1f5166768STheodore Ts'o // SPDX-License-Identifier: LGPL-2.1 2748de673SAkira Fujita /* 3748de673SAkira Fujita * Copyright (c) 2008,2009 NEC Software Tohoku, Ltd. 4748de673SAkira Fujita * Written by Takashi Sato <t-sato@yk.jp.nec.com> 5748de673SAkira Fujita * Akira Fujita <a-fujita@rs.jp.nec.com> 6748de673SAkira Fujita */ 7748de673SAkira Fujita 8748de673SAkira Fujita #include <linux/fs.h> 9748de673SAkira Fujita #include <linux/quotaops.h> 105a0e3ad6STejun Heo #include <linux/slab.h> 11748de673SAkira Fujita #include "ext4_jbd2.h" 12748de673SAkira Fujita #include "ext4.h" 134a092d73STheodore Ts'o #include "ext4_extents.h" 14748de673SAkira Fujita 15e8505970SAkira Fujita /** 16c60990b3STheodore Ts'o * get_ext_path() - Find an extent path for designated logical block number. 17c60990b3STheodore Ts'o * @inode: inode to be searched 18e8505970SAkira Fujita * @lblock: logical block number to find an extent path 19c60990b3STheodore Ts'o * @ppath: pointer to an extent path pointer (for output) 20e8505970SAkira Fujita * 21ed8a1a76STheodore Ts'o * ext4_find_extent wrapper. Return 0 on success, or a negative error value 22e8505970SAkira Fujita * on failure. 23e8505970SAkira Fujita */ 24e8505970SAkira Fujita static inline int 25e8505970SAkira Fujita get_ext_path(struct inode *inode, ext4_lblk_t lblock, 263bdf14b4STheodore Ts'o struct ext4_ext_path **ppath) 27e8505970SAkira Fujita { 280e401101SDmitry Monakhov struct ext4_ext_path *path; 29e8505970SAkira Fujita 30ed8a1a76STheodore Ts'o path = ext4_find_extent(inode, lblock, ppath, EXT4_EX_NOCACHE); 310e401101SDmitry Monakhov if (IS_ERR(path)) 323bdf14b4STheodore Ts'o return PTR_ERR(path); 333bdf14b4STheodore Ts'o if (path[ext_depth(inode)].p_ext == NULL) { 343bdf14b4STheodore Ts'o ext4_ext_drop_refs(path); 353bdf14b4STheodore Ts'o kfree(path); 363bdf14b4STheodore Ts'o *ppath = NULL; 373bdf14b4STheodore Ts'o return -ENODATA; 383bdf14b4STheodore Ts'o } 393bdf14b4STheodore Ts'o *ppath = path; 403bdf14b4STheodore Ts'o return 0; 41e8505970SAkira Fujita } 42748de673SAkira Fujita 43748de673SAkira Fujita /** 44c60990b3STheodore Ts'o * ext4_double_down_write_data_sem() - write lock two inodes's i_data_sem 45c60990b3STheodore Ts'o * @first: inode to be locked 46c60990b3STheodore Ts'o * @second: inode to be locked 47748de673SAkira Fujita * 4803bd8b9bSDmitry Monakhov * Acquire write lock of i_data_sem of the two inodes 49748de673SAkira Fujita */ 50393d1d1dSDr. Tilmann Bubeck void 51393d1d1dSDr. Tilmann Bubeck ext4_double_down_write_data_sem(struct inode *first, struct inode *second) 52748de673SAkira Fujita { 5303bd8b9bSDmitry Monakhov if (first < second) { 54748de673SAkira Fujita down_write(&EXT4_I(first)->i_data_sem); 55daf647d2STheodore Ts'o down_write_nested(&EXT4_I(second)->i_data_sem, I_DATA_SEM_OTHER); 5603bd8b9bSDmitry Monakhov } else { 5703bd8b9bSDmitry Monakhov down_write(&EXT4_I(second)->i_data_sem); 58daf647d2STheodore Ts'o down_write_nested(&EXT4_I(first)->i_data_sem, I_DATA_SEM_OTHER); 5903bd8b9bSDmitry Monakhov 6003bd8b9bSDmitry Monakhov } 61748de673SAkira Fujita } 62748de673SAkira Fujita 63748de673SAkira Fujita /** 64393d1d1dSDr. Tilmann Bubeck * ext4_double_up_write_data_sem - Release two inodes' write lock of i_data_sem 65748de673SAkira Fujita * 66748de673SAkira Fujita * @orig_inode: original inode structure to be released its lock first 67748de673SAkira Fujita * @donor_inode: donor inode structure to be released its lock second 68fc04cb49SAkira Fujita * Release write lock of i_data_sem of two inodes (orig and donor). 69748de673SAkira Fujita */ 70393d1d1dSDr. Tilmann Bubeck void 71393d1d1dSDr. Tilmann Bubeck ext4_double_up_write_data_sem(struct inode *orig_inode, 72393d1d1dSDr. Tilmann Bubeck struct inode *donor_inode) 73748de673SAkira Fujita { 74748de673SAkira Fujita up_write(&EXT4_I(orig_inode)->i_data_sem); 75748de673SAkira Fujita up_write(&EXT4_I(donor_inode)->i_data_sem); 76748de673SAkira Fujita } 77748de673SAkira Fujita 78748de673SAkira Fujita /** 798c854473SDmitry Monakhov * mext_check_coverage - Check that all extents in range has the same type 808c854473SDmitry Monakhov * 818c854473SDmitry Monakhov * @inode: inode in question 828c854473SDmitry Monakhov * @from: block offset of inode 838c854473SDmitry Monakhov * @count: block count to be checked 84556615dcSLukas Czerner * @unwritten: extents expected to be unwritten 858c854473SDmitry Monakhov * @err: pointer to save error value 868c854473SDmitry Monakhov * 878c854473SDmitry Monakhov * Return 1 if all extents in range has expected type, and zero otherwise. 