1ac27a0ecSDave Kleikamp /* 2617ba13bSMingming Cao * linux/fs/ext4/inode.c 3ac27a0ecSDave Kleikamp * 4ac27a0ecSDave Kleikamp * Copyright (C) 1992, 1993, 1994, 1995 5ac27a0ecSDave Kleikamp * Remy Card (card@masi.ibp.fr) 6ac27a0ecSDave Kleikamp * Laboratoire MASI - Institut Blaise Pascal 7ac27a0ecSDave Kleikamp * Universite Pierre et Marie Curie (Paris VI) 8ac27a0ecSDave Kleikamp * 9ac27a0ecSDave Kleikamp * from 10ac27a0ecSDave Kleikamp * 11ac27a0ecSDave Kleikamp * linux/fs/minix/inode.c 12ac27a0ecSDave Kleikamp * 13ac27a0ecSDave Kleikamp * Copyright (C) 1991, 1992 Linus Torvalds 14ac27a0ecSDave Kleikamp * 15ac27a0ecSDave Kleikamp * 64-bit file support on 64-bit platforms by Jakub Jelinek 16ac27a0ecSDave Kleikamp * (jj@sunsite.ms.mff.cuni.cz) 17ac27a0ecSDave Kleikamp * 18617ba13bSMingming Cao * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000 19ac27a0ecSDave Kleikamp */ 20ac27a0ecSDave Kleikamp 21ac27a0ecSDave Kleikamp #include <linux/fs.h> 22ac27a0ecSDave Kleikamp #include <linux/time.h> 23dab291afSMingming Cao #include <linux/jbd2.h> 24ac27a0ecSDave Kleikamp #include <linux/highuid.h> 25ac27a0ecSDave Kleikamp #include <linux/pagemap.h> 26ac27a0ecSDave Kleikamp #include <linux/quotaops.h> 27ac27a0ecSDave Kleikamp #include <linux/string.h> 28ac27a0ecSDave Kleikamp #include <linux/buffer_head.h> 29ac27a0ecSDave Kleikamp #include <linux/writeback.h> 3064769240SAlex Tomas #include <linux/pagevec.h> 31ac27a0ecSDave Kleikamp #include <linux/mpage.h> 32e83c1397SDuane Griffin #include <linux/namei.h> 33ac27a0ecSDave Kleikamp #include <linux/uio.h> 34ac27a0ecSDave Kleikamp #include <linux/bio.h> 354c0425ffSMingming Cao #include <linux/workqueue.h> 36744692dcSJiaying Zhang #include <linux/kernel.h> 376db26ffcSAndrew Morton #include <linux/printk.h> 385a0e3ad6STejun Heo #include <linux/slab.h> 39a8901d34STheodore Ts'o #include <linux/ratelimit.h> 40a27bb332SKent Overstreet #include <linux/aio.h> 419bffad1eSTheodore Ts'o 423dcf5451SChristoph Hellwig #include "ext4_jbd2.h" 43ac27a0ecSDave Kleikamp #include "xattr.h" 44ac27a0ecSDave Kleikamp #include "acl.h" 459f125d64STheodore Ts'o #include "truncate.h" 46ac27a0ecSDave Kleikamp 479bffad1eSTheodore Ts'o #include <trace/events/ext4.h> 489bffad1eSTheodore Ts'o 49a1d6cc56SAneesh Kumar K.V #define MPAGE_DA_EXTENT_TAIL 0x01 50a1d6cc56SAneesh Kumar K.V 51814525f4SDarrick J. Wong static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, 52814525f4SDarrick J. Wong struct ext4_inode_info *ei) 53814525f4SDarrick J. Wong { 54814525f4SDarrick J. Wong struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 55814525f4SDarrick J. Wong __u16 csum_lo; 56814525f4SDarrick J. Wong __u16 csum_hi = 0; 57814525f4SDarrick J. Wong __u32 csum; 58814525f4SDarrick J. Wong 59171a7f21SDmitry Monakhov csum_lo = le16_to_cpu(raw->i_checksum_lo); 60814525f4SDarrick J. Wong raw->i_checksum_lo = 0; 61814525f4SDarrick J. Wong if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && 62814525f4SDarrick J. Wong EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { 63171a7f21SDmitry Monakhov csum_hi = le16_to_cpu(raw->i_checksum_hi); 64814525f4SDarrick J. Wong raw->i_checksum_hi = 0; 65814525f4SDarrick J. Wong } 66814525f4SDarrick J. Wong 67814525f4SDarrick J. Wong csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, 68814525f4SDarrick J. Wong EXT4_INODE_SIZE(inode->i_sb)); 69814525f4SDarrick J. Wong 70171a7f21SDmitry Monakhov raw->i_checksum_lo = cpu_to_le16(csum_lo); 71814525f4SDarrick J. Wong if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && 72814525f4SDarrick J. Wong EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) 73171a7f21SDmitry Monakhov raw->i_checksum_hi = cpu_to_le16(csum_hi); 74814525f4SDarrick J. Wong 75814525f4SDarrick J. Wong return csum; 76814525f4SDarrick J. Wong } 77814525f4SDarrick J. Wong 78814525f4SDarrick J. Wong static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, 79814525f4SDarrick J. Wong struct ext4_inode_info *ei) 80814525f4SDarrick J. Wong { 81814525f4SDarrick J. Wong __u32 provided, calculated; 82814525f4SDarrick J. Wong 83814525f4SDarrick J. Wong if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != 84814525f4SDarrick J. Wong cpu_to_le32(EXT4_OS_LINUX) || 85814525f4SDarrick J. Wong !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, 86814525f4SDarrick J. Wong EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) 87814525f4SDarrick J. Wong return 1; 88814525f4SDarrick J. Wong 89814525f4SDarrick J. Wong provided = le16_to_cpu(raw->i_checksum_lo); 90814525f4SDarrick J. Wong calculated = ext4_inode_csum(inode, raw, ei); 91814525f4SDarrick J. Wong if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && 92814525f4SDarrick J. Wong EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) 93814525f4SDarrick J. Wong provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; 94814525f4SDarrick J. Wong else 95814525f4SDarrick J. Wong calculated &= 0xFFFF; 96814525f4SDarrick J. Wong 97814525f4SDarrick J. Wong return provided == calculated; 98814525f4SDarrick J. Wong } 99814525f4SDarrick J. Wong 100814525f4SDarrick J. Wong static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, 101814525f4SDarrick J. Wong struct ext4_inode_info *ei) 102814525f4SDarrick J. Wong { 103814525f4SDarrick J. Wong __u32 csum; 104814525f4SDarrick J. Wong 105814525f4SDarrick J. Wong if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != 106814525f4SDarrick J. Wong cpu_to_le32(EXT4_OS_LINUX) || 107814525f4SDarrick J. Wong !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, 108814525f4SDarrick J. Wong EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) 109814525f4SDarrick J. Wong return; 110814525f4SDarrick J. Wong 111814525f4SDarrick J. Wong csum = ext4_inode_csum(inode, raw, ei); 112814525f4SDarrick J. Wong raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF); 113814525f4SDarrick J. Wong if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && 114814525f4SDarrick J. Wong EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) 115814525f4SDarrick J. Wong raw->i_checksum_hi = cpu_to_le16(csum >> 16); 116814525f4SDarrick J. Wong } 117814525f4SDarrick J. Wong 118678aaf48SJan Kara static inline int ext4_begin_ordered_truncate(struct inode *inode, 119678aaf48SJan Kara loff_t new_size) 120678aaf48SJan Kara { 1217ff9c073STheodore Ts'o trace_ext4_begin_ordered_truncate(inode, new_size); 1228aefcd55STheodore Ts'o /* 1238aefcd55STheodore Ts'o * If jinode is zero, then we never opened the file for 1248aefcd55STheodore Ts'o * writing, so there's no need to call 1258aefcd55STheodore Ts'o * jbd2_journal_begin_ordered_truncate() since there's no 1268aefcd55STheodore Ts'o * outstanding writes we need to flush. 1278aefcd55STheodore Ts'o */ 1288aefcd55STheodore Ts'o if (!EXT4_I(inode)->jinode) 1298aefcd55STheodore Ts'o return 0; 1308aefcd55STheodore Ts'o return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), 1318aefcd55STheodore Ts'o EXT4_I(inode)->jinode, 132678aaf48SJan Kara new_size); 133678aaf48SJan Kara } 134678aaf48SJan Kara 135d47992f8SLukas Czerner static void ext4_invalidatepage(struct page *page, unsigned int offset, 136d47992f8SLukas Czerner unsigned int length); 137cb20d518STheodore Ts'o static int __ext4_journalled_writepage(struct page *page, unsigned int len); 138cb20d518STheodore Ts'o static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); 139fffb2739SJan Kara static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, 140fffb2739SJan Kara int pextents); 14164769240SAlex Tomas 142ac27a0ecSDave Kleikamp /* 143ac27a0ecSDave Kleikamp * Test whether an inode is a fast symlink. 144ac27a0ecSDave Kleikamp */ 145617ba13bSMingming Cao static int ext4_inode_is_fast_symlink(struct inode *inode) 146ac27a0ecSDave Kleikamp { 147617ba13bSMingming Cao int ea_blocks = EXT4_I(inode)->i_file_acl ? 148ac27a0ecSDave Kleikamp (inode->i_sb->s_blocksize >> 9) : 0; 149ac27a0ecSDave Kleikamp 150ac27a0ecSDave Kleikamp return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); 151ac27a0ecSDave Kleikamp } 152ac27a0ecSDave Kleikamp 153ac27a0ecSDave Kleikamp /* 154ac27a0ecSDave Kleikamp * Restart the transaction associated with *handle. This does a commit, 155ac27a0ecSDave Kleikamp * so before we call here everything must be consistently dirtied against 156ac27a0ecSDave Kleikamp * this transaction. 157ac27a0ecSDave Kleikamp */ 158487caeefSJan Kara int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, 159487caeefSJan Kara int nblocks) 160ac27a0ecSDave Kleikamp { 161487caeefSJan Kara int ret; 162487caeefSJan Kara 163487caeefSJan Kara /* 164e35fd660STheodore Ts'o * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this 165487caeefSJan Kara * moment, get_block can be called only for blocks inside i_size since 166487caeefSJan Kara * page cache has been already dropped and writes are blocked by 167487caeefSJan Kara * i_mutex. So we can safely drop the i_data_sem here. 168487caeefSJan Kara */ 1690390131bSFrank Mayhar BUG_ON(EXT4_JOURNAL(inode) == NULL); 170ac27a0ecSDave Kleikamp jbd_debug(2, "restarting handle %p\n", handle); 171487caeefSJan Kara up_write(&EXT4_I(inode)->i_data_sem); 1728e8eaabeSAmir Goldstein ret = ext4_journal_restart(handle, nblocks); 173487caeefSJan Kara down_write(&EXT4_I(inode)->i_data_sem); 174fa5d1113SAneesh Kumar K.V ext4_discard_preallocations(inode); 175487caeefSJan Kara 176487caeefSJan Kara return ret; 177ac27a0ecSDave Kleikamp } 178ac27a0ecSDave Kleikamp 179ac27a0ecSDave Kleikamp /* 180ac27a0ecSDave Kleikamp * Called at the last iput() if i_nlink is zero. 181ac27a0ecSDave Kleikamp */ 1820930fcc1SAl Viro void ext4_evict_inode(struct inode *inode) 183ac27a0ecSDave Kleikamp { 184ac27a0ecSDave Kleikamp handle_t *handle; 185bc965ab3STheodore Ts'o int err; 186ac27a0ecSDave Kleikamp 1877ff9c073STheodore Ts'o trace_ext4_evict_inode(inode); 1882581fdc8SJiaying Zhang 1890930fcc1SAl Viro if (inode->i_nlink) { 1902d859db3SJan Kara /* 1912d859db3SJan Kara * When journalling data dirty buffers are tracked only in the 1922d859db3SJan Kara * journal. So although mm thinks everything is clean and 1932d859db3SJan Kara * ready for reaping the inode might still have some pages to 1942d859db3SJan Kara * write in the running transaction or waiting to be 1952d859db3SJan Kara * checkpointed. Thus calling jbd2_journal_invalidatepage() 1962d859db3SJan Kara * (via truncate_inode_pages()) to discard these buffers can 1972d859db3SJan Kara * cause data loss. Also even if we did not discard these 1982d859db3SJan Kara * buffers, we would have no way to find them after the inode 1992d859db3SJan Kara * is reaped and thus user could see stale data if he tries to 2002d859db3SJan Kara * read them before the transaction is checkpointed. So be 2012d859db3SJan Kara * careful and force everything to disk here... We use 2022d859db3SJan Kara * ei->i_datasync_tid to store the newest transaction 2032d859db3SJan Kara * containing inode's data. 2042d859db3SJan Kara * 2052d859db3SJan Kara * Note that directories do not have this problem because they 2062d859db3SJan Kara * don't use page cache. 2072d859db3SJan Kara */ 2082d859db3SJan Kara if (ext4_should_journal_data(inode) && 2092b405bfaSTheodore Ts'o (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) && 2102b405bfaSTheodore Ts'o inode->i_ino != EXT4_JOURNAL_INO) { 2112d859db3SJan Kara journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; 2122d859db3SJan Kara tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; 2132d859db3SJan Kara 214d76a3a77STheodore Ts'o jbd2_complete_transaction(journal, commit_tid); 2152d859db3SJan Kara filemap_write_and_wait(&inode->i_data); 2162d859db3SJan Kara } 2170930fcc1SAl Viro truncate_inode_pages(&inode->i_data, 0); 2185dc23bddSJan Kara 2195dc23bddSJan Kara WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count)); 2200930fcc1SAl Viro goto no_delete; 2210930fcc1SAl Viro } 2220930fcc1SAl Viro 223907f4554SChristoph Hellwig if (!is_bad_inode(inode)) 224871a2931SChristoph Hellwig dquot_initialize(inode); 225907f4554SChristoph Hellwig 226678aaf48SJan Kara if (ext4_should_order_data(inode)) 227678aaf48SJan Kara ext4_begin_ordered_truncate(inode, 0); 228ac27a0ecSDave Kleikamp truncate_inode_pages(&inode->i_data, 0); 229ac27a0ecSDave Kleikamp 2305dc23bddSJan Kara WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count)); 231ac27a0ecSDave Kleikamp if (is_bad_inode(inode)) 232ac27a0ecSDave Kleikamp goto no_delete; 233ac27a0ecSDave Kleikamp 2348e8ad8a5SJan Kara /* 2358e8ad8a5SJan Kara * Protect us against freezing - iput() caller didn't have to have any 2368e8ad8a5SJan Kara * protection against it 2378e8ad8a5SJan Kara */ 2388e8ad8a5SJan Kara sb_start_intwrite(inode->i_sb); 2399924a92aSTheodore Ts'o handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, 2409924a92aSTheodore Ts'o ext4_blocks_for_truncate(inode)+3); 241ac27a0ecSDave Kleikamp if (IS_ERR(handle)) { 242bc965ab3STheodore Ts'o ext4_std_error(inode->i_sb, PTR_ERR(handle)); 243ac27a0ecSDave Kleikamp /* 244ac27a0ecSDave Kleikamp * If we're going to skip the normal cleanup, we still need to 245ac27a0ecSDave Kleikamp * make sure that the in-core orphan linked list is properly 246ac27a0ecSDave Kleikamp * cleaned up. 247ac27a0ecSDave Kleikamp */ 248617ba13bSMingming Cao ext4_orphan_del(NULL, inode); 2498e8ad8a5SJan Kara sb_end_intwrite(inode->i_sb); 250ac27a0ecSDave Kleikamp goto no_delete; 251ac27a0ecSDave Kleikamp } 252ac27a0ecSDave Kleikamp 253ac27a0ecSDave Kleikamp if (IS_SYNC(inode)) 2540390131bSFrank Mayhar ext4_handle_sync(handle); 255ac27a0ecSDave Kleikamp inode->i_size = 0; 256bc965ab3STheodore Ts'o err = ext4_mark_inode_dirty(handle, inode); 257bc965ab3STheodore Ts'o if (err) { 25812062dddSEric Sandeen ext4_warning(inode->i_sb, 259bc965ab3STheodore Ts'o "couldn't mark inode dirty (err %d)", err); 260bc965ab3STheodore Ts'o goto stop_handle; 261bc965ab3STheodore Ts'o } 262ac27a0ecSDave Kleikamp if (inode->i_blocks) 263617ba13bSMingming Cao ext4_truncate(inode); 264bc965ab3STheodore Ts'o 265bc965ab3STheodore Ts'o /* 266bc965ab3STheodore Ts'o * ext4_ext_truncate() doesn't reserve any slop when it 267bc965ab3STheodore Ts'o * restarts journal transactions; therefore there may not be 268bc965ab3STheodore Ts'o * enough credits left in the handle to remove the inode from 269bc965ab3STheodore Ts'o * the orphan list and set the dtime field. 