1bd2d0210STheodore Ts'o /* 2bd2d0210STheodore Ts'o * linux/fs/ext4/page-io.c 3bd2d0210STheodore Ts'o * 4bd2d0210STheodore Ts'o * This contains the new page_io functions for ext4 5bd2d0210STheodore Ts'o * 6bd2d0210STheodore Ts'o * Written by Theodore Ts'o, 2010. 7bd2d0210STheodore Ts'o */ 8bd2d0210STheodore Ts'o 9bd2d0210STheodore Ts'o #include <linux/module.h> 10bd2d0210STheodore Ts'o #include <linux/fs.h> 11bd2d0210STheodore Ts'o #include <linux/time.h> 12bd2d0210STheodore Ts'o #include <linux/jbd2.h> 13bd2d0210STheodore Ts'o #include <linux/highuid.h> 14bd2d0210STheodore Ts'o #include <linux/pagemap.h> 15bd2d0210STheodore Ts'o #include <linux/quotaops.h> 16bd2d0210STheodore Ts'o #include <linux/string.h> 17bd2d0210STheodore Ts'o #include <linux/buffer_head.h> 18bd2d0210STheodore Ts'o #include <linux/writeback.h> 19bd2d0210STheodore Ts'o #include <linux/pagevec.h> 20bd2d0210STheodore Ts'o #include <linux/mpage.h> 21bd2d0210STheodore Ts'o #include <linux/namei.h> 22bd2d0210STheodore Ts'o #include <linux/uio.h> 23bd2d0210STheodore Ts'o #include <linux/bio.h> 24bd2d0210STheodore Ts'o #include <linux/workqueue.h> 25bd2d0210STheodore Ts'o #include <linux/kernel.h> 26bd2d0210STheodore Ts'o #include <linux/slab.h> 27bd2d0210STheodore Ts'o 28bd2d0210STheodore Ts'o #include "ext4_jbd2.h" 29bd2d0210STheodore Ts'o #include "xattr.h" 30bd2d0210STheodore Ts'o #include "acl.h" 31bd2d0210STheodore Ts'o #include "ext4_extents.h" 32bd2d0210STheodore Ts'o 33bd2d0210STheodore Ts'o static struct kmem_cache *io_page_cachep, *io_end_cachep; 34bd2d0210STheodore Ts'o 35f7ad6d2eSTheodore Ts'o #define WQ_HASH_SZ 37 36f7ad6d2eSTheodore Ts'o #define to_ioend_wq(v) (&ioend_wq[((unsigned long)v) % WQ_HASH_SZ]) 37f7ad6d2eSTheodore Ts'o static wait_queue_head_t ioend_wq[WQ_HASH_SZ]; 38f7ad6d2eSTheodore Ts'o 395dabfc78STheodore Ts'o int __init ext4_init_pageio(void) 40bd2d0210STheodore Ts'o { 41f7ad6d2eSTheodore Ts'o int i; 42f7ad6d2eSTheodore Ts'o 43bd2d0210STheodore Ts'o io_page_cachep = KMEM_CACHE(ext4_io_page, SLAB_RECLAIM_ACCOUNT); 44bd2d0210STheodore Ts'o if (io_page_cachep == NULL) 45bd2d0210STheodore Ts'o return -ENOMEM; 46bd2d0210STheodore Ts'o io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT); 47bd2d0210STheodore Ts'o if (io_page_cachep == NULL) { 48bd2d0210STheodore Ts'o kmem_cache_destroy(io_page_cachep); 49bd2d0210STheodore Ts'o return -ENOMEM; 50bd2d0210STheodore Ts'o } 51f7ad6d2eSTheodore Ts'o for (i = 0; i < WQ_HASH_SZ; i++) 52f7ad6d2eSTheodore Ts'o init_waitqueue_head(&ioend_wq[i]); 53bd2d0210STheodore Ts'o 54bd2d0210STheodore Ts'o return 0; 55bd2d0210STheodore Ts'o } 56bd2d0210STheodore Ts'o 575dabfc78STheodore Ts'o void ext4_exit_pageio(void) 58bd2d0210STheodore Ts'o { 59bd2d0210STheodore Ts'o kmem_cache_destroy(io_end_cachep); 60bd2d0210STheodore Ts'o kmem_cache_destroy(io_page_cachep); 61bd2d0210STheodore Ts'o } 62bd2d0210STheodore Ts'o 63f7ad6d2eSTheodore Ts'o void ext4_ioend_wait(struct inode *inode) 64f7ad6d2eSTheodore Ts'o { 65f7ad6d2eSTheodore Ts'o wait_queue_head_t *wq = to_ioend_wq(inode); 