| 1 |
// SPDX-License-Identifier: GPL-2.0 |
| 2 |
/* |
| 3 |
* Interface between ext4 and JBD |
| 4 |
*/ |
| 5 |
|
| 6 |
#include "ext4_jbd2.h" |
| 7 |
|
| 8 |
#include <trace/events/ext4.h> |
| 9 |
|
| 10 |
int ext4_inode_journal_mode(struct inode *inode) |
| 11 |
{ |
| 12 |
if (EXT4_JOURNAL(inode) == NULL) |
| 13 |
return EXT4_INODE_WRITEBACK_DATA_MODE; /* writeback */ |
| 14 |
/* We do not support data journalling with delayed allocation */ |
| 15 |
if (!S_ISREG(inode->i_mode) || |
| 16 |
ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE) || |
| 17 |
test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA || |
| 18 |
(ext4_test_inode_flag(inode, EXT4_INODE_JOURNAL_DATA) && |
| 19 |
!test_opt(inode->i_sb, DELALLOC))) { |
| 20 |
/* We do not support data journalling for encrypted data */ |
| 21 |
if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode)) |
| 22 |
return EXT4_INODE_ORDERED_DATA_MODE; /* ordered */ |
| 23 |
return EXT4_INODE_JOURNAL_DATA_MODE; /* journal data */ |
| 24 |
} |
| 25 |
if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA) |
| 26 |
return EXT4_INODE_ORDERED_DATA_MODE; /* ordered */ |
| 27 |
if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA) |
| 28 |
return EXT4_INODE_WRITEBACK_DATA_MODE; /* writeback */ |
| 29 |
BUG(); |
| 30 |
} |
| 31 |
|
| 32 |
/* Just increment the non-pointer handle value */ |
| 33 |
static handle_t *ext4_get_nojournal(void) |
| 34 |
{ |
| 35 |
handle_t *handle = current->journal_info; |
| 36 |
unsigned long ref_cnt = (unsigned long)handle; |
| 37 |
|
| 38 |
BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT); |
| 39 |
|
| 40 |
ref_cnt++; |
| 41 |
handle = (handle_t *)ref_cnt; |
| 42 |
|
| 43 |
current->journal_info = handle; |
| 44 |
return handle; |
| 45 |
} |
| 46 |
|
| 47 |
|
| 48 |
/* Decrement the non-pointer handle value */ |
| 49 |
static void ext4_put_nojournal(handle_t *handle) |
| 50 |
{ |
| 51 |
unsigned long ref_cnt = (unsigned long)handle; |
| 52 |
|
| 53 |
BUG_ON(ref_cnt == 0); |
| 54 |
|
| 55 |
ref_cnt--; |
| 56 |
handle = (handle_t *)ref_cnt; |
| 57 |
|
| 58 |
current->journal_info = handle; |
| 59 |
} |
| 60 |
|
| 61 |
/* |
| 62 |
* Wrappers for jbd2_journal_start/end. |
| 63 |
*/ |
| 64 |
static int ext4_journal_check_start(struct super_block *sb) |
| 65 |
{ |
| 66 |
int ret; |
| 67 |
journal_t *journal; |
| 68 |
|
| 69 |
might_sleep(); |
| 70 |
|
| 71 |
ret = ext4_emergency_state(sb); |
| 72 |
if (unlikely(ret)) |
| 73 |
return ret; |
| 74 |
|
| 75 |
if (WARN_ON_ONCE(sb_rdonly(sb))) |
| 76 |
return -EROFS; |
| 77 |
|
| 78 |
WARN_ON(sb->s_writers.frozen == SB_FREEZE_COMPLETE); |
| 79 |
journal = EXT4_SB(sb)->s_journal; |
| 80 |
/* |
| 81 |
* Special case here: if the journal has aborted behind our |
| 82 |
* backs (eg. EIO in the commit thread), then we still need to |
