885 | } | 3163 | if (buffer_journaled(bh)) { |
2886 | 3164 | PROC_INFO_INC(p_s_sb, journal.mark_dirty_already); | |
2887 | /* this must be turned into a panic instead of a warning. We can't allow | 3165 | return 0; |
2888 | ** a dirty or journal_dirty or locked buffer to be logged, as some changes | 3166 | } |
2889 | ** could get to disk too early. NOT GOOD. | 3167 | |
2890 | */ | 3168 | /* this must be turned into a panic instead of a warning. We can't allow |
2891 | if (!prepared || buffer_dirty(bh)) { | 3169 | ** a dirty or journal_dirty or locked buffer to be logged, as some changes |
2892 | reiserfs_warning (p_s_sb, "journal-1777: buffer %llu bad state " | 3170 | ** could get to disk too early. NOT GOOD. |
2893 | "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT", | 3171 | */ |
2894 | (unsigned long long)bh->b_blocknr, prepared ? ' ' : '!', | 3172 | if (!prepared || buffer_dirty(bh)) { |
2895 | buffer_locked(bh) ? ' ' : '!', | 3173 | reiserfs_warning(p_s_sb, "journal-1777: buffer %llu bad state " |
2896 | buffer_dirty(bh) ? ' ' : '!', | 3174 | "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT", |
2897 | buffer_journal_dirty(bh) ? ' ' : '!') ; | 3175 | (unsigned long long)bh->b_blocknr, |
2898 | } | 3176 | prepared ? ' ' : '!', |
2899 | 3177 | buffer_locked(bh) ? ' ' : '!', | |
2900 | if (atomic_read(&(journal->j_wcount)) <= 0) { | 3178 | buffer_dirty(bh) ? ' ' : '!', |
2901 | reiserfs_warning (p_s_sb, "journal-1409: journal_mark_dirty returning because j_wcount was %d", atomic_read(&(journal->j_wcount))) ; | 3179 | buffer_journal_dirty(bh) ? ' ' : '!'); |
2902 | return 1 ; | 3180 | } |
2903 | } | 3181 | |
2904 | /* this error means I've screwed up, and we've overflowed the transaction. | 3182 | if (atomic_read(&(journal->j_wcount)) <= 0) { |
2905 | ** Nothing can be done here, except make the FS readonly or panic. | 3183 | reiserfs_warning(p_s_sb, |
2906 | */ | 3184 | "journal-1409: journal_mark_dirty returning because j_wcount was %d", |
2907 | if (journal->j_len >= journal->j_trans_max) { | 3185 | atomic_read(&(journal->j_wcount))); |
2908 | reiserfs_panic(th->t_super, "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n", journal->j_len) ; | 3186 | return 1; |
2909 | } | 3187 | } |
2910 | 3188 | /* this error means I've screwed up, and we've overflowed the transaction. | |
2911 | if (buffer_journal_dirty(bh)) { | 3189 | ** Nothing can be done here, except make the FS readonly or panic. |
2912 | count_already_incd = 1 ; | 3190 | */ |
2913 | PROC_INFO_INC( p_s_sb, journal.mark_dirty_notjournal ); | 3191 | if (journal->j_len >= journal->j_trans_max) { |
2914 | clear_buffer_journal_dirty (bh); | 3192 | reiserfs_panic(th->t_super, |
2915 | } | 3193 | "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n", |
2916 | 3194 | journal->j_len); | |
2917 | if (journal->j_len > journal->j_len_alloc) { | 3195 | } |
2918 | journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT ; | 3196 | |
2919 | } | 3197 | if (buffer_journal_dirty(bh)) { |
2920 | 3198 | count_already_incd = 1; | |
2921 | set_buffer_journaled (bh); | 3199 | PROC_INFO_INC(p_s_sb, journal.mark_dirty_notjournal); |
2922 | 3200 | clear_buffer_journal_dirty(bh); | |
2923 | /* now put this guy on the end */ | 3201 | } |
2924 | if (!cn) { | 3202 | |
2925 | cn = get_cnode(p_s_sb) ; | 3203 | if (journal->j_len > journal->j_len_alloc) { |
2926 | if (!cn) { | 3204 | journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT; |
2927 | reiserfs_panic(p_s_sb, "get_cnode failed!\n"); | 3205 | } |
2928 | } | 3206 | |
2929 | 3207 | set_buffer_journaled(bh); | |
2930 | if (th->t_blocks_logged == th->t_blocks_allocated) { | 3208 | |
2931 | th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT ; | 3209 | /* now put this guy on the end */ |
2932 | journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT ; | 3210 | if (!cn) { |
2933 | } | 3211 | cn = get_cnode(p_s_sb); |
2934 | th->t_blocks_logged++ ; | 3212 | if (!cn) { |
2935 | journal->j_len++ ; | 3213 | reiserfs_panic(p_s_sb, "get_cnode failed!\n"); |
2936 | 3214 | } | |
2937 | cn->bh = bh ; | 3215 | |
2938 | cn->blocknr = bh->b_blocknr ; | 3216 | if (th->t_blocks_logged == th->t_blocks_allocated) { |
2939 | cn->sb = p_s_sb; | 3217 | th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT; |
2940 | cn->jlist = NULL ; | 3218 | journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT; |
2941 | insert_journal_hash(journal->j_hash_table, cn) ; | 3219 | } |
2942 | if (!count_already_incd) { | 3220 | th->t_blocks_logged++; |
2943 | get_bh(bh) ; | 3221 | journal->j_len++; |
2944 | } | 3222 | |
2945 | } | 3223 | cn->bh = bh; |
2946 | cn->next = NULL ; | 3224 | cn->blocknr = bh->b_blocknr; |
2947 | cn->prev = journal->j_last ; | 3225 | cn->sb = p_s_sb; |
2948 | cn->bh = bh ; | 3226 | cn->jlist = NULL; |
2949 | if (journal->j_last) { | 3227 | insert_journal_hash(journal->j_hash_table, cn); |
2950 | journal->j_last->next = cn ; | 3228 | if (!count_already_incd) { |
2951 | journal->j_last = cn ; | 3229 | get_bh(bh); |
2952 | } else { | 3230 | } |
2953 | journal->j_first = cn ; | 3231 | } |
2954 | journal->j_last = cn ; | 3232 | cn->next = NULL; |
2955 | } | 3233 | cn->prev = journal->j_last; |
2956 | return 0 ; | 3234 | cn->bh = bh; |
2957 | } | 3235 | if (journal->j_last) { |
2958 | 3236 | journal->j_last->next = cn; | |
2959 | int journal_end(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long nblocks) { | 3237 | journal->j_last = cn; |
2960 | if (!current->journal_info && th->t_refcount > 1) | 3238 | } else { |
2961 | reiserfs_warning (p_s_sb, "REISER-NESTING: th NULL, refcount %d", | 3239 | journal->j_first = cn; |
2962 | th->t_refcount); | 3240 | journal->j_last = cn; |
2963 | 3241 | } | |
2964 | if (!th->t_trans_id) { | 3242 | return 0; |
2965 | WARN_ON (1); | 3243 | } |
2966 | return -EIO; | 3244 | |
2967 | } | 3245 | int journal_end(struct reiserfs_transaction_handle *th, |
2968 | 3246 | struct super_block *p_s_sb, unsigned long nblocks) | |
2969 | th->t_refcount--; | 3247 | { |
2970 | if (th->t_refcount > 0) { | 3248 | if (!current->journal_info && th->t_refcount > 1) |
2971 | struct reiserfs_transaction_handle *cur_th = current->journal_info ; | 3249 | reiserfs_warning(p_s_sb, "REISER-NESTING: th NULL, refcount %d", |
2972 | 3250 | th->t_refcount); | |
2973 | /* we aren't allowed to close a nested transaction on a different | 3251 | |
2974 | ** filesystem from the one in the task struct | 3252 | if (!th->t_trans_id) { |
2975 | */ | 3253 | WARN_ON(1); |
2976 | if (cur_th->t_super != th->t_super) | 3254 | return -EIO; |
2977 | BUG() ; | 3255 | } |
2978 | 3256 | ||
2979 | if (th != cur_th) { | 3257 | th->t_refcount--; |
2980 | memcpy(current->journal_info, th, sizeof(*th)); | 3258 | if (th->t_refcount > 0) { |
2981 | th->t_trans_id = 0; | 3259 | struct reiserfs_transaction_handle *cur_th = |
2982 | } | 3260 | current->journal_info; |
2983 | return 0; | 3261 | |
2984 | } else { | 3262 | /* we aren't allowed to close a nested transaction on a different |
2985 | return do_journal_end(th, p_s_sb, nblocks, 0) ; | 3263 | ** filesystem from the one in the task struct |
2986 | } | 3264 | */ |
3265 | if (cur_th->t_super != th->t_super) | ||
3266 | BUG(); | ||
3267 | |||
3268 | if (th != cur_th) { | ||
3269 | memcpy(current->journal_info, th, sizeof(*th)); | ||
3270 | th->t_trans_id = 0; | ||
3271 | } | ||
3272 | return 0; | ||
3273 | } else { | ||
3274 | return do_journal_end(th, p_s_sb, nblocks, 0); | ||
3275 | } | ||
2987 | } | 3276 | } |
2988 | 3277 | ||
2989 | /* removes from the current transaction, relsing and descrementing any counters. | 3278 | /* removes from the current transaction, relsing and descrementing any counters. |
@@ -2993,47 +3282,51 @@ int journal_end(struct reiserfs_transaction_handle *th, struct super_block *p_s_ | |||
2993 | ** | 3282 | ** |
2994 | ** returns 1 if it cleaned and relsed the buffer. 0 otherwise | 3283 | ** returns 1 if it cleaned and relsed the buffer. 0 otherwise |
2995 | */ | 3284 | */ |
2996 | static int remove_from_transaction(struct super_block *p_s_sb, b_blocknr_t blocknr, int already_cleaned) { | 3285 | static int remove_from_transaction(struct super_block *p_s_sb, |
2997 | struct buffer_head *bh ; | 3286 | b_blocknr_t blocknr, int already_cleaned) |
2998 | struct reiserfs_journal_cnode *cn ; | 3287 | { |
2999 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | 3288 | struct buffer_head *bh; |
3000 | int ret = 0; | 3289 | struct reiserfs_journal_cnode *cn; |
3001 | 3290 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); | |
3002 | cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr) ; | 3291 | int ret = 0; |
3003 | if (!cn || !cn->bh) { | 3292 | |
3004 | return ret ; | 3293 | cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr); |
3005 | } | 3294 | if (!cn || !cn->bh) { |
3006 | bh = cn->bh ; | 3295 | return ret; |
3007 | if (cn->prev) { | 3296 | } |
3008 | cn->prev->next = cn->next ; | 3297 | bh = cn->bh; |
3009 | } | 3298 | if (cn->prev) { |
3010 | if (cn->next) { | 3299 | cn->prev->next = cn->next; |
3011 | cn->next->prev = cn->prev ; | 3300 | } |
3012 | } | 3301 | if (cn->next) { |
3013 | if (cn == journal->j_first) { | 3302 | cn->next->prev = cn->prev; |
3014 | journal->j_first = cn->next ; | 3303 | } |
3015 | } | 3304 | if (cn == journal->j_first) { |
3016 | if (cn == journal->j_last) { | 3305 | journal->j_first = cn->next; |