270bc965ab3STheodore Ts'o */ 2710390131bSFrank Mayhar if (!ext4_handle_has_enough_credits(handle, 3)) { 272bc965ab3STheodore Ts'o err = ext4_journal_extend(handle, 3); 273bc965ab3STheodore Ts'o if (err > 0) 274bc965ab3STheodore Ts'o err = ext4_journal_restart(handle, 3); 275bc965ab3STheodore Ts'o if (err != 0) { 27612062dddSEric Sandeen ext4_warning(inode->i_sb, 277bc965ab3STheodore Ts'o "couldn't extend journal (err %d)", err); 278bc965ab3STheodore Ts'o stop_handle: 279bc965ab3STheodore Ts'o ext4_journal_stop(handle); 28045388219STheodore Ts'o ext4_orphan_del(NULL, inode); 2818e8ad8a5SJan Kara sb_end_intwrite(inode->i_sb); 282bc965ab3STheodore Ts'o goto no_delete; 283bc965ab3STheodore Ts'o } 284bc965ab3STheodore Ts'o } 285bc965ab3STheodore Ts'o 286ac27a0ecSDave Kleikamp /* 287617ba13bSMingming Cao * Kill off the orphan record which ext4_truncate created. 288ac27a0ecSDave Kleikamp * AKPM: I think this can be inside the above `if'. 289617ba13bSMingming Cao * Note that ext4_orphan_del() has to be able to cope with the 290ac27a0ecSDave Kleikamp * deletion of a non-existent orphan - this is because we don't 291617ba13bSMingming Cao * know if ext4_truncate() actually created an orphan record. 292ac27a0ecSDave Kleikamp * (Well, we could do this if we need to, but heck - it works) 293ac27a0ecSDave Kleikamp */ 294617ba13bSMingming Cao ext4_orphan_del(handle, inode); 295617ba13bSMingming Cao EXT4_I(inode)->i_dtime = get_seconds(); 296ac27a0ecSDave Kleikamp 297ac27a0ecSDave Kleikamp /* 298ac27a0ecSDave Kleikamp * One subtle ordering requirement: if anything has gone wrong 299ac27a0ecSDave Kleikamp * (transaction abort, IO errors, whatever), then we can still 300ac27a0ecSDave Kleikamp * do these next steps (the fs will already have been marked as 301ac27a0ecSDave Kleikamp * having errors), but we can't free the inode if the mark_dirty 302ac27a0ecSDave Kleikamp * fails. 303ac27a0ecSDave Kleikamp */ 304617ba13bSMingming Cao if (ext4_mark_inode_dirty(handle, inode)) 305ac27a0ecSDave Kleikamp /* If that failed, just do the required in-core inode clear. */ 3060930fcc1SAl Viro ext4_clear_inode(inode); 307ac27a0ecSDave Kleikamp else 308617ba13bSMingming Cao ext4_free_inode(handle, inode); 309617ba13bSMingming Cao ext4_journal_stop(handle); 3108e8ad8a5SJan Kara sb_end_intwrite(inode->i_sb); 311ac27a0ecSDave Kleikamp return; 312ac27a0ecSDave Kleikamp no_delete: 3130930fcc1SAl Viro ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ 314ac27a0ecSDave Kleikamp } 315ac27a0ecSDave Kleikamp 316a9e7f447SDmitry Monakhov #ifdef CONFIG_QUOTA 317a9e7f447SDmitry Monakhov qsize_t *ext4_get_reserved_space(struct inode *inode) 31860e58e0fSMingming Cao { 319a9e7f447SDmitry Monakhov return &EXT4_I(inode)->i_reserved_quota; 32060e58e0fSMingming Cao } 321a9e7f447SDmitry Monakhov #endif 3229d0be502STheodore Ts'o 32312219aeaSAneesh Kumar K.V /* 32412219aeaSAneesh Kumar K.V * Calculate the number of metadata blocks need to reserve 3259d0be502STheodore Ts'o * to allocate a block located at @lblock 32612219aeaSAneesh Kumar K.V */ 32701f49d0bSTheodore Ts'o static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) 32812219aeaSAneesh Kumar K.V { 32912e9b892SDmitry Monakhov if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 3309d0be502STheodore Ts'o return ext4_ext_calc_metadata_amount(inode, lblock); 33112219aeaSAneesh Kumar K.V 3328bb2b247SAmir Goldstein return ext4_ind_calc_metadata_amount(inode, lblock); 33312219aeaSAneesh Kumar K.V } 33412219aeaSAneesh Kumar K.V 3350637c6f4STheodore Ts'o /* 3360637c6f4STheodore Ts'o * Called with i_data_sem down, which is important since we can call 3370637c6f4STheodore Ts'o * ext4_discard_preallocations() from here. 3380637c6f4STheodore Ts'o */ 3395f634d06SAneesh Kumar K.V void ext4_da_update_reserve_space(struct inode *inode, 3405f634d06SAneesh Kumar K.V int used, int quota_claim) 34112219aeaSAneesh Kumar K.V { 34212219aeaSAneesh Kumar K.V struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 3430637c6f4STheodore Ts'o struct ext4_inode_info *ei = EXT4_I(inode); 34412219aeaSAneesh Kumar K.V 3450637c6f4STheodore Ts'o spin_lock(&ei->i_block_reservation_lock); 346d8990240SAditya Kali trace_ext4_da_update_reserve_space(inode, used, quota_claim); 3470637c6f4STheodore Ts'o if (unlikely(used > ei->i_reserved_data_blocks)) { 3488de5c325STheodore Ts'o ext4_warning(inode->i_sb, "%s: ino %lu, used %d " 3491084f252STheodore Ts'o "with only %d reserved data blocks", 3500637c6f4STheodore Ts'o __func__, inode->i_ino, used, 3510637c6f4STheodore Ts'o ei->i_reserved_data_blocks); 3520637c6f4STheodore Ts'o WARN_ON(1); 3530637c6f4STheodore Ts'o used = ei->i_reserved_data_blocks; 3546bc6e63fSAneesh Kumar K.V } 35512219aeaSAneesh Kumar K.V 35697795d2aSBrian Foster if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) { 35701a523ebSTheodore Ts'o ext4_warning(inode->i_sb, "ino %lu, allocated %d " 35801a523ebSTheodore Ts'o "with only %d reserved metadata blocks " 35901a523ebSTheodore Ts'o "(releasing %d blocks with reserved %d data blocks)", 36097795d2aSBrian Foster inode->i_ino, ei->i_allocated_meta_blocks, 36101a523ebSTheodore Ts'o ei->i_reserved_meta_blocks, used, 36201a523ebSTheodore Ts'o ei->i_reserved_data_blocks); 36397795d2aSBrian Foster WARN_ON(1); 36497795d2aSBrian Foster ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks; 36597795d2aSBrian Foster } 36697795d2aSBrian Foster 3670637c6f4STheodore Ts'o /* Update per-inode reservations */ 3680637c6f4STheodore Ts'o ei->i_reserved_data_blocks -= used; 3690637c6f4STheodore Ts'o ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks; 37057042651STheodore Ts'o percpu_counter_sub(&sbi->s_dirtyclusters_counter, 37172b8ab9dSEric Sandeen used + ei->i_allocated_meta_blocks); 3720637c6f4STheodore Ts'o ei->i_allocated_meta_blocks = 0; 3730637c6f4STheodore Ts'o 3740637c6f4STheodore Ts'o if (ei->i_reserved_data_blocks == 0) { 3750637c6f4STheodore Ts'o /* 3760637c6f4STheodore Ts'o * We can release all of the reserved metadata blocks 3770637c6f4STheodore Ts'o * only when we have written all of the delayed 3780637c6f4STheodore Ts'o * allocation blocks. 3790637c6f4STheodore Ts'o */ 38057042651STheodore Ts'o percpu_counter_sub(&sbi->s_dirtyclusters_counter, 38172b8ab9dSEric Sandeen ei->i_reserved_meta_blocks); 382ee5f4d9cSTheodore Ts'o ei->i_reserved_meta_blocks = 0; 3839d0be502STheodore Ts'o ei->i_da_metadata_calc_len = 0; 3840637c6f4STheodore Ts'o } 38512219aeaSAneesh Kumar K.V spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); 38660e58e0fSMingming Cao 38772b8ab9dSEric Sandeen /* Update quota subsystem for data blocks */ 38872b8ab9dSEric Sandeen if (quota_claim) 3897b415bf6SAditya Kali dquot_claim_block(inode, EXT4_C2B(sbi, used)); 39072b8ab9dSEric Sandeen else { 3915f634d06SAneesh Kumar K.V /* 3925f634d06SAneesh Kumar K.V * We did fallocate with an offset that is already delayed 3935f634d06SAneesh Kumar K.V * allocated. So on delayed allocated writeback we should 39472b8ab9dSEric Sandeen * not re-claim the quota for fallocated blocks. 3955f634d06SAneesh Kumar K.V */ 3967b415bf6SAditya Kali dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); 3975f634d06SAneesh Kumar K.V } 398d6014301SAneesh Kumar K.V 399d6014301SAneesh Kumar K.V /* 400d6014301SAneesh Kumar K.V * If we have done all the pending block allocations and if 401d6014301SAneesh Kumar K.V * there aren't any writers on the inode, we can discard the 402d6014301SAneesh Kumar K.V * inode's preallocations. 403d6014301SAneesh Kumar K.V */ 4040637c6f4STheodore Ts'o if ((ei->i_reserved_data_blocks == 0) && 4050637c6f4STheodore Ts'o (atomic_read(&inode->i_writecount) == 0)) 406d6014301SAneesh Kumar K.V ext4_discard_preallocations(inode); 40712219aeaSAneesh Kumar K.V } 40812219aeaSAneesh Kumar K.V 409e29136f8STheodore Ts'o static int __check_block_validity(struct inode *inode, const char *func, 410c398eda0STheodore Ts'o unsigned int line, 41124676da4STheodore Ts'o struct ext4_map_blocks *map) 4126fd058f7STheodore Ts'o { 41324676da4STheodore Ts'o if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, 41424676da4STheodore Ts'o map->m_len)) { 415c398eda0STheodore Ts'o ext4_error_inode(inode, func, line, map->m_pblk, 416c398eda0STheodore Ts'o "lblock %lu mapped to illegal pblock " 41724676da4STheodore Ts'o "(length %d)", (unsigned long) map->m_lblk, 418c398eda0STheodore Ts'o map->m_len); 4196fd058f7STheodore Ts'o return -EIO; 4206fd058f7STheodore Ts'o } 4216fd058f7STheodore Ts'o return 0; 4226fd058f7STheodore Ts'o } 4236fd058f7STheodore Ts'o 424e29136f8STheodore Ts'o #define check_block_validity(inode, map) \ 425c398eda0STheodore Ts'o __check_block_validity((inode), __func__, __LINE__, (map)) 426e29136f8STheodore Ts'o 427921f266bSDmitry Monakhov #ifdef ES_AGGRESSIVE_TEST 428921f266bSDmitry Monakhov static void ext4_map_blocks_es_recheck(handle_t *handle, 429921f266bSDmitry Monakhov struct inode *inode, 430921f266bSDmitry Monakhov struct ext4_map_blocks *es_map, 431921f266bSDmitry Monakhov struct ext4_map_blocks *map, 432921f266bSDmitry Monakhov int flags) 433921f266bSDmitry Monakhov { 434921f266bSDmitry Monakhov int retval; 435921f266bSDmitry Monakhov 436921f266bSDmitry Monakhov map->m_flags = 0; 437921f266bSDmitry Monakhov /* 438921f266bSDmitry Monakhov * There is a race window that the result is not the same. 439921f266bSDmitry Monakhov * e.g. xfstests #223 when dioread_nolock enables. The reason 440921f266bSDmitry Monakhov * is that we lookup a block mapping in extent status tree with 441921f266bSDmitry Monakhov * out taking i_data_sem. So at the time the unwritten extent 442921f266bSDmitry Monakhov * could be converted. 443921f266bSDmitry Monakhov */ 444921f266bSDmitry Monakhov if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) 445921f266bSDmitry Monakhov down_read((&EXT4_I(inode)->i_data_sem)); 446921f266bSDmitry Monakhov if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { 447921f266bSDmitry Monakhov retval = ext4_ext_map_blocks(handle, inode, map, flags & 448921f266bSDmitry Monakhov EXT4_GET_BLOCKS_KEEP_SIZE); 449921f266bSDmitry Monakhov } else { 450921f266bSDmitry Monakhov retval = ext4_ind_map_blocks(handle, inode, map, flags & 451921f266bSDmitry Monakhov EXT4_GET_BLOCKS_KEEP_SIZE); 452921f266bSDmitry Monakhov } 453921f266bSDmitry Monakhov if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) 454921f266bSDmitry Monakhov up_read((&EXT4_I(inode)->i_data_sem)); 455921f266bSDmitry Monakhov /* 456921f266bSDmitry Monakhov * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag 457921f266bSDmitry Monakhov * because it shouldn't be marked in es_map->m_flags. 458921f266bSDmitry Monakhov */ 459921f266bSDmitry Monakhov map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY); 460921f266bSDmitry Monakhov 461921f266bSDmitry Monakhov /* 462921f266bSDmitry Monakhov * We don't check m_len because extent will be collpased in status 463921f266bSDmitry Monakhov * tree. So the m_len might not equal. 464921f266bSDmitry Monakhov */ 465921f266bSDmitry Monakhov if (es_map->m_lblk != map->m_lblk || 466921f266bSDmitry Monakhov es_map->m_flags != map->m_flags || 467921f266bSDmitry Monakhov es_map->m_pblk != map->m_pblk) { 468bdafe42aSTheodore Ts'o printk("ES cache assertion failed for inode: %lu " 469921f266bSDmitry Monakhov "es_cached ex [%d/%d/%llu/%x] != " 470921f266bSDmitry Monakhov "found ex [%d/%d/%llu/%x] retval %d flags %x\n", 471921f266bSDmitry Monakhov inode->i_ino, es_map->m_lblk, es_map->m_len, 472921f266bSDmitry Monakhov es_map->m_pblk, es_map->m_flags, map->m_lblk, 473921f266bSDmitry Monakhov map->m_len, map->m_pblk, map->m_flags, 474921f266bSDmitry Monakhov retval, flags); 475921f266bSDmitry Monakhov } 476921f266bSDmitry Monakhov } 477921f266bSDmitry Monakhov #endif /* ES_AGGRESSIVE_TEST */ 478921f266bSDmitry Monakhov 47955138e0bSTheodore Ts'o /* 480e35fd660STheodore Ts'o * The ext4_map_blocks() function tries to look up the requested blocks, 4812b2d6d01STheodore Ts'o * and returns if the blocks are already mapped. 482f5ab0d1fSMingming Cao * 483f5ab0d1fSMingming Cao * Otherwise it takes the write lock of the i_data_sem and allocate blocks 484f5ab0d1fSMingming Cao * and store the allocated blocks in the result buffer head and mark it 485f5ab0d1fSMingming Cao * mapped. 486f5ab0d1fSMingming Cao * 487e35fd660STheodore Ts'o * If file type is extents based, it will call ext4_ext_map_blocks(), 488e35fd660STheodore Ts'o * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping 489f5ab0d1fSMingming Cao * based files 490f5ab0d1fSMingming Cao * 491f5ab0d1fSMingming Cao * On success, it returns the number of blocks being mapped or allocate. 