66f7ad6d2eSTheodore Ts'o 67f7ad6d2eSTheodore Ts'o wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_ioend_count) == 0)); 68f7ad6d2eSTheodore Ts'o } 69f7ad6d2eSTheodore Ts'o 70*83668e71STheodore Ts'o static void put_io_page(struct ext4_io_page *io_page) 71*83668e71STheodore Ts'o { 72*83668e71STheodore Ts'o if (atomic_dec_and_test(&io_page->p_count)) { 73*83668e71STheodore Ts'o end_page_writeback(io_page->p_page); 74*83668e71STheodore Ts'o put_page(io_page->p_page); 75*83668e71STheodore Ts'o kmem_cache_free(io_page_cachep, io_page); 76*83668e71STheodore Ts'o } 77*83668e71STheodore Ts'o } 78*83668e71STheodore Ts'o 79bd2d0210STheodore Ts'o void ext4_free_io_end(ext4_io_end_t *io) 80bd2d0210STheodore Ts'o { 81bd2d0210STheodore Ts'o int i; 82f7ad6d2eSTheodore Ts'o wait_queue_head_t *wq; 83bd2d0210STheodore Ts'o 84bd2d0210STheodore Ts'o BUG_ON(!io); 85bd2d0210STheodore Ts'o if (io->page) 86bd2d0210STheodore Ts'o put_page(io->page); 87*83668e71STheodore Ts'o for (i = 0; i < io->num_io_pages; i++) 88*83668e71STheodore Ts'o put_io_page(io->pages[i]); 89bd2d0210STheodore Ts'o io->num_io_pages = 0; 90f7ad6d2eSTheodore Ts'o wq = to_ioend_wq(io->inode); 91f7ad6d2eSTheodore Ts'o if (atomic_dec_and_test(&EXT4_I(io->inode)->i_ioend_count) && 92f7ad6d2eSTheodore Ts'o waitqueue_active(wq)) 93f7ad6d2eSTheodore Ts'o wake_up_all(wq); 94bd2d0210STheodore Ts'o kmem_cache_free(io_end_cachep, io); 95bd2d0210STheodore Ts'o } 96bd2d0210STheodore Ts'o 97bd2d0210STheodore Ts'o /* 98bd2d0210STheodore Ts'o * check a range of space and convert unwritten extents to written. 99bd2d0210STheodore Ts'o */ 100bd2d0210STheodore Ts'o int ext4_end_io_nolock(ext4_io_end_t *io) 101bd2d0210STheodore Ts'o { 102bd2d0210STheodore Ts'o struct inode *inode = io->inode; 103bd2d0210STheodore Ts'o loff_t offset = io->offset; 104bd2d0210STheodore Ts'o ssize_t size = io->size; 105bd2d0210STheodore Ts'o int ret = 0; 106bd2d0210STheodore Ts'o 107bd2d0210STheodore Ts'o ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p," 108bd2d0210STheodore Ts'o "list->prev 0x%p\n", 109bd2d0210STheodore Ts'o io, inode->i_ino, io->list.next, io->list.prev); 110bd2d0210STheodore Ts'o 111bd2d0210STheodore Ts'o if (list_empty(&io->list)) 112bd2d0210STheodore Ts'o return ret; 113bd2d0210STheodore Ts'o 114bd2d0210STheodore Ts'o if (!(io->flag & EXT4_IO_END_UNWRITTEN)) 115bd2d0210STheodore Ts'o return ret; 116bd2d0210STheodore Ts'o 117bd2d0210STheodore Ts'o ret = ext4_convert_unwritten_extents(inode, offset, size); 118bd2d0210STheodore Ts'o if (ret < 0) { 119bd2d0210STheodore Ts'o printk(KERN_EMERG "%s: failed to convert unwritten " 120bd2d0210STheodore Ts'o "extents to written extents, error is %d " 121bd2d0210STheodore Ts'o "io is still on inode %lu aio dio list\n", 122bd2d0210STheodore Ts'o __func__, ret, inode->i_ino); 123bd2d0210STheodore Ts'o return ret; 124bd2d0210STheodore Ts'o } 125bd2d0210STheodore Ts'o 126bd2d0210STheodore Ts'o if (io->iocb) 127bd2d0210STheodore Ts'o aio_complete(io->iocb, io->result, 0); 128bd2d0210STheodore Ts'o /* clear the DIO