3017 | journal->j_last = cn->prev ; | 3306 | } |
3018 | } | 3307 | if (cn == journal->j_last) { |
3019 | if (bh) | 3308 | journal->j_last = cn->prev; |
3020 | remove_journal_hash(p_s_sb, journal->j_hash_table, NULL, bh->b_blocknr, 0) ; | 3309 | } |
3021 | clear_buffer_journaled (bh); /* don't log this one */ | 3310 | if (bh) |
3022 | 3311 | remove_journal_hash(p_s_sb, journal->j_hash_table, NULL, | |
3023 | if (!already_cleaned) { | 3312 | bh->b_blocknr, 0); |
3024 | clear_buffer_journal_dirty (bh); | 3313 | clear_buffer_journaled(bh); /* don't log this one */ |
3025 | clear_buffer_dirty(bh); | 3314 | |
3026 | clear_buffer_journal_test (bh); | 3315 | if (!already_cleaned) { |
3027 | put_bh(bh) ; | 3316 | clear_buffer_journal_dirty(bh); |
3028 | if (atomic_read(&(bh->b_count)) < 0) { | 3317 | clear_buffer_dirty(bh); |
3029 | reiserfs_warning (p_s_sb, "journal-1752: remove from trans, b_count < 0"); | 3318 | clear_buffer_journal_test(bh); |
3030 | } | 3319 | put_bh(bh); |
3031 | ret = 1 ; | 3320 | if (atomic_read(&(bh->b_count)) < 0) { |
3032 | } | 3321 | reiserfs_warning(p_s_sb, |
3033 | journal->j_len-- ; | 3322 | "journal-1752: remove from trans, b_count < 0"); |
3034 | journal->j_len_alloc-- ; | 3323 | } |
3035 | free_cnode(p_s_sb, cn) ; | 3324 | ret = 1; |
3036 | return ret ; | 3325 | } |
3326 | journal->j_len--; | ||
3327 | journal->j_len_alloc--; | ||
3328 | free_cnode(p_s_sb, cn); | ||
3329 | return ret; | ||
3037 | } | 3330 | } |
3038 | 3331 | ||
3039 | /* | 3332 | /* |
@@ -3046,120 +3339,129 @@ static int remove_from_transaction(struct super_block *p_s_sb, b_blocknr_t block | |||
3046 | ** blocks for a given transaction on disk | 3339 | ** blocks for a given transaction on disk |
3047 | ** | 3340 | ** |
3048 | */ | 3341 | */ |
3049 | static int can_dirty(struct reiserfs_journal_cnode *cn) { | 3342 | static int can_dirty(struct reiserfs_journal_cnode *cn) |
3050 | struct super_block *sb = cn->sb; | 3343 | { |
3051 | b_blocknr_t blocknr = cn->blocknr ; | 3344 | struct super_block *sb = cn->sb; |
3052 | struct reiserfs_journal_cnode *cur = cn->hprev ; | 3345 | b_blocknr_t blocknr = cn->blocknr; |
3053 | int can_dirty = 1 ; | 3346 | struct reiserfs_journal_cnode *cur = cn->hprev; |
3054 | 3347 | int can_dirty = 1; | |
3055 | /* first test hprev. These are all newer than cn, so any node here | 3348 | |
3056 | ** with the same block number and dev means this node can't be sent | 3349 | /* first test hprev. These are all newer than cn, so any node here |
3057 | ** to disk right now. | 3350 | ** with the same block number and dev means this node can't be sent |
3058 | */ | 3351 | ** to disk right now. |
3059 | while(cur && can_dirty) { | 3352 | */ |
3060 | if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb && | 3353 | while (cur && can_dirty) { |
3061 | cur->blocknr == blocknr) { | 3354 | if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb && |
3062 | can_dirty = 0 ; | 3355 | cur->blocknr == blocknr) { |
3063 | } | 3356 | can_dirty = 0; |
3064 | cur = cur->hprev ; | 3357 | } |
3065 | } | 3358 | cur = cur->hprev; |
3066 | /* then test hnext. These are all older than cn. As long as they | 3359 | } |
3067 | ** are committed to the log, it is safe to write cn to disk | 3360 | /* then test hnext. These are all older than cn. As long as they |
3068 | */ | 3361 | ** are committed to the log, it is safe to write cn to disk |
3069 | cur = cn->hnext ; | 3362 | */ |
3070 | while(cur && can_dirty) { | 3363 | cur = cn->hnext; |
3071 | if (cur->jlist && cur->jlist->j_len > 0 && | 3364 | while (cur && can_dirty) { |
3072 | atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh && | 3365 | if (cur->jlist && cur->jlist->j_len > 0 && |
3073 | cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) { | 3366 | atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh && |
3074 | can_dirty = 0 ; | 3367 | cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) { |
3075 | } | 3368 | can_dirty = 0; |
3076 | cur = cur->hnext ; | 3369 | } |
3077 | } | 3370 | cur = cur->hnext; |
3078 | return can_dirty ; | 3371 | } |
3372 | return can_dirty; | ||
3079 | } | 3373 | } |
3080 | 3374 | ||
3081 | /* syncs the commit blocks, but does not force the real buffers to disk | 3375 | /* syncs the commit blocks, but does not force the real buffers to disk |
3082 | ** will wait until the current transaction is done/commited before returning | 3376 | ** will wait until the current transaction is done/commited before returning |
3083 | */ | 3377 | */ |
3084 | int journal_end_sync(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, unsigned long nblocks) { | 3378 | int journal_end_sync(struct reiserfs_transaction_handle *th, |
3085 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | 3379 | struct super_block *p_s_sb, unsigned long nblocks) |
3380 | { | ||
3381 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); | ||
3086 | 3382 | ||
3087 | BUG_ON (!th->t_trans_id); | 3383 | BUG_ON(!th->t_trans_id); |
3088 | /* you can sync while nested, very, very bad */ | 3384 | /* you can sync while nested, very, very bad */ |
3089 | if (th->t_refcount > 1) { | 3385 | if (th->t_refcount > 1) { |
3090 | BUG() ; | 3386 | BUG(); |
3091 | } | 3387 | } |
3092 | if (journal->j_len == 0) { | 3388 | if (journal->j_len == 0) { |
3093 | reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ; | 3389 | reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), |
3094 | journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ; | 3390 | 1); |
3095 | } | 3391 | journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)); |
3096 | return do_journal_end(th, p_s_sb, nblocks, COMMIT_NOW | WAIT) ; | 3392 | } |
3393 | return do_journal_end(th, p_s_sb, nblocks, COMMIT_NOW | WAIT); | ||
3097 | } | 3394 | } |
3098 | 3395 | ||
3099 | /* | 3396 | /* |
3100 | ** writeback the pending async commits to disk | 3397 | ** writeback the pending async commits to disk |
3101 | */ | 3398 | */ |
3102 | static void flush_async_commits(void *p) { | 3399 | static void flush_async_commits(void *p) |
3103 | struct super_block *p_s_sb = p; | 3400 | { |
3104 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | 3401 | struct super_block *p_s_sb = p; |
3105 | struct reiserfs_journal_list *jl; | 3402 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); |
3106 | struct list_head *entry; | 3403 | struct reiserfs_journal_list *jl; |
3107 | 3404 | struct list_head *entry; | |
3108 | lock_kernel(); | 3405 | |
3109 | if (!list_empty(&journal->j_journal_list)) { | 3406 | lock_kernel(); |
3110 | /* last entry is the youngest, commit it and you get everything */ | 3407 | if (!list_empty(&journal->j_journal_list)) { |
3111 | entry = journal->j_journal_list.prev; | 3408 | /* last entry is the youngest, commit it and you get everything */ |
3112 | jl = JOURNAL_LIST_ENTRY(entry); | 3409 | entry = journal->j_journal_list.prev; |
3113 | flush_commit_list(p_s_sb, jl, 1); | 3410 | jl = JOURNAL_LIST_ENTRY(entry); |
3114 | } | 3411 | flush_commit_list(p_s_sb, jl, 1); |
3115 | unlock_kernel(); | 3412 | } |
3116 | /* | 3413 | unlock_kernel(); |
3117 | * this is a little racey, but there's no harm in missing | 3414 | /* |
3118 | * the filemap_fdata_write | 3415 | * this is a little racey, but there's no harm in missing |
3119 | */ | 3416 | * the filemap_fdata_write |
3120 | if (!atomic_read(&journal->j_async_throttle) && !reiserfs_is_journal_aborted (journal)) { | 3417 | */ |
3121 | atomic_inc(&journal->j_async_throttle); | 3418 | if (!atomic_read(&journal->j_async_throttle) |
3122 | filemap_fdatawrite(p_s_sb->s_bdev->bd_inode->i_mapping); | 3419 | && !reiserfs_is_journal_aborted(journal)) { |
3123 | atomic_dec(&journal->j_async_throttle); | 3420 | atomic_inc(&journal->j_async_throttle); |
3124 | } | 3421 | filemap_fdatawrite(p_s_sb->s_bdev->bd_inode->i_mapping); |
3422 | atomic_dec(&journal->j_async_throttle); | ||
3423 | } | ||
3125 | } | 3424 | } |
3126 | 3425 | ||
3127 | /* | 3426 | /* |
3128 | ** flushes any old transactions to disk | 3427 | ** flushes any old transactions to disk |
3129 | ** ends the current transaction if it is too old | 3428 | ** ends the current transaction if it is too old |
3130 | */ | 3429 | */ |
3131 | int reiserfs_flush_old_commits(struct super_block *p_s_sb) { | 3430 | int reiserfs_flush_old_commits(struct super_block *p_s_sb) |
3132 | time_t now ; | 3431 | { |
3133 | struct reiserfs_transaction_handle th ; | 3432 | time_t now; |
3134 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | 3433 | struct reiserfs_transaction_handle th; |
3135 | 3434 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); | |
3136 | now = get_seconds(); | 3435 | |
3137 | /* safety check so we don't flush while we are replaying the log during | 3436 | now = get_seconds(); |
3138 | * mount | 3437 | /* safety check so we don't flush while we are replaying the log during |
3139 | */ | 3438 | * mount |
3140 | if (list_empty(&journal->j_journal_list)) { | 3439 | */ |
3141 | return 0 ; | 3440 | if (list_empty(&journal->j_journal_list)) { |
3142 | } | 3441 | return 0; |
3143 | 3442 | } | |
3144 | /* check the current transaction. If there are no writers, and it is | 3443 | |
3145 | * too old, finish it, and force the commit blocks to disk | 3444 | /* check the current transaction. If there are no writers, and it is |