888c854473SDmitry Monakhov */ 898c854473SDmitry Monakhov static int 908c854473SDmitry Monakhov mext_check_coverage(struct inode *inode, ext4_lblk_t from, ext4_lblk_t count, 91556615dcSLukas Czerner int unwritten, int *err) 928c854473SDmitry Monakhov { 938c854473SDmitry Monakhov struct ext4_ext_path *path = NULL; 948c854473SDmitry Monakhov struct ext4_extent *ext; 950e401101SDmitry Monakhov int ret = 0; 968c854473SDmitry Monakhov ext4_lblk_t last = from + count; 978c854473SDmitry Monakhov while (from < last) { 988c854473SDmitry Monakhov *err = get_ext_path(inode, from, &path); 998c854473SDmitry Monakhov if (*err) 1000e401101SDmitry Monakhov goto out; 1018c854473SDmitry Monakhov ext = path[ext_depth(inode)].p_ext; 102556615dcSLukas Czerner if (unwritten != ext4_ext_is_unwritten(ext)) 1030e401101SDmitry Monakhov goto out; 1048c854473SDmitry Monakhov from += ext4_ext_get_actual_len(ext); 1058c854473SDmitry Monakhov ext4_ext_drop_refs(path); 1068c854473SDmitry Monakhov } 1070e401101SDmitry Monakhov ret = 1; 1080e401101SDmitry Monakhov out: 1090e401101SDmitry Monakhov ext4_ext_drop_refs(path); 1100e401101SDmitry Monakhov kfree(path); 1110e401101SDmitry Monakhov return ret; 1128c854473SDmitry Monakhov } 1138c854473SDmitry Monakhov 1148c854473SDmitry Monakhov /** 115bb557488SDmitry Monakhov * mext_page_double_lock - Grab and lock pages on both @inode1 and @inode2 116bb557488SDmitry Monakhov * 117bb557488SDmitry Monakhov * @inode1: the inode structure 118bb557488SDmitry Monakhov * @inode2: the inode structure 11965dd8327SXiaoguang Wang * @index1: page index 12065dd8327SXiaoguang Wang * @index2: page index 121bb557488SDmitry Monakhov * @page: result page vector 122bb557488SDmitry Monakhov * 123bb557488SDmitry Monakhov * Grab two locked pages for inode's by inode order 124bb557488SDmitry Monakhov */ 125bb557488SDmitry Monakhov static int 126bb557488SDmitry Monakhov mext_page_double_lock(struct inode *inode1, struct inode *inode2, 127fcf6b1b7SDmitry Monakhov pgoff_t index1, pgoff_t index2, struct page *page[2]) 128bb557488SDmitry Monakhov { 129bb557488SDmitry Monakhov struct address_space *mapping[2]; 130bb557488SDmitry Monakhov unsigned fl = AOP_FLAG_NOFS; 131bb557488SDmitry Monakhov 132bb557488SDmitry Monakhov BUG_ON(!inode1 || !inode2); 133bb557488SDmitry Monakhov if (inode1 < inode2) { 134bb557488SDmitry Monakhov mapping[0] = inode1->i_mapping; 135bb557488SDmitry Monakhov mapping[1] = inode2->i_mapping; 136bb557488SDmitry Monakhov } else { 13762bbdd99SGustavo A. R. Silva swap(index1, index2); 138bb557488SDmitry Monakhov mapping[0] = inode2->i_mapping; 139bb557488SDmitry Monakhov mapping[1] = inode1->i_mapping; 140bb557488SDmitry Monakhov } 141bb557488SDmitry Monakhov 142fcf6b1b7SDmitry Monakhov page[0] = grab_cache_page_write_begin(mapping[0], index1, fl); 143bb557488SDmitry Monakhov if (!page[0]) 144bb557488SDmitry Monakhov return -ENOMEM; 145bb557488SDmitry Monakhov 146fcf6b1b7SDmitry Monakhov page[1] = grab_cache_page_write_begin(mapping[1], index2, fl); 147bb557488SDmitry Monakhov if (!page[1]) { 148bb557488SDmitry Monakhov unlock_page(page[0]); 14909cbfeafSKirill A. Shutemov put_page(page[0]); 150bb557488SDmitry Monakhov return -ENOMEM; 151bb557488SDmitry Monakhov } 1527e8b12c6SDmitry Monakhov /* 1537e8b12c6SDmitry Monakhov * grab_cache_page_write_begin() may not wait on page's writeback if 1547e8b12c6SDmitry Monakhov * BDI not demand that. But it is reasonable to be very conservative 1557e8b12c6SDmitry Monakhov * here and explicitly wait on page's writeback 1567e8b12c6SDmitry Monakhov */ 1577e8b12c6SDmitry Monakhov wait_on_page_writeback(page[0]); 1587e8b12c6SDmitry Monakhov wait_on_page_writeback(page[1]); 159bf865467SFabian Frederick if (inode1 > inode2) 160bf865467SFabian Frederick swap(page[0], page[1]); 161bf865467SFabian Frederick 162bb557488SDmitry Monakhov return 0; 163bb557488SDmitry Monakhov } 164bb557488SDmitry Monakhov 165bb557488SDmitry Monakhov /* Force page buffers uptodate w/o dropping page's lock */ 166bb557488SDmitry Monakhov static int 167bb557488SDmitry Monakhov mext_page_mkuptodate(struct page *page, unsigned from, unsigned to) 168bb557488SDmitry Monakhov { 169bb557488SDmitry Monakhov struct inode *inode = page->mapping->host; 170bb557488SDmitry Monakhov sector_t block; 171bb557488SDmitry Monakhov struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE]; 172bb557488SDmitry Monakhov unsigned int blocksize, block_start, block_end; 173bb557488SDmitry Monakhov int i, err, nr = 0, partial = 0; 174bb557488SDmitry Monakhov BUG_ON(!PageLocked(page)); 175bb557488SDmitry Monakhov BUG_ON(PageWriteback(page)); 176bb557488SDmitry Monakhov 177bb557488SDmitry Monakhov if (PageUptodate(page)) 178bb557488SDmitry Monakhov return 0; 179bb557488SDmitry Monakhov 18093407472SFabian Frederick blocksize = i_blocksize(inode); 181bb557488SDmitry Monakhov if (!page_has_buffers(page)) 182bb557488SDmitry Monakhov create_empty_buffers(page, blocksize, 0); 183bb557488SDmitry Monakhov 184bb557488SDmitry Monakhov head = page_buffers(page); 18509cbfeafSKirill A. Shutemov block = (sector_t)page->index << (PAGE_SHIFT - inode->i_blkbits); 186bb557488SDmitry Monakhov for (bh = head, block_start = 0; bh != head || !block_start; 187bb557488SDmitry Monakhov block++, block_start = block_end, bh = bh->b_this_page) { 188bb557488SDmitry Monakhov block_end = block_start + blocksize; 189bb557488SDmitry Monakhov if (block_end <= from || block_start >= to) { 190bb557488SDmitry Monakhov if (!buffer_uptodate(bh)) 191bb557488SDmitry Monakhov partial = 1; 192bb557488SDmitry Monakhov continue; 193bb557488SDmitry Monakhov } 194bb557488SDmitry Monakhov if (buffer_uptodate(bh)) 195bb557488SDmitry Monakhov continue; 196bb557488SDmitry Monakhov if (!buffer_mapped(bh)) { 197bb557488SDmitry Monakhov err = ext4_get_block(inode, block, bh, 0); 198bb557488SDmitry Monakhov if (err) { 199bb557488SDmitry Monakhov SetPageError(page); 200bb557488SDmitry Monakhov return err; 201bb557488SDmitry Monakhov } 202bb557488SDmitry Monakhov if (!buffer_mapped(bh)) { 203bb557488SDmitry Monakhov zero_user(page, block_start, blocksize); 204bb557488SDmitry Monakhov set_buffer_uptodate(bh); 205bb557488SDmitry Monakhov continue; 206bb557488SDmitry Monakhov } 207bb557488SDmitry Monakhov } 208bb557488SDmitry Monakhov BUG_ON(nr >= MAX_BUF_PER_PAGE); 209bb557488SDmitry Monakhov arr[nr++] = bh; 210bb557488SDmitry Monakhov } 211bb557488SDmitry Monakhov /* No io required */ 212bb557488SDmitry Monakhov if (!nr) 213bb557488SDmitry Monakhov goto out; 214bb557488SDmitry Monakhov 215bb557488SDmitry Monakhov for (i = 0; i < nr; i++) { 216bb557488SDmitry Monakhov bh = arr[i]; 217bb557488SDmitry Monakhov if (!bh_uptodate_or_lock(bh)) { 218bb557488SDmitry Monakhov err = bh_submit_read(bh); 219bb557488SDmitry Monakhov if (err) 220bb557488SDmitry Monakhov return err; 221bb557488SDmitry Monakhov } 222bb557488SDmitry Monakhov } 223bb557488SDmitry Monakhov out: 224bb557488SDmitry Monakhov if (!partial) 225bb557488SDmitry Monakhov SetPageUptodate(page); 226bb557488SDmitry Monakhov return 0; 227bb557488SDmitry Monakhov } 228bb557488SDmitry Monakhov 229bb557488SDmitry Monakhov /** 230748de673SAkira Fujita * move_extent_per_page - Move extent data per page 231748de673SAkira Fujita * 232748de673SAkira Fujita * @o_filp: file structure of original file 233748de673SAkira Fujita * @donor_inode: donor inode 234748de673SAkira Fujita * @orig_page_offset: page index on original file 23565dd8327SXiaoguang Wang * @donor_page_offset: page index on donor file 236748de673SAkira Fujita * @data_offset_in_page: block index where data swapping starts 237748de673SAkira Fujita * @block_len_in_page: the number of blocks to be swapped 238556615dcSLukas Czerner * @unwritten: orig extent is unwritten or not 239f868a48dSAkira Fujita * @err: pointer to save return value 240748de673SAkira Fujita * 241748de673SAkira Fujita * Save the data in original inode blocks and replace original inode extents 24265dd8327SXiaoguang Wang * with donor inode extents by calling ext4_swap_extents(). 243f868a48dSAkira Fujita * Finally, write out the saved data in new original inode blocks. Return 244f868a48dSAkira Fujita * replaced block count. 245748de673SAkira Fujita */ 246748de673SAkira Fujita static int 24744fc48f7SAkira Fujita move_extent_per_page(struct file *o_filp, struct inode *donor_inode, 248fcf6b1b7SDmitry Monakhov pgoff_t orig_page_offset, pgoff_t donor_page_offset, 249fcf6b1b7SDmitry Monakhov int data_offset_in_page, 250556615dcSLukas Czerner int block_len_in_page, int unwritten, int *err) 251748de673SAkira Fujita { 252496ad9aaSAl Viro struct inode *orig_inode = file_inode(o_filp); 253bb557488SDmitry Monakhov struct page *pagep[2] = {NULL, NULL}; 254748de673SAkira Fujita handle_t *handle; 255fcf6b1b7SDmitry Monakhov ext4_lblk_t orig_blk_offset, donor_blk_offset; 256748de673SAkira Fujita unsigned long blocksize = orig_inode->i_sb->s_blocksize; 257f868a48dSAkira Fujita unsigned int tmp_data_size, data_size, replaced_size; 258bcff2488SEryu Guan int i, err2, jblocks, retries = 0; 259f868a48dSAkira Fujita int replaced_count = 0; 260bb557488SDmitry Monakhov int from = data_offset_in_page << orig_inode->i_blkbits; 26109cbfeafSKirill A. Shutemov int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits; 26288c6b61fSDmitry Monakhov struct super_block *sb = orig_inode->i_sb; 263bcff2488SEryu Guan struct buffer_head *bh = NULL; 264748de673SAkira Fujita 265748de673SAkira Fujita /* 266748de673SAkira Fujita * It needs twice the amount of ordinary journal buffers because 267748de673SAkira Fujita * inode and donor_inode may change each different metadata blocks. 