| 1 |
// SPDX-License-Identifier: GPL-2.0-or-later |
| 2 |
/* |
| 3 |
* Copyright (C) International Business Machines Corp., 2000-2004 |
| 4 |
* Portions Copyright (C) Christoph Hellwig, 2001-2002 |
| 5 |
*/ |
| 6 |
|
| 7 |
#include <linux/fs.h> |
| 8 |
#include <linux/mpage.h> |
| 9 |
#include <linux/buffer_head.h> |
| 10 |
#include <linux/pagemap.h> |
| 11 |
#include <linux/quotaops.h> |
| 12 |
#include <linux/uio.h> |
| 13 |
#include <linux/writeback.h> |
| 14 |
#include "jfs_incore.h" |
| 15 |
#include "jfs_inode.h" |
| 16 |
#include "jfs_filsys.h" |
| 17 |
#include "jfs_imap.h" |
| 18 |
#include "jfs_extent.h" |
| 19 |
#include "jfs_unicode.h" |
| 20 |
#include "jfs_debug.h" |
| 21 |
#include "jfs_dmap.h" |
| 22 |
|
| 23 |
|
| 24 |
struct inode *jfs_iget(struct super_block *sb, unsigned long ino) |
| 25 |
{ |
| 26 |
struct inode *inode; |
| 27 |
int ret; |
| 28 |
|
| 29 |
inode = iget_locked(sb, ino); |
| 30 |
if (!inode) |
| 31 |
return ERR_PTR(-ENOMEM); |
| 32 |
if (!(inode_state_read_once(inode) & I_NEW)) |
| 33 |
return inode; |
| 34 |
|
| 35 |
ret = diRead(inode); |
| 36 |
if (ret < 0) { |
| 37 |
iget_failed(inode); |
| 38 |
return ERR_PTR(ret); |
| 39 |
} |
| 40 |
|
| 41 |
if (S_ISREG(inode->i_mode)) { |
| 42 |
inode->i_op = &jfs_file_inode_operations; |
| 43 |
inode->i_fop = &jfs_file_operations; |
| 44 |
inode->i_mapping->a_ops = &jfs_aops; |
| 45 |
} else if (S_ISDIR(inode->i_mode)) { |
| 46 |
inode->i_op = &jfs_dir_inode_operations; |
| 47 |
inode->i_fop = &jfs_dir_operations; |
| 48 |
} else if (S_ISLNK(inode->i_mode)) { |
| 49 |
if (inode->i_size >= IDATASIZE) { |
| 50 |
inode->i_op = &page_symlink_inode_operations; |
| 51 |
inode_nohighmem(inode); |
| 52 |
inode->i_mapping->a_ops = &jfs_aops; |
| 53 |
} else { |
| 54 |
inode->i_op = &jfs_fast_symlink_inode_operations; |
| 55 |
inode->i_link = JFS_IP(inode)->i_inline; |
| 56 |
/* |
| 57 |
* The inline data should be null-terminated, but |
| 58 |
* don't let on-disk corruption crash the kernel |
| 59 |
*/ |
| 60 |
inode->i_link[inode->i_size] = '\0'; |
| 61 |
} |
| 62 |
} else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
| 63 |
S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { |
| 64 |
inode->i_op = &jfs_file_inode_operations; |
| 65 |
init_special_inode(inode, inode->i_mode, inode->i_rdev); |
| 66 |
} else { |
| 67 |
printk(KERN_DEBUG "JFS: Invalid file type 0%04o for inode %lu.\n", |
| 68 |
inode->i_mode, inode->i_ino); |
| 69 |
iget_failed(inode); |
| 70 |
return ERR_PTR(-EIO); |
| 71 |
} |
| 72 |
unlock_new_inode(inode); |
| 73 |
return inode; |
| 74 |
} |
| 75 |
|
| 76 |
/* |
| 77 |
* Workhorse of both fsync & write_inode |
| 78 |
*/ |
| 79 |
int jfs_commit_inode(struct inode *inode, int wait) |