3146 | */ | 3445 | * too old, finish it, and force the commit blocks to disk |
3147 | if (atomic_read(&journal->j_wcount) <= 0 && | 3446 | */ |
3148 | journal->j_trans_start_time > 0 && | 3447 | if (atomic_read(&journal->j_wcount) <= 0 && |
3149 | journal->j_len > 0 && | 3448 | journal->j_trans_start_time > 0 && |
3150 | (now - journal->j_trans_start_time) > journal->j_max_trans_age) | 3449 | journal->j_len > 0 && |
3151 | { | 3450 | (now - journal->j_trans_start_time) > journal->j_max_trans_age) { |
3152 | if (!journal_join(&th, p_s_sb, 1)) { | 3451 | if (!journal_join(&th, p_s_sb, 1)) { |
3153 | reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ; | 3452 | reiserfs_prepare_for_journal(p_s_sb, |
3154 | journal_mark_dirty(&th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ; | 3453 | SB_BUFFER_WITH_SB(p_s_sb), |
3155 | 3454 | 1); | |
3156 | /* we're only being called from kreiserfsd, it makes no sense to do | 3455 | journal_mark_dirty(&th, p_s_sb, |
3157 | ** an async commit so that kreiserfsd can do it later | 3456 | SB_BUFFER_WITH_SB(p_s_sb)); |
3158 | */ | 3457 | |
3159 | do_journal_end(&th, p_s_sb,1, COMMIT_NOW | WAIT) ; | 3458 | /* we're only being called from kreiserfsd, it makes no sense to do |
3160 | } | 3459 | ** an async commit so that kreiserfsd can do it later |
3161 | } | 3460 | */ |
3162 | return p_s_sb->s_dirt; | 3461 | do_journal_end(&th, p_s_sb, 1, COMMIT_NOW | WAIT); |
3462 | } | ||
3463 | } | ||
3464 | return p_s_sb->s_dirt; | ||
3163 | } | 3465 | } |
3164 | 3466 | ||
3165 | /* | 3467 | /* |
@@ -3173,101 +3475,108 @@ int reiserfs_flush_old_commits(struct super_block *p_s_sb) { | |||
3173 | ** | 3475 | ** |
3174 | ** Note, we can't allow the journal_end to proceed while there are still writers in the log. | 3476 | ** Note, we can't allow the journal_end to proceed while there are still writers in the log. |
3175 | */ | 3477 | */ |
3176 | static int check_journal_end(struct reiserfs_transaction_handle *th, struct super_block * p_s_sb, | 3478 | static int check_journal_end(struct reiserfs_transaction_handle *th, |
3177 | unsigned long nblocks, int flags) { | 3479 | struct super_block *p_s_sb, unsigned long nblocks, |
3178 | 3480 | int flags) | |
3179 | time_t now ; | 3481 | { |
3180 | int flush = flags & FLUSH_ALL ; | 3482 | |
3181 | int commit_now = flags & COMMIT_NOW ; | 3483 | time_t now; |
3182 | int wait_on_commit = flags & WAIT ; | 3484 | int flush = flags & FLUSH_ALL; |
3183 | struct reiserfs_journal_list *jl; | 3485 | int commit_now = flags & COMMIT_NOW; |
3184 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | 3486 | int wait_on_commit = flags & WAIT; |
3185 | 3487 | struct reiserfs_journal_list *jl; | |
3186 | BUG_ON (!th->t_trans_id); | 3488 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); |
3187 | 3489 | ||
3188 | if (th->t_trans_id != journal->j_trans_id) { | 3490 | BUG_ON(!th->t_trans_id); |
3189 | reiserfs_panic(th->t_super, "journal-1577: handle trans id %ld != current trans id %ld\n", | 3491 | |
3190 | th->t_trans_id, journal->j_trans_id); | 3492 | if (th->t_trans_id != journal->j_trans_id) { |
3191 | } | 3493 | reiserfs_panic(th->t_super, |
3192 | 3494 | "journal-1577: handle trans id %ld != current trans id %ld\n", | |
3193 | journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged) ; | 3495 | th->t_trans_id, journal->j_trans_id); |
3194 | if (atomic_read(&(journal->j_wcount)) > 0) { /* <= 0 is allowed. unmounting might not call begin */ | 3496 | } |
3195 | atomic_dec(&(journal->j_wcount)) ; | 3497 | |
3196 | } | 3498 | journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged); |
3197 | 3499 | if (atomic_read(&(journal->j_wcount)) > 0) { /* <= 0 is allowed. unmounting might not call begin */ | |
3198 | /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released | 3500 | atomic_dec(&(journal->j_wcount)); |
3199 | ** will be dealt with by next transaction that actually writes something, but should be taken | 3501 | } |
3200 | ** care of in this trans | 3502 | |
3201 | */ | 3503 | /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released |
3202 | if (journal->j_len == 0) { | 3504 | ** will be dealt with by next transaction that actually writes something, but should be taken |
3203 | BUG(); | 3505 | ** care of in this trans |
3204 | } | 3506 | */ |
3205 | /* if wcount > 0, and we are called to with flush or commit_now, | 3507 | if (journal->j_len == 0) { |
3206 | ** we wait on j_join_wait. We will wake up when the last writer has | 3508 | BUG(); |
3207 | ** finished the transaction, and started it on its way to the disk. | 3509 | } |
3208 | ** Then, we flush the commit or journal list, and just return 0 | 3510 | /* if wcount > 0, and we are called to with flush or commit_now, |
3209 | ** because the rest of journal end was already done for this transaction. | 3511 | ** we wait on j_join_wait. We will wake up when the last writer has |
3210 | */ | 3512 | ** finished the transaction, and started it on its way to the disk. |
3211 | if (atomic_read(&(journal->j_wcount)) > 0) { | 3513 | ** Then, we flush the commit or journal list, and just return 0 |
3212 | if (flush || commit_now) { | 3514 | ** because the rest of journal end was already done for this transaction. |
3213 | unsigned trans_id ; | 3515 | */ |
3214 | 3516 | if (atomic_read(&(journal->j_wcount)) > 0) { | |
3215 | jl = journal->j_current_jl; | 3517 | if (flush || commit_now) { |
3216 | trans_id = jl->j_trans_id; | 3518 | unsigned trans_id; |
3217 | if (wait_on_commit) | 3519 | |
3218 | jl->j_state |= LIST_COMMIT_PENDING; | 3520 | jl = journal->j_current_jl; |
3219 | atomic_set(&(journal->j_jlock), 1) ; | 3521 | trans_id = jl->j_trans_id; |
3220 | if (flush) { | 3522 | if (wait_on_commit) |
3221 | journal->j_next_full_flush = 1 ; | 3523 | jl->j_state |= LIST_COMMIT_PENDING; |
3222 | } | 3524 | atomic_set(&(journal->j_jlock), 1); |
3223 | unlock_journal(p_s_sb) ; | 3525 | if (flush) { |
3224 | 3526 | journal->j_next_full_flush = 1; | |
3225 | /* sleep while the current transaction is still j_jlocked */ | 3527 | } |
3226 | while(journal->j_trans_id == trans_id) { | 3528 | unlock_journal(p_s_sb); |
3227 | if (atomic_read(&journal->j_jlock)) { | 3529 | |
3228 | queue_log_writer(p_s_sb); | 3530 | /* sleep while the current transaction is still j_jlocked */ |
3229 | } else { | 3531 | while (journal->j_trans_id == trans_id) { |
3230 | lock_journal(p_s_sb); | 3532 | if (atomic_read(&journal->j_jlock)) { |
3231 | if (journal->j_trans_id == trans_id) { | 3533 | queue_log_writer(p_s_sb); |
3232 | atomic_set(&(journal->j_jlock), 1) ; | 3534 | } else { |
3233 | } | 3535 | lock_journal(p_s_sb); |
3234 | unlock_journal(p_s_sb); | 3536 | if (journal->j_trans_id == trans_id) { |
3235 | } | 3537 | atomic_set(&(journal->j_jlock), |
3236 | } | 3538 | 1); |
3237 | if (journal->j_trans_id == trans_id) { | 3539 | } |
3238 | BUG(); | 3540 | unlock_journal(p_s_sb); |
3239 | } | 3541 | } |
3240 | if (commit_now && journal_list_still_alive(p_s_sb, trans_id) && | 3542 | } |
3241 | wait_on_commit) | 3543 | if (journal->j_trans_id == trans_id) { |
3242 | { | 3544 | BUG(); |
3243 | flush_commit_list(p_s_sb, jl, 1) ; | 3545 | } |
3244 | } | 3546 | if (commit_now |
3245 | return 0 ; | 3547 | && journal_list_still_alive(p_s_sb, trans_id) |
3246 | } | 3548 | && wait_on_commit) { |
3247 | unlock_journal(p_s_sb) ; | 3549 | flush_commit_list(p_s_sb, jl, 1); |
3248 | return 0 ; | 3550 | } |
3249 | } | 3551 | return 0; |
3250 | 3552 | } | |
3251 | /* deal with old transactions where we are the last writers */ | 3553 | unlock_journal(p_s_sb); |
3252 | now = get_seconds(); | 3554 | return 0; |
3253 | if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) { | 3555 | } |
3254 | commit_now = 1 ; | 3556 | |
3255 | journal->j_next_async_flush = 1 ; | 3557 | /* deal with old transactions where we are the last writers */ |
3256 | } | 3558 | now = get_seconds(); |
3257 | /* don't batch when someone is waiting on j_join_wait */ | 3559 | if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) { |
3258 | /* don't batch when syncing the commit or flushing the whole trans */ | 3560 | commit_now = 1; |
3259 | if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock))) && !flush && !commit_now && | 3561 | journal->j_next_async_flush = 1; |
3260 | (journal->j_len < journal->j_max_batch) && | 3562 | } |
3261 | journal->j_len_alloc < journal->j_max_batch && journal->j_cnode_free > (journal->j_trans_max * 3)) { | 3563 | /* don't batch when someone is waiting on j_join_wait */ |
3262 | journal->j_bcount++ ; | 3564 | /* don't batch when syncing the commit or flushing the whole trans */ |
3263 | unlock_journal(p_s_sb) ; | 3565 | if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock))) |
3264 | return 0 ; | 3566 | && !flush && !commit_now && (journal->j_len < journal->j_max_batch) |
3265 | } | 3567 | && journal->j_len_alloc < journal->j_max_batch |
3266 | 3568 | && journal->j_cnode_free > (journal->j_trans_max * 3)) { | |
3267 | if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(p_s_sb)) { | 3569 | journal->j_bcount++; |
3268 | reiserfs_panic(p_s_sb, "journal-003: journal_end: j_start (%ld) is too high\n", journal->j_start) ; | 3570 | unlock_journal(p_s_sb); |
3269 | } | 3571 | return 0; |
3270 | return 1 ; | 3572 | } |
3573 | |||
3574 | if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(p_s_sb)) { | ||
3575 | reiserfs_panic(p_s_sb, | ||