268748de673SAkira Fujita */ 269bb557488SDmitry Monakhov again: 270bb557488SDmitry Monakhov *err = 0; 271748de673SAkira Fujita jblocks = ext4_writepage_trans_blocks(orig_inode) * 2; 2729924a92aSTheodore Ts'o handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks); 273748de673SAkira Fujita if (IS_ERR(handle)) { 274f868a48dSAkira Fujita *err = PTR_ERR(handle); 275f868a48dSAkira Fujita return 0; 276748de673SAkira Fujita } 277748de673SAkira Fujita 278748de673SAkira Fujita orig_blk_offset = orig_page_offset * blocks_per_page + 279748de673SAkira Fujita data_offset_in_page; 280748de673SAkira Fujita 281fcf6b1b7SDmitry Monakhov donor_blk_offset = donor_page_offset * blocks_per_page + 282fcf6b1b7SDmitry Monakhov data_offset_in_page; 283fcf6b1b7SDmitry Monakhov 284f868a48dSAkira Fujita /* Calculate data_size */ 285748de673SAkira Fujita if ((orig_blk_offset + block_len_in_page - 1) == 286748de673SAkira Fujita ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) { 287748de673SAkira Fujita /* Replace the last block */ 288f868a48dSAkira Fujita tmp_data_size = orig_inode->i_size & (blocksize - 1); 289748de673SAkira Fujita /* 290f868a48dSAkira Fujita * If data_size equal zero, it shows data_size is multiples of 291748de673SAkira Fujita * blocksize. So we set appropriate value. 292748de673SAkira Fujita */ 293f868a48dSAkira Fujita if (tmp_data_size == 0) 294f868a48dSAkira Fujita tmp_data_size = blocksize; 295748de673SAkira Fujita 296f868a48dSAkira Fujita data_size = tmp_data_size + 297748de673SAkira Fujita ((block_len_in_page - 1) << orig_inode->i_blkbits); 298f868a48dSAkira Fujita } else 299f868a48dSAkira Fujita data_size = block_len_in_page << orig_inode->i_blkbits; 300748de673SAkira Fujita 301f868a48dSAkira Fujita replaced_size = data_size; 302f868a48dSAkira Fujita 303bb557488SDmitry Monakhov *err = mext_page_double_lock(orig_inode, donor_inode, orig_page_offset, 304fcf6b1b7SDmitry Monakhov donor_page_offset, pagep); 305f868a48dSAkira Fujita if (unlikely(*err < 0)) 306bb557488SDmitry Monakhov goto stop_journal; 3078c854473SDmitry Monakhov /* 308556615dcSLukas Czerner * If orig extent was unwritten it can become initialized 3098c854473SDmitry Monakhov * at any time after i_data_sem was dropped, in order to 3108c854473SDmitry Monakhov * serialize with delalloc we have recheck extent while we 3118c854473SDmitry Monakhov * hold page's lock, if it is still the case data copy is not 3128c854473SDmitry Monakhov * necessary, just swap data blocks between orig and donor. 3138c854473SDmitry Monakhov */ 314556615dcSLukas Czerner if (unwritten) { 315393d1d1dSDr. Tilmann Bubeck ext4_double_down_write_data_sem(orig_inode, donor_inode); 3168c854473SDmitry Monakhov /* If any of extents in range became initialized we have to 3178c854473SDmitry Monakhov * fallback to data copying */ 318556615dcSLukas Czerner unwritten = mext_check_coverage(orig_inode, orig_blk_offset, 3198c854473SDmitry Monakhov block_len_in_page, 1, err); 3208c854473SDmitry Monakhov if (*err) 3218c854473SDmitry Monakhov goto drop_data_sem; 322748de673SAkira Fujita 323fcf6b1b7SDmitry Monakhov unwritten &= mext_check_coverage(donor_inode, donor_blk_offset, 3248c854473SDmitry Monakhov block_len_in_page, 1, err); 3258c854473SDmitry Monakhov if (*err) 3268c854473SDmitry Monakhov goto drop_data_sem; 3278c854473SDmitry Monakhov 328556615dcSLukas Czerner if (!unwritten) { 329393d1d1dSDr. Tilmann Bubeck ext4_double_up_write_data_sem(orig_inode, donor_inode); 3308c854473SDmitry Monakhov goto data_copy; 3318c854473SDmitry Monakhov } 3328c854473SDmitry Monakhov if ((page_has_private(pagep[0]) && 3338c854473SDmitry Monakhov !try_to_release_page(pagep[0], 0)) || 3348c854473SDmitry Monakhov (page_has_private(pagep[1]) && 3358c854473SDmitry Monakhov !try_to_release_page(pagep[1], 0))) { 3368c854473SDmitry Monakhov *err = -EBUSY; 3378c854473SDmitry Monakhov goto drop_data_sem; 3388c854473SDmitry Monakhov } 339fcf6b1b7SDmitry Monakhov replaced_count = ext4_swap_extents(handle, orig_inode, 3408c854473SDmitry Monakhov donor_inode, orig_blk_offset, 341fcf6b1b7SDmitry Monakhov donor_blk_offset, 342fcf6b1b7SDmitry Monakhov block_len_in_page, 1, err); 3438c854473SDmitry Monakhov drop_data_sem: 344393d1d1dSDr. Tilmann Bubeck ext4_double_up_write_data_sem(orig_inode, donor_inode); 3458c854473SDmitry Monakhov goto unlock_pages; 3468c854473SDmitry Monakhov } 3478c854473SDmitry Monakhov data_copy: 348bb557488SDmitry Monakhov *err = mext_page_mkuptodate(pagep[0], from, from + replaced_size); 349bb557488SDmitry Monakhov if (*err) 350bb557488SDmitry Monakhov goto unlock_pages; 351bb557488SDmitry Monakhov 352bb557488SDmitry Monakhov /* At this point all buffers in range are uptodate, old mapping layout 353bb557488SDmitry Monakhov * is no longer required, try to drop it now. */ 354bb557488SDmitry Monakhov if ((page_has_private(pagep[0]) && !try_to_release_page(pagep[0], 0)) || 355bb557488SDmitry Monakhov (page_has_private(pagep[1]) && !try_to_release_page(pagep[1], 0))) { 356bb557488SDmitry Monakhov *err = -EBUSY; 357bb557488SDmitry Monakhov goto unlock_pages; 358748de673SAkira Fujita } 3596e263146SDmitry