AIO unwritten flag */ 129bd2d0210STheodore Ts'o io->flag &= ~EXT4_IO_END_UNWRITTEN; 130bd2d0210STheodore Ts'o return ret; 131bd2d0210STheodore Ts'o } 132bd2d0210STheodore Ts'o 133bd2d0210STheodore Ts'o /* 134bd2d0210STheodore Ts'o * work on completed aio dio IO, to convert unwritten extents to extents 135bd2d0210STheodore Ts'o */ 136bd2d0210STheodore Ts'o static void ext4_end_io_work(struct work_struct *work) 137bd2d0210STheodore Ts'o { 138bd2d0210STheodore Ts'o ext4_io_end_t *io = container_of(work, ext4_io_end_t, work); 139bd2d0210STheodore Ts'o struct inode *inode = io->inode; 140bd2d0210STheodore Ts'o struct ext4_inode_info *ei = EXT4_I(inode); 141bd2d0210STheodore Ts'o unsigned long flags; 142bd2d0210STheodore Ts'o int ret; 143bd2d0210STheodore Ts'o 144bd2d0210STheodore Ts'o mutex_lock(&inode->i_mutex); 145bd2d0210STheodore Ts'o ret = ext4_end_io_nolock(io); 146bd2d0210STheodore Ts'o if (ret < 0) { 147bd2d0210STheodore Ts'o mutex_unlock(&inode->i_mutex); 148bd2d0210STheodore Ts'o return; 149bd2d0210STheodore Ts'o } 150bd2d0210STheodore Ts'o 151bd2d0210STheodore Ts'o spin_lock_irqsave(&ei->i_completed_io_lock, flags); 152bd2d0210STheodore Ts'o if (!list_empty(&io->list)) 153bd2d0210STheodore Ts'o list_del_init(&io->list); 154bd2d0210STheodore Ts'o spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); 155bd2d0210STheodore Ts'o mutex_unlock(&inode->i_mutex); 156bd2d0210STheodore Ts'o ext4_free_io_end(io); 157bd2d0210STheodore Ts'o } 158bd2d0210STheodore Ts'o 159bd2d0210STheodore Ts'o ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags) 160bd2d0210STheodore Ts'o { 161bd2d0210STheodore Ts'o ext4_io_end_t *io = NULL; 162bd2d0210STheodore Ts'o 163bd2d0210STheodore Ts'o io = kmem_cache_alloc(io_end_cachep, flags); 164bd2d0210STheodore Ts'o if (io) { 165bd2d0210STheodore Ts'o memset(io, 0, sizeof(*io)); 166f7ad6d2eSTheodore Ts'o atomic_inc(&EXT4_I(inode)->i_ioend_count); 167f7ad6d2eSTheodore Ts'o io->inode = inode; 168bd2d0210STheodore Ts'o INIT_WORK(&io->work, ext4_end_io_work); 169bd2d0210STheodore Ts'o INIT_LIST_HEAD(&io->list); 170bd2d0210STheodore Ts'o } 171bd2d0210STheodore Ts'o return io; 172bd2d0210STheodore Ts'o } 173bd2d0210STheodore Ts'o 174bd2d0210STheodore Ts'o /* 175bd2d0210STheodore Ts'o * Print an buffer I/O error compatible with the fs/buffer.c. This 176bd2d0210STheodore Ts'o * provides compatibility with dmesg scrapers that look for a specific 177bd2d0210STheodore Ts'o * buffer I/O error message. We really need a unified error reporting 178bd2d0210STheodore Ts'o * structure to userspace ala Digital Unix's uerf system, but it's 179bd2d0210STheodore Ts'o * probably not going to happen in my lifetime, due to LKML politics... 