1 /* 2 * linux/fs/ext4/page-io.c 3 * 4 * This contains the new page_io functions for ext4 5 * 6 * Written by Theodore Ts'o, 2010. 7 */ 8 9 #include <linux/fs.h> 10 #include <linux/time.h> 11 #include <linux/jbd2.h> 12 #include <linux/highuid.h> 13 #include <linux/pagemap.h> 14 #include <linux/quotaops.h> 15 #include <linux/string.h> 16 #include <linux/buffer_head.h> 17 #include <linux/writeback.h> 18 #include <linux/pagevec.h> 19 #include <linux/mpage.h> 20 #include <linux/namei.h> 21 #include <linux/uio.h> 22 #include <linux/bio.h> 23 #include <linux/workqueue.h> 24 #include <linux/kernel.h> 25 #include <linux/slab.h> 26 27 #include "ext4_jbd2.h" 28 #include "xattr.h" 29 #include "acl.h" 30 #include "ext4_extents.h" 31 32 static struct kmem_cache *io_page_cachep, *io_end_cachep; 33 34 int __init ext4_init_pageio(void) 35 { 36 io_page_cachep = KMEM_CACHE(ext4_io_page, SLAB_RECLAIM_ACCOUNT); 37 if (io_page_cachep == NULL) 38 return -ENOMEM; 39 io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT); 40 if (io_end_cachep == NULL) { 41 kmem_cache_destroy(io_page_cachep); 42 return -ENOMEM; 43 } 44 return 0; 45 } 46 47 void ext4_exit_pageio(void) 48 { 49 kmem_cache_destroy(io_end_cachep); 50 kmem_cache_destroy(io_page_cachep); 51 } 52 53 void ext4_ioend_wait(struct inode *inode) 54 { 55 wait_queue_head_t *wq = ext4_ioend_wq(inode); 56 57 wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_ioend_count) == 0)); 58 } 59 60 static void put_io_page(struct ext4_io_page *io_page) 61 { 62 if (atomic_dec_and_test(&io_page->p_count)) { 63 end_page_writeback(io_page->p_page); 64 put_page(io_page->p_page); 65 kmem_cache_free(io_page_cachep, io_page); 66 } 67 } 68 69 void ext4_free_io_end(ext4_io_end_t *io) 70 { 71 int i; 72 73 BUG_ON(!io); 74 BUG_ON(!list_empty(&io->list)); 75 BUG_ON(io->flag & EXT4_IO_END_UNWRITTEN); 76 77 if (io->page) 78 put_page(io->page); 79 for (i = 0; i < io->num_io_pages; i++) 80 put_io_page(io->pages[i]); 81 io->num_io_pages = 0; 82 if (atomic_dec_and_test(&EXT4_I(io->inode)->i_ioend_count)) 83 wake_up_all(ext4_ioend_wq(io->inode)); 84 kmem_cache_free(io_end_cachep, io); 85 } 86 87 /* check a range of space and convert unwritten extents to written. */ 88 static int ext4_end_io(ext4_io_end_t *io) 89 { 90 struct inode *inode = io->inode; 91 loff_t offset = io->offset; 92 ssize_t size = io->size; 93 int ret = 0; 94 95 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p," 96 "list->prev 0x%p\n", 97 io, inode->i_ino, io->list.next, io->list.prev); 98 99 ret = ext4_convert_unwritten_extents(inode, offset, size); 100 if (ret < 0) { 101 ext4_msg(inode->i_sb, KERN_EMERG, 102 "failed to convert unwritten extents to written " 103 "extents -- potential data loss! " 104 "(inode %lu, offset %llu, size %zd, error %d)", 105 inode->i_ino, offset, size, ret); 106 } 107 if (io->iocb) 108 aio_complete(io->iocb, io->result, 0); 109 110 if (io->flag & EXT4_IO_END_DIRECT) 111 inode_dio_done(inode); 112 /* Wake up anyone waiting on unwritten extent conversion */ 113 if (atomic_dec_and_test(&EXT4_I(inode)->i_unwritten)) 114 wake_up_all(ext4_ioend_wq(io->inode)); 115 return ret; 116 } 117 118 static void dump_completed_IO(struct inode *inode) 119 { 120 #ifdef EXT4FS_DEBUG 121 struct list_head *cur, *before, *after; 122 ext4_io_end_t *io, *io0, *io1; 123 unsigned long flags; 124 125 if (list_empty(&EXT4_I(inode)->i_completed_io_list)) { 126 ext4_debug("inode %lu completed_io list is empty\n", 127 inode->i_ino); 128 return; 129 } 130 131 ext4_debug("Dump inode %lu completed_io list\n", inode->i_ino); 132 