| 1 |
// SPDX-License-Identifier: GPL-2.0 |
| 2 |
/* |
| 3 |
* linux/fs/ext4/inode.c |
| 4 |
* |
| 5 |
* Copyright (C) 1992, 1993, 1994, 1995 |
| 6 |
* Remy Card ([email protected]) |
| 7 |
* Laboratoire MASI - Institut Blaise Pascal |
| 8 |
* Universite Pierre et Marie Curie (Paris VI) |
| 9 |
* |
| 10 |
* from |
| 11 |
* |
| 12 |
* linux/fs/minix/inode.c |
| 13 |
* |
| 14 |
* Copyright (C) 1991, 1992 Linus Torvalds |
| 15 |
* |
| 16 |
* 64-bit file support on 64-bit platforms by Jakub Jelinek |
| 17 |
* ([email protected]) |
| 18 |
* |
| 19 |
* Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000 |
| 20 |
*/ |
| 21 |
|
| 22 |
#include <linux/fs.h> |
| 23 |
#include <linux/mount.h> |
| 24 |
#include <linux/time.h> |
| 25 |
#include <linux/highuid.h> |
| 26 |
#include <linux/pagemap.h> |
| 27 |
#include <linux/dax.h> |
| 28 |
#include <linux/quotaops.h> |
| 29 |
#include <linux/string.h> |
| 30 |
#include <linux/buffer_head.h> |
| 31 |
#include <linux/writeback.h> |
| 32 |
#include <linux/pagevec.h> |
| 33 |
#include <linux/mpage.h> |
| 34 |
#include <linux/rmap.h> |
| 35 |
#include <linux/namei.h> |
| 36 |
#include <linux/uio.h> |
| 37 |
#include <linux/bio.h> |
| 38 |
#include <linux/workqueue.h> |
| 39 |
#include <linux/kernel.h> |
| 40 |
#include <linux/printk.h> |
| 41 |
#include <linux/slab.h> |
| 42 |
#include <linux/bitops.h> |
| 43 |
#include <linux/iomap.h> |
| 44 |
#include <linux/iversion.h> |
| 45 |
|
| 46 |
#include "ext4_jbd2.h" |
| 47 |
#include "xattr.h" |
| 48 |
#include "acl.h" |
| 49 |
#include "truncate.h" |
| 50 |
|
| 51 |
#include <kunit/static_stub.h> |
| 52 |
|
| 53 |
#include <trace/events/ext4.h> |
| 54 |
|
| 55 |
static void ext4_journalled_zero_new_buffers(handle_t *handle, |
| 56 |
struct inode *inode, |
| 57 |
struct folio *folio, |
| 58 |
unsigned from, unsigned to); |
| 59 |
|
| 60 |
static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, |
| 61 |
struct ext4_inode_info *ei) |
| 62 |
{ |
| 63 |
__u32 csum; |
| 64 |
__u16 dummy_csum = 0; |
| 65 |
int offset = offsetof(struct ext4_inode, i_checksum_lo); |
| 66 |
unsigned int csum_size = sizeof(dummy_csum); |
| 67 |
|
| 68 |
csum = ext4_chksum(ei->i_csum_seed, (__u8 *)raw, offset); |
| 69 |
csum = ext4_chksum(csum, (__u8 *)&dummy_csum, csum_size); |
| 70 |
offset += csum_size; |
| 71 |
csum = ext4_chksum(csum, (__u8 *)raw + offset, |
| 72 |
EXT4_GOOD_OLD_INODE_SIZE - offset); |
| 73 |
|
| 74 |
if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
| 75 |
offset = offsetof(struct ext4_inode, i_checksum_hi); |
| 76 |
csum = ext4_chksum(csum, (__u8 *)raw + EXT4_GOOD_OLD_INODE_SIZE, |
| 77 |
offset - EXT4_GOOD_OLD_INODE_SIZE); |
| 78 |
if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { |
| 79 |
csum = ext4_chksum(csum, (__u8 *)&dummy_csum, |
| 80 |
csum_size); |
| 81 |
offset += csum_size; |
| 82 |
} |
| 83 |
csum = ext4_chksum(csum, (__u8 *)raw + offset, |
| 84 |
EXT4_INODE_SIZE(inode->i_sb) - offset); |
| 85 |
} |
| 86 |
|
| 87 |
return csum; |
| 88 |
} |
| 89 |
|
| 90 |
static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, |
| 91 |
struct ext4_inode_info *ei) |
| 92 |
{ |
| 93 |
__u32 provided, calculated; |
| 94 |
|
| 95 |
if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
| 96 |
cpu_to_le32(EXT4_OS_LINUX) || |
| 97 |
!ext4_has_feature_metadata_csum(inode->i_sb)) |
| 98 |
return 1; |
| 99 |
|
| 100 |
provided = le16_to_cpu(raw->i_checksum_lo); |
| 101 |
calculated = ext4_inode_csum(inode, raw, ei); |
| 102 |
if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && |
| 103 |
EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) |
| 104 |
provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; |
| 105 |
else |
| 106 |
calculated &= 0xFFFF; |
| 107 |
|
| 108 |
return provided == calculated; |
| 109 |
} |
| 110 |
|
| 111 |
void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, |
| 112 |
struct ext4_inode_info *ei) |
| 113 |
{ |
| 114 |
__u32 csum; |
| 115 |
|
| 116 |
if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
| 117 |
cpu_to_le32(EXT4_OS_LINUX) || |
| 118 |
!ext4_has_feature_metadata_csum(inode->i_sb)) |
| 119 |
return; |
| 120 |
|
| 121 |
csum = ext4_inode_csum(inode, raw, ei); |
| 122 |
raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF); |
| 123 |
if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && |
| 124 |
EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) |
| 125 |
raw->i_checksum_hi = cpu_to_le16(csum >> 16); |
| 126 |
} |
| 127 |
|
| 128 |
static inline int ext4_begin_ordered_truncate(struct inode *inode, |
| 129 |
loff_t new_size) |
| 130 |
{ |
| 131 |
struct jbd2_inode *jinode = READ_ONCE(EXT4_I(inode)->jinode); |
| 132 |
|
| 133 |
trace_ext4_begin_ordered_truncate(inode, new_size); |
| 134 |
/* |
| 135 |
* If jinode is zero, then we never opened the file for |
| 136 |
* writing, so there's no need to call |
| 137 |
* jbd2_journal_begin_ordered_truncate() since there's no |
| 138 |
* outstanding writes we need to flush. |
| 139 |
*/ |
| 140 |
if (!jinode) |
| 141 |
return 0; |
| 142 |
return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), |
| 143 |
jinode, |
| 144 |
new_size); |
| 145 |
} |
| 146 |
|
| 147 |
/* |
| 148 |
* Test whether an inode is a fast symlink. |
| 149 |
* A fast symlink has its symlink data stored in ext4_inode_info->i_data. |
| 150 |
*/ |
| 151 |
int ext4_inode_is_fast_symlink(struct inode *inode) |
| 152 |
{ |
| 153 |
if (!ext4_has_feature_ea_inode(inode->i_sb)) { |
| 154 |
int ea_blocks = EXT4_I(inode)->i_file_acl ? |
| 155 |
EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0; |
| 156 |
|
| 157 |
if (ext4_has_inline_data(inode)) |
| 158 |
return 0; |
| 159 |
|
| 160 |
return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); |
| 161 |
} |
| 162 |
return S_ISLNK(inode->i_mode) && inode->i_size && |
| 163 |
(inode->i_size < EXT4_N_BLOCKS * 4); |
| 164 |
} |
| 165 |
|
| 166 |
/* |
| 167 |
* Called at the last iput() if i_nlink is zero. |
| 168 |
*/ |
| 169 |
void ext4_evict_inode(struct inode *inode) |
| 170 |
{ |
| 171 |
handle_t *handle; |
| 172 |
int err; |
| 173 |
/* |
| 174 |
* Credits for final inode cleanup and freeing: |
| 175 |
* sb + inode (ext4_orphan_del()), block bitmap, group descriptor |
| 176 |
* (xattr block freeing), bitmap, group descriptor (inode freeing) |
| 177 |
*/ |
| 178 |
int extra_credits = 6; |
| 179 |
struct ext4_xattr_inode_array *ea_inode_array = NULL; |
| 180 |
bool freeze_protected = false; |
| 181 |
|
| 182 |
trace_ext4_evict_inode(inode); |
| 183 |
|
| 184 |
dax_break_layout_final(inode); |
| 185 |
|
| 186 |
if (EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL) |
| 187 |
ext4_evict_ea_inode(inode); |
| 188 |
if (inode->i_nlink) { |
| 189 |
/* |
| 190 |
* If there's dirty page will lead to data loss, user |
| 191 |
* could see stale data. |
| 192 |
*/ |
| 193 |
if (unlikely(!ext4_emergency_state(inode->i_sb) && |
| 194 |
mapping_tagged(&inode->i_data, PAGECACHE_TAG_DIRTY))) |
| 195 |
ext4_warning_inode(inode, "data will be lost"); |
| 196 |
|
| 197 |
truncate_inode_pages_final(&inode->i_data); |
| 198 |
|
| 199 |
goto no_delete; |
| 200 |
} |
| 201 |
|
| 202 |
if (is_bad_inode(inode)) |
| 203 |
goto no_delete; |
| 204 |
dquot_initialize(inode); |
| 205 |
|
| 206 |
if (ext4_should_order_data(inode)) |
| 207 |
ext4_begin_ordered_truncate(inode, 0); |
| 208 |
truncate_inode_pages_final(&inode->i_data); |
| 209 |
|
| 210 |
/* |
| 211 |
* For inodes with journalled data, transaction commit could have |
| 212 |
* dirtied the inode. And for inodes with dioread_nolock, unwritten |
| 213 |
* extents converting worker could merge extents and also have dirtied |
| 214 |
* the inode. Flush worker is ignoring it because of I_FREEING flag but |
| 215 |
* we still need to remove the inode from the writeback lists. |
| 216 |
*/ |
| 217 |
inode_io_list_del(inode); |
| 218 |
|
| 219 |
/* |
| 220 |
* Protect us against freezing - iput() caller didn't have to have any |
| 221 |
* protection against it. When we are in a running transaction though, |
| 222 |
* we are already protected against freezing and we cannot grab further |
| 223 |
* protection due to lock ordering constraints. |
| 224 |
*/ |
| 225 |
if (!ext4_journal_current_handle()) { |
| 226 |
sb_start_intwrite(inode->i_sb); |
| 227 |
freeze_protected = true; |
| 228 |
} |
| 229 |
|
| 230 |
if (!IS_NOQUOTA(inode)) |
| 231 |
extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb); |
| 232 |
|
| 233 |
/* |
| 234 |
* Block bitmap, group descriptor, and inode are accounted in both |
| 235 |
* ext4_blocks_for_truncate() and extra_credits. So subtract 3. |
| 236 |
*/ |
| 237 |
handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, |
| 238 |
ext4_blocks_for_truncate(inode) + extra_credits - 3); |
| 239 |
if (IS_ERR(handle)) { |
| 240 |
ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
| 241 |
/* |
| 242 |
* If we're going to skip the normal cleanup, we still need to |
| 243 |
* make sure that the in-core orphan linked list is properly |
| 244 |
* cleaned up. |
| 245 |
*/ |
| 246 |
ext4_orphan_del(NULL, inode); |
| 247 |
if (freeze_protected) |
| 248 |
sb_end_intwrite(inode->i_sb); |
| 249 |
goto no_delete; |
| 250 |
} |
| 251 |
|
| 252 |
if (IS_SYNC(inode)) |
| 253 |
ext4_handle_sync(handle); |
| 254 |
|
| 255 |
/* |
| 256 |
* Set inode->i_size to 0 before calling ext4_truncate(). We need |
| 257 |
* special handling of symlinks here because i_size is used to |
| 258 |
* determine whether ext4_inode_info->i_data contains symlink data or |
| 259 |
* block mappings. Setting i_size to 0 will remove its fast symlink |
| 260 |
* status. Erase i_data so that it becomes a valid empty block map. |
| 261 |
*/ |
| 262 |
if (ext4_inode_is_fast_symlink(inode)) |
| 263 |
memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data)); |
| 264 |
inode->i_size = 0; |
| 265 |
err = ext4_mark_inode_dirty(handle, inode); |
| 266 |
if (err) { |
| 267 |
ext4_warning(inode->i_sb, |
| 268 |
"couldn't mark inode dirty (err %d)", err); |
| 269 |
goto stop_handle; |
| 270 |
} |
| 271 |
if (inode->i_blocks) { |
| 272 |
err = ext4_truncate(inode); |
| 273 |
if (err) { |
| 274 |
ext4_error_err(inode->i_sb, -err, |
| 275 |
"couldn't truncate inode %lu (err %d)", |
| 276 |
inode->i_ino, err); |
| 277 |
goto stop_handle; |
| 278 |
} |
| 279 |
} |
| 280 |
|
| 281 |
/* Remove xattr references. */ |
| 282 |
err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array, |
| 283 |
extra_credits); |
| 284 |
if (err) { |
| 285 |
ext4_warning(inode->i_sb, "xattr delete (err %d)", err); |
| 286 |
stop_handle: |
| 287 |
ext4_journal_stop(handle); |
| 288 |
ext4_orphan_del(NULL, inode); |
| 289 |
if (freeze_protected) |
| 290 |
sb_end_intwrite(inode->i_sb); |
| 291 |
ext4_xattr_inode_array_free(ea_inode_array); |
| 292 |
goto no_delete; |
| 293 |
} |
| 294 |
|
| 295 |
/* |
| 296 |
* Kill off the orphan record which ext4_truncate created. |
| 297 |
* AKPM: I think this can be inside the above `if'. |
| 298 |
* Note that ext4_orphan_del() has to be able to cope with the |
| 299 |
* deletion of a non-existent orphan - this is because we don't |
| 300 |
* know if ext4_truncate() actually created an orphan record. |
| 301 |
* (Well, we could do this if we need to, but heck - it works) |
| 302 |
*/ |
| 303 |
ext4_orphan_del(handle, inode); |
| 304 |
EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds(); |
| 305 |
|
| 306 |
/* |
| 307 |
* One subtle ordering requirement: if anything has gone wrong |
| 308 |
* (transaction abort, IO errors, whatever), then we can still |
| 309 |
* do these next steps (the fs will already have been marked as |
| 310 |
* having errors), but we can't free the inode if the mark_dirty |
| 311 |
* fails. |
| 312 |
*/ |
| 313 |
if (ext4_mark_inode_dirty(handle, inode)) |
| 314 |
/* If that failed, just do the required in-core inode clear. */ |
| 315 |
ext4_clear_inode(inode); |
| 316 |
else |
| 317 |
ext4_free_inode(handle, inode); |
| 318 |
ext4_journal_stop(handle); |
| 319 |
if (freeze_protected) |
| 320 |
sb_end_intwrite(inode->i_sb); |
| 321 |
ext4_xattr_inode_array_free(ea_inode_array); |
| 322 |
return; |
| 323 |
no_delete: |
| 324 |
/* |
| 325 |
* Check out some where else accidentally dirty the evicting inode, |
| 326 |
* which may probably cause inode use-after-free issues later. |
| 327 |
*/ |
| 328 |
WARN_ON_ONCE(!list_empty_careful(&inode->i_io_list)); |
| 329 |
|
| 330 |
if (!list_empty(&EXT4_I(inode)->i_fc_list)) |
| 331 |
ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM, NULL); |
| 332 |
ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
| 333 |
} |
| 334 |
|
| 335 |
#ifdef CONFIG_QUOTA |
| 336 |
qsize_t *ext4_get_reserved_space(struct inode *inode) |
| 337 |
{ |
| 338 |
return &EXT4_I(inode)->i_reserved_quota; |
| 339 |
} |
| 340 |
#endif |
| 341 |
|
| 342 |
/* |
| 343 |
* Called with i_data_sem down, which is important since we can call |
| 344 |
* ext4_discard_preallocations() from here. |
| 345 |
*/ |
| 346 |
void ext4_da_update_reserve_space(struct inode *inode, |
| 347 |
int used, int quota_claim) |
| 348 |
{ |
| 349 |
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 350 |
struct ext4_inode_info *ei = EXT4_I(inode); |
| 351 |
|
| 352 |
spin_lock(&ei->i_block_reservation_lock); |
| 353 |
trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
| 354 |
if (unlikely(used > ei->i_reserved_data_blocks)) { |
| 355 |
ext4_warning(inode->i_sb, "%s: ino %lu, used %d " |
| 356 |
"with only %d reserved data blocks", |
| 357 |
__func__, inode->i_ino, used, |
| 358 |
ei->i_reserved_data_blocks); |
| 359 |
WARN_ON(1); |
| 360 |
used = ei->i_reserved_data_blocks; |
| 361 |
} |
| 362 |
|
| 363 |
/* Update per-inode reservations */ |
| 364 |
ei->i_reserved_data_blocks -= used; |
| 365 |
percpu_counter_sub(&sbi->s_dirtyclusters_counter, used); |
| 366 |
|
| 367 |
spin_unlock(&ei->i_block_reservation_lock); |
| 368 |
|
| 369 |
/* Update quota subsystem for data blocks */ |
| 370 |
if (quota_claim) |
| 371 |
dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
| 372 |
else { |
| 373 |
/* |
| 374 |
* We did fallocate with an offset that is already delayed |
| 375 |
* allocated. So on delayed allocated writeback we should |
| 376 |
* not re-claim the quota for fallocated blocks. |
| 377 |
*/ |
| 378 |
dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
| 379 |
} |
| 380 |
|
| 381 |
/* |
| 382 |
* If we have done all the pending block allocations and if |
| 383 |
* there aren't any writers on the inode, we can discard the |
| 384 |
* inode's preallocations. |
| 385 |
*/ |
| 386 |
if ((ei->i_reserved_data_blocks == 0) && |
| 387 |
!inode_is_open_for_write(inode)) |
| 388 |
ext4_discard_preallocations(inode); |
| 389 |
} |
| 390 |
|
| 391 |
static int __check_block_validity(struct inode *inode, const char *func, |
| 392 |
unsigned int line, |
| 393 |
struct ext4_map_blocks *map) |
| 394 |
{ |
| 395 |
journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
| 396 |
|
| 397 |
if (journal && inode == journal->j_inode) |
| 398 |
return 0; |
| 399 |
|
| 400 |
if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) { |
| 401 |
ext4_error_inode(inode, func, line, map->m_pblk, |
| 402 |
"lblock %lu mapped to illegal pblock %llu " |
| 403 |
"(length %d)", (unsigned long) map->m_lblk, |
| 404 |
map->m_pblk, map->m_len); |
| 405 |
return -EFSCORRUPTED; |
| 406 |
} |
| 407 |
return 0; |
| 408 |
} |
| 409 |
|
| 410 |
int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk, |
| 411 |
ext4_lblk_t len) |
| 412 |
{ |
| 413 |
int ret; |
| 414 |
|
| 415 |
KUNIT_STATIC_STUB_REDIRECT(ext4_issue_zeroout, inode, lblk, pblk, len); |
| 416 |
|
| 417 |
if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) |
| 418 |
return fscrypt_zeroout_range(inode, lblk, pblk, len); |
| 419 |
|
| 420 |
ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS); |
| 421 |
if (ret > 0) |
| 422 |
ret = 0; |
| 423 |
|
| 424 |
return ret; |
| 425 |
} |
| 426 |
|
| 427 |
/* |
| 428 |
* For generic regular files, when updating the extent tree, Ext4 should |
| 429 |
* hold the i_rwsem and invalidate_lock exclusively. This ensures |
| 430 |
* exclusion against concurrent page faults, as well as reads and writes. |
| 431 |
*/ |
| 432 |
#ifdef CONFIG_EXT4_DEBUG |
| 433 |
void ext4_check_map_extents_env(struct inode *inode) |
| 434 |
{ |
| 435 |
if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) |
| 436 |
return; |
| 437 |
|
| 438 |
if (!S_ISREG(inode->i_mode) || |
| 439 |
IS_NOQUOTA(inode) || IS_VERITY(inode) || |
| 440 |
is_special_ino(inode->i_sb, inode->i_ino) || |
| 441 |
(inode_state_read_once(inode) & (I_FREEING | I_WILL_FREE | I_NEW)) || |
| 442 |
ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE) || |
| 443 |
ext4_verity_in_progress(inode)) |
| 444 |
return; |
| 445 |
|
| 446 |
WARN_ON_ONCE(!inode_is_locked(inode) && |
| 447 |
!rwsem_is_locked(&inode->i_mapping->invalidate_lock)); |
| 448 |
} |
| 449 |
#else |
| 450 |
void ext4_check_map_extents_env(struct inode *inode) {} |
| 451 |
#endif |
| 452 |
|
| 453 |
#define check_block_validity(inode, map) \ |
| 454 |
__check_block_validity((inode), __func__, __LINE__, (map)) |
| 455 |
|
| 456 |
#ifdef ES_AGGRESSIVE_TEST |
| 457 |
static void ext4_map_blocks_es_recheck(handle_t *handle, |
| 458 |
struct inode *inode, |
| 459 |
struct ext4_map_blocks *es_map, |
| 460 |
struct ext4_map_blocks *map, |
| 461 |
int flags) |
| 462 |
{ |
| 463 |
int retval; |
| 464 |
|
| 465 |
map->m_flags = 0; |
| 466 |
/* |
| 467 |
* There is a race window that the result is not the same. |
| 468 |
* e.g. xfstests #223 when dioread_nolock enables. The reason |
| 469 |
* is that we lookup a block mapping in extent status tree with |
| 470 |
* out taking i_data_sem. So at the time the unwritten extent |
| 471 |
* could be converted. |
| 472 |
*/ |
| 473 |
down_read(&EXT4_I(inode)->i_data_sem); |
| 474 |
if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
| 475 |
retval = ext4_ext_map_blocks(handle, inode, map, 0); |
| 476 |
} else { |
| 477 |
retval = ext4_ind_map_blocks(handle, inode, map, 0); |
| 478 |
} |
| 479 |
up_read((&EXT4_I(inode)->i_data_sem)); |
| 480 |
|
| 481 |
/* |
| 482 |
* We don't check m_len because extent will be collpased in status |
| 483 |
* tree. So the m_len might not equal. |
| 484 |
*/ |
| 485 |
if (es_map->m_lblk != map->m_lblk || |
| 486 |
es_map->m_flags != map->m_flags || |
| 487 |
es_map->m_pblk != map->m_pblk) { |
| 488 |
printk("ES cache assertion failed for inode: %lu " |
| 489 |
"es_cached ex [%d/%d/%llu/%x] != " |
| 490 |
"found ex [%d/%d/%llu/%x] retval %d flags %x\n", |
| 491 |
inode->i_ino, es_map->m_lblk, es_map->m_len, |
| 492 |
es_map->m_pblk, es_map->m_flags, map->m_lblk, |
| 493 |
map->m_len, map->m_pblk, map->m_flags, |
| 494 |
retval, flags); |
| 495 |
} |
| 496 |
} |
| 497 |
#endif /* ES_AGGRESSIVE_TEST */ |
| 498 |
|
| 499 |
static int ext4_map_query_blocks_next_in_leaf(handle_t *handle, |
| 500 |
struct inode *inode, struct ext4_map_blocks *map, |
| 501 |
unsigned int orig_mlen) |
| 502 |
{ |
| 503 |
struct ext4_map_blocks map2; |
| 504 |
unsigned int status, status2; |
| 505 |
int retval; |
| 506 |
|
| 507 |
status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
| 508 |
EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; |
| 509 |
|
| 510 |
WARN_ON_ONCE(!(map->m_flags & EXT4_MAP_QUERY_LAST_IN_LEAF)); |
| 511 |
WARN_ON_ONCE(orig_mlen <= map->m_len); |
| 512 |
|
| 513 |
/* Prepare map2 for lookup in next leaf block */ |
| 514 |
map2.m_lblk = map->m_lblk + map->m_len; |
| 515 |
map2.m_len = orig_mlen - map->m_len; |
| 516 |
map2.m_flags = 0; |
| 517 |
retval = ext4_ext_map_blocks(handle, inode, &map2, 0); |
| 518 |
|
| 519 |
if (retval <= 0) { |
| 520 |
ext4_es_cache_extent(inode, map->m_lblk, map->m_len, |
| 521 |
map->m_pblk, status); |
| 522 |
return map->m_len; |
| 523 |
} |
| 524 |
|
| 525 |
if (unlikely(retval != map2.m_len)) { |
| 526 |
ext4_warning(inode->i_sb, |
| 527 |
"ES len assertion failed for inode " |
| 528 |
"%lu: retval %d != map->m_len %d", |
| 529 |
inode->i_ino, retval, map2.m_len); |
| 530 |
WARN_ON(1); |
| 531 |
} |
| 532 |
|
| 533 |
status2 = map2.m_flags & EXT4_MAP_UNWRITTEN ? |
| 534 |
EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; |
| 535 |
|
| 536 |
/* |
| 537 |
* If map2 is contiguous with map, then let's insert it as a single |
| 538 |
* extent in es cache and return the combined length of both the maps. |
| 539 |
*/ |
| 540 |
if (map->m_pblk + map->m_len == map2.m_pblk && |
| 541 |
status == status2) { |
| 542 |
ext4_es_cache_extent(inode, map->m_lblk, |
| 543 |
map->m_len + map2.m_len, map->m_pblk, |
| 544 |
status); |
| 545 |
map->m_len += map2.m_len; |
| 546 |
} else { |
| 547 |
ext4_es_cache_extent(inode, map->m_lblk, map->m_len, |
| 548 |
map->m_pblk, status); |
| 549 |
} |
| 550 |
|
| 551 |
return map->m_len; |
| 552 |
} |
| 553 |
|
| 554 |
int ext4_map_query_blocks(handle_t *handle, struct inode *inode, |
| 555 |
struct ext4_map_blocks *map, int flags) |
| 556 |
{ |
| 557 |
unsigned int status; |
| 558 |
int retval; |
| 559 |
unsigned int orig_mlen = map->m_len; |
| 560 |
|
| 561 |
flags &= EXT4_EX_QUERY_FILTER; |
| 562 |
if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 563 |
retval = ext4_ext_map_blocks(handle, inode, map, flags); |
| 564 |
else |
| 565 |
retval = ext4_ind_map_blocks(handle, inode, map, flags); |
| 566 |
if (retval < 0) |
| 567 |
return retval; |
| 568 |
|
| 569 |
/* A hole? */ |
| 570 |
if (retval == 0) |
| 571 |
goto out; |
| 572 |
|
| 573 |
if (unlikely(retval != map->m_len)) { |
| 574 |
ext4_warning(inode->i_sb, |
| 575 |
"ES len assertion failed for inode " |
| 576 |
"%lu: retval %d != map->m_len %d", |
| 577 |
inode->i_ino, retval, map->m_len); |
| 578 |
WARN_ON(1); |
| 579 |
} |
| 580 |
|
| 581 |
/* |
| 582 |
* No need to query next in leaf: |
| 583 |
* - if returned extent is not last in leaf or |
| 584 |
* - if the last in leaf is the full requested range |
| 585 |
*/ |
| 586 |
if (!(map->m_flags & EXT4_MAP_QUERY_LAST_IN_LEAF) || |
| 587 |
map->m_len == orig_mlen) { |
| 588 |
status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
| 589 |
EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; |
| 590 |
ext4_es_cache_extent(inode, map->m_lblk, map->m_len, |
| 591 |
map->m_pblk, status); |
| 592 |
} else { |
| 593 |
retval = ext4_map_query_blocks_next_in_leaf(handle, inode, map, |
| 594 |
orig_mlen); |
| 595 |
} |
| 596 |
out: |
| 597 |
map->m_seq = READ_ONCE(EXT4_I(inode)->i_es_seq); |
| 598 |
return retval; |
| 599 |
} |
| 600 |
|
| 601 |
int ext4_map_create_blocks(handle_t *handle, struct inode *inode, |
| 602 |
struct ext4_map_blocks *map, int flags) |
| 603 |
{ |
| 604 |
unsigned int status; |
| 605 |
int err, retval = 0; |
| 606 |
|
| 607 |
/* |
| 608 |
* We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE |
| 609 |
* indicates that the blocks and quotas has already been |
| 610 |
* checked when the data was copied into the page cache. |
| 611 |
*/ |
| 612 |
if (map->m_flags & EXT4_MAP_DELAYED) |
| 613 |
flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; |
| 614 |
|
| 615 |
/* |
| 616 |
* Here we clear m_flags because after allocating an new extent, |
| 617 |
* it will be set again. |
| 618 |
*/ |
| 619 |
map->m_flags &= ~EXT4_MAP_FLAGS; |
| 620 |
|
| 621 |
/* |
| 622 |
* We need to check for EXT4 here because migrate could have |
| 623 |
* changed the inode type in between. |
| 624 |
*/ |
| 625 |
if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
| 626 |
retval = ext4_ext_map_blocks(handle, inode, map, flags); |
| 627 |
} else { |
| 628 |
retval = ext4_ind_map_blocks(handle, inode, map, flags); |
| 629 |
|
| 630 |
/* |
| 631 |
* We allocated new blocks which will result in i_data's |
| 632 |
* format changing. Force the migrate to fail by clearing |
| 633 |
* migrate flags. |
| 634 |
*/ |
| 635 |
if (retval > 0 && map->m_flags & EXT4_MAP_NEW) |
| 636 |
ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
| 637 |
} |
| 638 |
if (retval <= 0) |
| 639 |
return retval; |
| 640 |
|
| 641 |
if (unlikely(retval != map->m_len)) { |
| 642 |
ext4_warning(inode->i_sb, |
| 643 |
"ES len assertion failed for inode %lu: " |
| 644 |
"retval %d != map->m_len %d", |
| 645 |
inode->i_ino, retval, map->m_len); |
| 646 |
WARN_ON(1); |
| 647 |
} |
| 648 |
|
| 649 |
/* |
| 650 |
* We have to zeroout blocks before inserting them into extent |
| 651 |
* status tree. Otherwise someone could look them up there and |
| 652 |
* use them before they are really zeroed. We also have to |
| 653 |
* unmap metadata before zeroing as otherwise writeback can |
| 654 |
* overwrite zeros with stale data from block device. |
| 655 |
*/ |
| 656 |
if (flags & EXT4_GET_BLOCKS_ZERO && |
| 657 |
map->m_flags & EXT4_MAP_MAPPED && map->m_flags & EXT4_MAP_NEW) { |
| 658 |
err = ext4_issue_zeroout(inode, map->m_lblk, map->m_pblk, |
| 659 |
map->m_len); |
| 660 |
if (err) |
| 661 |
return err; |
| 662 |
} |
| 663 |
|
| 664 |
status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
| 665 |
EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; |
| 666 |
ext4_es_insert_extent(inode, map->m_lblk, map->m_len, map->m_pblk, |
| 667 |
status, flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE); |
| 668 |
map->m_seq = READ_ONCE(EXT4_I(inode)->i_es_seq); |
| 669 |
|
| 670 |
return retval; |
| 671 |
} |
| 672 |
|
| 673 |
/* |
| 674 |
* The ext4_map_blocks() function tries to look up the requested blocks, |
| 675 |
* and returns if the blocks are already mapped. |
| 676 |
* |
| 677 |
* Otherwise it takes the write lock of the i_data_sem and allocate blocks |
| 678 |
* and store the allocated blocks in the result buffer head and mark it |
| 679 |
* mapped. |
| 680 |
* |
| 681 |
* If file type is extents based, it will call ext4_ext_map_blocks(), |
| 682 |
* Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping |
| 683 |
* based files |
| 684 |
* |
| 685 |
* On success, it returns the number of blocks being mapped or allocated. |
| 686 |
* If flags doesn't contain EXT4_GET_BLOCKS_CREATE the blocks are |
| 687 |
* pre-allocated and unwritten, the resulting @map is marked as unwritten. |
| 688 |
* If the flags contain EXT4_GET_BLOCKS_CREATE, it will mark @map as mapped. |
| 689 |
* |
| 690 |
* It returns 0 if plain look up failed (blocks have not been allocated), in |
| 691 |
* that case, @map is returned as unmapped but we still do fill map->m_len to |
| 692 |
* indicate the length of a hole starting at map->m_lblk. |
| 693 |
* |
| 694 |
* It returns the error in case of allocation failure. |
| 695 |
*/ |
| 696 |
int ext4_map_blocks(handle_t *handle, struct inode *inode, |
| 697 |
struct ext4_map_blocks *map, int flags) |
| 698 |
{ |
| 699 |
struct extent_status es; |
| 700 |
int retval; |
| 701 |
int ret = 0; |
| 702 |
unsigned int orig_mlen = map->m_len; |
| 703 |
#ifdef ES_AGGRESSIVE_TEST |
| 704 |
struct ext4_map_blocks orig_map; |
| 705 |
|
| 706 |
memcpy(&orig_map, map, sizeof(*map)); |
| 707 |
#endif |
| 708 |
|
| 709 |
map->m_flags = 0; |
| 710 |
ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n", |
| 711 |
flags, map->m_len, (unsigned long) map->m_lblk); |
| 712 |
|
| 713 |
/* |
| 714 |
* ext4_map_blocks returns an int, and m_len is an unsigned int |
| 715 |
*/ |
| 716 |
if (unlikely(map->m_len > INT_MAX)) |
| 717 |
map->m_len = INT_MAX; |
| 718 |
|
| 719 |
/* We can handle the block number less than EXT_MAX_BLOCKS */ |
| 720 |
if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS)) |
| 721 |
return -EFSCORRUPTED; |
| 722 |
|
| 723 |
/* |
| 724 |
* Callers from the context of data submission are the only exceptions |
| 725 |
* for regular files that do not hold the i_rwsem or invalidate_lock. |
| 726 |
* However, caching unrelated ranges is not permitted. |
| 727 |
*/ |
| 728 |
if (flags & EXT4_GET_BLOCKS_IO_SUBMIT) |
| 729 |
WARN_ON_ONCE(!(flags & EXT4_EX_NOCACHE)); |
| 730 |
else |
| 731 |
ext4_check_map_extents_env(inode); |
| 732 |
|
| 733 |
/* Lookup extent status tree firstly */ |
| 734 |
if (ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es, &map->m_seq)) { |
| 735 |
if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { |
| 736 |
map->m_pblk = ext4_es_pblock(&es) + |
| 737 |
map->m_lblk - es.es_lblk; |
| 738 |
map->m_flags |= ext4_es_is_written(&es) ? |
| 739 |
EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; |
| 740 |
retval = es.es_len - (map->m_lblk - es.es_lblk); |
| 741 |
if (retval > map->m_len) |
| 742 |
retval = map->m_len; |
| 743 |
map->m_len = retval; |
| 744 |
} else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { |
| 745 |
map->m_pblk = 0; |
| 746 |
map->m_flags |= ext4_es_is_delayed(&es) ? |
| 747 |
EXT4_MAP_DELAYED : 0; |
| 748 |
retval = es.es_len - (map->m_lblk - es.es_lblk); |
| 749 |
if (retval > map->m_len) |
| 750 |
retval = map->m_len; |
| 751 |
map->m_len = retval; |
| 752 |
retval = 0; |
| 753 |
} else { |
| 754 |
BUG(); |
| 755 |
} |
| 756 |
|
| 757 |
if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT) |
| 758 |
return retval; |
| 759 |
#ifdef ES_AGGRESSIVE_TEST |
| 760 |
ext4_map_blocks_es_recheck(handle, inode, map, |
| 761 |
&orig_map, flags); |
| 762 |
#endif |
| 763 |
if (!(flags & EXT4_GET_BLOCKS_QUERY_LAST_IN_LEAF) || |
| 764 |
orig_mlen == map->m_len) |
| 765 |
goto found; |
| 766 |
|
| 767 |
map->m_len = orig_mlen; |
| 768 |
} |
| 769 |
/* |
| 770 |
* In the query cache no-wait mode, nothing we can do more if we |
| 771 |
* cannot find extent in the cache. |
| 772 |
*/ |
| 773 |
if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT) |
| 774 |
return 0; |
| 775 |
|
| 776 |
/* |
| 777 |
* Try to see if we can get the block without requesting a new |
| 778 |
* file system block. |
| 779 |
*/ |
| 780 |
down_read(&EXT4_I(inode)->i_data_sem); |
| 781 |
retval = ext4_map_query_blocks(handle, inode, map, flags); |
| 782 |
up_read((&EXT4_I(inode)->i_data_sem)); |
| 783 |
|
| 784 |
found: |
| 785 |
if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
| 786 |
ret = check_block_validity(inode, map); |
| 787 |
if (ret != 0) |
| 788 |
return ret; |
| 789 |
} |
| 790 |
|
| 791 |
/* If it is only a block(s) look up */ |
| 792 |
if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
| 793 |
return retval; |
| 794 |
|
| 795 |
/* |
| 796 |
* Returns if the blocks have already allocated |
| 797 |
* |
| 798 |
* Note that if blocks have been preallocated |
| 799 |
* ext4_ext_map_blocks() returns with buffer head unmapped |
| 800 |
*/ |
| 801 |
if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
| 802 |
/* |
| 803 |
* If we need to convert extent to unwritten |
| 804 |
* we continue and do the actual work in |
| 805 |
* ext4_ext_map_blocks() |
| 806 |
*/ |
| 807 |
if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) |
| 808 |
return retval; |
| 809 |
|
| 810 |
|
| 811 |
ext4_fc_track_inode(handle, inode); |
| 812 |
/* |
| 813 |
* New blocks allocate and/or writing to unwritten extent |
| 814 |
* will possibly result in updating i_data, so we take |
| 815 |
* the write lock of i_data_sem, and call get_block() |
| 816 |
* with create == 1 flag. |
| 817 |
*/ |
| 818 |
down_write(&EXT4_I(inode)->i_data_sem); |
| 819 |
retval = ext4_map_create_blocks(handle, inode, map, flags); |
| 820 |
up_write((&EXT4_I(inode)->i_data_sem)); |
| 821 |
|
| 822 |
if (retval < 0) |
| 823 |
ext_debug(inode, "failed with err %d\n", retval); |
| 824 |
if (retval <= 0) |
| 825 |
return retval; |
| 826 |
|
| 827 |
if (map->m_flags & EXT4_MAP_MAPPED) { |
| 828 |
ret = check_block_validity(inode, map); |
| 829 |
if (ret != 0) |
| 830 |
return ret; |
| 831 |
|
| 832 |
/* |
| 833 |
* Inodes with freshly allocated blocks where contents will be |
| 834 |
* visible after transaction commit must be on transaction's |
| 835 |
* ordered data list. |
| 836 |
*/ |
| 837 |
if (map->m_flags & EXT4_MAP_NEW && |
| 838 |
!(map->m_flags & EXT4_MAP_UNWRITTEN) && |
| 839 |
!(flags & EXT4_GET_BLOCKS_ZERO) && |
| 840 |
!ext4_is_quota_file(inode) && |
| 841 |
ext4_should_order_data(inode)) { |
| 842 |
loff_t start_byte = EXT4_LBLK_TO_B(inode, map->m_lblk); |
| 843 |
loff_t length = EXT4_LBLK_TO_B(inode, map->m_len); |
| 844 |
|
| 845 |
if (flags & EXT4_GET_BLOCKS_IO_SUBMIT) |
| 846 |
ret = ext4_jbd2_inode_add_wait(handle, inode, |
| 847 |
start_byte, length); |
| 848 |
else |
| 849 |
ret = ext4_jbd2_inode_add_write(handle, inode, |
| 850 |
start_byte, length); |
| 851 |
if (ret) |
| 852 |
return ret; |
| 853 |
} |
| 854 |
} |
| 855 |