492f5ab0d1fSMingming Cao * if create==0 and the blocks are pre-allocated and uninitialized block, 493f5ab0d1fSMingming Cao * the result buffer head is unmapped. If the create ==1, it will make sure 494f5ab0d1fSMingming Cao * the buffer head is mapped. 495f5ab0d1fSMingming Cao * 496f5ab0d1fSMingming Cao * It returns 0 if plain look up failed (blocks have not been allocated), in 497df3ab170STao Ma * that case, buffer head is unmapped 498f5ab0d1fSMingming Cao * 499f5ab0d1fSMingming Cao * It returns the error in case of allocation failure. 500f5ab0d1fSMingming Cao */ 501e35fd660STheodore Ts'o int ext4_map_blocks(handle_t *handle, struct inode *inode, 502e35fd660STheodore Ts'o struct ext4_map_blocks *map, int flags) 5030e855ac8SAneesh Kumar K.V { 504d100eef2SZheng Liu struct extent_status es; 5050e855ac8SAneesh Kumar K.V int retval; 506921f266bSDmitry Monakhov #ifdef ES_AGGRESSIVE_TEST 507921f266bSDmitry Monakhov struct ext4_map_blocks orig_map; 508921f266bSDmitry Monakhov 509921f266bSDmitry Monakhov memcpy(&orig_map, map, sizeof(*map)); 510921f266bSDmitry Monakhov #endif 511f5ab0d1fSMingming Cao 512e35fd660STheodore Ts'o map->m_flags = 0; 513e35fd660STheodore Ts'o ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," 514e35fd660STheodore Ts'o "logical block %lu\n", inode->i_ino, flags, map->m_len, 515e35fd660STheodore Ts'o (unsigned long) map->m_lblk); 516d100eef2SZheng Liu 517d100eef2SZheng Liu /* Lookup extent status tree firstly */ 518d100eef2SZheng Liu if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) { 51963b99968STheodore Ts'o ext4_es_lru_add(inode); 520d100eef2SZheng Liu if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { 521d100eef2SZheng Liu map->m_pblk = ext4_es_pblock(&es) + 522d100eef2SZheng Liu map->m_lblk - es.es_lblk; 523d100eef2SZheng Liu map->m_flags |= ext4_es_is_written(&es) ? 524d100eef2SZheng Liu EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; 525d100eef2SZheng Liu retval = es.es_len - (map->m_lblk - es.es_lblk); 526d100eef2SZheng Liu if (retval > map->m_len) 527d100eef2SZheng Liu retval = map->m_len; 528d100eef2SZheng Liu map->m_len = retval; 529d100eef2SZheng Liu } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { 530d100eef2SZheng Liu retval = 0; 531d100eef2SZheng Liu } else { 532d100eef2SZheng Liu BUG_ON(1); 533d100eef2SZheng Liu } 534921f266bSDmitry Monakhov #ifdef ES_AGGRESSIVE_TEST 535921f266bSDmitry Monakhov ext4_map_blocks_es_recheck(handle, inode, map, 536921f266bSDmitry Monakhov &orig_map, flags); 537921f266bSDmitry Monakhov #endif 538d100eef2SZheng Liu goto found; 539d100eef2SZheng Liu } 540d100eef2SZheng Liu 5414df3d265SAneesh Kumar K.V /* 542b920c755STheodore Ts'o * Try to see if we can get the block without requesting a new 543b920c755STheodore Ts'o * file system block. 5444df3d265SAneesh Kumar K.V */ 545729f52c6SZheng Liu if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) 5460e855ac8SAneesh Kumar K.V down_read((&EXT4_I(inode)->i_data_sem)); 54712e9b892SDmitry Monakhov if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { 548a4e5d88bSDmitry Monakhov retval = ext4_ext_map_blocks(handle, inode, map, flags & 549a4e5d88bSDmitry Monakhov EXT4_GET_BLOCKS_KEEP_SIZE); 5504df3d265SAneesh Kumar K.V } else { 551a4e5d88bSDmitry Monakhov retval = ext4_ind_map_blocks(handle, inode, map, flags & 552a4e5d88bSDmitry Monakhov EXT4_GET_BLOCKS_KEEP_SIZE); 5530e855ac8SAneesh Kumar K.V } 554f7fec032SZheng Liu if (retval > 0) { 555f7fec032SZheng Liu int ret; 5563be78c73STheodore Ts'o unsigned int status; 557f7fec032SZheng Liu 55844fb851dSZheng Liu if (unlikely(retval != map->m_len)) { 55944fb851dSZheng Liu ext4_warning(inode->i_sb, 56044fb851dSZheng Liu "ES len assertion failed for inode " 56144fb851dSZheng Liu "%lu: retval %d != map->m_len %d", 56244fb851dSZheng Liu inode->i_ino, retval, map->m_len); 56344fb851dSZheng Liu WARN_ON(1); 564921f266bSDmitry Monakhov } 565921f266bSDmitry Monakhov 566f7fec032SZheng Liu status = map->m_flags & EXT4_MAP_UNWRITTEN ? 567f7fec032SZheng Liu EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; 568f7fec032SZheng Liu if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && 569f7fec032SZheng Liu ext4_find_delalloc_range(inode, map->m_lblk, 570f7fec032SZheng Liu map->m_lblk + map->m_len - 1)) 571f7fec032SZheng Liu status |= EXTENT_STATUS_DELAYED; 572f7fec032SZheng Liu ret = ext4_es_insert_extent(inode, map->m_lblk, 573f7fec032SZheng Liu map->m_len, map->m_pblk, status); 574f7fec032SZheng Liu if (ret < 0) 575f7fec032SZheng Liu retval = ret; 576f7fec032SZheng Liu } 577729f52c6SZheng Liu if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) 5784df3d265SAneesh Kumar K.V up_read((&EXT4_I(inode)->i_data_sem)); 579f5ab0d1fSMingming Cao 580d100eef2SZheng Liu found: 581e35fd660STheodore Ts'o if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { 582f7fec032SZheng Liu int ret = check_block_validity(inode, map); 5836fd058f7STheodore Ts'o if (ret != 0) 5846fd058f7STheodore Ts'o return ret; 5856fd058f7STheodore Ts'o } 5866fd058f7STheodore Ts'o 587f5ab0d1fSMingming Cao /* If it is only a block(s) look up */ 588c2177057STheodore Ts'o if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) 5894df3d265SAneesh Kumar K.V return retval; 5904df3d265SAneesh Kumar K.V 5914df3d265SAneesh Kumar K.V /* 592f5ab0d1fSMingming Cao * Returns if the blocks have already allocated 593f5ab0d1fSMingming Cao * 594f5ab0d1fSMingming Cao * Note that if blocks have been preallocated 595df3ab170STao Ma * ext4_ext_get_block() returns the create = 0 596f5ab0d1fSMingming Cao * with buffer head unmapped. 597f5ab0d1fSMingming Cao */ 598e35fd660STheodore Ts'o if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) 599f5ab0d1fSMingming Cao return retval; 600f5ab0d1fSMingming Cao 601f5ab0d1fSMingming Cao /* 602a25a4e1aSZheng Liu * Here we clear m_flags because after allocating an new extent, 603a25a4e1aSZheng Liu * it will be set again. 6042a8964d6SAneesh Kumar K.V */ 605a25a4e1aSZheng Liu map->m_flags &= ~EXT4_MAP_FLAGS; 6062a8964d6SAneesh Kumar K.V 6072a8964d6SAneesh Kumar K.V /* 608f5ab0d1fSMingming Cao * New blocks allocate and/or writing to uninitialized extent 609f5ab0d1fSMingming Cao * will possibly result in updating i_data, so we take 610f5ab0d1fSMingming Cao * the write lock of i_data_sem, and call get_blocks() 611f5ab0d1fSMingming Cao * with create == 1 flag. 6124df3d265SAneesh Kumar K.V */ 6134df3d265SAneesh Kumar K.V down_write((&EXT4_I(inode)->i_data_sem)); 614d2a17637SMingming Cao 615d2a17637SMingming Cao /* 616d2a17637SMingming Cao * if the caller is from delayed allocation writeout path 617d2a17637SMingming Cao * we have already reserved fs blocks for allocation 618d2a17637SMingming Cao * let the underlying get_block() function know to 619d2a17637SMingming Cao * avoid double accounting 620d2a17637SMingming Cao */ 621c2177057STheodore Ts'o if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) 622f2321097STheodore Ts'o ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); 6234df3d265SAneesh Kumar K.V /* 6244df3d265SAneesh Kumar K.V * We need to check for EXT4 here because migrate 6254df3d265SAneesh Kumar K.V * could have changed the inode type in between 6264df3d265SAneesh Kumar K.V */ 62712e9b892SDmitry Monakhov if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { 628e35fd660STheodore Ts'o retval = ext4_ext_map_blocks(handle, inode, map, flags); 6290e855ac8SAneesh Kumar K.V } else { 630e35fd660STheodore Ts'o retval = ext4_ind_map_blocks(handle, inode, map, flags); 631267e4db9SAneesh Kumar K.V 632e35fd660STheodore Ts'o if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { 633267e4db9SAneesh Kumar K.V /* 634267e4db9SAneesh Kumar K.V * We allocated new blocks which will result in 635267e4db9SAneesh Kumar K.V * i_data's format changing. Force the migrate 636267e4db9SAneesh Kumar K.V * to fail by clearing migrate flags 637267e4db9SAneesh Kumar K.V */ 63819f5fb7aSTheodore Ts'o ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); 639267e4db9SAneesh Kumar K.V } 6402ac3b6e0STheodore Ts'o 641d2a17637SMingming Cao /* 6422ac3b6e0STheodore Ts'o * Update reserved blocks/metadata blocks after successful 6435f634d06SAneesh Kumar K.V * block allocation which had been deferred till now. We don't 6445f634d06SAneesh Kumar K.V * support fallocate for non extent files. So we can update 6455f634d06SAneesh Kumar K.V * reserve space here. 646d2a17637SMingming Cao */ 6475f634d06SAneesh Kumar K.V if ((retval > 0) && 6481296cc85SAneesh Kumar K.V (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) 6495f634d06SAneesh Kumar K.V ext4_da_update_reserve_space(inode, retval, 1); 6505f634d06SAneesh Kumar K.V } 651f7fec032SZheng Liu if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) 652f2321097STheodore Ts'o ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); 653d2a17637SMingming Cao 654f7fec032SZheng Liu if (retval > 0) { 65551865fdaSZheng Liu int ret; 6563be78c73STheodore Ts'o unsigned int status; 657f7fec032SZheng Liu 65844fb851dSZheng Liu if (unlikely(retval != map->m_len)) { 65944fb851dSZheng Liu ext4_warning(inode->i_sb, 66044fb851dSZheng Liu "ES len assertion failed for inode " 66144fb851dSZheng Liu "%lu: retval %d != map->m_len %d", 66244fb851dSZheng Liu inode->i_ino, retval, map->m_len); 66344fb851dSZheng Liu WARN_ON(1); 664921f266bSDmitry Monakhov } 665921f266bSDmitry Monakhov 666adb23551SZheng Liu /* 667adb23551SZheng Liu * If the extent has been zeroed out, we don't need to update 668adb23551SZheng Liu * extent status tree. 669adb23551SZheng Liu */ 670adb23551SZheng Liu if ((flags & EXT4_GET_BLOCKS_PRE_IO) && 671adb23551SZheng Liu ext4_es_lookup_extent(inode, map->m_lblk, &es)) { 672adb23551SZheng Liu if (ext4_es_is_written(&es)) 673adb23551SZheng Liu goto has_zeroout; 674adb23551SZheng Liu } 675f7fec032SZheng Liu status = map->m_flags & EXT4_MAP_UNWRITTEN ? 676f7fec032SZheng Liu EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; 677f7fec032SZheng Liu if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && 678f7fec032SZheng Liu ext4_find_delalloc_range(inode, map->m_lblk, 679f7fec032SZheng Liu map->m_lblk + map->m_len - 1)) 680f7fec032SZheng Liu status |= EXTENT_STATUS_DELAYED; 681f7fec032SZheng Liu ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, 682f7fec032SZheng Liu map->m_pblk, status); 68351865fdaSZheng Liu if (ret < 0) 68451865fdaSZheng Liu retval = ret; 68551865fdaSZheng Liu } 6865356f261SAditya Kali 687adb23551SZheng Liu has_zeroout: 6880e855ac8SAneesh Kumar K.V up_write((&EXT4_I(inode)->i_data_sem)); 689e35fd660STheodore Ts'o if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { 690e29136f8STheodore Ts'o int ret = check_block_validity(inode, map); 6916fd058f7STheodore Ts'o if (ret != 0) 6926fd058f7STheodore Ts'o return ret; 6936fd058f7STheodore Ts'o } 6940e855ac8SAneesh Kumar K.V return retval; 6950e855ac8SAneesh Kumar K.V } 6960e855ac8SAneesh Kumar K.V 697f3bd1f3fSMingming Cao /* Maximum number of blocks we map for direct IO at once. */ 698f3bd1f3fSMingming Cao #define DIO_MAX_BLOCKS 4096 699f3bd1f3fSMingming Cao 7002ed88685STheodore Ts'o static int _ext4_get_block(struct inode *inode, sector_t iblock, 7012ed88685STheodore Ts'o struct buffer_head *bh, int flags) 702ac27a0ecSDave Kleikamp { 7033e4fdaf8SDmitriy Monakhov handle_t *handle = ext4_journal_current_handle(); 7042ed88685STheodore Ts'o struct ext4_map_blocks map; 7057fb5409dSJan Kara int ret = 0, started = 0; 706f3bd1f3fSMingming Cao int dio_credits; 707ac27a0ecSDave Kleikamp 70846c7f254STao Ma if (ext4_has_inline_data(inode)) 70946c7f254STao Ma return -ERANGE; 71046c7f254STao Ma 7112ed88685STheodore Ts'o map.m_lblk = iblock; 7122ed88685STheodore Ts'o map.m_len = bh->b_size >> inode->i_blkbits; 7132ed88685STheodore Ts'o 7148b0f165fSAnatol Pomozov if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) { 7157fb5409dSJan Kara /* Direct IO write... */ 7162ed88685STheodore Ts'o if (map.m_len > DIO_MAX_BLOCKS) 7172ed88685STheodore Ts'o map.m_len = DIO_MAX_BLOCKS; 7182ed88685STheodore Ts'o dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); 7199924a92aSTheodore Ts'o handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, 7209924a92aSTheodore Ts'o dio_credits); 7217fb5409dSJan Kara if (IS_ERR(handle)) { 722ac27a0ecSDave Kleikamp ret = PTR_ERR(handle); 7232ed88685STheodore Ts'o return ret; 7247fb5409dSJan Kara } 7257fb5409dSJan Kara started = 1; 726ac27a0ecSDave Kleikamp } 727ac27a0ecSDave Kleikamp 7282ed88685STheodore Ts'o ret = ext4_map_blocks(handle, inode, &map, flags); 729ac27a0ecSDave Kleikamp if (ret > 0) { 7307b7a8665SChristoph Hellwig ext4_io_end_t *io_end = ext4_inode_aio(inode); 7317b7a8665SChristoph Hellwig 7322ed88685STheodore Ts'o map_bh(bh, inode->i_sb, map.m_pblk); 7332ed88685STheodore Ts'o bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; 7347b7a8665SChristoph Hellwig if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN) 7357b7a8665SChristoph Hellwig set_buffer_defer_completion(bh); 7362ed88685STheodore Ts'o bh->b_size = inode->i_sb->s_blocksize * map.m_len; 737ac27a0ecSDave Kleikamp ret = 0; 738ac27a0ecSDave Kleikamp } 7397fb5409dSJan Kara if (started) 7407fb5409dSJan Kara ext4_journal_stop(handle); 741ac27a0ecSDave Kleikamp return ret; 742ac27a0ecSDave Kleikamp } 743ac27a0ecSDave Kleikamp 7442ed88685STheodore Ts'o int ext4_get_block(struct inode *inode, sector_t iblock, 7452ed88685STheodore Ts'o struct buffer_head *bh, int create) 7462ed88685STheodore Ts'o { 7472ed88685STheodore Ts'o return _ext4_get_block(inode, iblock, bh, 7482ed88685STheodore Ts'o create ? EXT4_GET_BLOCKS_CREATE : 0); 7492ed88685STheodore Ts'o } 7502ed88685STheodore Ts'o 751ac27a0ecSDave Kleikamp /* 752ac27a0ecSDave Kleikamp * `handle' can be NULL if create is zero 753ac27a0ecSDave Kleikamp */ 754617ba13bSMingming Cao struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, 755725d26d3SAneesh Kumar K.V ext4_lblk_t block, int create, int *errp) 756ac27a0ecSDave Kleikamp { 7572ed88685STheodore Ts'o struct ext4_map_blocks map; 7582ed88685STheodore Ts'o struct buffer_head *bh; 759ac27a0ecSDave Kleikamp int fatal = 0, err; 760ac27a0ecSDave Kleikamp 761ac27a0ecSDave Kleikamp J_ASSERT(handle != NULL || create == 0); 762ac27a0ecSDave Kleikamp 7632ed88685STheodore Ts'o map.m_lblk = block; 7642ed88685STheodore Ts'o map.m_len = 1; 7652ed88685STheodore Ts'o err = ext4_map_blocks(handle, inode, &map, 7662ed88685STheodore Ts'o create ? EXT4_GET_BLOCKS_CREATE : 0); 7672ed88685STheodore Ts'o 76890b0a973SCarlos Maiolino /* ensure we send some value back into *errp */ 76990b0a973SCarlos Maiolino *errp = 0; 77090b0a973SCarlos Maiolino 7710f70b406STheodore Ts'o if (create && err == 0) 7720f70b406STheodore Ts'o err = -ENOSPC; /* should never happen */ 7732ed88685STheodore Ts'o if (err < 0) 774ac27a0ecSDave Kleikamp *errp = err; 7752ed88685STheodore Ts'o if (err <= 0) 7762ed88685STheodore Ts'o return NULL; 7772ed88685STheodore Ts'o 7782ed88685STheodore Ts'o bh = sb_getblk(inode->i_sb, map.m_pblk); 779aebf0243SWang Shilong if (unlikely(!bh)) { 780860d21e2STheodore Ts'o *errp = -ENOMEM; 7812ed88685STheodore Ts'o return NULL; 782ac27a0ecSDave Kleikamp } 7832ed88685STheodore Ts'o if (map.m_flags & EXT4_MAP_NEW) { 784ac27a0ecSDave Kleikamp J_ASSERT(create != 0); 785ac39849dSAneesh Kumar K.V J_ASSERT(handle != NULL); 786ac27a0ecSDave Kleikamp 787ac27a0ecSDave Kleikamp /* 788ac27a0ecSDave Kleikamp * Now that we do not always journal data, we should 789ac27a0ecSDave Kleikamp * keep in mind whether this should always journal the 790ac27a0ecSDave Kleikamp * new buffer as metadata. For now, regular file 791617ba13bSMingming Cao * writes use ext4_get_block instead, so it's not a 792ac27a0ecSDave Kleikamp * problem. 793ac27a0ecSDave Kleikamp */ 794ac27a0ecSDave Kleikamp lock_buffer(bh); 795ac27a0ecSDave Kleikamp BUFFER_TRACE(bh, "call get_create_access"); 796617ba13bSMingming Cao fatal = ext4_journal_get_create_access(handle, bh); 797ac27a0ecSDave Kleikamp if (!fatal && !buffer_uptodate(bh)) { 798ac27a0ecSDave Kleikamp memset(bh->b_data, 0, inode->i_sb->s_blocksize); 799ac27a0ecSDave Kleikamp set_buffer_uptodate(bh); 800ac27a0ecSDave Kleikamp } 801ac27a0ecSDave Kleikamp unlock_buffer(bh); 8020390131bSFrank Mayhar BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); 8030390131bSFrank Mayhar err = ext4_handle_dirty_metadata(handle, inode, bh); 804ac27a0ecSDave Kleikamp if (!fatal) 805ac27a0ecSDave Kleikamp fatal = err; 806ac27a0ecSDave Kleikamp } else { 807ac27a0ecSDave Kleikamp BUFFER_TRACE(bh, "not a new buffer"); 808ac27a0ecSDave Kleikamp } 809ac27a0ecSDave Kleikamp if (fatal) { 810ac27a0ecSDave Kleikamp *errp = fatal; 811ac27a0ecSDave Kleikamp brelse(bh); 812ac27a0ecSDave Kleikamp bh = NULL; 813ac27a0ecSDave Kleikamp } 814ac27a0ecSDave Kleikamp return bh; 815ac27a0ecSDave Kleikamp } 816ac27a0ecSDave Kleikamp 817617ba13bSMingming Cao struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, 818725d26d3SAneesh Kumar K.V ext4_lblk_t block, int create, int *err) 819ac27a0ecSDave Kleikamp { 820ac27a0ecSDave Kleikamp struct buffer_head *bh; 821ac27a0ecSDave Kleikamp 822617ba13bSMingming Cao bh = ext4_getblk(handle, inode, block, create, err); 823ac27a0ecSDave Kleikamp if (!bh) 824ac27a0ecSDave Kleikamp return bh; 825ac27a0ecSDave Kleikamp if (buffer_uptodate(bh)) 826ac27a0ecSDave Kleikamp return bh; 82765299a3bSChristoph Hellwig ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh); 828ac27a0ecSDave Kleikamp wait_on_buffer(bh); 829ac27a0ecSDave Kleikamp if (buffer_uptodate(bh)) 830ac27a0ecSDave Kleikamp return bh; 831ac27a0ecSDave Kleikamp put_bh(bh); 832ac27a0ecSDave Kleikamp *err = -EIO; 833ac27a0ecSDave Kleikamp return NULL; 834ac27a0ecSDave Kleikamp } 835ac27a0ecSDave Kleikamp 836f19d5870STao Ma int ext4_walk_page_buffers(handle_t *handle, 837ac27a0ecSDave Kleikamp struct buffer_head *head, 838ac27a0ecSDave Kleikamp unsigned from, 839ac27a0ecSDave Kleikamp unsigned to, 840ac27a0ecSDave Kleikamp int *partial, 841ac27a0ecSDave Kleikamp int (*fn)(handle_t *handle, 842ac27a0ecSDave Kleikamp struct buffer_head *bh)) 843ac27a0ecSDave Kleikamp { 844ac27a0ecSDave Kleikamp struct buffer_head *bh; 845ac27a0ecSDave Kleikamp unsigned block_start, block_end; 846ac27a0ecSDave Kleikamp unsigned blocksize = head->b_size; 847ac27a0ecSDave Kleikamp int err, ret = 0; 848ac27a0ecSDave Kleikamp struct buffer_head *next; 849ac27a0ecSDave Kleikamp 850ac27a0ecSDave Kleikamp for (bh = head, block_start = 0; 851ac27a0ecSDave Kleikamp ret == 0 && (bh != head || !block_start); 852de9a55b8STheodore Ts'o block_start = block_end, bh = next) { 853ac27a0ecSDave Kleikamp next = bh->b_this_page; 854ac27a0ecSDave Kleikamp block_end = block_start + blocksize; 855ac27a0ecSDave Kleikamp if (block_end <= from || block_start >= to) { 856ac27a0ecSDave Kleikamp if (partial && !buffer_uptodate(bh)) 857ac27a0ecSDave Kleikamp *partial = 1; 858ac27a0ecSDave Kleikamp continue; 859ac27a0ecSDave Kleikamp } 860ac27a0ecSDave Kleikamp err = (*fn)(handle, bh); 861ac27a0ecSDave Kleikamp if (!ret) 862ac27a0ecSDave Kleikamp ret = err; 863ac27a0ecSDave Kleikamp } 864ac27a0ecSDave Kleikamp return ret; 865ac27a0ecSDave Kleikamp } 866ac27a0ecSDave Kleikamp 867ac27a0ecSDave Kleikamp /* 868ac27a0ecSDave Kleikamp * To preserve ordering, it is essential that the hole instantiation and 869ac27a0ecSDave Kleikamp * the data write be encapsulated in a single transaction. We cannot 870617ba13bSMingming Cao * close off a transaction and start a new one between the ext4_get_block() 871dab291afSMingming Cao * and the commit_write(). So doing the jbd2_journal_start at the start of 872ac27a0ecSDave Kleikamp * prepare_write() is the right place. 873ac27a0ecSDave Kleikamp * 87436ade451SJan Kara * Also, this function can nest inside ext4_writepage(). In that case, we 87536ade451SJan Kara * *know* that ext4_writepage() has generated enough buffer credits to do the 87636ade451SJan Kara * whole page. So we won't block on the journal in that case, which is good, 87736ade451SJan Kara * because the caller may be PF_MEMALLOC. 878ac27a0ecSDave Kleikamp * 879617ba13bSMingming Cao * By accident, ext4 can be reentered when a transaction is open via 880ac27a0ecSDave Kleikamp * quota file writes. If we were to commit the transaction while thus 881ac27a0ecSDave Kleikamp * reentered, there can be a deadlock - we would be holding a quota 882ac27a0ecSDave Kleikamp * lock, and the commit would never complete if another thread had a 883ac27a0ecSDave Kleikamp * transaction open and was blocking on the quota lock - a ranking 884ac27a0ecSDave Kleikamp * violation. 885ac27a0ecSDave Kleikamp * 886dab291afSMingming Cao * So what we do is to rely on the fact that jbd2_journal_stop/journal_start 887ac27a0ecSDave Kleikamp * will _not_ run commit under these circumstances because handle->h_ref 888ac27a0ecSDave Kleikamp * is elevated. We'll still have enough credits for the tiny quotafile 889ac27a0ecSDave Kleikamp * write. 890ac27a0ecSDave Kleikamp */ 891f19d5870STao Ma int do_journal_get_write_access(handle_t *handle, 892ac27a0ecSDave Kleikamp struct buffer_head *bh) 893ac27a0ecSDave Kleikamp { 89456d35a4cSJan Kara int dirty = buffer_dirty(bh); 89556d35a4cSJan Kara int ret; 89656d35a4cSJan Kara 897ac27a0ecSDave Kleikamp if (!buffer_mapped(bh) || buffer_freed(bh)) 898ac27a0ecSDave Kleikamp return 0; 89956d35a4cSJan Kara /* 900ebdec241SChristoph Hellwig * __block_write_begin() could have dirtied some buffers. Clean 90156d35a4cSJan Kara * the dirty bit as jbd2_journal_get_write_access() could complain 90256d35a4cSJan Kara * otherwise about fs integrity issues. Setting of the dirty bit 903ebdec241SChristoph Hellwig * by __block_write_begin() isn't a real problem here as we clear 90456d35a4cSJan Kara * the bit before releasing a page lock and thus writeback cannot 90556d35a4cSJan Kara * ever write the buffer. 90656d35a4cSJan Kara */ 90756d35a4cSJan Kara if (dirty) 90856d35a4cSJan Kara clear_buffer_dirty(bh); 90956d35a4cSJan Kara ret = ext4_journal_get_write_access(handle, bh); 91056d35a4cSJan Kara if (!ret && dirty) 91156d35a4cSJan Kara ret = ext4_handle_dirty_metadata(handle, NULL, bh); 91256d35a4cSJan Kara return ret; 913ac27a0ecSDave Kleikamp } 914ac27a0ecSDave Kleikamp 9158b0f165fSAnatol Pomozov static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, 9168b0f165fSAnatol Pomozov struct buffer_head *bh_result, int create); 917bfc1af65SNick Piggin static int ext4_write_begin(struct file *file, struct address_space *mapping, 918bfc1af65SNick Piggin loff_t pos, unsigned len, unsigned flags, 919bfc1af65SNick Piggin struct page **pagep, void **fsdata) 920ac27a0ecSDave Kleikamp { 921bfc1af65SNick Piggin struct inode *inode = mapping->host; 9221938a150SAneesh Kumar K.V int ret, needed_blocks; 923ac27a0ecSDave Kleikamp handle_t *handle; 924ac27a0ecSDave Kleikamp int retries = 0; 925bfc1af65SNick Piggin struct page *page; 926bfc1af65SNick Piggin pgoff_t index; 927bfc1af65SNick Piggin unsigned from, to; 928bfc1af65SNick Piggin 9299bffad1eSTheodore Ts'o trace_ext4_write_begin(inode, pos, len, flags); 9301938a150SAneesh Kumar K.V /* 9311938a150SAneesh Kumar K.V * Reserve one block more for addition to orphan list in case 9321938a150SAneesh Kumar K.V * we allocate blocks but write fails for some reason 9331938a150SAneesh Kumar K.V */ 9341938a150SAneesh Kumar K.V needed_blocks = ext4_writepage_trans_blocks(inode) + 1; 935bfc1af65SNick Piggin index = pos >> PAGE_CACHE_SHIFT; 936bfc1af65SNick Piggin from = pos & (PAGE_CACHE_SIZE - 1); 937bfc1af65SNick Piggin to = from + len; 938ac27a0ecSDave Kleikamp 939f19d5870STao Ma if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { 940f19d5870STao Ma ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, 941f19d5870STao Ma flags, pagep); 942f19d5870STao Ma if (ret < 0) 94347564bfbSTheodore Ts'o return ret; 94447564bfbSTheodore Ts'o if (ret == 1) 94547564bfbSTheodore Ts'o return 0; 946f19d5870STao Ma } 947f19d5870STao Ma 94847564bfbSTheodore Ts'o /* 94947564bfbSTheodore Ts'o * grab_cache_page_write_begin() can take a long time if the 95047564bfbSTheodore Ts'o * system is thrashing due to memory pressure, or if the page 95147564bfbSTheodore Ts'o * is being written back. So grab it first before we start 95247564bfbSTheodore Ts'o * the transaction handle. This also allows us to allocate 95347564bfbSTheodore Ts'o * the page (if needed) without using GFP_NOFS. 95447564bfbSTheodore Ts'o */ 95547564bfbSTheodore Ts'o retry_grab: 95654566b2cSNick Piggin page = grab_cache_page_write_begin(mapping, index, flags); 95747564bfbSTheodore Ts'o if (!page) 95847564bfbSTheodore Ts'o return -ENOMEM; 95947564bfbSTheodore Ts'o unlock_page(page); 96047564bfbSTheodore Ts'o 96147564bfbSTheodore Ts'o retry_journal: 9629924a92aSTheodore Ts'o handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); 963ac27a0ecSDave Kleikamp if (IS_ERR(handle)) { 96447564bfbSTheodore Ts'o page_cache_release(page); 96547564bfbSTheodore Ts'o return PTR_ERR(handle); 966cf108bcaSJan Kara } 967f19d5870STao Ma 96847564bfbSTheodore Ts'o lock_page(page); 96947564bfbSTheodore Ts'o if (page->mapping != mapping) { 97047564bfbSTheodore Ts'o /* The page got truncated from under us */ 97147564bfbSTheodore Ts'o unlock_page(page); 97247564bfbSTheodore Ts'o page_cache_release(page); 973cf108bcaSJan Kara ext4_journal_stop(handle); 97447564bfbSTheodore Ts'o goto retry_grab; 975cf108bcaSJan Kara } 9767afe5aa5SDmitry Monakhov /* In case writeback began while the page was unlocked */ 9777afe5aa5SDmitry Monakhov wait_for_stable_page(page); 978cf108bcaSJan Kara 979744692dcSJiaying Zhang if (ext4_should_dioread_nolock(inode)) 9806e1db88dSChristoph Hellwig ret = __block_write_begin(page, pos, len, ext4_get_block_write); 981744692dcSJiaying Zhang else 9826e1db88dSChristoph Hellwig ret = __block_write_begin(page, pos, len, ext4_get_block); 983bfc1af65SNick Piggin 984bfc1af65SNick Piggin if (!ret && ext4_should_journal_data(inode)) { 985f19d5870STao Ma ret = ext4_walk_page_buffers(handle, page_buffers(page), 986f19d5870STao Ma from, to, NULL, 987f19d5870STao Ma do_journal_get_write_access); 988b46be050SAndrey Savochkin } 989bfc1af65SNick Piggin 990bfc1af65SNick Piggin if (ret) { 991bfc1af65SNick Piggin unlock_page(page); 992ae4d5372SAneesh Kumar K.V /* 9936e1db88dSChristoph Hellwig * __block_write_begin may have instantiated a few blocks 994ae4d5372SAneesh Kumar K.V * outside i_size. Trim these off again. Don't need 995ae4d5372SAneesh Kumar K.V * i_size_read because we hold i_mutex. 