Monakhov ext4_double_down_write_data_sem(orig_inode, donor_inode); 360fcf6b1b7SDmitry Monakhov replaced_count = ext4_swap_extents(handle, orig_inode, donor_inode, 361fcf6b1b7SDmitry Monakhov orig_blk_offset, donor_blk_offset, 362fcf6b1b7SDmitry Monakhov block_len_in_page, 1, err); 3636e263146SDmitry Monakhov ext4_double_up_write_data_sem(orig_inode, donor_inode); 364bb557488SDmitry Monakhov if (*err) { 365f868a48dSAkira Fujita if (replaced_count) { 366f868a48dSAkira Fujita block_len_in_page = replaced_count; 367f868a48dSAkira Fujita replaced_size = 368f868a48dSAkira Fujita block_len_in_page << orig_inode->i_blkbits; 369ac48b0a1SAkira Fujita } else 370bb557488SDmitry Monakhov goto unlock_pages; 371f868a48dSAkira Fujita } 372bb557488SDmitry Monakhov /* Perform all necessary steps similar write_begin()/write_end() 373bb557488SDmitry Monakhov * but keeping in mind that i_size will not change */ 374bcff2488SEryu Guan if (!page_has_buffers(pagep[0])) 375bcff2488SEryu Guan create_empty_buffers(pagep[0], 1 << orig_inode->i_blkbits, 0); 376bcff2488SEryu Guan bh = page_buffers(pagep[0]); 377bcff2488SEryu Guan for (i = 0; i < data_offset_in_page; i++) 378bcff2488SEryu Guan bh = bh->b_this_page; 379bcff2488SEryu Guan for (i = 0; i < block_len_in_page; i++) { 380bcff2488SEryu Guan *err = ext4_get_block(orig_inode, orig_blk_offset + i, bh, 0); 381bcff2488SEryu Guan if (*err < 0) 382bcff2488SEryu Guan break; 3836ffe77baSEryu Guan bh = bh->b_this_page; 384bcff2488SEryu Guan } 385bb557488SDmitry Monakhov if (!*err) 386bb557488SDmitry Monakhov *err = block_commit_write(pagep[0], from, from + replaced_size); 387fc04cb49SAkira Fujita 388bb557488SDmitry Monakhov if (unlikely(*err < 0)) 389bb557488SDmitry Monakhov goto repair_branches; 390748de673SAkira Fujita 391bb557488SDmitry Monakhov /* Even in case of data=writeback it is reasonable to pin 392bb557488SDmitry Monakhov * inode to transaction, to prevent unexpected data loss */ 39373131fbbSRoss Zwisler *err = ext4_jbd2_inode_add_write(handle, orig_inode, 39473131fbbSRoss Zwisler (loff_t)orig_page_offset << PAGE_SHIFT, replaced_size); 395748de673SAkira Fujita 396bb557488SDmitry Monakhov unlock_pages: 397bb557488SDmitry Monakhov unlock_page(pagep[0]); 39809cbfeafSKirill A. Shutemov put_page(pagep[0]); 399bb557488SDmitry Monakhov unlock_page(pagep[1]); 40009cbfeafSKirill A. Shutemov put_page(pagep[1]); 401bb557488SDmitry Monakhov stop_journal: 40291cc219aSPeng Tao ext4_journal_stop(handle); 40388c6b61fSDmitry Monakhov if (*err == -ENOSPC && 40488c6b61fSDmitry Monakhov ext4_should_retry_alloc(sb, &retries)) 40588c6b61fSDmitry Monakhov goto again; 406bb557488SDmitry Monakhov /* Buffer was busy because probably is pinned to journal transaction, 407bb557488SDmitry Monakhov * force transaction commit may help to free it. */ 40888c6b61fSDmitry Monakhov if (*err == -EBUSY && retries++ < 4 && EXT4_SB(sb)->s_journal && 40988c6b61fSDmitry Monakhov jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal)) 410bb557488SDmitry Monakhov goto again; 411f868a48dSAkira Fujita return replaced_count; 412bb557488SDmitry Monakhov 413bb557488SDmitry Monakhov repair_branches: 414bb557488SDmitry Monakhov /* 415bb557488SDmitry Monakhov * This should never ever happen! 416bb557488SDmitry Monakhov * Extents are swapped already, but we are not able to copy data. 417bb557488SDmitry Monakhov * Try to swap extents to it's original places 418bb557488SDmitry Monakhov */ 419393d1d1dSDr. Tilmann Bubeck ext4_double_down_write_data_sem(orig_inode, donor_inode); 420fcf6b1b7SDmitry Monakhov replaced_count = ext4_swap_extents(handle, donor_inode, orig_inode, 421fcf6b1b7SDmitry Monakhov orig_blk_offset, donor_blk_offset, 422fcf6b1b7SDmitry Monakhov block_len_in_page, 0, &err2); 423393d1d1dSDr. Tilmann Bubeck ext4_double_up_write_data_sem(orig_inode, donor_inode); 424bb557488SDmitry Monakhov if (replaced_count != block_len_in_page) { 425bb557488SDmitry Monakhov EXT4_ERROR_INODE_BLOCK(orig_inode, (sector_t)(orig_blk_offset), 426bb557488SDmitry Monakhov "Unable to copy data block," 427bb557488SDmitry Monakhov " data will be lost."); 428bb557488SDmitry Monakhov *err = -EIO; 429bb557488SDmitry Monakhov } 430bb557488SDmitry Monakhov replaced_count = 0; 431bb557488SDmitry Monakhov goto unlock_pages; 432748de673SAkira Fujita } 433748de673SAkira Fujita 434748de673SAkira Fujita /** 435c437b273SAkira Fujita * mext_check_arguments - Check whether move extent can be done 436748de673SAkira Fujita * 437748de673SAkira Fujita * @orig_inode: original inode 438748de673SAkira Fujita * @donor_inode: donor inode 439748de673SAkira Fujita * @orig_start: logical start offset in block for orig 440748de673SAkira Fujita * @donor_start: logical start offset in block for donor 441748de673SAkira Fujita * @len: the number of blocks to be moved 442748de673SAkira Fujita * 443748de673SAkira Fujita * Check the arguments of ext4_move_extents() whether the files can be 444748de673SAkira Fujita * exchanged with each other. 