ext4_fc_track_range(handle, inode, map->m_lblk, map->m_lblk + |
| 856 |
map->m_len - 1); |
| 857 |
return retval; |
| 858 |
} |
| 859 |
|
| 860 |
/* |
| 861 |
* Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages |
| 862 |
* we have to be careful as someone else may be manipulating b_state as well. |
| 863 |
*/ |
| 864 |
static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags) |
| 865 |
{ |
| 866 |
unsigned long old_state; |
| 867 |
unsigned long new_state; |
| 868 |
|
| 869 |
flags &= EXT4_MAP_FLAGS; |
| 870 |
|
| 871 |
/* Dummy buffer_head? Set non-atomically. */ |
| 872 |
if (!bh->b_folio) { |
| 873 |
bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags; |
| 874 |
return; |
| 875 |
} |
| 876 |
/* |
| 877 |
* Someone else may be modifying b_state. Be careful! This is ugly but |
| 878 |
* once we get rid of using bh as a container for mapping information |
| 879 |
* to pass to / from get_block functions, this can go away. |
| 880 |
*/ |
| 881 |
old_state = READ_ONCE(bh->b_state); |
| 882 |
do { |
| 883 |
new_state = (old_state & ~EXT4_MAP_FLAGS) | flags; |
| 884 |
} while (unlikely(!try_cmpxchg(&bh->b_state, &old_state, new_state))); |
| 885 |
} |
| 886 |
|
| 887 |
/* |
| 888 |
* Make sure that the current journal transaction has enough credits to map |
| 889 |
* one extent. Return -EAGAIN if it cannot extend the current running |
| 890 |
* transaction. |
| 891 |
*/ |
| 892 |
static inline int ext4_journal_ensure_extent_credits(handle_t *handle, |
| 893 |
struct inode *inode) |
| 894 |
{ |
| 895 |
int credits; |
| 896 |
int ret; |
| 897 |
|
| 898 |
/* Called from ext4_da_write_begin() which has no handle started? */ |
| 899 |
if (!handle) |
| 900 |
return 0; |
| 901 |
|
| 902 |
credits = ext4_chunk_trans_blocks(inode, 1); |
| 903 |
ret = __ext4_journal_ensure_credits(handle, credits, credits, 0); |
| 904 |
return ret <= 0 ? ret : -EAGAIN; |
| 905 |
} |
| 906 |
|
| 907 |
static int _ext4_get_block(struct inode *inode, sector_t iblock, |
| 908 |
struct buffer_head *bh, int flags) |
| 909 |
{ |
| 910 |
struct ext4_map_blocks map; |
| 911 |
int ret = 0; |
| 912 |
|
| 913 |
if (ext4_has_inline_data(inode)) |
| 914 |
return -ERANGE; |
| 915 |
|
| 916 |
map.m_lblk = iblock; |
| 917 |
map.m_len = bh->b_size >> inode->i_blkbits; |
| 918 |
|
| 919 |
ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map, |
| 920 |
flags); |
| 921 |
if (ret > 0) { |
| 922 |
map_bh(bh, inode->i_sb, map.m_pblk); |
| 923 |
ext4_update_bh_state(bh, map.m_flags); |
| 924 |
bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
| 925 |
ret = 0; |
| 926 |
} else if (ret == 0) { |
| 927 |
/* hole case, need to fill in bh->b_size */ |
| 928 |
bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
| 929 |
} |
| 930 |
return ret; |
| 931 |
} |
| 932 |
|
| 933 |
int ext4_get_block(struct inode *inode, sector_t iblock, |
| 934 |
struct buffer_head *bh, int create) |
| 935 |
{ |
| 936 |
return _ext4_get_block(inode, iblock, bh, |
| 937 |
create ? EXT4_GET_BLOCKS_CREATE : 0); |
| 938 |
} |
| 939 |
|
| 940 |
/* |
| 941 |
* Get block function used when preparing for buffered write if we require |
| 942 |
* creating an unwritten extent if blocks haven't been allocated. The extent |
| 943 |
* will be converted to written after the IO is complete. |
| 944 |
*/ |
| 945 |
int ext4_get_block_unwritten(struct inode *inode, sector_t iblock, |
| 946 |
struct buffer_head *bh_result, int create) |
| 947 |
{ |
| 948 |
int ret = 0; |
| 949 |
|
| 950 |
ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n", |
| 951 |
inode->i_ino, create); |
| 952 |
ret = _ext4_get_block(inode, iblock, bh_result, |
| 953 |
EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT); |
| 954 |
|
| 955 |
/* |
| 956 |
* If the buffer is marked unwritten, mark it as new to make sure it is |
| 957 |
* zeroed out correctly in case of partial writes. Otherwise, there is |
| 958 |
* a chance of stale data getting exposed. |
| 959 |
*/ |
| 960 |
if (ret == 0 && buffer_unwritten(bh_result)) |
| 961 |
set_buffer_new(bh_result); |
| 962 |
|
| 963 |
return ret; |
| 964 |
} |
| 965 |
|
| 966 |
/* Maximum number of blocks we map for direct IO at once. */ |
| 967 |
#define DIO_MAX_BLOCKS 4096 |
| 968 |
|
| 969 |
/* |
| 970 |
* `handle' can be NULL if create is zero |
| 971 |
*/ |
| 972 |
struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
| 973 |
ext4_lblk_t block, int map_flags) |
| 974 |
{ |
| 975 |
struct ext4_map_blocks map; |
| 976 |
struct buffer_head *bh; |
| 977 |
int create = map_flags & EXT4_GET_BLOCKS_CREATE; |
| 978 |
bool nowait = map_flags & EXT4_GET_BLOCKS_CACHED_NOWAIT; |
| 979 |
int err; |
| 980 |
|
| 981 |
ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) |
| 982 |
|| handle != NULL || create == 0); |
| 983 |
ASSERT(create == 0 || !nowait); |
| 984 |
|
| 985 |
map.m_lblk = block; |
| 986 |
map.m_len = 1; |
| 987 |
err = ext4_map_blocks(handle, inode, &map, map_flags); |
| 988 |
|
| 989 |
if (err == 0) |
| 990 |
return create ? ERR_PTR(-ENOSPC) : NULL; |
| 991 |
if (err < 0) |
| 992 |
return ERR_PTR(err); |
| 993 |
|
| 994 |
if (nowait) |
| 995 |
return sb_find_get_block(inode->i_sb, map.m_pblk); |
| 996 |
|
| 997 |
/* |
| 998 |
* Since bh could introduce extra ref count such as referred by |
| 999 |
* journal_head etc. Try to avoid using __GFP_MOVABLE here |
| 1000 |
* as it may fail the migration when journal_head remains. |
| 1001 |
*/ |
| 1002 |
bh = getblk_unmovable(inode->i_sb->s_bdev, map.m_pblk, |
| 1003 |
inode->i_sb->s_blocksize); |
| 1004 |
|
| 1005 |
if (unlikely(!bh)) |
| 1006 |
return ERR_PTR(-ENOMEM); |
| 1007 |
if (map.m_flags & EXT4_MAP_NEW) { |
| 1008 |
ASSERT(create != 0); |
| 1009 |
ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) |
| 1010 |
|| (handle != NULL)); |
| 1011 |
|
| 1012 |
/* |
| 1013 |
* Now that we do not always journal data, we should |
| 1014 |
* keep in mind whether this should always journal the |
| 1015 |
* new buffer as metadata. For now, regular file |
| 1016 |
* writes use ext4_get_block instead, so it's not a |
| 1017 |
* problem. |
| 1018 |
*/ |
| 1019 |
lock_buffer(bh); |
| 1020 |
BUFFER_TRACE(bh, "call get_create_access"); |
| 1021 |
err = ext4_journal_get_create_access(handle, inode->i_sb, bh, |
| 1022 |
EXT4_JTR_NONE); |
| 1023 |
if (unlikely(err)) { |
| 1024 |
unlock_buffer(bh); |
| 1025 |
goto errout; |
| 1026 |
} |
| 1027 |
if (!buffer_uptodate(bh)) { |
| 1028 |
memset(bh->b_data, 0, inode->i_sb->s_blocksize); |
| 1029 |
set_buffer_uptodate(bh); |
| 1030 |
} |
| 1031 |
unlock_buffer(bh); |
| 1032 |
BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
| 1033 |
err = ext4_handle_dirty_metadata(handle, inode, bh); |
| 1034 |
if (unlikely(err)) |
| 1035 |
goto errout; |
| 1036 |
} else |
| 1037 |
BUFFER_TRACE(bh, "not a new buffer"); |
| 1038 |
return bh; |
| 1039 |
errout: |
| 1040 |
brelse(bh); |
| 1041 |
return ERR_PTR(err); |
| 1042 |
} |
| 1043 |
|
| 1044 |
struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
| 1045 |
ext4_lblk_t block, int map_flags) |
| 1046 |
{ |
| 1047 |
struct buffer_head *bh; |
| 1048 |
int ret; |
| 1049 |
|
| 1050 |
bh = ext4_getblk(handle, inode, block, map_flags); |
| 1051 |
if (IS_ERR(bh)) |
| 1052 |
return bh; |
| 1053 |
if (!bh || ext4_buffer_uptodate(bh)) |
| 1054 |
return bh; |
| 1055 |
|
| 1056 |
ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true); |
| 1057 |
if (ret) { |
| 1058 |
put_bh(bh); |
| 1059 |
return ERR_PTR(ret); |
| 1060 |
} |
| 1061 |
return bh; |
| 1062 |
} |
| 1063 |
|
| 1064 |
/* Read a contiguous batch of blocks. */ |
| 1065 |
int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count, |
| 1066 |
bool wait, struct buffer_head **bhs) |
| 1067 |
{ |
| 1068 |
int i, err; |
| 1069 |
|
| 1070 |
for (i = 0; i < bh_count; i++) { |
| 1071 |
bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */); |
| 1072 |
if (IS_ERR(bhs[i])) { |
| 1073 |
err = PTR_ERR(bhs[i]); |
| 1074 |
bh_count = i; |
| 1075 |
goto out_brelse; |
| 1076 |
} |
| 1077 |
} |
| 1078 |
|
| 1079 |
for (i = 0; i < bh_count; i++) |
| 1080 |
/* Note that NULL bhs[i] is valid because of holes. */ |
| 1081 |
if (bhs[i] && !ext4_buffer_uptodate(bhs[i])) |
| 1082 |
ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false); |
| 1083 |
|
| 1084 |
if (!wait) |
| 1085 |
return 0; |
| 1086 |
|
| 1087 |
for (i = 0; i < bh_count; i++) |
| 1088 |
if (bhs[i]) |
| 1089 |
wait_on_buffer(bhs[i]); |
| 1090 |
|
| 1091 |
for (i = 0; i < bh_count; i++) { |
| 1092 |
if (bhs[i] && !buffer_uptodate(bhs[i])) { |
| 1093 |
err = -EIO; |
| 1094 |
goto out_brelse; |
| 1095 |
} |
| 1096 |
} |
| 1097 |
return 0; |
| 1098 |
|
| 1099 |
out_brelse: |
| 1100 |
for (i = 0; i < bh_count; i++) { |
| 1101 |
brelse(bhs[i]); |
| 1102 |
bhs[i] = NULL; |
| 1103 |
} |
| 1104 |
return err; |
| 1105 |
} |
| 1106 |
|
| 1107 |
int ext4_walk_page_buffers(handle_t *handle, struct inode *inode, |
| 1108 |
struct buffer_head *head, |
| 1109 |
unsigned from, |
| 1110 |
unsigned to, |
| 1111 |
int *partial, |
| 1112 |
int (*fn)(handle_t *handle, struct inode *inode, |
| 1113 |
struct buffer_head *bh)) |
| 1114 |
{ |
| 1115 |
struct buffer_head *bh; |
| 1116 |
unsigned block_start, block_end; |
| 1117 |
unsigned blocksize = head->b_size; |
| 1118 |
int err, ret = 0; |
| 1119 |
struct buffer_head *next; |
| 1120 |
|
| 1121 |
for (bh = head, block_start = 0; |
| 1122 |
ret == 0 && (bh != head || !block_start); |
| 1123 |
block_start = block_end, bh = next) { |
| 1124 |
next = bh->b_this_page; |
| 1125 |
block_end = block_start + blocksize; |
| 1126 |
if (block_end <= from || block_start >= to) { |
| 1127 |
if (partial && !buffer_uptodate(bh)) |
| 1128 |
*partial = 1; |
| 1129 |
continue; |
| 1130 |
} |
| 1131 |
err = (*fn)(handle, inode, bh); |
| 1132 |
if (!ret) |
| 1133 |
ret = err; |
| 1134 |
} |
| 1135 |
return ret; |
| 1136 |
} |
| 1137 |
|
| 1138 |
/* |
| 1139 |
* Helper for handling dirtying of journalled data. We also mark the folio as |
| 1140 |
* dirty so that writeback code knows about this page (and inode) contains |
| 1141 |
* dirty data. ext4_writepages() then commits appropriate transaction to |
| 1142 |
* make data stable. |
| 1143 |
*/ |
| 1144 |
static int ext4_dirty_journalled_data(handle_t *handle, struct buffer_head *bh) |
| 1145 |
{ |
| 1146 |
struct folio *folio = bh->b_folio; |
| 1147 |
struct inode *inode = folio->mapping->host; |
| 1148 |
|
| 1149 |
/* only regular files have a_ops */ |
| 1150 |
if (S_ISREG(inode->i_mode)) |
| 1151 |
folio_mark_dirty(folio); |
| 1152 |
return ext4_handle_dirty_metadata(handle, NULL, bh); |
| 1153 |
} |
| 1154 |
|
| 1155 |
int do_journal_get_write_access(handle_t *handle, struct inode *inode, |
| 1156 |
struct buffer_head *bh) |
| 1157 |
{ |
| 1158 |
if (!buffer_mapped(bh) || buffer_freed(bh)) |
| 1159 |
return 0; |
| 1160 |
BUFFER_TRACE(bh, "get write access"); |
| 1161 |
return ext4_journal_get_write_access(handle, inode->i_sb, bh, |
| 1162 |
EXT4_JTR_NONE); |
| 1163 |
} |
| 1164 |
|
| 1165 |
int ext4_block_write_begin(handle_t *handle, struct folio *folio, |
| 1166 |
loff_t pos, unsigned len, |
| 1167 |
get_block_t *get_block) |
| 1168 |
{ |
| 1169 |
unsigned int from = offset_in_folio(folio, pos); |
| 1170 |
unsigned to = from + len; |
| 1171 |
struct inode *inode = folio->mapping->host; |
| 1172 |
unsigned block_start, block_end; |
| 1173 |
sector_t block; |
| 1174 |
int err = 0; |
| 1175 |
unsigned int blocksize = i_blocksize(inode); |
| 1176 |
struct buffer_head *bh, *head, *wait[2]; |
| 1177 |
int nr_wait = 0; |
| 1178 |
int i; |
| 1179 |
bool should_journal_data = ext4_should_journal_data(inode); |
| 1180 |
|
| 1181 |
BUG_ON(!folio_test_locked(folio)); |
| 1182 |
BUG_ON(to > folio_size(folio)); |
| 1183 |
BUG_ON(from > to); |
| 1184 |
WARN_ON_ONCE(blocksize > folio_size(folio)); |
| 1185 |
|
| 1186 |
head = folio_buffers(folio); |
| 1187 |
if (!head) |
| 1188 |
head = create_empty_buffers(folio, blocksize, 0); |
| 1189 |
block = EXT4_PG_TO_LBLK(inode, folio->index); |
| 1190 |
|
| 1191 |
for (bh = head, block_start = 0; bh != head || !block_start; |
| 1192 |
block++, block_start = block_end, bh = bh->b_this_page) { |
| 1193 |
block_end = block_start + blocksize; |
| 1194 |
if (block_end <= from || block_start >= to) { |
| 1195 |
if (folio_test_uptodate(folio)) { |
| 1196 |
set_buffer_uptodate(bh); |
| 1197 |
} |
| 1198 |
continue; |
| 1199 |
} |
| 1200 |
if (WARN_ON_ONCE(buffer_new(bh))) |
| 1201 |
clear_buffer_new(bh); |
| 1202 |
if (!buffer_mapped(bh)) { |
| 1203 |
WARN_ON(bh->b_size != blocksize); |
| 1204 |
err = ext4_journal_ensure_extent_credits(handle, inode); |
| 1205 |
if (!err) |
| 1206 |
err = get_block(inode, block, bh, 1); |
| 1207 |
if (err) |
| 1208 |
break; |
| 1209 |
if (buffer_new(bh)) { |
| 1210 |
/* |
| 1211 |
* We may be zeroing partial buffers or all new |
| 1212 |
* buffers in case of failure. Prepare JBD2 for |
| 1213 |
* that. |
| 1214 |
*/ |
| 1215 |
if (should_journal_data) |
| 1216 |
do_journal_get_write_access(handle, |
| 1217 |
inode, bh); |
| 1218 |
if (folio_test_uptodate(folio)) { |
| 1219 |
/* |
| 1220 |
* Unlike __block_write_begin() we leave |
| 1221 |
* dirtying of new uptodate buffers to |
| 1222 |
* ->write_end() time or |
| 1223 |
* folio_zero_new_buffers(). |
| 1224 |
*/ |
| 1225 |
set_buffer_uptodate(bh); |
| 1226 |
continue; |
| 1227 |
} |
| 1228 |
if (block_end > to || block_start < from) |
| 1229 |
folio_zero_segments(folio, to, |
| 1230 |
block_end, |
| 1231 |
block_start, from); |
| 1232 |
continue; |
| 1233 |
} |
| 1234 |
} |
| 1235 |
if (folio_test_uptodate(folio)) { |
| 1236 |
set_buffer_uptodate(bh); |
| 1237 |
continue; |
| 1238 |
} |
| 1239 |
if (!buffer_uptodate(bh) && !buffer_delay(bh) && |
| 1240 |
!buffer_unwritten(bh) && |
| 1241 |
(block_start < from || block_end > to)) { |
| 1242 |
ext4_read_bh_lock(bh, 0, false); |
| 1243 |
wait[nr_wait++] = bh; |
| 1244 |
} |
| 1245 |
} |
| 1246 |
/* |
| 1247 |
* If we issued read requests, let them complete. |
| 1248 |
*/ |
| 1249 |
for (i = 0; i < nr_wait; i++) { |
| 1250 |
wait_on_buffer(wait[i]); |
| 1251 |
if (!buffer_uptodate(wait[i])) |
| 1252 |
err = -EIO; |
| 1253 |
} |
| 1254 |
if (unlikely(err)) { |
| 1255 |
if (should_journal_data) |
| 1256 |
ext4_journalled_zero_new_buffers(handle, inode, folio, |
| 1257 |
from, to); |
| 1258 |
else |
| 1259 |
folio_zero_new_buffers(folio, from, to); |
| 1260 |
} else if (fscrypt_inode_uses_fs_layer_crypto(inode)) { |
| 1261 |
for (i = 0; i < nr_wait; i++) { |
| 1262 |
int err2; |
| 1263 |
|
| 1264 |
err2 = fscrypt_decrypt_pagecache_blocks(folio, |
| 1265 |
blocksize, bh_offset(wait[i])); |
| 1266 |
if (err2) { |
| 1267 |
clear_buffer_uptodate(wait[i]); |
| 1268 |
err = err2; |
| 1269 |
} |
| 1270 |
} |
| 1271 |
} |
| 1272 |
|
| 1273 |
return err; |
| 1274 |
} |
| 1275 |
|
| 1276 |
/* |
| 1277 |
* To preserve ordering, it is essential that the hole instantiation and |
| 1278 |
* the data write be encapsulated in a single transaction. We cannot |
| 1279 |
* close off a transaction and start a new one between the ext4_get_block() |
| 1280 |
* and the ext4_write_end(). So doing the jbd2_journal_start at the start of |
| 1281 |
* ext4_write_begin() is the right place. |
| 1282 |
*/ |
| 1283 |
static int ext4_write_begin(const struct kiocb *iocb, |
| 1284 |
struct address_space *mapping, |
| 1285 |
loff_t pos, unsigned len, |
| 1286 |
struct folio **foliop, void **fsdata) |
| 1287 |
{ |
| 1288 |
struct inode *inode = mapping->host; |
| 1289 |
int ret, needed_blocks; |
| 1290 |
handle_t *handle; |
| 1291 |
int retries = 0; |
| 1292 |
struct folio *folio; |
| 1293 |
pgoff_t index; |
| 1294 |
unsigned from, to; |
| 1295 |
|
| 1296 |
ret = ext4_emergency_state(inode->i_sb); |
| 1297 |
if (unlikely(ret)) |
| 1298 |
return ret; |
| 1299 |
|
| 1300 |
trace_ext4_write_begin(inode, pos, len); |
| 1301 |
/* |
| 1302 |
* Reserve one block more for addition to orphan list in case |
| 1303 |
* we allocate blocks but write fails for some reason |
| 1304 |
*/ |
| 1305 |
needed_blocks = ext4_chunk_trans_extent(inode, |
| 1306 |
ext4_journal_blocks_per_folio(inode)) + 1; |
| 1307 |
index = pos >> PAGE_SHIFT; |
| 1308 |
|
| 1309 |
if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
| 1310 |
ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, |
| 1311 |
foliop); |
| 1312 |
if (ret < 0) |
| 1313 |
return ret; |
| 1314 |
if (ret == 1) |
| 1315 |
return 0; |
| 1316 |
} |
| 1317 |
|
| 1318 |
/* |
| 1319 |
* write_begin_get_folio() can take a long time if the |
| 1320 |
* system is thrashing due to memory pressure, or if the folio |
| 1321 |
* is being written back. So grab it first before we start |
| 1322 |
* the transaction handle. This also allows us to allocate |
| 1323 |
* the folio (if needed) without using GFP_NOFS. |
| 1324 |
*/ |
| 1325 |
retry_grab: |
| 1326 |
folio = write_begin_get_folio(iocb, mapping, index, len); |
| 1327 |
if (IS_ERR(folio)) |
| 1328 |
return PTR_ERR(folio); |
| 1329 |
|
| 1330 |
if (len > folio_next_pos(folio) - pos) |
| 1331 |
len = folio_next_pos(folio) - pos; |
| 1332 |
|
| 1333 |
from = offset_in_folio(folio, pos); |
| 1334 |
to = from + len; |
| 1335 |
|
| 1336 |
/* |
| 1337 |
* The same as page allocation, we prealloc buffer heads before |
| 1338 |
* starting the handle. |
| 1339 |
*/ |
| 1340 |
if (!folio_buffers(folio)) |
| 1341 |
create_empty_buffers(folio, inode->i_sb->s_blocksize, 0); |
| 1342 |
|
| 1343 |
folio_unlock(folio); |
| 1344 |
|
| 1345 |
retry_journal: |
| 1346 |
handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
| 1347 |
if (IS_ERR(handle)) { |
| 1348 |
folio_put(folio); |
| 1349 |
return PTR_ERR(handle); |
| 1350 |
} |
| 1351 |
|
| 1352 |
folio_lock(folio); |
| 1353 |
if (folio->mapping != mapping) { |
| 1354 |
/* The folio got truncated from under us */ |
| 1355 |
folio_unlock(folio); |
| 1356 |
folio_put(folio); |
| 1357 |
ext4_journal_stop(handle); |
| 1358 |
goto retry_grab; |
| 1359 |
} |
| 1360 |
/* In case writeback began while the folio was unlocked */ |
| 1361 |
folio_wait_stable(folio); |
| 1362 |
|
| 1363 |
if (ext4_should_dioread_nolock(inode)) |
| 1364 |
ret = ext4_block_write_begin(handle, folio, pos, len, |
| 1365 |
ext4_get_block_unwritten); |
| 1366 |
else |
| 1367 |
ret = ext4_block_write_begin(handle, folio, pos, len, |
| 1368 |
ext4_get_block); |
| 1369 |
if (!ret && ext4_should_journal_data(inode)) { |
| 1370 |
ret = ext4_walk_page_buffers(handle, inode, |
| 1371 |
folio_buffers(folio), from, to, |
| 1372 |
NULL, do_journal_get_write_access); |
| 1373 |
} |
| 1374 |
|
| 1375 |
if (ret) { |
| 1376 |
bool extended = (pos + len > inode->i_size) && |
| 1377 |
!ext4_verity_in_progress(inode); |
| 1378 |
|
| 1379 |
folio_unlock(folio); |
| 1380 |
/* |
| 1381 |
* ext4_block_write_begin may have instantiated a few blocks |
| 1382 |
* outside i_size. Trim these off again. Don't need |
| 1383 |
* i_size_read because we hold i_rwsem. |
| 1384 |
* |
| 1385 |
* Add inode to orphan list in case we crash before |
| 1386 |
* truncate finishes |
| 1387 |
*/ |
| 1388 |
if (extended && ext4_can_truncate(inode)) |
| 1389 |
ext4_orphan_add(handle, inode); |
| 1390 |
|
| 1391 |
ext4_journal_stop(handle); |
| 1392 |
if (extended) { |
| 1393 |
ext4_truncate_failed_write(inode); |
| 1394 |
/* |
| 1395 |
* If truncate failed early the inode might |
| 1396 |
* still be on the orphan list; we need to |
| 1397 |
* make sure the inode is removed from the |
| 1398 |
* orphan list in that case. |
| 1399 |
*/ |
| 1400 |
if (inode->i_nlink) |
| 1401 |
ext4_orphan_del(NULL, inode); |
| 1402 |
} |
| 1403 |
|
| 1404 |
if (ret == -EAGAIN || |
| 1405 |
(ret == -ENOSPC && |
| 1406 |
ext4_should_retry_alloc(inode->i_sb, &retries))) |
| 1407 |
goto retry_journal; |
| 1408 |
folio_put(folio); |
| 1409 |
return ret; |
| 1410 |
} |
| 1411 |
*foliop = folio; |
| 1412 |
return ret; |
| 1413 |
} |
| 1414 |
|
| 1415 |
/* For write_end() in data=journal mode */ |
| 1416 |
static int write_end_fn(handle_t *handle, struct inode *inode, |
| 1417 |
struct buffer_head *bh) |
| 1418 |
{ |
| 1419 |
int ret; |
| 1420 |
if (!buffer_mapped(bh) || buffer_freed(bh)) |
| 1421 |
return 0; |
| 1422 |
set_buffer_uptodate(bh); |
| 1423 |
ret = ext4_dirty_journalled_data(handle, bh); |
| 1424 |
clear_buffer_meta(bh); |
| 1425 |
clear_buffer_prio(bh); |
| 1426 |
clear_buffer_new(bh); |
| 1427 |
return ret; |
| 1428 |
} |
| 1429 |
|
| 1430 |
/* |
| 1431 |
* We need to pick up the new inode size which generic_commit_write gave us |
| 1432 |
* `iocb` can be NULL - eg, when called from page_symlink(). |
| 1433 |
* |
| 1434 |
* ext4 never places buffers on inode->i_mapping->i_private_list. metadata |
| 1435 |
* buffers are managed internally. |
| 1436 |
*/ |
| 1437 |
static int ext4_write_end(const struct kiocb *iocb, |
| 1438 |
struct address_space *mapping, |
| 1439 |
loff_t pos, unsigned len, unsigned copied, |
| 1440 |
struct folio *folio, void *fsdata) |
| 1441 |
{ |
| 1442 |
handle_t *handle = ext4_journal_current_handle(); |
| 1443 |
struct inode *inode = mapping->host; |
| 1444 |
loff_t old_size = inode->i_size; |
| 1445 |
int ret = 0, ret2; |
| 1446 |
int i_size_changed = 0; |
| 1447 |
bool verity = ext4_verity_in_progress(inode); |
| 1448 |
|
| 1449 |
trace_ext4_write_end(inode, pos, len, copied); |
| 1450 |
|
| 1451 |
if (ext4_has_inline_data(inode) && |
| 1452 |
ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) |
| 1453 |
return ext4_write_inline_data_end(inode, pos, len, copied, |
| 1454 |
folio); |
| 1455 |
|
| 1456 |
copied = block_write_end(pos, len, copied, folio); |
| 1457 |
/* |
| 1458 |
* it's important to update i_size while still holding folio lock: |
| 1459 |
* page writeout could otherwise come in and zero beyond i_size. |
| 1460 |
* |
| 1461 |
* If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree |
| 1462 |
* blocks are being written past EOF, so skip the i_size update. |
| 1463 |
*/ |
| 1464 |
if (!verity) |
| 1465 |
i_size_changed = ext4_update_inode_size(inode, pos + copied); |
| 1466 |
folio_unlock(folio); |
| 1467 |
folio_put(folio); |
| 1468 |
|
| 1469 |
if (old_size < pos && !verity) { |
| 1470 |
pagecache_isize_extended(inode, old_size, pos); |
| 1471 |
ext4_zero_partial_blocks(handle, inode, old_size, pos - old_size); |
| 1472 |
} |
| 1473 |
/* |
| 1474 |
* Don't mark the inode dirty under folio lock. First, it unnecessarily |
| 1475 |
* makes the holding time of folio lock longer. Second, it forces lock |
| 1476 |
* ordering of folio lock and transaction start for journaling |
| 1477 |
* filesystems. |
| 1478 |
*/ |
| 1479 |
if (i_size_changed) |
| 1480 |
ret = ext4_mark_inode_dirty(handle, inode); |
| 1481 |
|
| 1482 |
if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode)) |
| 1483 |
/* if we have allocated more blocks and copied |
| 1484 |
* less. We will have blocks allocated outside |
| 1485 |
* inode->i_size. So truncate them |
| 1486 |
*/ |
| 1487 |
ext4_orphan_add(handle, inode); |
| 1488 |
|
| 1489 |
ret2 = ext4_journal_stop(handle); |
| 1490 |
if (!ret) |
| 1491 |
ret = ret2; |
| 1492 |
|
| 1493 |
if (pos + len > inode->i_size && !verity) { |
| 1494 |
ext4_truncate_failed_write(inode); |
| 1495 |
/* |
| 1496 |
* If truncate failed early the inode might still be |
| 1497 |
* on the orphan list; we need to make sure the inode |
| 1498 |
* is removed from the orphan list in that case. |
| 1499 |
*/ |
| 1500 |
if (inode->i_nlink) |
| 1501 |
ext4_orphan_del(NULL, inode); |
| 1502 |
} |
| 1503 |
|
| 1504 |
return ret ? ret : copied; |
| 1505 |
} |
| 1506 |
|
| 1507 |
/* |
| 1508 |
* This is a private version of folio_zero_new_buffers() which doesn't |
| 1509 |
* set the buffer to be dirty, since in data=journalled mode we need |
| 1510 |
* to call ext4_dirty_journalled_data() instead. |
| 1511 |
*/ |
| 1512 |
static void ext4_journalled_zero_new_buffers(handle_t *handle, |
| 1513 |
struct inode *inode, |
| 1514 |
struct folio *folio, |
| 1515 |
unsigned from, unsigned to) |
| 1516 |
{ |
| 1517 |
unsigned int block_start = 0, block_end; |
| 1518 |
struct buffer_head *head, *bh; |
| 1519 |
|
| 1520 |
bh = head = folio_buffers(folio); |
| 1521 |
do { |
| 1522 |
block_end = block_start + bh->b_size; |
| 1523 |
if (buffer_new(bh)) { |
| 1524 |
if (block_end > from && block_start < to) { |
| 1525 |
if (!folio_test_uptodate(folio)) { |
| 1526 |
unsigned start, size; |
| 1527 |
|
| 1528 |
start = max(from, block_start); |
| 1529 |
size = min(to, block_end) - start; |
| 1530 |
|
| 1531 |
folio_zero_range(folio, start, size); |
| 1532 |
} |
| 1533 |
clear_buffer_new(bh); |
| 1534 |
write_end_fn(handle, inode, bh); |
| 1535 |
} |
| 1536 |
} |
| 1537 |
block_start = block_end; |
| 1538 |
bh = bh->b_this_page; |
| 1539 |
} while (bh != head); |
| 1540 |
} |
| 1541 |
|
| 1542 |
static int ext4_journalled_write_end(const struct kiocb *iocb, |
| 1543 |
struct address_space *mapping, |
| 1544 |
loff_t pos, unsigned len, unsigned copied, |
| 1545 |
struct folio *folio, void *fsdata) |
| 1546 |
{ |
| 1547 |
handle_t *handle = ext4_journal_current_handle(); |
| 1548 |
struct inode *inode = mapping->host; |
| 1549 |
loff_t old_size = inode->i_size; |
| 1550 |
int ret = 0, ret2; |
| 1551 |
int partial = 0; |
| 1552 |
unsigned from, to; |
| 1553 |
int size_changed = 0; |
| 1554 |
bool verity = ext4_verity_in_progress(inode); |
| 1555 |
|
| 1556 |
trace_ext4_journalled_write_end(inode, pos, len, copied); |
| 1557 |
from = pos & (PAGE_SIZE - 1); |
| 1558 |
to = from + len; |
| 1559 |
|
| 1560 |
BUG_ON(!ext4_handle_valid(handle)); |
| 1561 |
|
| 1562 |
if (ext4_has_inline_data(inode)) |
| 1563 |
return ext4_write_inline_data_end(inode, pos, len, copied, |
| 1564 |
folio); |
| 1565 |
|
| 1566 |
if (unlikely(copied < len) && !folio_test_uptodate(folio)) { |
| 1567 |
copied = 0; |
| 1568 |
ext4_journalled_zero_new_buffers(handle, inode, folio, |
| 1569 |
from, to); |
| 1570 |
} else { |
| 1571 |
if (unlikely(copied < len)) |
| 1572 |
ext4_journalled_zero_new_buffers(handle, inode, folio, |
| 1573 |
from + copied, to); |
| 1574 |
ret = ext4_walk_page_buffers(handle, inode, |
| 1575 |
folio_buffers(folio), |
| 1576 |
from, from + copied, &partial, |
| 1577 |
write_end_fn); |
| 1578 |
if (!partial) |
| 1579 |
folio_mark_uptodate(folio); |
| 1580 |
} |
| 1581 |
if (!verity) |
| 1582 |
size_changed = ext4_update_inode_size(inode, pos + copied); |
| 1583 |
EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
| 1584 |
folio_unlock(folio); |
| 1585 |
folio_put(folio); |
| 1586 |
|
| 1587 |
if (old_size < pos && !verity) { |
| 1588 |
pagecache_isize_extended(inode, old_size, pos); |
| 1589 |
ext4_zero_partial_blocks(handle, inode, old_size, pos - old_size); |
| 1590 |
} |
| 1591 |
|
| 1592 |
if (size_changed) { |
| 1593 |
ret2 = ext4_mark_inode_dirty(handle, inode); |
| 1594 |
if (!ret) |
| 1595 |
ret = ret2; |
| 1596 |
} |
| 1597 |
|
| 1598 |
if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode)) |
| 1599 |
/* if we have allocated more blocks and copied |
| 1600 |
* less. We will have blocks allocated outside |
| 1601 |
* inode->i_size. So truncate them |
| 1602 |
*/ |
| 1603 |
ext4_orphan_add(handle, inode); |
| 1604 |
|
| 1605 |
ret2 = ext4_journal_stop(handle); |
| 1606 |
if (!ret) |
| 1607 |
ret = ret2; |
| 1608 |
if (pos + len > inode->i_size && !verity) { |
| 1609 |
ext4_truncate_failed_write(inode); |
| 1610 |
/* |
| 1611 |
* If truncate failed early the inode might still be |
| 1612 |
* on the orphan list; we need to make sure the inode |
| 1613 |
* is removed from the orphan list in that case. |
| 1614 |
*/ |
| 1615 |
if (inode->i_nlink) |
| 1616 |
ext4_orphan_del(NULL, inode); |
| 1617 |
} |
| 1618 |
|
| 1619 |
return ret ? ret : copied; |
| 1620 |
} |
| 1621 |
|
| 1622 |
/* |
| 1623 |
* Reserve space for 'nr_resv' clusters |
| 1624 |
*/ |
| 1625 |
static int ext4_da_reserve_space(struct inode *inode, int nr_resv) |
| 1626 |
{ |
| 1627 |
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 1628 |
struct ext4_inode_info *ei = EXT4_I(inode); |
| 1629 |
int ret; |
| 1630 |
|
| 1631 |
/* |
| 1632 |
* We will charge metadata quota at writeout time; this saves |
| 1633 |
* us from metadata over-estimation, though we may go over by |
| 1634 |
* a small amount in the end. Here we just reserve for data. |
| 1635 |
*/ |
| 1636 |
ret = dquot_reserve_block(inode, EXT4_C2B(sbi, nr_resv)); |
| 1637 |
if (ret) |
| 1638 |
return ret; |
| 1639 |
|
| 1640 |
spin_lock(&ei->i_block_reservation_lock); |
| 1641 |
if (ext4_claim_free_clusters(sbi, nr_resv, 0)) { |
| 1642 |
spin_unlock(&ei->i_block_reservation_lock); |
| 1643 |
dquot_release_reservation_block(inode, EXT4_C2B(sbi, nr_resv)); |
| 1644 |
return -ENOSPC; |
| 1645 |
} |
| 1646 |
ei->i_reserved_data_blocks += nr_resv; |
| 1647 |
trace_ext4_da_reserve_space(inode, nr_resv); |
| 1648 |
spin_unlock(&ei->i_block_reservation_lock); |
| 1649 |
|
| 1650 |
return 0; /* success */ |
| 1651 |
} |
| 1652 |
|
| 1653 |
void ext4_da_release_space(struct inode *inode, int to_free) |
| 1654 |
{ |
| 1655 |
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 1656 |
struct ext4_inode_info *ei = EXT4_I(inode); |
| 1657 |
|
| 1658 |
if (!to_free) |
| 1659 |
return; /* Nothing to release, exit */ |
| 1660 |
|
| 1661 |
spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
| 1662 |
|
| 1663 |
trace_ext4_da_release_space(inode, to_free); |
| 1664 |
if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
| 1665 |
/* |
| 1666 |
* if there aren't enough reserved blocks, then the |
| 1667 |
* counter is messed up somewhere. Since this |
| 1668 |
* function is called from invalidate page, it's |
| 1669 |
* harmless to return without any action. |
| 1670 |
*/ |
| 1671 |
ext4_warning(inode->i_sb, "ext4_da_release_space: " |
| 1672 |
"ino %lu, to_free %d with only %d reserved " |
| 1673 |
"data blocks", inode->i_ino, to_free, |
| 1674 |
ei->i_reserved_data_blocks); |
| 1675 |
WARN_ON(1); |
| 1676 |
to_free = ei->i_reserved_data_blocks; |
| 1677 |
} |
| 1678 |
ei->i_reserved_data_blocks -= to_free; |
| 1679 |
|
| 1680 |
/* update fs dirty data blocks counter */ |
| 1681 |
percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
| 1682 |
|
| 1683 |
spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
| 1684 |
|
| 1685 |
dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
| 1686 |
} |
| 1687 |
|
| 1688 |
/* |
| 1689 |
* Delayed allocation stuff |
| 1690 |
*/ |
| 1691 |
|
| 1692 |
struct mpage_da_data { |
| 1693 |
/* These are input fields for ext4_do_writepages() */ |
| 1694 |
struct inode *inode; |
| 1695 |
struct writeback_control *wbc; |
| 1696 |
unsigned int can_map:1; /* Can writepages call map blocks? */ |
| 1697 |
|
| 1698 |
/* These are internal state of ext4_do_writepages() */ |
| 1699 |
loff_t start_pos; /* The start pos to write */ |
| 1700 |
loff_t next_pos; /* Current pos to examine */ |
| 1701 |
loff_t end_pos; /* Last pos to examine */ |
| 1702 |
|
| 1703 |
/* |
| 1704 |
* Extent to map - this can be after start_pos because that can be |
| 1705 |
* fully mapped. We somewhat abuse m_flags to store whether the extent |
| 1706 |
* is delalloc or unwritten. |
| 1707 |
*/ |
| 1708 |
struct ext4_map_blocks map; |
| 1709 |
struct ext4_io_submit io_submit; /* IO submission data */ |
| 1710 |
unsigned int do_map:1; |
| 1711 |
unsigned int scanned_until_end:1; |
| 1712 |
unsigned int journalled_more_data:1; |
| 1713 |
}; |
| 1714 |
|
| 1715 |
static void mpage_release_unused_pages(struct mpage_da_data *mpd, |
| 1716 |
bool invalidate) |
| 1717 |
{ |
| 1718 |
unsigned nr, i; |
| 1719 |
pgoff_t index, end; |
| 1720 |
struct folio_batch fbatch; |
| 1721 |
struct inode *inode = mpd->inode; |
| 1722 |
struct address_space *mapping = inode->i_mapping; |
| 1723 |
|
| 1724 |
/* This is necessary when next_pos == 0. */ |
| 1725 |
if (mpd->start_pos >= mpd->next_pos) |
| 1726 |
return; |
| 1727 |
|
| 1728 |
mpd->scanned_until_end = 0; |
| 1729 |
if (invalidate) { |
| 1730 |
ext4_lblk_t start, last; |
| 1731 |
start = EXT4_B_TO_LBLK(inode, mpd->start_pos); |
| 1732 |
last = mpd->next_pos >> inode->i_blkbits; |
| 1733 |
|
| 1734 |
/* |
| 1735 |
* avoid racing with extent status tree scans made by |
| 1736 |
* ext4_insert_delayed_block() |
| 1737 |
*/ |
| 1738 |
down_write(&EXT4_I(inode)->i_data_sem); |
| 1739 |
ext4_es_remove_extent(inode, start, last - start); |
| 1740 |
up_write(&EXT4_I(inode)->i_data_sem); |
| 1741 |
} |
| 1742 |
|
| 1743 |
folio_batch_init(&fbatch); |
| 1744 |
index = mpd->start_pos >> PAGE_SHIFT; |
| 1745 |
end = mpd->next_pos >> PAGE_SHIFT; |
| 1746 |