9961938a150SAneesh Kumar K.V * 9971938a150SAneesh Kumar K.V * Add inode to orphan list in case we crash before 9981938a150SAneesh Kumar K.V * truncate finishes 999ae4d5372SAneesh Kumar K.V */ 1000ffacfa7aSJan Kara if (pos + len > inode->i_size && ext4_can_truncate(inode)) 10011938a150SAneesh Kumar K.V ext4_orphan_add(handle, inode); 10021938a150SAneesh Kumar K.V 10031938a150SAneesh Kumar K.V ext4_journal_stop(handle); 10041938a150SAneesh Kumar K.V if (pos + len > inode->i_size) { 1005b9a4207dSJan Kara ext4_truncate_failed_write(inode); 10061938a150SAneesh Kumar K.V /* 1007ffacfa7aSJan Kara * If truncate failed early the inode might 10081938a150SAneesh Kumar K.V * still be on the orphan list; we need to 10091938a150SAneesh Kumar K.V * make sure the inode is removed from the 10101938a150SAneesh Kumar K.V * orphan list in that case. 10111938a150SAneesh Kumar K.V */ 10121938a150SAneesh Kumar K.V if (inode->i_nlink) 10131938a150SAneesh Kumar K.V ext4_orphan_del(NULL, inode); 10141938a150SAneesh Kumar K.V } 1015bfc1af65SNick Piggin 101647564bfbSTheodore Ts'o if (ret == -ENOSPC && 101747564bfbSTheodore Ts'o ext4_should_retry_alloc(inode->i_sb, &retries)) 101847564bfbSTheodore Ts'o goto retry_journal; 101947564bfbSTheodore Ts'o page_cache_release(page); 102047564bfbSTheodore Ts'o return ret; 102147564bfbSTheodore Ts'o } 102247564bfbSTheodore Ts'o *pagep = page; 1023ac27a0ecSDave Kleikamp return ret; 1024ac27a0ecSDave Kleikamp } 1025ac27a0ecSDave Kleikamp 1026bfc1af65SNick Piggin /* For write_end() in data=journal mode */ 1027bfc1af65SNick Piggin static int write_end_fn(handle_t *handle, struct buffer_head *bh) 1028ac27a0ecSDave Kleikamp { 102913fca323STheodore Ts'o int ret; 1030ac27a0ecSDave Kleikamp if (!buffer_mapped(bh) || buffer_freed(bh)) 1031ac27a0ecSDave Kleikamp return 0; 1032ac27a0ecSDave Kleikamp set_buffer_uptodate(bh); 103313fca323STheodore Ts'o ret = ext4_handle_dirty_metadata(handle, NULL, bh); 103413fca323STheodore Ts'o clear_buffer_meta(bh); 103513fca323STheodore Ts'o clear_buffer_prio(bh); 103613fca323STheodore Ts'o return ret; 1037ac27a0ecSDave Kleikamp } 1038ac27a0ecSDave Kleikamp 1039eed4333fSZheng Liu /* 1040eed4333fSZheng Liu * We need to pick up the new inode size which generic_commit_write gave us 1041eed4333fSZheng Liu * `file' can be NULL - eg, when called from page_symlink(). 1042eed4333fSZheng Liu * 1043eed4333fSZheng Liu * ext4 never places buffers on inode->i_mapping->private_list. metadata 1044eed4333fSZheng Liu * buffers are managed internally. 1045eed4333fSZheng Liu */ 1046eed4333fSZheng Liu static int ext4_write_end(struct file *file, 1047f8514083SAneesh Kumar K.V struct address_space *mapping, 1048f8514083SAneesh Kumar K.V loff_t pos, unsigned len, unsigned copied, 1049f8514083SAneesh Kumar K.V struct page *page, void *fsdata) 1050f8514083SAneesh Kumar K.V { 1051f8514083SAneesh Kumar K.V handle_t *handle = ext4_journal_current_handle(); 1052eed4333fSZheng Liu struct inode *inode = mapping->host; 1053eed4333fSZheng Liu int ret = 0, ret2; 1054eed4333fSZheng Liu int i_size_changed = 0; 1055eed4333fSZheng Liu 1056eed4333fSZheng Liu trace_ext4_write_end(inode, pos, len, copied); 1057eed4333fSZheng Liu if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) { 1058eed4333fSZheng Liu ret = ext4_jbd2_file_inode(handle, inode); 1059eed4333fSZheng Liu if (ret) { 1060eed4333fSZheng Liu unlock_page(page); 1061eed4333fSZheng Liu page_cache_release(page); 1062eed4333fSZheng Liu goto errout; 1063eed4333fSZheng Liu } 1064eed4333fSZheng Liu } 1065f8514083SAneesh Kumar K.V 106642c832deSTheodore Ts'o if (ext4_has_inline_data(inode)) { 106742c832deSTheodore Ts'o ret = ext4_write_inline_data_end(inode, pos, len, 1068f19d5870STao Ma copied, page); 106942c832deSTheodore Ts'o if (ret < 0) 107042c832deSTheodore Ts'o goto errout; 107142c832deSTheodore Ts'o copied = ret; 107242c832deSTheodore Ts'o } else 1073f19d5870STao Ma copied = block_write_end(file, mapping, pos, 1074f19d5870STao Ma len, copied, page, fsdata); 1075f8514083SAneesh Kumar K.V 1076f8514083SAneesh Kumar K.V /* 1077f8514083SAneesh Kumar K.V * No need to use i_size_read() here, the i_size 1078eed4333fSZheng Liu * cannot change under us because we hole i_mutex. 1079f8514083SAneesh Kumar K.V * 1080f8514083SAneesh Kumar K.V * But it's important to update i_size while still holding page lock: 1081f8514083SAneesh Kumar K.V * page writeout could otherwise come in and zero beyond i_size. 1082f8514083SAneesh Kumar K.V */ 1083f8514083SAneesh Kumar K.V if (pos + copied > inode->i_size) { 1084f8514083SAneesh Kumar K.V i_size_write(inode, pos + copied); 1085f8514083SAneesh Kumar K.V i_size_changed = 1; 1086f8514083SAneesh Kumar K.V } 1087f8514083SAneesh Kumar K.V 1088f8514083SAneesh Kumar K.V if (pos + copied > EXT4_I(inode)->i_disksize) { 1089f8514083SAneesh Kumar K.V /* We need to mark inode dirty even if 1090f8514083SAneesh Kumar K.V * new_i_size is less that inode->i_size 1091eed4333fSZheng Liu * but greater than i_disksize. (hint delalloc) 1092f8514083SAneesh Kumar K.V */ 1093f8514083SAneesh Kumar K.V ext4_update_i_disksize(inode, (pos + copied)); 1094f8514083SAneesh Kumar K.V i_size_changed = 1; 1095f8514083SAneesh Kumar K.V } 1096f8514083SAneesh Kumar K.V unlock_page(page); 1097f8514083SAneesh Kumar K.V page_cache_release(page); 1098f8514083SAneesh Kumar K.V 1099f8514083SAneesh Kumar K.V /* 1100f8514083SAneesh Kumar K.V * Don't mark the inode dirty under page lock. First, it unnecessarily 1101f8514083SAneesh Kumar K.V * makes the holding time of page lock longer. Second, it forces lock 1102f8514083SAneesh Kumar K.V * ordering of page lock and transaction start for journaling 1103f8514083SAneesh Kumar K.V * filesystems. 1104f8514083SAneesh Kumar K.V */ 1105f8514083SAneesh Kumar K.V if (i_size_changed) 1106f8514083SAneesh Kumar K.V ext4_mark_inode_dirty(handle, inode); 1107f8514083SAneesh Kumar K.V 1108ffacfa7aSJan Kara if (pos + len > inode->i_size && ext4_can_truncate(inode)) 1109f8514083SAneesh Kumar K.V /* if we have allocated more blocks and copied 1110f8514083SAneesh Kumar K.V * less. We will have blocks allocated outside 1111f8514083SAneesh Kumar K.V * inode->i_size. So truncate them 1112f8514083SAneesh Kumar K.V */ 1113f8514083SAneesh Kumar K.V ext4_orphan_add(handle, inode); 111474d553aaSTheodore Ts'o errout: 1115617ba13bSMingming Cao ret2 = ext4_journal_stop(handle); 1116ac27a0ecSDave Kleikamp if (!ret) 1117ac27a0ecSDave Kleikamp ret = ret2; 1118bfc1af65SNick Piggin 1119f8514083SAneesh Kumar K.V if (pos + len > inode->i_size) { 1120b9a4207dSJan Kara ext4_truncate_failed_write(inode); 1121f8514083SAneesh Kumar K.V /* 1122ffacfa7aSJan Kara * If truncate failed early the inode might still be 1123f8514083SAneesh Kumar K.V * on the orphan list; we need to make sure the inode 1124f8514083SAneesh Kumar K.V * is removed from the orphan list in that case. 1125f8514083SAneesh Kumar K.V */ 1126f8514083SAneesh Kumar K.V if (inode->i_nlink) 1127f8514083SAneesh Kumar K.V ext4_orphan_del(NULL, inode); 1128f8514083SAneesh Kumar K.V } 1129f8514083SAneesh Kumar K.V 1130bfc1af65SNick Piggin return ret ? ret : copied; 1131ac27a0ecSDave Kleikamp } 1132ac27a0ecSDave Kleikamp 1133bfc1af65SNick Piggin static int ext4_journalled_write_end(struct file *file, 1134bfc1af65SNick Piggin struct address_space *mapping, 1135bfc1af65SNick Piggin loff_t pos, unsigned len, unsigned copied, 1136bfc1af65SNick Piggin struct page *page, void *fsdata) 1137ac27a0ecSDave Kleikamp { 1138617ba13bSMingming Cao handle_t *handle = ext4_journal_current_handle(); 1139bfc1af65SNick Piggin struct inode *inode = mapping->host; 1140ac27a0ecSDave Kleikamp int ret = 0, ret2; 1141ac27a0ecSDave Kleikamp int partial = 0; 1142bfc1af65SNick Piggin unsigned from, to; 1143cf17fea6SAneesh Kumar K.V loff_t new_i_size; 1144ac27a0ecSDave Kleikamp 11459bffad1eSTheodore Ts'o trace_ext4_journalled_write_end(inode, pos, len, copied); 1146bfc1af65SNick Piggin from = pos & (PAGE_CACHE_SIZE - 1); 1147bfc1af65SNick Piggin to = from + len; 1148bfc1af65SNick Piggin 1149441c8508SCurt Wohlgemuth BUG_ON(!ext4_handle_valid(handle)); 1150441c8508SCurt Wohlgemuth 11513fdcfb66STao Ma if (ext4_has_inline_data(inode)) 11523fdcfb66STao Ma copied = ext4_write_inline_data_end(inode, pos, len, 11533fdcfb66STao Ma copied, page); 11543fdcfb66STao Ma else { 1155bfc1af65SNick Piggin if (copied < len) { 1156bfc1af65SNick Piggin if (!PageUptodate(page)) 1157bfc1af65SNick Piggin copied = 0; 1158bfc1af65SNick Piggin page_zero_new_buffers(page, from+copied, to); 1159bfc1af65SNick Piggin } 1160ac27a0ecSDave Kleikamp 1161f19d5870STao Ma ret = ext4_walk_page_buffers(handle, page_buffers(page), from, 1162bfc1af65SNick Piggin to, &partial, write_end_fn); 1163ac27a0ecSDave Kleikamp if (!partial) 1164ac27a0ecSDave Kleikamp SetPageUptodate(page); 11653fdcfb66STao Ma } 1166cf17fea6SAneesh Kumar K.V new_i_size = pos + copied; 1167cf17fea6SAneesh Kumar K.V if (new_i_size > inode->i_size) 1168bfc1af65SNick Piggin i_size_write(inode, pos+copied); 116919f5fb7aSTheodore Ts'o ext4_set_inode_state(inode, EXT4_STATE_JDATA); 11702d859db3SJan Kara EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; 1171cf17fea6SAneesh Kumar K.V if (new_i_size > EXT4_I(inode)->i_disksize) { 1172cf17fea6SAneesh Kumar K.V ext4_update_i_disksize(inode, new_i_size); 1173617ba13bSMingming Cao ret2 = ext4_mark_inode_dirty(handle, inode); 1174ac27a0ecSDave Kleikamp if (!ret) 1175ac27a0ecSDave Kleikamp ret = ret2; 1176ac27a0ecSDave Kleikamp } 1177bfc1af65SNick Piggin 1178cf108bcaSJan Kara unlock_page(page); 1179f8514083SAneesh Kumar K.V page_cache_release(page); 1180ffacfa7aSJan Kara if (pos + len > inode->i_size && ext4_can_truncate(inode)) 1181f8514083SAneesh Kumar K.V /* if we have allocated more blocks and copied 1182f8514083SAneesh Kumar K.V * less. We will have blocks allocated outside 1183f8514083SAneesh Kumar K.V * inode->i_size. So truncate them 1184f8514083SAneesh Kumar K.V */ 1185f8514083SAneesh Kumar K.V ext4_orphan_add(handle, inode); 1186f8514083SAneesh Kumar K.V 1187617ba13bSMingming Cao ret2 = ext4_journal_stop(handle); 1188ac27a0ecSDave Kleikamp if (!ret) 1189ac27a0ecSDave Kleikamp ret = ret2; 1190f8514083SAneesh Kumar K.V if (pos + len > inode->i_size) { 1191b9a4207dSJan Kara ext4_truncate_failed_write(inode); 1192f8514083SAneesh Kumar K.V /* 1193ffacfa7aSJan Kara * If truncate failed early the inode might still be 1194f8514083SAneesh Kumar K.V * on the orphan list; we need to make sure the inode 1195f8514083SAneesh Kumar K.V * is removed from the orphan list in that case. 1196f8514083SAneesh Kumar K.V */ 1197f8514083SAneesh Kumar K.V if (inode->i_nlink) 1198f8514083SAneesh Kumar K.V ext4_orphan_del(NULL, inode); 1199f8514083SAneesh Kumar K.V } 1200bfc1af65SNick Piggin 1201bfc1af65SNick Piggin return ret ? ret : copied; 1202ac27a0ecSDave Kleikamp } 1203d2a17637SMingming Cao 12049d0be502STheodore Ts'o /* 1205386ad67cSLukas Czerner * Reserve a metadata for a single block located at lblock 1206386ad67cSLukas Czerner */ 1207386ad67cSLukas Czerner static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock) 1208386ad67cSLukas Czerner { 1209386ad67cSLukas Czerner int retries = 0; 1210386ad67cSLukas Czerner struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 1211386ad67cSLukas Czerner struct ext4_inode_info *ei = EXT4_I(inode); 1212386ad67cSLukas Czerner unsigned int md_needed; 1213386ad67cSLukas Czerner ext4_lblk_t save_last_lblock; 1214386ad67cSLukas Czerner int save_len; 1215386ad67cSLukas Czerner 1216386ad67cSLukas Czerner /* 1217386ad67cSLukas Czerner * recalculate the amount of metadata blocks to reserve 1218386ad67cSLukas Czerner * in order to allocate nrblocks 1219386ad67cSLukas Czerner * worse case is one extent per block 1220386ad67cSLukas Czerner */ 1221386ad67cSLukas Czerner repeat: 1222386ad67cSLukas Czerner spin_lock(&ei->i_block_reservation_lock); 1223386ad67cSLukas Czerner /* 1224386ad67cSLukas Czerner * ext4_calc_metadata_amount() has side effects, which we have 1225386ad67cSLukas Czerner * to be prepared undo if we fail to claim space. 1226386ad67cSLukas Czerner */ 1227386ad67cSLukas Czerner save_len = ei->i_da_metadata_calc_len; 1228386ad67cSLukas Czerner save_last_lblock = ei->i_da_metadata_calc_last_lblock; 1229386ad67cSLukas Czerner md_needed = EXT4_NUM_B2C(sbi, 1230386ad67cSLukas Czerner ext4_calc_metadata_amount(inode, lblock)); 1231386ad67cSLukas Czerner trace_ext4_da_reserve_space(inode, md_needed); 1232386ad67cSLukas Czerner 1233386ad67cSLukas Czerner /* 1234386ad67cSLukas Czerner * We do still charge estimated metadata to the sb though; 1235386ad67cSLukas Czerner * we cannot afford to run out of free blocks. 1236386ad67cSLukas Czerner */ 1237386ad67cSLukas Czerner if (ext4_claim_free_clusters(sbi, md_needed, 0)) { 1238386ad67cSLukas Czerner ei->i_da_metadata_calc_len = save_len; 1239386ad67cSLukas Czerner ei->i_da_metadata_calc_last_lblock = save_last_lblock; 1240386ad67cSLukas Czerner spin_unlock(&ei->i_block_reservation_lock); 1241386ad67cSLukas Czerner if (ext4_should_retry_alloc(inode->i_sb, &retries)) { 1242386ad67cSLukas Czerner cond_resched(); 1243386ad67cSLukas Czerner goto repeat; 1244386ad67cSLukas Czerner } 1245386ad67cSLukas Czerner return -ENOSPC; 1246386ad67cSLukas Czerner } 1247386ad67cSLukas Czerner ei->i_reserved_meta_blocks += md_needed; 1248386ad67cSLukas Czerner spin_unlock(&ei->i_block_reservation_lock); 1249386ad67cSLukas Czerner 1250386ad67cSLukas Czerner return 0; /* success */ 1251386ad67cSLukas Czerner } 1252386ad67cSLukas Czerner 1253386ad67cSLukas Czerner /* 12547b415bf6SAditya Kali * Reserve a single cluster located at lblock 12559d0be502STheodore Ts'o */ 125601f49d0bSTheodore Ts'o static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock) 1257d2a17637SMingming Cao { 1258030ba6bcSAneesh Kumar K.V int retries = 0; 1259d2a17637SMingming Cao struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 12600637c6f4STheodore Ts'o struct ext4_inode_info *ei = EXT4_I(inode); 12617b415bf6SAditya Kali unsigned int md_needed; 12625dd4056dSChristoph Hellwig int ret; 126303179fe9STheodore Ts'o ext4_lblk_t save_last_lblock; 126403179fe9STheodore Ts'o int save_len; 1265d2a17637SMingming Cao 126660e58e0fSMingming Cao /* 126772b8ab9dSEric Sandeen * We will charge metadata quota at writeout time; this saves 126872b8ab9dSEric Sandeen * us from metadata over-estimation, though we may go over by 126972b8ab9dSEric Sandeen * a small amount in the end. Here we just reserve for data. 127060e58e0fSMingming Cao */ 12717b415bf6SAditya Kali ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); 12725dd4056dSChristoph Hellwig if (ret) 12735dd4056dSChristoph Hellwig return ret; 127403179fe9STheodore Ts'o 127503179fe9STheodore Ts'o /* 127603179fe9STheodore Ts'o * recalculate the amount of metadata blocks to reserve 127703179fe9STheodore Ts'o * in order to allocate nrblocks 127803179fe9STheodore Ts'o * worse case is one extent per block 127903179fe9STheodore Ts'o */ 128003179fe9STheodore Ts'o repeat: 128103179fe9STheodore Ts'o spin_lock(&ei->i_block_reservation_lock); 128203179fe9STheodore Ts'o /* 128303179fe9STheodore Ts'o * ext4_calc_metadata_amount() has side effects, which we have 128403179fe9STheodore Ts'o * to be prepared undo if we fail to claim space. 