3576 | "journal-003: journal_end: j_start (%ld) is too high\n", | ||
3577 | journal->j_start); | ||
3578 | } | ||
3579 | return 1; | ||
3271 | } | 3580 | } |
3272 | 3581 | ||
3273 | /* | 3582 | /* |
@@ -3284,83 +3593,95 @@ static int check_journal_end(struct reiserfs_transaction_handle *th, struct supe | |||
3284 | ** | 3593 | ** |
3285 | ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list. | 3594 | ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list. |
3286 | */ | 3595 | */ |
3287 | int journal_mark_freed(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, b_blocknr_t blocknr) { | 3596 | int journal_mark_freed(struct reiserfs_transaction_handle *th, |
3288 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | 3597 | struct super_block *p_s_sb, b_blocknr_t blocknr) |
3289 | struct reiserfs_journal_cnode *cn = NULL ; | 3598 | { |
3290 | struct buffer_head *bh = NULL ; | 3599 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); |
3291 | struct reiserfs_list_bitmap *jb = NULL ; | 3600 | struct reiserfs_journal_cnode *cn = NULL; |
3292 | int cleaned = 0 ; | 3601 | struct buffer_head *bh = NULL; |
3293 | BUG_ON (!th->t_trans_id); | 3602 | struct reiserfs_list_bitmap *jb = NULL; |
3294 | 3603 | int cleaned = 0; | |
3295 | cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr); | 3604 | BUG_ON(!th->t_trans_id); |
3296 | if (cn && cn->bh) { | 3605 | |
3297 | bh = cn->bh ; | 3606 | cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr); |
3298 | get_bh(bh) ; | 3607 | if (cn && cn->bh) { |
3299 | } | 3608 | bh = cn->bh; |
3300 | /* if it is journal new, we just remove it from this transaction */ | 3609 | get_bh(bh); |
3301 | if (bh && buffer_journal_new(bh)) { | 3610 | } |
3302 | clear_buffer_journal_new (bh); | 3611 | /* if it is journal new, we just remove it from this transaction */ |
3303 | clear_prepared_bits(bh) ; | 3612 | if (bh && buffer_journal_new(bh)) { |
3304 | reiserfs_clean_and_file_buffer(bh) ; | 3613 | clear_buffer_journal_new(bh); |
3305 | cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned) ; | 3614 | clear_prepared_bits(bh); |
3306 | } else { | 3615 | reiserfs_clean_and_file_buffer(bh); |
3307 | /* set the bit for this block in the journal bitmap for this transaction */ | 3616 | cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned); |
3308 | jb = journal->j_current_jl->j_list_bitmap; | 3617 | } else { |
3309 | if (!jb) { | 3618 | /* set the bit for this block in the journal bitmap for this transaction */ |
3310 | reiserfs_panic(p_s_sb, "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n") ; | 3619 | jb = journal->j_current_jl->j_list_bitmap; |
3311 | } | 3620 | if (!jb) { |
3312 | set_bit_in_list_bitmap(p_s_sb, blocknr, jb) ; | 3621 | reiserfs_panic(p_s_sb, |
3313 | 3622 | "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n"); | |
3314 | /* Note, the entire while loop is not allowed to schedule. */ | 3623 | } |
3315 | 3624 | set_bit_in_list_bitmap(p_s_sb, blocknr, jb); | |
3316 | if (bh) { | 3625 | |
3317 | clear_prepared_bits(bh) ; | 3626 | /* Note, the entire while loop is not allowed to schedule. */ |
3318 | reiserfs_clean_and_file_buffer(bh) ; | 3627 | |
3319 | } | 3628 | if (bh) { |
3320 | cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned) ; | 3629 | clear_prepared_bits(bh); |
3321 | 3630 | reiserfs_clean_and_file_buffer(bh); | |
3322 | /* find all older transactions with this block, make sure they don't try to write it out */ | 3631 | } |
3323 | cn = get_journal_hash_dev(p_s_sb,journal->j_list_hash_table, blocknr) ; | 3632 | cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned); |
3324 | while (cn) { | 3633 | |
3325 | if (p_s_sb == cn->sb && blocknr == cn->blocknr) { | 3634 | /* find all older transactions with this block, make sure they don't try to write it out */ |
3326 | set_bit(BLOCK_FREED, &cn->state) ; | 3635 | cn = get_journal_hash_dev(p_s_sb, journal->j_list_hash_table, |
3327 | if (cn->bh) { | 3636 | blocknr); |
3328 | if (!cleaned) { | 3637 | while (cn) { |
3329 | /* remove_from_transaction will brelse the buffer if it was | 3638 | if (p_s_sb == cn->sb && blocknr == cn->blocknr) { |
3330 | ** in the current trans | 3639 | set_bit(BLOCK_FREED, &cn->state); |
3331 | */ | 3640 | if (cn->bh) { |
3332 | clear_buffer_journal_dirty (cn->bh); | 3641 | if (!cleaned) { |
3333 | clear_buffer_dirty(cn->bh); | 3642 | /* remove_from_transaction will brelse the buffer if it was |
3334 | clear_buffer_journal_test(cn->bh); | 3643 | ** in the current trans |
3335 | cleaned = 1 ; | 3644 | */ |
3336 | put_bh(cn->bh) ; | 3645 | clear_buffer_journal_dirty(cn-> |
3337 | if (atomic_read(&(cn->bh->b_count)) < 0) { | 3646 | bh); |
3338 | reiserfs_warning (p_s_sb, "journal-2138: cn->bh->b_count < 0"); | 3647 | clear_buffer_dirty(cn->bh); |
3339 | } | 3648 | clear_buffer_journal_test(cn-> |
3340 | } | 3649 | bh); |
3341 | if (cn->jlist) { /* since we are clearing the bh, we MUST dec nonzerolen */ | 3650 | cleaned = 1; |
3342 | atomic_dec(&(cn->jlist->j_nonzerolen)) ; | 3651 | put_bh(cn->bh); |
3343 | } | 3652 | if (atomic_read |
3344 | cn->bh = NULL ; | 3653 | (&(cn->bh->b_count)) < 0) { |
3345 | } | 3654 | reiserfs_warning(p_s_sb, |
3346 | } | 3655 | "journal-2138: cn->bh->b_count < 0"); |
3347 | cn = cn->hnext ; | 3656 | } |
3348 | } | 3657 | } |
3349 | } | 3658 | if (cn->jlist) { /* since we are clearing the bh, we MUST dec nonzerolen */ |
3350 | 3659 | atomic_dec(& | |
3351 | if (bh) { | 3660 | (cn->jlist-> |
3352 | put_bh(bh) ; /* get_hash grabs the buffer */ | 3661 | j_nonzerolen)); |
3353 | if (atomic_read(&(bh->b_count)) < 0) { | 3662 | } |
3354 | reiserfs_warning (p_s_sb, "journal-2165: bh->b_count < 0"); | 3663 | cn->bh = NULL; |
3355 | } | 3664 | } |
3356 | } | 3665 | } |
3357 | return 0 ; | 3666 | cn = cn->hnext; |
3358 | } | 3667 | } |
3359 | 3668 | } | |
3360 | void reiserfs_update_inode_transaction(struct inode *inode) { | 3669 | |
3361 | struct reiserfs_journal *journal = SB_JOURNAL (inode->i_sb); | 3670 | if (bh) { |
3362 | REISERFS_I(inode)->i_jl = journal->j_current_jl; | 3671 | put_bh(bh); /* get_hash grabs the buffer */ |
3363 | REISERFS_I(inode)->i_trans_id = journal->j_trans_id ; | 3672 | if (atomic_read(&(bh->b_count)) < 0) { |
3673 | reiserfs_warning(p_s_sb, | ||
3674 | "journal-2165: bh->b_count < 0"); | ||
3675 | } | ||
3676 | } | ||
3677 | return 0; | ||
3678 | } | ||
3679 | |||
3680 | void reiserfs_update_inode_transaction(struct inode *inode) | ||
3681 | { | ||
3682 | struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb); | ||
3683 | REISERFS_I(inode)->i_jl = journal->j_current_jl; | ||
3684 | REISERFS_I(inode)->i_trans_id = journal->j_trans_id; | ||
3364 | } | 3685 | } |
3365 | 3686 | ||
3366 | /* | 3687 | /* |
@@ -3368,99 +3689,102 @@ void reiserfs_update_inode_transaction(struct inode *inode) { | |||
3368 | * if a transaction was actually committed and the barrier was done | 3689 | * if a transaction was actually committed and the barrier was done |
3369 | */ | 3690 | */ |
3370 | static int __commit_trans_jl(struct inode *inode, unsigned long id, | 3691 | static int __commit_trans_jl(struct inode *inode, unsigned long id, |
3371 | struct reiserfs_journal_list *jl) | 3692 | struct reiserfs_journal_list *jl) |
3372 | { | 3693 | { |
3373 | struct reiserfs_transaction_handle th ; | 3694 | struct reiserfs_transaction_handle th; |
3374 | struct super_block *sb = inode->i_sb ; | 3695 | struct super_block *sb = inode->i_sb; |
3375 | struct reiserfs_journal *journal = SB_JOURNAL (sb); | 3696 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
3376 | int ret = 0; | 3697 | int ret = 0; |
3698 | |||
3699 | /* is it from the current transaction, or from an unknown transaction? */ | ||
3700 | if (id == journal->j_trans_id) { | ||
3701 | jl = journal->j_current_jl; | ||
3702 | /* try to let other writers come in and grow this transaction */ | ||
3703 | let_transaction_grow(sb, id); | ||
3704 | if (journal->j_trans_id != id) { | ||
3705 | goto flush_commit_only; | ||
3706 | } | ||
3377 | 3707 | ||
3378 | /* is it from the current transaction, or from an unknown transaction? */ | 3708 | ret = journal_begin(&th, sb, 1); |
3379 | if (id == journal->j_trans_id) { | 3709 | if (ret) |
3380 | jl = journal->j_current_jl; | 3710 | return ret; |
3381 | /* try to let other writers come in and grow this transaction */ | 3711 | |
3382 | let_transaction_grow(sb, id); | 3712 | /* someone might have ended this transaction while we joined */ |
3383 | if (journal->j_trans_id != id) { | 3713 | if (journal->j_trans_id != id) { |
3384 | goto flush_commit_only; | 3714 | reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), |
3385 | } | 3715 | 1); |
3716 | journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb)); | ||
3717 | ret = journal_end(&th, sb, 1); | ||
3718 | goto flush_commit_only; | ||
3719 | } | ||
3386 | 3720 | ||
3387 | ret = journal_begin(&th, sb, 1) ; | 3721 | ret = journal_end_sync(&th, sb, 1); |
3388 | if (ret) | 3722 | if (!ret) |
3389 | return ret; | 3723 | ret = 1; |
3390 | 3724 | ||
3391 | /* someone might have ended this transaction while we joined */ | 3725 | } else { |
3392 | if (journal->j_trans_id != id) { | 3726 | /* this gets tricky, we have to make sure the journal list in |
3393 | reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb), 1) ; | 3727 | * the inode still exists. We know the list is still around |