| 80 |
{ |
| 81 |
int rc = 0; |
| 82 |
tid_t tid; |
| 83 |
static int noisy = 5; |
| 84 |
|
| 85 |
jfs_info("In jfs_commit_inode, inode = 0x%p", inode); |
| 86 |
|
| 87 |
/* |
| 88 |
* Don't commit if inode has been committed since last being |
| 89 |
* marked dirty, or if it has been deleted. |
| 90 |
*/ |
| 91 |
if (inode->i_nlink == 0 || !test_cflag(COMMIT_Dirty, inode)) |
| 92 |
return 0; |
| 93 |
|
| 94 |
if (isReadOnly(inode)) { |
| 95 |
/* kernel allows writes to devices on read-only |
| 96 |
* partitions and may think inode is dirty |
| 97 |
*/ |
| 98 |
if (!special_file(inode->i_mode) && noisy) { |
| 99 |
jfs_err("jfs_commit_inode(0x%p) called on read-only volume", |
| 100 |
inode); |
| 101 |
jfs_err("Is remount racy?"); |
| 102 |
noisy--; |
| 103 |
} |
| 104 |
return 0; |
| 105 |
} |
| 106 |
|
| 107 |
tid = txBegin(inode->i_sb, COMMIT_INODE); |
| 108 |
mutex_lock(&JFS_IP(inode)->commit_mutex); |
| 109 |
|
| 110 |
/* |
| 111 |
* Retest inode state after taking commit_mutex |
| 112 |
*/ |
| 113 |
if (inode->i_nlink && test_cflag(COMMIT_Dirty, inode)) |
| 114 |
rc = txCommit(tid, 1, &inode, wait ? COMMIT_SYNC : 0); |
| 115 |
|
| 116 |
txEnd(tid); |
| 117 |
mutex_unlock(&JFS_IP(inode)->commit_mutex); |
| 118 |
return rc; |
| 119 |
} |
| 120 |
|
| 121 |
int jfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
| 122 |
{ |
| 123 |
int wait = wbc->sync_mode == WB_SYNC_ALL; |
| 124 |
|
| 125 |
if (inode->i_nlink == 0) |
| 126 |
return 0; |
| 127 |
/* |
| 128 |
* If COMMIT_DIRTY is not set, the inode isn't really dirty. |
| 129 |
* It has been committed since the last change, but was still |
| 130 |
* on the dirty inode list. |
| 131 |
*/ |
| 132 |
if (!test_cflag(COMMIT_Dirty, inode)) { |
| 133 |
/* Make sure committed changes hit the disk */ |
| 134 |
jfs_flush_journal(JFS_SBI(inode->i_sb)->log, wait); |
| 135 |
return 0; |
| 136 |
} |
| 137 |
|
| 138 |
if (jfs_commit_inode(inode, wait)) { |
| 139 |
jfs_err("jfs_write_inode: jfs_commit_inode failed!"); |
| 140 |
return -EIO; |
| 141 |
} else |
| 142 |
return 0; |
| 143 |
} |
| 144 |
|
| 145 |
void jfs_evict_inode(struct inode *inode) |
| 146 |
{ |
| 147 |
struct jfs_inode_info *ji = JFS_IP(inode); |
| 148 |
|
| 149 |
jfs_info("In jfs_evict_inode, inode = 0x%p", inode); |
| 150 |
|
| 151 |
if (!inode->i_nlink && !is_bad_inode(inode)) { |
| 152 |
dquot_initialize(inode); |
| 153 |
|
| 154 |
truncate_inode_pages_final(&inode->i_data); |
| 155 |
if (JFS_IP(inode)->fileset == FILESYSTEM_I) { |
| 156 |
struct inode *ipimap = JFS_SBI(inode->i_sb)->ipimap; |
| 157 |
|
| 158 |
if (test_cflag(COMMIT_Freewmap, inode)) |
| 159 |
jfs_free_zero_link(inode); |
| 160 |
|
| 161 |
if (ipimap && JFS_IP(ipimap)->i_imap) |
| 162 |
diFree(inode); |