128503179fe9STheodore Ts'o */ 128603179fe9STheodore Ts'o save_len = ei->i_da_metadata_calc_len; 128703179fe9STheodore Ts'o save_last_lblock = ei->i_da_metadata_calc_last_lblock; 128803179fe9STheodore Ts'o md_needed = EXT4_NUM_B2C(sbi, 128903179fe9STheodore Ts'o ext4_calc_metadata_amount(inode, lblock)); 129003179fe9STheodore Ts'o trace_ext4_da_reserve_space(inode, md_needed); 129103179fe9STheodore Ts'o 129272b8ab9dSEric Sandeen /* 129372b8ab9dSEric Sandeen * We do still charge estimated metadata to the sb though; 129472b8ab9dSEric Sandeen * we cannot afford to run out of free blocks. 129572b8ab9dSEric Sandeen */ 1296e7d5f315STheodore Ts'o if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) { 129703179fe9STheodore Ts'o ei->i_da_metadata_calc_len = save_len; 129803179fe9STheodore Ts'o ei->i_da_metadata_calc_last_lblock = save_last_lblock; 129903179fe9STheodore Ts'o spin_unlock(&ei->i_block_reservation_lock); 1300030ba6bcSAneesh Kumar K.V if (ext4_should_retry_alloc(inode->i_sb, &retries)) { 1301bb8b20edSLukas Czerner cond_resched(); 1302030ba6bcSAneesh Kumar K.V goto repeat; 1303030ba6bcSAneesh Kumar K.V } 130403179fe9STheodore Ts'o dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); 1305d2a17637SMingming Cao return -ENOSPC; 1306d2a17637SMingming Cao } 13079d0be502STheodore Ts'o ei->i_reserved_data_blocks++; 13080637c6f4STheodore Ts'o ei->i_reserved_meta_blocks += md_needed; 13090637c6f4STheodore Ts'o spin_unlock(&ei->i_block_reservation_lock); 131039bc680aSDmitry Monakhov 1311d2a17637SMingming Cao return 0; /* success */ 1312d2a17637SMingming Cao } 1313d2a17637SMingming Cao 131412219aeaSAneesh Kumar K.V static void ext4_da_release_space(struct inode *inode, int to_free) 1315d2a17637SMingming Cao { 1316d2a17637SMingming Cao struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 13170637c6f4STheodore Ts'o struct ext4_inode_info *ei = EXT4_I(inode); 1318d2a17637SMingming Cao 1319cd213226SMingming Cao if (!to_free) 1320cd213226SMingming Cao return; /* Nothing to release, exit */ 1321cd213226SMingming Cao 1322d2a17637SMingming Cao spin_lock(&EXT4_I(inode)->i_block_reservation_lock); 1323cd213226SMingming Cao 13245a58ec87SLi Zefan trace_ext4_da_release_space(inode, to_free); 13250637c6f4STheodore Ts'o if (unlikely(to_free > ei->i_reserved_data_blocks)) { 1326cd213226SMingming Cao /* 13270637c6f4STheodore Ts'o * if there aren't enough reserved blocks, then the 13280637c6f4STheodore Ts'o * counter is messed up somewhere. Since this 13290637c6f4STheodore Ts'o * function is called from invalidate page, it's 13300637c6f4STheodore Ts'o * harmless to return without any action. 1331cd213226SMingming Cao */ 13328de5c325STheodore Ts'o ext4_warning(inode->i_sb, "ext4_da_release_space: " 13330637c6f4STheodore Ts'o "ino %lu, to_free %d with only %d reserved " 13341084f252STheodore Ts'o "data blocks", inode->i_ino, to_free, 13350637c6f4STheodore Ts'o ei->i_reserved_data_blocks); 13360637c6f4STheodore Ts'o WARN_ON(1); 13370637c6f4STheodore Ts'o to_free = ei->i_reserved_data_blocks; 13380637c6f4STheodore Ts'o } 13390637c6f4STheodore Ts'o ei->i_reserved_data_blocks -= to_free; 13400637c6f4STheodore Ts'o 13410637c6f4STheodore Ts'o if (ei->i_reserved_data_blocks == 0) { 13420637c6f4STheodore Ts'o /* 13430637c6f4STheodore Ts'o * We can release all of the reserved metadata blocks 13440637c6f4STheodore Ts'o * only when we have written all of the delayed 13450637c6f4STheodore Ts'o * allocation blocks. 13467b415bf6SAditya Kali * Note that in case of bigalloc, i_reserved_meta_blocks, 13477b415bf6SAditya Kali * i_reserved_data_blocks, etc. refer to number of clusters. 13480637c6f4STheodore Ts'o */ 134957042651STheodore Ts'o percpu_counter_sub(&sbi->s_dirtyclusters_counter, 135072b8ab9dSEric Sandeen ei->i_reserved_meta_blocks); 1351ee5f4d9cSTheodore Ts'o ei->i_reserved_meta_blocks = 0; 13529d0be502STheodore Ts'o ei->i_da_metadata_calc_len = 0; 1353cd213226SMingming Cao } 1354cd213226SMingming Cao 135572b8ab9dSEric Sandeen /* update fs dirty data blocks counter */ 135657042651STheodore Ts'o percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); 1357d2a17637SMingming Cao 1358d2a17637SMingming Cao spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); 135960e58e0fSMingming Cao 13607b415bf6SAditya Kali dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); 1361d2a17637SMingming Cao } 1362d2a17637SMingming Cao 1363d2a17637SMingming Cao static void ext4_da_page_release_reservation(struct page *page, 1364ca99fdd2SLukas Czerner unsigned int offset, 1365ca99fdd2SLukas Czerner unsigned int length) 1366d2a17637SMingming Cao { 1367d2a17637SMingming Cao int to_release = 0; 1368d2a17637SMingming Cao struct buffer_head *head, *bh; 1369d2a17637SMingming Cao unsigned int curr_off = 0; 13707b415bf6SAditya Kali struct inode *inode = page->mapping->host; 13717b415bf6SAditya Kali struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 1372ca99fdd2SLukas Czerner unsigned int stop = offset + length; 13737b415bf6SAditya Kali int num_clusters; 137451865fdaSZheng Liu ext4_fsblk_t lblk; 1375d2a17637SMingming Cao 1376ca99fdd2SLukas Czerner BUG_ON(stop > PAGE_CACHE_SIZE || stop < length); 1377ca99fdd2SLukas Czerner 1378d2a17637SMingming Cao head = page_buffers(page); 1379d2a17637SMingming Cao bh = head; 1380d2a17637SMingming Cao do { 1381d2a17637SMingming Cao unsigned int next_off = curr_off + bh->b_size; 1382d2a17637SMingming Cao 1383ca99fdd2SLukas Czerner if (next_off > stop) 1384ca99fdd2SLukas Czerner break; 1385ca99fdd2SLukas Czerner 1386d2a17637SMingming Cao if ((offset <= curr_off) && (buffer_delay(bh))) { 1387d2a17637SMingming Cao to_release++; 1388d2a17637SMingming Cao clear_buffer_delay(bh); 1389d2a17637SMingming Cao } 1390d2a17637SMingming Cao curr_off = next_off; 1391d2a17637SMingming Cao } while ((bh = bh->b_this_page) != head); 13927b415bf6SAditya Kali 139351865fdaSZheng Liu if (to_release) { 139451865fdaSZheng Liu lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 139551865fdaSZheng Liu ext4_es_remove_extent(inode, lblk, to_release); 139651865fdaSZheng Liu } 139751865fdaSZheng Liu 13987b415bf6SAditya Kali /* If we have released all the blocks belonging to a cluster, then we 13997b415bf6SAditya Kali * need to release the reserved space for that cluster. */ 14007b415bf6SAditya Kali num_clusters = EXT4_NUM_B2C(sbi, to_release); 14017b415bf6SAditya Kali while (num_clusters > 0) { 14027b415bf6SAditya Kali lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) + 14037b415bf6SAditya Kali ((num_clusters - 1) << sbi->s_cluster_bits); 14047b415bf6SAditya Kali if (sbi->s_cluster_ratio == 1 || 14057d1b1fbcSZheng Liu !ext4_find_delalloc_cluster(inode, lblk)) 14067b415bf6SAditya Kali ext4_da_release_space(inode, 1); 14077b415bf6SAditya Kali 14087b415bf6SAditya Kali num_clusters--; 14097b415bf6SAditya Kali } 1410d2a17637SMingming Cao } 1411ac27a0ecSDave Kleikamp 1412ac27a0ecSDave Kleikamp /* 141364769240SAlex Tomas * Delayed allocation stuff 141464769240SAlex Tomas */ 141564769240SAlex Tomas 14164e7ea81dSJan Kara struct mpage_da_data { 14174e7ea81dSJan Kara struct inode *inode; 14184e7ea81dSJan Kara struct writeback_control *wbc; 14196b523df4SJan Kara 14204e7ea81dSJan Kara pgoff_t first_page; /* The first page to write */ 14214e7ea81dSJan Kara pgoff_t next_page; /* Current page to examine */ 14224e7ea81dSJan Kara pgoff_t last_page; /* Last page to examine */ 142364769240SAlex Tomas /* 14244e7ea81dSJan Kara * Extent to map - this can be after first_page because that can be 14254e7ea81dSJan Kara * fully mapped. We somewhat abuse m_flags to store whether the extent 14264e7ea81dSJan Kara * is delalloc or unwritten. 142764769240SAlex Tomas */ 14284e7ea81dSJan Kara struct ext4_map_blocks map; 14294e7ea81dSJan Kara struct ext4_io_submit io_submit; /* IO submission data */ 14304e7ea81dSJan Kara }; 143164769240SAlex Tomas 14324e7ea81dSJan Kara static void mpage_release_unused_pages(struct mpage_da_data *mpd, 14334e7ea81dSJan Kara bool invalidate) 1434c4a0c46eSAneesh Kumar K.V { 1435c4a0c46eSAneesh Kumar K.V int nr_pages, i; 1436c4a0c46eSAneesh Kumar K.V pgoff_t index, end; 1437c4a0c46eSAneesh Kumar K.V struct pagevec pvec; 1438c4a0c46eSAneesh Kumar K.V struct inode *inode = mpd->inode; 1439c4a0c46eSAneesh Kumar K.V struct address_space *mapping = inode->i_mapping; 14404e7ea81dSJan Kara 14414e7ea81dSJan Kara /* This is necessary when next_page == 0. */ 14424e7ea81dSJan Kara if (mpd->first_page >= mpd->next_page) 14434e7ea81dSJan Kara return; 1444c4a0c46eSAneesh Kumar K.V 1445c7f5938aSCurt Wohlgemuth index = mpd->first_page; 1446c7f5938aSCurt Wohlgemuth end = mpd->next_page - 1; 14474e7ea81dSJan Kara if (invalidate) { 14484e7ea81dSJan Kara ext4_lblk_t start, last; 144951865fdaSZheng Liu start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 145051865fdaSZheng Liu last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits); 145151865fdaSZheng Liu ext4_es_remove_extent(inode, start, last - start + 1); 14524e7ea81dSJan Kara } 145351865fdaSZheng Liu 145466bea92cSEric Sandeen pagevec_init(&pvec, 0); 1455c4a0c46eSAneesh Kumar K.V while (index <= end) { 1456c4a0c46eSAneesh Kumar K.V nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); 1457c4a0c46eSAneesh Kumar K.V if (nr_pages == 0) 1458c4a0c46eSAneesh Kumar K.V break; 1459c4a0c46eSAneesh Kumar K.V for (i = 0; i < nr_pages; i++) { 1460c4a0c46eSAneesh Kumar K.V struct page *page = pvec.pages[i]; 14619b1d0998SJan Kara if (page->index > end) 1462c4a0c46eSAneesh Kumar K.V break; 1463c4a0c46eSAneesh Kumar K.V BUG_ON(!PageLocked(page)); 1464c4a0c46eSAneesh Kumar K.V BUG_ON(PageWriteback(page)); 14654e7ea81dSJan Kara if (invalidate) { 1466d47992f8SLukas Czerner block_invalidatepage(page, 0, PAGE_CACHE_SIZE); 1467c4a0c46eSAneesh Kumar K.V ClearPageUptodate(page); 14684e7ea81dSJan Kara } 1469c4a0c46eSAneesh Kumar K.V unlock_page(page); 1470c4a0c46eSAneesh Kumar K.V } 14719b1d0998SJan Kara index = pvec.pages[nr_pages - 1]->index + 1; 14729b1d0998SJan Kara pagevec_release(&pvec); 1473c4a0c46eSAneesh Kumar K.V } 1474c4a0c46eSAneesh Kumar K.V } 1475c4a0c46eSAneesh Kumar K.V 1476df22291fSAneesh Kumar K.V static void ext4_print_free_blocks(struct inode *inode) 1477df22291fSAneesh Kumar K.V { 1478df22291fSAneesh Kumar K.V struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 147992b97816STheodore Ts'o struct super_block *sb = inode->i_sb; 1480f78ee70dSLukas Czerner struct ext4_inode_info *ei = EXT4_I(inode); 148192b97816STheodore Ts'o 148292b97816STheodore Ts'o ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", 14835dee5437STheodore Ts'o EXT4_C2B(EXT4_SB(inode->i_sb), 1484f78ee70dSLukas Czerner ext4_count_free_clusters(sb))); 148592b97816STheodore Ts'o ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); 148692b97816STheodore Ts'o ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", 1487f78ee70dSLukas Czerner (long long) EXT4_C2B(EXT4_SB(sb), 148857042651STheodore Ts'o percpu_counter_sum(&sbi->s_freeclusters_counter))); 148992b97816STheodore Ts'o ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", 1490f78ee70dSLukas Czerner (long long) EXT4_C2B(EXT4_SB(sb), 14917b415bf6SAditya Kali percpu_counter_sum(&sbi->s_dirtyclusters_counter))); 149292b97816STheodore Ts'o ext4_msg(sb, KERN_CRIT, "Block reservation details"); 149392b97816STheodore Ts'o ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", 1494f78ee70dSLukas Czerner ei->i_reserved_data_blocks); 149592b97816STheodore Ts'o ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u", 1496f78ee70dSLukas Czerner ei->i_reserved_meta_blocks); 1497f78ee70dSLukas Czerner ext4_msg(sb, KERN_CRIT, "i_allocated_meta_blocks=%u", 1498f78ee70dSLukas Czerner ei->i_allocated_meta_blocks); 1499df22291fSAneesh Kumar K.V return; 1500df22291fSAneesh Kumar K.V } 1501df22291fSAneesh Kumar K.V 1502c364b22cSAneesh Kumar K.V static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) 150329fa89d0SAneesh Kumar K.V { 1504c364b22cSAneesh Kumar K.V return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); 150529fa89d0SAneesh Kumar K.V } 150629fa89d0SAneesh Kumar K.V 150764769240SAlex Tomas /* 15085356f261SAditya Kali * This function is grabs code from the very beginning of 15095356f261SAditya Kali * ext4_map_blocks, but assumes that the caller is from delayed write 15105356f261SAditya Kali * time. This function looks up the requested blocks and sets the 15115356f261SAditya Kali * buffer delay bit under the protection of i_data_sem. 15125356f261SAditya Kali */ 15135356f261SAditya Kali static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, 15145356f261SAditya Kali struct ext4_map_blocks *map, 15155356f261SAditya Kali struct buffer_head *bh) 15165356f261SAditya Kali { 1517d100eef2SZheng Liu struct extent_status es; 15185356f261SAditya Kali int retval; 15195356f261SAditya Kali sector_t invalid_block = ~((sector_t) 0xffff); 1520921f266bSDmitry Monakhov #ifdef ES_AGGRESSIVE_TEST 1521921f266bSDmitry Monakhov struct ext4_map_blocks orig_map; 1522921f266bSDmitry Monakhov 1523921f266bSDmitry Monakhov memcpy(&orig_map, map, sizeof(*map)); 1524921f266bSDmitry Monakhov #endif 15255356f261SAditya Kali 15265356f261SAditya Kali if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) 15275356f261SAditya Kali invalid_block = ~0; 15285356f261SAditya Kali 15295356f261SAditya Kali map->m_flags = 0; 15305356f261SAditya Kali ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," 15315356f261SAditya Kali "logical block %lu\n", inode->i_ino, map->m_len, 15325356f261SAditya Kali (unsigned long) map->m_lblk); 1533d100eef2SZheng Liu 1534d100eef2SZheng Liu /* Lookup extent status tree firstly */ 1535d100eef2SZheng Liu if (ext4_es_lookup_extent(inode, iblock, &es)) { 153663b99968STheodore Ts'o ext4_es_lru_add(inode); 1537d100eef2SZheng Liu if (ext4_es_is_hole(&es)) { 1538d100eef2SZheng Liu retval = 0; 1539d100eef2SZheng Liu down_read((&EXT4_I(inode)->i_data_sem)); 1540d100eef2SZheng Liu goto add_delayed; 1541d100eef2SZheng Liu } 1542d100eef2SZheng Liu 1543d100eef2SZheng Liu /* 1544d100eef2SZheng Liu * Delayed extent could be allocated by fallocate. 