3394 | journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb)) ; | 3728 | * if we've got a larger transaction id than the oldest list |
3395 | ret = journal_end(&th, sb, 1) ; | 3729 | */ |
3396 | goto flush_commit_only; | 3730 | flush_commit_only: |
3731 | if (journal_list_still_alive(inode->i_sb, id)) { | ||
3732 | /* | ||
3733 | * we only set ret to 1 when we know for sure | ||
3734 | * the barrier hasn't been started yet on the commit | ||
3735 | * block. | ||
3736 | */ | ||
3737 | if (atomic_read(&jl->j_commit_left) > 1) | ||
3738 | ret = 1; | ||
3739 | flush_commit_list(sb, jl, 1); | ||
3740 | if (journal->j_errno) | ||
3741 | ret = journal->j_errno; | ||
3742 | } | ||
3397 | } | 3743 | } |
3744 | /* otherwise the list is gone, and long since committed */ | ||
3745 | return ret; | ||
3746 | } | ||
3398 | 3747 | ||
3399 | ret = journal_end_sync(&th, sb, 1) ; | 3748 | int reiserfs_commit_for_inode(struct inode *inode) |
3400 | if (!ret) | 3749 | { |
3401 | ret = 1; | 3750 | unsigned long id = REISERFS_I(inode)->i_trans_id; |
3751 | struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl; | ||
3402 | 3752 | ||
3403 | } else { | 3753 | /* for the whole inode, assume unset id means it was |
3404 | /* this gets tricky, we have to make sure the journal list in | 3754 | * changed in the current transaction. More conservative |
3405 | * the inode still exists. We know the list is still around | ||
3406 | * if we've got a larger transaction id than the oldest list | ||
3407 | */ | 3755 | */ |
3408 | flush_commit_only: | 3756 | if (!id || !jl) { |
3409 | if (journal_list_still_alive(inode->i_sb, id)) { | 3757 | reiserfs_update_inode_transaction(inode); |
3410 | /* | 3758 | id = REISERFS_I(inode)->i_trans_id; |
3411 | * we only set ret to 1 when we know for sure | 3759 | /* jl will be updated in __commit_trans_jl */ |
3412 | * the barrier hasn't been started yet on the commit | 3760 | } |
3413 | * block. | 3761 | |
3414 | */ | 3762 | return __commit_trans_jl(inode, id, jl); |
3415 | if (atomic_read(&jl->j_commit_left) > 1) | 3763 | } |
3416 | ret = 1; | 3764 | |
3417 | flush_commit_list(sb, jl, 1) ; | 3765 | void reiserfs_restore_prepared_buffer(struct super_block *p_s_sb, |
3418 | if (journal->j_errno) | 3766 | struct buffer_head *bh) |
3419 | ret = journal->j_errno; | 3767 | { |
3420 | } | 3768 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); |
3421 | } | 3769 | PROC_INFO_INC(p_s_sb, journal.restore_prepared); |
3422 | /* otherwise the list is gone, and long since committed */ | 3770 | if (!bh) { |
3423 | return ret; | 3771 | return; |
3424 | } | 3772 | } |
3425 | 3773 | if (test_clear_buffer_journal_restore_dirty(bh) && | |
3426 | int reiserfs_commit_for_inode(struct inode *inode) { | 3774 | buffer_journal_dirty(bh)) { |
3427 | unsigned long id = REISERFS_I(inode)->i_trans_id; | 3775 | struct reiserfs_journal_cnode *cn; |
3428 | struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl; | 3776 | cn = get_journal_hash_dev(p_s_sb, |
3429 | 3777 | journal->j_list_hash_table, | |
3430 | /* for the whole inode, assume unset id means it was | 3778 | bh->b_blocknr); |
3431 | * changed in the current transaction. More conservative | 3779 | if (cn && can_dirty(cn)) { |
3432 | */ | 3780 | set_buffer_journal_test(bh); |
3433 | if (!id || !jl) { | 3781 | mark_buffer_dirty(bh); |
3434 | reiserfs_update_inode_transaction(inode) ; | 3782 | } |
3435 | id = REISERFS_I(inode)->i_trans_id; | 3783 | } |
3436 | /* jl will be updated in __commit_trans_jl */ | 3784 | clear_buffer_journal_prepared(bh); |
3437 | } | 3785 | } |
3438 | 3786 | ||
3439 | return __commit_trans_jl(inode, id, jl); | 3787 | extern struct tree_balance *cur_tb; |
3440 | } | ||
3441 | |||
3442 | void reiserfs_restore_prepared_buffer(struct super_block *p_s_sb, | ||
3443 | struct buffer_head *bh) { | ||
3444 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | ||
3445 | PROC_INFO_INC( p_s_sb, journal.restore_prepared ); | ||
3446 | if (!bh) { | ||
3447 | return ; | ||
3448 | } | ||
3449 | if (test_clear_buffer_journal_restore_dirty (bh) && | ||
3450 | buffer_journal_dirty(bh)) { | ||
3451 | struct reiserfs_journal_cnode *cn; | ||
3452 | cn = get_journal_hash_dev(p_s_sb, | ||
3453 | journal->j_list_hash_table, | ||
3454 | bh->b_blocknr); | ||
3455 | if (cn && can_dirty(cn)) { | ||
3456 | set_buffer_journal_test (bh); | ||
3457 | mark_buffer_dirty(bh); | ||
3458 | } | ||
3459 | } | ||
3460 | clear_buffer_journal_prepared (bh); | ||
3461 | } | ||
3462 | |||
3463 | extern struct tree_balance *cur_tb ; | ||
3464 | /* | 3788 | /* |
3465 | ** before we can change a metadata block, we have to make sure it won't | 3789 | ** before we can change a metadata block, we have to make sure it won't |
3466 | ** be written to disk while we are altering it. So, we must: | 3790 | ** be written to disk while we are altering it. So, we must: |
@@ -3469,39 +3793,41 @@ extern struct tree_balance *cur_tb ; | |||
3469 | ** | 3793 | ** |
3470 | */ | 3794 | */ |
3471 | int reiserfs_prepare_for_journal(struct super_block *p_s_sb, | 3795 | int reiserfs_prepare_for_journal(struct super_block *p_s_sb, |
3472 | struct buffer_head *bh, int wait) { | 3796 | struct buffer_head *bh, int wait) |
3473 | PROC_INFO_INC( p_s_sb, journal.prepare ); | 3797 | { |
3474 | 3798 | PROC_INFO_INC(p_s_sb, journal.prepare); | |
3475 | if (test_set_buffer_locked(bh)) { | 3799 | |
3476 | if (!wait) | 3800 | if (test_set_buffer_locked(bh)) { |
3477 | return 0; | 3801 | if (!wait) |
3478 | lock_buffer(bh); | 3802 | return 0; |
3479 | } | 3803 | lock_buffer(bh); |
3480 | set_buffer_journal_prepared (bh); | 3804 | } |
3481 | if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) { | 3805 | set_buffer_journal_prepared(bh); |
3482 | clear_buffer_journal_test (bh); | 3806 | if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) { |
3483 | set_buffer_journal_restore_dirty (bh); | 3807 | clear_buffer_journal_test(bh); |
3484 | } | 3808 | set_buffer_journal_restore_dirty(bh); |
3485 | unlock_buffer(bh); | 3809 | } |
3486 | return 1; | 3810 | unlock_buffer(bh); |
3487 | } | 3811 | return 1; |
3488 | 3812 | } | |
3489 | static void flush_old_journal_lists(struct super_block *s) { | 3813 | |
3490 | struct reiserfs_journal *journal = SB_JOURNAL (s); | 3814 | static void flush_old_journal_lists(struct super_block *s) |
3491 | struct reiserfs_journal_list *jl; | 3815 | { |
3492 | struct list_head *entry; | 3816 | struct reiserfs_journal *journal = SB_JOURNAL(s); |
3493 | time_t now = get_seconds(); | 3817 | struct reiserfs_journal_list *jl; |
3494 | 3818 | struct list_head *entry; | |
3495 | while(!list_empty(&journal->j_journal_list)) { | 3819 | time_t now = get_seconds(); |
3496 | entry = journal->j_journal_list.next; | 3820 | |
3497 | jl = JOURNAL_LIST_ENTRY(entry); | 3821 | while (!list_empty(&journal->j_journal_list)) { |
3498 | /* this check should always be run, to send old lists to disk */ | 3822 | entry = journal->j_journal_list.next; |
3499 | if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4))) { | 3823 | jl = JOURNAL_LIST_ENTRY(entry); |
3500 | flush_used_journal_lists(s, jl); | 3824 | /* this check should always be run, to send old lists to disk */ |
3501 | } else { | 3825 | if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4))) { |
3502 | break; | 3826 | flush_used_journal_lists(s, jl); |
3827 | } else { | ||
3828 | break; | ||
3829 | } | ||
3503 | } | 3830 | } |
3504 | } | ||
3505 | } | 3831 | } |
3506 | 3832 | ||
3507 | /* | 3833 | /* |
@@ -3514,375 +3840,390 @@ static void flush_old_journal_lists(struct super_block *s) { | |||
3514 | ** If the journal is aborted, we just clean up. Things like flushing | 3840 | ** If the journal is aborted, we just clean up. Things like flushing |
3515 | ** journal lists, etc just won't happen. | 3841 | ** journal lists, etc just won't happen. |
3516 | */ | 3842 | */ |
3517 | static int do_journal_end(struct reiserfs_transaction_handle *th, struct super_block * p_s_sb, unsigned long nblocks, | 3843 | static int do_journal_end(struct reiserfs_transaction_handle *th, |
3518 | int flags) { | 3844 | struct super_block *p_s_sb, unsigned long nblocks, |
3519 | struct reiserfs_journal *journal = SB_JOURNAL (p_s_sb); | 3845 | int flags) |
3520 | struct reiserfs_journal_cnode *cn, *next, *jl_cn; | 3846 | { |
3521 | struct reiserfs_journal_cnode *last_cn = NULL; | 3847 | struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb); |
3522 | struct reiserfs_journal_desc *desc ; | 3848 | struct reiserfs_journal_cnode *cn, *next, *jl_cn; |
3523 | struct reiserfs_journal_commit *commit ; | 3849 | struct reiserfs_journal_cnode *last_cn = NULL; |
3524 | struct buffer_head *c_bh ; /* commit bh */ | 3850 | struct reiserfs_journal_desc *desc; |
3525 | struct buffer_head *d_bh ; /* desc bh */ | 3851 | struct reiserfs_journal_commit *commit; |
3526 | int cur_write_start = 0 ; /* start index of current log write */ | 3852 | struct buffer_head *c_bh; /* commit bh */ |
3527 | int old_start ; | 3853 | struct buffer_head *d_bh; /* desc bh */ |
3528 | int i ; | 3854 | int cur_write_start = 0; /* start index of current log write */ |
3529 | int flush = flags & FLUSH_ALL ; | 3855 | int old_start; |
3530 | int wait_on_commit = flags & WAIT ; | 3856 | int i; |
3531 | struct reiserfs_journal_list *jl, *temp_jl; | 3857 | int flush = flags & FLUSH_ALL; |