| 83 |
* take the FS itself readonly cleanly. |
| 84 |
*/ |
| 85 |
if (journal && is_journal_aborted(journal)) { |
| 86 |
ext4_abort(sb, -journal->j_errno, "Detected aborted journal"); |
| 87 |
return -EROFS; |
| 88 |
} |
| 89 |
return 0; |
| 90 |
} |
| 91 |
|
| 92 |
handle_t *__ext4_journal_start_sb(struct inode *inode, |
| 93 |
struct super_block *sb, unsigned int line, |
| 94 |
int type, int blocks, int rsv_blocks, |
| 95 |
int revoke_creds) |
| 96 |
{ |
| 97 |
journal_t *journal; |
| 98 |
int err; |
| 99 |
if (inode) |
| 100 |
trace_ext4_journal_start_inode(inode, blocks, rsv_blocks, |
| 101 |
revoke_creds, type, |
| 102 |
_RET_IP_); |
| 103 |
else |
| 104 |
trace_ext4_journal_start_sb(sb, blocks, rsv_blocks, |
| 105 |
revoke_creds, type, |
| 106 |
_RET_IP_); |
| 107 |
err = ext4_journal_check_start(sb); |
| 108 |
if (err < 0) |
| 109 |
return ERR_PTR(err); |
| 110 |
|
| 111 |
journal = EXT4_SB(sb)->s_journal; |
| 112 |
if (!journal || (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY)) |
| 113 |
return ext4_get_nojournal(); |
| 114 |
return jbd2__journal_start(journal, blocks, rsv_blocks, revoke_creds, |
| 115 |
GFP_NOFS, type, line); |
| 116 |
} |
| 117 |
|
| 118 |
int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle) |
| 119 |
{ |
| 120 |
struct super_block *sb; |
| 121 |
int err; |
| 122 |
int rc; |
| 123 |
|
| 124 |
if (!ext4_handle_valid(handle)) { |
| 125 |
ext4_put_nojournal(handle); |
| 126 |
return 0; |
| 127 |
} |
| 128 |
|
| 129 |
err = handle->h_err; |
| 130 |
if (!handle->h_transaction) { |
| 131 |
rc = jbd2_journal_stop(handle); |
| 132 |
return err ? err : rc; |
| 133 |
} |
| 134 |
|
| 135 |
sb = handle->h_transaction->t_journal->j_private; |
| 136 |
rc = jbd2_journal_stop(handle); |
| 137 |
|
| 138 |
if (!err) |
| 139 |
err = rc; |
| 140 |
if (err) |
| 141 |
__ext4_std_error(sb, where, line, err); |
| 142 |
return err; |
| 143 |
} |
| 144 |
|
| 145 |
handle_t *__ext4_journal_start_reserved(handle_t *handle, unsigned int line, |
| 146 |
int type) |
| 147 |
{ |
| 148 |
struct super_block *sb; |
| 149 |
int err; |
| 150 |
|
| 151 |
if (!ext4_handle_valid(handle)) |
| 152 |
return ext4_get_nojournal(); |
| 153 |
|
| 154 |
sb = handle->h_journal->j_private; |
| 155 |
trace_ext4_journal_start_reserved(sb, |
| 156 |
jbd2_handle_buffer_credits(handle), _RET_IP_); |
| 157 |
err = ext4_journal_check_start(sb); |
| 158 |
if (err < 0) { |
| 159 |
jbd2_journal_free_reserved(handle); |
| 160 |
return ERR_PTR(err); |
| 161 |
} |
| 162 |
|
| 163 |
err = jbd2_journal_start_reserved(handle, type, line); |
| 164 |
if (err < 0) |
| 165 |
return ERR_PTR(err); |
| 166 |
return handle; |
| 167 |
} |
| 168 |
|
| 169 |
int __ext4_journal_ensure_credits(handle_t *handle, int check_cred, |
| 170 |
int extend_cred, int revoke_cred) |
| 171 |
{ |
| 172 |
