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/module.h> 10 #include <linux/fs.h> 11 #include <linux/time.h> 12 #include <linux/jbd2.h> 13 #include <linux/highuid.h> 14 #include <linux/pagemap.h> 15 #include <linux/quotaops.h> 16 #include <linux/string.h> 17 #include <linux/buffer_head.h> 18 #include <linux/writeback.h> 19 #include <linux/pagevec.h> 20 #include <linux/mpage.h> 21 #include <linux/namei.h> 22 #include <linux/uio.h> 23 #include <linux/bio.h> 24 #include <linux/workqueue.h> 25 #include <linux/kernel.h> 26 #include <linux/slab.h> 27 28 #include "ext4_jbd2.h" 29 #include "xattr.h" 30 #include "acl.h" 31 #include "ext4_extents.h" 32 33 static struct kmem_cache *io_page_cachep, *io_end_cachep; 34 35 #define WQ_HASH_SZ 37 36 #define to_ioend_wq(v) (&ioend_wq[((unsigned long)v) % WQ_HASH_SZ]) 37 static wait_queue_head_t ioend_wq[WQ_HASH_SZ]; 38 39 int __init ext4_init_pageio(void) 40 { 41 int i; 42 43 io_page_cachep = KMEM_CACHE(ext4_io_page, SLAB_RECLAIM_ACCOUNT); 44 if (io_page_cachep == NULL) 45 return -ENOMEM; 46 io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT); 47 if (io_end_cachep == NULL) { 48 kmem_cache_destroy(io_page_cachep); 49 return -ENOMEM; 50 } 51 for (i = 0; i < WQ_HASH_SZ; i++) 52 init_waitqueue_head(&ioend_wq[i]); 53 54 return 0; 55 } 56 57 void ext4_exit_pageio(void) 58 { 59 kmem_cache_destroy(io_end_cachep); 60 kmem_cache_destroy(io_page_cachep); 61 } 62 63 void ext4_ioend_wait(struct inode *inode) 64 { 65 wait_queue_head_t *wq = to_ioend_wq(inode); 66 67 wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_ioend_count) == 0)); 68 } 69 70 static void put_io_page(struct ext4_io_page *io_page) 71 { 72 if (atomic_dec_and_test(&io_page->p_count)) { 73 end_page_writeback(io_page->p_page); 74 put_page(io_page->p_page); 75 kmem_cache_free(io_page_cachep, io_page); 76 } 77 } 78 79 void ext4_free_io_end(ext4_io_end_t *io) 80 { 81 int i; 82 wait_queue_head_t *wq; 83 84 BUG_ON(!io); 85 if (io->page) 86 put_page(io->page); 87 for (i = 0; i < io->num_io_pages; i++) 88 put_io_page(io->pages[i]); 89 io->num_io_pages = 0; 90 wq = to_ioend_wq(io->inode); 91 if (atomic_dec_and_test(&EXT4_I(io->inode)->i_ioend_count) && 92 waitqueue_active(wq)) 93 wake_up_all(wq); 94 kmem_cache_free(io_end_cachep, io); 95 } 96 97 /* 98 * check a range of space and convert unwritten extents to written. 99 */ 100 int ext4_end_io_nolock(ext4_io_end_t *io) 101 { 102 struct inode *inode = io->inode; 103 loff_t offset = io->offset; 104 ssize_t size = io->size; 105 int ret = 0; 106 107 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p," 108 "list->prev 0x%p\n", 109 io, inode->i_ino, io->list.next, io->list.prev); 110 111 if (list_empty(&io->list)) 112 return ret; 113 114 if (!(io->flag & EXT4_IO_END_UNWRITTEN)) 115 return ret; 116 117 ret = ext4_convert_unwritten_extents(inode, offset, size); 118 if (ret < 0) { 119 printk(KERN_EMERG "%s: failed to convert unwritten " 120 "extents to written extents, error is %d " 121 "io is still on inode %lu aio dio list\n", 122 __func__, ret, inode->i_ino); 123 return ret; 124 } 125 126 if (io->iocb) 127 aio_complete(io->iocb, io->result, 0); 128 /* clear the DIO AIO unwritten flag */ 129 io->flag &= ~EXT4_IO_END_UNWRITTEN; 130 return ret; 131 } 132 133 /* 134 * work on completed aio dio IO, to convert unwritten extents to extents 135 */ 136 static void ext4_end_io_work(struct work_struct *work) 137 { 138 ext4_io_end_t *io = container_of(work, ext4_io_end_t, work); 139 struct inode *inode = io->inode; 140 struct ext4_inode_info *ei = EXT4_I(inode); 141 unsigned long flags; 142 int ret; 143 144 mutex_lock(&inode->i_mutex); 145 ret = ext4_end_io_nolock(io); 146 if (ret < 0) { 147 mutex_unlock(&inode->i_mutex); 148 return; 149 } 150 151 spin_lock_irqsave(&ei->i_completed_io_lock, flags); 152 if (!list_empty(&io->list)) 153 list_del_init(&io->list); 154 