3532 | struct list_head *entry, *safe; | 3858 | int wait_on_commit = flags & WAIT; |
3533 | unsigned long jindex; | 3859 | struct reiserfs_journal_list *jl, *temp_jl; |
3534 | unsigned long commit_trans_id; | 3860 | struct list_head *entry, *safe; |
3535 | int trans_half; | 3861 | unsigned long jindex; |
3536 | 3862 | unsigned long commit_trans_id; | |
3537 | BUG_ON (th->t_refcount > 1); | 3863 | int trans_half; |
3538 | BUG_ON (!th->t_trans_id); | 3864 | |
3539 | 3865 | BUG_ON(th->t_refcount > 1); | |
3540 | put_fs_excl(); | 3866 | BUG_ON(!th->t_trans_id); |
3541 | current->journal_info = th->t_handle_save; | 3867 | |
3542 | reiserfs_check_lock_depth(p_s_sb, "journal end"); | 3868 | put_fs_excl(); |
3543 | if (journal->j_len == 0) { | 3869 | current->journal_info = th->t_handle_save; |
3544 | reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ; | 3870 | reiserfs_check_lock_depth(p_s_sb, "journal end"); |
3545 | journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ; | 3871 | if (journal->j_len == 0) { |
3546 | } | 3872 | reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), |
3547 | 3873 | 1); | |
3548 | lock_journal(p_s_sb) ; | 3874 | journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)); |
3549 | if (journal->j_next_full_flush) { | 3875 | } |
3550 | flags |= FLUSH_ALL ; | ||
3551 | flush = 1 ; | ||
3552 | } | ||
3553 | if (journal->j_next_async_flush) { | ||
3554 | flags |= COMMIT_NOW | WAIT; | ||
3555 | wait_on_commit = 1; | ||
3556 | } | ||
3557 | |||
3558 | /* check_journal_end locks the journal, and unlocks if it does not return 1 | ||
3559 | ** it tells us if we should continue with the journal_end, or just return | ||
3560 | */ | ||
3561 | if (!check_journal_end(th, p_s_sb, nblocks, flags)) { | ||
3562 | p_s_sb->s_dirt = 1; | ||
3563 | wake_queued_writers(p_s_sb); | ||
3564 | reiserfs_async_progress_wait(p_s_sb); | ||
3565 | goto out ; | ||
3566 | } | ||
3567 | |||
3568 | /* check_journal_end might set these, check again */ | ||
3569 | if (journal->j_next_full_flush) { | ||
3570 | flush = 1 ; | ||
3571 | } | ||
3572 | |||
3573 | /* | ||
3574 | ** j must wait means we have to flush the log blocks, and the real blocks for | ||
3575 | ** this transaction | ||
3576 | */ | ||
3577 | if (journal->j_must_wait > 0) { | ||
3578 | flush = 1 ; | ||
3579 | } | ||
3580 | 3876 | ||
3877 | lock_journal(p_s_sb); | ||
3878 | if (journal->j_next_full_flush) { | ||
3879 | flags |= FLUSH_ALL; | ||
3880 | flush = 1; | ||
3881 | } | ||
3882 | if (journal->j_next_async_flush) { | ||
3883 | flags |= COMMIT_NOW | WAIT; | ||
3884 | wait_on_commit = 1; | ||
3885 | } | ||
3886 | |||
3887 | /* check_journal_end locks the journal, and unlocks if it does not return 1 | ||
3888 | ** it tells us if we should continue with the journal_end, or just return | ||
3889 | */ | ||
3890 | if (!check_journal_end(th, p_s_sb, nblocks, flags)) { | ||
3891 | p_s_sb->s_dirt = 1; | ||
3892 | wake_queued_writers(p_s_sb); | ||
3893 | reiserfs_async_progress_wait(p_s_sb); | ||
3894 | goto out; | ||
3895 | } | ||
3896 | |||
3897 | /* check_journal_end might set these, check again */ | ||
3898 | if (journal->j_next_full_flush) { | ||
3899 | flush = 1; | ||
3900 | } | ||
3901 | |||
3902 | /* | ||
3903 | ** j must wait means we have to flush the log blocks, and the real blocks for | ||
3904 | ** this transaction | ||
3905 | */ | ||
3906 | if (journal->j_must_wait > 0) { | ||
3907 | flush = 1; | ||
3908 | } | ||
3581 | #ifdef REISERFS_PREALLOCATE | 3909 | #ifdef REISERFS_PREALLOCATE |
3582 | /* quota ops might need to nest, setup the journal_info pointer for them */ | 3910 | /* quota ops might need to nest, setup the journal_info pointer for them */ |
3583 | current->journal_info = th ; | 3911 | current->journal_info = th; |
3584 | reiserfs_discard_all_prealloc(th); /* it should not involve new blocks into | 3912 | reiserfs_discard_all_prealloc(th); /* it should not involve new blocks into |
3585 | * the transaction */ | 3913 | * the transaction */ |
3586 | current->journal_info = th->t_handle_save ; | 3914 | current->journal_info = th->t_handle_save; |
3587 | #endif | 3915 | #endif |
3588 | 3916 | ||
3589 | /* setup description block */ | 3917 | /* setup description block */ |
3590 | d_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + journal->j_start) ; | 3918 | d_bh = |
3591 | set_buffer_uptodate(d_bh); | 3919 | journal_getblk(p_s_sb, |
3592 | desc = (struct reiserfs_journal_desc *)(d_bh)->b_data ; | 3920 | SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + |
3593 | memset(d_bh->b_data, 0, d_bh->b_size) ; | 3921 | journal->j_start); |
3594 | memcpy(get_journal_desc_magic (d_bh), JOURNAL_DESC_MAGIC, 8) ; | 3922 | set_buffer_uptodate(d_bh); |
3595 | set_desc_trans_id(desc, journal->j_trans_id) ; | 3923 | desc = (struct reiserfs_journal_desc *)(d_bh)->b_data; |
3596 | 3924 | memset(d_bh->b_data, 0, d_bh->b_size); | |
3597 | /* setup commit block. Don't write (keep it clean too) this one until after everyone else is written */ | 3925 | memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8); |
3598 | c_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + | 3926 | set_desc_trans_id(desc, journal->j_trans_id); |
3599 | ((journal->j_start + journal->j_len + 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb))) ; | 3927 | |
3600 | commit = (struct reiserfs_journal_commit *)c_bh->b_data ; | 3928 | /* setup commit block. Don't write (keep it clean too) this one until after everyone else is written */ |
3601 | memset(c_bh->b_data, 0, c_bh->b_size) ; | 3929 | c_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + |
3602 | set_commit_trans_id(commit, journal->j_trans_id) ; | 3930 | ((journal->j_start + journal->j_len + |
3603 | set_buffer_uptodate(c_bh) ; | 3931 | 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb))); |
3604 | 3932 | commit = (struct reiserfs_journal_commit *)c_bh->b_data; | |
3605 | /* init this journal list */ | 3933 | memset(c_bh->b_data, 0, c_bh->b_size); |
3606 | jl = journal->j_current_jl; | 3934 | set_commit_trans_id(commit, journal->j_trans_id); |
3607 | 3935 | set_buffer_uptodate(c_bh); | |
3608 | /* we lock the commit before doing anything because | 3936 | |
3609 | * we want to make sure nobody tries to run flush_commit_list until | 3937 | /* init this journal list */ |
3610 | * the new transaction is fully setup, and we've already flushed the | 3938 | jl = journal->j_current_jl; |
3611 | * ordered bh list | 3939 | |
3612 | */ | 3940 | /* we lock the commit before doing anything because |
3613 | down(&jl->j_commit_lock); | 3941 | * we want to make sure nobody tries to run flush_commit_list until |
3614 | 3942 | * the new transaction is fully setup, and we've already flushed the | |
3615 | /* save the transaction id in case we need to commit it later */ | 3943 | * ordered bh list |
3616 | commit_trans_id = jl->j_trans_id; | 3944 | */ |
3617 | 3945 | down(&jl->j_commit_lock); | |
3618 | atomic_set(&jl->j_older_commits_done, 0) ; | 3946 | |
3619 | jl->j_trans_id = journal->j_trans_id ; | 3947 | /* save the transaction id in case we need to commit it later */ |
3620 | jl->j_timestamp = journal->j_trans_start_time ; | 3948 | commit_trans_id = jl->j_trans_id; |
3621 | jl->j_commit_bh = c_bh ; | 3949 | |
3622 | jl->j_start = journal->j_start ; | 3950 | atomic_set(&jl->j_older_commits_done, 0); |
3623 | jl->j_len = journal->j_len ; | 3951 | jl->j_trans_id = journal->j_trans_id; |
3624 | atomic_set(&jl->j_nonzerolen, journal->j_len) ; | 3952 | jl->j_timestamp = journal->j_trans_start_time; |
3625 | atomic_set(&jl->j_commit_left, journal->j_len + 2); | 3953 | jl->j_commit_bh = c_bh; |
3626 | jl->j_realblock = NULL ; | 3954 | jl->j_start = journal->j_start; |
3627 | 3955 | jl->j_len = journal->j_len; | |
3628 | /* The ENTIRE FOR LOOP MUST not cause schedule to occur. | 3956 | atomic_set(&jl->j_nonzerolen, journal->j_len); |
3629 | ** for each real block, add it to the journal list hash, | 3957 | atomic_set(&jl->j_commit_left, journal->j_len + 2); |
3630 | ** copy into real block index array in the commit or desc block | 3958 | jl->j_realblock = NULL; |
3631 | */ | 3959 | |
3632 | trans_half = journal_trans_half(p_s_sb->s_blocksize); | 3960 | /* The ENTIRE FOR LOOP MUST not cause schedule to occur. |
3633 | for (i = 0, cn = journal->j_first ; cn ; cn = cn->next, i++) { | 3961 | ** for each real block, add it to the journal list hash, |
3634 | if (buffer_journaled (cn->bh)) { | 3962 | ** copy into real block index array in the commit or desc block |
3635 | jl_cn = get_cnode(p_s_sb) ; | 3963 | */ |
3636 | if (!jl_cn) { | 3964 | trans_half = journal_trans_half(p_s_sb->s_blocksize); |
3637 | reiserfs_panic(p_s_sb, "journal-1676, get_cnode returned NULL\n") ; | 3965 | for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) { |
3638 | } | 3966 | if (buffer_journaled(cn->bh)) { |
3639 | if (i == 0) { | 3967 | jl_cn = get_cnode(p_s_sb); |
3640 | jl->j_realblock = jl_cn ; | 3968 | if (!jl_cn) { |
3641 | } | 3969 | reiserfs_panic(p_s_sb, |
3642 | jl_cn->prev = last_cn ; | 3970 | "journal-1676, get_cnode returned NULL\n"); |
3643 | jl_cn->next = NULL ; | 3971 | } |
3644 | if (last_cn) { | 3972 | if (i == 0) { |
3645 | last_cn->next = jl_cn ; | 3973 | jl->j_realblock = jl_cn; |
3646 | } | 3974 | } |
3647 | last_cn = jl_cn ; | 3975 | jl_cn->prev = last_cn; |