while (index < end) { |
| 1747 |
nr = filemap_get_folios(mapping, &index, end - 1, &fbatch); |
| 1748 |
if (nr == 0) |
| 1749 |
break; |
| 1750 |
for (i = 0; i < nr; i++) { |
| 1751 |
struct folio *folio = fbatch.folios[i]; |
| 1752 |
|
| 1753 |
if (folio_pos(folio) < mpd->start_pos) |
| 1754 |
continue; |
| 1755 |
if (folio_next_index(folio) > end) |
| 1756 |
continue; |
| 1757 |
BUG_ON(!folio_test_locked(folio)); |
| 1758 |
BUG_ON(folio_test_writeback(folio)); |
| 1759 |
if (invalidate) { |
| 1760 |
if (folio_mapped(folio)) |
| 1761 |
folio_clear_dirty_for_io(folio); |
| 1762 |
block_invalidate_folio(folio, 0, |
| 1763 |
folio_size(folio)); |
| 1764 |
folio_clear_uptodate(folio); |
| 1765 |
} |
| 1766 |
folio_unlock(folio); |
| 1767 |
} |
| 1768 |
folio_batch_release(&fbatch); |
| 1769 |
} |
| 1770 |
} |
| 1771 |
|
| 1772 |
static void ext4_print_free_blocks(struct inode *inode) |
| 1773 |
{ |
| 1774 |
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 1775 |
struct super_block *sb = inode->i_sb; |
| 1776 |
struct ext4_inode_info *ei = EXT4_I(inode); |
| 1777 |
|
| 1778 |
ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", |
| 1779 |
EXT4_C2B(EXT4_SB(inode->i_sb), |
| 1780 |
ext4_count_free_clusters(sb))); |
| 1781 |
ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
| 1782 |
ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", |
| 1783 |
(long long) EXT4_C2B(EXT4_SB(sb), |
| 1784 |
percpu_counter_sum(&sbi->s_freeclusters_counter))); |
| 1785 |
ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
| 1786 |
(long long) EXT4_C2B(EXT4_SB(sb), |
| 1787 |
percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
| 1788 |
ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
| 1789 |
ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", |
| 1790 |
ei->i_reserved_data_blocks); |
| 1791 |
return; |
| 1792 |
} |
| 1793 |
|
| 1794 |
/* |
| 1795 |
* Check whether the cluster containing lblk has been allocated or has |
| 1796 |
* delalloc reservation. |
| 1797 |
* |
| 1798 |
* Returns 0 if the cluster doesn't have either, 1 if it has delalloc |
| 1799 |
* reservation, 2 if it's already been allocated, negative error code on |
| 1800 |
* failure. |
| 1801 |
*/ |
| 1802 |
static int ext4_clu_alloc_state(struct inode *inode, ext4_lblk_t lblk) |
| 1803 |
{ |
| 1804 |
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 1805 |
int ret; |
| 1806 |
|
| 1807 |
/* Has delalloc reservation? */ |
| 1808 |
if (ext4_es_scan_clu(inode, &ext4_es_is_delayed, lblk)) |
| 1809 |
return 1; |
| 1810 |
|
| 1811 |
/* Already been allocated? */ |
| 1812 |
if (ext4_es_scan_clu(inode, &ext4_es_is_mapped, lblk)) |
| 1813 |
return 2; |
| 1814 |
ret = ext4_clu_mapped(inode, EXT4_B2C(sbi, lblk)); |
| 1815 |
if (ret < 0) |
| 1816 |
return ret; |
| 1817 |
if (ret > 0) |
| 1818 |
return 2; |
| 1819 |
|
| 1820 |
return 0; |
| 1821 |
} |
| 1822 |
|
| 1823 |
/* |
| 1824 |
* ext4_insert_delayed_blocks - adds a multiple delayed blocks to the extents |
| 1825 |
* status tree, incrementing the reserved |
| 1826 |
* cluster/block count or making pending |
| 1827 |
* reservations where needed |
| 1828 |
* |
| 1829 |
* @inode - file containing the newly added block |
| 1830 |
* @lblk - start logical block to be added |
| 1831 |
* @len - length of blocks to be added |
| 1832 |
* |
| 1833 |
* Returns 0 on success, negative error code on failure. |
| 1834 |
*/ |
| 1835 |
static int ext4_insert_delayed_blocks(struct inode *inode, ext4_lblk_t lblk, |
| 1836 |
ext4_lblk_t len) |
| 1837 |
{ |
| 1838 |
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 1839 |
int ret; |
| 1840 |
bool lclu_allocated = false; |
| 1841 |
bool end_allocated = false; |
| 1842 |
ext4_lblk_t resv_clu; |
| 1843 |
ext4_lblk_t end = lblk + len - 1; |
| 1844 |
|
| 1845 |
/* |
| 1846 |
* If the cluster containing lblk or end is shared with a delayed, |
| 1847 |
* written, or unwritten extent in a bigalloc file system, it's |
| 1848 |
* already been accounted for and does not need to be reserved. |
| 1849 |
* A pending reservation must be made for the cluster if it's |
| 1850 |
* shared with a written or unwritten extent and doesn't already |
| 1851 |
* have one. Written and unwritten extents can be purged from the |
| 1852 |
* extents status tree if the system is under memory pressure, so |
| 1853 |
* it's necessary to examine the extent tree if a search of the |
| 1854 |
* extents status tree doesn't get a match. |
| 1855 |
*/ |
| 1856 |
if (sbi->s_cluster_ratio == 1) { |
| 1857 |
ret = ext4_da_reserve_space(inode, len); |
| 1858 |
if (ret != 0) /* ENOSPC */ |
| 1859 |
return ret; |
| 1860 |
} else { /* bigalloc */ |
| 1861 |
resv_clu = EXT4_B2C(sbi, end) - EXT4_B2C(sbi, lblk) + 1; |
| 1862 |
|
| 1863 |
ret = ext4_clu_alloc_state(inode, lblk); |
| 1864 |
if (ret < 0) |
| 1865 |
return ret; |
| 1866 |
if (ret > 0) { |
| 1867 |
resv_clu--; |
| 1868 |
lclu_allocated = (ret == 2); |
| 1869 |
} |
| 1870 |
|
| 1871 |
if (EXT4_B2C(sbi, lblk) != EXT4_B2C(sbi, end)) { |
| 1872 |
ret = ext4_clu_alloc_state(inode, end); |
| 1873 |
if (ret < 0) |
| 1874 |
return ret; |
| 1875 |
if (ret > 0) { |
| 1876 |
resv_clu--; |
| 1877 |
end_allocated = (ret == 2); |
| 1878 |
} |
| 1879 |
} |
| 1880 |
|
| 1881 |
if (resv_clu) { |
| 1882 |
ret = ext4_da_reserve_space(inode, resv_clu); |
| 1883 |
if (ret != 0) /* ENOSPC */ |
| 1884 |
return ret; |
| 1885 |
} |
| 1886 |
} |
| 1887 |
|
| 1888 |
ext4_es_insert_delayed_extent(inode, lblk, len, lclu_allocated, |
| 1889 |
end_allocated); |
| 1890 |
return 0; |
| 1891 |
} |
| 1892 |
|
| 1893 |
/* |
| 1894 |
* Looks up the requested blocks and sets the delalloc extent map. |
| 1895 |
* First try to look up for the extent entry that contains the requested |
| 1896 |
* blocks in the extent status tree without i_data_sem, then try to look |
| 1897 |
* up for the ondisk extent mapping with i_data_sem in read mode, |
| 1898 |
* finally hold i_data_sem in write mode, looks up again and add a |
| 1899 |
* delalloc extent entry if it still couldn't find any extent. Pass out |
| 1900 |
* the mapped extent through @map and return 0 on success. |
| 1901 |
*/ |
| 1902 |
static int ext4_da_map_blocks(struct inode *inode, struct ext4_map_blocks *map) |
| 1903 |
{ |
| 1904 |
struct extent_status es; |
| 1905 |
int retval; |
| 1906 |
#ifdef ES_AGGRESSIVE_TEST |
| 1907 |
struct ext4_map_blocks orig_map; |
| 1908 |
|
| 1909 |
memcpy(&orig_map, map, sizeof(*map)); |
| 1910 |
#endif |
| 1911 |
|
| 1912 |
map->m_flags = 0; |
| 1913 |
ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len, |
| 1914 |
(unsigned long) map->m_lblk); |
| 1915 |
|
| 1916 |
ext4_check_map_extents_env(inode); |
| 1917 |
|
| 1918 |
/* Lookup extent status tree firstly */ |
| 1919 |
if (ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es, NULL)) { |
| 1920 |
map->m_len = min_t(unsigned int, map->m_len, |
| 1921 |
es.es_len - (map->m_lblk - es.es_lblk)); |
| 1922 |
|
| 1923 |
if (ext4_es_is_hole(&es)) |
| 1924 |
goto add_delayed; |
| 1925 |
|
| 1926 |
found: |
| 1927 |
/* |
| 1928 |
* Delayed extent could be allocated by fallocate. |
| 1929 |
* So we need to check it. |
| 1930 |
*/ |
| 1931 |
if (ext4_es_is_delayed(&es)) { |
| 1932 |
map->m_flags |= EXT4_MAP_DELAYED; |
| 1933 |
return 0; |
| 1934 |
} |
| 1935 |
|
| 1936 |
map->m_pblk = ext4_es_pblock(&es) + map->m_lblk - es.es_lblk; |
| 1937 |
if (ext4_es_is_written(&es)) |
| 1938 |
map->m_flags |= EXT4_MAP_MAPPED; |
| 1939 |
else if (ext4_es_is_unwritten(&es)) |
| 1940 |
map->m_flags |= EXT4_MAP_UNWRITTEN; |
| 1941 |
else |
| 1942 |
BUG(); |
| 1943 |
|
| 1944 |
#ifdef ES_AGGRESSIVE_TEST |
| 1945 |
ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); |
| 1946 |
#endif |
| 1947 |
return 0; |
| 1948 |
} |
| 1949 |
|
| 1950 |
/* |
| 1951 |
* Try to see if we can get the block without requesting a new |
| 1952 |
* file system block. |
| 1953 |
*/ |
| 1954 |
down_read(&EXT4_I(inode)->i_data_sem); |
| 1955 |
if (ext4_has_inline_data(inode)) |
| 1956 |
retval = 0; |
| 1957 |
else |
| 1958 |
retval = ext4_map_query_blocks(NULL, inode, map, 0); |
| 1959 |
up_read(&EXT4_I(inode)->i_data_sem); |
| 1960 |
if (retval) |
| 1961 |
return retval < 0 ? retval : 0; |
| 1962 |
|
| 1963 |
add_delayed: |
| 1964 |
down_write(&EXT4_I(inode)->i_data_sem); |
| 1965 |
/* |
| 1966 |
* Page fault path (ext4_page_mkwrite does not take i_rwsem) |
| 1967 |
* and fallocate path (no folio lock) can race. Make sure we |
| 1968 |
* lookup the extent status tree here again while i_data_sem |
| 1969 |
* is held in write mode, before inserting a new da entry in |
| 1970 |
* the extent status tree. |
| 1971 |
*/ |
| 1972 |
if (ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es, NULL)) { |
| 1973 |
map->m_len = min_t(unsigned int, map->m_len, |
| 1974 |
es.es_len - (map->m_lblk - es.es_lblk)); |
| 1975 |
|
| 1976 |
if (!ext4_es_is_hole(&es)) { |
| 1977 |
up_write(&EXT4_I(inode)->i_data_sem); |
| 1978 |
goto found; |
| 1979 |
} |
| 1980 |
} else if (!ext4_has_inline_data(inode)) { |
| 1981 |
retval = ext4_map_query_blocks(NULL, inode, map, 0); |
| 1982 |
if (retval) { |
| 1983 |
up_write(&EXT4_I(inode)->i_data_sem); |
| 1984 |
return retval < 0 ? retval : 0; |
| 1985 |
} |
| 1986 |
} |
| 1987 |
|
| 1988 |
map->m_flags |= EXT4_MAP_DELAYED; |
| 1989 |
retval = ext4_insert_delayed_blocks(inode, map->m_lblk, map->m_len); |
| 1990 |
if (!retval) |
| 1991 |
map->m_seq = READ_ONCE(EXT4_I(inode)->i_es_seq); |
| 1992 |
up_write(&EXT4_I(inode)->i_data_sem); |
| 1993 |
|
| 1994 |
return retval; |
| 1995 |
} |
| 1996 |
|
| 1997 |
/* |
| 1998 |
* This is a special get_block_t callback which is used by |
| 1999 |
* ext4_da_write_begin(). It will either return mapped block or |
| 2000 |
* reserve space for a single block. |
| 2001 |
* |
| 2002 |
* For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. |
| 2003 |
* We also have b_blocknr = -1 and b_bdev initialized properly |
| 2004 |
* |
| 2005 |
* For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. |
| 2006 |
* We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev |
| 2007 |
* initialized properly. |
| 2008 |
*/ |
| 2009 |
int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
| 2010 |
struct buffer_head *bh, int create) |
| 2011 |
{ |
| 2012 |
struct ext4_map_blocks map; |
| 2013 |
sector_t invalid_block = ~((sector_t) 0xffff); |
| 2014 |
int ret = 0; |
| 2015 |
|
| 2016 |
BUG_ON(create == 0); |
| 2017 |
BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
| 2018 |
|
| 2019 |
if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) |
| 2020 |
invalid_block = ~0; |
| 2021 |
|
| 2022 |
map.m_lblk = iblock; |
| 2023 |
map.m_len = 1; |
| 2024 |
|
| 2025 |
/* |
| 2026 |
* first, we need to know whether the block is allocated already |
| 2027 |
* preallocated blocks are unmapped but should treated |
| 2028 |
* the same as allocated blocks. |
| 2029 |
*/ |
| 2030 |
ret = ext4_da_map_blocks(inode, &map); |
| 2031 |
if (ret < 0) |
| 2032 |
return ret; |
| 2033 |
|
| 2034 |
if (map.m_flags & EXT4_MAP_DELAYED) { |
| 2035 |
map_bh(bh, inode->i_sb, invalid_block); |
| 2036 |
set_buffer_new(bh); |
| 2037 |
set_buffer_delay(bh); |
| 2038 |
return 0; |
| 2039 |
} |
| 2040 |
|
| 2041 |
map_bh(bh, inode->i_sb, map.m_pblk); |
| 2042 |
ext4_update_bh_state(bh, map.m_flags); |
| 2043 |
|
| 2044 |
if (buffer_unwritten(bh)) { |
| 2045 |
/* A delayed write to unwritten bh should be marked |
| 2046 |
* new and mapped. Mapped ensures that we don't do |
| 2047 |
* get_block multiple times when we write to the same |
| 2048 |
* offset and new ensures that we do proper zero out |
| 2049 |
* for partial write. |
| 2050 |
*/ |
| 2051 |
set_buffer_new(bh); |
| 2052 |
set_buffer_mapped(bh); |
| 2053 |
} |
| 2054 |
return 0; |
| 2055 |
} |
| 2056 |
|
| 2057 |
static void mpage_folio_done(struct mpage_da_data *mpd, struct folio *folio) |
| 2058 |
{ |
| 2059 |
mpd->start_pos += folio_size(folio); |
| 2060 |
mpd->wbc->nr_to_write -= folio_nr_pages(folio); |
| 2061 |
folio_unlock(folio); |
| 2062 |
} |
| 2063 |
|
| 2064 |
static int mpage_submit_folio(struct mpage_da_data *mpd, struct folio *folio) |
| 2065 |
{ |
| 2066 |
size_t len; |
| 2067 |
loff_t size; |
| 2068 |
int err; |
| 2069 |
|
| 2070 |
WARN_ON_ONCE(folio_pos(folio) != mpd->start_pos); |
| 2071 |
folio_clear_dirty_for_io(folio); |
| 2072 |
/* |
| 2073 |
* We have to be very careful here! Nothing protects writeback path |
| 2074 |
* against i_size changes and the page can be writeably mapped into |
| 2075 |
* page tables. So an application can be growing i_size and writing |
| 2076 |
* data through mmap while writeback runs. folio_clear_dirty_for_io() |
| 2077 |
* write-protects our page in page tables and the page cannot get |
| 2078 |
* written to again until we release folio lock. So only after |
| 2079 |
* folio_clear_dirty_for_io() we are safe to sample i_size for |
| 2080 |
* ext4_bio_write_folio() to zero-out tail of the written page. We rely |
| 2081 |
* on the barrier provided by folio_test_clear_dirty() in |
| 2082 |
* folio_clear_dirty_for_io() to make sure i_size is really sampled only |
| 2083 |
* after page tables are updated. |
| 2084 |
*/ |
| 2085 |
size = i_size_read(mpd->inode); |
| 2086 |
len = folio_size(folio); |
| 2087 |
if (folio_pos(folio) + len > size && |
| 2088 |
!ext4_verity_in_progress(mpd->inode)) |
| 2089 |
len = size & (len - 1); |
| 2090 |
err = ext4_bio_write_folio(&mpd->io_submit, folio, len); |
| 2091 |
|
| 2092 |
return err; |
| 2093 |
} |
| 2094 |
|
| 2095 |
#define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay)) |
| 2096 |
|
| 2097 |
/* |
| 2098 |
* mballoc gives us at most this number of blocks... |
| 2099 |
* XXX: That seems to be only a limitation of ext4_mb_normalize_request(). |
| 2100 |
* The rest of mballoc seems to handle chunks up to full group size. |
| 2101 |
*/ |
| 2102 |
#define MAX_WRITEPAGES_EXTENT_LEN 2048 |
| 2103 |
|
| 2104 |
/* |
| 2105 |
* mpage_add_bh_to_extent - try to add bh to extent of blocks to map |
| 2106 |
* |
| 2107 |
* @mpd - extent of blocks |
| 2108 |
* @lblk - logical number of the block in the file |
| 2109 |
* @bh - buffer head we want to add to the extent |
| 2110 |
* |
| 2111 |
* The function is used to collect contig. blocks in the same state. If the |
| 2112 |
* buffer doesn't require mapping for writeback and we haven't started the |
| 2113 |
* extent of buffers to map yet, the function returns 'true' immediately - the |
| 2114 |
* caller can write the buffer right away. Otherwise the function returns true |
| 2115 |
* if the block has been added to the extent, false if the block couldn't be |
| 2116 |
* added. |
| 2117 |
*/ |
| 2118 |
static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk, |
| 2119 |
struct buffer_head *bh) |
| 2120 |
{ |
| 2121 |
struct ext4_map_blocks *map = &mpd->map; |
| 2122 |
|
| 2123 |
/* Buffer that doesn't need mapping for writeback? */ |
| 2124 |
if (!buffer_dirty(bh) || !buffer_mapped(bh) || |
| 2125 |
(!buffer_delay(bh) && !buffer_unwritten(bh))) { |
| 2126 |
/* So far no extent to map => we write the buffer right away */ |
| 2127 |
if (map->m_len == 0) |
| 2128 |
return true; |
| 2129 |
return false; |
| 2130 |
} |
| 2131 |
|
| 2132 |
/* First block in the extent? */ |
| 2133 |
if (map->m_len == 0) { |
| 2134 |
/* We cannot map unless handle is started... */ |
| 2135 |
if (!mpd->do_map) |
| 2136 |
return false; |
| 2137 |
map->m_lblk = lblk; |
| 2138 |
map->m_len = 1; |
| 2139 |
map->m_flags = bh->b_state & BH_FLAGS; |
| 2140 |
return true; |
| 2141 |
} |
| 2142 |
|
| 2143 |
/* Don't go larger than mballoc is willing to allocate */ |
| 2144 |
if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN) |
| 2145 |
return false; |
| 2146 |
|
| 2147 |
/* Can we merge the block to our big extent? */ |
| 2148 |
if (lblk == map->m_lblk + map->m_len && |
| 2149 |
(bh->b_state & BH_FLAGS) == map->m_flags) { |
| 2150 |
map->m_len++; |
| 2151 |
return true; |
| 2152 |
} |
| 2153 |
return false; |
| 2154 |
} |
| 2155 |
|
| 2156 |
/* |
| 2157 |
* mpage_process_page_bufs - submit page buffers for IO or add them to extent |
| 2158 |
* |
| 2159 |
* @mpd - extent of blocks for mapping |
| 2160 |
* @head - the first buffer in the page |
| 2161 |
* @bh - buffer we should start processing from |
| 2162 |
* @lblk - logical number of the block in the file corresponding to @bh |
| 2163 |
* |
| 2164 |
* Walk through page buffers from @bh upto @head (exclusive) and either submit |
| 2165 |
* the page for IO if all buffers in this page were mapped and there's no |
| 2166 |
* accumulated extent of buffers to map or add buffers in the page to the |
| 2167 |
* extent of buffers to map. The function returns 1 if the caller can continue |
| 2168 |
* by processing the next page, 0 if it should stop adding buffers to the |
| 2169 |
* extent to map because we cannot extend it anymore. It can also return value |
| 2170 |
* < 0 in case of error during IO submission. |
| 2171 |
*/ |
| 2172 |
static int mpage_process_page_bufs(struct mpage_da_data *mpd, |
| 2173 |
struct buffer_head *head, |
| 2174 |
struct buffer_head *bh, |
| 2175 |
ext4_lblk_t lblk) |
| 2176 |
{ |
| 2177 |
struct inode *inode = mpd->inode; |
| 2178 |
int err; |
| 2179 |
ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1) |
| 2180 |
>> inode->i_blkbits; |
| 2181 |
|
| 2182 |
if (ext4_verity_in_progress(inode)) |
| 2183 |
blocks = EXT_MAX_BLOCKS; |
| 2184 |
|
| 2185 |
do { |
| 2186 |
BUG_ON(buffer_locked(bh)); |
| 2187 |
|
| 2188 |
if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) { |
| 2189 |
/* Found extent to map? */ |
| 2190 |
if (mpd->map.m_len) |
| 2191 |
return 0; |
| 2192 |
/* Buffer needs mapping and handle is not started? */ |
| 2193 |
if (!mpd->do_map) |
| 2194 |
return 0; |
| 2195 |
/* Everything mapped so far and we hit EOF */ |
| 2196 |
break; |
| 2197 |
} |
| 2198 |
} while (lblk++, (bh = bh->b_this_page) != head); |
| 2199 |
/* So far everything mapped? Submit the page for IO. */ |
| 2200 |
if (mpd->map.m_len == 0) { |
| 2201 |
err = mpage_submit_folio(mpd, head->b_folio); |
| 2202 |
if (err < 0) |
| 2203 |
return err; |
| 2204 |
mpage_folio_done(mpd, head->b_folio); |
| 2205 |
} |
| 2206 |
if (lblk >= blocks) { |
| 2207 |
mpd->scanned_until_end = 1; |
| 2208 |
return 0; |
| 2209 |
} |
| 2210 |
return 1; |
| 2211 |
} |
| 2212 |
|
| 2213 |
/* |
| 2214 |
* mpage_process_folio - update folio buffers corresponding to changed extent |
| 2215 |
* and may submit fully mapped page for IO |
| 2216 |
* @mpd: description of extent to map, on return next extent to map |
| 2217 |
* @folio: Contains these buffers. |
| 2218 |
* @m_lblk: logical block mapping. |
| 2219 |
* @m_pblk: corresponding physical mapping. |
| 2220 |
* @map_bh: determines on return whether this page requires any further |
| 2221 |
* mapping or not. |
| 2222 |
* |
| 2223 |
* Scan given folio buffers corresponding to changed extent and update buffer |
| 2224 |
* state according to new extent state. |
| 2225 |
* We map delalloc buffers to their physical location, clear unwritten bits. |
| 2226 |
* If the given folio is not fully mapped, we update @mpd to the next extent in |
| 2227 |
* the given folio that needs mapping & return @map_bh as true. |
| 2228 |
*/ |
| 2229 |
static int mpage_process_folio(struct mpage_da_data *mpd, struct folio *folio, |
| 2230 |
ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk, |
| 2231 |
bool *map_bh) |
| 2232 |
{ |
| 2233 |
struct buffer_head *head, *bh; |
| 2234 |
ext4_io_end_t *io_end = mpd->io_submit.io_end; |
| 2235 |
ext4_lblk_t lblk = *m_lblk; |
| 2236 |
ext4_fsblk_t pblock = *m_pblk; |
| 2237 |
int err = 0; |
| 2238 |
ssize_t io_end_size = 0; |
| 2239 |
struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end); |
| 2240 |
|
| 2241 |
bh = head = folio_buffers(folio); |
| 2242 |
do { |
| 2243 |
if (lblk < mpd->map.m_lblk) |
| 2244 |
continue; |
| 2245 |
if (lblk >= mpd->map.m_lblk + mpd->map.m_len) { |
| 2246 |
/* |
| 2247 |
* Buffer after end of mapped extent. |
| 2248 |
* Find next buffer in the folio to map. |
| 2249 |
*/ |
| 2250 |
mpd->map.m_len = 0; |
| 2251 |
mpd->map.m_flags = 0; |
| 2252 |
io_end_vec->size += io_end_size; |
| 2253 |
|
| 2254 |
err = mpage_process_page_bufs(mpd, head, bh, lblk); |
| 2255 |
if (err > 0) |
| 2256 |
err = 0; |
| 2257 |
if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) { |
| 2258 |
io_end_vec = ext4_alloc_io_end_vec(io_end); |
| 2259 |
if (IS_ERR(io_end_vec)) { |
| 2260 |
err = PTR_ERR(io_end_vec); |
| 2261 |
goto out; |
| 2262 |
} |
| 2263 |
io_end_vec->offset = EXT4_LBLK_TO_B(mpd->inode, |
| 2264 |
mpd->map.m_lblk); |
| 2265 |
} |
| 2266 |
*map_bh = true; |
| 2267 |
goto out; |
| 2268 |
} |
| 2269 |
if (buffer_delay(bh)) { |
| 2270 |
clear_buffer_delay(bh); |
| 2271 |
bh->b_blocknr = pblock++; |
| 2272 |
} |
| 2273 |
clear_buffer_unwritten(bh); |
| 2274 |
io_end_size += i_blocksize(mpd->inode); |
| 2275 |
} while (lblk++, (bh = bh->b_this_page) != head); |
| 2276 |
|
| 2277 |
io_end_vec->size += io_end_size; |
| 2278 |
*map_bh = false; |
| 2279 |
out: |
| 2280 |
*m_lblk = lblk; |
| 2281 |
*m_pblk = pblock; |
| 2282 |
return err; |
| 2283 |
} |
| 2284 |
|
| 2285 |
/* |
| 2286 |
* mpage_map_buffers - update buffers corresponding to changed extent and |
| 2287 |
* submit fully mapped pages for IO |
| 2288 |
* |
| 2289 |
* @mpd - description of extent to map, on return next extent to map |
| 2290 |
* |
| 2291 |
* Scan buffers corresponding to changed extent (we expect corresponding pages |
| 2292 |
* to be already locked) and update buffer state according to new extent state. |
| 2293 |
* We map delalloc buffers to their physical location, clear unwritten bits, |
| 2294 |
* and mark buffers as uninit when we perform writes to unwritten extents |
| 2295 |
* and do extent conversion after IO is finished. If the last page is not fully |
| 2296 |
* mapped, we update @map to the next extent in the last page that needs |
| 2297 |
* mapping. Otherwise we submit the page for IO. |
| 2298 |
*/ |
| 2299 |
static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd) |
| 2300 |
{ |
| 2301 |
struct folio_batch fbatch; |
| 2302 |
unsigned nr, i; |
| 2303 |
struct inode *inode = mpd->inode; |
| 2304 |
pgoff_t start, end; |
| 2305 |
ext4_lblk_t lblk; |
| 2306 |
ext4_fsblk_t pblock; |
| 2307 |
int err; |
| 2308 |
bool map_bh = false; |
| 2309 |
|
| 2310 |
start = EXT4_LBLK_TO_PG(inode, mpd->map.m_lblk); |
| 2311 |
end = EXT4_LBLK_TO_PG(inode, mpd->map.m_lblk + mpd->map.m_len - 1); |
| 2312 |
pblock = mpd->map.m_pblk; |
| 2313 |
|
| 2314 |
folio_batch_init(&fbatch); |
| 2315 |
while (start <= end) { |
| 2316 |
nr = filemap_get_folios(inode->i_mapping, &start, end, &fbatch); |
| 2317 |
if (nr == 0) |
| 2318 |
break; |
| 2319 |
for (i = 0; i < nr; i++) { |
| 2320 |
struct folio *folio = fbatch.folios[i]; |
| 2321 |
|
| 2322 |
lblk = EXT4_PG_TO_LBLK(inode, folio->index); |
| 2323 |
err = mpage_process_folio(mpd, folio, &lblk, &pblock, |
| 2324 |
&map_bh); |
| 2325 |
/* |
| 2326 |
* If map_bh is true, means page may require further bh |
| 2327 |
* mapping, or maybe the page was submitted for IO. |
| 2328 |
* So we return to call further extent mapping. |
| 2329 |
*/ |
| 2330 |
if (err < 0 || map_bh) |
| 2331 |
goto out; |
| 2332 |
/* Page fully mapped - let IO run! */ |
| 2333 |
err = mpage_submit_folio(mpd, folio); |
| 2334 |
if (err < 0) |
| 2335 |
goto out; |
| 2336 |
mpage_folio_done(mpd, folio); |
| 2337 |
} |
| 2338 |
folio_batch_release(&fbatch); |
| 2339 |
} |
| 2340 |
/* Extent fully mapped and matches with page boundary. We are done. */ |
| 2341 |
mpd->map.m_len = 0; |
| 2342 |
mpd->map.m_flags = 0; |
| 2343 |
return 0; |
| 2344 |
out: |
| 2345 |
folio_batch_release(&fbatch); |
| 2346 |
return err; |
| 2347 |
} |
| 2348 |
|
| 2349 |
static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd) |
| 2350 |
{ |
| 2351 |
struct inode *inode = mpd->inode; |
| 2352 |
struct ext4_map_blocks *map = &mpd->map; |
| 2353 |
int get_blocks_flags; |
| 2354 |
int err, dioread_nolock; |
| 2355 |
|
| 2356 |
/* Make sure transaction has enough credits for this extent */ |
| 2357 |
err = ext4_journal_ensure_extent_credits(handle, inode); |
| 2358 |
if (err < 0) |
| 2359 |
return err; |
| 2360 |
|
| 2361 |
trace_ext4_da_write_pages_extent(inode, map); |
| 2362 |
/* |
| 2363 |
* Call ext4_map_blocks() to allocate any delayed allocation blocks, or |
| 2364 |
* to convert an unwritten extent to be initialized (in the case |
| 2365 |
* where we have written into one or more preallocated blocks). It is |
| 2366 |
* possible that we're going to need more metadata blocks than |
| 2367 |
* previously reserved. However we must not fail because we're in |
| 2368 |
* writeback and there is nothing we can do about it so it might result |
| 2369 |
* in data loss. So use reserved blocks to allocate metadata if |
| 2370 |
* possible. In addition, do not cache any unrelated extents, as it |
| 2371 |
* only holds the folio lock but does not hold the i_rwsem or |
| 2372 |
* invalidate_lock, which could corrupt the extent status tree. |
| 2373 |
*/ |
| 2374 |
get_blocks_flags = EXT4_GET_BLOCKS_CREATE | |
| 2375 |
EXT4_GET_BLOCKS_METADATA_NOFAIL | |
| 2376 |
EXT4_GET_BLOCKS_IO_SUBMIT | |
| 2377 |
EXT4_EX_NOCACHE; |
| 2378 |
|
| 2379 |
dioread_nolock = ext4_should_dioread_nolock(inode); |
| 2380 |
if (dioread_nolock) |
| 2381 |
get_blocks_flags |= EXT4_GET_BLOCKS_UNWRIT_EXT; |
| 2382 |
|
| 2383 |
err = ext4_map_blocks(handle, inode, map, get_blocks_flags); |
| 2384 |
if (err < 0) |
| 2385 |
return err; |
| 2386 |
if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) { |
| 2387 |
if (!mpd->io_submit.io_end->handle && |
| 2388 |
ext4_handle_valid(handle)) { |
| 2389 |
mpd->io_submit.io_end->handle = handle->h_rsv_handle; |
| 2390 |
handle->h_rsv_handle = NULL; |
| 2391 |
} |
| 2392 |
ext4_set_io_unwritten_flag(mpd->io_submit.io_end); |
| 2393 |
} |
| 2394 |
|
| 2395 |
BUG_ON(map->m_len == 0); |
| 2396 |
return 0; |
| 2397 |
} |
| 2398 |
|
| 2399 |
/* |
| 2400 |
* This is used to submit mapped buffers in a single folio that is not fully |
| 2401 |
* mapped for various reasons, such as insufficient space or journal credits. |
| 2402 |
*/ |
| 2403 |
static int mpage_submit_partial_folio(struct mpage_da_data *mpd) |
| 2404 |
{ |
| 2405 |
struct inode *inode = mpd->inode; |
| 2406 |
struct folio *folio; |
| 2407 |
loff_t pos; |
| 2408 |
int ret; |
| 2409 |
|
| 2410 |
folio = filemap_get_folio(inode->i_mapping, |
| 2411 |
mpd->start_pos >> PAGE_SHIFT); |
| 2412 |
if (IS_ERR(folio)) |
| 2413 |
return PTR_ERR(folio); |
| 2414 |
/* |
| 2415 |
* The mapped position should be within the current processing folio |
| 2416 |
* but must not be the folio start position. |
| 2417 |
*/ |
| 2418 |
pos = ((loff_t)mpd->map.m_lblk) << inode->i_blkbits; |
| 2419 |
if (WARN_ON_ONCE((folio_pos(folio) == pos) || |
| 2420 |
!folio_contains(folio, pos >> PAGE_SHIFT))) |
| 2421 |
return -EINVAL; |
| 2422 |
|
| 2423 |
ret = mpage_submit_folio(mpd, folio); |
| 2424 |
if (ret) |
| 2425 |
goto out; |
| 2426 |
/* |
| 2427 |
* Update start_pos to prevent this folio from being released in |
| 2428 |
* mpage_release_unused_pages(), it will be reset to the aligned folio |
| 2429 |
* pos when this folio is written again in the next round. Additionally, |
| 2430 |
* do not update wbc->nr_to_write here, as it will be updated once the |
| 2431 |
* entire folio has finished processing. |
| 2432 |
*/ |
| 2433 |
mpd->start_pos = pos; |
| 2434 |
out: |
| 2435 |
folio_unlock(folio); |
| 2436 |
folio_put(folio); |
| 2437 |
return ret; |
| 2438 |
} |
| 2439 |
|
| 2440 |
/* |
| 2441 |
* mpage_map_and_submit_extent - map extent starting at mpd->lblk of length |
| 2442 |
* mpd->len and submit pages underlying it for IO |
| 2443 |
* |
| 2444 |
* @handle - handle for journal operations |
| 2445 |
* @mpd - extent to map |
| 2446 |
* @give_up_on_write - we set this to true iff there is a fatal error and there |
| 2447 |
* is no hope of writing the data. The caller should discard |
| 2448 |
* dirty pages to avoid infinite loops. |
| 2449 |
* |
| 2450 |
* The function maps extent starting at mpd->lblk of length mpd->len. If it is |
| 2451 |
* delayed, blocks are allocated, if it is unwritten, we may need to convert |
| 2452 |
* them to initialized or split the described range from larger unwritten |
| 2453 |
* extent. Note that we need not map all the described range since allocation |
| 2454 |
* can return less blocks or the range is covered by more unwritten extents. We |
| 2455 |
* cannot map more because we are limited by reserved transaction credits. On |
| 2456 |
* the other hand we always make sure that the last touched page is fully |
| 2457 |
* mapped so that it can be written out (and thus forward progress is |
| 2458 |
* guaranteed). After mapping we submit all mapped pages for IO. |
| 2459 |
*/ |
| 2460 |
static int mpage_map_and_submit_extent(handle_t *handle, |
| 2461 |
struct mpage_da_data *mpd, |
| 2462 |
bool *give_up_on_write) |
| 2463 |
{ |
| 2464 |
struct inode *inode = mpd->inode; |
| 2465 |
struct ext4_map_blocks *map = &mpd->map; |
| 2466 |
int err; |
| 2467 |
loff_t disksize; |
| 2468 |
int progress = 0; |
| 2469 |
ext4_io_end_t *io_end = mpd->io_submit.io_end; |
| 2470 |
struct ext4_io_end_vec *io_end_vec; |
| 2471 |
|
| 2472 |
io_end_vec = ext4_alloc_io_end_vec(io_end); |
| 2473 |
if (IS_ERR(io_end_vec)) |
| 2474 |
return PTR_ERR(io_end_vec); |
| 2475 |
io_end_vec->offset = EXT4_LBLK_TO_B(inode, map->m_lblk); |
| 2476 |
do { |
| 2477 |
err = mpage_map_one_extent(handle, mpd); |
| 2478 |
if (err < 0) { |
| 2479 |
struct super_block *sb = inode->i_sb; |
| 2480 |
|
| 2481 |
if (ext4_emergency_state(sb)) |
| 2482 |
goto invalidate_dirty_pages; |
| 2483 |
/* |
| 2484 |
* Let the uper layers retry transient errors. |
| 2485 |
* In the case of ENOSPC, if ext4_count_free_blocks() |
| 2486 |
* is non-zero, a commit should free up blocks. |
| 2487 |
*/ |
| 2488 |
if ((err == -ENOMEM) || (err == -EAGAIN) || |
| 2489 |
(err == -ENOSPC && ext4_count_free_clusters(sb))) { |
| 2490 |
/* |
| 2491 |
* We may have already allocated extents for |
| 2492 |
* some bhs inside the folio, issue the |
| 2493 |
* corresponding data to prevent stale data. |
| 2494 |
*/ |
| 2495 |
if (progress) { |
| 2496 |
if (mpage_submit_partial_folio(mpd)) |
| 2497 |
goto invalidate_dirty_pages; |
| 2498 |
goto update_disksize; |
| 2499 |
} |
| 2500 |
return err; |
| 2501 |
} |
| 2502 |
ext4_msg(sb, KERN_CRIT, |
| 2503 |
"Delayed block allocation failed for " |
| 2504 |
"inode %lu at logical offset %llu with" |
| 2505 |
" max blocks %u with error %d", |
| 2506 |
inode->i_ino, |
| 2507 |
(unsigned long long)map->m_lblk, |
| 2508 |
(unsigned)map->m_len, -err); |
| 2509 |
ext4_msg(sb, KERN_CRIT, |
| 2510 |
"This should not happen!! Data will " |
| 2511 |
"be lost\n"); |
| 2512 |
if (err == -ENOSPC) |
| 2513 |
ext4_print_free_blocks(inode); |
| 2514 |
invalidate_dirty_pages: |
| 2515 |
*give_up_on_write = true; |
| 2516 |
return err; |
| 2517 |
} |
| 2518 |
progress = 1; |
| 2519 |
/* |
| 2520 |
* Update buffer state, submit mapped pages, and get us new |
| 2521 |
* extent to map |
| 2522 |
*/ |
| 2523 |
err = mpage_map_and_submit_buffers(mpd); |
| 2524 |
if (err < 0) |
| 2525 |
goto update_disksize; |
| 2526 |
} while (map->m_len); |
| 2527 |
|
| 2528 |
update_disksize: |
| 2529 |
/* |
| 2530 |
* Update on-disk size after IO is submitted. Races with |
| 2531 |
* truncate are avoided by checking i_size under i_data_sem. |
| 2532 |
*/ |
| 2533 |
disksize = mpd->start_pos; |
| 2534 |
if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) { |
| 2535 |
int err2; |
| 2536 |
loff_t i_size; |
| 2537 |
|
| 2538 |
down_write(&EXT4_I(inode)->i_data_sem); |
| 2539 |
i_size = i_size_read(inode); |
| 2540 |
if (disksize > i_size) |
| 2541 |
disksize = i_size; |
| 2542 |
if (disksize > EXT4_I(inode)->i_disksize) |
| 2543 |
EXT4_I(inode)->i_disksize = disksize; |
| 2544 |
up_write(&EXT4_I(inode)->i_data_sem); |
| 2545 |
err2 = ext4_mark_inode_dirty(handle, inode); |
| 2546 |
if (err2) { |
| 2547 |
ext4_error_err(inode->i_sb, -err2, |
| 2548 |
"Failed to mark inode %lu dirty", |
| 2549 |
inode->i_ino); |
| 2550 |
} |
| 2551 |
if (!err) |
| 2552 |
err = err2; |
| 2553 |
} |
| 2554 |
return err; |
| 2555 |
} |
| 2556 |
|
| 2557 |
static int ext4_journal_folio_buffers(handle_t *handle, struct folio *folio, |
| 2558 |
size_t len) |
| 2559 |
{ |
| 2560 |
struct buffer_head *page_bufs = folio_buffers(folio); |