if (!ext4_handle_valid(handle)) |
| 173 |
return 0; |
| 174 |
if (is_handle_aborted(handle)) |
| 175 |
return -EROFS; |
| 176 |
if (jbd2_handle_buffer_credits(handle) >= check_cred && |
| 177 |
handle->h_revoke_credits >= revoke_cred) |
| 178 |
return 0; |
| 179 |
extend_cred = max(0, extend_cred - jbd2_handle_buffer_credits(handle)); |
| 180 |
revoke_cred = max(0, revoke_cred - handle->h_revoke_credits); |
| 181 |
return ext4_journal_extend(handle, extend_cred, revoke_cred); |
| 182 |
} |
| 183 |
|
| 184 |
static void ext4_journal_abort_handle(const char *caller, unsigned int line, |
| 185 |
const char *err_fn, |
| 186 |
struct buffer_head *bh, |
| 187 |
handle_t *handle, int err) |
| 188 |
{ |
| 189 |
char nbuf[16]; |
| 190 |
const char *errstr = ext4_decode_error(NULL, err, nbuf); |
| 191 |
|
| 192 |
BUG_ON(!ext4_handle_valid(handle)); |
| 193 |
|
| 194 |
if (bh) |
| 195 |
BUFFER_TRACE(bh, "abort"); |
| 196 |
|
| 197 |
if (!handle->h_err) |
| 198 |
handle->h_err = err; |
| 199 |
|
| 200 |
if (is_handle_aborted(handle)) |
| 201 |
return; |
| 202 |
|
| 203 |
printk(KERN_ERR "EXT4-fs: %s:%d: aborting transaction: %s in %s\n", |
| 204 |
caller, line, errstr, err_fn); |
| 205 |
|
| 206 |
jbd2_journal_abort_handle(handle); |
| 207 |
} |
| 208 |
|
| 209 |
static void ext4_check_bdev_write_error(struct super_block *sb) |
| 210 |
{ |
| 211 |
struct address_space *mapping = sb->s_bdev->bd_mapping; |
| 212 |
struct ext4_sb_info *sbi = EXT4_SB(sb); |
| 213 |
int err; |
| 214 |
|
| 215 |
/* |
| 216 |
* If the block device has write error flag, it may have failed to |
| 217 |
* async write out metadata buffers in the background. In this case, |
| 218 |
* we could read old data from disk and write it out again, which |
| 219 |
* may lead to on-disk filesystem inconsistency. |
| 220 |
*/ |
| 221 |
if (errseq_check(&mapping->wb_err, READ_ONCE(sbi->s_bdev_wb_err))) { |
| 222 |
spin_lock(&sbi->s_bdev_wb_lock); |
| 223 |
err = errseq_check_and_advance(&mapping->wb_err, &sbi->s_bdev_wb_err); |
| 224 |
spin_unlock(&sbi->s_bdev_wb_lock); |
| 225 |
if (err) |
| 226 |
ext4_error_err(sb, -err, |
| 227 |
"Error while async write back metadata"); |
| 228 |
} |
| 229 |
} |
| 230 |
|
| 231 |
int __ext4_journal_get_write_access(const char *where, unsigned int line, |
| 232 |
handle_t *handle, struct super_block *sb, |
| 233 |
struct buffer_head *bh, |
| 234 |
enum ext4_journal_trigger_type trigger_type) |
| 235 |
{ |
| 236 |
int err; |
| 237 |
|
| 238 |
might_sleep(); |
| 239 |
|
| 240 |
if (ext4_handle_valid(handle)) { |
| 241 |
err = jbd2_journal_get_write_access(handle, bh); |
| 242 |
if (err) { |
| 243 |
ext4_journal_abort_handle(where, line, __func__, bh, |
| 244 |
handle, err); |
| 245 |
return err; |
| 246 |
} |
| 247 |
} else |
| 248 |
ext4_check_bdev_write_error(sb); |
| 249 |