list_for_each_entry(io, &EXT4_I(inode)->i_completed_io_list, list) { 133 cur = &io->list; 134 before = cur->prev; 135 io0 = container_of(before, ext4_io_end_t, list); 136 after = cur->next; 137 io1 = container_of(after, ext4_io_end_t, list); 138 139 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n", 140 io, inode->i_ino, io0, io1); 141 } 142 #endif 143 } 144 145 /* Add the io_end to per-inode completed end_io list. */ 146 void ext4_add_complete_io(ext4_io_end_t *io_end) 147 { 148 struct ext4_inode_info *ei = EXT4_I(io_end->inode); 149 struct workqueue_struct *wq; 150 unsigned long flags; 151 152 BUG_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN)); 153 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq; 154 155 spin_lock_irqsave(&ei->i_completed_io_lock, flags); 156 if (list_empty(&ei->i_completed_io_list)) { 157 io_end->flag |= EXT4_IO_END_QUEUED; 158 queue_work(wq, &io_end->work); 159 } 160 list_add_tail(&io_end->list, &ei->i_completed_io_list); 161 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); 162 } 163 164 static int ext4_do_flush_completed_IO(struct inode *inode, 165 ext4_io_end_t *work_io) 166 { 167 ext4_io_end_t *io; 168 struct list_head unwritten, complete, to_free; 169 unsigned long flags; 170 struct ext4_inode_info *ei = EXT4_I(inode); 171 int err, ret = 0; 172 173 INIT_LIST_HEAD(&complete); 174 INIT_LIST_HEAD(&to_free); 175 176 spin_lock_irqsave(&ei->i_completed_io_lock, flags); 177 dump_completed_IO(inode); 178 list_replace_init(&ei->i_completed_io_list, &unwritten); 179 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); 180 181 while (!list_empty(&unwritten)) { 182 io = list_entry(unwritten.next, ext4_io_end_t, list); 183 BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN)); 184 list_del_init(&io->list); 185 186 err = ext4_end_io(io); 187 if (unlikely(!ret && err)) 188 ret = err; 189 190 list_add_tail(&io->list, &complete); 191 } 192 spin_lock_irqsave(&ei->i_completed_io_lock, flags); 193 while (!list_empty(&complete)) { 194 io = list_entry(complete.next, ext4_io_end_t, list); 195 io->flag &= ~EXT4_IO_END_UNWRITTEN; 196 /* end_io context can not be destroyed now because it still 197 * used by queued worker. Worker thread will destroy it later */ 198 if (io->flag & EXT4_IO_END_QUEUED) 199 list_del_init(&io->list); 200 else 201 list_move(&io->list, &to_free); 202 } 203 /* If we are called from worker context, it is time to clear queued 204 * flag, and destroy it's end_io if it was converted already */ 205 if (work_io) { 206 work_io->flag &= ~EXT4_IO_END_QUEUED; 207 if (!(work_io->flag & EXT4_IO_END_UNWRITTEN)) 208 list_add_tail(&work_io->list, &to_free); 209 } 210 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); 211 212 while (!list_empty(&to_free)) { 213 io = list_entry(to_free.next, ext4_io_end_t, list); 214 list_del_init(&io->list); 215 ext4_free_io_end(io); 216 } 217 return ret; 218 } 219 220 /* 221 * work on completed aio dio IO, to convert unwritten extents to extents 222 */ 223 static void ext4_end_io_work(struct work_struct *work) 224 { 225 ext4_io_end_t *io = container_of(work, ext4_io_end_t, work); 226 ext4_do_flush_completed_IO(io->inode, io); 227 } 228 229 int ext4_flush_unwritten_io(struct inode *inode) 230 { 231 int ret; 232 WARN_ON_ONCE(!mutex_is_locked(&inode->i_mutex) && 233 !