| 163 |
|
| 164 |
/* |
| 165 |
* Free the inode from the quota allocation. |
| 166 |
*/ |
| 167 |
dquot_free_inode(inode); |
| 168 |
} |
| 169 |
} else { |
| 170 |
truncate_inode_pages_final(&inode->i_data); |
| 171 |
} |
| 172 |
clear_inode(inode); |
| 173 |
dquot_drop(inode); |
| 174 |
|
| 175 |
BUG_ON(!list_empty(&ji->anon_inode_list)); |
| 176 |
|
| 177 |
spin_lock_irq(&ji->ag_lock); |
| 178 |
if (ji->active_ag != -1) { |
| 179 |
struct bmap *bmap = JFS_SBI(inode->i_sb)->bmap; |
| 180 |
atomic_dec(&bmap->db_active[ji->active_ag]); |
| 181 |
ji->active_ag = -1; |
| 182 |
} |
| 183 |
spin_unlock_irq(&ji->ag_lock); |
| 184 |
} |
| 185 |
|
| 186 |
void jfs_dirty_inode(struct inode *inode, int flags) |
| 187 |
{ |
| 188 |
static int noisy = 5; |
| 189 |
|
| 190 |
if (isReadOnly(inode)) { |
| 191 |
if (!special_file(inode->i_mode) && noisy) { |
| 192 |
/* kernel allows writes to devices on read-only |
| 193 |
* partitions and may try to mark inode dirty |
| 194 |
*/ |
| 195 |
jfs_err("jfs_dirty_inode called on read-only volume"); |
| 196 |
jfs_err("Is remount racy?"); |
| 197 |
noisy--; |
| 198 |
} |
| 199 |
return; |
| 200 |
} |
| 201 |
|
| 202 |
set_cflag(COMMIT_Dirty, inode); |
| 203 |
} |
| 204 |
|
| 205 |
int jfs_get_block(struct inode *ip, sector_t lblock, |
| 206 |
struct buffer_head *bh_result, int create) |
| 207 |
{ |
| 208 |
s64 lblock64 = lblock; |
| 209 |
int rc = 0; |
| 210 |
xad_t xad; |
| 211 |
s64 xaddr; |
| 212 |
int xflag; |
| 213 |
s32 xlen = bh_result->b_size >> ip->i_blkbits; |
| 214 |
|
| 215 |
/* |
| 216 |
* Take appropriate lock on inode |
| 217 |
*/ |
| 218 |
if (create) |
| 219 |
IWRITE_LOCK(ip, RDWRLOCK_NORMAL); |
| 220 |
else |
| 221 |
IREAD_LOCK(ip, RDWRLOCK_NORMAL); |
| 222 |
|
| 223 |
if (((lblock64 << ip->i_sb->s_blocksize_bits) < ip->i_size) && |
| 224 |
(!xtLookup(ip, lblock64, xlen, &xflag, &xaddr, &xlen, 0)) && |
| 225 |
xaddr) { |
| 226 |
if (xflag & XAD_NOTRECORDED) { |
| 227 |
if (!create) |
| 228 |
/* |
| 229 |
* Allocated but not recorded, read treats |
| 230 |
* this as a hole |
| 231 |
*/ |
| 232 |
goto unlock; |
| 233 |
XADoffset(&xad, lblock64); |
| 234 |
XADlength(&xad, xlen); |
| 235 |
XADaddress(&xad, xaddr); |
| 236 |
rc = extRecord(ip, &xad); |
| 237 |
if (rc) |
| 238 |
goto unlock; |
| 239 |
set_buffer_new(bh_result); |
| 240 |
} |
| 241 |
|
| 242 |
map_bh(bh_result, ip->i_sb, xaddr); |
| 243 |
bh_result->b_size = xlen << ip->i_blkbits; |
| 244 |
goto unlock; |
| 245 |
} |
| 246 |
if (!create) |
| 247 |
goto unlock; |
| 248 |
|
| 249 |
/* |
| 250 |
* Allocate a new block |
| 251 |
*/ |
| 252 |
if ((rc = extHint(ip, lblock64 << ip->i_sb->s_blocksize_bits, &xad))) |
| 253 |
goto unlock; |
| 254 |
rc = extAlloc(ip, xlen, lblock64, &xad, false); |