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); 155 mutex_unlock(&inode->i_mutex); 156 ext4_free_io_end(io); 157 } 158 159 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags) 160 { 161 ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags); 162 if (io) { 163 atomic_inc(&EXT4_I(inode)->i_ioend_count); 164 io->inode = inode; 165 INIT_WORK(&io->work, ext4_end_io_work); 166 INIT_LIST_HEAD(&io->list); 167 } 168 return io; 169 } 170 171 /* 172 * Print an buffer I/O error compatible with the fs/buffer.c. This 173 * provides compatibility with dmesg scrapers that look for a specific 174 * buffer I/O error message. We really need a unified error reporting 175 * structure to userspace ala Digital Unix's uerf system, but it's 176 * probably not going to happen in my lifetime, due to LKML politics... 177 */ 178 static void buffer_io_error(struct buffer_head *bh) 179 { 180 char b[BDEVNAME_SIZE]; 181 printk(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n", 182 bdevname(bh->b_bdev, b), 183 (unsigned long long)bh->b_blocknr); 184 } 185 186 static void ext4_end_bio(struct bio *bio, int error) 187 { 188 ext4_io_end_t *io_end = bio->bi_private; 189 struct workqueue_struct *wq; 190 struct inode *inode; 191 unsigned long flags; 192 int i; 193 194 BUG_ON(!io_end); 195 bio->bi_private = NULL; 196 bio->bi_end_io = NULL; 197 if (test_bit(BIO_UPTODATE, &bio->bi_flags)) 198 error = 0; 199 bio_put(bio); 200 201 for (i = 0; i < io_end->num_io_pages; i++) { 202 struct page *page = io_end->pages[i]->p_page; 203 struct buffer_head *bh, *head; 204 int partial_write = 0; 205 206 head = page_buffers(page); 207 if (error) 208 SetPageError(page); 209 BUG_ON(!head); 210 if (head->b_size == PAGE_CACHE_SIZE) 211 clear_buffer_dirty(head); 212 else { 213 loff_t offset; 214 loff_t io_end_offset = io_end->offset + io_end->size; 215 216 offset = (sector_t) page->index << PAGE_CACHE_SHIFT; 217 bh = head; 218 do { 219 if ((offset >= io_end->offset) && 220 (offset+bh->b_size <= io_end_offset)) { 221 if (error) 222 buffer_io_error(bh); 223 224 clear_buffer_dirty(bh); 225 } 226 if (buffer_delay(bh)) 227 partial_write = 1; 228 else if (!buffer_mapped(bh)) 229 clear_buffer_dirty(bh); 230 else if (buffer_dirty(bh)) 231 partial_write = 1; 232 offset += bh->b_size; 233 bh = bh->b_this_page; 234 } while (bh != head); 235 } 236 237 /* 238 * If this is a partial write which happened to make 239 * all buffers uptodate then we can optimize away a 240 * bogus readpage() for the next read(). Here we 241 * 'discover' whether the page went uptodate as a 242 * result of this (potentially partial) write. 243 */ 244 if (!partial_write) 245 SetPageUptodate(page); 246 247 put_io_page(io_end->pages[i]); 248 } 249 io_end->num_io_pages = 0; 250 inode = io_end->inode; 251 252 if (error) { 253 io_end->flag |= EXT4_IO_END_ERROR; 254 ext4_warning(inode->i_sb, "I/O error writing to inode %lu " 255 "(offset %llu size %ld starting block %llu)", 256 inode->i_ino, 257 (unsigned long long) io_end->offset, 258 (long) io_end->size, 259 (unsigned long long) 260 bio->bi_sector >> (inode->i_blkbits - 9)); 261 } 262 263 /* Add the io_end to per-inode completed io list*/ 264 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags); 265 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list); 266 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags); 267 268 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq; 269 /* queue the work to convert unwritten extents to written */ 270 queue_work(wq, &io_end->work); 271 } 272 273 void ext4_io_submit(struct ext4_io_submit *io) 274 { 275 struct bio *bio = io->io_bio; 276 277 if (bio) { 278 bio_get(io->io_bio); 279 submit_bio(io->io_op, io->io_bio); 280 BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP)); 281 bio_put(io->io_bio); 282 } 283 io->io_bio = 0; 284 io->io_op = 0; 285 io->io_end = 0; 286 } 287 288 static int io_submit_init(struct ext4_io_submit *io, 289 struct inode *inode, 290 struct writeback_control *wbc, 291 struct buffer_head *bh) 292 { 293 ext4_io_end_t *io_end; 294 struct page *page = bh->b_page; 295 int nvecs = bio_get_nr_vecs(bh->b_bdev); 296 struct bio *bio; 297 298 io_end = ext4_init_io_end(inode, GFP_NOFS); 299 if (!io_end) 300 return -ENOMEM; 301 do { 302 bio = bio_alloc(GFP_NOIO, nvecs); 303 nvecs >>= 1; 304 } while (bio == NULL); 305 306 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); 307 bio->bi_bdev = bh->b_bdev; 308 bio->bi_private = io->io_end = io_end; 309 bio->bi_end_io = ext4_end_bio; 310 311 io_end->offset = (page->index << PAGE_CACHE_SHIFT) + bh_offset(bh); 312 313 io->io_bio = bio; 314 io->io_op = (wbc->sync_mode == WB_SYNC_ALL ? 315 WRITE_SYNC_PLUG : WRITE); 316 io->io_next_block = bh->b_blocknr; 317 return 0; 318 } 319 320 static int io_submit_add_bh(struct ext4_io_submit *io, 321 struct ext4_io_page *io_page, 322 struct inode *inode, 323 struct writeback_control *wbc, 324 struct buffer_head *bh) 325 { 326 ext4_io_end_t *io_end; 327 int ret; 328 329 if (buffer_new(bh)) { 330 clear_buffer_new(bh); 331 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); 332 } 333 334 if (!buffer_mapped(bh) || buffer_delay(bh)) { 335 if (!buffer_mapped(bh)) 336 clear_buffer_dirty(bh); 337 if (io->io_bio) 338 ext4_io_submit(io); 339 return 0; 340 } 341 342 if (io->io_bio && bh->b_blocknr != io->io_next_block) { 343 submit_and_retry: 344 ext4_io_submit(io); 345 } 346 if (io->io_bio == NULL) { 347 ret = io_submit_init(io, inode, wbc, bh); 348 if (ret) 349 return ret; 350 } 351 io_end = io->io_end; 352 if ((io_end->num_io_pages >= MAX_IO_PAGES) && 353 (io_end->pages[io_end->num_io_pages-1] != io_page)) 354 goto submit_and_retry; 355 if (buffer_uninit(bh)) 356 io->io_end->flag |= EXT4_IO_END_UNWRITTEN; 357 io->io_end->size += bh->b_size; 358 io->io_next_block++; 359 ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh)); 360 if (ret != bh->b_size) 361 goto submit_and_retry; 362 if ((io_end->num_io_pages == 0) || 363 (io_end->pages[io_end->num_io_pages-1] != io_page)) { 364 io_end->pages[io_end->num_io_pages++] = io_page; 365 atomic_inc(&io_page->p_count); 366 } 367 return 0; 368 } 369 370 int ext4_bio_write_page(struct ext4_io_submit *io, 371 struct page *page, 372 int len, 373 struct writeback_control *wbc) 374 { 375 struct inode *inode = page->mapping->host; 376 unsigned block_start, block_end, blocksize; 377 struct ext4_io_page *io_page; 378 struct buffer_head *bh, *head; 379 int ret = 0; 380 381 blocksize = 1 << inode->i_blkbits; 382 383 BUG_ON(PageWriteback(page)); 384 set_page_writeback(page); 385 ClearPageError(page); 386 387 io_page = kmem_cache_alloc(io_page_cachep, GFP_NOFS); 388 if (!io_page) { 389 set_page_dirty(page); 390 unlock_page(page); 391 return -ENOMEM; 392 } 393 io_page->p_page = page; 394 atomic_set(&io_page->p_count, 1); 395 get_page(page); 396 397 for (bh = head = page_buffers(page), block_start = 0; 398 bh != head || !block_start; 399 block_start = block_end, bh = bh->b_this_page) { 400 block_end = block_start + blocksize; 401 if (block_start >= len) { 402 clear_buffer_dirty(bh); 403 set_buffer_uptodate(bh); 404 continue; 405 } 406 ret = io_submit_add_bh(io, io_page, inode, wbc, bh); 407 if (ret) { 408 /* 409 * We only get here on ENOMEM. Not much else 410 * we can do but mark the page as dirty, and 411 * better luck next time. 412 */ 413 set_page_dirty(page); 414 break; 415 } 416 } 417 unlock_page(page); 418 /* 419 * If the page was truncated before we could do the writeback, 420 * or we had a memory allocation error while trying to write 421 * the first buffer head, we won't have submitted any pages for 422 * I/O. In that case we need to make sure we've cleared the 423 * PageWriteback bit from the page to prevent the system from 424 * wedging later on. 425 */ 426 put_io_page(io_page); 427 return ret; 428 } 429