| 2561 |
struct inode *inode = folio->mapping->host; |
| 2562 |
int ret, err; |
| 2563 |
|
| 2564 |
ret = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len, |
| 2565 |
NULL, do_journal_get_write_access); |
| 2566 |
err = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len, |
| 2567 |
NULL, write_end_fn); |
| 2568 |
if (ret == 0) |
| 2569 |
ret = err; |
| 2570 |
err = ext4_jbd2_inode_add_write(handle, inode, folio_pos(folio), len); |
| 2571 |
if (ret == 0) |
| 2572 |
ret = err; |
| 2573 |
EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
| 2574 |
|
| 2575 |
return ret; |
| 2576 |
} |
| 2577 |
|
| 2578 |
static int mpage_journal_page_buffers(handle_t *handle, |
| 2579 |
struct mpage_da_data *mpd, |
| 2580 |
struct folio *folio) |
| 2581 |
{ |
| 2582 |
struct inode *inode = mpd->inode; |
| 2583 |
loff_t size = i_size_read(inode); |
| 2584 |
size_t len = folio_size(folio); |
| 2585 |
|
| 2586 |
folio_clear_checked(folio); |
| 2587 |
mpd->wbc->nr_to_write -= folio_nr_pages(folio); |
| 2588 |
|
| 2589 |
if (folio_pos(folio) + len > size && |
| 2590 |
!ext4_verity_in_progress(inode)) |
| 2591 |
len = size & (len - 1); |
| 2592 |
|
| 2593 |
return ext4_journal_folio_buffers(handle, folio, len); |
| 2594 |
} |
| 2595 |
|
| 2596 |
/* |
| 2597 |
* mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages |
| 2598 |
* needing mapping, submit mapped pages |
| 2599 |
* |
| 2600 |
* @mpd - where to look for pages |
| 2601 |
* |
| 2602 |
* Walk dirty pages in the mapping. If they are fully mapped, submit them for |
| 2603 |
* IO immediately. If we cannot map blocks, we submit just already mapped |
| 2604 |
* buffers in the page for IO and keep page dirty. When we can map blocks and |
| 2605 |
* we find a page which isn't mapped we start accumulating extent of buffers |
| 2606 |
* underlying these pages that needs mapping (formed by either delayed or |
| 2607 |
* unwritten buffers). We also lock the pages containing these buffers. The |
| 2608 |
* extent found is returned in @mpd structure (starting at mpd->lblk with |
| 2609 |
* length mpd->len blocks). |
| 2610 |
* |
| 2611 |
* Note that this function can attach bios to one io_end structure which are |
| 2612 |
* neither logically nor physically contiguous. Although it may seem as an |
| 2613 |
* unnecessary complication, it is actually inevitable in blocksize < pagesize |
| 2614 |
* case as we need to track IO to all buffers underlying a page in one io_end. |
| 2615 |
*/ |
| 2616 |
static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd) |
| 2617 |
{ |
| 2618 |
struct address_space *mapping = mpd->inode->i_mapping; |
| 2619 |
struct folio_batch fbatch; |
| 2620 |
unsigned int nr_folios; |
| 2621 |
pgoff_t index = mpd->start_pos >> PAGE_SHIFT; |
| 2622 |
pgoff_t end = mpd->end_pos >> PAGE_SHIFT; |
| 2623 |
xa_mark_t tag; |
| 2624 |
int i, err = 0; |
| 2625 |
ext4_lblk_t lblk; |
| 2626 |
struct buffer_head *head; |
| 2627 |
handle_t *handle = NULL; |
| 2628 |
int bpp = ext4_journal_blocks_per_folio(mpd->inode); |
| 2629 |
|
| 2630 |
tag = wbc_to_tag(mpd->wbc); |
| 2631 |
|
| 2632 |
mpd->map.m_len = 0; |
| 2633 |
mpd->next_pos = mpd->start_pos; |
| 2634 |
if (ext4_should_journal_data(mpd->inode)) { |
| 2635 |
handle = ext4_journal_start(mpd->inode, EXT4_HT_WRITE_PAGE, |
| 2636 |
bpp); |
| 2637 |
if (IS_ERR(handle)) |
| 2638 |
return PTR_ERR(handle); |
| 2639 |
} |
| 2640 |
folio_batch_init(&fbatch); |
| 2641 |
while (index <= end) { |
| 2642 |
nr_folios = filemap_get_folios_tag(mapping, &index, end, |
| 2643 |
tag, &fbatch); |
| 2644 |
if (nr_folios == 0) |
| 2645 |
break; |
| 2646 |
|
| 2647 |
for (i = 0; i < nr_folios; i++) { |
| 2648 |
struct folio *folio = fbatch.folios[i]; |
| 2649 |
|
| 2650 |
/* |
| 2651 |
* Accumulated enough dirty pages? This doesn't apply |
| 2652 |
* to WB_SYNC_ALL mode. For integrity sync we have to |
| 2653 |
* keep going because someone may be concurrently |
| 2654 |
* dirtying pages, and we might have synced a lot of |
| 2655 |
* newly appeared dirty pages, but have not synced all |
| 2656 |
* of the old dirty pages. |
| 2657 |
*/ |
| 2658 |
if (mpd->wbc->sync_mode == WB_SYNC_NONE && |
| 2659 |
mpd->wbc->nr_to_write <= |
| 2660 |
EXT4_LBLK_TO_PG(mpd->inode, mpd->map.m_len)) |
| 2661 |
goto out; |
| 2662 |
|
| 2663 |
/* If we can't merge this page, we are done. */ |
| 2664 |
if (mpd->map.m_len > 0 && |
| 2665 |
mpd->next_pos != folio_pos(folio)) |
| 2666 |
goto out; |
| 2667 |
|
| 2668 |
if (handle) { |
| 2669 |
err = ext4_journal_ensure_credits(handle, bpp, |
| 2670 |
0); |
| 2671 |
if (err < 0) |
| 2672 |
goto out; |
| 2673 |
} |
| 2674 |
|
| 2675 |
folio_lock(folio); |
| 2676 |
/* |
| 2677 |
* If the page is no longer dirty, or its mapping no |
| 2678 |
* longer corresponds to inode we are writing (which |
| 2679 |
* means it has been truncated or invalidated), or the |
| 2680 |
* page is already under writeback and we are not doing |
| 2681 |
* a data integrity writeback, skip the page |
| 2682 |
*/ |
| 2683 |
if (!folio_test_dirty(folio) || |
| 2684 |
(folio_test_writeback(folio) && |
| 2685 |
(mpd->wbc->sync_mode == WB_SYNC_NONE)) || |
| 2686 |
unlikely(folio->mapping != mapping)) { |
| 2687 |
folio_unlock(folio); |
| 2688 |
continue; |
| 2689 |
} |
| 2690 |
|
| 2691 |
folio_wait_writeback(folio); |
| 2692 |
BUG_ON(folio_test_writeback(folio)); |
| 2693 |
|
| 2694 |
/* |
| 2695 |
* Should never happen but for buggy code in |
| 2696 |
* other subsystems that call |
| 2697 |
* set_page_dirty() without properly warning |
| 2698 |
* the file system first. See [1] for more |
| 2699 |
* information. |
| 2700 |
* |
| 2701 |
* [1] https://lore.kernel.org/linux-mm/[email protected] |
| 2702 |
*/ |
| 2703 |
if (!folio_buffers(folio)) { |
| 2704 |
ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", folio->index); |
| 2705 |
folio_clear_dirty(folio); |
| 2706 |
folio_unlock(folio); |
| 2707 |
continue; |
| 2708 |
} |
| 2709 |
|
| 2710 |
if (mpd->map.m_len == 0) |
| 2711 |
mpd->start_pos = folio_pos(folio); |
| 2712 |
mpd->next_pos = folio_next_pos(folio); |
| 2713 |
/* |
| 2714 |
* Writeout when we cannot modify metadata is simple. |
| 2715 |
* Just submit the page. For data=journal mode we |
| 2716 |
* first handle writeout of the page for checkpoint and |
| 2717 |
* only after that handle delayed page dirtying. This |
| 2718 |
* makes sure current data is checkpointed to the final |
| 2719 |
* location before possibly journalling it again which |
| 2720 |
* is desirable when the page is frequently dirtied |
| 2721 |
* through a pin. |
| 2722 |
*/ |
| 2723 |
if (!mpd->can_map) { |
| 2724 |
err = mpage_submit_folio(mpd, folio); |
| 2725 |
if (err < 0) |
| 2726 |
goto out; |
| 2727 |
/* Pending dirtying of journalled data? */ |
| 2728 |
if (folio_test_checked(folio)) { |
| 2729 |
err = mpage_journal_page_buffers(handle, |
| 2730 |
mpd, folio); |
| 2731 |
if (err < 0) |
| 2732 |
goto out; |
| 2733 |
mpd->journalled_more_data = 1; |
| 2734 |
} |
| 2735 |
mpage_folio_done(mpd, folio); |
| 2736 |
} else { |
| 2737 |
/* Add all dirty buffers to mpd */ |
| 2738 |
lblk = EXT4_PG_TO_LBLK(mpd->inode, folio->index); |
| 2739 |
head = folio_buffers(folio); |
| 2740 |
err = mpage_process_page_bufs(mpd, head, head, |
| 2741 |
lblk); |
| 2742 |
if (err <= 0) |
| 2743 |
goto out; |
| 2744 |
err = 0; |
| 2745 |
} |
| 2746 |
} |
| 2747 |
folio_batch_release(&fbatch); |
| 2748 |
cond_resched(); |
| 2749 |
} |
| 2750 |
mpd->scanned_until_end = 1; |
| 2751 |
if (handle) |
| 2752 |
ext4_journal_stop(handle); |
| 2753 |
return 0; |
| 2754 |
out: |
| 2755 |
folio_batch_release(&fbatch); |
| 2756 |
if (handle) |
| 2757 |
ext4_journal_stop(handle); |
| 2758 |
return err; |
| 2759 |
} |
| 2760 |
|
| 2761 |
static int ext4_do_writepages(struct mpage_da_data *mpd) |
| 2762 |
{ |
| 2763 |
struct writeback_control *wbc = mpd->wbc; |
| 2764 |
pgoff_t writeback_index = 0; |
| 2765 |
long nr_to_write = wbc->nr_to_write; |
| 2766 |
int range_whole = 0; |
| 2767 |
int cycled = 1; |
| 2768 |
handle_t *handle = NULL; |
| 2769 |
struct inode *inode = mpd->inode; |
| 2770 |
struct address_space *mapping = inode->i_mapping; |
| 2771 |
int needed_blocks, rsv_blocks = 0, ret = 0; |
| 2772 |
struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
| 2773 |
struct blk_plug plug; |
| 2774 |
bool give_up_on_write = false; |
| 2775 |
|
| 2776 |
trace_ext4_writepages(inode, wbc); |
| 2777 |
|
| 2778 |
/* |
| 2779 |
* No pages to write? This is mainly a kludge to avoid starting |
| 2780 |
* a transaction for special inodes like journal inode on last iput() |
| 2781 |
* because that could violate lock ordering on umount |
| 2782 |
*/ |
| 2783 |
if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
| 2784 |
goto out_writepages; |
| 2785 |
|
| 2786 |
/* |
| 2787 |
* If the filesystem has aborted, it is read-only, so return |
| 2788 |
* right away instead of dumping stack traces later on that |
| 2789 |
* will obscure the real source of the problem. We test |
| 2790 |
* fs shutdown state instead of sb->s_flag's SB_RDONLY because |
| 2791 |
* the latter could be true if the filesystem is mounted |
| 2792 |
* read-only, and in that case, ext4_writepages should |
| 2793 |
* *never* be called, so if that ever happens, we would want |
| 2794 |
* the stack trace. |
| 2795 |
*/ |
| 2796 |
ret = ext4_emergency_state(mapping->host->i_sb); |
| 2797 |
if (unlikely(ret)) |
| 2798 |
goto out_writepages; |
| 2799 |
|
| 2800 |
/* |
| 2801 |
* If we have inline data and arrive here, it means that |
| 2802 |
* we will soon create the block for the 1st page, so |
| 2803 |
* we'd better clear the inline data here. |
| 2804 |
*/ |
| 2805 |
if (ext4_has_inline_data(inode)) { |
| 2806 |
/* Just inode will be modified... */ |
| 2807 |
handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
| 2808 |
if (IS_ERR(handle)) { |
| 2809 |
ret = PTR_ERR(handle); |
| 2810 |
goto out_writepages; |
| 2811 |
} |
| 2812 |
BUG_ON(ext4_test_inode_state(inode, |
| 2813 |
EXT4_STATE_MAY_INLINE_DATA)); |
| 2814 |
ext4_destroy_inline_data(handle, inode); |
| 2815 |
ext4_journal_stop(handle); |
| 2816 |
} |
| 2817 |
|
| 2818 |
/* |
| 2819 |
* data=journal mode does not do delalloc so we just need to writeout / |
| 2820 |
* journal already mapped buffers. On the other hand we need to commit |
| 2821 |
* transaction to make data stable. We expect all the data to be |
| 2822 |
* already in the journal (the only exception are DMA pinned pages |
| 2823 |
* dirtied behind our back) so we commit transaction here and run the |
| 2824 |
* writeback loop to checkpoint them. The checkpointing is not actually |
| 2825 |
* necessary to make data persistent *but* quite a few places (extent |
| 2826 |
* shifting operations, fsverity, ...) depend on being able to drop |
| 2827 |
* pagecache pages after calling filemap_write_and_wait() and for that |
| 2828 |
* checkpointing needs to happen. |
| 2829 |
*/ |
| 2830 |
if (ext4_should_journal_data(inode)) { |
| 2831 |
mpd->can_map = 0; |
| 2832 |
if (wbc->sync_mode == WB_SYNC_ALL) |
| 2833 |
ext4_fc_commit(sbi->s_journal, |
| 2834 |
EXT4_I(inode)->i_datasync_tid); |
| 2835 |
} |
| 2836 |
mpd->journalled_more_data = 0; |
| 2837 |
|
| 2838 |
if (ext4_should_dioread_nolock(inode)) { |
| 2839 |
int bpf = ext4_journal_blocks_per_folio(inode); |
| 2840 |
/* |
| 2841 |
* We may need to convert up to one extent per block in |
| 2842 |
* the folio and we may dirty the inode. |
| 2843 |
*/ |
| 2844 |
rsv_blocks = 1 + ext4_ext_index_trans_blocks(inode, bpf); |
| 2845 |
} |
| 2846 |
|
| 2847 |
if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
| 2848 |
range_whole = 1; |
| 2849 |
|
| 2850 |
if (wbc->range_cyclic) { |
| 2851 |
writeback_index = mapping->writeback_index; |
| 2852 |
if (writeback_index) |
| 2853 |
cycled = 0; |
| 2854 |
mpd->start_pos = writeback_index << PAGE_SHIFT; |
| 2855 |
mpd->end_pos = LLONG_MAX; |
| 2856 |
} else { |
| 2857 |
mpd->start_pos = wbc->range_start; |
| 2858 |
mpd->end_pos = wbc->range_end; |
| 2859 |
} |
| 2860 |
|
| 2861 |
ext4_io_submit_init(&mpd->io_submit, wbc); |
| 2862 |
retry: |
| 2863 |
if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
| 2864 |
tag_pages_for_writeback(mapping, mpd->start_pos >> PAGE_SHIFT, |
| 2865 |
mpd->end_pos >> PAGE_SHIFT); |
| 2866 |
blk_start_plug(&plug); |
| 2867 |
|
| 2868 |
/* |
| 2869 |
* First writeback pages that don't need mapping - we can avoid |
| 2870 |
* starting a transaction unnecessarily and also avoid being blocked |
| 2871 |
* in the block layer on device congestion while having transaction |
| 2872 |
* started. |
| 2873 |
*/ |
| 2874 |
mpd->do_map = 0; |
| 2875 |
mpd->scanned_until_end = 0; |
| 2876 |
mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); |
| 2877 |
if (!mpd->io_submit.io_end) { |
| 2878 |
ret = -ENOMEM; |
| 2879 |
goto unplug; |
| 2880 |
} |
| 2881 |
ret = mpage_prepare_extent_to_map(mpd); |
| 2882 |
/* Unlock pages we didn't use */ |
| 2883 |
mpage_release_unused_pages(mpd, false); |
| 2884 |
/* Submit prepared bio */ |
| 2885 |
ext4_io_submit(&mpd->io_submit); |
| 2886 |
ext4_put_io_end_defer(mpd->io_submit.io_end); |
| 2887 |
mpd->io_submit.io_end = NULL; |
| 2888 |
if (ret < 0) |
| 2889 |
goto unplug; |
| 2890 |
|
| 2891 |
while (!mpd->scanned_until_end && wbc->nr_to_write > 0) { |
| 2892 |
/* For each extent of pages we use new io_end */ |
| 2893 |
mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); |
| 2894 |
if (!mpd->io_submit.io_end) { |
| 2895 |
ret = -ENOMEM; |
| 2896 |
break; |
| 2897 |
} |
| 2898 |
|
| 2899 |
WARN_ON_ONCE(!mpd->can_map); |
| 2900 |
/* |
| 2901 |
* We have two constraints: We find one extent to map and we |
| 2902 |
* must always write out whole page (makes a difference when |
| 2903 |
* blocksize < pagesize) so that we don't block on IO when we |
| 2904 |
* try to write out the rest of the page. Journalled mode is |
| 2905 |
* not supported by delalloc. |
| 2906 |
*/ |
| 2907 |
BUG_ON(ext4_should_journal_data(inode)); |
| 2908 |
/* |
| 2909 |
* Calculate the number of credits needed to reserve for one |
| 2910 |
* extent of up to MAX_WRITEPAGES_EXTENT_LEN blocks. It will |
| 2911 |
* attempt to extend the transaction or start a new iteration |
| 2912 |
* if the reserved credits are insufficient. |
| 2913 |
*/ |
| 2914 |
needed_blocks = ext4_chunk_trans_blocks(inode, |
| 2915 |
MAX_WRITEPAGES_EXTENT_LEN); |
| 2916 |
/* start a new transaction */ |
| 2917 |
handle = ext4_journal_start_with_reserve(inode, |
| 2918 |
EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks); |
| 2919 |
if (IS_ERR(handle)) { |
| 2920 |
ret = PTR_ERR(handle); |
| 2921 |
ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
| 2922 |
"%ld pages, ino %lu; err %d", __func__, |
| 2923 |
wbc->nr_to_write, inode->i_ino, ret); |
| 2924 |
/* Release allocated io_end */ |
| 2925 |
ext4_put_io_end(mpd->io_submit.io_end); |
| 2926 |
mpd->io_submit.io_end = NULL; |
| 2927 |
break; |
| 2928 |
} |
| 2929 |
mpd->do_map = 1; |
| 2930 |
|
| 2931 |
trace_ext4_da_write_folios_start(inode, mpd->start_pos, |
| 2932 |
mpd->next_pos, wbc); |
| 2933 |
ret = mpage_prepare_extent_to_map(mpd); |
| 2934 |
if (!ret && mpd->map.m_len) |
| 2935 |
ret = mpage_map_and_submit_extent(handle, mpd, |
| 2936 |
&give_up_on_write); |
| 2937 |
/* |
| 2938 |
* Caution: If the handle is synchronous, |
| 2939 |
* ext4_journal_stop() can wait for transaction commit |
| 2940 |
* to finish which may depend on writeback of pages to |
| 2941 |
* complete or on page lock to be released. In that |
| 2942 |
* case, we have to wait until after we have |
| 2943 |
* submitted all the IO, released page locks we hold, |
| 2944 |
* and dropped io_end reference (for extent conversion |
| 2945 |
* to be able to complete) before stopping the handle. |
| 2946 |
*/ |
| 2947 |
if (!ext4_handle_valid(handle) || handle->h_sync == 0) { |
| 2948 |
ext4_journal_stop(handle); |
| 2949 |
handle = NULL; |
| 2950 |
mpd->do_map = 0; |
| 2951 |
} |
| 2952 |
/* Unlock pages we didn't use */ |
| 2953 |
mpage_release_unused_pages(mpd, give_up_on_write); |
| 2954 |
/* Submit prepared bio */ |
| 2955 |
ext4_io_submit(&mpd->io_submit); |
| 2956 |
|
| 2957 |
/* |
| 2958 |
* Drop our io_end reference we got from init. We have |
| 2959 |
* to be careful and use deferred io_end finishing if |
| 2960 |
* we are still holding the transaction as we can |
| 2961 |
* release the last reference to io_end which may end |
| 2962 |
* up doing unwritten extent conversion. |
| 2963 |
*/ |
| 2964 |
if (handle) { |
| 2965 |
ext4_put_io_end_defer(mpd->io_submit.io_end); |
| 2966 |
ext4_journal_stop(handle); |
| 2967 |
} else |
| 2968 |
ext4_put_io_end(mpd->io_submit.io_end); |
| 2969 |
mpd->io_submit.io_end = NULL; |
| 2970 |
trace_ext4_da_write_folios_end(inode, mpd->start_pos, |
| 2971 |
mpd->next_pos, wbc, ret); |
| 2972 |
|
| 2973 |
if (ret == -ENOSPC && sbi->s_journal) { |
| 2974 |
/* |
| 2975 |
* Commit the transaction which would |
| 2976 |
* free blocks released in the transaction |
| 2977 |
* and try again |
| 2978 |
*/ |
| 2979 |
jbd2_journal_force_commit_nested(sbi->s_journal); |
| 2980 |
ret = 0; |
| 2981 |
continue; |
| 2982 |
} |
| 2983 |
if (ret == -EAGAIN) |
| 2984 |
ret = 0; |
| 2985 |
/* Fatal error - ENOMEM, EIO... */ |
| 2986 |
if (ret) |
| 2987 |
break; |
| 2988 |
} |
| 2989 |
unplug: |
| 2990 |
blk_finish_plug(&plug); |
| 2991 |
if (!ret && !cycled && wbc->nr_to_write > 0) { |
| 2992 |
cycled = 1; |
| 2993 |
mpd->end_pos = (writeback_index << PAGE_SHIFT) - 1; |
| 2994 |
mpd->start_pos = 0; |
| 2995 |
goto retry; |
| 2996 |
} |
| 2997 |
|
| 2998 |
/* Update index */ |
| 2999 |
if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
| 3000 |
/* |
| 3001 |
* Set the writeback_index so that range_cyclic |
| 3002 |
* mode will write it back later |
| 3003 |
*/ |
| 3004 |
mapping->writeback_index = mpd->start_pos >> PAGE_SHIFT; |
| 3005 |
|
| 3006 |
out_writepages: |
| 3007 |
trace_ext4_writepages_result(inode, wbc, ret, |
| 3008 |
nr_to_write - wbc->nr_to_write); |
| 3009 |
return ret; |
| 3010 |
} |
| 3011 |
|
| 3012 |
static int ext4_writepages(struct address_space *mapping, |
| 3013 |
struct writeback_control *wbc) |
| 3014 |
{ |
| 3015 |
struct super_block *sb = mapping->host->i_sb; |
| 3016 |
struct mpage_da_data mpd = { |
| 3017 |
.inode = mapping->host, |
| 3018 |
.wbc = wbc, |
| 3019 |
.can_map = 1, |
| 3020 |
}; |
| 3021 |
int ret; |
| 3022 |
int alloc_ctx; |
| 3023 |
|
| 3024 |
ret = ext4_emergency_state(sb); |
| 3025 |
if (unlikely(ret)) |
| 3026 |
return ret; |
| 3027 |
|
| 3028 |
alloc_ctx = ext4_writepages_down_read(sb); |
| 3029 |
ret = ext4_do_writepages(&mpd); |
| 3030 |
/* |
| 3031 |
* For data=journal writeback we could have come across pages marked |
| 3032 |
* for delayed dirtying (PageChecked) which were just added to the |
| 3033 |
* running transaction. Try once more to get them to stable storage. |
| 3034 |
*/ |
| 3035 |
if (!ret && mpd.journalled_more_data) |
| 3036 |
ret = ext4_do_writepages(&mpd); |
| 3037 |
ext4_writepages_up_read(sb, alloc_ctx); |
| 3038 |
|
| 3039 |
return ret; |
| 3040 |
} |
| 3041 |
|
| 3042 |
int ext4_normal_submit_inode_data_buffers(struct jbd2_inode *jinode) |
| 3043 |
{ |
| 3044 |
struct writeback_control wbc = { |
| 3045 |
.sync_mode = WB_SYNC_ALL, |
| 3046 |
.nr_to_write = LONG_MAX, |
| 3047 |
.range_start = jinode->i_dirty_start, |
| 3048 |
.range_end = jinode->i_dirty_end, |
| 3049 |
}; |
| 3050 |
struct mpage_da_data mpd = { |
| 3051 |
.inode = jinode->i_vfs_inode, |
| 3052 |
.wbc = &wbc, |
| 3053 |
.can_map = 0, |
| 3054 |
}; |
| 3055 |
return ext4_do_writepages(&mpd); |
| 3056 |
} |
| 3057 |
|
| 3058 |
static int ext4_dax_writepages(struct address_space *mapping, |
| 3059 |
struct writeback_control *wbc) |
| 3060 |
{ |
| 3061 |
int ret; |
| 3062 |
long nr_to_write = wbc->nr_to_write; |
| 3063 |
struct inode *inode = mapping->host; |
| 3064 |
int alloc_ctx; |
| 3065 |
|
| 3066 |
ret = ext4_emergency_state(inode->i_sb); |
| 3067 |
if (unlikely(ret)) |
| 3068 |
return ret; |
| 3069 |
|
| 3070 |
alloc_ctx = ext4_writepages_down_read(inode->i_sb); |
| 3071 |
trace_ext4_writepages(inode, wbc); |
| 3072 |
|
| 3073 |
ret = dax_writeback_mapping_range(mapping, |
| 3074 |
EXT4_SB(inode->i_sb)->s_daxdev, wbc); |
| 3075 |
trace_ext4_writepages_result(inode, wbc, ret, |
| 3076 |
nr_to_write - wbc->nr_to_write); |
| 3077 |
ext4_writepages_up_read(inode->i_sb, alloc_ctx); |
| 3078 |
return ret; |
| 3079 |
} |
| 3080 |
|
| 3081 |
static int ext4_nonda_switch(struct super_block *sb) |
| 3082 |
{ |
| 3083 |
s64 free_clusters, dirty_clusters; |
| 3084 |
struct ext4_sb_info *sbi = EXT4_SB(sb); |
| 3085 |
|
| 3086 |
/* |
| 3087 |
* switch to non delalloc mode if we are running low |
| 3088 |
* on free block. The free block accounting via percpu |
| 3089 |
* counters can get slightly wrong with percpu_counter_batch getting |
| 3090 |
* accumulated on each CPU without updating global counters |
| 3091 |
* Delalloc need an accurate free block accounting. So switch |
| 3092 |
* to non delalloc when we are near to error range. |
| 3093 |
*/ |
| 3094 |
free_clusters = |
| 3095 |
percpu_counter_read_positive(&sbi->s_freeclusters_counter); |
| 3096 |
dirty_clusters = |
| 3097 |
percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); |
| 3098 |
/* |
| 3099 |
* Start pushing delalloc when 1/2 of free blocks are dirty. |
| 3100 |
*/ |
| 3101 |
if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
| 3102 |
try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
| 3103 |
|
| 3104 |
if (2 * free_clusters < 3 * dirty_clusters || |
| 3105 |
free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { |
| 3106 |
/* |
| 3107 |
* free block count is less than 150% of dirty blocks |
| 3108 |
* or free blocks is less than watermark |
| 3109 |
*/ |
| 3110 |
return 1; |
| 3111 |
} |
| 3112 |
return 0; |
| 3113 |
} |
| 3114 |
|
| 3115 |
static int ext4_da_write_begin(const struct kiocb *iocb, |
| 3116 |
struct address_space *mapping, |
| 3117 |
loff_t pos, unsigned len, |
| 3118 |
struct folio **foliop, void **fsdata) |
| 3119 |
{ |
| 3120 |
int ret, retries = 0; |
| 3121 |
struct folio *folio; |
| 3122 |
pgoff_t index; |
| 3123 |
struct inode *inode = mapping->host; |
| 3124 |
|
| 3125 |
ret = ext4_emergency_state(inode->i_sb); |
| 3126 |
if (unlikely(ret)) |
| 3127 |
return ret; |
| 3128 |
|
| 3129 |
index = pos >> PAGE_SHIFT; |
| 3130 |
|
| 3131 |
if (ext4_nonda_switch(inode->i_sb) || ext4_verity_in_progress(inode)) { |
| 3132 |
*fsdata = (void *)FALL_BACK_TO_NONDELALLOC; |
| 3133 |
return ext4_write_begin(iocb, mapping, pos, |
| 3134 |
len, foliop, fsdata); |
| 3135 |
} |
| 3136 |
*fsdata = (void *)0; |
| 3137 |
trace_ext4_da_write_begin(inode, pos, len); |
| 3138 |
|
| 3139 |
if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
| 3140 |
ret = ext4_generic_write_inline_data(mapping, inode, pos, len, |
| 3141 |
foliop, fsdata, true); |
| 3142 |
if (ret < 0) |
| 3143 |
return ret; |
| 3144 |
if (ret == 1) |
| 3145 |
return 0; |
| 3146 |
} |
| 3147 |
|
| 3148 |
retry: |
| 3149 |
folio = write_begin_get_folio(iocb, mapping, index, len); |
| 3150 |
if (IS_ERR(folio)) |
| 3151 |
return PTR_ERR(folio); |
| 3152 |
|
| 3153 |
if (len > folio_next_pos(folio) - pos) |
| 3154 |
len = folio_next_pos(folio) - pos; |
| 3155 |
|
| 3156 |
ret = ext4_block_write_begin(NULL, folio, pos, len, |
| 3157 |
ext4_da_get_block_prep); |
| 3158 |
if (ret < 0) { |
| 3159 |
folio_unlock(folio); |
| 3160 |
folio_put(folio); |
| 3161 |
/* |
| 3162 |
* ext4_block_write_begin may have instantiated a few blocks |
| 3163 |
* outside i_size. Trim these off again. Don't need |
| 3164 |
* i_size_read because we hold inode lock. |
| 3165 |
*/ |
| 3166 |
if (pos + len > inode->i_size) |
| 3167 |
ext4_truncate_failed_write(inode); |
| 3168 |
|
| 3169 |
if (ret == -ENOSPC && |
| 3170 |
ext4_should_retry_alloc(inode->i_sb, &retries)) |
| 3171 |
goto retry; |
| 3172 |
return ret; |
| 3173 |
} |
| 3174 |
|
| 3175 |
*foliop = folio; |
| 3176 |
return ret; |
| 3177 |
} |
| 3178 |
|
| 3179 |
/* |
| 3180 |
* Check if we should update i_disksize |
| 3181 |
* when write to the end of file but not require block allocation |
| 3182 |
*/ |
| 3183 |
static int ext4_da_should_update_i_disksize(struct folio *folio, |
| 3184 |
unsigned long offset) |
| 3185 |
{ |
| 3186 |
struct buffer_head *bh; |
| 3187 |
struct inode *inode = folio->mapping->host; |
| 3188 |
unsigned int idx; |
| 3189 |
int i; |
| 3190 |
|
| 3191 |
bh = folio_buffers(folio); |
| 3192 |
idx = offset >> inode->i_blkbits; |
| 3193 |
|
| 3194 |
for (i = 0; i < idx; i++) |
| 3195 |
bh = bh->b_this_page; |
| 3196 |
|
| 3197 |
if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
| 3198 |
return 0; |
| 3199 |
return 1; |
| 3200 |
} |
| 3201 |
|
| 3202 |
static int ext4_da_do_write_end(struct address_space *mapping, |
| 3203 |
loff_t pos, unsigned len, unsigned copied, |
| 3204 |
struct folio *folio) |
| 3205 |
{ |
| 3206 |
struct inode *inode = mapping->host; |
| 3207 |
loff_t old_size = inode->i_size; |
| 3208 |
bool disksize_changed = false; |
| 3209 |
loff_t new_i_size, zero_len = 0; |
| 3210 |
handle_t *handle; |
| 3211 |
|
| 3212 |
if (unlikely(!folio_buffers(folio))) { |
| 3213 |
folio_unlock(folio); |
| 3214 |
folio_put(folio); |
| 3215 |
return -EIO; |
| 3216 |
} |
| 3217 |
/* |
| 3218 |
* block_write_end() will mark the inode as dirty with I_DIRTY_PAGES |
| 3219 |
* flag, which all that's needed to trigger page writeback. |
| 3220 |
*/ |
| 3221 |
copied = block_write_end(pos, len, copied, folio); |
| 3222 |
new_i_size = pos + copied; |
| 3223 |
|
| 3224 |
/* |
| 3225 |
* It's important to update i_size while still holding folio lock, |
| 3226 |
* because folio writeout could otherwise come in and zero beyond |
| 3227 |
* i_size. |
| 3228 |
* |
| 3229 |
* Since we are holding inode lock, we are sure i_disksize <= |
| 3230 |
* i_size. We also know that if i_disksize < i_size, there are |
| 3231 |
* delalloc writes pending in the range up to i_size. If the end of |
| 3232 |
* the current write is <= i_size, there's no need to touch |
| 3233 |
* i_disksize since writeback will push i_disksize up to i_size |
| 3234 |
* eventually. If the end of the current write is > i_size and |
| 3235 |
* inside an allocated block which ext4_da_should_update_i_disksize() |
| 3236 |
* checked, we need to update i_disksize here as certain |
| 3237 |
* ext4_writepages() paths not allocating blocks and update i_disksize. |
| 3238 |
*/ |
| 3239 |
if (new_i_size > inode->i_size) { |
| 3240 |
unsigned long end; |
| 3241 |
|
| 3242 |
i_size_write(inode, new_i_size); |
| 3243 |
end = offset_in_folio(folio, new_i_size - 1); |
| 3244 |
if (copied && ext4_da_should_update_i_disksize(folio, end)) { |
| 3245 |
ext4_update_i_disksize(inode, new_i_size); |
| 3246 |
disksize_changed = true; |
| 3247 |
} |
| 3248 |
} |
| 3249 |
|
| 3250 |
folio_unlock(folio); |
| 3251 |
folio_put(folio); |
| 3252 |
|
| 3253 |
if (pos > old_size) { |
| 3254 |
pagecache_isize_extended(inode, old_size, pos); |
| 3255 |
zero_len = pos - old_size; |
| 3256 |
} |
| 3257 |
|
| 3258 |
if (!disksize_changed && !zero_len) |
| 3259 |
return copied; |
| 3260 |
|
| 3261 |
handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
| 3262 |
if (IS_ERR(handle)) |
| 3263 |
return PTR_ERR(handle); |
| 3264 |
if (zero_len) |
| 3265 |
ext4_zero_partial_blocks(handle, inode, old_size, zero_len); |
| 3266 |
ext4_mark_inode_dirty(handle, inode); |
| 3267 |
ext4_journal_stop(handle); |
| 3268 |
|
| 3269 |
return copied; |
| 3270 |
} |
| 3271 |
|
| 3272 |
static int ext4_da_write_end(const struct kiocb *iocb, |
| 3273 |
struct address_space *mapping, |
| 3274 |
loff_t pos, unsigned len, unsigned copied, |
| 3275 |
struct folio *folio, void *fsdata) |
| 3276 |
{ |
| 3277 |
struct inode *inode = mapping->host; |
| 3278 |
int write_mode = (int)(unsigned long)fsdata; |
| 3279 |
|
| 3280 |
if (write_mode == FALL_BACK_TO_NONDELALLOC) |
| 3281 |
return ext4_write_end(iocb, mapping, pos, |
| 3282 |
len, copied, folio, fsdata); |
| 3283 |
|
| 3284 |
trace_ext4_da_write_end(inode, pos, len, copied); |
| 3285 |
|
| 3286 |
if (write_mode != CONVERT_INLINE_DATA && |
| 3287 |
ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && |
| 3288 |
ext4_has_inline_data(inode)) |
| 3289 |
return ext4_write_inline_data_end(inode, pos, len, copied, |
| 3290 |
folio); |
| 3291 |
|
| 3292 |
if (unlikely(copied < len) && !folio_test_uptodate(folio)) |
| 3293 |
copied = 0; |
| 3294 |
|
| 3295 |
return ext4_da_do_write_end(mapping, pos, len, copied, folio); |
| 3296 |
} |
| 3297 |
|
| 3298 |
/* |
| 3299 |
* Force all delayed allocation blocks to be allocated for a given inode. |
| 3300 |
*/ |
| 3301 |
int ext4_alloc_da_blocks(struct inode *inode) |
| 3302 |
{ |
| 3303 |
trace_ext4_alloc_da_blocks(inode); |
| 3304 |
|
| 3305 |
if (!EXT4_I(inode)->i_reserved_data_blocks) |
| 3306 |
return 0; |
| 3307 |
|
| 3308 |
/* |
| 3309 |
* We do something simple for now. The filemap_flush() will |
| 3310 |
* also start triggering a write of the data blocks, which is |
| 3311 |
* not strictly speaking necessary. However, to do otherwise |
| 3312 |
* would require replicating code paths in: |
| 3313 |
* |
| 3314 |
* ext4_writepages() -> |
| 3315 |
* write_cache_pages() ---> (via passed in callback function) |
| 3316 |
* __mpage_da_writepage() --> |
| 3317 |
* mpage_add_bh_to_extent() |
| 3318 |
* mpage_da_map_blocks() |
| 3319 |
* |
| 3320 |
* The problem is that write_cache_pages(), located in |
| 3321 |
* mm/page-writeback.c, marks pages clean in preparation for |
| 3322 |
* doing I/O, which is not desirable if we're not planning on |
| 3323 |
* doing I/O at all. |
| 3324 |
* |
| 3325 |
* We could call write_cache_pages(), and then redirty all of |
| 3326 |
* the pages by calling redirty_page_for_writepage() but that |
| 3327 |
* would be ugly in the extreme. So instead we would need to |
| 3328 |
* replicate parts of the code in the above functions, |
| 3329 |
* simplifying them because we wouldn't actually intend to |
| 3330 |
* write out the pages, but rather only collect contiguous |
| 3331 |
* logical block extents, call the multi-block allocator, and |
| 3332 |
* then update the buffer heads with the block allocations. |
| 3333 |
* |
| 3334 |
* For now, though, we'll cheat by calling filemap_flush(), |
| 3335 |
* which will map the blocks, and start the I/O, but not |
| 3336 |
* actually wait for the I/O to complete. |
| 3337 |
*/ |
| 3338 |
return filemap_flush(inode->i_mapping); |
| 3339 |
} |
| 3340 |
|
| 3341 |
/* |
| 3342 |
* bmap() is special. It gets used by applications such as lilo and by |
| 3343 |
* the swapper to find the on-disk block of a specific piece of data. |
| 3344 |
* |
| 3345 |
* Naturally, this is dangerous if the block concerned is still in the |
| 3346 |
* journal. If somebody makes a swapfile on an ext4 data-journaling |
| 3347 |
* filesystem and enables swap, then they may get a nasty shock when the |
| 3348 |
* data getting swapped to that swapfile suddenly gets overwritten by |
| 3349 |
* the original zero's written out previously to the journal and |
| 3350 |
* awaiting writeback in the kernel's buffer cache. |
| 3351 |
* |
| 3352 |
* So, if we see any bmap calls here on a modified, data-journaled file, |
| 3353 |
* take extra steps to flush any blocks which might be in the cache. |
| 3354 |
*/ |
| 3355 |
static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
| 3356 |
{ |
| 3357 |
struct inode *inode = mapping->host; |
| 3358 |
sector_t ret = 0; |
| 3359 |
|
| 3360 |
inode_lock_shared(inode); |
| 3361 |
/* |
| 3362 |
* We can get here for an inline file via the FIBMAP ioctl |
| 3363 |
*/ |
| 3364 |
if (ext4_has_inline_data(inode)) |
| 3365 |
goto out; |
| 3366 |
|
| 3367 |
if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
| 3368 |
(test_opt(inode->i_sb, DELALLOC) || |
| 3369 |
ext4_should_journal_data(inode))) { |
| 3370 |
/* |
| 3371 |
* With delalloc or journalled data we want to sync the file so |
| 3372 |
* that we can make sure we allocate blocks for file and data |
| 3373 |
* is in place for the user to see it |
| 3374 |
*/ |
| 3375 |
filemap_write_and_wait(mapping); |
| 3376 |
} |
| 3377 |
|
| 3378 |
ret = iomap_bmap(mapping, block, &ext4_iomap_ops); |
| 3379 |
|
| 3380 |
out: |
| 3381 |
inode_unlock_shared(inode); |
| 3382 |
return ret; |
| 3383 |
} |
| 3384 |
|
| 3385 |
static void ext4_invalidate_folio(struct folio *folio, size_t offset, |