1545d100eef2SZheng Liu * So we need to check it. 1546d100eef2SZheng Liu */ 1547d100eef2SZheng Liu if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { 1548d100eef2SZheng Liu map_bh(bh, inode->i_sb, invalid_block); 1549d100eef2SZheng Liu set_buffer_new(bh); 1550d100eef2SZheng Liu set_buffer_delay(bh); 1551d100eef2SZheng Liu return 0; 1552d100eef2SZheng Liu } 1553d100eef2SZheng Liu 1554d100eef2SZheng Liu map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; 1555d100eef2SZheng Liu retval = es.es_len - (iblock - es.es_lblk); 1556d100eef2SZheng Liu if (retval > map->m_len) 1557d100eef2SZheng Liu retval = map->m_len; 1558d100eef2SZheng Liu map->m_len = retval; 1559d100eef2SZheng Liu if (ext4_es_is_written(&es)) 1560d100eef2SZheng Liu map->m_flags |= EXT4_MAP_MAPPED; 1561d100eef2SZheng Liu else if (ext4_es_is_unwritten(&es)) 1562d100eef2SZheng Liu map->m_flags |= EXT4_MAP_UNWRITTEN; 1563d100eef2SZheng Liu else 1564d100eef2SZheng Liu BUG_ON(1); 1565d100eef2SZheng Liu 1566921f266bSDmitry Monakhov #ifdef ES_AGGRESSIVE_TEST 1567921f266bSDmitry Monakhov ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); 1568921f266bSDmitry Monakhov #endif 1569d100eef2SZheng Liu return retval; 1570d100eef2SZheng Liu } 1571d100eef2SZheng Liu 15725356f261SAditya Kali /* 15735356f261SAditya Kali * Try to see if we can get the block without requesting a new 15745356f261SAditya Kali * file system block. 15755356f261SAditya Kali */ 15765356f261SAditya Kali down_read((&EXT4_I(inode)->i_data_sem)); 15779c3569b5STao Ma if (ext4_has_inline_data(inode)) { 15789c3569b5STao Ma /* 15799c3569b5STao Ma * We will soon create blocks for this page, and let 15809c3569b5STao Ma * us pretend as if the blocks aren't allocated yet. 15819c3569b5STao Ma * In case of clusters, we have to handle the work 15829c3569b5STao Ma * of mapping from cluster so that the reserved space 15839c3569b5STao Ma * is calculated properly. 15849c3569b5STao Ma */ 15859c3569b5STao Ma if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) && 15869c3569b5STao Ma ext4_find_delalloc_cluster(inode, map->m_lblk)) 15879c3569b5STao Ma map->m_flags |= EXT4_MAP_FROM_CLUSTER; 15889c3569b5STao Ma retval = 0; 15899c3569b5STao Ma } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 1590d100eef2SZheng Liu retval = ext4_ext_map_blocks(NULL, inode, map, 1591d100eef2SZheng Liu EXT4_GET_BLOCKS_NO_PUT_HOLE); 15925356f261SAditya Kali else 1593d100eef2SZheng Liu retval = ext4_ind_map_blocks(NULL, inode, map, 1594d100eef2SZheng Liu EXT4_GET_BLOCKS_NO_PUT_HOLE); 15955356f261SAditya Kali 1596d100eef2SZheng Liu add_delayed: 15975356f261SAditya Kali if (retval == 0) { 1598f7fec032SZheng Liu int ret; 15995356f261SAditya Kali /* 16005356f261SAditya Kali * XXX: __block_prepare_write() unmaps passed block, 16015356f261SAditya Kali * is it OK? 16025356f261SAditya Kali */ 1603386ad67cSLukas Czerner /* 1604386ad67cSLukas Czerner * If the block was allocated from previously allocated cluster, 1605386ad67cSLukas Czerner * then we don't need to reserve it again. However we still need 1606386ad67cSLukas Czerner * to reserve metadata for every block we're going to write. 1607386ad67cSLukas Czerner */ 16085356f261SAditya Kali if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) { 1609f7fec032SZheng Liu ret = ext4_da_reserve_space(inode, iblock); 1610f7fec032SZheng Liu if (ret) { 16115356f261SAditya Kali /* not enough space to reserve */ 1612f7fec032SZheng Liu retval = ret; 16135356f261SAditya Kali goto out_unlock; 16145356f261SAditya Kali } 1615386ad67cSLukas Czerner } else { 1616386ad67cSLukas Czerner ret = ext4_da_reserve_metadata(inode, iblock); 1617386ad67cSLukas Czerner if (ret) { 1618386ad67cSLukas Czerner /* not enough space to reserve */ 1619386ad67cSLukas Czerner retval = ret; 1620386ad67cSLukas Czerner goto out_unlock; 1621386ad67cSLukas Czerner } 1622f7fec032SZheng Liu } 16235356f261SAditya Kali 1624f7fec032SZheng Liu ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, 1625fdc0212eSZheng Liu ~0, EXTENT_STATUS_DELAYED); 1626f7fec032SZheng Liu if (ret) { 1627f7fec032SZheng Liu retval = ret; 162851865fdaSZheng Liu goto out_unlock; 1629f7fec032SZheng Liu } 163051865fdaSZheng Liu 16315356f261SAditya Kali /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served 16325356f261SAditya Kali * and it should not appear on the bh->b_state. 16335356f261SAditya Kali */ 16345356f261SAditya Kali map->m_flags &= ~EXT4_MAP_FROM_CLUSTER; 16355356f261SAditya Kali 16365356f261SAditya Kali map_bh(bh, inode->i_sb, invalid_block); 16375356f261SAditya Kali set_buffer_new(bh); 16385356f261SAditya Kali set_buffer_delay(bh); 1639f7fec032SZheng Liu } else if (retval > 0) { 1640f7fec032SZheng Liu int ret; 16413be78c73STheodore Ts'o unsigned int status; 1642f7fec032SZheng Liu 164344fb851dSZheng Liu if (unlikely(retval != map->m_len)) { 164444fb851dSZheng Liu ext4_warning(inode->i_sb, 164544fb851dSZheng Liu "ES len assertion failed for inode " 164644fb851dSZheng Liu "%lu: retval %d != map->m_len %d", 164744fb851dSZheng Liu inode->i_ino, retval, map->m_len); 164844fb851dSZheng Liu WARN_ON(1); 1649921f266bSDmitry Monakhov } 1650921f266bSDmitry Monakhov 1651f7fec032SZheng Liu status = map->m_flags & EXT4_MAP_UNWRITTEN ? 1652f7fec032SZheng Liu EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; 1653f7fec032SZheng Liu ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, 1654f7fec032SZheng Liu map->m_pblk, status); 1655f7fec032SZheng Liu if (ret != 0) 1656f7fec032SZheng Liu retval = ret; 16575356f261SAditya Kali } 16585356f261SAditya Kali 16595356f261SAditya Kali out_unlock: 16605356f261SAditya Kali up_read((&EXT4_I(inode)->i_data_sem)); 16615356f261SAditya Kali 16625356f261SAditya Kali return retval; 16635356f261SAditya Kali } 16645356f261SAditya Kali 16655356f261SAditya Kali /* 1666b920c755STheodore Ts'o * This is a special get_blocks_t callback which is used by 1667b920c755STheodore Ts'o * ext4_da_write_begin(). It will either return mapped block or 1668b920c755STheodore Ts'o * reserve space for a single block. 166929fa89d0SAneesh Kumar K.V * 167029fa89d0SAneesh Kumar K.V * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. 167129fa89d0SAneesh Kumar K.V * We also have b_blocknr = -1 and b_bdev initialized properly 167229fa89d0SAneesh Kumar K.V * 167329fa89d0SAneesh Kumar K.V * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. 167429fa89d0SAneesh Kumar K.V * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev 167529fa89d0SAneesh Kumar K.V * initialized properly. 167664769240SAlex Tomas */ 16779c3569b5STao Ma int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, 16782ed88685STheodore Ts'o struct buffer_head *bh, int create) 167964769240SAlex Tomas { 16802ed88685STheodore Ts'o struct ext4_map_blocks map; 168164769240SAlex Tomas int ret = 0; 168264769240SAlex Tomas 168364769240SAlex Tomas BUG_ON(create == 0); 16842ed88685STheodore Ts'o BUG_ON(bh->b_size != inode->i_sb->s_blocksize); 16852ed88685STheodore Ts'o 16862ed88685STheodore Ts'o map.m_lblk = iblock; 16872ed88685STheodore Ts'o map.m_len = 1; 168864769240SAlex Tomas 168964769240SAlex Tomas /* 169064769240SAlex Tomas * first, we need to know whether the block is allocated already 169164769240SAlex Tomas * preallocated blocks are unmapped but should treated 169264769240SAlex Tomas * the same as allocated blocks. 169364769240SAlex Tomas */ 16945356f261SAditya Kali ret = ext4_da_map_blocks(inode, iblock, &map, bh); 16955356f261SAditya Kali if (ret <= 0) 16962ed88685STheodore Ts'o return ret; 169764769240SAlex Tomas 16982ed88685STheodore Ts'o map_bh(bh, inode->i_sb, map.m_pblk); 16992ed88685STheodore Ts'o bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; 17002ed88685STheodore Ts'o 17012ed88685STheodore Ts'o if (buffer_unwritten(bh)) { 17022ed88685STheodore Ts'o /* A delayed write to unwritten bh should be marked 17032ed88685STheodore Ts'o * new and mapped. Mapped ensures that we don't do 17042ed88685STheodore Ts'o * get_block multiple times when we write to the same 17052ed88685STheodore Ts'o * offset and new ensures that we do proper zero out 17062ed88685STheodore Ts'o * for partial write. 17072ed88685STheodore Ts'o */ 17082ed88685STheodore Ts'o set_buffer_new(bh); 1709c8205636STheodore Ts'o set_buffer_mapped(bh); 17102ed88685STheodore Ts'o } 17112ed88685STheodore Ts'o return 0; 171264769240SAlex Tomas } 171361628a3fSMingming Cao 171462e086beSAneesh Kumar K.V static int bget_one(handle_t *handle, struct buffer_head *bh) 171562e086beSAneesh Kumar K.V { 171662e086beSAneesh Kumar K.V get_bh(bh); 171762e086beSAneesh Kumar K.V return 0; 171862e086beSAneesh Kumar K.V } 171962e086beSAneesh Kumar K.V 172062e086beSAneesh Kumar K.V static int bput_one(handle_t *handle, struct buffer_head *bh) 172162e086beSAneesh Kumar K.V { 172262e086beSAneesh Kumar K.V put_bh(bh); 172362e086beSAneesh Kumar K.V return 0; 172462e086beSAneesh Kumar K.V } 172562e086beSAneesh Kumar K.V 172662e086beSAneesh Kumar K.V static int __ext4_journalled_writepage(struct page *page, 172762e086beSAneesh Kumar K.V unsigned int len) 172862e086beSAneesh Kumar K.V { 172962e086beSAneesh Kumar K.V struct address_space *mapping = page->mapping; 173062e086beSAneesh Kumar K.V struct inode *inode = mapping->host; 17313fdcfb66STao Ma struct buffer_head *page_bufs = NULL; 173262e086beSAneesh Kumar K.V handle_t *handle = NULL; 17333fdcfb66STao Ma int ret = 0, err = 0; 17343fdcfb66STao Ma int inline_data = ext4_has_inline_data(inode); 17353fdcfb66STao Ma struct buffer_head *inode_bh = NULL; 173662e086beSAneesh Kumar K.V 1737cb20d518STheodore Ts'o ClearPageChecked(page); 17383fdcfb66STao Ma 17393fdcfb66STao Ma if (inline_data) { 17403fdcfb66STao Ma BUG_ON(page->index != 0); 17413fdcfb66STao Ma BUG_ON(len > ext4_get_max_inline_size(inode)); 17423fdcfb66STao Ma inode_bh = ext4_journalled_write_inline_data(inode, len, page); 17433fdcfb66STao Ma if (inode_bh == NULL) 17443fdcfb66STao Ma goto out; 17453fdcfb66STao Ma } else { 174662e086beSAneesh Kumar K.V page_bufs = page_buffers(page); 17473fdcfb66STao Ma if (!page_bufs) { 17483fdcfb66STao Ma BUG(); 17493fdcfb66STao Ma goto out; 17503fdcfb66STao Ma } 17513fdcfb66STao Ma ext4_walk_page_buffers(handle, page_bufs, 0, len, 17523fdcfb66STao Ma NULL, bget_one); 17533fdcfb66STao Ma } 175462e086beSAneesh Kumar K.V /* As soon as we unlock the page, it can go away, but we have 175562e086beSAneesh Kumar K.V * references to buffers so we are safe */ 175662e086beSAneesh Kumar K.V unlock_page(page); 175762e086beSAneesh Kumar K.V 17589924a92aSTheodore Ts'o handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 17599924a92aSTheodore Ts'o ext4_writepage_trans_blocks(inode)); 176062e086beSAneesh Kumar K.V if (IS_ERR(handle)) { 176162e086beSAneesh Kumar K.V ret = PTR_ERR(handle); 176262e086beSAneesh Kumar K.V goto out; 176362e086beSAneesh Kumar K.V } 176462e086beSAneesh Kumar K.V 1765441c8508SCurt Wohlgemuth BUG_ON(!ext4_handle_valid(handle)); 1766441c8508SCurt Wohlgemuth 17673fdcfb66STao Ma if (inline_data) { 17683fdcfb66STao Ma ret = ext4_journal_get_write_access(handle, inode_bh); 17693fdcfb66STao Ma 17703fdcfb66STao Ma err = ext4_handle_dirty_metadata(handle, inode, inode_bh); 17713fdcfb66STao Ma 17723fdcfb66STao Ma } else { 1773f19d5870STao Ma ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, 177462e086beSAneesh Kumar K.V do_journal_get_write_access); 177562e086beSAneesh Kumar K.V 1776f19d5870STao Ma err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, 177762e086beSAneesh Kumar K.V write_end_fn); 17783fdcfb66STao Ma } 177962e086beSAneesh Kumar K.V if (ret == 0) 178062e086beSAneesh Kumar K.V ret = err; 17812d859db3SJan Kara EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; 178262e086beSAneesh Kumar K.V err = ext4_journal_stop(handle); 178362e086beSAneesh Kumar K.V if (!ret) 178462e086beSAneesh Kumar K.V ret = err; 178562e086beSAneesh Kumar K.V 17863fdcfb66STao Ma if (!ext4_has_inline_data(inode)) 17873fdcfb66STao Ma ext4_walk_page_buffers(handle, page_bufs, 0, len, 17883fdcfb66STao Ma NULL, bput_one); 178919f5fb7aSTheodore Ts'o ext4_set_inode_state(inode, EXT4_STATE_JDATA); 179062e086beSAneesh Kumar K.V out: 17913fdcfb66STao Ma brelse(inode_bh); 179262e086beSAneesh Kumar K.V return ret; 179362e086beSAneesh Kumar K.V } 179462e086beSAneesh Kumar K.V 179561628a3fSMingming Cao /* 179643ce1d23SAneesh Kumar K.V * Note that we don't need to start a transaction unless we're journaling data 179743ce1d23SAneesh Kumar K.V * because we should have holes filled from ext4_page_mkwrite(). We even don't 179843ce1d23SAneesh Kumar K.V * need to file the inode to the transaction's list in ordered mode because if 179943ce1d23SAneesh Kumar K.V * we are writing back data added by write(), the inode is already there and if 180043ce1d23SAneesh Kumar K.V * we are writing back data modified via mmap(), no one guarantees in which 180143ce1d23SAneesh Kumar K.V * transaction the data will hit the disk. In case we are journaling data, we 180243ce1d23SAneesh Kumar K.V * cannot start transaction directly because transaction start ranks above page 180343ce1d23SAneesh Kumar K.V * lock so we have to do some magic. 