180bd2d0210STheodore Ts'o */ 181bd2d0210STheodore Ts'o static void buffer_io_error(struct buffer_head *bh) 182bd2d0210STheodore Ts'o { 183bd2d0210STheodore Ts'o char b[BDEVNAME_SIZE]; 184bd2d0210STheodore Ts'o printk(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n", 185bd2d0210STheodore Ts'o bdevname(bh->b_bdev, b), 186bd2d0210STheodore Ts'o (unsigned long long)bh->b_blocknr); 187bd2d0210STheodore Ts'o } 188bd2d0210STheodore Ts'o 189bd2d0210STheodore Ts'o static void ext4_end_bio(struct bio *bio, int error) 190bd2d0210STheodore Ts'o { 191bd2d0210STheodore Ts'o ext4_io_end_t *io_end = bio->bi_private; 192bd2d0210STheodore Ts'o struct workqueue_struct *wq; 193bd2d0210STheodore Ts'o struct inode *inode; 194bd2d0210STheodore Ts'o unsigned long flags; 195bd2d0210STheodore Ts'o int i; 196bd2d0210STheodore Ts'o 197bd2d0210STheodore Ts'o BUG_ON(!io_end); 198bd2d0210STheodore Ts'o bio->bi_private = NULL; 199bd2d0210STheodore Ts'o bio->bi_end_io = NULL; 200bd2d0210STheodore Ts'o if (test_bit(BIO_UPTODATE, &bio->bi_flags)) 201bd2d0210STheodore Ts'o error = 0; 202bd2d0210STheodore Ts'o bio_put(bio); 203bd2d0210STheodore Ts'o 204bd2d0210STheodore Ts'o for (i = 0; i < io_end->num_io_pages; i++) { 205bd2d0210STheodore Ts'o struct page *page = io_end->pages[i]->p_page; 206bd2d0210STheodore Ts'o struct buffer_head *bh, *head; 207bd2d0210STheodore Ts'o int partial_write = 0; 208bd2d0210STheodore Ts'o 209bd2d0210STheodore Ts'o head = page_buffers(page); 210bd2d0210STheodore Ts'o if (error) 211bd2d0210STheodore Ts'o SetPageError(page); 212bd2d0210STheodore Ts'o BUG_ON(!head); 213bd2d0210STheodore Ts'o if (head->b_size == PAGE_CACHE_SIZE) 214bd2d0210STheodore Ts'o clear_buffer_dirty(head); 215bd2d0210STheodore Ts'o else { 216bd2d0210STheodore Ts'o loff_t offset; 217bd2d0210STheodore Ts'o loff_t io_end_offset = io_end->offset + io_end->size; 218bd2d0210STheodore Ts'o 219bd2d0210STheodore Ts'o offset = (sector_t) page->index << PAGE_CACHE_SHIFT; 220bd2d0210STheodore Ts'o bh = head; 221bd2d0210STheodore Ts'o do { 222bd2d0210STheodore Ts'o if ((offset >= io_end->offset) && 223bd2d0210STheodore Ts'o (offset+bh->b_size <= io_end_offset)) { 224bd2d0210STheodore Ts'o if (error) 225bd2d0210STheodore Ts'o buffer_io_error(bh); 226bd2d0210STheodore Ts'o 227bd2d0210STheodore Ts'o clear_buffer_dirty(bh); 228bd2d0210STheodore Ts'o } 229bd2d0210STheodore Ts'o if (buffer_delay(bh)) 230bd2d0210STheodore Ts'o partial_write = 1; 231bd2d0210STheodore Ts'o else if (!buffer_mapped(bh)) 232bd2d0210STheodore Ts'o clear_buffer_dirty(bh); 233bd2d0210STheodore Ts'o else if (buffer_dirty(bh)) 234bd2d0210STheodore Ts'o partial_write = 1; 235bd2d0210STheodore Ts'o offset += bh->b_size; 236bd2d0210STheodore Ts'o bh = bh->b_this_page; 237bd2d0210STheodore Ts'o } while (bh != head); 238bd2d0210STheodore Ts'o } 239bd2d0210STheodore Ts'o 240*83668e71STheodore Ts'o put_io_page(io_end->pages[i]); 241bd2d0210STheodore Ts'o 242bd2d0210STheodore Ts'o /* 243bd2d0210STheodore Ts'o * If this is a partial write which happened to make 244bd2d0210STheodore Ts'o * all buffers uptodate then we can optimize away a 245bd2d0210STheodore Ts'o * bogus readpage() for the next read(). Here we 246bd2d0210STheodore Ts'o * 'discover' whether the page went uptodate as a 247bd2d0210STheodore Ts'o * result of this (potentially partial) write. 