| 3386 |
size_t length) |
| 3387 |
{ |
| 3388 |
trace_ext4_invalidate_folio(folio, offset, length); |
| 3389 |
|
| 3390 |
/* No journalling happens on data buffers when this function is used */ |
| 3391 |
WARN_ON(folio_buffers(folio) && buffer_jbd(folio_buffers(folio))); |
| 3392 |
|
| 3393 |
block_invalidate_folio(folio, offset, length); |
| 3394 |
} |
| 3395 |
|
| 3396 |
static int __ext4_journalled_invalidate_folio(struct folio *folio, |
| 3397 |
size_t offset, size_t length) |
| 3398 |
{ |
| 3399 |
journal_t *journal = EXT4_JOURNAL(folio->mapping->host); |
| 3400 |
|
| 3401 |
trace_ext4_journalled_invalidate_folio(folio, offset, length); |
| 3402 |
|
| 3403 |
/* |
| 3404 |
* If it's a full truncate we just forget about the pending dirtying |
| 3405 |
*/ |
| 3406 |
if (offset == 0 && length == folio_size(folio)) |
| 3407 |
folio_clear_checked(folio); |
| 3408 |
|
| 3409 |
return jbd2_journal_invalidate_folio(journal, folio, offset, length); |
| 3410 |
} |
| 3411 |
|
| 3412 |
/* Wrapper for aops... */ |
| 3413 |
static void ext4_journalled_invalidate_folio(struct folio *folio, |
| 3414 |
size_t offset, |
| 3415 |
size_t length) |
| 3416 |
{ |
| 3417 |
WARN_ON(__ext4_journalled_invalidate_folio(folio, offset, length) < 0); |
| 3418 |
} |
| 3419 |
|
| 3420 |
static bool ext4_release_folio(struct folio *folio, gfp_t wait) |
| 3421 |
{ |
| 3422 |
struct inode *inode = folio->mapping->host; |
| 3423 |
journal_t *journal = EXT4_JOURNAL(inode); |
| 3424 |
|
| 3425 |
trace_ext4_release_folio(inode, folio); |
| 3426 |
|
| 3427 |
/* Page has dirty journalled data -> cannot release */ |
| 3428 |
if (folio_test_checked(folio)) |
| 3429 |
return false; |
| 3430 |
if (journal) |
| 3431 |
return jbd2_journal_try_to_free_buffers(journal, folio); |
| 3432 |
else |
| 3433 |
return try_to_free_buffers(folio); |
| 3434 |
} |
| 3435 |
|
| 3436 |
static bool ext4_inode_datasync_dirty(struct inode *inode) |
| 3437 |
{ |
| 3438 |
journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
| 3439 |
|
| 3440 |
if (journal) { |
| 3441 |
if (jbd2_transaction_committed(journal, |
| 3442 |
EXT4_I(inode)->i_datasync_tid)) |
| 3443 |
return false; |
| 3444 |
if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT)) |
| 3445 |
return !list_empty(&EXT4_I(inode)->i_fc_list); |
| 3446 |
return true; |
| 3447 |
} |
| 3448 |
|
| 3449 |
/* Any metadata buffers to write? */ |
| 3450 |
if (!list_empty(&inode->i_mapping->i_private_list)) |
| 3451 |
return true; |
| 3452 |
return inode_state_read_once(inode) & I_DIRTY_DATASYNC; |
| 3453 |
} |
| 3454 |
|
| 3455 |
static void ext4_set_iomap(struct inode *inode, struct iomap *iomap, |
| 3456 |
struct ext4_map_blocks *map, loff_t offset, |
| 3457 |
loff_t length, unsigned int flags) |
| 3458 |
{ |
| 3459 |
u8 blkbits = inode->i_blkbits; |
| 3460 |
|
| 3461 |
/* |
| 3462 |
* Writes that span EOF might trigger an I/O size update on completion, |
| 3463 |
* so consider them to be dirty for the purpose of O_DSYNC, even if |
| 3464 |
* there is no other metadata changes being made or are pending. |
| 3465 |
*/ |
| 3466 |
iomap->flags = 0; |
| 3467 |
if (ext4_inode_datasync_dirty(inode) || |
| 3468 |
offset + length > i_size_read(inode)) |
| 3469 |
iomap->flags |= IOMAP_F_DIRTY; |
| 3470 |
|
| 3471 |
if (map->m_flags & EXT4_MAP_NEW) |
| 3472 |
iomap->flags |= IOMAP_F_NEW; |
| 3473 |
|
| 3474 |
/* HW-offload atomics are always used */ |
| 3475 |
if (flags & IOMAP_ATOMIC) |
| 3476 |
iomap->flags |= IOMAP_F_ATOMIC_BIO; |
| 3477 |
|
| 3478 |
if (flags & IOMAP_DAX) |
| 3479 |
iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev; |
| 3480 |
else |
| 3481 |
iomap->bdev = inode->i_sb->s_bdev; |
| 3482 |
iomap->offset = EXT4_LBLK_TO_B(inode, map->m_lblk); |
| 3483 |
iomap->length = EXT4_LBLK_TO_B(inode, map->m_len); |
| 3484 |
|
| 3485 |
if ((map->m_flags & EXT4_MAP_MAPPED) && |
| 3486 |
!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 3487 |
iomap->flags |= IOMAP_F_MERGED; |
| 3488 |
|
| 3489 |
/* |
| 3490 |
* Flags passed to ext4_map_blocks() for direct I/O writes can result |
| 3491 |
* in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits |
| 3492 |
* set. In order for any allocated unwritten extents to be converted |
| 3493 |
* into written extents correctly within the ->end_io() handler, we |
| 3494 |
* need to ensure that the iomap->type is set appropriately. Hence, the |
| 3495 |
* reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has |
| 3496 |
* been set first. |
| 3497 |
*/ |
| 3498 |
if (map->m_flags & EXT4_MAP_UNWRITTEN) { |
| 3499 |
iomap->type = IOMAP_UNWRITTEN; |
| 3500 |
iomap->addr = (u64) map->m_pblk << blkbits; |
| 3501 |
if (flags & IOMAP_DAX) |
| 3502 |
iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off; |
| 3503 |
} else if (map->m_flags & EXT4_MAP_MAPPED) { |
| 3504 |
iomap->type = IOMAP_MAPPED; |
| 3505 |
iomap->addr = (u64) map->m_pblk << blkbits; |
| 3506 |
if (flags & IOMAP_DAX) |
| 3507 |
iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off; |
| 3508 |
} else if (map->m_flags & EXT4_MAP_DELAYED) { |
| 3509 |
iomap->type = IOMAP_DELALLOC; |
| 3510 |
iomap->addr = IOMAP_NULL_ADDR; |
| 3511 |
} else { |
| 3512 |
iomap->type = IOMAP_HOLE; |
| 3513 |
iomap->addr = IOMAP_NULL_ADDR; |
| 3514 |
} |
| 3515 |
} |
| 3516 |
|
| 3517 |
static int ext4_map_blocks_atomic_write_slow(handle_t *handle, |
| 3518 |
struct inode *inode, struct ext4_map_blocks *map) |
| 3519 |
{ |
| 3520 |
ext4_lblk_t m_lblk = map->m_lblk; |
| 3521 |
unsigned int m_len = map->m_len; |
| 3522 |
unsigned int mapped_len = 0, m_flags = 0; |
| 3523 |
ext4_fsblk_t next_pblk = 0; |
| 3524 |
bool check_next_pblk = false; |
| 3525 |
int ret = 0; |
| 3526 |
|
| 3527 |
WARN_ON_ONCE(!ext4_has_feature_bigalloc(inode->i_sb)); |
| 3528 |
|
| 3529 |
/* |
| 3530 |
* This is a slow path in case of mixed mapping. We use |
| 3531 |
* EXT4_GET_BLOCKS_CREATE_ZERO flag here to make sure we get a single |
| 3532 |
* contiguous mapped mapping. This will ensure any unwritten or hole |
| 3533 |
* regions within the requested range is zeroed out and we return |
| 3534 |
* a single contiguous mapped extent. |
| 3535 |
*/ |
| 3536 |
m_flags = EXT4_GET_BLOCKS_CREATE_ZERO; |
| 3537 |
|
| 3538 |
do { |
| 3539 |
ret = ext4_map_blocks(handle, inode, map, m_flags); |
| 3540 |
if (ret < 0 && ret != -ENOSPC) |
| 3541 |
goto out_err; |
| 3542 |
/* |
| 3543 |
* This should never happen, but let's return an error code to |
| 3544 |
* avoid an infinite loop in here. |
| 3545 |
*/ |
| 3546 |
if (ret == 0) { |
| 3547 |
ret = -EFSCORRUPTED; |
| 3548 |
ext4_warning_inode(inode, |
| 3549 |
"ext4_map_blocks() couldn't allocate blocks m_flags: 0x%x, ret:%d", |
| 3550 |
m_flags, ret); |
| 3551 |
goto out_err; |
| 3552 |
} |
| 3553 |
/* |
| 3554 |
* With bigalloc we should never get ENOSPC nor discontiguous |
| 3555 |
* physical extents. |
| 3556 |
*/ |
| 3557 |
if ((check_next_pblk && next_pblk != map->m_pblk) || |
| 3558 |
ret == -ENOSPC) { |
| 3559 |
ext4_warning_inode(inode, |
| 3560 |
"Non-contiguous allocation detected: expected %llu, got %llu, " |
| 3561 |
"or ext4_map_blocks() returned out of space ret: %d", |
| 3562 |
next_pblk, map->m_pblk, ret); |
| 3563 |
ret = -EFSCORRUPTED; |
| 3564 |
goto out_err; |
| 3565 |
} |
| 3566 |
next_pblk = map->m_pblk + map->m_len; |
| 3567 |
check_next_pblk = true; |
| 3568 |
|
| 3569 |
mapped_len += map->m_len; |
| 3570 |
map->m_lblk += map->m_len; |
| 3571 |
map->m_len = m_len - mapped_len; |
| 3572 |
} while (mapped_len < m_len); |
| 3573 |
|
| 3574 |
/* |
| 3575 |
* We might have done some work in above loop, so we need to query the |
| 3576 |
* start of the physical extent, based on the origin m_lblk and m_len. |
| 3577 |
* Let's also ensure we were able to allocate the required range for |
| 3578 |
* mixed mapping case. |
| 3579 |
*/ |
| 3580 |
map->m_lblk = m_lblk; |
| 3581 |
map->m_len = m_len; |
| 3582 |
map->m_flags = 0; |
| 3583 |
|
| 3584 |
ret = ext4_map_blocks(handle, inode, map, |
| 3585 |
EXT4_GET_BLOCKS_QUERY_LAST_IN_LEAF); |
| 3586 |
if (ret != m_len) { |
| 3587 |
ext4_warning_inode(inode, |
| 3588 |
"allocation failed for atomic write request m_lblk:%u, m_len:%u, ret:%d\n", |
| 3589 |
m_lblk, m_len, ret); |
| 3590 |
ret = -EINVAL; |
| 3591 |
} |
| 3592 |
return ret; |
| 3593 |
|
| 3594 |
out_err: |
| 3595 |
/* reset map before returning an error */ |
| 3596 |
map->m_lblk = m_lblk; |
| 3597 |
map->m_len = m_len; |
| 3598 |
map->m_flags = 0; |
| 3599 |
return ret; |
| 3600 |
} |
| 3601 |
|
| 3602 |
/* |
| 3603 |
* ext4_map_blocks_atomic: Helper routine to ensure the entire requested |
| 3604 |
* range in @map [lblk, lblk + len) is one single contiguous extent with no |
| 3605 |
* mixed mappings. |
| 3606 |
* |
| 3607 |
* We first use m_flags passed to us by our caller (ext4_iomap_alloc()). |
| 3608 |
* We only call EXT4_GET_BLOCKS_ZERO in the slow path, when the underlying |
| 3609 |
* physical extent for the requested range does not have a single contiguous |
| 3610 |
* mapping type i.e. (Hole, Mapped, or Unwritten) throughout. |
| 3611 |
* In that case we will loop over the requested range to allocate and zero out |
| 3612 |
* the unwritten / holes in between, to get a single mapped extent from |
| 3613 |
* [m_lblk, m_lblk + m_len). Note that this is only possible because we know |
| 3614 |
* this can be called only with bigalloc enabled filesystem where the underlying |
| 3615 |
* cluster is already allocated. This avoids allocating discontiguous extents |
| 3616 |
* in the slow path due to multiple calls to ext4_map_blocks(). |
| 3617 |
* The slow path is mostly non-performance critical path, so it should be ok to |
| 3618 |
* loop using ext4_map_blocks() with appropriate flags to allocate & zero the |
| 3619 |
* underlying short holes/unwritten extents within the requested range. |
| 3620 |
*/ |
| 3621 |
static int ext4_map_blocks_atomic_write(handle_t *handle, struct inode *inode, |
| 3622 |
struct ext4_map_blocks *map, int m_flags, |
| 3623 |
bool *force_commit) |
| 3624 |
{ |
| 3625 |
ext4_lblk_t m_lblk = map->m_lblk; |
| 3626 |
unsigned int m_len = map->m_len; |
| 3627 |
int ret = 0; |
| 3628 |
|
| 3629 |
WARN_ON_ONCE(m_len > 1 && !ext4_has_feature_bigalloc(inode->i_sb)); |
| 3630 |
|
| 3631 |
ret = ext4_map_blocks(handle, inode, map, m_flags); |
| 3632 |
if (ret < 0 || ret == m_len) |
| 3633 |
goto out; |
| 3634 |
/* |
| 3635 |
* This is a mixed mapping case where we were not able to allocate |
| 3636 |
* a single contiguous extent. In that case let's reset requested |
| 3637 |
* mapping and call the slow path. |
| 3638 |
*/ |
| 3639 |
map->m_lblk = m_lblk; |
| 3640 |
map->m_len = m_len; |
| 3641 |
map->m_flags = 0; |
| 3642 |
|
| 3643 |
/* |
| 3644 |
* slow path means we have mixed mapping, that means we will need |
| 3645 |
* to force txn commit. |
| 3646 |
*/ |
| 3647 |
*force_commit = true; |
| 3648 |
return ext4_map_blocks_atomic_write_slow(handle, inode, map); |
| 3649 |
out: |
| 3650 |
return ret; |
| 3651 |
} |
| 3652 |
|
| 3653 |
static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map, |
| 3654 |
unsigned int flags) |
| 3655 |
{ |
| 3656 |
handle_t *handle; |
| 3657 |
int ret, dio_credits, m_flags = 0, retries = 0; |
| 3658 |
bool force_commit = false; |
| 3659 |
|
| 3660 |
/* |
| 3661 |
* Trim the mapping request to the maximum value that we can map at |
| 3662 |
* once for direct I/O. |
| 3663 |
*/ |
| 3664 |
if (map->m_len > DIO_MAX_BLOCKS) |
| 3665 |
map->m_len = DIO_MAX_BLOCKS; |
| 3666 |
|
| 3667 |
/* |
| 3668 |
* journal credits estimation for atomic writes. We call |
| 3669 |
* ext4_map_blocks(), to find if there could be a mixed mapping. If yes, |
| 3670 |
* then let's assume the no. of pextents required can be m_len i.e. |
| 3671 |
* every alternate block can be unwritten and hole. |
| 3672 |
*/ |
| 3673 |
if (flags & IOMAP_ATOMIC) { |
| 3674 |
unsigned int orig_mlen = map->m_len; |
| 3675 |
|
| 3676 |
ret = ext4_map_blocks(NULL, inode, map, 0); |
| 3677 |
if (ret < 0) |
| 3678 |
return ret; |
| 3679 |
if (map->m_len < orig_mlen) { |
| 3680 |
map->m_len = orig_mlen; |
| 3681 |
dio_credits = ext4_meta_trans_blocks(inode, orig_mlen, |
| 3682 |
map->m_len); |
| 3683 |
} else { |
| 3684 |
dio_credits = ext4_chunk_trans_blocks(inode, |
| 3685 |
map->m_len); |
| 3686 |
} |
| 3687 |
} else { |
| 3688 |
dio_credits = ext4_chunk_trans_blocks(inode, map->m_len); |
| 3689 |
} |
| 3690 |
|
| 3691 |
retry: |
| 3692 |
/* |
| 3693 |
* Either we allocate blocks and then don't get an unwritten extent, so |
| 3694 |
* in that case we have reserved enough credits. Or, the blocks are |
| 3695 |
* already allocated and unwritten. In that case, the extent conversion |
| 3696 |
* fits into the credits as well. |
| 3697 |
*/ |
| 3698 |
handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits); |
| 3699 |
if (IS_ERR(handle)) |
| 3700 |
return PTR_ERR(handle); |
| 3701 |
|
| 3702 |
/* |
| 3703 |
* DAX and direct I/O are the only two operations that are currently |
| 3704 |
* supported with IOMAP_WRITE. |
| 3705 |
*/ |
| 3706 |
WARN_ON(!(flags & (IOMAP_DAX | IOMAP_DIRECT))); |
| 3707 |
if (flags & IOMAP_DAX) |
| 3708 |
m_flags = EXT4_GET_BLOCKS_CREATE_ZERO; |
| 3709 |
/* |
| 3710 |
* We use i_size instead of i_disksize here because delalloc writeback |
| 3711 |
* can complete at any point during the I/O and subsequently push the |
| 3712 |
* i_disksize out to i_size. This could be beyond where direct I/O is |
| 3713 |
* happening and thus expose allocated blocks to direct I/O reads. |
| 3714 |
*/ |
| 3715 |
else if (EXT4_LBLK_TO_B(inode, map->m_lblk) >= i_size_read(inode)) |
| 3716 |
m_flags = EXT4_GET_BLOCKS_CREATE; |
| 3717 |
else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 3718 |
m_flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT; |
| 3719 |
|
| 3720 |
if (flags & IOMAP_ATOMIC) |
| 3721 |
ret = ext4_map_blocks_atomic_write(handle, inode, map, m_flags, |
| 3722 |
&force_commit); |
| 3723 |
else |
| 3724 |
ret = ext4_map_blocks(handle, inode, map, m_flags); |
| 3725 |
|
| 3726 |
/* |
| 3727 |
* We cannot fill holes in indirect tree based inodes as that could |
| 3728 |
* expose stale data in the case of a crash. Use the magic error code |
| 3729 |
* to fallback to buffered I/O. |
| 3730 |
*/ |
| 3731 |
if (!m_flags && !ret) |
| 3732 |
ret = -ENOTBLK; |
| 3733 |
|
| 3734 |
ext4_journal_stop(handle); |
| 3735 |
if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
| 3736 |
goto retry; |
| 3737 |
|
| 3738 |
/* |
| 3739 |
* Force commit the current transaction if the allocation spans a mixed |
| 3740 |
* mapping range. This ensures any pending metadata updates (like |
| 3741 |
* unwritten to written extents conversion) in this range are in |
| 3742 |
* consistent state with the file data blocks, before performing the |
| 3743 |
* actual write I/O. If the commit fails, the whole I/O must be aborted |
| 3744 |
* to prevent any possible torn writes. |
| 3745 |
*/ |
| 3746 |
if (ret > 0 && force_commit) { |
| 3747 |
int ret2; |
| 3748 |
|
| 3749 |
ret2 = ext4_force_commit(inode->i_sb); |
| 3750 |
if (ret2) |
| 3751 |
return ret2; |
| 3752 |
} |
| 3753 |
|
| 3754 |
return ret; |
| 3755 |
} |
| 3756 |
|
| 3757 |
|
| 3758 |
static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length, |
| 3759 |
unsigned flags, struct iomap *iomap, struct iomap *srcmap) |
| 3760 |
{ |
| 3761 |
int ret; |
| 3762 |
struct ext4_map_blocks map; |
| 3763 |
u8 blkbits = inode->i_blkbits; |
| 3764 |
unsigned int orig_mlen; |
| 3765 |
|
| 3766 |
if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK) |
| 3767 |
return -EINVAL; |
| 3768 |
|
| 3769 |
if (WARN_ON_ONCE(ext4_has_inline_data(inode))) |
| 3770 |
return -ERANGE; |
| 3771 |
|
| 3772 |
/* |
| 3773 |
* Calculate the first and last logical blocks respectively. |
| 3774 |
*/ |
| 3775 |
map.m_lblk = offset >> blkbits; |
| 3776 |
map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits, |
| 3777 |
EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1; |
| 3778 |
orig_mlen = map.m_len; |
| 3779 |
|
| 3780 |
if (flags & IOMAP_WRITE) { |
| 3781 |
/* |
| 3782 |
* We check here if the blocks are already allocated, then we |
| 3783 |
* don't need to start a journal txn and we can directly return |
| 3784 |
* the mapping information. This could boost performance |
| 3785 |
* especially in multi-threaded overwrite requests. |
| 3786 |
*/ |
| 3787 |
if (offset + length <= i_size_read(inode)) { |
| 3788 |
ret = ext4_map_blocks(NULL, inode, &map, 0); |
| 3789 |
/* |
| 3790 |
* For DAX we convert extents to initialized ones before |
| 3791 |
* copying the data, otherwise we do it after I/O so |
| 3792 |
* there's no need to call into ext4_iomap_alloc(). |
| 3793 |
*/ |
| 3794 |
if ((map.m_flags & EXT4_MAP_MAPPED) || |
| 3795 |
(!(flags & IOMAP_DAX) && |
| 3796 |
(map.m_flags & EXT4_MAP_UNWRITTEN))) { |
| 3797 |
/* |
| 3798 |
* For atomic writes the entire requested |
| 3799 |
* length should be mapped. |
| 3800 |
*/ |
| 3801 |
if (ret == orig_mlen || |
| 3802 |
(!(flags & IOMAP_ATOMIC) && ret > 0)) |
| 3803 |
goto out; |
| 3804 |
} |
| 3805 |
map.m_len = orig_mlen; |
| 3806 |
} |
| 3807 |
ret = ext4_iomap_alloc(inode, &map, flags); |
| 3808 |
} else { |
| 3809 |
ret = ext4_map_blocks(NULL, inode, &map, 0); |
| 3810 |
} |
| 3811 |
|
| 3812 |
if (ret < 0) |
| 3813 |
return ret; |
| 3814 |
out: |
| 3815 |
/* |
| 3816 |
* When inline encryption is enabled, sometimes I/O to an encrypted file |
| 3817 |
* has to be broken up to guarantee DUN contiguity. Handle this by |
| 3818 |
* limiting the length of the mapping returned. |
| 3819 |
*/ |
| 3820 |
map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len); |
| 3821 |
|
| 3822 |
/* |
| 3823 |
* Before returning to iomap, let's ensure the allocated mapping |
| 3824 |
* covers the entire requested length for atomic writes. |
| 3825 |
*/ |
| 3826 |
if (flags & IOMAP_ATOMIC) { |
| 3827 |
if (map.m_len < (length >> blkbits)) { |
| 3828 |
WARN_ON_ONCE(1); |
| 3829 |
return -EINVAL; |
| 3830 |
} |
| 3831 |
} |
| 3832 |
ext4_set_iomap(inode, iomap, &map, offset, length, flags); |
| 3833 |
|
| 3834 |
return 0; |
| 3835 |
} |
| 3836 |
|
| 3837 |
const struct iomap_ops ext4_iomap_ops = { |
| 3838 |
.iomap_begin = ext4_iomap_begin, |
| 3839 |
}; |
| 3840 |
|
| 3841 |
static int ext4_iomap_begin_report(struct inode *inode, loff_t offset, |
| 3842 |
loff_t length, unsigned int flags, |
| 3843 |
struct iomap *iomap, struct iomap *srcmap) |
| 3844 |
{ |
| 3845 |
int ret; |
| 3846 |
struct ext4_map_blocks map; |
| 3847 |
u8 blkbits = inode->i_blkbits; |
| 3848 |
|
| 3849 |
if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK) |
| 3850 |
return -EINVAL; |
| 3851 |
|
| 3852 |
if (ext4_has_inline_data(inode)) { |
| 3853 |
ret = ext4_inline_data_iomap(inode, iomap); |
| 3854 |
if (ret != -EAGAIN) { |
| 3855 |
if (ret == 0 && offset >= iomap->length) |
| 3856 |
ret = -ENOENT; |
| 3857 |
return ret; |
| 3858 |
} |
| 3859 |
} |
| 3860 |
|
| 3861 |
/* |
| 3862 |
* Calculate the first and last logical block respectively. |
| 3863 |
*/ |
| 3864 |
map.m_lblk = offset >> blkbits; |
| 3865 |
map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits, |
| 3866 |
EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1; |
| 3867 |
|
| 3868 |
/* |
| 3869 |
* Fiemap callers may call for offset beyond s_bitmap_maxbytes. |
| 3870 |
* So handle it here itself instead of querying ext4_map_blocks(). |
| 3871 |
* Since ext4_map_blocks() will warn about it and will return |
| 3872 |
* -EIO error. |
| 3873 |
*/ |
| 3874 |
if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
| 3875 |
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 3876 |
|
| 3877 |
if (offset >= sbi->s_bitmap_maxbytes) { |
| 3878 |
map.m_flags = 0; |
| 3879 |
goto set_iomap; |
| 3880 |
} |
| 3881 |
} |
| 3882 |
|
| 3883 |
ret = ext4_map_blocks(NULL, inode, &map, 0); |
| 3884 |
if (ret < 0) |
| 3885 |
return ret; |
| 3886 |
set_iomap: |
| 3887 |
ext4_set_iomap(inode, iomap, &map, offset, length, flags); |
| 3888 |
|
| 3889 |
return 0; |
| 3890 |
} |
| 3891 |
|
| 3892 |
const struct iomap_ops ext4_iomap_report_ops = { |
| 3893 |
.iomap_begin = ext4_iomap_begin_report, |
| 3894 |
}; |
| 3895 |
|
| 3896 |
/* |
| 3897 |
* For data=journal mode, folio should be marked dirty only when it was |
| 3898 |
* writeably mapped. When that happens, it was already attached to the |
| 3899 |
* transaction and marked as jbddirty (we take care of this in |
| 3900 |
* ext4_page_mkwrite()). On transaction commit, we writeprotect page mappings |
| 3901 |
* so we should have nothing to do here, except for the case when someone |
| 3902 |
* had the page pinned and dirtied the page through this pin (e.g. by doing |
| 3903 |
* direct IO to it). In that case we'd need to attach buffers here to the |
| 3904 |
* transaction but we cannot due to lock ordering. We cannot just dirty the |
| 3905 |
* folio and leave attached buffers clean, because the buffers' dirty state is |
| 3906 |
* "definitive". We cannot just set the buffers dirty or jbddirty because all |
| 3907 |
* the journalling code will explode. So what we do is to mark the folio |
| 3908 |
* "pending dirty" and next time ext4_writepages() is called, attach buffers |
| 3909 |
* to the transaction appropriately. |
| 3910 |
*/ |
| 3911 |
static bool ext4_journalled_dirty_folio(struct address_space *mapping, |
| 3912 |
struct folio *folio) |
| 3913 |
{ |
| 3914 |
WARN_ON_ONCE(!folio_buffers(folio)); |
| 3915 |
if (folio_maybe_dma_pinned(folio)) |
| 3916 |
folio_set_checked(folio); |
| 3917 |
return filemap_dirty_folio(mapping, folio); |
| 3918 |
} |
| 3919 |
|
| 3920 |
static bool ext4_dirty_folio(struct address_space *mapping, struct folio *folio) |
| 3921 |
{ |
| 3922 |
WARN_ON_ONCE(!folio_test_locked(folio) && !folio_test_dirty(folio)); |
| 3923 |
WARN_ON_ONCE(!folio_buffers(folio)); |
| 3924 |
return block_dirty_folio(mapping, folio); |
| 3925 |
} |
| 3926 |
|
| 3927 |
static int ext4_iomap_swap_activate(struct swap_info_struct *sis, |
| 3928 |
struct file *file, sector_t *span) |
| 3929 |
{ |
| 3930 |
return iomap_swapfile_activate(sis, file, span, |
| 3931 |
&ext4_iomap_report_ops); |
| 3932 |
} |
| 3933 |
|
| 3934 |
static const struct address_space_operations ext4_aops = { |
| 3935 |
.read_folio = ext4_read_folio, |
| 3936 |
.readahead = ext4_readahead, |
| 3937 |
.writepages = ext4_writepages, |
| 3938 |
.write_begin = ext4_write_begin, |
| 3939 |
.write_end = ext4_write_end, |
| 3940 |
.dirty_folio = ext4_dirty_folio, |
| 3941 |
.bmap = ext4_bmap, |
| 3942 |
.invalidate_folio = ext4_invalidate_folio, |
| 3943 |
.release_folio = ext4_release_folio, |
| 3944 |
.migrate_folio = buffer_migrate_folio, |
| 3945 |
.is_partially_uptodate = block_is_partially_uptodate, |
| 3946 |
.error_remove_folio = generic_error_remove_folio, |
| 3947 |
.swap_activate = ext4_iomap_swap_activate, |
| 3948 |
}; |
| 3949 |
|
| 3950 |
static const struct address_space_operations ext4_journalled_aops = { |
| 3951 |
.read_folio = ext4_read_folio, |
| 3952 |
.readahead = ext4_readahead, |
| 3953 |
.writepages = ext4_writepages, |
| 3954 |
.write_begin = ext4_write_begin, |
| 3955 |
.write_end = ext4_journalled_write_end, |
| 3956 |
.dirty_folio = ext4_journalled_dirty_folio, |
| 3957 |
.bmap = ext4_bmap, |
| 3958 |
.invalidate_folio = ext4_journalled_invalidate_folio, |
| 3959 |
.release_folio = ext4_release_folio, |
| 3960 |
.migrate_folio = buffer_migrate_folio_norefs, |
| 3961 |
.is_partially_uptodate = block_is_partially_uptodate, |
| 3962 |
.error_remove_folio = generic_error_remove_folio, |
| 3963 |
.swap_activate = ext4_iomap_swap_activate, |
| 3964 |
}; |
| 3965 |
|
| 3966 |
static const struct address_space_operations ext4_da_aops = { |
| 3967 |
.read_folio = ext4_read_folio, |
| 3968 |
.readahead = ext4_readahead, |
| 3969 |
.writepages = ext4_writepages, |
| 3970 |
.write_begin = ext4_da_write_begin, |
| 3971 |
.write_end = ext4_da_write_end, |
| 3972 |
.dirty_folio = ext4_dirty_folio, |
| 3973 |
.bmap = ext4_bmap, |
| 3974 |
.invalidate_folio = ext4_invalidate_folio, |
| 3975 |
.release_folio = ext4_release_folio, |
| 3976 |
.migrate_folio = buffer_migrate_folio, |
| 3977 |
.is_partially_uptodate = block_is_partially_uptodate, |
| 3978 |
.error_remove_folio = generic_error_remove_folio, |
| 3979 |
.swap_activate = ext4_iomap_swap_activate, |
| 3980 |
}; |
| 3981 |
|
| 3982 |
static const struct address_space_operations ext4_dax_aops = { |
| 3983 |
.writepages = ext4_dax_writepages, |
| 3984 |
.dirty_folio = noop_dirty_folio, |
| 3985 |
.bmap = ext4_bmap, |
| 3986 |
.swap_activate = ext4_iomap_swap_activate, |
| 3987 |
}; |
| 3988 |
|
| 3989 |
void ext4_set_aops(struct inode *inode) |
| 3990 |
{ |
| 3991 |
switch (ext4_inode_journal_mode(inode)) { |
| 3992 |
case EXT4_INODE_ORDERED_DATA_MODE: |
| 3993 |
case EXT4_INODE_WRITEBACK_DATA_MODE: |
| 3994 |
break; |
| 3995 |
case EXT4_INODE_JOURNAL_DATA_MODE: |
| 3996 |
inode->i_mapping->a_ops = &ext4_journalled_aops; |
| 3997 |
return; |
| 3998 |
default: |
| 3999 |
BUG(); |
| 4000 |
} |
| 4001 |
if (IS_DAX(inode)) |
| 4002 |
inode->i_mapping->a_ops = &ext4_dax_aops; |
| 4003 |
else if (test_opt(inode->i_sb, DELALLOC)) |
| 4004 |
inode->i_mapping->a_ops = &ext4_da_aops; |
| 4005 |
else |
| 4006 |
inode->i_mapping->a_ops = &ext4_aops; |
| 4007 |
} |
| 4008 |
|
| 4009 |
/* |
| 4010 |
* Here we can't skip an unwritten buffer even though it usually reads zero |
| 4011 |
* because it might have data in pagecache (eg, if called from ext4_zero_range, |
| 4012 |
* ext4_punch_hole, etc) which needs to be properly zeroed out. Otherwise a |
| 4013 |
* racing writeback can come later and flush the stale pagecache to disk. |
| 4014 |
*/ |
| 4015 |
static int __ext4_block_zero_page_range(handle_t *handle, |
| 4016 |
struct address_space *mapping, loff_t from, loff_t length) |
| 4017 |
{ |
| 4018 |
unsigned int offset, blocksize, pos; |
| 4019 |
ext4_lblk_t iblock; |
| 4020 |
struct inode *inode = mapping->host; |
| 4021 |
struct buffer_head *bh; |
| 4022 |
struct folio *folio; |
| 4023 |
int err = 0; |
| 4024 |
|
| 4025 |
folio = __filemap_get_folio(mapping, from >> PAGE_SHIFT, |
| 4026 |
FGP_LOCK | FGP_ACCESSED | FGP_CREAT, |
| 4027 |
mapping_gfp_constraint(mapping, ~__GFP_FS)); |
| 4028 |
if (IS_ERR(folio)) |
| 4029 |
return PTR_ERR(folio); |
| 4030 |
|
| 4031 |
blocksize = inode->i_sb->s_blocksize; |
| 4032 |
|
| 4033 |
iblock = EXT4_PG_TO_LBLK(inode, folio->index); |
| 4034 |
|
| 4035 |
bh = folio_buffers(folio); |
| 4036 |
if (!bh) |
| 4037 |
bh = create_empty_buffers(folio, blocksize, 0); |
| 4038 |
|
| 4039 |
/* Find the buffer that contains "offset" */ |
| 4040 |
offset = offset_in_folio(folio, from); |
| 4041 |
pos = blocksize; |
| 4042 |
while (offset >= pos) { |
| 4043 |
bh = bh->b_this_page; |
| 4044 |
iblock++; |
| 4045 |
pos += blocksize; |
| 4046 |
} |
| 4047 |
if (buffer_freed(bh)) { |
| 4048 |
BUFFER_TRACE(bh, "freed: skip"); |
| 4049 |
goto unlock; |
| 4050 |
} |
| 4051 |
if (!buffer_mapped(bh)) { |
| 4052 |
BUFFER_TRACE(bh, "unmapped"); |
| 4053 |
ext4_get_block(inode, iblock, bh, 0); |
| 4054 |
/* unmapped? It's a hole - nothing to do */ |
| 4055 |
if (!buffer_mapped(bh)) { |
| 4056 |
BUFFER_TRACE(bh, "still unmapped"); |
| 4057 |
goto unlock; |
| 4058 |
} |
| 4059 |
} |
| 4060 |
|
| 4061 |
/* Ok, it's mapped. Make sure it's up-to-date */ |
| 4062 |
if (folio_test_uptodate(folio)) |
| 4063 |
set_buffer_uptodate(bh); |
| 4064 |
|
| 4065 |
if (!buffer_uptodate(bh)) { |
| 4066 |
err = ext4_read_bh_lock(bh, 0, true); |
| 4067 |
if (err) |
| 4068 |
goto unlock; |
| 4069 |
if (fscrypt_inode_uses_fs_layer_crypto(inode)) { |
| 4070 |
/* We expect the key to be set. */ |
| 4071 |
BUG_ON(!fscrypt_has_encryption_key(inode)); |
| 4072 |
err = fscrypt_decrypt_pagecache_blocks(folio, |
| 4073 |
blocksize, |
| 4074 |
bh_offset(bh)); |
| 4075 |
if (err) { |
| 4076 |
clear_buffer_uptodate(bh); |
| 4077 |
goto unlock; |
| 4078 |
} |
| 4079 |
} |
| 4080 |
} |
| 4081 |
if (ext4_should_journal_data(inode)) { |
| 4082 |
BUFFER_TRACE(bh, "get write access"); |
| 4083 |
err = ext4_journal_get_write_access(handle, inode->i_sb, bh, |
| 4084 |
EXT4_JTR_NONE); |
| 4085 |
if (err) |
| 4086 |
goto unlock; |
| 4087 |
} |
| 4088 |
folio_zero_range(folio, offset, length); |
| 4089 |
BUFFER_TRACE(bh, "zeroed end of block"); |
| 4090 |
|
| 4091 |
if (ext4_should_journal_data(inode)) { |
| 4092 |
err = ext4_dirty_journalled_data(handle, bh); |
| 4093 |
} else { |
| 4094 |
mark_buffer_dirty(bh); |
| 4095 |
/* |
| 4096 |
* Only the written block requires ordered data to prevent |
| 4097 |
* exposing stale data. |
| 4098 |
*/ |
| 4099 |
if (!buffer_unwritten(bh) && !buffer_delay(bh) && |
| 4100 |
ext4_should_order_data(inode)) |
| 4101 |
err = ext4_jbd2_inode_add_write(handle, inode, from, |
| 4102 |
length); |
| 4103 |
} |
| 4104 |
|
| 4105 |
unlock: |
| 4106 |
folio_unlock(folio); |
| 4107 |
folio_put(folio); |
| 4108 |
return err; |
| 4109 |
} |
| 4110 |
|
| 4111 |
/* |
| 4112 |
* ext4_block_zero_page_range() zeros out a mapping of length 'length' |
| 4113 |
* starting from file offset 'from'. The range to be zero'd must |
| 4114 |
* be contained with in one block. If the specified range exceeds |
| 4115 |
* the end of the block it will be shortened to end of the block |
| 4116 |
* that corresponds to 'from' |
| 4117 |
*/ |
| 4118 |
static int ext4_block_zero_page_range(handle_t *handle, |
| 4119 |
struct address_space *mapping, loff_t from, loff_t length) |
| 4120 |
{ |
| 4121 |
struct inode *inode = mapping->host; |
| 4122 |
unsigned blocksize = inode->i_sb->s_blocksize; |
| 4123 |
unsigned int max = blocksize - (from & (blocksize - 1)); |
| 4124 |
|
| 4125 |
/* |
| 4126 |
* correct length if it does not fall between |
| 4127 |
* 'from' and the end of the block |
| 4128 |
*/ |
| 4129 |
if (length > max || length < 0) |
| 4130 |
length = max; |
| 4131 |
|
| 4132 |
if (IS_DAX(inode)) { |
| 4133 |
return dax_zero_range(inode, from, length, NULL, |
| 4134 |
&ext4_iomap_ops); |
| 4135 |
} |
| 4136 |
return __ext4_block_zero_page_range(handle, mapping, from, length); |
| 4137 |
} |
| 4138 |
|
| 4139 |
/* |
| 4140 |
* ext4_block_truncate_page() zeroes out a mapping from file offset `from' |
| 4141 |
* up to the end of the block which corresponds to `from'. |
| 4142 |
* This required during truncate. We need to physically zero the tail end |
| 4143 |
* of that block so it doesn't yield old data if the file is later grown. |
| 4144 |
*/ |
| 4145 |
static int ext4_block_truncate_page(handle_t *handle, |
| 4146 |
struct address_space *mapping, loff_t from) |
| 4147 |
{ |
| 4148 |
unsigned length; |
| 4149 |
unsigned blocksize; |
| 4150 |
struct inode *inode = mapping->host; |
| 4151 |
|
| 4152 |
/* If we are processing an encrypted inode during orphan list handling */ |
| 4153 |
if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode)) |
| 4154 |
return 0; |
| 4155 |
|
| 4156 |
blocksize = i_blocksize(inode); |
| 4157 |
length = blocksize - (from & (blocksize - 1)); |
| 4158 |
|
| 4159 |
return ext4_block_zero_page_range(handle, mapping, from, length); |
| 4160 |
} |
| 4161 |
|
| 4162 |
int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, |
| 4163 |
loff_t lstart, loff_t length) |
| 4164 |
{ |
| 4165 |
struct super_block *sb = inode->i_sb; |
| 4166 |
struct address_space *mapping = inode->i_mapping; |
| 4167 |
unsigned partial_start, partial_end; |
| 4168 |
ext4_fsblk_t start, end; |
| 4169 |
loff_t byte_end = (lstart + length - 1); |
| 4170 |
int err = 0; |
| 4171 |
|
| 4172 |
partial_start = lstart & (sb->s_blocksize - 1); |
| 4173 |
partial_end = byte_end & (sb->s_blocksize - 1); |
| 4174 |
|
| 4175 |
start = lstart >> sb->s_blocksize_bits; |
| 4176 |
end = byte_end >> sb->s_blocksize_bits; |
| 4177 |
|
| 4178 |
/* Handle partial zero within the single block */ |
| 4179 |
if (start == end && |
| 4180 |
(partial_start || (partial_end != sb->s_blocksize - 1))) { |
| 4181 |
err = ext4_block_zero_page_range(handle, mapping, |
| 4182 |
lstart, length); |
| 4183 |
return err; |
| 4184 |
} |
| 4185 |
/* Handle partial zero out on the start of the range */ |
| 4186 |
if (partial_start) { |
| 4187 |
err = ext4_block_zero_page_range(handle, mapping, |
| 4188 |
lstart, sb->s_blocksize); |
| 4189 |
if (err) |
| 4190 |
return err; |
| 4191 |
} |
| 4192 |
/* Handle partial zero out on the end of the range */ |
| 4193 |
if (partial_end != sb->s_blocksize - 1) |