445748de673SAkira Fujita * Return 0 on success, or a negative error value on failure. 446748de673SAkira Fujita */ 447748de673SAkira Fujita static int 448748de673SAkira Fujita mext_check_arguments(struct inode *orig_inode, 449748de673SAkira Fujita struct inode *donor_inode, __u64 orig_start, 450446aaa6eSKazuya Mio __u64 donor_start, __u64 *len) 451748de673SAkira Fujita { 452fcf6b1b7SDmitry Monakhov __u64 orig_eof, donor_eof; 45370d5d3dcSAkira Fujita unsigned int blkbits = orig_inode->i_blkbits; 45470d5d3dcSAkira Fujita unsigned int blocksize = 1 << blkbits; 45570d5d3dcSAkira Fujita 456fcf6b1b7SDmitry Monakhov orig_eof = (i_size_read(orig_inode) + blocksize - 1) >> blkbits; 457fcf6b1b7SDmitry Monakhov donor_eof = (i_size_read(donor_inode) + blocksize - 1) >> blkbits; 458fcf6b1b7SDmitry Monakhov 459fcf6b1b7SDmitry Monakhov 4604a58579bSAkira Fujita if (donor_inode->i_mode & (S_ISUID|S_ISGID)) { 4614a58579bSAkira Fujita ext4_debug("ext4 move extent: suid or sgid is set" 4624a58579bSAkira Fujita " to donor file [ino:orig %lu, donor %lu]\n", 4634a58579bSAkira Fujita orig_inode->i_ino, donor_inode->i_ino); 4644a58579bSAkira Fujita return -EINVAL; 4654a58579bSAkira Fujita } 4664a58579bSAkira Fujita 4671f5a81e4STheodore Ts'o if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode)) 4681f5a81e4STheodore Ts'o return -EPERM; 4691f5a81e4STheodore Ts'o 470748de673SAkira Fujita /* Ext4 move extent does not support swapfile */ 471748de673SAkira Fujita if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) { 472748de673SAkira Fujita ext4_debug("ext4 move extent: The argument files should " 473748de673SAkira Fujita "not be swapfile [ino:orig %lu, donor %lu]\n", 474748de673SAkira Fujita orig_inode->i_ino, donor_inode->i_ino); 475fcf6b1b7SDmitry Monakhov return -EBUSY; 476748de673SAkira Fujita } 477748de673SAkira Fujita 47802749a4cSTahsin Erdogan if (ext4_is_quota_file(orig_inode) && ext4_is_quota_file(donor_inode)) { 479daf647d2STheodore Ts'o ext4_debug("ext4 move extent: The argument files should " 480daf647d2STheodore Ts'o "not be quota files [ino:orig %lu, donor %lu]\n", 481daf647d2STheodore Ts'o orig_inode->i_ino, donor_inode->i_ino); 482daf647d2STheodore Ts'o return -EBUSY; 483daf647d2STheodore Ts'o } 484daf647d2STheodore Ts'o 485748de673SAkira Fujita /* Ext4 move extent supports only extent based file */ 48612e9b892SDmitry Monakhov if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) { 487748de673SAkira Fujita ext4_debug("ext4 move extent: orig file is not extents " 488748de673SAkira Fujita "based file [ino:orig %lu]\n", orig_inode->i_ino); 489748de673SAkira Fujita return -EOPNOTSUPP; 49012e9b892SDmitry Monakhov } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) { 491748de673SAkira Fujita ext4_debug("ext4 move extent: donor file is not extents " 492748de673SAkira Fujita "based file [ino:donor %lu]\n", donor_inode->i_ino); 493748de673SAkira Fujita return -EOPNOTSUPP; 494748de673SAkira Fujita } 495748de673SAkira Fujita 496748de673SAkira Fujita if ((!orig_inode->i_size) || (!donor_inode->i_size)) { 497748de673SAkira Fujita ext4_debug("ext4 move extent: File size is 0 byte\n"); 498748de673SAkira Fujita return -EINVAL; 499748de673SAkira Fujita } 500748de673SAkira Fujita 501748de673SAkira Fujita /* Start offset should be same */ 502fcf6b1b7SDmitry Monakhov if ((orig_start & ~(PAGE_MASK >> orig_inode->i_blkbits)) != 503fcf6b1b7SDmitry Monakhov (donor_start & ~(PAGE_MASK >> orig_inode->i_blkbits))) { 504748de673SAkira Fujita ext4_debug("ext4 move extent: orig and donor's start " 505d67d64f4STheodore Ts'o "offsets are not aligned [ino:orig %lu, donor %lu]\n", 506748de673SAkira Fujita orig_inode->i_ino, donor_inode->i_ino); 507748de673SAkira Fujita return -EINVAL; 508748de673SAkira Fujita } 509748de673SAkira Fujita 510f17722f9SLukas Czerner if ((orig_start >= EXT_MAX_BLOCKS) || 511fcf6b1b7SDmitry Monakhov (donor_start >= EXT_MAX_BLOCKS) || 512f17722f9SLukas Czerner (*len > EXT_MAX_BLOCKS) || 513fcf6b1b7SDmitry Monakhov (donor_start + *len >= EXT_MAX_BLOCKS) || 514f17722f9SLukas Czerner (orig_start + *len >= EXT_MAX_BLOCKS)) { 5150a80e986SEric Sandeen ext4_debug("ext4 move extent: Can't handle over [%u] blocks " 516f17722f9SLukas Czerner "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS, 517748de673SAkira Fujita orig_inode->i_ino, donor_inode->i_ino); 518748de673SAkira Fujita return -EINVAL; 519748de673SAkira Fujita } 520f18b2b83STheodore Ts'o if (orig_eof <= orig_start) 521f18b2b83STheodore Ts'o *len = 0; 522f18b2b83STheodore Ts'o else if (orig_eof < orig_start + *len - 1) 523fcf6b1b7SDmitry Monakhov *len = orig_eof - orig_start; 524f18b2b83STheodore Ts'o if (donor_eof <= donor_start) 525f18b2b83STheodore Ts'o *len = 0; 526f18b2b83STheodore Ts'o else if (donor_eof < donor_start + *len - 1) 527fcf6b1b7SDmitry Monakhov *len = donor_eof - donor_start; 528748de673SAkira Fujita if (!