248bd2d0210STheodore Ts'o */ 249bd2d0210STheodore Ts'o if (!partial_write) 250bd2d0210STheodore Ts'o SetPageUptodate(page); 251bd2d0210STheodore Ts'o } 252bd2d0210STheodore Ts'o io_end->num_io_pages = 0; 253f7ad6d2eSTheodore Ts'o inode = io_end->inode; 254f7ad6d2eSTheodore Ts'o 255f7ad6d2eSTheodore Ts'o if (error) { 256f7ad6d2eSTheodore Ts'o io_end->flag |= EXT4_IO_END_ERROR; 257f7ad6d2eSTheodore Ts'o ext4_warning(inode->i_sb, "I/O error writing to inode %lu " 258f7ad6d2eSTheodore Ts'o "(offset %llu size %ld starting block %llu)", 259f7ad6d2eSTheodore Ts'o inode->i_ino, 260f7ad6d2eSTheodore Ts'o (unsigned long long) io_end->offset, 261f7ad6d2eSTheodore Ts'o (long) io_end->size, 262f7ad6d2eSTheodore Ts'o (unsigned long long) 263f7ad6d2eSTheodore Ts'o bio->bi_sector >> (inode->i_blkbits - 9)); 264f7ad6d2eSTheodore Ts'o } 265bd2d0210STheodore Ts'o 266bd2d0210STheodore Ts'o /* Add the io_end to per-inode completed io list*/ 267bd2d0210STheodore Ts'o spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags); 268bd2d0210STheodore Ts'o list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list); 269bd2d0210STheodore Ts'o spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags); 270bd2d0210STheodore Ts'o 271bd2d0210STheodore Ts'o wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq; 272bd2d0210STheodore Ts'o /* queue the work to convert unwritten extents to written */ 273bd2d0210STheodore Ts'o queue_work(wq, &io_end->work); 274bd2d0210STheodore Ts'o } 275bd2d0210STheodore Ts'o 276bd2d0210STheodore Ts'o void ext4_io_submit(struct ext4_io_submit *io) 277bd2d0210STheodore Ts'o { 278bd2d0210STheodore Ts'o struct bio *bio = io->io_bio; 279bd2d0210STheodore Ts'o 280bd2d0210STheodore Ts'o if (bio) { 281bd2d0210STheodore Ts'o bio_get(io->io_bio); 282bd2d0210STheodore Ts'o submit_bio(io->io_op, io->io_bio); 283bd2d0210STheodore Ts'o BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP)); 284bd2d0210STheodore Ts'o bio_put(io->io_bio); 285bd2d0210STheodore Ts'o } 286bd2d0210STheodore Ts'o io->io_bio = 0; 287bd2d0210STheodore Ts'o io->io_op = 0; 288bd2d0210STheodore Ts'o io->io_end = 0; 289bd2d0210STheodore Ts'o } 290bd2d0210STheodore Ts'o 291bd2d0210STheodore Ts'o static int io_submit_init(struct ext4_io_submit *io, 292bd2d0210STheodore Ts'o struct inode *inode, 293bd2d0210STheodore Ts'o struct writeback_control *wbc, 294bd2d0210STheodore Ts'o struct buffer_head *bh) 295bd2d0210STheodore Ts'o { 296bd2d0210STheodore Ts'o ext4_io_end_t *io_end; 297bd2d0210STheodore Ts'o struct page *page = bh->b_page; 298bd2d0210STheodore Ts'o int nvecs = bio_get_nr_vecs(bh->b_bdev); 299bd2d0210STheodore Ts'o struct bio *bio; 300bd2d0210STheodore Ts'o 301bd2d0210STheodore Ts'o io_end = ext4_init_io_end(inode, GFP_NOFS); 302bd2d0210STheodore Ts'o if (!io_end) 303bd2d0210STheodore Ts'o return -ENOMEM; 304bd2d0210STheodore Ts'o do { 305bd2d0210STheodore Ts'o bio = bio_alloc(GFP_NOIO, nvecs); 306bd2d0210STheodore Ts'o nvecs >>= 1; 307bd2d0210STheodore Ts'o } while (bio == NULL); 308bd2d0210STheodore Ts'o 309bd2d0210STheodore Ts'o bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); 310bd2d0210STheodore Ts'o bio->bi_bdev = bh->b_bdev; 311bd2d0210STheodore Ts'o bio->bi_private = io->io_end = io_end; 312bd2d0210STheodore Ts'o bio->bi_end_io = ext4_end_bio; 313bd2d0210STheodore Ts'o 314bd2d0210STheodore Ts'o io_end->offset = (page->index << PAGE_CACHE_SHIFT) + bh_offset(bh); 315bd2d0210STheodore Ts'o 316bd2d0210STheodore Ts'o io->io_bio = bio; 317bd2d0210STheodore Ts'o io->io_op = (wbc->sync_mode == WB_SYNC_ALL ? 