| 4194 |
err = ext4_block_zero_page_range(handle, mapping, |
| 4195 |
byte_end - partial_end, |
| 4196 |
partial_end + 1); |
| 4197 |
return err; |
| 4198 |
} |
| 4199 |
|
| 4200 |
int ext4_can_truncate(struct inode *inode) |
| 4201 |
{ |
| 4202 |
if (S_ISREG(inode->i_mode)) |
| 4203 |
return 1; |
| 4204 |
if (S_ISDIR(inode->i_mode)) |
| 4205 |
return 1; |
| 4206 |
if (S_ISLNK(inode->i_mode)) |
| 4207 |
return !ext4_inode_is_fast_symlink(inode); |
| 4208 |
return 0; |
| 4209 |
} |
| 4210 |
|
| 4211 |
/* |
| 4212 |
* We have to make sure i_disksize gets properly updated before we truncate |
| 4213 |
* page cache due to hole punching or zero range. Otherwise i_disksize update |
| 4214 |
* can get lost as it may have been postponed to submission of writeback but |
| 4215 |
* that will never happen if we remove the folio containing i_size from the |
| 4216 |
* page cache. Also if we punch hole within i_size but above i_disksize, |
| 4217 |
* following ext4_page_mkwrite() may mistakenly allocate written blocks over |
| 4218 |
* the hole and thus introduce allocated blocks beyond i_disksize which is |
| 4219 |
* not allowed (e2fsck would complain in case of crash). |
| 4220 |
*/ |
| 4221 |
int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset, |
| 4222 |
loff_t len) |
| 4223 |
{ |
| 4224 |
handle_t *handle; |
| 4225 |
int ret; |
| 4226 |
|
| 4227 |
loff_t size = i_size_read(inode); |
| 4228 |
|
| 4229 |
WARN_ON(!inode_is_locked(inode)); |
| 4230 |
if (offset > size) |
| 4231 |
return 0; |
| 4232 |
|
| 4233 |
if (offset + len < size) |
| 4234 |
size = offset + len; |
| 4235 |
if (EXT4_I(inode)->i_disksize >= size) |
| 4236 |
return 0; |
| 4237 |
|
| 4238 |
handle = ext4_journal_start(inode, EXT4_HT_MISC, 1); |
| 4239 |
if (IS_ERR(handle)) |
| 4240 |
return PTR_ERR(handle); |
| 4241 |
ext4_update_i_disksize(inode, size); |
| 4242 |
ret = ext4_mark_inode_dirty(handle, inode); |
| 4243 |
ext4_journal_stop(handle); |
| 4244 |
|
| 4245 |
return ret; |
| 4246 |
} |
| 4247 |
|
| 4248 |
static inline void ext4_truncate_folio(struct inode *inode, |
| 4249 |
loff_t start, loff_t end) |
| 4250 |
{ |
| 4251 |
unsigned long blocksize = i_blocksize(inode); |
| 4252 |
struct folio *folio; |
| 4253 |
|
| 4254 |
/* Nothing to be done if no complete block needs to be truncated. */ |
| 4255 |
if (round_up(start, blocksize) >= round_down(end, blocksize)) |
| 4256 |
return; |
| 4257 |
|
| 4258 |
folio = filemap_lock_folio(inode->i_mapping, start >> PAGE_SHIFT); |
| 4259 |
if (IS_ERR(folio)) |
| 4260 |
return; |
| 4261 |
|
| 4262 |
if (folio_mkclean(folio)) |
| 4263 |
folio_mark_dirty(folio); |
| 4264 |
folio_unlock(folio); |
| 4265 |
folio_put(folio); |
| 4266 |
} |
| 4267 |
|
| 4268 |
int ext4_truncate_page_cache_block_range(struct inode *inode, |
| 4269 |
loff_t start, loff_t end) |
| 4270 |
{ |
| 4271 |
unsigned long blocksize = i_blocksize(inode); |
| 4272 |
int ret; |
| 4273 |
|
| 4274 |
/* |
| 4275 |
* For journalled data we need to write (and checkpoint) pages |
| 4276 |
* before discarding page cache to avoid inconsitent data on disk |
| 4277 |
* in case of crash before freeing or unwritten converting trans |
| 4278 |
* is committed. |
| 4279 |
*/ |
| 4280 |
if (ext4_should_journal_data(inode)) { |
| 4281 |
ret = filemap_write_and_wait_range(inode->i_mapping, start, |
| 4282 |
end - 1); |
| 4283 |
if (ret) |
| 4284 |
return ret; |
| 4285 |
goto truncate_pagecache; |
| 4286 |
} |
| 4287 |
|
| 4288 |
/* |
| 4289 |
* If the block size is less than the page size, the file's mapped |
| 4290 |
* blocks within one page could be freed or converted to unwritten. |
| 4291 |
* So it's necessary to remove writable userspace mappings, and then |
| 4292 |
* ext4_page_mkwrite() can be called during subsequent write access |
| 4293 |
* to these partial folios. |
| 4294 |
*/ |
| 4295 |
if (!IS_ALIGNED(start | end, PAGE_SIZE) && |
| 4296 |
blocksize < PAGE_SIZE && start < inode->i_size) { |
| 4297 |
loff_t page_boundary = round_up(start, PAGE_SIZE); |
| 4298 |
|
| 4299 |
ext4_truncate_folio(inode, start, min(page_boundary, end)); |
| 4300 |
if (end > page_boundary) |
| 4301 |
ext4_truncate_folio(inode, |
| 4302 |
round_down(end, PAGE_SIZE), end); |
| 4303 |
} |
| 4304 |
|
| 4305 |
truncate_pagecache: |
| 4306 |
truncate_pagecache_range(inode, start, end - 1); |
| 4307 |
return 0; |
| 4308 |
} |
| 4309 |
|
| 4310 |
static void ext4_wait_dax_page(struct inode *inode) |
| 4311 |
{ |
| 4312 |
filemap_invalidate_unlock(inode->i_mapping); |
| 4313 |
schedule(); |
| 4314 |
filemap_invalidate_lock(inode->i_mapping); |
| 4315 |
} |
| 4316 |
|
| 4317 |
int ext4_break_layouts(struct inode *inode) |
| 4318 |
{ |
| 4319 |
if (WARN_ON_ONCE(!rwsem_is_locked(&inode->i_mapping->invalidate_lock))) |
| 4320 |
return -EINVAL; |
| 4321 |
|
| 4322 |
return dax_break_layout_inode(inode, ext4_wait_dax_page); |
| 4323 |
} |
| 4324 |
|
| 4325 |
/* |
| 4326 |
* ext4_punch_hole: punches a hole in a file by releasing the blocks |
| 4327 |
* associated with the given offset and length |
| 4328 |
* |
| 4329 |
* @inode: File inode |
| 4330 |
* @offset: The offset where the hole will begin |
| 4331 |
* @len: The length of the hole |
| 4332 |
* |
| 4333 |
* Returns: 0 on success or negative on failure |
| 4334 |
*/ |
| 4335 |
|
| 4336 |
int ext4_punch_hole(struct file *file, loff_t offset, loff_t length) |
| 4337 |
{ |
| 4338 |
struct inode *inode = file_inode(file); |
| 4339 |
struct super_block *sb = inode->i_sb; |
| 4340 |
ext4_lblk_t start_lblk, end_lblk; |
| 4341 |
loff_t max_end = sb->s_maxbytes; |
| 4342 |
loff_t end = offset + length; |
| 4343 |
handle_t *handle; |
| 4344 |
unsigned int credits; |
| 4345 |
int ret; |
| 4346 |
|
| 4347 |
trace_ext4_punch_hole(inode, offset, length, 0); |
| 4348 |
WARN_ON_ONCE(!inode_is_locked(inode)); |
| 4349 |
|
| 4350 |
/* |
| 4351 |
* For indirect-block based inodes, make sure that the hole within |
| 4352 |
* one block before last range. |
| 4353 |
*/ |
| 4354 |
if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 4355 |
max_end = EXT4_SB(sb)->s_bitmap_maxbytes - sb->s_blocksize; |
| 4356 |
|
| 4357 |
/* No need to punch hole beyond i_size */ |
| 4358 |
if (offset >= inode->i_size || offset >= max_end) |
| 4359 |
return 0; |
| 4360 |
|
| 4361 |
/* |
| 4362 |
* If the hole extends beyond i_size, set the hole to end after |
| 4363 |
* the block that contains i_size to save pointless tail block zeroing. |
| 4364 |
*/ |
| 4365 |
if (end >= inode->i_size) |
| 4366 |
end = round_up(inode->i_size, sb->s_blocksize); |
| 4367 |
if (end > max_end) |
| 4368 |
end = max_end; |
| 4369 |
length = end - offset; |
| 4370 |
|
| 4371 |
/* |
| 4372 |
* Attach jinode to inode for jbd2 if we do any zeroing of partial |
| 4373 |
* block. |
| 4374 |
*/ |
| 4375 |
if (!IS_ALIGNED(offset | end, sb->s_blocksize)) { |
| 4376 |
ret = ext4_inode_attach_jinode(inode); |
| 4377 |
if (ret < 0) |
| 4378 |
return ret; |
| 4379 |
} |
| 4380 |
|
| 4381 |
|
| 4382 |
ret = ext4_update_disksize_before_punch(inode, offset, length); |
| 4383 |
if (ret) |
| 4384 |
return ret; |
| 4385 |
|
| 4386 |
/* Now release the pages and zero block aligned part of pages*/ |
| 4387 |
ret = ext4_truncate_page_cache_block_range(inode, offset, end); |
| 4388 |
if (ret) |
| 4389 |
return ret; |
| 4390 |
|
| 4391 |
if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 4392 |
credits = ext4_chunk_trans_extent(inode, 2); |
| 4393 |
else |
| 4394 |
credits = ext4_blocks_for_truncate(inode); |
| 4395 |
handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); |
| 4396 |
if (IS_ERR(handle)) { |
| 4397 |
ret = PTR_ERR(handle); |
| 4398 |
ext4_std_error(sb, ret); |
| 4399 |
return ret; |
| 4400 |
} |
| 4401 |
|
| 4402 |
ret = ext4_zero_partial_blocks(handle, inode, offset, length); |
| 4403 |
if (ret) |
| 4404 |
goto out_handle; |
| 4405 |
|
| 4406 |
/* If there are blocks to remove, do it */ |
| 4407 |
start_lblk = EXT4_B_TO_LBLK(inode, offset); |
| 4408 |
end_lblk = end >> inode->i_blkbits; |
| 4409 |
|
| 4410 |
if (end_lblk > start_lblk) { |
| 4411 |
ext4_lblk_t hole_len = end_lblk - start_lblk; |
| 4412 |
|
| 4413 |
ext4_fc_track_inode(handle, inode); |
| 4414 |
ext4_check_map_extents_env(inode); |
| 4415 |
down_write(&EXT4_I(inode)->i_data_sem); |
| 4416 |
ext4_discard_preallocations(inode); |
| 4417 |
|
| 4418 |
ext4_es_remove_extent(inode, start_lblk, hole_len); |
| 4419 |
|
| 4420 |
if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 4421 |
ret = ext4_ext_remove_space(inode, start_lblk, |
| 4422 |
end_lblk - 1); |
| 4423 |
else |
| 4424 |
ret = ext4_ind_remove_space(handle, inode, start_lblk, |
| 4425 |
end_lblk); |
| 4426 |
if (ret) { |
| 4427 |
up_write(&EXT4_I(inode)->i_data_sem); |
| 4428 |
goto out_handle; |
| 4429 |
} |
| 4430 |
|
| 4431 |
ext4_es_insert_extent(inode, start_lblk, hole_len, ~0, |
| 4432 |
EXTENT_STATUS_HOLE, 0); |
| 4433 |
up_write(&EXT4_I(inode)->i_data_sem); |
| 4434 |
} |
| 4435 |
ext4_fc_track_range(handle, inode, start_lblk, end_lblk); |
| 4436 |
|
| 4437 |
ret = ext4_mark_inode_dirty(handle, inode); |
| 4438 |
if (unlikely(ret)) |
| 4439 |
goto out_handle; |
| 4440 |
|
| 4441 |
ext4_update_inode_fsync_trans(handle, inode, 1); |
| 4442 |
if (IS_SYNC(inode)) |
| 4443 |
ext4_handle_sync(handle); |
| 4444 |
out_handle: |
| 4445 |
ext4_journal_stop(handle); |
| 4446 |
return ret; |
| 4447 |
} |
| 4448 |
|
| 4449 |
int ext4_inode_attach_jinode(struct inode *inode) |
| 4450 |
{ |
| 4451 |
struct ext4_inode_info *ei = EXT4_I(inode); |
| 4452 |
struct jbd2_inode *jinode; |
| 4453 |
|
| 4454 |
if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal) |
| 4455 |
return 0; |
| 4456 |
|
| 4457 |
jinode = jbd2_alloc_inode(GFP_KERNEL); |
| 4458 |
spin_lock(&inode->i_lock); |
| 4459 |
if (!ei->jinode) { |
| 4460 |
if (!jinode) { |
| 4461 |
spin_unlock(&inode->i_lock); |
| 4462 |
return -ENOMEM; |
| 4463 |
} |
| 4464 |
jbd2_journal_init_jbd_inode(jinode, inode); |
| 4465 |
/* |
| 4466 |
* Publish ->jinode only after it is fully initialized so that |
| 4467 |
* readers never observe a partially initialized jbd2_inode. |
| 4468 |
*/ |
| 4469 |
smp_wmb(); |
| 4470 |
WRITE_ONCE(ei->jinode, jinode); |
| 4471 |
jinode = NULL; |
| 4472 |
} |
| 4473 |
spin_unlock(&inode->i_lock); |
| 4474 |
if (unlikely(jinode != NULL)) |
| 4475 |
jbd2_free_inode(jinode); |
| 4476 |
return 0; |
| 4477 |
} |
| 4478 |
|
| 4479 |
/* |
| 4480 |
* ext4_truncate() |
| 4481 |
* |
| 4482 |
* We block out ext4_get_block() block instantiations across the entire |
| 4483 |
* transaction, and VFS/VM ensures that ext4_truncate() cannot run |
| 4484 |
* simultaneously on behalf of the same inode. |
| 4485 |
* |
| 4486 |
* As we work through the truncate and commit bits of it to the journal there |
| 4487 |
* is one core, guiding principle: the file's tree must always be consistent on |
| 4488 |
* disk. We must be able to restart the truncate after a crash. |
| 4489 |
* |
| 4490 |
* The file's tree may be transiently inconsistent in memory (although it |
| 4491 |
* probably isn't), but whenever we close off and commit a journal transaction, |
| 4492 |
* the contents of (the filesystem + the journal) must be consistent and |
| 4493 |
* restartable. It's pretty simple, really: bottom up, right to left (although |
| 4494 |
* left-to-right works OK too). |
| 4495 |
* |
| 4496 |
* Note that at recovery time, journal replay occurs *before* the restart of |
| 4497 |
* truncate against the orphan inode list. |
| 4498 |
* |
| 4499 |
* The committed inode has the new, desired i_size (which is the same as |
| 4500 |
* i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
| 4501 |
* that this inode's truncate did not complete and it will again call |
| 4502 |
* ext4_truncate() to have another go. So there will be instantiated blocks |
| 4503 |
* to the right of the truncation point in a crashed ext4 filesystem. But |
| 4504 |
* that's fine - as long as they are linked from the inode, the post-crash |
| 4505 |
* ext4_truncate() run will find them and release them. |
| 4506 |
*/ |
| 4507 |
int ext4_truncate(struct inode *inode) |
| 4508 |
{ |
| 4509 |
struct ext4_inode_info *ei = EXT4_I(inode); |
| 4510 |
unsigned int credits; |
| 4511 |
int err = 0, err2; |
| 4512 |
handle_t *handle; |
| 4513 |
struct address_space *mapping = inode->i_mapping; |
| 4514 |
|
| 4515 |
/* |
| 4516 |
* There is a possibility that we're either freeing the inode |
| 4517 |
* or it's a completely new inode. In those cases we might not |
| 4518 |
* have i_rwsem locked because it's not necessary. |
| 4519 |
*/ |
| 4520 |
if (!(inode_state_read_once(inode) & (I_NEW | I_FREEING))) |
| 4521 |
WARN_ON(!inode_is_locked(inode)); |
| 4522 |
trace_ext4_truncate_enter(inode); |
| 4523 |
|
| 4524 |
if (!ext4_can_truncate(inode)) |
| 4525 |
goto out_trace; |
| 4526 |
|
| 4527 |
if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
| 4528 |
ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
| 4529 |
|
| 4530 |
if (ext4_has_inline_data(inode)) { |
| 4531 |
int has_inline = 1; |
| 4532 |
|
| 4533 |
err = ext4_inline_data_truncate(inode, &has_inline); |
| 4534 |
if (err || has_inline) |
| 4535 |
goto out_trace; |
| 4536 |
} |
| 4537 |
|
| 4538 |
/* If we zero-out tail of the page, we have to create jinode for jbd2 */ |
| 4539 |
if (inode->i_size & (inode->i_sb->s_blocksize - 1)) { |
| 4540 |
err = ext4_inode_attach_jinode(inode); |
| 4541 |
if (err) |
| 4542 |
goto out_trace; |
| 4543 |
} |
| 4544 |
|
| 4545 |
if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 4546 |
credits = ext4_chunk_trans_extent(inode, 1); |
| 4547 |
else |
| 4548 |
credits = ext4_blocks_for_truncate(inode); |
| 4549 |
|
| 4550 |
handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); |
| 4551 |
if (IS_ERR(handle)) { |
| 4552 |
err = PTR_ERR(handle); |
| 4553 |
goto out_trace; |
| 4554 |
} |
| 4555 |
|
| 4556 |
if (inode->i_size & (inode->i_sb->s_blocksize - 1)) |
| 4557 |
ext4_block_truncate_page(handle, mapping, inode->i_size); |
| 4558 |
|
| 4559 |
/* |
| 4560 |
* We add the inode to the orphan list, so that if this |
| 4561 |
* truncate spans multiple transactions, and we crash, we will |
| 4562 |
* resume the truncate when the filesystem recovers. It also |
| 4563 |
* marks the inode dirty, to catch the new size. |
| 4564 |
* |
| 4565 |
* Implication: the file must always be in a sane, consistent |
| 4566 |
* truncatable state while each transaction commits. |
| 4567 |
*/ |
| 4568 |
err = ext4_orphan_add(handle, inode); |
| 4569 |
if (err) |
| 4570 |
goto out_stop; |
| 4571 |
|
| 4572 |
ext4_fc_track_inode(handle, inode); |
| 4573 |
ext4_check_map_extents_env(inode); |
| 4574 |
|
| 4575 |
down_write(&EXT4_I(inode)->i_data_sem); |
| 4576 |
ext4_discard_preallocations(inode); |
| 4577 |
|
| 4578 |
if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 4579 |
err = ext4_ext_truncate(handle, inode); |
| 4580 |
else |
| 4581 |
ext4_ind_truncate(handle, inode); |
| 4582 |
|
| 4583 |
up_write(&ei->i_data_sem); |
| 4584 |
if (err) |
| 4585 |
goto out_stop; |
| 4586 |
|
| 4587 |
if (IS_SYNC(inode)) |
| 4588 |
ext4_handle_sync(handle); |
| 4589 |
|
| 4590 |
out_stop: |
| 4591 |
/* |
| 4592 |
* If this was a simple ftruncate() and the file will remain alive, |
| 4593 |
* then we need to clear up the orphan record which we created above. |
| 4594 |
* However, if this was a real unlink then we were called by |
| 4595 |
* ext4_evict_inode(), and we allow that function to clean up the |
| 4596 |
* orphan info for us. |
| 4597 |
*/ |
| 4598 |
if (inode->i_nlink) |
| 4599 |
ext4_orphan_del(handle, inode); |
| 4600 |
|
| 4601 |
inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode)); |
| 4602 |
err2 = ext4_mark_inode_dirty(handle, inode); |
| 4603 |
if (unlikely(err2 && !err)) |
| 4604 |
err = err2; |
| 4605 |
ext4_journal_stop(handle); |
| 4606 |
|
| 4607 |
out_trace: |
| 4608 |
trace_ext4_truncate_exit(inode); |
| 4609 |
return err; |
| 4610 |
} |
| 4611 |
|
| 4612 |
static inline u64 ext4_inode_peek_iversion(const struct inode *inode) |
| 4613 |
{ |
| 4614 |
if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) |
| 4615 |
return inode_peek_iversion_raw(inode); |
| 4616 |
else |
| 4617 |
return inode_peek_iversion(inode); |
| 4618 |
} |
| 4619 |
|
| 4620 |
static int ext4_inode_blocks_set(struct ext4_inode *raw_inode, |
| 4621 |
struct ext4_inode_info *ei) |
| 4622 |
{ |
| 4623 |
struct inode *inode = &(ei->vfs_inode); |
| 4624 |
u64 i_blocks = READ_ONCE(inode->i_blocks); |
| 4625 |
struct super_block *sb = inode->i_sb; |
| 4626 |
|
| 4627 |
if (i_blocks <= ~0U) { |
| 4628 |
/* |
| 4629 |
* i_blocks can be represented in a 32 bit variable |
| 4630 |
* as multiple of 512 bytes |
| 4631 |
*/ |
| 4632 |
raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
| 4633 |
raw_inode->i_blocks_high = 0; |
| 4634 |
ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
| 4635 |
return 0; |
| 4636 |
} |
| 4637 |
|
| 4638 |
/* |
| 4639 |
* This should never happen since sb->s_maxbytes should not have |
| 4640 |
* allowed this, sb->s_maxbytes was set according to the huge_file |
| 4641 |
* feature in ext4_fill_super(). |
| 4642 |
*/ |
| 4643 |
if (!ext4_has_feature_huge_file(sb)) |
| 4644 |
return -EFSCORRUPTED; |
| 4645 |
|
| 4646 |
if (i_blocks <= 0xffffffffffffULL) { |
| 4647 |
/* |
| 4648 |
* i_blocks can be represented in a 48 bit variable |
| 4649 |
* as multiple of 512 bytes |
| 4650 |
*/ |
| 4651 |
raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
| 4652 |
raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
| 4653 |
ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
| 4654 |
} else { |
| 4655 |
ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
| 4656 |
/* i_block is stored in file system block size */ |
| 4657 |
i_blocks = i_blocks >> (inode->i_blkbits - 9); |
| 4658 |
raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
| 4659 |
raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
| 4660 |
} |
| 4661 |
return 0; |
| 4662 |
} |
| 4663 |
|
| 4664 |
static int ext4_fill_raw_inode(struct inode *inode, struct ext4_inode *raw_inode) |
| 4665 |
{ |
| 4666 |
struct ext4_inode_info *ei = EXT4_I(inode); |
| 4667 |
uid_t i_uid; |
| 4668 |
gid_t i_gid; |
| 4669 |
projid_t i_projid; |
| 4670 |
int block; |
| 4671 |
int err; |
| 4672 |
|
| 4673 |
err = ext4_inode_blocks_set(raw_inode, ei); |
| 4674 |
|
| 4675 |
raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
| 4676 |
i_uid = i_uid_read(inode); |
| 4677 |
i_gid = i_gid_read(inode); |
| 4678 |
i_projid = from_kprojid(&init_user_ns, ei->i_projid); |
| 4679 |
if (!(test_opt(inode->i_sb, NO_UID32))) { |
| 4680 |
raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
| 4681 |
raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); |
| 4682 |
/* |
| 4683 |
* Fix up interoperability with old kernels. Otherwise, |
| 4684 |
* old inodes get re-used with the upper 16 bits of the |
| 4685 |
* uid/gid intact. |
| 4686 |
*/ |
| 4687 |
if (ei->i_dtime && !ext4_inode_orphan_tracked(inode)) { |
| 4688 |
raw_inode->i_uid_high = 0; |
| 4689 |
raw_inode->i_gid_high = 0; |
| 4690 |
} else { |
| 4691 |
raw_inode->i_uid_high = |
| 4692 |
cpu_to_le16(high_16_bits(i_uid)); |
| 4693 |
raw_inode->i_gid_high = |
| 4694 |
cpu_to_le16(high_16_bits(i_gid)); |
| 4695 |
} |
| 4696 |
} else { |
| 4697 |
raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
| 4698 |
raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); |
| 4699 |
raw_inode->i_uid_high = 0; |
| 4700 |
raw_inode->i_gid_high = 0; |
| 4701 |
} |
| 4702 |
raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); |
| 4703 |
|
| 4704 |
EXT4_INODE_SET_CTIME(inode, raw_inode); |
| 4705 |
EXT4_INODE_SET_MTIME(inode, raw_inode); |
| 4706 |
EXT4_INODE_SET_ATIME(inode, raw_inode); |
| 4707 |
EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); |
| 4708 |
|
| 4709 |
raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
| 4710 |
raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
| 4711 |
if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) |
| 4712 |
raw_inode->i_file_acl_high = |
| 4713 |
cpu_to_le16(ei->i_file_acl >> 32); |
| 4714 |
raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
| 4715 |
ext4_isize_set(raw_inode, ei->i_disksize); |
| 4716 |
|
| 4717 |
raw_inode->i_generation = cpu_to_le32(inode->i_generation); |
| 4718 |
if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { |
| 4719 |
if (old_valid_dev(inode->i_rdev)) { |
| 4720 |
raw_inode->i_block[0] = |
| 4721 |
cpu_to_le32(old_encode_dev(inode->i_rdev)); |
| 4722 |
raw_inode->i_block[1] = 0; |
| 4723 |
} else { |
| 4724 |
raw_inode->i_block[0] = 0; |
| 4725 |
raw_inode->i_block[1] = |
| 4726 |
cpu_to_le32(new_encode_dev(inode->i_rdev)); |
| 4727 |
raw_inode->i_block[2] = 0; |
| 4728 |
} |
| 4729 |
} else if (!ext4_has_inline_data(inode)) { |
| 4730 |
for (block = 0; block < EXT4_N_BLOCKS; block++) |
| 4731 |
raw_inode->i_block[block] = ei->i_data[block]; |
| 4732 |
} |
| 4733 |
|
| 4734 |
if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
| 4735 |
u64 ivers = ext4_inode_peek_iversion(inode); |
| 4736 |
|
| 4737 |
raw_inode->i_disk_version = cpu_to_le32(ivers); |
| 4738 |
if (ei->i_extra_isize) { |
| 4739 |
if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) |
| 4740 |
raw_inode->i_version_hi = |
| 4741 |
cpu_to_le32(ivers >> 32); |
| 4742 |
raw_inode->i_extra_isize = |
| 4743 |
cpu_to_le16(ei->i_extra_isize); |
| 4744 |
} |
| 4745 |
} |
| 4746 |
|
| 4747 |
if (i_projid != EXT4_DEF_PROJID && |
| 4748 |
!ext4_has_feature_project(inode->i_sb)) |
| 4749 |
err = err ?: -EFSCORRUPTED; |
| 4750 |
|
| 4751 |
if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && |
| 4752 |
EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) |
| 4753 |
raw_inode->i_projid = cpu_to_le32(i_projid); |
| 4754 |
|
| 4755 |
ext4_inode_csum_set(inode, raw_inode, ei); |
| 4756 |
return err; |
| 4757 |
} |
| 4758 |
|
| 4759 |
/* |
| 4760 |
* ext4_get_inode_loc returns with an extra refcount against the inode's |
| 4761 |
* underlying buffer_head on success. If we pass 'inode' and it does not |
| 4762 |
* have in-inode xattr, we have all inode data in memory that is needed |
| 4763 |
* to recreate the on-disk version of this inode. |
| 4764 |
*/ |
| 4765 |
static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino, |
| 4766 |
struct inode *inode, struct ext4_iloc *iloc, |
| 4767 |
ext4_fsblk_t *ret_block) |
| 4768 |
{ |
| 4769 |
struct ext4_group_desc *gdp; |
| 4770 |
struct buffer_head *bh; |
| 4771 |
ext4_fsblk_t block; |
| 4772 |
struct blk_plug plug; |
| 4773 |
int inodes_per_block, inode_offset; |
| 4774 |
|
| 4775 |
iloc->bh = NULL; |
| 4776 |
if (ino < EXT4_ROOT_INO || |
| 4777 |
ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)) |
| 4778 |
return -EFSCORRUPTED; |
| 4779 |
|
| 4780 |
iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
| 4781 |
gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); |
| 4782 |
if (!gdp) |
| 4783 |
return -EIO; |
| 4784 |
|
| 4785 |
/* |
| 4786 |
* Figure out the offset within the block group inode table |
| 4787 |
*/ |
| 4788 |
inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
| 4789 |
inode_offset = ((ino - 1) % |
| 4790 |
EXT4_INODES_PER_GROUP(sb)); |
| 4791 |
iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); |
| 4792 |
|
| 4793 |
block = ext4_inode_table(sb, gdp); |
| 4794 |
if ((block <= le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) || |
| 4795 |
(block >= ext4_blocks_count(EXT4_SB(sb)->s_es))) { |
| 4796 |
ext4_error(sb, "Invalid inode table block %llu in " |
| 4797 |
"block_group %u", block, iloc->block_group); |
| 4798 |
return -EFSCORRUPTED; |
| 4799 |
} |
| 4800 |
block += (inode_offset / inodes_per_block); |
| 4801 |
|
| 4802 |
bh = sb_getblk(sb, block); |
| 4803 |
if (unlikely(!bh)) |
| 4804 |
return -ENOMEM; |
| 4805 |
if (ext4_buffer_uptodate(bh)) |
| 4806 |
goto has_buffer; |
| 4807 |
|
| 4808 |
lock_buffer(bh); |
| 4809 |
if (ext4_buffer_uptodate(bh)) { |
| 4810 |
/* Someone brought it uptodate while we waited */ |
| 4811 |
unlock_buffer(bh); |
| 4812 |
goto has_buffer; |
| 4813 |
} |
| 4814 |
|
| 4815 |
/* |
| 4816 |
* If we have all information of the inode in memory and this |
| 4817 |
* is the only valid inode in the block, we need not read the |
| 4818 |
* block. |
| 4819 |
*/ |
| 4820 |
if (inode && !ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { |
| 4821 |
struct buffer_head *bitmap_bh; |
| 4822 |
int i, start; |
| 4823 |
|
| 4824 |
start = inode_offset & ~(inodes_per_block - 1); |
| 4825 |
|
| 4826 |
/* Is the inode bitmap in cache? */ |
| 4827 |
bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); |
| 4828 |
if (unlikely(!bitmap_bh)) |
| 4829 |
goto make_io; |
| 4830 |
|
| 4831 |
/* |
| 4832 |
* If the inode bitmap isn't in cache then the |
| 4833 |
* optimisation may end up performing two reads instead |
| 4834 |
* of one, so skip it. |
| 4835 |
*/ |
| 4836 |
if (!buffer_uptodate(bitmap_bh)) { |
| 4837 |
brelse(bitmap_bh); |
| 4838 |
goto make_io; |
| 4839 |
} |
| 4840 |
for (i = start; i < start + inodes_per_block; i++) { |
| 4841 |
if (i == inode_offset) |
| 4842 |
continue; |
| 4843 |
if (ext4_test_bit(i, bitmap_bh->b_data)) |
| 4844 |
break; |
| 4845 |
} |
| 4846 |
brelse(bitmap_bh); |
| 4847 |
if (i == start + inodes_per_block) { |
| 4848 |
struct ext4_inode *raw_inode = |
| 4849 |
(struct ext4_inode *) (bh->b_data + iloc->offset); |
| 4850 |
|
| 4851 |
/* all other inodes are free, so skip I/O */ |
| 4852 |
memset(bh->b_data, 0, bh->b_size); |
| 4853 |
if (!ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
| 4854 |
ext4_fill_raw_inode(inode, raw_inode); |
| 4855 |
set_buffer_uptodate(bh); |
| 4856 |
unlock_buffer(bh); |
| 4857 |
goto has_buffer; |
| 4858 |
} |
| 4859 |
} |
| 4860 |
|
| 4861 |
make_io: |
| 4862 |
/* |
| 4863 |
* If we need to do any I/O, try to pre-readahead extra |
| 4864 |
* blocks from the inode table. |
| 4865 |
*/ |
| 4866 |
blk_start_plug(&plug); |
| 4867 |
if (EXT4_SB(sb)->s_inode_readahead_blks) { |
| 4868 |
ext4_fsblk_t b, end, table; |
| 4869 |
unsigned num; |
| 4870 |
__u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; |
| 4871 |
|
| 4872 |
table = ext4_inode_table(sb, gdp); |
| 4873 |
/* s_inode_readahead_blks is always a power of 2 */ |
| 4874 |
b = block & ~((ext4_fsblk_t) ra_blks - 1); |
| 4875 |
if (table > b) |
| 4876 |
b = table; |
| 4877 |
end = b + ra_blks; |
| 4878 |
num = EXT4_INODES_PER_GROUP(sb); |
| 4879 |
if (ext4_has_group_desc_csum(sb)) |
| 4880 |
num -= ext4_itable_unused_count(sb, gdp); |
| 4881 |
table += num / inodes_per_block; |
| 4882 |
if (end > table) |
| 4883 |
end = table; |
| 4884 |
while (b <= end) |
| 4885 |
ext4_sb_breadahead_unmovable(sb, b++); |
| 4886 |
} |
| 4887 |
|
| 4888 |
/* |
| 4889 |
* There are other valid inodes in the buffer, this inode |
| 4890 |
* has in-inode xattrs, or we don't have this inode in memory. |
| 4891 |
* Read the block from disk. |
| 4892 |
*/ |
| 4893 |
trace_ext4_load_inode(sb, ino); |
| 4894 |
ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL, |
| 4895 |
ext4_simulate_fail(sb, EXT4_SIM_INODE_EIO)); |
| 4896 |
blk_finish_plug(&plug); |
| 4897 |
wait_on_buffer(bh); |
| 4898 |
if (!buffer_uptodate(bh)) { |
| 4899 |
if (ret_block) |
| 4900 |
*ret_block = block; |
| 4901 |
brelse(bh); |
| 4902 |
return -EIO; |
| 4903 |
} |
| 4904 |
has_buffer: |
| 4905 |
iloc->bh = bh; |
| 4906 |
return 0; |
| 4907 |
} |
| 4908 |
|
| 4909 |
static int __ext4_get_inode_loc_noinmem(struct inode *inode, |
| 4910 |
struct ext4_iloc *iloc) |
| 4911 |
{ |
| 4912 |
ext4_fsblk_t err_blk = 0; |
| 4913 |
int ret; |
| 4914 |
|
| 4915 |
ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, NULL, iloc, |
| 4916 |
&err_blk); |
| 4917 |
|
| 4918 |
if (ret == -EIO) |
| 4919 |
ext4_error_inode_block(inode, err_blk, EIO, |
| 4920 |
"unable to read itable block"); |
| 4921 |
|
| 4922 |
return ret; |
| 4923 |
} |
| 4924 |
|
| 4925 |
int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
| 4926 |
{ |
| 4927 |
ext4_fsblk_t err_blk = 0; |
| 4928 |
int ret; |
| 4929 |
|
| 4930 |
ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, inode, iloc, |
| 4931 |
&err_blk); |
| 4932 |
|
| 4933 |
if (ret == -EIO) |
| 4934 |
ext4_error_inode_block(inode, err_blk, EIO, |
| 4935 |
"unable to read itable block"); |
| 4936 |
|
| 4937 |
return ret; |
| 4938 |
} |
| 4939 |
|
| 4940 |
|
| 4941 |
int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino, |
| 4942 |
struct ext4_iloc *iloc) |
| 4943 |
{ |
| 4944 |
return __ext4_get_inode_loc(sb, ino, NULL, iloc, NULL); |
| 4945 |
} |
| 4946 |
|
| 4947 |
static bool ext4_should_enable_dax(struct inode *inode) |
| 4948 |
{ |
| 4949 |
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 4950 |
|
| 4951 |
if (test_opt2(inode->i_sb, DAX_NEVER)) |
| 4952 |
return false; |
| 4953 |
if (!S_ISREG(inode->i_mode)) |
| 4954 |
return false; |
| 4955 |
if (ext4_should_journal_data(inode)) |
| 4956 |
return false; |
| 4957 |
if (ext4_has_inline_data(inode)) |
| 4958 |
return false; |
| 4959 |
if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT)) |
| 4960 |
return false; |
| 4961 |
if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY)) |
| 4962 |
return false; |
| 4963 |
if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) |
| 4964 |
return false; |
| 4965 |
if (test_opt(inode->i_sb, DAX_ALWAYS)) |
| 4966 |
return true; |
| 4967 |
|
| 4968 |
return ext4_test_inode_flag(inode, EXT4_INODE_DAX); |
| 4969 |
} |
| 4970 |
|
| 4971 |
void ext4_set_inode_flags(struct inode *inode, bool init) |
| 4972 |
{ |
| 4973 |
unsigned int flags = EXT4_I(inode)->i_flags; |
| 4974 |
unsigned int new_fl = 0; |
| 4975 |
|
| 4976 |
WARN_ON_ONCE(IS_DAX(inode) && init); |
| 4977 |
|
| 4978 |
if (flags & EXT4_SYNC_FL) |
| 4979 |
new_fl |= S_SYNC; |
| 4980 |
if (flags & EXT4_APPEND_FL) |
| 4981 |
new_fl |= S_APPEND; |
| 4982 |
if (flags & EXT4_IMMUTABLE_FL) |
| 4983 |
new_fl |= S_IMMUTABLE; |
| 4984 |
if (flags & EXT4_NOATIME_FL) |
| 4985 |
new_fl |= S_NOATIME; |
| 4986 |
if (flags & EXT4_DIRSYNC_FL) |
| 4987 |
new_fl |= S_DIRSYNC; |
| 4988 |
|
| 4989 |
/* Because of the way inode_set_flags() works we must preserve S_DAX |
| 4990 |
* here if already set. */ |
| 4991 |
new_fl |= (inode->i_flags & S_DAX); |
| 4992 |
if (init && ext4_should_enable_dax(inode)) |
| 4993 |
new_fl |= S_DAX; |
| 4994 |
|
| 4995 |
if (flags & EXT4_ENCRYPT_FL) |
| 4996 |
new_fl |= S_ENCRYPTED; |
| 4997 |
if (flags & EXT4_CASEFOLD_FL) |
| 4998 |
new_fl |= S_CASEFOLD; |
| 4999 |
if (flags & EXT4_VERITY_FL) |
| 5000 |
new_fl |= S_VERITY; |
| 5001 |
inode_set_flags(inode, new_fl, |
| 5002 |
S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX| |
| 5003 |
S_ENCRYPTED|S_CASEFOLD|S_VERITY); |
| 5004 |
} |
| 5005 |
|
| 5006 |
static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
| 5007 |
struct ext4_inode_info *ei) |
| 5008 |
{ |
| 5009 |
blkcnt_t i_blocks ; |
| 5010 |
struct inode *inode = &(ei->vfs_inode); |
| 5011 |
struct super_block *sb = inode->i_sb; |
| 5012 |
|
| 5013 |
if (ext4_has_feature_huge_file(sb)) { |
| 5014 |
/* we are using combined 48 bit field */ |
| 5015 |
i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | |
| 5016 |
le32_to_cpu(raw_inode->i_blocks_lo); |
| 5017 |
if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
| 5018 |
/* i_blocks represent file system block size */ |
| 5019 |
return i_blocks << (inode->i_blkbits - 9); |
| 5020 |
} else { |
| 5021 |
return i_blocks; |
| 5022 |
} |
| 5023 |
} else { |
| 5024 |
return le32_to_cpu(raw_inode->i_blocks_lo); |
| 5025 |
} |
| 5026 |
} |
| 5027 |
|
| 5028 |
static inline int ext4_iget_extra_inode(struct inode *inode, |
| 5029 |
struct ext4_inode *raw_inode, |
| 5030 |
struct ext4_inode_info *ei) |
| 5031 |
{ |
| 5032 |
__le32 *magic = (void *)raw_inode + |
| 5033 |
EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; |
| 5034 |
|
| 5035 |
if (EXT4_INODE_HAS_XATTR_SPACE(inode) && |
| 5036 |
*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { |
| 5037 |
int err; |
| 5038 |
|
| 5039 |
err = xattr_check_inode(inode, IHDR(inode, raw_inode), |
| 5040 |
ITAIL(inode, raw_inode)); |
| 5041 |
if (err) |
| 5042 |
return err; |
| 5043 |
|
| 5044 |
ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
| 5045 |
err = ext4_find_inline_data_nolock(inode); |
| 5046 |
if (!err && ext4_has_inline_data(inode)) |
| 5047 |
ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); |
| 5048 |
return err; |
| 5049 |
} else |
| 5050 |
EXT4_I(inode)->i_inline_off = 0; |