*len) { 52992c28159SAkira Fujita ext4_debug("ext4 move extent: len should not be 0 " 530748de673SAkira Fujita "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino, 531748de673SAkira Fujita donor_inode->i_ino); 532748de673SAkira Fujita return -EINVAL; 533748de673SAkira Fujita } 534748de673SAkira Fujita 535748de673SAkira Fujita return 0; 536748de673SAkira Fujita } 537748de673SAkira Fujita 538748de673SAkira Fujita /** 539748de673SAkira Fujita * ext4_move_extents - Exchange the specified range of a file 540748de673SAkira Fujita * 541748de673SAkira Fujita * @o_filp: file structure of the original file 542748de673SAkira Fujita * @d_filp: file structure of the donor file 54365dd8327SXiaoguang Wang * @orig_blk: start offset in block for orig 54465dd8327SXiaoguang Wang * @donor_blk: start offset in block for donor 545748de673SAkira Fujita * @len: the number of blocks to be moved 546748de673SAkira Fujita * @moved_len: moved block length 547748de673SAkira Fujita * 548748de673SAkira Fujita * This function returns 0 and moved block length is set in moved_len 549748de673SAkira Fujita * if succeed, otherwise returns error value. 550748de673SAkira Fujita * 551748de673SAkira Fujita */ 552748de673SAkira Fujita int 553fcf6b1b7SDmitry Monakhov ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk, 554fcf6b1b7SDmitry Monakhov __u64 donor_blk, __u64 len, __u64 *moved_len) 555748de673SAkira Fujita { 556496ad9aaSAl Viro struct inode *orig_inode = file_inode(o_filp); 557496ad9aaSAl Viro struct inode *donor_inode = file_inode(d_filp); 558fcf6b1b7SDmitry Monakhov struct ext4_ext_path *path = NULL; 55909cbfeafSKirill A. Shutemov int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits; 560fcf6b1b7SDmitry Monakhov ext4_lblk_t o_end, o_start = orig_blk; 561fcf6b1b7SDmitry Monakhov ext4_lblk_t d_start = donor_blk; 562fcf6b1b7SDmitry Monakhov int ret; 563748de673SAkira Fujita 56403bd8b9bSDmitry Monakhov if (orig_inode->i_sb != donor_inode->i_sb) { 56503bd8b9bSDmitry Monakhov ext4_debug("ext4 move extent: The argument files " 56603bd8b9bSDmitry Monakhov "should be in same FS [ino:orig %lu, donor %lu]\n", 56703bd8b9bSDmitry Monakhov orig_inode->i_ino, donor_inode->i_ino); 56803bd8b9bSDmitry Monakhov return -EINVAL; 56903bd8b9bSDmitry Monakhov } 57003bd8b9bSDmitry Monakhov 57103bd8b9bSDmitry Monakhov /* orig and donor should be different inodes */ 57203bd8b9bSDmitry Monakhov if (orig_inode == donor_inode) { 573f3ce8064STheodore Ts'o ext4_debug("ext4 move extent: The argument files should not " 57403bd8b9bSDmitry Monakhov "be same inode [ino:orig %lu, donor %lu]\n", 575f3ce8064STheodore Ts'o orig_inode->i_ino, donor_inode->i_ino); 576f3ce8064STheodore Ts'o return -EINVAL; 577f3ce8064STheodore Ts'o } 578f3ce8064STheodore Ts'o 5797247c0caSAkira Fujita /* Regular file check */ 5807247c0caSAkira Fujita if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) { 5817247c0caSAkira Fujita ext4_debug("ext4 move extent: The argument files should be " 5827247c0caSAkira Fujita "regular file [ino:orig %lu, donor %lu]\n", 5837247c0caSAkira Fujita orig_inode->i_ino, donor_inode->i_ino); 5847247c0caSAkira Fujita return -EINVAL; 5857247c0caSAkira Fujita } 58604e22412SEric Whitney 58704e22412SEric Whitney /* TODO: it's not obvious how to swap blocks for inodes with full 58804e22412SEric Whitney journaling enabled */ 589f066055aSDmitry Monakhov if (ext4_should_journal_data(orig_inode) || 590f066055aSDmitry Monakhov ext4_should_journal_data(donor_inode)) { 59104e22412SEric Whitney ext4_msg(orig_inode->i_sb, KERN_ERR, 59204e22412SEric Whitney "Online defrag not supported with data journaling"); 59304e22412SEric Whitney return -EOPNOTSUPP; 594f066055aSDmitry Monakhov } 59504e22412SEric Whitney 596592ddec7SChandan Rajendra if (IS_ENCRYPTED(orig_inode) || IS_ENCRYPTED(donor_inode)) { 59714fbd4aaSEric Whitney ext4_msg(orig_inode->i_sb, KERN_ERR, 59814fbd4aaSEric Whitney "Online defrag not supported for encrypted files"); 59914fbd4aaSEric Whitney return -EOPNOTSUPP; 60014fbd4aaSEric Whitney } 60114fbd4aaSEric Whitney 602fc04cb49SAkira Fujita /* Protect orig and donor inodes against a truncate */ 603375e289eSJ. Bruce Fields lock_two_nondirectories(orig_inode, donor_inode); 604748de673SAkira Fujita 60517335dccSDmitry Monakhov /* Wait for all existing dio workers */ 60617335dccSDmitry Monakhov inode_dio_wait(orig_inode); 60717335dccSDmitry Monakhov inode_dio_wait(donor_inode); 60817335dccSDmitry Monakhov 609fc04cb49SAkira Fujita /* Protect extent tree against block allocations via delalloc */ 610393d1d1dSDr. Tilmann Bubeck ext4_double_down_write_data_sem(orig_inode, donor_inode); 