if (trigger_type == EXT4_JTR_NONE || |
| 250 |
!ext4_has_feature_metadata_csum(sb)) |
| 251 |
return 0; |
| 252 |
BUG_ON(trigger_type >= EXT4_JOURNAL_TRIGGER_COUNT); |
| 253 |
jbd2_journal_set_triggers(bh, |
| 254 |
&EXT4_SB(sb)->s_journal_triggers[trigger_type].tr_triggers); |
| 255 |
return 0; |
| 256 |
} |
| 257 |
|
| 258 |
/* |
| 259 |
* The ext4 forget function must perform a revoke if we are freeing data |
| 260 |
* which has been journaled. Metadata (eg. indirect blocks) must be |
| 261 |
* revoked in all cases. |
| 262 |
* |
| 263 |
* "bh" may be NULL: a metadata block may have been freed from memory |
| 264 |
* but there may still be a record of it in the journal, and that record |
| 265 |
* still needs to be revoked. |
| 266 |
*/ |
| 267 |
int __ext4_forget(const char *where, unsigned int line, handle_t *handle, |
| 268 |
int is_metadata, struct inode *inode, |
| 269 |
struct buffer_head *bh, ext4_fsblk_t blocknr) |
| 270 |
{ |
| 271 |
int err; |
| 272 |
|
| 273 |
might_sleep(); |
| 274 |
|
| 275 |
trace_ext4_forget(inode, is_metadata, blocknr); |
| 276 |
BUFFER_TRACE(bh, "enter"); |
| 277 |
|
| 278 |
ext4_debug("forgetting bh %p: is_metadata=%d, mode %o, data mode %x\n", |
| 279 |
bh, is_metadata, inode->i_mode, |
| 280 |
test_opt(inode->i_sb, DATA_FLAGS)); |
| 281 |
|
| 282 |
/* |
| 283 |
* In the no journal case, we should wait for the ongoing buffer |
| 284 |
* to complete and do a forget. |
| 285 |
*/ |
| 286 |
if (!ext4_handle_valid(handle)) { |
| 287 |
if (bh) { |
| 288 |
clear_buffer_dirty(bh); |
| 289 |
wait_on_buffer(bh); |
| 290 |
__bforget(bh); |
| 291 |
} |
| 292 |
return 0; |
| 293 |
} |
| 294 |
|
| 295 |
/* Never use the revoke function if we are doing full data |
| 296 |
* journaling: there is no need to, and a V1 superblock won't |
| 297 |
* support it. Otherwise, only skip the revoke on un-journaled |
| 298 |
* data blocks. */ |
| 299 |
|
| 300 |
if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA || |
| 301 |
(!is_metadata && !ext4_should_journal_data(inode))) { |
| 302 |
if (bh) { |
| 303 |
BUFFER_TRACE(bh, "call jbd2_journal_forget"); |
| 304 |
err = jbd2_journal_forget(handle, bh); |
| 305 |
if (err) |
| 306 |
ext4_journal_abort_handle(where, line, __func__, |
| 307 |
bh, handle, err); |
| 308 |
return err; |
| 309 |
} |
| 310 |
return 0; |
| 311 |
} |
| 312 |
|
| 313 |
/* |
| 314 |
* data!=journal && (is_metadata || should_journal_data(inode)) |
| 315 |
*/ |
| 316 |
BUFFER_TRACE(bh, "call jbd2_journal_revoke"); |
| 317 |
err = jbd2_journal_revoke(handle, blocknr, bh); |
| 318 |
if (err) { |
| 319 |
ext4_journal_abort_handle(where, line, __func__, |
| 320 |
bh, handle, err); |
| 321 |
__ext4_error(inode->i_sb, where, line, true, -err, 0, |
| 322 |
"error %d when attempting revoke", err); |
| 323 |
} |
| 324 |
BUFFER_TRACE(bh, "exit"); |