| 5051 |
return 0; |
| 5052 |
} |
| 5053 |
|
| 5054 |
int ext4_get_projid(struct inode *inode, kprojid_t *projid) |
| 5055 |
{ |
| 5056 |
if (!ext4_has_feature_project(inode->i_sb)) |
| 5057 |
return -EOPNOTSUPP; |
| 5058 |
*projid = EXT4_I(inode)->i_projid; |
| 5059 |
return 0; |
| 5060 |
} |
| 5061 |
|
| 5062 |
/* |
| 5063 |
* ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of |
| 5064 |
* refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag |
| 5065 |
* set. |
| 5066 |
*/ |
| 5067 |
static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val) |
| 5068 |
{ |
| 5069 |
if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) |
| 5070 |
inode_set_iversion_raw(inode, val); |
| 5071 |
else |
| 5072 |
inode_set_iversion_queried(inode, val); |
| 5073 |
} |
| 5074 |
|
| 5075 |
static int check_igot_inode(struct inode *inode, ext4_iget_flags flags, |
| 5076 |
const char *function, unsigned int line) |
| 5077 |
{ |
| 5078 |
const char *err_str; |
| 5079 |
|
| 5080 |
if (flags & EXT4_IGET_EA_INODE) { |
| 5081 |
if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) { |
| 5082 |
err_str = "missing EA_INODE flag"; |
| 5083 |
goto error; |
| 5084 |
} |
| 5085 |
if (ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
| 5086 |
EXT4_I(inode)->i_file_acl) { |
| 5087 |
err_str = "ea_inode with extended attributes"; |
| 5088 |
goto error; |
| 5089 |
} |
| 5090 |
} else { |
| 5091 |
if ((EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) { |
| 5092 |
/* |
| 5093 |
* open_by_handle_at() could provide an old inode number |
| 5094 |
* that has since been reused for an ea_inode; this does |
| 5095 |
* not indicate filesystem corruption |
| 5096 |
*/ |
| 5097 |
if (flags & EXT4_IGET_HANDLE) |
| 5098 |
return -ESTALE; |
| 5099 |
err_str = "unexpected EA_INODE flag"; |
| 5100 |
goto error; |
| 5101 |
} |
| 5102 |
} |
| 5103 |
if (is_bad_inode(inode) && !(flags & EXT4_IGET_BAD)) { |
| 5104 |
err_str = "unexpected bad inode w/o EXT4_IGET_BAD"; |
| 5105 |
goto error; |
| 5106 |
} |
| 5107 |
return 0; |
| 5108 |
|
| 5109 |
error: |
| 5110 |
ext4_error_inode(inode, function, line, 0, "%s", err_str); |
| 5111 |
return -EFSCORRUPTED; |
| 5112 |
} |
| 5113 |
|
| 5114 |
void ext4_set_inode_mapping_order(struct inode *inode) |
| 5115 |
{ |
| 5116 |
struct super_block *sb = inode->i_sb; |
| 5117 |
u16 min_order, max_order; |
| 5118 |
|
| 5119 |
max_order = EXT4_SB(sb)->s_max_folio_order; |
| 5120 |
if (!max_order) |
| 5121 |
return; |
| 5122 |
|
| 5123 |
min_order = EXT4_SB(sb)->s_min_folio_order; |
| 5124 |
if (!min_order && !S_ISREG(inode->i_mode)) |
| 5125 |
return; |
| 5126 |
|
| 5127 |
if (ext4_test_inode_flag(inode, EXT4_INODE_JOURNAL_DATA)) |
| 5128 |
max_order = min_order; |
| 5129 |
|
| 5130 |
mapping_set_folio_order_range(inode->i_mapping, min_order, max_order); |
| 5131 |
} |
| 5132 |
|
| 5133 |
struct inode *__ext4_iget(struct super_block *sb, unsigned long ino, |
| 5134 |
ext4_iget_flags flags, const char *function, |
| 5135 |
unsigned int line) |
| 5136 |
{ |
| 5137 |
struct ext4_iloc iloc; |
| 5138 |
struct ext4_inode *raw_inode; |
| 5139 |
struct ext4_inode_info *ei; |
| 5140 |
struct ext4_super_block *es = EXT4_SB(sb)->s_es; |
| 5141 |
struct inode *inode; |
| 5142 |
journal_t *journal = EXT4_SB(sb)->s_journal; |
| 5143 |
long ret; |
| 5144 |
loff_t size; |
| 5145 |
int block; |
| 5146 |
uid_t i_uid; |
| 5147 |
gid_t i_gid; |
| 5148 |
projid_t i_projid; |
| 5149 |
|
| 5150 |
if ((!(flags & EXT4_IGET_SPECIAL) && is_special_ino(sb, ino)) || |
| 5151 |
(ino < EXT4_ROOT_INO) || |
| 5152 |
(ino > le32_to_cpu(es->s_inodes_count))) { |
| 5153 |
if (flags & EXT4_IGET_HANDLE) |
| 5154 |
return ERR_PTR(-ESTALE); |
| 5155 |
__ext4_error(sb, function, line, false, EFSCORRUPTED, 0, |
| 5156 |
"inode #%lu: comm %s: iget: illegal inode #", |
| 5157 |
ino, current->comm); |
| 5158 |
return ERR_PTR(-EFSCORRUPTED); |
| 5159 |
} |
| 5160 |
|
| 5161 |
inode = iget_locked(sb, ino); |
| 5162 |
if (!inode) |
| 5163 |
return ERR_PTR(-ENOMEM); |
| 5164 |
if (!(inode_state_read_once(inode) & I_NEW)) { |
| 5165 |
ret = check_igot_inode(inode, flags, function, line); |
| 5166 |
if (ret) { |
| 5167 |
iput(inode); |
| 5168 |
return ERR_PTR(ret); |
| 5169 |
} |
| 5170 |
return inode; |
| 5171 |
} |
| 5172 |
|
| 5173 |
ei = EXT4_I(inode); |
| 5174 |
iloc.bh = NULL; |
| 5175 |
|
| 5176 |
ret = __ext4_get_inode_loc_noinmem(inode, &iloc); |
| 5177 |
if (ret < 0) |
| 5178 |
goto bad_inode; |
| 5179 |
raw_inode = ext4_raw_inode(&iloc); |
| 5180 |
|
| 5181 |
if ((flags & EXT4_IGET_HANDLE) && |
| 5182 |
(raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) { |
| 5183 |
ret = -ESTALE; |
| 5184 |
goto bad_inode; |
| 5185 |
} |
| 5186 |
|
| 5187 |
if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
| 5188 |
ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); |
| 5189 |
if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > |
| 5190 |
EXT4_INODE_SIZE(inode->i_sb) || |
| 5191 |
(ei->i_extra_isize & 3)) { |
| 5192 |
ext4_error_inode(inode, function, line, 0, |
| 5193 |
"iget: bad extra_isize %u " |
| 5194 |
"(inode size %u)", |
| 5195 |
ei->i_extra_isize, |
| 5196 |
EXT4_INODE_SIZE(inode->i_sb)); |
| 5197 |
ret = -EFSCORRUPTED; |
| 5198 |
goto bad_inode; |
| 5199 |
} |
| 5200 |
} else |
| 5201 |
ei->i_extra_isize = 0; |
| 5202 |
|
| 5203 |
/* Precompute checksum seed for inode metadata */ |
| 5204 |
if (ext4_has_feature_metadata_csum(sb)) { |
| 5205 |
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 5206 |
__u32 csum; |
| 5207 |
__le32 inum = cpu_to_le32(inode->i_ino); |
| 5208 |
__le32 gen = raw_inode->i_generation; |
| 5209 |
csum = ext4_chksum(sbi->s_csum_seed, (__u8 *)&inum, |
| 5210 |
sizeof(inum)); |
| 5211 |
ei->i_csum_seed = ext4_chksum(csum, (__u8 *)&gen, sizeof(gen)); |
| 5212 |
} |
| 5213 |
|
| 5214 |
if ((!ext4_inode_csum_verify(inode, raw_inode, ei) || |
| 5215 |
ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) && |
| 5216 |
(!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) { |
| 5217 |
ext4_error_inode_err(inode, function, line, 0, |
| 5218 |
EFSBADCRC, "iget: checksum invalid"); |
| 5219 |
ret = -EFSBADCRC; |
| 5220 |
goto bad_inode; |
| 5221 |
} |
| 5222 |
|
| 5223 |
inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
| 5224 |
i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
| 5225 |
i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); |
| 5226 |
if (ext4_has_feature_project(sb) && |
| 5227 |
EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE && |
| 5228 |
EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) |
| 5229 |
i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid); |
| 5230 |
else |
| 5231 |
i_projid = EXT4_DEF_PROJID; |
| 5232 |
|
| 5233 |
if (!(test_opt(inode->i_sb, NO_UID32))) { |
| 5234 |
i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
| 5235 |
i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; |
| 5236 |
} |
| 5237 |
i_uid_write(inode, i_uid); |
| 5238 |
i_gid_write(inode, i_gid); |
| 5239 |
ei->i_projid = make_kprojid(&init_user_ns, i_projid); |
| 5240 |
set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
| 5241 |
|
| 5242 |
ei->i_inline_off = 0; |
| 5243 |
ei->i_dir_start_lookup = 0; |
| 5244 |
ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); |
| 5245 |
/* We now have enough fields to check if the inode was active or not. |
| 5246 |
* This is needed because nfsd might try to access dead inodes |
| 5247 |
* the test is that same one that e2fsck uses |
| 5248 |
* NeilBrown 1999oct15 |
| 5249 |
*/ |
| 5250 |
if (inode->i_nlink == 0) { |
| 5251 |
if ((inode->i_mode == 0 || flags & EXT4_IGET_SPECIAL || |
| 5252 |
!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && |
| 5253 |
ino != EXT4_BOOT_LOADER_INO) { |
| 5254 |
/* this inode is deleted or unallocated */ |
| 5255 |
if (flags & EXT4_IGET_SPECIAL) { |
| 5256 |
ext4_error_inode(inode, function, line, 0, |
| 5257 |
"iget: special inode unallocated"); |
| 5258 |
ret = -EFSCORRUPTED; |
| 5259 |
} else |
| 5260 |
ret = -ESTALE; |
| 5261 |
goto bad_inode; |
| 5262 |
} |
| 5263 |
/* The only unlinked inodes we let through here have |
| 5264 |
* valid i_mode and are being read by the orphan |
| 5265 |
* recovery code: that's fine, we're about to complete |
| 5266 |
* the process of deleting those. |
| 5267 |
* OR it is the EXT4_BOOT_LOADER_INO which is |
| 5268 |
* not initialized on a new filesystem. */ |
| 5269 |
} |
| 5270 |
ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
| 5271 |
ext4_set_inode_flags(inode, true); |
| 5272 |
/* Detect invalid flag combination - can't have both inline data and extents */ |
| 5273 |
if (ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA) && |
| 5274 |
ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
| 5275 |
ext4_error_inode(inode, function, line, 0, |
| 5276 |
"inode has both inline data and extents flags"); |
| 5277 |
ret = -EFSCORRUPTED; |
| 5278 |
goto bad_inode; |
| 5279 |
} |
| 5280 |
inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
| 5281 |
ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
| 5282 |
if (ext4_has_feature_64bit(sb)) |
| 5283 |
ei->i_file_acl |= |
| 5284 |
((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; |
| 5285 |
inode->i_size = ext4_isize(sb, raw_inode); |
| 5286 |
size = i_size_read(inode); |
| 5287 |
if (size < 0 || size > ext4_get_maxbytes(inode)) { |
| 5288 |
ext4_error_inode(inode, function, line, 0, |
| 5289 |
"iget: bad i_size value: %lld", size); |
| 5290 |
ret = -EFSCORRUPTED; |
| 5291 |
goto bad_inode; |
| 5292 |
} |
| 5293 |
/* |
| 5294 |
* If dir_index is not enabled but there's dir with INDEX flag set, |
| 5295 |
* we'd normally treat htree data as empty space. But with metadata |
| 5296 |
* checksumming that corrupts checksums so forbid that. |
| 5297 |
*/ |
| 5298 |
if (!ext4_has_feature_dir_index(sb) && |
| 5299 |
ext4_has_feature_metadata_csum(sb) && |
| 5300 |
ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) { |
| 5301 |
ext4_error_inode(inode, function, line, 0, |
| 5302 |
"iget: Dir with htree data on filesystem without dir_index feature."); |
| 5303 |
ret = -EFSCORRUPTED; |
| 5304 |
goto bad_inode; |
| 5305 |
} |
| 5306 |
ei->i_disksize = inode->i_size; |
| 5307 |
#ifdef CONFIG_QUOTA |
| 5308 |
ei->i_reserved_quota = 0; |
| 5309 |
#endif |
| 5310 |
inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
| 5311 |
ei->i_block_group = iloc.block_group; |
| 5312 |
ei->i_last_alloc_group = ~0; |
| 5313 |
/* |
| 5314 |
* NOTE! The in-memory inode i_data array is in little-endian order |
| 5315 |
* even on big-endian machines: we do NOT byteswap the block numbers! |
| 5316 |
*/ |
| 5317 |
for (block = 0; block < EXT4_N_BLOCKS; block++) |
| 5318 |
ei->i_data[block] = raw_inode->i_block[block]; |
| 5319 |
INIT_LIST_HEAD(&ei->i_orphan); |
| 5320 |
ext4_fc_init_inode(&ei->vfs_inode); |
| 5321 |
|
| 5322 |
/* |
| 5323 |
* Set transaction id's of transactions that have to be committed |
| 5324 |
* to finish f[data]sync. We set them to currently running transaction |
| 5325 |
* as we cannot be sure that the inode or some of its metadata isn't |
| 5326 |
* part of the transaction - the inode could have been reclaimed and |
| 5327 |
* now it is reread from disk. |
| 5328 |
*/ |
| 5329 |
if (journal) { |
| 5330 |
transaction_t *transaction; |
| 5331 |
tid_t tid; |
| 5332 |
|
| 5333 |
read_lock(&journal->j_state_lock); |
| 5334 |
if (journal->j_running_transaction) |
| 5335 |
transaction = journal->j_running_transaction; |
| 5336 |
else |
| 5337 |
transaction = journal->j_committing_transaction; |
| 5338 |
if (transaction) |
| 5339 |
tid = transaction->t_tid; |
| 5340 |
else |
| 5341 |
tid = journal->j_commit_sequence; |
| 5342 |
read_unlock(&journal->j_state_lock); |
| 5343 |
ei->i_sync_tid = tid; |
| 5344 |
ei->i_datasync_tid = tid; |
| 5345 |
} |
| 5346 |
|
| 5347 |
if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
| 5348 |
if (ei->i_extra_isize == 0) { |
| 5349 |
/* The extra space is currently unused. Use it. */ |
| 5350 |
BUILD_BUG_ON(sizeof(struct ext4_inode) & 3); |
| 5351 |
ei->i_extra_isize = sizeof(struct ext4_inode) - |
| 5352 |
EXT4_GOOD_OLD_INODE_SIZE; |
| 5353 |
} else { |
| 5354 |
ret = ext4_iget_extra_inode(inode, raw_inode, ei); |
| 5355 |
if (ret) |
| 5356 |
goto bad_inode; |
| 5357 |
} |
| 5358 |
} |
| 5359 |
|
| 5360 |
EXT4_INODE_GET_CTIME(inode, raw_inode); |
| 5361 |
EXT4_INODE_GET_ATIME(inode, raw_inode); |
| 5362 |
EXT4_INODE_GET_MTIME(inode, raw_inode); |
| 5363 |
EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); |
| 5364 |
|
| 5365 |
if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
| 5366 |
u64 ivers = le32_to_cpu(raw_inode->i_disk_version); |
| 5367 |
|
| 5368 |
if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
| 5369 |
if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) |
| 5370 |
ivers |= |
| 5371 |
(__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; |
| 5372 |
} |
| 5373 |
ext4_inode_set_iversion_queried(inode, ivers); |
| 5374 |
} |
| 5375 |
|
| 5376 |
ret = 0; |
| 5377 |
if (ei->i_file_acl && |
| 5378 |
!ext4_inode_block_valid(inode, ei->i_file_acl, 1)) { |
| 5379 |
ext4_error_inode(inode, function, line, 0, |
| 5380 |
"iget: bad extended attribute block %llu", |
| 5381 |
ei->i_file_acl); |
| 5382 |
ret = -EFSCORRUPTED; |
| 5383 |
goto bad_inode; |
| 5384 |
} else if (!ext4_has_inline_data(inode)) { |
| 5385 |
/* validate the block references in the inode */ |
| 5386 |
if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) && |
| 5387 |
(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
| 5388 |
(S_ISLNK(inode->i_mode) && |
| 5389 |
!ext4_inode_is_fast_symlink(inode)))) { |
| 5390 |
if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 5391 |
ret = ext4_ext_check_inode(inode); |
| 5392 |
else |
| 5393 |
ret = ext4_ind_check_inode(inode); |
| 5394 |
} |
| 5395 |
} |
| 5396 |
if (ret) |
| 5397 |
goto bad_inode; |
| 5398 |
|
| 5399 |
if (S_ISREG(inode->i_mode)) { |
| 5400 |
inode->i_op = &ext4_file_inode_operations; |
| 5401 |
inode->i_fop = &ext4_file_operations; |
| 5402 |
ext4_set_aops(inode); |
| 5403 |
} else if (S_ISDIR(inode->i_mode)) { |
| 5404 |
inode->i_op = &ext4_dir_inode_operations; |
| 5405 |
inode->i_fop = &ext4_dir_operations; |
| 5406 |
} else if (S_ISLNK(inode->i_mode)) { |
| 5407 |
/* VFS does not allow setting these so must be corruption */ |
| 5408 |
if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) { |
| 5409 |
ext4_error_inode(inode, function, line, 0, |
| 5410 |
"iget: immutable or append flags " |
| 5411 |
"not allowed on symlinks"); |
| 5412 |
ret = -EFSCORRUPTED; |
| 5413 |
goto bad_inode; |
| 5414 |
} |
| 5415 |
if (IS_ENCRYPTED(inode)) { |
| 5416 |
inode->i_op = &ext4_encrypted_symlink_inode_operations; |
| 5417 |
} else if (ext4_inode_is_fast_symlink(inode)) { |
| 5418 |
inode->i_op = &ext4_fast_symlink_inode_operations; |
| 5419 |
|
| 5420 |
/* |
| 5421 |
* Orphan cleanup can see inodes with i_size == 0 |
| 5422 |
* and i_data uninitialized. Skip size checks in |
| 5423 |
* that case. This is safe because the first thing |
| 5424 |
* ext4_evict_inode() does for fast symlinks is |
| 5425 |
* clearing of i_data and i_size. |
| 5426 |
*/ |
| 5427 |
if ((EXT4_SB(sb)->s_mount_state & EXT4_ORPHAN_FS)) { |
| 5428 |
if (inode->i_nlink != 0) { |
| 5429 |
ext4_error_inode(inode, function, line, 0, |
| 5430 |
"invalid orphan symlink nlink %d", |
| 5431 |
inode->i_nlink); |
| 5432 |
ret = -EFSCORRUPTED; |
| 5433 |
goto bad_inode; |
| 5434 |
} |
| 5435 |
} else { |
| 5436 |
if (inode->i_size == 0 || |
| 5437 |
inode->i_size >= sizeof(ei->i_data) || |
| 5438 |
strnlen((char *)ei->i_data, inode->i_size + 1) != |
| 5439 |
inode->i_size) { |
| 5440 |
ext4_error_inode(inode, function, line, 0, |
| 5441 |
"invalid fast symlink length %llu", |
| 5442 |
(unsigned long long)inode->i_size); |
| 5443 |
ret = -EFSCORRUPTED; |
| 5444 |
goto bad_inode; |
| 5445 |
} |
| 5446 |
inode_set_cached_link(inode, (char *)ei->i_data, |
| 5447 |
inode->i_size); |
| 5448 |
} |
| 5449 |
} else { |
| 5450 |
inode->i_op = &ext4_symlink_inode_operations; |
| 5451 |
} |
| 5452 |
} else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
| 5453 |
S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { |
| 5454 |
inode->i_op = &ext4_special_inode_operations; |
| 5455 |
if (raw_inode->i_block[0]) |
| 5456 |
init_special_inode(inode, inode->i_mode, |
| 5457 |
old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); |
| 5458 |
else |
| 5459 |
init_special_inode(inode, inode->i_mode, |
| 5460 |
new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); |
| 5461 |
} else if (ino == EXT4_BOOT_LOADER_INO) { |
| 5462 |
make_bad_inode(inode); |
| 5463 |
} else { |
| 5464 |
ret = -EFSCORRUPTED; |
| 5465 |
ext4_error_inode(inode, function, line, 0, |
| 5466 |
"iget: bogus i_mode (%o)", inode->i_mode); |
| 5467 |
goto bad_inode; |
| 5468 |
} |
| 5469 |
if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb)) { |
| 5470 |
ext4_error_inode(inode, function, line, 0, |
| 5471 |
"casefold flag without casefold feature"); |
| 5472 |
ret = -EFSCORRUPTED; |
| 5473 |
goto bad_inode; |
| 5474 |
} |
| 5475 |
|
| 5476 |
ext4_set_inode_mapping_order(inode); |
| 5477 |
|
| 5478 |
ret = check_igot_inode(inode, flags, function, line); |
| 5479 |
/* |
| 5480 |
* -ESTALE here means there is nothing inherently wrong with the inode, |
| 5481 |
* it's just not an inode we can return for an fhandle lookup. |
| 5482 |
*/ |
| 5483 |
if (ret == -ESTALE) { |
| 5484 |
brelse(iloc.bh); |
| 5485 |
unlock_new_inode(inode); |
| 5486 |
iput(inode); |
| 5487 |
return ERR_PTR(-ESTALE); |
| 5488 |
} |
| 5489 |
if (ret) |
| 5490 |
goto bad_inode; |
| 5491 |
brelse(iloc.bh); |
| 5492 |
/* Initialize the "no ACL's" state for the simple cases */ |
| 5493 |
if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) && !ei->i_file_acl) |
| 5494 |
cache_no_acl(inode); |
| 5495 |
unlock_new_inode(inode); |
| 5496 |
return inode; |
| 5497 |
|
| 5498 |
bad_inode: |
| 5499 |
brelse(iloc.bh); |
| 5500 |
iget_failed(inode); |
| 5501 |
return ERR_PTR(ret); |
| 5502 |
} |
| 5503 |
|
| 5504 |
static void __ext4_update_other_inode_time(struct super_block *sb, |
| 5505 |
unsigned long orig_ino, |
| 5506 |
unsigned long ino, |
| 5507 |
struct ext4_inode *raw_inode) |
| 5508 |
{ |
| 5509 |
struct inode *inode; |
| 5510 |
|
| 5511 |
inode = find_inode_by_ino_rcu(sb, ino); |
| 5512 |
if (!inode) |
| 5513 |
return; |
| 5514 |
|
| 5515 |
if (!inode_is_dirtytime_only(inode)) |
| 5516 |
return; |
| 5517 |
|
| 5518 |
spin_lock(&inode->i_lock); |
| 5519 |
if (inode_is_dirtytime_only(inode)) { |
| 5520 |
struct ext4_inode_info *ei = EXT4_I(inode); |
| 5521 |
|
| 5522 |
inode_state_clear(inode, I_DIRTY_TIME); |
| 5523 |
spin_unlock(&inode->i_lock); |
| 5524 |
|
| 5525 |
spin_lock(&ei->i_raw_lock); |
| 5526 |
EXT4_INODE_SET_CTIME(inode, raw_inode); |
| 5527 |
EXT4_INODE_SET_MTIME(inode, raw_inode); |
| 5528 |
EXT4_INODE_SET_ATIME(inode, raw_inode); |
| 5529 |
ext4_inode_csum_set(inode, raw_inode, ei); |
| 5530 |
spin_unlock(&ei->i_raw_lock); |
| 5531 |
trace_ext4_other_inode_update_time(inode, orig_ino); |
| 5532 |
return; |
| 5533 |
} |
| 5534 |
spin_unlock(&inode->i_lock); |
| 5535 |
} |
| 5536 |
|
| 5537 |
/* |
| 5538 |
* Opportunistically update the other time fields for other inodes in |
| 5539 |
* the same inode table block. |
| 5540 |
*/ |
| 5541 |
static void ext4_update_other_inodes_time(struct super_block *sb, |
| 5542 |
unsigned long orig_ino, char *buf) |
| 5543 |
{ |
| 5544 |
unsigned long ino; |
| 5545 |
int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
| 5546 |
int inode_size = EXT4_INODE_SIZE(sb); |
| 5547 |
|
| 5548 |
/* |
| 5549 |
* Calculate the first inode in the inode table block. Inode |
| 5550 |
* numbers are one-based. That is, the first inode in a block |
| 5551 |
* (assuming 4k blocks and 256 byte inodes) is (n*16 + 1). |
| 5552 |
*/ |
| 5553 |
ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1; |
| 5554 |
rcu_read_lock(); |
| 5555 |
for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) { |
| 5556 |
if (ino == orig_ino) |
| 5557 |
continue; |
| 5558 |
__ext4_update_other_inode_time(sb, orig_ino, ino, |
| 5559 |
(struct ext4_inode *)buf); |
| 5560 |
} |
| 5561 |
rcu_read_unlock(); |
| 5562 |
} |
| 5563 |
|
| 5564 |
/* |
| 5565 |
* Post the struct inode info into an on-disk inode location in the |
| 5566 |
* buffer-cache. This gobbles the caller's reference to the |
| 5567 |
* buffer_head in the inode location struct. |
| 5568 |
* |
| 5569 |
* The caller must have write access to iloc->bh. |
| 5570 |
*/ |
| 5571 |
static int ext4_do_update_inode(handle_t *handle, |
| 5572 |
struct inode *inode, |
| 5573 |
struct ext4_iloc *iloc) |
| 5574 |
{ |
| 5575 |
struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
| 5576 |
struct ext4_inode_info *ei = EXT4_I(inode); |
| 5577 |
struct buffer_head *bh = iloc->bh; |
| 5578 |
struct super_block *sb = inode->i_sb; |
| 5579 |
int err; |
| 5580 |
int need_datasync = 0, set_large_file = 0; |
| 5581 |
|
| 5582 |
spin_lock(&ei->i_raw_lock); |
| 5583 |
|
| 5584 |
/* |
| 5585 |
* For fields not tracked in the in-memory inode, initialise them |
| 5586 |
* to zero for new inodes. |
| 5587 |
*/ |
| 5588 |
if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
| 5589 |
memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
| 5590 |
|
| 5591 |
if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode)) |
| 5592 |
need_datasync = 1; |
| 5593 |
if (ei->i_disksize > 0x7fffffffULL) { |
| 5594 |
if (!ext4_has_feature_large_file(sb) || |
| 5595 |
EXT4_SB(sb)->s_es->s_rev_level == cpu_to_le32(EXT4_GOOD_OLD_REV)) |
| 5596 |
set_large_file = 1; |
| 5597 |
} |
| 5598 |
|
| 5599 |
err = ext4_fill_raw_inode(inode, raw_inode); |
| 5600 |
spin_unlock(&ei->i_raw_lock); |
| 5601 |
if (err) { |
| 5602 |
EXT4_ERROR_INODE(inode, "corrupted inode contents"); |
| 5603 |
goto out_brelse; |
| 5604 |
} |
| 5605 |
|
| 5606 |
if (inode->i_sb->s_flags & SB_LAZYTIME) |
| 5607 |
ext4_update_other_inodes_time(inode->i_sb, inode->i_ino, |
| 5608 |
bh->b_data); |
| 5609 |
|
| 5610 |
BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
| 5611 |
err = ext4_handle_dirty_metadata(handle, NULL, bh); |
| 5612 |
if (err) |
| 5613 |
goto out_error; |
| 5614 |
ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
| 5615 |
if (set_large_file) { |
| 5616 |
BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access"); |
| 5617 |
err = ext4_journal_get_write_access(handle, sb, |
| 5618 |
EXT4_SB(sb)->s_sbh, |
| 5619 |
EXT4_JTR_NONE); |
| 5620 |
if (err) |
| 5621 |
goto out_error; |
| 5622 |
lock_buffer(EXT4_SB(sb)->s_sbh); |
| 5623 |
ext4_set_feature_large_file(sb); |
| 5624 |
ext4_superblock_csum_set(sb); |
| 5625 |
unlock_buffer(EXT4_SB(sb)->s_sbh); |
| 5626 |
ext4_handle_sync(handle); |
| 5627 |
err = ext4_handle_dirty_metadata(handle, NULL, |
| 5628 |
EXT4_SB(sb)->s_sbh); |
| 5629 |
} |
| 5630 |
ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
| 5631 |
out_error: |
| 5632 |
ext4_std_error(inode->i_sb, err); |
| 5633 |
out_brelse: |
| 5634 |
brelse(bh); |
| 5635 |
return err; |
| 5636 |
} |
| 5637 |
|
| 5638 |
/* |
| 5639 |
* ext4_write_inode() |
| 5640 |
* |
| 5641 |
* We are called from a few places: |
| 5642 |
* |
| 5643 |
* - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files. |
| 5644 |
* Here, there will be no transaction running. We wait for any running |
| 5645 |
* transaction to commit. |
| 5646 |
* |
| 5647 |
* - Within flush work (sys_sync(), kupdate and such). |
| 5648 |
* We wait on commit, if told to. |
| 5649 |
* |
| 5650 |
* - Within iput_final() -> write_inode_now() |
| 5651 |
* We wait on commit, if told to. |
| 5652 |
* |
| 5653 |
* In all cases it is actually safe for us to return without doing anything, |
| 5654 |
* because the inode has been copied into a raw inode buffer in |
| 5655 |
* ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL |
| 5656 |
* writeback. |
| 5657 |
* |
| 5658 |
* Note that we are absolutely dependent upon all inode dirtiers doing the |
| 5659 |
* right thing: they *must* call mark_inode_dirty() after dirtying info in |
| 5660 |
* which we are interested. |
| 5661 |
* |
| 5662 |
* It would be a bug for them to not do this. The code: |
| 5663 |
* |
| 5664 |
* mark_inode_dirty(inode) |
| 5665 |
* stuff(); |
| 5666 |
* inode->i_size = expr; |
| 5667 |
* |
| 5668 |
* is in error because write_inode() could occur while `stuff()' is running, |
| 5669 |
* and the new i_size will be lost. Plus the inode will no longer be on the |
| 5670 |
* superblock's dirty inode list. |
| 5671 |
*/ |
| 5672 |
int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
| 5673 |
{ |
| 5674 |
int err; |
| 5675 |
|
| 5676 |
if (WARN_ON_ONCE(current->flags & PF_MEMALLOC)) |
| 5677 |
return 0; |
| 5678 |
|
| 5679 |
err = ext4_emergency_state(inode->i_sb); |
| 5680 |
if (unlikely(err)) |
| 5681 |
return err; |
| 5682 |
|
| 5683 |
if (EXT4_SB(inode->i_sb)->s_journal) { |
| 5684 |
if (ext4_journal_current_handle()) { |
| 5685 |
ext4_debug("called recursively, non-PF_MEMALLOC!\n"); |
| 5686 |
dump_stack(); |
| 5687 |
return -EIO; |
| 5688 |
} |
| 5689 |
|
| 5690 |
/* |
| 5691 |
* No need to force transaction in WB_SYNC_NONE mode. Also |
| 5692 |
* ext4_sync_fs() will force the commit after everything is |
| 5693 |
* written. |
| 5694 |
*/ |
| 5695 |
if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync) |
| 5696 |
return 0; |
| 5697 |
|
| 5698 |
err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal, |
| 5699 |
EXT4_I(inode)->i_sync_tid); |
| 5700 |
} else { |
| 5701 |
struct ext4_iloc iloc; |
| 5702 |
|
| 5703 |
err = __ext4_get_inode_loc_noinmem(inode, &iloc); |
| 5704 |
if (err) |
| 5705 |
return err; |
| 5706 |
/* |
| 5707 |
* sync(2) will flush the whole buffer cache. No need to do |
| 5708 |
* it here separately for each inode. |
| 5709 |
*/ |
| 5710 |
if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) |
| 5711 |
sync_dirty_buffer(iloc.bh); |
| 5712 |
if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { |
| 5713 |
ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO, |
| 5714 |
"IO error syncing inode"); |
| 5715 |
err = -EIO; |
| 5716 |
} |
| 5717 |
brelse(iloc.bh); |
| 5718 |
} |
| 5719 |
return err; |
| 5720 |
} |
| 5721 |
|
| 5722 |
/* |
| 5723 |
* In data=journal mode ext4_journalled_invalidate_folio() may fail to invalidate |
| 5724 |
* buffers that are attached to a folio straddling i_size and are undergoing |
| 5725 |
* commit. In that case we have to wait for commit to finish and try again. |
| 5726 |
*/ |
| 5727 |
static void ext4_wait_for_tail_page_commit(struct inode *inode) |
| 5728 |
{ |
| 5729 |
unsigned offset; |
| 5730 |
journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
| 5731 |
tid_t commit_tid; |
| 5732 |
int ret; |
| 5733 |
bool has_transaction; |
| 5734 |
|
| 5735 |
offset = inode->i_size & (PAGE_SIZE - 1); |
| 5736 |
/* |
| 5737 |
* If the folio is fully truncated, we don't need to wait for any commit |
| 5738 |
* (and we even should not as __ext4_journalled_invalidate_folio() may |
| 5739 |
* strip all buffers from the folio but keep the folio dirty which can then |
| 5740 |
* confuse e.g. concurrent ext4_writepages() seeing dirty folio without |
| 5741 |
* buffers). Also we don't need to wait for any commit if all buffers in |
| 5742 |
* the folio remain valid. This is most beneficial for the common case of |
| 5743 |
* blocksize == PAGESIZE. |
| 5744 |
*/ |
| 5745 |
if (!offset || offset > (PAGE_SIZE - i_blocksize(inode))) |
| 5746 |
return; |
| 5747 |
while (1) { |
| 5748 |
struct folio *folio = filemap_lock_folio(inode->i_mapping, |
| 5749 |
inode->i_size >> PAGE_SHIFT); |
| 5750 |
if (IS_ERR(folio)) |
| 5751 |
return; |
| 5752 |
ret = __ext4_journalled_invalidate_folio(folio, offset, |
| 5753 |
folio_size(folio) - offset); |
| 5754 |
folio_unlock(folio); |
| 5755 |
folio_put(folio); |
| 5756 |
if (ret != -EBUSY) |
| 5757 |
return; |
| 5758 |
has_transaction = false; |
| 5759 |
read_lock(&journal->j_state_lock); |
| 5760 |
if (journal->j_committing_transaction) { |
| 5761 |
commit_tid = journal->j_committing_transaction->t_tid; |
| 5762 |
has_transaction = true; |
| 5763 |
} |
| 5764 |
read_unlock(&journal->j_state_lock); |
| 5765 |
if (has_transaction) |
| 5766 |
jbd2_log_wait_commit(journal, commit_tid); |
| 5767 |
} |
| 5768 |
} |
| 5769 |
|
| 5770 |
/* |
| 5771 |
* ext4_setattr() |
| 5772 |
* |
| 5773 |
* Called from notify_change. |
| 5774 |
* |
| 5775 |
* We want to trap VFS attempts to truncate the file as soon as |
| 5776 |
* possible. In particular, we want to make sure that when the VFS |
| 5777 |
* shrinks i_size, we put the inode on the orphan list and modify |
| 5778 |
* i_disksize immediately, so that during the subsequent flushing of |
| 5779 |
* dirty pages and freeing of disk blocks, we can guarantee that any |
| 5780 |
* commit will leave the blocks being flushed in an unused state on |
| 5781 |
* disk. (On recovery, the inode will get truncated and the blocks will |
| 5782 |
* be freed, so we have a strong guarantee that no future commit will |
| 5783 |
* leave these blocks visible to the user.) |
| 5784 |
* |
| 5785 |
* Another thing we have to assure is that if we are in ordered mode |
| 5786 |
* and inode is still attached to the committing transaction, we must |
| 5787 |
* we start writeout of all the dirty pages which are being truncated. |
| 5788 |
* This way we are sure that all the data written in the previous |
| 5789 |
* transaction are already on disk (truncate waits for pages under |
| 5790 |
* writeback). |
| 5791 |
* |
| 5792 |
* Called with inode->i_rwsem down. |
| 5793 |
*/ |
| 5794 |
int ext4_setattr(struct mnt_idmap *idmap, struct dentry *dentry, |
| 5795 |
struct iattr *attr) |
| 5796 |
{ |
| 5797 |
struct inode *inode = d_inode(dentry); |
| 5798 |
int error, rc = 0; |
| 5799 |
int orphan = 0; |
| 5800 |
const unsigned int ia_valid = attr->ia_valid; |
| 5801 |
bool inc_ivers = true; |
| 5802 |
|
| 5803 |
error = ext4_emergency_state(inode->i_sb); |
| 5804 |
if (unlikely(error)) |
| 5805 |
return error; |
| 5806 |
|
| 5807 |
if (unlikely(IS_IMMUTABLE(inode))) |
| 5808 |
return -EPERM; |
| 5809 |
|
| 5810 |
if (unlikely(IS_APPEND(inode) && |
| 5811 |
(ia_valid & (ATTR_MODE | ATTR_UID | |
| 5812 |
ATTR_GID | ATTR_TIMES_SET)))) |
| 5813 |
return -EPERM; |
| 5814 |
|
| 5815 |
error = setattr_prepare(idmap, dentry, attr); |
| 5816 |
if (error) |
| 5817 |
return error; |
| 5818 |
|
| 5819 |
error = fscrypt_prepare_setattr(dentry, attr); |
| 5820 |
if (error) |
| 5821 |
return error; |
| 5822 |
|
| 5823 |
if (is_quota_modification(idmap, inode, attr)) { |
| 5824 |
error = dquot_initialize(inode); |
| 5825 |
if (error) |
| 5826 |
return error; |
| 5827 |
} |
| 5828 |
|
| 5829 |
if (i_uid_needs_update(idmap, attr, inode) || |
| 5830 |
i_gid_needs_update(idmap, attr, inode)) { |
| 5831 |
handle_t *handle; |
| 5832 |
|
| 5833 |
/* (user+group)*(old+new) structure, inode write (sb, |
| 5834 |
* inode block, ? - but truncate inode update has it) */ |
| 5835 |
handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
| 5836 |
(EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + |
| 5837 |
EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); |
| 5838 |
if (IS_ERR(handle)) { |
| 5839 |
error = PTR_ERR(handle); |
| 5840 |
goto err_out; |
| 5841 |
} |
| 5842 |
|
| 5843 |
/* dquot_transfer() calls back ext4_get_inode_usage() which |
| 5844 |
* counts xattr inode references. |
| 5845 |
*/ |
| 5846 |
down_read(&EXT4_I(inode)->xattr_sem); |
| 5847 |
error = dquot_transfer(idmap, inode, attr); |
| 5848 |
up_read(&EXT4_I(inode)->xattr_sem); |
| 5849 |
|
| 5850 |
if (error) { |
| 5851 |
ext4_journal_stop(handle); |
| 5852 |
return error; |
| 5853 |
} |
| 5854 |
/* Update corresponding info in inode so that everything is in |
| 5855 |
* one transaction */ |
| 5856 |
i_uid_update(idmap, attr, inode); |
| 5857 |
i_gid_update(idmap, attr, inode); |
| 5858 |
error = ext4_mark_inode_dirty(handle, inode); |
| 5859 |
ext4_journal_stop(handle); |
| 5860 |
if (unlikely(error)) { |
| 5861 |
return error; |
| 5862 |
} |
| 5863 |
} |
| 5864 |
|
| 5865 |
if (attr->ia_valid & ATTR_SIZE) { |
| 5866 |
handle_t *handle; |
| 5867 |
loff_t oldsize = inode->i_size; |
| 5868 |
loff_t old_disksize; |
| 5869 |
int shrink = (attr->ia_size < inode->i_size); |
| 5870 |
|
| 5871 |
if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