611748de673SAkira Fujita /* Check the filesystem environment whether move_extent can be done */ 612fcf6b1b7SDmitry Monakhov ret = mext_check_arguments(orig_inode, donor_inode, orig_blk, 613fcf6b1b7SDmitry Monakhov donor_blk, &len); 61403bd8b9bSDmitry Monakhov if (ret) 615347fa6f1SAkira Fujita goto out; 616fcf6b1b7SDmitry Monakhov o_end = o_start + len; 617748de673SAkira Fujita 618fcf6b1b7SDmitry Monakhov while (o_start < o_end) { 619fcf6b1b7SDmitry Monakhov struct ext4_extent *ex; 620fcf6b1b7SDmitry Monakhov ext4_lblk_t cur_blk, next_blk; 621fcf6b1b7SDmitry Monakhov pgoff_t orig_page_index, donor_page_index; 622fcf6b1b7SDmitry Monakhov int offset_in_page; 623fcf6b1b7SDmitry Monakhov int unwritten, cur_len; 624748de673SAkira Fujita 625fcf6b1b7SDmitry Monakhov ret = get_ext_path(orig_inode, o_start, &path); 62603bd8b9bSDmitry Monakhov if (ret) 627347fa6f1SAkira Fujita goto out; 628fcf6b1b7SDmitry Monakhov ex = path[path->p_depth].p_ext; 629fcf6b1b7SDmitry Monakhov next_blk = ext4_ext_next_allocated_block(path); 630fcf6b1b7SDmitry Monakhov cur_blk = le32_to_cpu(ex->ee_block); 631fcf6b1b7SDmitry Monakhov cur_len = ext4_ext_get_actual_len(ex); 632fcf6b1b7SDmitry Monakhov /* Check hole before the start pos */ 633fcf6b1b7SDmitry Monakhov if (cur_blk + cur_len - 1 < o_start) { 634fcf6b1b7SDmitry Monakhov if (next_blk == EXT_MAX_BLOCKS) { 635fcf6b1b7SDmitry Monakhov o_start = o_end; 636fcf6b1b7SDmitry Monakhov ret = -ENODATA; 637748de673SAkira Fujita goto out; 638748de673SAkira Fujita } 639fcf6b1b7SDmitry Monakhov d_start += next_blk - o_start; 640fcf6b1b7SDmitry Monakhov o_start = next_blk; 6413bdf14b4STheodore Ts'o continue; 642fcf6b1b7SDmitry Monakhov /* Check hole after the start pos */ 643fcf6b1b7SDmitry Monakhov } else if (cur_blk > o_start) { 644fcf6b1b7SDmitry Monakhov /* Skip hole */ 645fcf6b1b7SDmitry Monakhov d_start += cur_blk - o_start; 646fcf6b1b7SDmitry Monakhov o_start = cur_blk; 647fcf6b1b7SDmitry Monakhov /* Extent inside requested range ?*/ 648fcf6b1b7SDmitry Monakhov if (cur_blk >= o_end) 649748de673SAkira Fujita goto out; 650fcf6b1b7SDmitry Monakhov } else { /* in_range(o_start, o_blk, o_len) */ 651fcf6b1b7SDmitry Monakhov cur_len += cur_blk - o_start; 652748de673SAkira Fujita } 653fcf6b1b7SDmitry Monakhov unwritten = ext4_ext_is_unwritten(ex); 654fcf6b1b7SDmitry Monakhov if (o_end - o_start < cur_len) 655fcf6b1b7SDmitry Monakhov cur_len = o_end - o_start; 656748de673SAkira Fujita 65709cbfeafSKirill A. Shutemov orig_page_index = o_start >> (PAGE_SHIFT - 658fcf6b1b7SDmitry Monakhov orig_inode->i_blkbits); 65909cbfeafSKirill A. Shutemov donor_page_index = d_start >> (PAGE_SHIFT - 660fcf6b1b7SDmitry Monakhov donor_inode->i_blkbits); 661fcf6b1b7SDmitry Monakhov offset_in_page = o_start % blocks_per_page; 662fcf6b1b7SDmitry Monakhov if (cur_len > blocks_per_page- offset_in_page) 663fcf6b1b7SDmitry Monakhov cur_len = blocks_per_page - offset_in_page; 664fc04cb49SAkira Fujita /* 665fc04cb49SAkira Fujita * Up semaphore to avoid following problems: 666fc04cb49SAkira Fujita * a. transaction deadlock among ext4_journal_start, 667fc04cb49SAkira Fujita * ->write_begin via pagefault, and jbd2_journal_commit 668fc04cb49SAkira Fujita * b. racing with ->readpage, ->write_begin, and ext4_get_block 669fc04cb49SAkira Fujita * in move_extent_per_page 670fc04cb49SAkira Fujita */ 671393d1d1dSDr. Tilmann Bubeck ext4_double_up_write_data_sem(orig_inode, donor_inode); 672748de673SAkira Fujita /* Swap original branches with new branches */ 673fcf6b1b7SDmitry Monakhov move_extent_per_page(o_filp, donor_inode, 674fcf6b1b7SDmitry Monakhov orig_page_index, donor_page_index, 675fcf6b1b7SDmitry Monakhov offset_in_page, cur_len, 676556615dcSLukas Czerner unwritten, &ret); 677393d1d1dSDr. Tilmann Bubeck ext4_double_down_write_data_sem(orig_inode, donor_inode); 67803bd8b9bSDmitry Monakhov if (ret < 0) 679fc04cb49SAkira Fujita break; 680fcf6b1b7SDmitry Monakhov o_start += cur_len; 681fcf6b1b7SDmitry Monakhov d_start += cur_len; 682748de673SAkira Fujita } 683fcf6b1b7SDmitry Monakhov *moved_len = o_start - orig_blk; 684fcf6b1b7SDmitry Monakhov if (*moved_len > len) 685fcf6b1b7SDmitry Monakhov *moved_len = len; 686fcf6b1b7SDmitry Monakhov 687748de673SAkira Fujita out: 68894d7c16cSAkira Fujita if (*moved_len) { 68994d7c16cSAkira Fujita ext4_discard_preallocations(orig_inode); 69094d7c16cSAkira Fujita ext4_discard_preallocations(donor_inode); 69194d7c16cSAkira Fujita } 69294d7c16cSAkira Fujita 693fcf6b1b7SDmitry Monakhov ext4_ext_drop_refs(path); 694fcf6b1b7SDmitry Monakhov kfree(path); 695393d1d1dSDr. Tilmann Bubeck ext4_double_up_write_data_sem(orig_inode, donor_inode); 696375e289eSJ. Bruce Fields unlock_two_nondirectories(orig_inode, donor_inode); 697748de673SAkira Fujita 69803bd8b9bSDmitry Monakhov return ret; 699748de673SAkira Fujita } 700