3648 | /* make sure the block we are trying to log is not a block | 3976 | jl_cn->next = NULL; |
3649 | of journal or reserved area */ | 3977 | if (last_cn) { |
3650 | 3978 | last_cn->next = jl_cn; | |
3651 | if (is_block_in_log_or_reserved_area(p_s_sb, cn->bh->b_blocknr)) { | 3979 | } |
3652 | reiserfs_panic(p_s_sb, "journal-2332: Trying to log block %lu, which is a log block\n", cn->bh->b_blocknr) ; | 3980 | last_cn = jl_cn; |
3653 | } | 3981 | /* make sure the block we are trying to log is not a block |
3654 | jl_cn->blocknr = cn->bh->b_blocknr ; | 3982 | of journal or reserved area */ |
3655 | jl_cn->state = 0 ; | 3983 | |
3656 | jl_cn->sb = p_s_sb; | 3984 | if (is_block_in_log_or_reserved_area |
3657 | jl_cn->bh = cn->bh ; | 3985 | (p_s_sb, cn->bh->b_blocknr)) { |
3658 | jl_cn->jlist = jl; | 3986 | reiserfs_panic(p_s_sb, |
3659 | insert_journal_hash(journal->j_list_hash_table, jl_cn) ; | 3987 | "journal-2332: Trying to log block %lu, which is a log block\n", |
3660 | if (i < trans_half) { | 3988 | cn->bh->b_blocknr); |
3661 | desc->j_realblock[i] = cpu_to_le32(cn->bh->b_blocknr) ; | 3989 | } |
3662 | } else { | 3990 | jl_cn->blocknr = cn->bh->b_blocknr; |
3663 | commit->j_realblock[i - trans_half] = cpu_to_le32(cn->bh->b_blocknr) ; | 3991 | jl_cn->state = 0; |
3664 | } | 3992 | jl_cn->sb = p_s_sb; |
3665 | } else { | 3993 | jl_cn->bh = cn->bh; |
3666 | i-- ; | 3994 | jl_cn->jlist = jl; |
3667 | } | 3995 | insert_journal_hash(journal->j_list_hash_table, jl_cn); |
3668 | } | 3996 | if (i < trans_half) { |
3669 | set_desc_trans_len(desc, journal->j_len) ; | 3997 | desc->j_realblock[i] = |
3670 | set_desc_mount_id(desc, journal->j_mount_id) ; | 3998 | cpu_to_le32(cn->bh->b_blocknr); |
3671 | set_desc_trans_id(desc, journal->j_trans_id) ; | 3999 | } else { |
3672 | set_commit_trans_len(commit, journal->j_len); | 4000 | commit->j_realblock[i - trans_half] = |
3673 | 4001 | cpu_to_le32(cn->bh->b_blocknr); | |
3674 | /* special check in case all buffers in the journal were marked for not logging */ | 4002 | } |
3675 | if (journal->j_len == 0) { | 4003 | } else { |
3676 | BUG(); | 4004 | i--; |
3677 | } | 4005 | } |
3678 | 4006 | } | |
3679 | /* we're about to dirty all the log blocks, mark the description block | 4007 | set_desc_trans_len(desc, journal->j_len); |
3680 | * dirty now too. Don't mark the commit block dirty until all the | 4008 | set_desc_mount_id(desc, journal->j_mount_id); |
3681 | * others are on disk | 4009 | set_desc_trans_id(desc, journal->j_trans_id); |
3682 | */ | 4010 | set_commit_trans_len(commit, journal->j_len); |
3683 | mark_buffer_dirty(d_bh); | 4011 | |
3684 | 4012 | /* special check in case all buffers in the journal were marked for not logging */ | |
3685 | /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */ | 4013 | if (journal->j_len == 0) { |
3686 | cur_write_start = journal->j_start ; | 4014 | BUG(); |
3687 | cn = journal->j_first ; | 4015 | } |
3688 | jindex = 1 ; /* start at one so we don't get the desc again */ | 4016 | |
3689 | while(cn) { | 4017 | /* we're about to dirty all the log blocks, mark the description block |
3690 | clear_buffer_journal_new (cn->bh); | 4018 | * dirty now too. Don't mark the commit block dirty until all the |
3691 | /* copy all the real blocks into log area. dirty log blocks */ | 4019 | * others are on disk |
3692 | if (buffer_journaled (cn->bh)) { | 4020 | */ |
3693 | struct buffer_head *tmp_bh ; | 4021 | mark_buffer_dirty(d_bh); |
3694 | char *addr; | 4022 | |
3695 | struct page *page; | 4023 | /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */ |
3696 | tmp_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + | 4024 | cur_write_start = journal->j_start; |
3697 | ((cur_write_start + jindex) % SB_ONDISK_JOURNAL_SIZE(p_s_sb))) ; | 4025 | cn = journal->j_first; |
3698 | set_buffer_uptodate(tmp_bh); | 4026 | jindex = 1; /* start at one so we don't get the desc again */ |
3699 | page = cn->bh->b_page; | 4027 | while (cn) { |
3700 | addr = kmap(page); | 4028 | clear_buffer_journal_new(cn->bh); |
3701 | memcpy(tmp_bh->b_data, addr + offset_in_page(cn->bh->b_data), | 4029 | /* copy all the real blocks into log area. dirty log blocks */ |
3702 | cn->bh->b_size); | 4030 | if (buffer_journaled(cn->bh)) { |
3703 | kunmap(page); | 4031 | struct buffer_head *tmp_bh; |
3704 | mark_buffer_dirty(tmp_bh); | 4032 | char *addr; |
3705 | jindex++ ; | 4033 | struct page *page; |
3706 | set_buffer_journal_dirty (cn->bh); | 4034 | tmp_bh = |
3707 | clear_buffer_journaled (cn->bh); | 4035 | journal_getblk(p_s_sb, |
3708 | } else { | 4036 | SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + |
3709 | /* JDirty cleared sometime during transaction. don't log this one */ | 4037 | ((cur_write_start + |
3710 | reiserfs_warning(p_s_sb, "journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!") ; | 4038 | jindex) % |
3711 | brelse(cn->bh) ; | 4039 | SB_ONDISK_JOURNAL_SIZE(p_s_sb))); |
3712 | } | 4040 | set_buffer_uptodate(tmp_bh); |
3713 | next = cn->next ; | 4041 | page = cn->bh->b_page; |
3714 | free_cnode(p_s_sb, cn) ; | 4042 | addr = kmap(page); |
3715 | cn = next ; | 4043 | memcpy(tmp_bh->b_data, |
3716 | cond_resched(); | 4044 | addr + offset_in_page(cn->bh->b_data), |
3717 | } | 4045 | cn->bh->b_size); |
3718 | 4046 | kunmap(page); | |
3719 | /* we are done with both the c_bh and d_bh, but | 4047 | mark_buffer_dirty(tmp_bh); |
3720 | ** c_bh must be written after all other commit blocks, | 4048 | jindex++; |
3721 | ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1. | 4049 | set_buffer_journal_dirty(cn->bh); |
3722 | */ | 4050 | clear_buffer_journaled(cn->bh); |
3723 | 4051 | } else { | |
3724 | journal->j_current_jl = alloc_journal_list(p_s_sb); | 4052 | /* JDirty cleared sometime during transaction. don't log this one */ |
3725 | 4053 | reiserfs_warning(p_s_sb, | |
3726 | /* now it is safe to insert this transaction on the main list */ | 4054 | "journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!"); |
3727 | list_add_tail(&jl->j_list, &journal->j_journal_list); | 4055 | brelse(cn->bh); |
3728 | list_add_tail(&jl->j_working_list, &journal->j_working_list); | 4056 | } |
3729 | journal->j_num_work_lists++; | 4057 | next = cn->next; |
3730 | 4058 | free_cnode(p_s_sb, cn); | |
3731 | /* reset journal values for the next transaction */ | 4059 | cn = next; |
3732 | old_start = journal->j_start ; | 4060 | cond_resched(); |
3733 | journal->j_start = (journal->j_start + journal->j_len + 2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb); | 4061 | } |
3734 | atomic_set(&(journal->j_wcount), 0) ; | 4062 | |
3735 | journal->j_bcount = 0 ; | 4063 | /* we are done with both the c_bh and d_bh, but |
3736 | journal->j_last = NULL ; | 4064 | ** c_bh must be written after all other commit blocks, |
3737 | journal->j_first = NULL ; | 4065 | ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1. |
3738 | journal->j_len = 0 ; | 4066 | */ |
3739 | journal->j_trans_start_time = 0 ; | 4067 | |
3740 | journal->j_trans_id++ ; | 4068 | journal->j_current_jl = alloc_journal_list(p_s_sb); |
3741 | journal->j_current_jl->j_trans_id = journal->j_trans_id; | 4069 | |
3742 | journal->j_must_wait = 0 ; | 4070 | /* now it is safe to insert this transaction on the main list */ |
3743 | journal->j_len_alloc = 0 ; | 4071 | list_add_tail(&jl->j_list, &journal->j_journal_list); |
3744 | journal->j_next_full_flush = 0 ; | 4072 | list_add_tail(&jl->j_working_list, &journal->j_working_list); |
3745 | journal->j_next_async_flush = 0 ; | 4073 | journal->j_num_work_lists++; |
3746 | init_journal_hash(p_s_sb) ; | 4074 | |
3747 | 4075 | /* reset journal values for the next transaction */ | |
3748 | // make sure reiserfs_add_jh sees the new current_jl before we | 4076 | old_start = journal->j_start; |
3749 | // write out the tails | 4077 | journal->j_start = |
3750 | smp_mb(); | 4078 | (journal->j_start + journal->j_len + |
3751 | 4079 | 2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb); | |
3752 | /* tail conversion targets have to hit the disk before we end the | 4080 | atomic_set(&(journal->j_wcount), 0); |
3753 | * transaction. Otherwise a later transaction might repack the tail | 4081 | journal->j_bcount = 0; |
3754 | * before this transaction commits, leaving the data block unflushed and | 4082 | journal->j_last = NULL; |
3755 | * clean, if we crash before the later transaction commits, the data block | 4083 | journal->j_first = NULL; |
3756 | * is lost. | 4084 | journal->j_len = 0; |
3757 | */ | 4085 | journal->j_trans_start_time = 0; |
3758 | if (!list_empty(&jl->j_tail_bh_list)) { | 4086 | journal->j_trans_id++; |
3759 | unlock_kernel(); | 4087 | journal->j_current_jl->j_trans_id = journal->j_trans_id; |
3760 | write_ordered_buffers(&journal->j_dirty_buffers_lock, | 4088 | journal->j_must_wait = 0; |
3761 | journal, jl, &jl->j_tail_bh_list); | 4089 | journal->j_len_alloc = 0; |
3762 | lock_kernel(); | 4090 | journal->j_next_full_flush = 0; |
3763 | } | 4091 | journal->j_next_async_flush = 0; |
3764 | if (!list_empty(&jl->j_tail_bh_list)) | 4092 | init_journal_hash(p_s_sb); |
3765 | BUG(); | 4093 | |
3766 | up(&jl->j_commit_lock); | 4094 | // make sure reiserfs_add_jh sees the new current_jl before we |
3767 | 4095 | // write out the tails | |
3768 | /* honor the flush wishes from the caller, simple commits can | 4096 | smp_mb(); |