318bd2d0210STheodore Ts'o WRITE_SYNC_PLUG : WRITE); 319bd2d0210STheodore Ts'o io->io_next_block = bh->b_blocknr; 320bd2d0210STheodore Ts'o return 0; 321bd2d0210STheodore Ts'o } 322bd2d0210STheodore Ts'o 323bd2d0210STheodore Ts'o static int io_submit_add_bh(struct ext4_io_submit *io, 324bd2d0210STheodore Ts'o struct ext4_io_page *io_page, 325bd2d0210STheodore Ts'o struct inode *inode, 326bd2d0210STheodore Ts'o struct writeback_control *wbc, 327bd2d0210STheodore Ts'o struct buffer_head *bh) 328bd2d0210STheodore Ts'o { 329bd2d0210STheodore Ts'o ext4_io_end_t *io_end; 330bd2d0210STheodore Ts'o int ret; 331bd2d0210STheodore Ts'o 332bd2d0210STheodore Ts'o if (buffer_new(bh)) { 333bd2d0210STheodore Ts'o clear_buffer_new(bh); 334bd2d0210STheodore Ts'o unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); 335bd2d0210STheodore Ts'o } 336bd2d0210STheodore Ts'o 337bd2d0210STheodore Ts'o if (!buffer_mapped(bh) || buffer_delay(bh)) { 338bd2d0210STheodore Ts'o if (!buffer_mapped(bh)) 339bd2d0210STheodore Ts'o clear_buffer_dirty(bh); 340bd2d0210STheodore Ts'o if (io->io_bio) 341bd2d0210STheodore Ts'o ext4_io_submit(io); 342bd2d0210STheodore Ts'o return 0; 343bd2d0210STheodore Ts'o } 344bd2d0210STheodore Ts'o 345bd2d0210STheodore Ts'o if (io->io_bio && bh->b_blocknr != io->io_next_block) { 346bd2d0210STheodore Ts'o submit_and_retry: 347bd2d0210STheodore Ts'o ext4_io_submit(io); 348bd2d0210STheodore Ts'o } 349bd2d0210STheodore Ts'o if (io->io_bio == NULL) { 350bd2d0210STheodore Ts'o ret = io_submit_init(io, inode, wbc, bh); 351bd2d0210STheodore Ts'o if (ret) 352bd2d0210STheodore Ts'o return ret; 353bd2d0210STheodore Ts'o } 354bd2d0210STheodore Ts'o io_end = io->io_end; 355bd2d0210STheodore Ts'o if ((io_end->num_io_pages >= MAX_IO_PAGES) && 356bd2d0210STheodore Ts'o (io_end->pages[io_end->num_io_pages-1] != io_page)) 357bd2d0210STheodore Ts'o goto submit_and_retry; 358bd2d0210STheodore Ts'o if (buffer_uninit(bh)) 359bd2d0210STheodore Ts'o io->io_end->flag |= EXT4_IO_END_UNWRITTEN; 360bd2d0210STheodore Ts'o io->io_end->size += bh->b_size; 361bd2d0210STheodore Ts'o io->io_next_block++; 362bd2d0210STheodore Ts'o ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh)); 363bd2d0210STheodore Ts'o if (ret != bh->b_size) 364bd2d0210STheodore Ts'o goto submit_and_retry; 365bd2d0210STheodore Ts'o if ((io_end->num_io_pages == 0) || 366bd2d0210STheodore Ts'o (io_end->pages[io_end->num_io_pages-1] != io_page)) { 367bd2d0210STheodore Ts'o io_end->pages[io_end->num_io_pages++] = io_page; 368*83668e71STheodore Ts'o atomic_inc(&io_page->p_count); 369bd2d0210STheodore Ts'o } 370bd2d0210STheodore Ts'o