| 255 |
if (rc) |
| 256 |
goto unlock; |
| 257 |
|
| 258 |
set_buffer_new(bh_result); |
| 259 |
map_bh(bh_result, ip->i_sb, addressXAD(&xad)); |
| 260 |
bh_result->b_size = lengthXAD(&xad) << ip->i_blkbits; |
| 261 |
|
| 262 |
unlock: |
| 263 |
/* |
| 264 |
* Release lock on inode |
| 265 |
*/ |
| 266 |
if (create) |
| 267 |
IWRITE_UNLOCK(ip); |
| 268 |
else |
| 269 |
IREAD_UNLOCK(ip); |
| 270 |
return rc; |
| 271 |
} |
| 272 |
|
| 273 |
static int jfs_writepages(struct address_space *mapping, |
| 274 |
struct writeback_control *wbc) |
| 275 |
{ |
| 276 |
return mpage_writepages(mapping, wbc, jfs_get_block); |
| 277 |
} |
| 278 |
|
| 279 |
static int jfs_read_folio(struct file *file, struct folio *folio) |
| 280 |
{ |
| 281 |
return mpage_read_folio(folio, jfs_get_block); |
| 282 |
} |
| 283 |
|
| 284 |
static void jfs_readahead(struct readahead_control *rac) |
| 285 |
{ |
| 286 |
mpage_readahead(rac, jfs_get_block); |
| 287 |
} |
| 288 |
|
| 289 |
static void jfs_write_failed(struct address_space *mapping, loff_t to) |
| 290 |
{ |
| 291 |
struct inode *inode = mapping->host; |
| 292 |
|
| 293 |
if (to > inode->i_size) { |
| 294 |
truncate_pagecache(inode, inode->i_size); |
| 295 |
jfs_truncate(inode); |
| 296 |
} |
| 297 |
} |
| 298 |
|
| 299 |
static int jfs_write_begin(const struct kiocb *iocb, |
| 300 |
struct address_space *mapping, |
| 301 |
loff_t pos, unsigned len, |
| 302 |
struct folio **foliop, void **fsdata) |
| 303 |
{ |
| 304 |
int ret; |
| 305 |
|
| 306 |
ret = block_write_begin(mapping, pos, len, foliop, jfs_get_block); |
| 307 |
if (unlikely(ret)) |
| 308 |
jfs_write_failed(mapping, pos + len); |
| 309 |
|
| 310 |
return ret; |
| 311 |
} |
| 312 |
|
| 313 |
static int jfs_write_end(const struct kiocb *iocb, |
| 314 |
struct address_space *mapping, |
| 315 |
loff_t pos, unsigned len, unsigned copied, |
| 316 |
struct folio *folio, void *fsdata) |
| 317 |
{ |
| 318 |
int ret; |
| 319 |
|
| 320 |
ret = generic_write_end(iocb, mapping, pos, len, copied, folio, fsdata); |
| 321 |
if (ret < len) |
| 322 |
jfs_write_failed(mapping, pos + len); |
| 323 |
return ret; |
| 324 |
} |
| 325 |
|
| 326 |
static sector_t jfs_bmap(struct address_space *mapping, sector_t block) |
| 327 |
{ |
| 328 |
return generic_block_bmap(mapping, block, jfs_get_block); |
| 329 |
} |
| 330 |
|
| 331 |
static ssize_t jfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
| 332 |
{ |
| 333 |
struct file *file = iocb->ki_filp; |
| 334 |
struct address_space *mapping = file->f_mapping; |
| 335 |
struct inode *inode = file->f_mapping->host; |
| 336 |
size_t count = iov_iter_count(iter); |
| 337 |
ssize_t ret; |
| 338 |
|
| 339 |
ret = blockdev_direct_IO(iocb, inode, iter, jfs_get_block); |