3769 | ** be done outside the journal lock, they are done below | 4097 | |
3770 | ** | 4098 | /* tail conversion targets have to hit the disk before we end the |
3771 | ** if we don't flush the commit list right now, we put it into | 4099 | * transaction. Otherwise a later transaction might repack the tail |
3772 | ** the work queue so the people waiting on the async progress work | 4100 | * before this transaction commits, leaving the data block unflushed and |
3773 | ** queue don't wait for this proc to flush journal lists and such. | 4101 | * clean, if we crash before the later transaction commits, the data block |
3774 | */ | 4102 | * is lost. |
3775 | if (flush) { | 4103 | */ |
3776 | flush_commit_list(p_s_sb, jl, 1) ; | 4104 | if (!list_empty(&jl->j_tail_bh_list)) { |
3777 | flush_journal_list(p_s_sb, jl, 1) ; | 4105 | unlock_kernel(); |
3778 | } else if (!(jl->j_state & LIST_COMMIT_PENDING)) | 4106 | write_ordered_buffers(&journal->j_dirty_buffers_lock, |
3779 | queue_delayed_work(commit_wq, &journal->j_work, HZ/10); | 4107 | journal, jl, &jl->j_tail_bh_list); |
3780 | 4108 | lock_kernel(); | |
3781 | 4109 | } | |
3782 | /* if the next transaction has any chance of wrapping, flush | 4110 | if (!list_empty(&jl->j_tail_bh_list)) |
3783 | ** transactions that might get overwritten. If any journal lists are very | 4111 | BUG(); |
3784 | ** old flush them as well. | 4112 | up(&jl->j_commit_lock); |
3785 | */ | 4113 | |
3786 | first_jl: | 4114 | /* honor the flush wishes from the caller, simple commits can |
3787 | list_for_each_safe(entry, safe, &journal->j_journal_list) { | 4115 | ** be done outside the journal lock, they are done below |
3788 | temp_jl = JOURNAL_LIST_ENTRY(entry); | 4116 | ** |
3789 | if (journal->j_start <= temp_jl->j_start) { | 4117 | ** if we don't flush the commit list right now, we put it into |
3790 | if ((journal->j_start + journal->j_trans_max + 1) >= | 4118 | ** the work queue so the people waiting on the async progress work |
3791 | temp_jl->j_start) | 4119 | ** queue don't wait for this proc to flush journal lists and such. |
3792 | { | 4120 | */ |
3793 | flush_used_journal_lists(p_s_sb, temp_jl); | 4121 | if (flush) { |
3794 | goto first_jl; | 4122 | flush_commit_list(p_s_sb, jl, 1); |
3795 | } else if ((journal->j_start + | 4123 | flush_journal_list(p_s_sb, jl, 1); |
3796 | journal->j_trans_max + 1) < | 4124 | } else if (!(jl->j_state & LIST_COMMIT_PENDING)) |
3797 | SB_ONDISK_JOURNAL_SIZE(p_s_sb)) | 4125 | queue_delayed_work(commit_wq, &journal->j_work, HZ / 10); |
3798 | { | 4126 | |
3799 | /* if we don't cross into the next transaction and we don't | 4127 | /* if the next transaction has any chance of wrapping, flush |
3800 | * wrap, there is no way we can overlap any later transactions | 4128 | ** transactions that might get overwritten. If any journal lists are very |
3801 | * break now | 4129 | ** old flush them as well. |
3802 | */ | 4130 | */ |
3803 | break; | 4131 | first_jl: |
3804 | } | 4132 | list_for_each_safe(entry, safe, &journal->j_journal_list) { |
3805 | } else if ((journal->j_start + | 4133 | temp_jl = JOURNAL_LIST_ENTRY(entry); |
3806 | journal->j_trans_max + 1) > | 4134 | if (journal->j_start <= temp_jl->j_start) { |
3807 | SB_ONDISK_JOURNAL_SIZE(p_s_sb)) | 4135 | if ((journal->j_start + journal->j_trans_max + 1) >= |
3808 | { | 4136 | temp_jl->j_start) { |
3809 | if (((journal->j_start + journal->j_trans_max + 1) % | 4137 | flush_used_journal_lists(p_s_sb, temp_jl); |
3810 | SB_ONDISK_JOURNAL_SIZE(p_s_sb)) >= temp_jl->j_start) | 4138 | goto first_jl; |
3811 | { | 4139 | } else if ((journal->j_start + |
3812 | flush_used_journal_lists(p_s_sb, temp_jl); | 4140 | journal->j_trans_max + 1) < |
3813 | goto first_jl; | 4141 | SB_ONDISK_JOURNAL_SIZE(p_s_sb)) { |
3814 | } else { | 4142 | /* if we don't cross into the next transaction and we don't |
3815 | /* we don't overlap anything from out start to the end of the | 4143 | * wrap, there is no way we can overlap any later transactions |
3816 | * log, and our wrapped portion doesn't overlap anything at | 4144 | * break now |
3817 | * the start of the log. We can break | 4145 | */ |
3818 | */ | 4146 | break; |
3819 | break; | 4147 | } |
3820 | } | 4148 | } else if ((journal->j_start + |
3821 | } | 4149 | journal->j_trans_max + 1) > |
3822 | } | 4150 | SB_ONDISK_JOURNAL_SIZE(p_s_sb)) { |
3823 | flush_old_journal_lists(p_s_sb); | 4151 | if (((journal->j_start + journal->j_trans_max + 1) % |
3824 | 4152 | SB_ONDISK_JOURNAL_SIZE(p_s_sb)) >= | |
3825 | journal->j_current_jl->j_list_bitmap = get_list_bitmap(p_s_sb, journal->j_current_jl) ; | 4153 | temp_jl->j_start) { |
3826 | 4154 | flush_used_journal_lists(p_s_sb, temp_jl); | |
3827 | if (!(journal->j_current_jl->j_list_bitmap)) { | 4155 | goto first_jl; |
3828 | reiserfs_panic(p_s_sb, "journal-1996: do_journal_end, could not get a list bitmap\n") ; | 4156 | } else { |
3829 | } | 4157 | /* we don't overlap anything from out start to the end of the |
3830 | 4158 | * log, and our wrapped portion doesn't overlap anything at | |
3831 | atomic_set(&(journal->j_jlock), 0) ; | 4159 | * the start of the log. We can break |
3832 | unlock_journal(p_s_sb) ; | 4160 | */ |
3833 | /* wake up any body waiting to join. */ | 4161 | break; |
3834 | clear_bit(J_WRITERS_QUEUED, &journal->j_state); | 4162 | } |
3835 | wake_up(&(journal->j_join_wait)) ; | 4163 | } |
3836 | 4164 | } | |
3837 | if (!flush && wait_on_commit && | 4165 | flush_old_journal_lists(p_s_sb); |
3838 | journal_list_still_alive(p_s_sb, commit_trans_id)) { | 4166 | |
3839 | flush_commit_list(p_s_sb, jl, 1) ; | 4167 | journal->j_current_jl->j_list_bitmap = |
3840 | } | 4168 | get_list_bitmap(p_s_sb, journal->j_current_jl); |
3841 | out: | 4169 | |
3842 | reiserfs_check_lock_depth(p_s_sb, "journal end2"); | 4170 | if (!(journal->j_current_jl->j_list_bitmap)) { |
3843 | 4171 | reiserfs_panic(p_s_sb, | |
3844 | memset (th, 0, sizeof (*th)); | 4172 | "journal-1996: do_journal_end, could not get a list bitmap\n"); |
3845 | /* Re-set th->t_super, so we can properly keep track of how many | 4173 | } |
3846 | * persistent transactions there are. We need to do this so if this | 4174 | |
3847 | * call is part of a failed restart_transaction, we can free it later */ | 4175 | atomic_set(&(journal->j_jlock), 0); |
3848 | th->t_super = p_s_sb; | 4176 | unlock_journal(p_s_sb); |
3849 | 4177 | /* wake up any body waiting to join. */ | |
3850 | return journal->j_errno; | 4178 | clear_bit(J_WRITERS_QUEUED, &journal->j_state); |
3851 | } | 4179 | wake_up(&(journal->j_join_wait)); |
3852 | 4180 | ||
3853 | static void | 4181 | if (!flush && wait_on_commit && |
3854 | __reiserfs_journal_abort_hard (struct super_block *sb) | 4182 | journal_list_still_alive(p_s_sb, commit_trans_id)) { |
3855 | { | 4183 | flush_commit_list(p_s_sb, jl, 1); |
3856 | struct reiserfs_journal *journal = SB_JOURNAL (sb); | 4184 | } |
3857 | if (test_bit (J_ABORTED, &journal->j_state)) | 4185 | out: |
3858 | return; | 4186 | reiserfs_check_lock_depth(p_s_sb, "journal end2"); |
3859 | 4187 | ||
3860 | printk (KERN_CRIT "REISERFS: Aborting journal for filesystem on %s\n", | 4188 | memset(th, 0, sizeof(*th)); |
3861 | reiserfs_bdevname (sb)); | 4189 | /* Re-set th->t_super, so we can properly keep track of how many |
3862 | 4190 | * persistent transactions there are. We need to do this so if this | |
3863 | sb->s_flags |= MS_RDONLY; | 4191 | * call is part of a failed restart_transaction, we can free it later */ |
3864 | set_bit (J_ABORTED, &journal->j_state); | 4192 | th->t_super = p_s_sb; |
4193 | |||
4194 | return journal->j_errno; | ||
4195 | } | ||
4196 | |||
4197 | static void __reiserfs_journal_abort_hard(struct super_block *sb) | ||
4198 | { | ||
4199 | struct reiserfs_journal *journal = SB_JOURNAL(sb); | ||
4200 | if (test_bit(J_ABORTED, &journal->j_state)) | ||
4201 | return; | ||
4202 | |||
4203 | printk(KERN_CRIT "REISERFS: Aborting journal for filesystem on %s\n", | ||
4204 | reiserfs_bdevname(sb)); | ||
4205 | |||
4206 | sb->s_flags |= MS_RDONLY; | ||
4207 | set_bit(J_ABORTED, &journal->j_state); | ||
3865 | 4208 | ||
3866 | #ifdef CONFIG_REISERFS_CHECK | 4209 | #ifdef CONFIG_REISERFS_CHECK |
3867 | dump_stack(); | 4210 | dump_stack(); |
3868 | #endif | 4211 | #endif |
3869 | } | 4212 | } |
3870 | 4213 | ||
3871 | static void | 4214 | static void __reiserfs_journal_abort_soft(struct super_block *sb, int errno) |
3872 | __reiserfs_journal_abort_soft (struct super_block *sb, int errno) | ||
3873 | { | 4215 | { |
3874 | struct reiserfs_journal *journal = SB_JOURNAL (sb); | 4216 | struct reiserfs_journal *journal = SB_JOURNAL(sb); |
3875 | if (test_bit (J_ABORTED, &journal->j_state)) | 4217 | if (test_bit(J_ABORTED, &journal->j_state)) |
3876 | return; | 4218 | return; |
3877 | 4219 | ||
3878 | if (!journal->j_errno) | 4220 | if (!journal->j_errno) |
3879 | journal->j_errno = errno; | 4221 | journal->j_errno = errno; |
3880 | 4222 | ||
3881 | __reiserfs_journal_abort_hard (sb); | 4223 | __reiserfs_journal_abort_hard(sb); |
3882 | } | 4224 | } |
3883 | 4225 | ||
3884 | void | 4226 | void reiserfs_journal_abort(struct super_block *sb, int errno) |
3885 | reiserfs_journal_abort (struct super_block *sb, int errno) | ||
3886 | { | 4227 | { |
3887 | return __reiserfs_journal_abort_soft (sb, errno); | 4228 | return __reiserfs_journal_abort_soft(sb, errno); |
3888 | } | 4229 | } |