(inode->i_state & I_FREEING)); 234 ret = ext4_do_flush_completed_IO(inode, NULL); 235 ext4_unwritten_wait(inode); 236 return ret; 237 } 238 239 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags) 240 { 241 ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags); 242 if (io) { 243 atomic_inc(&EXT4_I(inode)->i_ioend_count); 244 io->inode = inode; 245 INIT_WORK(&io->work, ext4_end_io_work); 246 INIT_LIST_HEAD(&io->list); 247 } 248 return io; 249 } 250 251 /* 252 * Print an buffer I/O error compatible with the fs/buffer.c. This 253 * provides compatibility with dmesg scrapers that look for a specific 254 * buffer I/O error message. We really need a unified error reporting 255 * structure to userspace ala Digital Unix's uerf system, but it's 256 * probably not going to happen in my lifetime, due to LKML politics... 257 */ 258 static void buffer_io_error(struct buffer_head *bh) 259 { 260 char b[BDEVNAME_SIZE]; 261 printk(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n", 262 bdevname(bh->b_bdev, b), 263 (unsigned long long)bh->b_blocknr); 264 } 265 266 static void ext4_end_bio(struct bio *bio, int error) 267 { 268 ext4_io_end_t *io_end = bio->bi_private; 269 struct inode *inode; 270 int i; 271 sector_t bi_sector = bio->bi_sector; 272 273 BUG_ON(!io_end); 274 bio->bi_private = NULL; 275 bio->bi_end_io = NULL; 276 if (test_bit(BIO_UPTODATE, &bio->bi_flags)) 277 error = 0; 278 bio_put(bio); 279 280 for (i = 0; i < io_end->num_io_pages; i++) { 281 struct page *page = io_end->pages[i]->p_page; 282 struct buffer_head *bh, *head; 283 loff_t offset; 284 loff_t io_end_offset; 285 286 if (error) { 287 SetPageError(page); 288 set_bit(AS_EIO, &page->mapping->flags); 289 head = page_buffers(page); 290 BUG_ON(!head); 291 292 io_end_offset = io_end->offset + io_end->size; 293 294 offset = (sector_t) page->index << PAGE_CACHE_SHIFT; 295 bh = head; 296 do { 297 if ((offset >= io_end->offset) && 298 (offset+bh->b_size <= io_end_offset)) 299 buffer_io_error(bh); 300 301 offset += bh->b_size; 302 bh = bh->b_this_page; 303 } while (bh != head); 304 } 305 306 put_io_page(io_end->pages[i]); 307 } 308 io_end->num_io_pages = 0; 309 inode = io_end->inode; 310 311 if (error) { 312 io_end->flag |= EXT4_IO_END_ERROR; 313 ext4_warning(inode->i_sb, "I/O error writing to inode %lu " 314 "(offset %llu size %ld starting block %llu)", 315 inode->i_ino, 316 (unsigned long long) io_end->offset, 317 (long) io_end->size, 318 (unsigned long long) 319 bi_sector >> (inode->i_blkbits - 9)); 320 } 321 322 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) { 323 ext4_free_io_end(io_end); 324 return; 325 } 326 327 ext4_add_complete_io(io_end); 328 } 329 330 void ext4_io_submit(struct ext4_io_submit *io) 331 { 332 struct bio *bio = io->io_bio; 333 334 if (bio) { 335 bio_get(io->io_bio); 336 submit_bio(io->io_op, io->io_bio); 337 BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP)); 338 bio_put(io->io_bio); 339 } 340 io->io_bio = NULL; 341 io->io_op = 0; 342 io->io_end = NULL; 343 } 344 345 static int io_submit_init(struct ext4_io_submit *io, 346 struct inode *inode, 347 struct writeback_control *wbc, 348 struct buffer_head *bh) 349 { 350 ext4_io_end_t *io_end; 351 struct page *page = bh->b_page; 352 int nvecs = bio_get_nr_vecs(bh->b_bdev); 353 struct bio *bio; 354 355 io_end = ext4_init_io_end(inode, GFP_NOFS); 356 if (!io_end) 357 return -ENOMEM; 358 bio = bio_alloc(GFP_NOIO, min(nvecs, BIO_MAX_PAGES)); 359 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); 360 bio->bi_bdev = bh->b_bdev; 361 bio->bi_private = io->io_end = io_end; 362 bio->bi_end_io = ext4_end_bio; 363 364 io_end->offset = (page->index << PAGE_CACHE_SHIFT) + bh_offset(bh); 365 366 io->io_bio = bio; 367 io->io_op = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE); 368 io->io_next_block = bh->b_blocknr; 369 return 0; 370 } 371 372 static int io_submit_add_bh(struct