return 0; 371bd2d0210STheodore Ts'o } 372bd2d0210STheodore Ts'o 373bd2d0210STheodore Ts'o int ext4_bio_write_page(struct ext4_io_submit *io, 374bd2d0210STheodore Ts'o struct page *page, 375bd2d0210STheodore Ts'o int len, 376bd2d0210STheodore Ts'o struct writeback_control *wbc) 377bd2d0210STheodore Ts'o { 378bd2d0210STheodore Ts'o struct inode *inode = page->mapping->host; 379bd2d0210STheodore Ts'o unsigned block_start, block_end, blocksize; 380bd2d0210STheodore Ts'o struct ext4_io_page *io_page; 381bd2d0210STheodore Ts'o struct buffer_head *bh, *head; 382bd2d0210STheodore Ts'o int ret = 0; 383bd2d0210STheodore Ts'o 384bd2d0210STheodore Ts'o blocksize = 1 << inode->i_blkbits; 385bd2d0210STheodore Ts'o 386bd2d0210STheodore Ts'o BUG_ON(PageWriteback(page)); 387bd2d0210STheodore Ts'o set_page_writeback(page); 388bd2d0210STheodore Ts'o ClearPageError(page); 389bd2d0210STheodore Ts'o 390bd2d0210STheodore Ts'o io_page = kmem_cache_alloc(io_page_cachep, GFP_NOFS); 391bd2d0210STheodore Ts'o if (!io_page) { 392bd2d0210STheodore Ts'o set_page_dirty(page); 393bd2d0210STheodore Ts'o unlock_page(page); 394bd2d0210STheodore Ts'o return -ENOMEM; 395bd2d0210STheodore Ts'o } 396bd2d0210STheodore Ts'o io_page->p_page = page; 397*83668e71STheodore Ts'o atomic_set(&io_page->p_count, 1); 398bd2d0210STheodore Ts'o get_page(page); 399bd2d0210STheodore Ts'o 400bd2d0210STheodore Ts'o for (bh = head = page_buffers(page), block_start = 0; 401bd2d0210STheodore Ts'o bh != head || !block_start; 402bd2d0210STheodore Ts'o block_start = block_end, bh = bh->b_this_page) { 403bd2d0210STheodore Ts'o block_end = block_start + blocksize; 404bd2d0210STheodore Ts'o if (block_start >= len) { 405bd2d0210STheodore Ts'o clear_buffer_dirty(bh); 406bd2d0210STheodore Ts'o set_buffer_uptodate(bh); 407bd2d0210STheodore Ts'o continue; 408bd2d0210STheodore Ts'o } 409bd2d0210STheodore Ts'o ret = io_submit_add_bh(io, io_page, inode, wbc, bh); 410bd2d0210STheodore Ts'o if (ret) { 411bd2d0210STheodore Ts'o /* 412bd2d0210STheodore Ts'o * We only get here on ENOMEM. Not much else 413bd2d0210STheodore Ts'o * we can do but mark the page as dirty, and 414bd2d0210STheodore Ts'o * better luck next time. 415bd2d0210STheodore Ts'o */ 416bd2d0210STheodore Ts'o set_page_dirty(page); 417bd2d0210STheodore Ts'o break; 418bd2d0210STheodore Ts'o } 419bd2d0210STheodore Ts'o } 420bd2d0210STheodore Ts'o unlock_page(page); 421bd2d0210STheodore Ts'o /* 422bd2d0210STheodore Ts'o * If the page was truncated before we could do the writeback, 423bd2d0210STheodore Ts'o * or we had a memory allocation error while trying to write 424bd2d0210STheodore Ts'o * the first buffer head, we won't have submitted any pages for 425bd2d0210STheodore Ts'o * I/O. In that case we need to make sure we've cleared the 426bd2d0210STheodore Ts'o * PageWriteback bit from the page to prevent the system from 427bd2d0210STheodore Ts'o * wedging later on. 428bd2d0210STheodore Ts'o */ 429*83668e71STheodore Ts'o put_io_page(io_page); 430bd2d0210STheodore Ts'o return ret; 431bd2d0210STheodore Ts'o } 432