| 340 |
|
| 341 |
/* |
| 342 |
* In case of error extending write may have instantiated a few |
| 343 |
* blocks outside i_size. Trim these off again. |
| 344 |
*/ |
| 345 |
if (unlikely(iov_iter_rw(iter) == WRITE && ret < 0)) { |
| 346 |
loff_t isize = i_size_read(inode); |
| 347 |
loff_t end = iocb->ki_pos + count; |
| 348 |
|
| 349 |
if (end > isize) |
| 350 |
jfs_write_failed(mapping, end); |
| 351 |
} |
| 352 |
|
| 353 |
return ret; |
| 354 |
} |
| 355 |
|
| 356 |
const struct address_space_operations jfs_aops = { |
| 357 |
.dirty_folio = block_dirty_folio, |
| 358 |
.invalidate_folio = block_invalidate_folio, |
| 359 |
.read_folio = jfs_read_folio, |
| 360 |
.readahead = jfs_readahead, |
| 361 |
.writepages = jfs_writepages, |
| 362 |
.write_begin = jfs_write_begin, |
| 363 |
.write_end = jfs_write_end, |
| 364 |
.bmap = jfs_bmap, |
| 365 |
.direct_IO = jfs_direct_IO, |
| 366 |
.migrate_folio = buffer_migrate_folio, |
| 367 |
}; |
| 368 |
|
| 369 |
/* |
| 370 |
* Guts of jfs_truncate. Called with locks already held. Can be called |
| 371 |
* with directory for truncating directory index table. |
| 372 |
*/ |
| 373 |
void jfs_truncate_nolock(struct inode *ip, loff_t length) |
| 374 |
{ |
| 375 |
loff_t newsize; |
| 376 |
tid_t tid; |
| 377 |
|
| 378 |
ASSERT(length >= 0); |
| 379 |
|
| 380 |
if (test_cflag(COMMIT_Nolink, ip) || isReadOnly(ip)) { |
| 381 |
xtTruncate(0, ip, length, COMMIT_WMAP); |
| 382 |
return; |
| 383 |
} |
| 384 |
|
| 385 |
do { |
| 386 |
tid = txBegin(ip->i_sb, 0); |
| 387 |
|
| 388 |
/* |
| 389 |
* The commit_mutex cannot be taken before txBegin. |
| 390 |
* txBegin may block and there is a chance the inode |
| 391 |
* could be marked dirty and need to be committed |
| 392 |
* before txBegin unblocks |
| 393 |
*/ |
| 394 |
mutex_lock(&JFS_IP(ip)->commit_mutex); |
| 395 |
|
| 396 |
newsize = xtTruncate(tid, ip, length, |
| 397 |
COMMIT_TRUNCATE | COMMIT_PWMAP); |
| 398 |
if (newsize < 0) { |
| 399 |
txEnd(tid); |
| 400 |
mutex_unlock(&JFS_IP(ip)->commit_mutex); |
| 401 |
break; |
| 402 |
} |
| 403 |
|
| 404 |
inode_set_mtime_to_ts(ip, inode_set_ctime_current(ip)); |
| 405 |
mark_inode_dirty(ip); |
| 406 |
|
| 407 |
txCommit(tid, 1, &ip, 0); |
| 408 |
txEnd(tid); |
| 409 |
mutex_unlock(&JFS_IP(ip)->commit_mutex); |
| 410 |
} while (newsize > length); /* Truncate isn't always atomic */ |
| 411 |
} |
| 412 |
|
| 413 |
void jfs_truncate(struct inode *ip) |
| 414 |
{ |
| 415 |
jfs_info("jfs_truncate: size = 0x%lx", (ulong) ip->i_size); |
| 416 |
|
| 417 |
block_truncate_page(ip->i_mapping, ip->i_size, jfs_get_block); |
| 418 |
|
| 419 |
IWRITE_LOCK(ip, RDWRLOCK_NORMAL); |
| 420 |
jfs_truncate_nolock(ip, ip->i_size); |
| 421 |
IWRITE_UNLOCK(ip); |
| 422 |
} |