ext4_io_submit *io, 373 struct ext4_io_page *io_page, 374 struct inode *inode, 375 struct writeback_control *wbc, 376 struct buffer_head *bh) 377 { 378 ext4_io_end_t *io_end; 379 int ret; 380 381 if (buffer_new(bh)) { 382 clear_buffer_new(bh); 383 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); 384 } 385 386 if (!buffer_mapped(bh) || buffer_delay(bh)) { 387 if (!buffer_mapped(bh)) 388 clear_buffer_dirty(bh); 389 if (io->io_bio) 390 ext4_io_submit(io); 391 return 0; 392 } 393 394 if (io->io_bio && bh->b_blocknr != io->io_next_block) { 395 submit_and_retry: 396 ext4_io_submit(io); 397 } 398 if (io->io_bio == NULL) { 399 ret = io_submit_init(io, inode, wbc, bh); 400 if (ret) 401 return ret; 402 } 403 io_end = io->io_end; 404 if ((io_end->num_io_pages >= MAX_IO_PAGES) && 405 (io_end->pages[io_end->num_io_pages-1] != io_page)) 406 goto submit_and_retry; 407 if (buffer_uninit(bh)) 408 ext4_set_io_unwritten_flag(inode, io_end); 409 io->io_end->size += bh->b_size; 410 io->io_next_block++; 411 ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh)); 412 if (ret != bh->b_size) 413 goto submit_and_retry; 414 if ((io_end->num_io_pages == 0) || 415 (io_end->pages[io_end->num_io_pages-1] != io_page)) { 416 io_end->pages[io_end->num_io_pages++] = io_page; 417 atomic_inc(&io_page->p_count); 418 } 419 return 0; 420 } 421 422 int ext4_bio_write_page(struct ext4_io_submit *io, 423 struct page *page, 424 int len, 425 struct writeback_control *wbc) 426 { 427 struct inode *inode = page->mapping->host; 428 unsigned block_start, block_end, blocksize; 429 struct ext4_io_page *io_page; 430 struct buffer_head *bh, *head; 431 int ret = 0; 432 433 blocksize = 1 << inode->i_blkbits; 434 435 BUG_ON(!PageLocked(page)); 436 BUG_ON(PageWriteback(page)); 437 438 io_page = kmem_cache_alloc(io_page_cachep, GFP_NOFS); 439 if (!io_page) { 440 set_page_dirty(page); 441 unlock_page(page); 442 return -ENOMEM; 443 } 444 io_page->p_page = page; 445 atomic_set(&io_page->p_count, 1); 446 get_page(page); 447 set_page_writeback(page); 448 ClearPageError(page); 449 450 for (bh = head = page_buffers(page), block_start = 0; 451 bh != head || !block_start; 452 block_start = block_end, bh = bh->b_this_page) { 453 454 block_end = block_start + blocksize; 455 if (block_start >= len) { 456 /* 457 * Comments copied from block_write_full_page_endio: 458 * 459 * The page straddles i_size. It must be zeroed out on 460 * each and every writepage invocation because it may 461 * be mmapped. "A file is mapped in multiples of the 462 * page size. For a file that is not a multiple of 463 * the page size, the remaining memory is zeroed when 464 * mapped, and writes to that region are not written 465 * out to the file." 466 */ 467 zero_user_segment(page, block_start, block_end); 468 clear_buffer_dirty(bh); 469 set_buffer_uptodate(bh); 470 continue; 471 } 472 clear_buffer_dirty(bh); 473 ret = io_submit_add_bh(io, io_page, inode, wbc, bh); 474 if (ret) { 475 /* 476 * We only get here on ENOMEM. Not much else 477 * we can do but mark the page as dirty, and 478 * better luck next time. 479 */ 480 set_page_dirty(page); 481 break; 482 } 483 } 484 unlock_page(page); 485 /* 486 * If the page was truncated before we could do the writeback, 487 * or we had a memory allocation error while trying to write 488 * the first buffer head, we won't have submitted any pages for 489 * I/O. In that case we need to make sure we've cleared the 490 * PageWriteback bit from the page to prevent the system from 491 * wedging later on. 492 */ 493 put_io_page(io_page); 494 return ret; 495 } 496