| 325 |
return err; |
| 326 |
} |
| 327 |
|
| 328 |
int __ext4_journal_get_create_access(const char *where, unsigned int line, |
| 329 |
handle_t *handle, struct super_block *sb, |
| 330 |
struct buffer_head *bh, |
| 331 |
enum ext4_journal_trigger_type trigger_type) |
| 332 |
{ |
| 333 |
int err; |
| 334 |
|
| 335 |
if (!ext4_handle_valid(handle)) |
| 336 |
return 0; |
| 337 |
|
| 338 |
err = jbd2_journal_get_create_access(handle, bh); |
| 339 |
if (err) { |
| 340 |
ext4_journal_abort_handle(where, line, __func__, bh, handle, |
| 341 |
err); |
| 342 |
return err; |
| 343 |
} |
| 344 |
if (trigger_type == EXT4_JTR_NONE || |
| 345 |
!ext4_has_feature_metadata_csum(sb)) |
| 346 |
return 0; |
| 347 |
BUG_ON(trigger_type >= EXT4_JOURNAL_TRIGGER_COUNT); |
| 348 |
jbd2_journal_set_triggers(bh, |
| 349 |
&EXT4_SB(sb)->s_journal_triggers[trigger_type].tr_triggers); |
| 350 |
return 0; |
| 351 |
} |
| 352 |
|
| 353 |
int __ext4_handle_dirty_metadata(const char *where, unsigned int line, |
| 354 |
handle_t *handle, struct inode *inode, |
| 355 |
struct buffer_head *bh) |
| 356 |
{ |
| 357 |
int err = 0; |
| 358 |
|
| 359 |
might_sleep(); |
| 360 |
|
| 361 |
set_buffer_meta(bh); |
| 362 |
set_buffer_prio(bh); |
| 363 |
set_buffer_uptodate(bh); |
| 364 |
if (ext4_handle_valid(handle)) { |
| 365 |
err = jbd2_journal_dirty_metadata(handle, bh); |
| 366 |
/* Errors can only happen due to aborted journal or a nasty bug */ |
| 367 |
if (!is_handle_aborted(handle) && WARN_ON_ONCE(err)) { |
| 368 |
ext4_journal_abort_handle(where, line, __func__, bh, |
| 369 |
handle, err); |
| 370 |
if (inode == NULL) { |
| 371 |
pr_err("EXT4: jbd2_journal_dirty_metadata " |
| 372 |
"failed: handle type %u started at " |
| 373 |
"line %u, credits %u/%u, errcode %d", |
| 374 |
handle->h_type, |
| 375 |
handle->h_line_no, |
| 376 |
handle->h_requested_credits, |
| 377 |
jbd2_handle_buffer_credits(handle), err); |
| 378 |
return err; |
| 379 |
} |
| 380 |
ext4_error_inode(inode, where, line, |
| 381 |
bh->b_blocknr, |
| 382 |
"journal_dirty_metadata failed: " |
| 383 |
"handle type %u started at line %u, " |
| 384 |
"credits %u/%u, errcode %d", |
| 385 |
handle->h_type, |
| 386 |
handle->h_line_no, |
| 387 |
handle->h_requested_credits, |
| 388 |
jbd2_handle_buffer_credits(handle), |
| 389 |
err); |
| 390 |
} |
| 391 |
} else { |
| 392 |
if (inode) |
| 393 |
mark_buffer_dirty_inode(bh, inode); |
| 394 |
else |
| 395 |
mark_buffer_dirty(bh); |
| 396 |
if (inode && inode_needs_sync(inode)) { |
| 397 |
sync_dirty_buffer(bh); |
| 398 |
if (buffer_req(bh) && !buffer_uptodate(bh)) { |
| 399 |
ext4_error_inode_err(inode, where, line, |
| 400 |
bh->b_blocknr, EIO, |
| 401 |
"IO error syncing itable block"); |
| 402 |
err = -EIO; |
| 403 |
} |
| 404 |
} |
| 405 |
} |
| 406 |
return err; |
| 407 |
} |