| 5872 |
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 5873 |
|
| 5874 |
if (attr->ia_size > sbi->s_bitmap_maxbytes) { |
| 5875 |
return -EFBIG; |
| 5876 |
} |
| 5877 |
} |
| 5878 |
if (!S_ISREG(inode->i_mode)) { |
| 5879 |
return -EINVAL; |
| 5880 |
} |
| 5881 |
|
| 5882 |
if (attr->ia_size == inode->i_size) |
| 5883 |
inc_ivers = false; |
| 5884 |
|
| 5885 |
/* |
| 5886 |
* If file has inline data but new size exceeds inline capacity, |
| 5887 |
* convert to extent-based storage first to prevent inconsistent |
| 5888 |
* state (inline flag set but size exceeds inline capacity). |
| 5889 |
*/ |
| 5890 |
if (ext4_has_inline_data(inode) && |
| 5891 |
attr->ia_size > EXT4_I(inode)->i_inline_size) { |
| 5892 |
error = ext4_convert_inline_data(inode); |
| 5893 |
if (error) |
| 5894 |
goto err_out; |
| 5895 |
} |
| 5896 |
|
| 5897 |
if (shrink) { |
| 5898 |
if (ext4_should_order_data(inode)) { |
| 5899 |
error = ext4_begin_ordered_truncate(inode, |
| 5900 |
attr->ia_size); |
| 5901 |
if (error) |
| 5902 |
goto err_out; |
| 5903 |
} |
| 5904 |
/* |
| 5905 |
* Blocks are going to be removed from the inode. Wait |
| 5906 |
* for dio in flight. |
| 5907 |
*/ |
| 5908 |
inode_dio_wait(inode); |
| 5909 |
} |
| 5910 |
|
| 5911 |
filemap_invalidate_lock(inode->i_mapping); |
| 5912 |
|
| 5913 |
rc = ext4_break_layouts(inode); |
| 5914 |
if (rc) { |
| 5915 |
filemap_invalidate_unlock(inode->i_mapping); |
| 5916 |
goto err_out; |
| 5917 |
} |
| 5918 |
|
| 5919 |
if (attr->ia_size != inode->i_size) { |
| 5920 |
/* attach jbd2 jinode for EOF folio tail zeroing */ |
| 5921 |
if (attr->ia_size & (inode->i_sb->s_blocksize - 1) || |
| 5922 |
oldsize & (inode->i_sb->s_blocksize - 1)) { |
| 5923 |
error = ext4_inode_attach_jinode(inode); |
| 5924 |
if (error) |
| 5925 |
goto out_mmap_sem; |
| 5926 |
} |
| 5927 |
|
| 5928 |
handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); |
| 5929 |
if (IS_ERR(handle)) { |
| 5930 |
error = PTR_ERR(handle); |
| 5931 |
goto out_mmap_sem; |
| 5932 |
} |
| 5933 |
if (ext4_handle_valid(handle) && shrink) { |
| 5934 |
error = ext4_orphan_add(handle, inode); |
| 5935 |
orphan = 1; |
| 5936 |
} |
| 5937 |
/* |
| 5938 |
* Update c/mtime and tail zero the EOF folio on |
| 5939 |
* truncate up. ext4_truncate() handles the shrink case |
| 5940 |
* below. |
| 5941 |
*/ |
| 5942 |
if (!shrink) { |
| 5943 |
inode_set_mtime_to_ts(inode, |
| 5944 |
inode_set_ctime_current(inode)); |
| 5945 |
if (oldsize & (inode->i_sb->s_blocksize - 1)) |
| 5946 |
ext4_block_truncate_page(handle, |
| 5947 |
inode->i_mapping, oldsize); |
| 5948 |
} |
| 5949 |
|
| 5950 |
if (shrink) |
| 5951 |
ext4_fc_track_range(handle, inode, |
| 5952 |
(attr->ia_size > 0 ? attr->ia_size - 1 : 0) >> |
| 5953 |
inode->i_sb->s_blocksize_bits, |
| 5954 |
EXT_MAX_BLOCKS - 1); |
| 5955 |
else |
| 5956 |
ext4_fc_track_range( |
| 5957 |
handle, inode, |
| 5958 |
(oldsize > 0 ? oldsize - 1 : oldsize) >> |
| 5959 |
inode->i_sb->s_blocksize_bits, |
| 5960 |
(attr->ia_size > 0 ? attr->ia_size - 1 : 0) >> |
| 5961 |
inode->i_sb->s_blocksize_bits); |
| 5962 |
|
| 5963 |
down_write(&EXT4_I(inode)->i_data_sem); |
| 5964 |
old_disksize = EXT4_I(inode)->i_disksize; |
| 5965 |
EXT4_I(inode)->i_disksize = attr->ia_size; |
| 5966 |
|
| 5967 |
/* |
| 5968 |
* We have to update i_size under i_data_sem together |
| 5969 |
* with i_disksize to avoid races with writeback code |
| 5970 |
* running ext4_wb_update_i_disksize(). |
| 5971 |
*/ |
| 5972 |
if (!error) |
| 5973 |
i_size_write(inode, attr->ia_size); |
| 5974 |
else |
| 5975 |
EXT4_I(inode)->i_disksize = old_disksize; |
| 5976 |
up_write(&EXT4_I(inode)->i_data_sem); |
| 5977 |
rc = ext4_mark_inode_dirty(handle, inode); |
| 5978 |
if (!error) |
| 5979 |
error = rc; |
| 5980 |
ext4_journal_stop(handle); |
| 5981 |
if (error) |
| 5982 |
goto out_mmap_sem; |
| 5983 |
if (!shrink) { |
| 5984 |
pagecache_isize_extended(inode, oldsize, |
| 5985 |
inode->i_size); |
| 5986 |
} else if (ext4_should_journal_data(inode)) { |
| 5987 |
ext4_wait_for_tail_page_commit(inode); |
| 5988 |
} |
| 5989 |
} |
| 5990 |
|
| 5991 |
/* |
| 5992 |
* Truncate pagecache after we've waited for commit |
| 5993 |
* in data=journal mode to make pages freeable. |
| 5994 |
*/ |
| 5995 |
truncate_pagecache(inode, inode->i_size); |
| 5996 |
/* |
| 5997 |
* Call ext4_truncate() even if i_size didn't change to |
| 5998 |
* truncate possible preallocated blocks. |
| 5999 |
*/ |
| 6000 |
if (attr->ia_size <= oldsize) { |
| 6001 |
rc = ext4_truncate(inode); |
| 6002 |
if (rc) |
| 6003 |
error = rc; |
| 6004 |
} |
| 6005 |
out_mmap_sem: |
| 6006 |
filemap_invalidate_unlock(inode->i_mapping); |
| 6007 |
} |
| 6008 |
|
| 6009 |
if (!error) { |
| 6010 |
if (inc_ivers) |
| 6011 |
inode_inc_iversion(inode); |
| 6012 |
setattr_copy(idmap, inode, attr); |
| 6013 |
mark_inode_dirty(inode); |
| 6014 |
} |
| 6015 |
|
| 6016 |
/* |
| 6017 |
* If the call to ext4_truncate failed to get a transaction handle at |
| 6018 |
* all, we need to clean up the in-core orphan list manually. |
| 6019 |
*/ |
| 6020 |
if (orphan && inode->i_nlink) |
| 6021 |
ext4_orphan_del(NULL, inode); |
| 6022 |
|
| 6023 |
if (!error && (ia_valid & ATTR_MODE)) |
| 6024 |
rc = posix_acl_chmod(idmap, dentry, inode->i_mode); |
| 6025 |
|
| 6026 |
err_out: |
| 6027 |
if (error) |
| 6028 |
ext4_std_error(inode->i_sb, error); |
| 6029 |
if (!error) |
| 6030 |
error = rc; |
| 6031 |
return error; |
| 6032 |
} |
| 6033 |
|
| 6034 |
u32 ext4_dio_alignment(struct inode *inode) |
| 6035 |
{ |
| 6036 |
if (fsverity_active(inode)) |
| 6037 |
return 0; |
| 6038 |
if (ext4_should_journal_data(inode)) |
| 6039 |
return 0; |
| 6040 |
if (ext4_has_inline_data(inode)) |
| 6041 |
return 0; |
| 6042 |
if (IS_ENCRYPTED(inode)) { |
| 6043 |
if (!fscrypt_dio_supported(inode)) |
| 6044 |
return 0; |
| 6045 |
return i_blocksize(inode); |
| 6046 |
} |
| 6047 |
return 1; /* use the iomap defaults */ |
| 6048 |
} |
| 6049 |
|
| 6050 |
int ext4_getattr(struct mnt_idmap *idmap, const struct path *path, |
| 6051 |
struct kstat *stat, u32 request_mask, unsigned int query_flags) |
| 6052 |
{ |
| 6053 |
struct inode *inode = d_inode(path->dentry); |
| 6054 |
struct ext4_inode *raw_inode; |
| 6055 |
struct ext4_inode_info *ei = EXT4_I(inode); |
| 6056 |
unsigned int flags; |
| 6057 |
|
| 6058 |
if ((request_mask & STATX_BTIME) && |
| 6059 |
EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) { |
| 6060 |
stat->result_mask |= STATX_BTIME; |
| 6061 |
stat->btime.tv_sec = ei->i_crtime.tv_sec; |
| 6062 |
stat->btime.tv_nsec = ei->i_crtime.tv_nsec; |
| 6063 |
} |
| 6064 |
|
| 6065 |
/* |
| 6066 |
* Return the DIO alignment restrictions if requested. We only return |
| 6067 |
* this information when requested, since on encrypted files it might |
| 6068 |
* take a fair bit of work to get if the file wasn't opened recently. |
| 6069 |
*/ |
| 6070 |
if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->i_mode)) { |
| 6071 |
u32 dio_align = ext4_dio_alignment(inode); |
| 6072 |
|
| 6073 |
stat->result_mask |= STATX_DIOALIGN; |
| 6074 |
if (dio_align == 1) { |
| 6075 |
struct block_device *bdev = inode->i_sb->s_bdev; |
| 6076 |
|
| 6077 |
/* iomap defaults */ |
| 6078 |
stat->dio_mem_align = bdev_dma_alignment(bdev) + 1; |
| 6079 |
stat->dio_offset_align = bdev_logical_block_size(bdev); |
| 6080 |
} else { |
| 6081 |
stat->dio_mem_align = dio_align; |
| 6082 |
stat->dio_offset_align = dio_align; |
| 6083 |
} |
| 6084 |
} |
| 6085 |
|
| 6086 |
if ((request_mask & STATX_WRITE_ATOMIC) && S_ISREG(inode->i_mode)) { |
| 6087 |
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 6088 |
unsigned int awu_min = 0, awu_max = 0; |
| 6089 |
|
| 6090 |
if (ext4_inode_can_atomic_write(inode)) { |
| 6091 |
awu_min = sbi->s_awu_min; |
| 6092 |
awu_max = sbi->s_awu_max; |
| 6093 |
} |
| 6094 |
|
| 6095 |
generic_fill_statx_atomic_writes(stat, awu_min, awu_max, 0); |
| 6096 |
} |
| 6097 |
|
| 6098 |
flags = ei->i_flags & EXT4_FL_USER_VISIBLE; |
| 6099 |
if (flags & EXT4_APPEND_FL) |
| 6100 |
stat->attributes |= STATX_ATTR_APPEND; |
| 6101 |
if (flags & EXT4_COMPR_FL) |
| 6102 |
stat->attributes |= STATX_ATTR_COMPRESSED; |
| 6103 |
if (flags & EXT4_ENCRYPT_FL) |
| 6104 |
stat->attributes |= STATX_ATTR_ENCRYPTED; |
| 6105 |
if (flags & EXT4_IMMUTABLE_FL) |
| 6106 |
stat->attributes |= STATX_ATTR_IMMUTABLE; |
| 6107 |
if (flags & EXT4_NODUMP_FL) |
| 6108 |
stat->attributes |= STATX_ATTR_NODUMP; |
| 6109 |
if (flags & EXT4_VERITY_FL) |
| 6110 |
stat->attributes |= STATX_ATTR_VERITY; |
| 6111 |
|
| 6112 |
stat->attributes_mask |= (STATX_ATTR_APPEND | |
| 6113 |
STATX_ATTR_COMPRESSED | |
| 6114 |
STATX_ATTR_ENCRYPTED | |
| 6115 |
STATX_ATTR_IMMUTABLE | |
| 6116 |
STATX_ATTR_NODUMP | |
| 6117 |
STATX_ATTR_VERITY); |
| 6118 |
|
| 6119 |
generic_fillattr(idmap, request_mask, inode, stat); |
| 6120 |
return 0; |
| 6121 |
} |
| 6122 |
|
| 6123 |
int ext4_file_getattr(struct mnt_idmap *idmap, |
| 6124 |
const struct path *path, struct kstat *stat, |
| 6125 |
u32 request_mask, unsigned int query_flags) |
| 6126 |
{ |
| 6127 |
struct inode *inode = d_inode(path->dentry); |
| 6128 |
u64 delalloc_blocks; |
| 6129 |
|
| 6130 |
ext4_getattr(idmap, path, stat, request_mask, query_flags); |
| 6131 |
|
| 6132 |
/* |
| 6133 |
* If there is inline data in the inode, the inode will normally not |
| 6134 |
* have data blocks allocated (it may have an external xattr block). |
| 6135 |
* Report at least one sector for such files, so tools like tar, rsync, |
| 6136 |
* others don't incorrectly think the file is completely sparse. |
| 6137 |
*/ |
| 6138 |
if (unlikely(ext4_has_inline_data(inode))) |
| 6139 |
stat->blocks += (stat->size + 511) >> 9; |
| 6140 |
|
| 6141 |
/* |
| 6142 |
* We can't update i_blocks if the block allocation is delayed |
| 6143 |
* otherwise in the case of system crash before the real block |
| 6144 |
* allocation is done, we will have i_blocks inconsistent with |
| 6145 |
* on-disk file blocks. |
| 6146 |
* We always keep i_blocks updated together with real |
| 6147 |
* allocation. But to not confuse with user, stat |
| 6148 |
* will return the blocks that include the delayed allocation |
| 6149 |
* blocks for this file. |
| 6150 |
*/ |
| 6151 |
delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
| 6152 |
EXT4_I(inode)->i_reserved_data_blocks); |
| 6153 |
stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9); |
| 6154 |
return 0; |
| 6155 |
} |
| 6156 |
|
| 6157 |
static int ext4_index_trans_blocks(struct inode *inode, int lblocks, |
| 6158 |
int pextents) |
| 6159 |
{ |
| 6160 |
if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
| 6161 |
return ext4_ind_trans_blocks(inode, lblocks); |
| 6162 |
return ext4_ext_index_trans_blocks(inode, pextents); |
| 6163 |
} |
| 6164 |
|
| 6165 |
/* |
| 6166 |
* Account for index blocks, block groups bitmaps and block group |
| 6167 |
* descriptor blocks if modify datablocks and index blocks |
| 6168 |
* worse case, the indexs blocks spread over different block groups |
| 6169 |
* |
| 6170 |
* If datablocks are discontiguous, they are possible to spread over |
| 6171 |
* different block groups too. If they are contiguous, with flexbg, |
| 6172 |
* they could still across block group boundary. |
| 6173 |
* |
| 6174 |
* Also account for superblock, inode, quota and xattr blocks |
| 6175 |
*/ |
| 6176 |
int ext4_meta_trans_blocks(struct inode *inode, int lblocks, int pextents) |
| 6177 |
{ |
| 6178 |
ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
| 6179 |
int gdpblocks; |
| 6180 |
int idxblocks; |
| 6181 |
int ret; |
| 6182 |
|
| 6183 |
/* |
| 6184 |
* How many index and leaf blocks need to touch to map @lblocks |
| 6185 |
* logical blocks to @pextents physical extents? |
| 6186 |
*/ |
| 6187 |
idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents); |
| 6188 |
|
| 6189 |
/* |
| 6190 |
* Now let's see how many group bitmaps and group descriptors need |
| 6191 |
* to account |
| 6192 |
*/ |
| 6193 |
groups = idxblocks + pextents; |
| 6194 |
gdpblocks = groups; |
| 6195 |
if (groups > ngroups) |
| 6196 |
groups = ngroups; |
| 6197 |
if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
| 6198 |
gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; |
| 6199 |
|
| 6200 |
/* bitmaps and block group descriptor blocks */ |
| 6201 |
ret = idxblocks + groups + gdpblocks; |
| 6202 |
|
| 6203 |
/* Blocks for super block, inode, quota and xattr blocks */ |
| 6204 |
ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); |
| 6205 |
|
| 6206 |
return ret; |
| 6207 |
} |
| 6208 |
|
| 6209 |
/* |
| 6210 |
* Calculate the journal credits for modifying the number of blocks |
| 6211 |
* in a single extent within one transaction. 'nrblocks' is used only |
| 6212 |
* for non-extent inodes. For extent type inodes, 'nrblocks' can be |
| 6213 |
* zero if the exact number of blocks is unknown. |
| 6214 |
*/ |
| 6215 |
int ext4_chunk_trans_extent(struct inode *inode, int nrblocks) |
| 6216 |
{ |
| 6217 |
int ret; |
| 6218 |
|
| 6219 |
ret = ext4_meta_trans_blocks(inode, nrblocks, 1); |
| 6220 |
/* Account for data blocks for journalled mode */ |
| 6221 |
if (ext4_should_journal_data(inode)) |
| 6222 |
ret += nrblocks; |
| 6223 |
return ret; |
| 6224 |
} |
| 6225 |
|
| 6226 |
/* |
| 6227 |
* Calculate the journal credits for a chunk of data modification. |
| 6228 |
* |
| 6229 |
* This is called from DIO, fallocate or whoever calling |
| 6230 |
* ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
| 6231 |
* |
| 6232 |
* journal buffers for data blocks are not included here, as DIO |
| 6233 |
* and fallocate do no need to journal data buffers. |
| 6234 |
*/ |
| 6235 |
int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) |
| 6236 |
{ |
| 6237 |
return ext4_meta_trans_blocks(inode, nrblocks, 1); |
| 6238 |
} |
| 6239 |
|
| 6240 |
/* |
| 6241 |
* The caller must have previously called ext4_reserve_inode_write(). |
| 6242 |
* Give this, we know that the caller already has write access to iloc->bh. |
| 6243 |
*/ |
| 6244 |
int ext4_mark_iloc_dirty(handle_t *handle, |
| 6245 |
struct inode *inode, struct ext4_iloc *iloc) |
| 6246 |
{ |
| 6247 |
int err = 0; |
| 6248 |
|
| 6249 |
err = ext4_emergency_state(inode->i_sb); |
| 6250 |
if (unlikely(err)) { |
| 6251 |
put_bh(iloc->bh); |
| 6252 |
return err; |
| 6253 |
} |
| 6254 |
ext4_fc_track_inode(handle, inode); |
| 6255 |
|
| 6256 |
/* the do_update_inode consumes one bh->b_count */ |
| 6257 |
get_bh(iloc->bh); |
| 6258 |
|
| 6259 |
/* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
| 6260 |
err = ext4_do_update_inode(handle, inode, iloc); |
| 6261 |
put_bh(iloc->bh); |
| 6262 |
return err; |
| 6263 |
} |
| 6264 |
|
| 6265 |
/* |
| 6266 |
* On success, We end up with an outstanding reference count against |
| 6267 |
* iloc->bh. This _must_ be cleaned up later. |
| 6268 |
*/ |
| 6269 |
|
| 6270 |
int |
| 6271 |
ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
| 6272 |
struct ext4_iloc *iloc) |
| 6273 |
{ |
| 6274 |
int err; |
| 6275 |
|
| 6276 |
err = ext4_emergency_state(inode->i_sb); |
| 6277 |
if (unlikely(err)) |
| 6278 |
return err; |
| 6279 |
|
| 6280 |
err = ext4_get_inode_loc(inode, iloc); |
| 6281 |
if (!err) { |
| 6282 |
BUFFER_TRACE(iloc->bh, "get_write_access"); |
| 6283 |
err = ext4_journal_get_write_access(handle, inode->i_sb, |
| 6284 |
iloc->bh, EXT4_JTR_NONE); |
| 6285 |
if (err) { |
| 6286 |
brelse(iloc->bh); |
| 6287 |
iloc->bh = NULL; |
| 6288 |
} |
| 6289 |
ext4_fc_track_inode(handle, inode); |
| 6290 |
} |
| 6291 |
ext4_std_error(inode->i_sb, err); |
| 6292 |
return err; |
| 6293 |
} |
| 6294 |
|
| 6295 |
static int __ext4_expand_extra_isize(struct inode *inode, |
| 6296 |
unsigned int new_extra_isize, |
| 6297 |
struct ext4_iloc *iloc, |
| 6298 |
handle_t *handle, int *no_expand) |
| 6299 |
{ |
| 6300 |
struct ext4_inode *raw_inode; |
| 6301 |
struct ext4_xattr_ibody_header *header; |
| 6302 |
unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb); |
| 6303 |
struct ext4_inode_info *ei = EXT4_I(inode); |
| 6304 |
int error; |
| 6305 |
|
| 6306 |
/* this was checked at iget time, but double check for good measure */ |
| 6307 |
if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) || |
| 6308 |
(ei->i_extra_isize & 3)) { |
| 6309 |
EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)", |
| 6310 |
ei->i_extra_isize, |
| 6311 |
EXT4_INODE_SIZE(inode->i_sb)); |
| 6312 |
return -EFSCORRUPTED; |
| 6313 |
} |
| 6314 |
if ((new_extra_isize < ei->i_extra_isize) || |
| 6315 |
(new_extra_isize < 4) || |
| 6316 |
(new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE)) |
| 6317 |
return -EINVAL; /* Should never happen */ |
| 6318 |
|
| 6319 |
raw_inode = ext4_raw_inode(iloc); |
| 6320 |
|
| 6321 |
header = IHDR(inode, raw_inode); |
| 6322 |
|
| 6323 |
/* No extended attributes present */ |
| 6324 |
if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
| 6325 |
header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { |
| 6326 |
memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE + |
| 6327 |
EXT4_I(inode)->i_extra_isize, 0, |
| 6328 |
new_extra_isize - EXT4_I(inode)->i_extra_isize); |
| 6329 |
EXT4_I(inode)->i_extra_isize = new_extra_isize; |
| 6330 |
return 0; |
| 6331 |
} |
| 6332 |
|
| 6333 |
/* |
| 6334 |
* We may need to allocate external xattr block so we need quotas |
| 6335 |
* initialized. Here we can be called with various locks held so we |
| 6336 |
* cannot affort to initialize quotas ourselves. So just bail. |
| 6337 |
*/ |
| 6338 |
if (dquot_initialize_needed(inode)) |
| 6339 |
return -EAGAIN; |
| 6340 |
|
| 6341 |
/* try to expand with EAs present */ |
| 6342 |
error = ext4_expand_extra_isize_ea(inode, new_extra_isize, |
| 6343 |
raw_inode, handle); |
| 6344 |
if (error) { |
| 6345 |
/* |
| 6346 |
* Inode size expansion failed; don't try again |
| 6347 |
*/ |
| 6348 |
*no_expand = 1; |
| 6349 |
} |
| 6350 |
|
| 6351 |
return error; |
| 6352 |
} |
| 6353 |
|
| 6354 |
/* |
| 6355 |
* Expand an inode by new_extra_isize bytes. |
| 6356 |
* Returns 0 on success or negative error number on failure. |
| 6357 |
*/ |
| 6358 |
static int ext4_try_to_expand_extra_isize(struct inode *inode, |
| 6359 |
unsigned int new_extra_isize, |
| 6360 |
struct ext4_iloc iloc, |
| 6361 |
handle_t *handle) |
| 6362 |
{ |
| 6363 |
int no_expand; |
| 6364 |
int error; |
| 6365 |
|
| 6366 |
if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) |
| 6367 |
return -EOVERFLOW; |
| 6368 |
|
| 6369 |
/* |
| 6370 |
* In nojournal mode, we can immediately attempt to expand |
| 6371 |
* the inode. When journaled, we first need to obtain extra |
| 6372 |
* buffer credits since we may write into the EA block |
| 6373 |
* with this same handle. If journal_extend fails, then it will |
| 6374 |
* only result in a minor loss of functionality for that inode. |
| 6375 |
* If this is felt to be critical, then e2fsck should be run to |
| 6376 |
* force a large enough s_min_extra_isize. |
| 6377 |
*/ |
| 6378 |
if (ext4_journal_extend(handle, |
| 6379 |
EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0) |
| 6380 |
return -ENOSPC; |
| 6381 |
|
| 6382 |
if (ext4_write_trylock_xattr(inode, &no_expand) == 0) |
| 6383 |
return -EBUSY; |
| 6384 |
|
| 6385 |
error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc, |
| 6386 |
handle, &no_expand); |
| 6387 |
ext4_write_unlock_xattr(inode, &no_expand); |
| 6388 |
|
| 6389 |
return error; |
| 6390 |
} |
| 6391 |
|
| 6392 |
int ext4_expand_extra_isize(struct inode *inode, |
| 6393 |
unsigned int new_extra_isize, |
| 6394 |
struct ext4_iloc *iloc) |
| 6395 |
{ |
| 6396 |
handle_t *handle; |
| 6397 |
int no_expand; |
| 6398 |
int error, rc; |
| 6399 |
|
| 6400 |
if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
| 6401 |
brelse(iloc->bh); |
| 6402 |
return -EOVERFLOW; |
| 6403 |
} |
| 6404 |
|
| 6405 |
handle = ext4_journal_start(inode, EXT4_HT_INODE, |
| 6406 |
EXT4_DATA_TRANS_BLOCKS(inode->i_sb)); |
| 6407 |
if (IS_ERR(handle)) { |
| 6408 |
error = PTR_ERR(handle); |
| 6409 |
brelse(iloc->bh); |
| 6410 |
return error; |
| 6411 |
} |
| 6412 |
|
| 6413 |
ext4_write_lock_xattr(inode, &no_expand); |
| 6414 |
|
| 6415 |
BUFFER_TRACE(iloc->bh, "get_write_access"); |
| 6416 |
error = ext4_journal_get_write_access(handle, inode->i_sb, iloc->bh, |
| 6417 |
EXT4_JTR_NONE); |
| 6418 |
if (error) { |
| 6419 |
brelse(iloc->bh); |
| 6420 |
goto out_unlock; |
| 6421 |
} |
| 6422 |
|
| 6423 |
error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc, |
| 6424 |
handle, &no_expand); |
| 6425 |
|
| 6426 |
rc = ext4_mark_iloc_dirty(handle, inode, iloc); |
| 6427 |
if (!error) |
| 6428 |
error = rc; |
| 6429 |
|
| 6430 |
out_unlock: |
| 6431 |
ext4_write_unlock_xattr(inode, &no_expand); |
| 6432 |
ext4_journal_stop(handle); |
| 6433 |
return error; |
| 6434 |
} |
| 6435 |
|
| 6436 |
/* |
| 6437 |
* What we do here is to mark the in-core inode as clean with respect to inode |
| 6438 |
* dirtiness (it may still be data-dirty). |
| 6439 |
* This means that the in-core inode may be reaped by prune_icache |
| 6440 |
* without having to perform any I/O. This is a very good thing, |
| 6441 |
* because *any* task may call prune_icache - even ones which |
| 6442 |
* have a transaction open against a different journal. |
| 6443 |
* |
| 6444 |
* Is this cheating? Not really. Sure, we haven't written the |
| 6445 |
* inode out, but prune_icache isn't a user-visible syncing function. |
| 6446 |
* Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) |
| 6447 |
* we start and wait on commits. |
| 6448 |
*/ |
| 6449 |
int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode, |
| 6450 |
const char *func, unsigned int line) |
| 6451 |
{ |
| 6452 |
struct ext4_iloc iloc; |
| 6453 |
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 6454 |
int err; |
| 6455 |
|
| 6456 |
might_sleep(); |
| 6457 |
trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
| 6458 |
err = ext4_reserve_inode_write(handle, inode, &iloc); |
| 6459 |
if (err) |
| 6460 |
goto out; |
| 6461 |
|
| 6462 |
if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize) |
| 6463 |
ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize, |
| 6464 |
iloc, handle); |
| 6465 |
|
| 6466 |
err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
| 6467 |
out: |
| 6468 |
if (unlikely(err)) |
| 6469 |
ext4_error_inode_err(inode, func, line, 0, err, |
| 6470 |
"mark_inode_dirty error"); |
| 6471 |
return err; |
| 6472 |
} |
| 6473 |
|
| 6474 |
/* |
| 6475 |
* ext4_dirty_inode() is called from __mark_inode_dirty() |
| 6476 |
* |
| 6477 |
* We're really interested in the case where a file is being extended. |
| 6478 |
* i_size has been changed by generic_commit_write() and we thus need |
| 6479 |
* to include the updated inode in the current transaction. |
| 6480 |
* |
| 6481 |
* Also, dquot_alloc_block() will always dirty the inode when blocks |
| 6482 |
* are allocated to the file. |
| 6483 |
* |
| 6484 |
* If the inode is marked synchronous, we don't honour that here - doing |
| 6485 |
* so would cause a commit on atime updates, which we don't bother doing. |
| 6486 |
* We handle synchronous inodes at the highest possible level. |
| 6487 |
*/ |
| 6488 |
void ext4_dirty_inode(struct inode *inode, int flags) |
| 6489 |
{ |
| 6490 |
handle_t *handle; |
| 6491 |
|
| 6492 |
handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
| 6493 |
if (IS_ERR(handle)) |
| 6494 |
return; |
| 6495 |
ext4_mark_inode_dirty(handle, inode); |
| 6496 |
ext4_journal_stop(handle); |
| 6497 |
} |
| 6498 |
|
| 6499 |
int ext4_change_inode_journal_flag(struct inode *inode, int val) |
| 6500 |
{ |
| 6501 |
journal_t *journal; |
| 6502 |
handle_t *handle; |
| 6503 |
int err; |
| 6504 |
int alloc_ctx; |
| 6505 |
|
| 6506 |
/* |
| 6507 |
* We have to be very careful here: changing a data block's |
| 6508 |
* journaling status dynamically is dangerous. If we write a |
| 6509 |
* data block to the journal, change the status and then delete |
| 6510 |
* that block, we risk forgetting to revoke the old log record |
| 6511 |
* from the journal and so a subsequent replay can corrupt data. |
| 6512 |
* So, first we make sure that the journal is empty and that |
| 6513 |
* nobody is changing anything. |
| 6514 |
*/ |
| 6515 |
|
| 6516 |
journal = EXT4_JOURNAL(inode); |
| 6517 |
if (!journal) |
| 6518 |
return 0; |
| 6519 |
if (is_journal_aborted(journal)) |
| 6520 |
return -EROFS; |
| 6521 |
|
| 6522 |
/* Wait for all existing dio workers */ |
| 6523 |
inode_dio_wait(inode); |
| 6524 |
|
| 6525 |
/* |
| 6526 |
* Before flushing the journal and switching inode's aops, we have |
| 6527 |
* to flush all dirty data the inode has. There can be outstanding |
| 6528 |
* delayed allocations, there can be unwritten extents created by |
| 6529 |
* fallocate or buffered writes in dioread_nolock mode covered by |
| 6530 |
* dirty data which can be converted only after flushing the dirty |
| 6531 |
* data (and journalled aops don't know how to handle these cases). |
| 6532 |
*/ |
| 6533 |
filemap_invalidate_lock(inode->i_mapping); |
| 6534 |
err = filemap_write_and_wait(inode->i_mapping); |
| 6535 |
if (err < 0) { |
| 6536 |
filemap_invalidate_unlock(inode->i_mapping); |
| 6537 |
return err; |
| 6538 |
} |
| 6539 |
/* Before switch the inode journalling mode evict all the page cache. */ |
| 6540 |
truncate_pagecache(inode, 0); |
| 6541 |
|
| 6542 |
alloc_ctx = ext4_writepages_down_write(inode->i_sb); |
| 6543 |
jbd2_journal_lock_updates(journal); |
| 6544 |
|
| 6545 |
/* |
| 6546 |
* OK, there are no updates running now, and all cached data is |
| 6547 |
* synced to disk. We are now in a completely consistent state |
| 6548 |
* which doesn't have anything in the journal, and we know that |
| 6549 |
* no filesystem updates are running, so it is safe to modify |
| 6550 |
* the inode's in-core data-journaling state flag now. |
| 6551 |
*/ |
| 6552 |
|
| 6553 |
if (val) |
| 6554 |
ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
| 6555 |
else { |
| 6556 |
err = jbd2_journal_flush(journal, 0); |
| 6557 |
if (err < 0) { |
| 6558 |
jbd2_journal_unlock_updates(journal); |
| 6559 |
ext4_writepages_up_write(inode->i_sb, alloc_ctx); |
| 6560 |
filemap_invalidate_unlock(inode->i_mapping); |
| 6561 |
return err; |
| 6562 |
} |
| 6563 |
ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
| 6564 |
} |
| 6565 |
ext4_set_aops(inode); |
| 6566 |
ext4_set_inode_mapping_order(inode); |
| 6567 |
|
| 6568 |
jbd2_journal_unlock_updates(journal); |
| 6569 |
ext4_writepages_up_write(inode->i_sb, alloc_ctx); |
| 6570 |
filemap_invalidate_unlock(inode->i_mapping); |
| 6571 |
|
| 6572 |
/* Finally we can mark the inode as dirty. */ |
| 6573 |
|
| 6574 |
handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
| 6575 |
if (IS_ERR(handle)) |
| 6576 |
return PTR_ERR(handle); |
| 6577 |
|
| 6578 |
ext4_fc_mark_ineligible(inode->i_sb, |
| 6579 |
EXT4_FC_REASON_JOURNAL_FLAG_CHANGE, handle); |
| 6580 |
err = ext4_mark_inode_dirty(handle, inode); |
| 6581 |
ext4_handle_sync(handle); |
| 6582 |
ext4_journal_stop(handle); |
| 6583 |
ext4_std_error(inode->i_sb, err); |
| 6584 |
|
| 6585 |
return err; |
| 6586 |
} |
| 6587 |
|
| 6588 |
static int ext4_bh_unmapped(handle_t *handle, struct inode *inode, |
| 6589 |
struct buffer_head *bh) |
| 6590 |
{ |
| 6591 |
return !buffer_mapped(bh); |
| 6592 |
} |
| 6593 |
|
| 6594 |
static int ext4_block_page_mkwrite(struct inode *inode, struct folio *folio, |
| 6595 |
get_block_t get_block) |
| 6596 |
{ |
| 6597 |
handle_t *handle; |
| 6598 |
loff_t size; |
| 6599 |
unsigned long len; |
| 6600 |
int credits; |
| 6601 |
int ret; |
| 6602 |
|
| 6603 |
credits = ext4_chunk_trans_extent(inode, |
| 6604 |
ext4_journal_blocks_per_folio(inode)); |
| 6605 |
handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, credits); |
| 6606 |
if (IS_ERR(handle)) |
| 6607 |
return PTR_ERR(handle); |
| 6608 |
|
| 6609 |
folio_lock(folio); |
| 6610 |
size = i_size_read(inode); |
| 6611 |
/* Page got truncated from under us? */ |
| 6612 |
if (folio->mapping != inode->i_mapping || folio_pos(folio) > size) { |
| 6613 |
ret = -EFAULT; |
| 6614 |
goto out_error; |
| 6615 |
} |
| 6616 |
|
| 6617 |
len = folio_size(folio); |
| 6618 |
if (folio_pos(folio) + len > size) |
| 6619 |
len = size - folio_pos(folio); |
| 6620 |
|
| 6621 |
ret = ext4_block_write_begin(handle, folio, 0, len, get_block); |
| 6622 |
if (ret) |
| 6623 |
goto out_error; |
| 6624 |
|
| 6625 |
if (!ext4_should_journal_data(inode)) { |
| 6626 |
block_commit_write(folio, 0, len); |
| 6627 |
folio_mark_dirty(folio); |
| 6628 |
} else { |
| 6629 |
ret = ext4_journal_folio_buffers(handle, folio, len); |
| 6630 |
if (ret) |
| 6631 |
goto out_error; |
| 6632 |
} |
| 6633 |
ext4_journal_stop(handle); |
| 6634 |
folio_wait_stable(folio); |
| 6635 |
return ret; |
| 6636 |
|
| 6637 |
out_error: |
| 6638 |
folio_unlock(folio); |
| 6639 |
ext4_journal_stop(handle); |
| 6640 |
return ret; |
| 6641 |
} |
| 6642 |
|
| 6643 |
vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf) |
| 6644 |
{ |
| 6645 |
struct vm_area_struct *vma = vmf->vma; |
| 6646 |
struct folio *folio = page_folio(vmf->page); |
| 6647 |
loff_t size; |
| 6648 |
unsigned long len; |
| 6649 |
int err; |
| 6650 |
vm_fault_t ret; |
| 6651 |
struct file *file = vma->vm_file; |
| 6652 |
struct inode *inode = file_inode(file); |
| 6653 |
struct address_space *mapping = inode->i_mapping; |
| 6654 |
get_block_t *get_block = ext4_get_block; |
| 6655 |
int retries = 0; |
| 6656 |
|
| 6657 |
if (unlikely(IS_IMMUTABLE(inode))) |
| 6658 |
return VM_FAULT_SIGBUS; |
| 6659 |
|
| 6660 |
sb_start_pagefault(inode->i_sb); |
| 6661 |
file_update_time(vma->vm_file); |
| 6662 |
|
| 6663 |
filemap_invalidate_lock_shared(mapping); |
| 6664 |
|
| 6665 |
err = ext4_convert_inline_data(inode); |
| 6666 |
if (err) |
| 6667 |
goto out_ret; |
| 6668 |
|
| 6669 |
/* |
| 6670 |
* On data journalling we skip straight to the transaction handle: |
| 6671 |
* there's no delalloc; page truncated will be checked later; the |
| 6672 |
* early return w/ all buffers mapped (calculates size/len) can't |
| 6673 |
* be used; and there's no dioread_nolock, so only ext4_get_block. |
| 6674 |
*/ |
| 6675 |
if (ext4_should_journal_data(inode)) |
| 6676 |
goto retry_alloc; |
| 6677 |
|
| 6678 |
/* Delalloc case is easy... */ |
| 6679 |
if (test_opt(inode->i_sb, DELALLOC) && |
| 6680 |
!ext4_nonda_switch(inode->i_sb)) { |
| 6681 |
do { |
| 6682 |
err = block_page_mkwrite(vma, vmf, |
| 6683 |
ext4_da_get_block_prep); |
| 6684 |
} while (err == -ENOSPC && |
| 6685 |
ext4_should_retry_alloc(inode->i_sb, &retries)); |
| 6686 |
goto out_ret; |
| 6687 |
} |
| 6688 |
|
| 6689 |
folio_lock(folio); |
| 6690 |
size = i_size_read(inode); |
| 6691 |
/* Page got truncated from under us? */ |
| 6692 |
if (folio->mapping != mapping || folio_pos(folio) > size) { |
| 6693 |
folio_unlock(folio); |
| 6694 |
ret = VM_FAULT_NOPAGE; |
| 6695 |
goto out; |
| 6696 |
} |
| 6697 |
|
| 6698 |
len = folio_size(folio); |
| 6699 |
if (folio_pos(folio) + len > size) |
| 6700 |
len = size - folio_pos(folio); |
| 6701 |
/* |
| 6702 |
* Return if we have all the buffers mapped. This avoids the need to do |
| 6703 |
* journal_start/journal_stop which can block and take a long time |
| 6704 |
* |
| 6705 |
* This cannot be done for data journalling, as we have to add the |
| 6706 |
* inode to the transaction's list to writeprotect pages on commit. |
| 6707 |
*/ |
| 6708 |
if (folio_buffers(folio)) { |
| 6709 |
if (!ext4_walk_page_buffers(NULL, inode, folio_buffers(folio), |
| 6710 |
0, len, NULL, |
| 6711 |
ext4_bh_unmapped)) { |
| 6712 |
/* Wait so that we don't change page under IO */ |
| 6713 |
folio_wait_stable(folio); |
| 6714 |
ret = VM_FAULT_LOCKED; |
| 6715 |
goto out; |
| 6716 |
} |
| 6717 |
} |
| 6718 |
folio_unlock(folio); |
| 6719 |
/* OK, we need to fill the hole... */ |
| 6720 |
if (ext4_should_dioread_nolock(inode)) |
| 6721 |
get_block = ext4_get_block_unwritten; |
| 6722 |
retry_alloc: |
| 6723 |
/* Start journal and allocate blocks */ |
| 6724 |
err = ext4_block_page_mkwrite(inode, folio, get_block); |
| 6725 |
if (err == -EAGAIN || |
| 6726 |
(err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))) |
| 6727 |
goto retry_alloc; |
| 6728 |
out_ret: |
| 6729 |
ret = vmf_fs_error(err); |
| 6730 |
out: |
| 6731 |
filemap_invalidate_unlock_shared(mapping); |
| 6732 |
sb_end_pagefault(inode->i_sb); |
| 6733 |
return ret; |
| 6734 |
} |