180443ce1d23SAneesh Kumar K.V * 1805b920c755STheodore Ts'o * This function can get called via... 180620970ba6STheodore Ts'o * - ext4_writepages after taking page lock (have journal handle) 1807b920c755STheodore Ts'o * - journal_submit_inode_data_buffers (no journal handle) 1808f6463b0dSArtem Bityutskiy * - shrink_page_list via the kswapd/direct reclaim (no journal handle) 1809b920c755STheodore Ts'o * - grab_page_cache when doing write_begin (have journal handle) 181043ce1d23SAneesh Kumar K.V * 181143ce1d23SAneesh Kumar K.V * We don't do any block allocation in this function. If we have page with 181243ce1d23SAneesh Kumar K.V * multiple blocks we need to write those buffer_heads that are mapped. This 181343ce1d23SAneesh Kumar K.V * is important for mmaped based write. So if we do with blocksize 1K 181443ce1d23SAneesh Kumar K.V * truncate(f, 1024); 181543ce1d23SAneesh Kumar K.V * a = mmap(f, 0, 4096); 181643ce1d23SAneesh Kumar K.V * a[0] = 'a'; 181743ce1d23SAneesh Kumar K.V * truncate(f, 4096); 181843ce1d23SAneesh Kumar K.V * we have in the page first buffer_head mapped via page_mkwrite call back 181990802ed9SPaul Bolle * but other buffer_heads would be unmapped but dirty (dirty done via the 182043ce1d23SAneesh Kumar K.V * do_wp_page). So writepage should write the first block. If we modify 182143ce1d23SAneesh Kumar K.V * the mmap area beyond 1024 we will again get a page_fault and the 182243ce1d23SAneesh Kumar K.V * page_mkwrite callback will do the block allocation and mark the 182343ce1d23SAneesh Kumar K.V * buffer_heads mapped. 182443ce1d23SAneesh Kumar K.V * 182543ce1d23SAneesh Kumar K.V * We redirty the page if we have any buffer_heads that is either delay or 182643ce1d23SAneesh Kumar K.V * unwritten in the page. 182743ce1d23SAneesh Kumar K.V * 182843ce1d23SAneesh Kumar K.V * We can get recursively called as show below. 182943ce1d23SAneesh Kumar K.V * 183043ce1d23SAneesh Kumar K.V * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> 183143ce1d23SAneesh Kumar K.V * ext4_writepage() 183243ce1d23SAneesh Kumar K.V * 183343ce1d23SAneesh Kumar K.V * But since we don't do any block allocation we should not deadlock. 183443ce1d23SAneesh Kumar K.V * Page also have the dirty flag cleared so we don't get recurive page_lock. 183561628a3fSMingming Cao */ 183643ce1d23SAneesh Kumar K.V static int ext4_writepage(struct page *page, 183764769240SAlex Tomas struct writeback_control *wbc) 183864769240SAlex Tomas { 1839f8bec370SJan Kara int ret = 0; 184061628a3fSMingming Cao loff_t size; 1841498e5f24STheodore Ts'o unsigned int len; 1842744692dcSJiaying Zhang struct buffer_head *page_bufs = NULL; 184361628a3fSMingming Cao struct inode *inode = page->mapping->host; 184436ade451SJan Kara struct ext4_io_submit io_submit; 184564769240SAlex Tomas 1846a9c667f8SLukas Czerner trace_ext4_writepage(page); 184761628a3fSMingming Cao size = i_size_read(inode); 184861628a3fSMingming Cao if (page->index == size >> PAGE_CACHE_SHIFT) 184961628a3fSMingming Cao len = size & ~PAGE_CACHE_MASK; 185061628a3fSMingming Cao else 185161628a3fSMingming Cao len = PAGE_CACHE_SIZE; 185261628a3fSMingming Cao 1853f0e6c985SAneesh Kumar K.V page_bufs = page_buffers(page); 185464769240SAlex Tomas /* 1855fe386132SJan Kara * We cannot do block allocation or other extent handling in this 1856fe386132SJan Kara * function. If there are buffers needing that, we have to redirty 1857fe386132SJan Kara * the page. But we may reach here when we do a journal commit via 1858fe386132SJan Kara * journal_submit_inode_data_buffers() and in that case we must write 1859fe386132SJan Kara * allocated buffers to achieve data=ordered mode guarantees. 186064769240SAlex Tomas */ 1861f19d5870STao Ma if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL, 1862c364b22cSAneesh Kumar K.V ext4_bh_delay_or_unwritten)) { 186361628a3fSMingming Cao redirty_page_for_writepage(wbc, page); 1864fe386132SJan Kara if (current->flags & PF_MEMALLOC) { 1865fe386132SJan Kara /* 1866fe386132SJan Kara * For memory cleaning there's no point in writing only 1867fe386132SJan Kara * some buffers. So just bail out. Warn if we came here 1868fe386132SJan Kara * from direct reclaim. 1869fe386132SJan Kara */ 1870fe386132SJan Kara WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) 1871fe386132SJan Kara == PF_MEMALLOC); 187261628a3fSMingming Cao unlock_page(page); 187361628a3fSMingming Cao return 0; 187461628a3fSMingming Cao } 1875f0e6c985SAneesh Kumar K.V } 187664769240SAlex Tomas 1877cb20d518STheodore Ts'o if (PageChecked(page) && ext4_should_journal_data(inode)) 187843ce1d23SAneesh Kumar K.V /* 187943ce1d23SAneesh Kumar K.V * It's mmapped pagecache. Add buffers and journal it. There 188043ce1d23SAneesh Kumar K.V * doesn't seem much point in redirtying the page here. 188143ce1d23SAneesh Kumar K.V */ 18823f0ca309SWu Fengguang return __ext4_journalled_writepage(page, len); 188343ce1d23SAneesh Kumar K.V 188497a851edSJan Kara ext4_io_submit_init(&io_submit, wbc); 188597a851edSJan Kara io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS); 188697a851edSJan Kara if (!io_submit.io_end) { 188797a851edSJan Kara redirty_page_for_writepage(wbc, page); 188897a851edSJan Kara unlock_page(page); 188997a851edSJan Kara return -ENOMEM; 189097a851edSJan Kara } 189136ade451SJan Kara ret = ext4_bio_write_page(&io_submit, page, len, wbc); 189236ade451SJan Kara ext4_io_submit(&io_submit); 189397a851edSJan Kara /* Drop io_end reference we got from init */ 189497a851edSJan Kara ext4_put_io_end_defer(io_submit.io_end); 189564769240SAlex Tomas return ret; 189664769240SAlex Tomas } 189764769240SAlex Tomas 18985f1132b2SJan Kara static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page) 18995f1132b2SJan Kara { 19005f1132b2SJan Kara int len; 19015f1132b2SJan Kara loff_t size = i_size_read(mpd->inode); 19025f1132b2SJan Kara int err; 19035f1132b2SJan Kara 19045f1132b2SJan Kara BUG_ON(page->index != mpd->first_page); 19055f1132b2SJan Kara if (page->index == size >> PAGE_CACHE_SHIFT) 19065f1132b2SJan Kara len = size & ~PAGE_CACHE_MASK; 19075f1132b2SJan Kara else 19085f1132b2SJan Kara len = PAGE_CACHE_SIZE; 19095f1132b2SJan Kara clear_page_dirty_for_io(page); 19105f1132b2SJan Kara err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc); 19115f1132b2SJan Kara if (!err) 19125f1132b2SJan Kara mpd->wbc->nr_to_write--; 19135f1132b2SJan Kara mpd->first_page++; 19145f1132b2SJan Kara 19155f1132b2SJan Kara return err; 19165f1132b2SJan Kara } 19175f1132b2SJan Kara 19184e7ea81dSJan Kara #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay)) 19194e7ea81dSJan Kara 192061628a3fSMingming Cao /* 1921fffb2739SJan Kara * mballoc gives us at most this number of blocks... 1922fffb2739SJan Kara * XXX: That seems to be only a limitation of ext4_mb_normalize_request(). 1923fffb2739SJan Kara * The rest of mballoc seems to handle chunks up to full group size. 192461628a3fSMingming Cao */ 1925fffb2739SJan Kara #define MAX_WRITEPAGES_EXTENT_LEN 2048 1926525f4ed8SMingming Cao 1927525f4ed8SMingming Cao /* 19284e7ea81dSJan Kara * mpage_add_bh_to_extent - try to add bh to extent of blocks to map 19294e7ea81dSJan Kara * 19304e7ea81dSJan Kara * @mpd - extent of blocks 19314e7ea81dSJan Kara * @lblk - logical number of the block in the file 193209930042SJan Kara * @bh - buffer head we want to add to the extent 19334e7ea81dSJan Kara * 193409930042SJan Kara * The function is used to collect contig. blocks in the same state. If the 193509930042SJan Kara * buffer doesn't require mapping for writeback and we haven't started the 193609930042SJan Kara * extent of buffers to map yet, the function returns 'true' immediately - the 193709930042SJan Kara * caller can write the buffer right away. Otherwise the function returns true 193809930042SJan Kara * if the block has been added to the extent, false if the block couldn't be 193909930042SJan Kara * added. 19404e7ea81dSJan Kara */ 194109930042SJan Kara static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk, 194209930042SJan Kara struct buffer_head *bh) 19434e7ea81dSJan Kara { 19444e7ea81dSJan Kara struct ext4_map_blocks *map = &mpd->map; 19454e7ea81dSJan Kara 194609930042SJan Kara /* Buffer that doesn't need mapping for writeback? */ 194709930042SJan Kara if (!buffer_dirty(bh) || !buffer_mapped(bh) || 194809930042SJan Kara (!buffer_delay(bh) && !buffer_unwritten(bh))) { 194909930042SJan Kara /* So far no extent to map => we write the buffer right away */ 195009930042SJan Kara if (map->m_len == 0) 195109930042SJan Kara return true; 195209930042SJan Kara return false; 195309930042SJan Kara } 19544e7ea81dSJan Kara 19554e7ea81dSJan Kara /* First block in the extent? */ 19564e7ea81dSJan Kara if (map->m_len == 0) { 19574e7ea81dSJan Kara map->m_lblk = lblk; 19584e7ea81dSJan Kara map->m_len = 1; 195909930042SJan Kara map->m_flags = bh->b_state & BH_FLAGS; 196009930042SJan Kara return true; 19614e7ea81dSJan Kara } 19624e7ea81dSJan Kara 196309930042SJan Kara /* Don't go larger than mballoc is willing to allocate */ 196409930042SJan Kara if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN) 196509930042SJan Kara return false; 196609930042SJan Kara 19674e7ea81dSJan Kara /* Can we merge the block to our big extent? */ 19684e7ea81dSJan Kara if (lblk == map->m_lblk + map->m_len && 196909930042SJan Kara (bh->b_state & BH_FLAGS) == map->m_flags) { 19704e7ea81dSJan Kara map->m_len++; 197109930042SJan Kara return true; 19724e7ea81dSJan Kara } 197309930042SJan Kara return false; 19744e7ea81dSJan Kara } 19754e7ea81dSJan Kara 19765f1132b2SJan Kara /* 19775f1132b2SJan Kara * mpage_process_page_bufs - submit page buffers for IO or add them to extent 19785f1132b2SJan Kara * 19795f1132b2SJan Kara * @mpd - extent of blocks for mapping 19805f1132b2SJan Kara * @head - the first buffer in the page 19815f1132b2SJan Kara * @bh - buffer we should start processing from 19825f1132b2SJan Kara * @lblk - logical number of the block in the file corresponding to @bh 19835f1132b2SJan Kara * 19845f1132b2SJan Kara * Walk through page buffers from @bh upto @head (exclusive) and either submit 19855f1132b2SJan Kara * the page for IO if all buffers in this page were mapped and there's no 19865f1132b2SJan Kara * accumulated extent of buffers to map or add buffers in the page to the 19875f1132b2SJan Kara * extent of buffers to map. The function returns 1 if the caller can continue 19885f1132b2SJan Kara * by processing the next page, 0 if it should stop adding buffers to the 19895f1132b2SJan Kara * extent to map because we cannot extend it anymore. It can also return value 19905f1132b2SJan Kara * < 0 in case of error during IO submission. 19915f1132b2SJan Kara */ 19925f1132b2SJan Kara static int mpage_process_page_bufs(struct mpage_da_data *mpd, 19934e7ea81dSJan Kara struct buffer_head *head, 19944e7ea81dSJan Kara struct buffer_head *bh, 19954e7ea81dSJan Kara ext4_lblk_t lblk) 19964e7ea81dSJan Kara { 19974e7ea81dSJan Kara struct inode *inode = mpd->inode; 19985f1132b2SJan Kara int err; 19994e7ea81dSJan Kara ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1) 20004e7ea81dSJan Kara >> inode->i_blkbits; 20014e7ea81dSJan Kara 20024e7ea81dSJan Kara do { 20034e7ea81dSJan Kara BUG_ON(buffer_locked(bh)); 20044e7ea81dSJan Kara 200509930042SJan Kara if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) { 20064e7ea81dSJan Kara /* Found extent to map? */ 20074e7ea81dSJan Kara if (mpd->map.m_len) 20085f1132b2SJan Kara return 0; 200909930042SJan Kara /* Everything mapped so far and we hit EOF */ 20105f1132b2SJan Kara break; 20114e7ea81dSJan Kara } 20124e7ea81dSJan Kara } while (lblk++, (bh = bh->b_this_page) != head); 20135f1132b2SJan Kara /* So far everything mapped? Submit the page for IO. */ 20145f1132b2SJan Kara if (mpd->map.m_len == 0) { 20155f1132b2SJan Kara err = mpage_submit_page(mpd, head->b_page); 20165f1132b2SJan Kara if (err < 0) 20174e7ea81dSJan Kara return err; 20184e7ea81dSJan Kara } 20195f1132b2SJan Kara return lblk < blocks; 20205f1132b2SJan Kara } 20214e7ea81dSJan Kara 20224e7ea81dSJan Kara /* 20234e7ea81dSJan Kara * mpage_map_buffers - update buffers corresponding to changed extent and 20244e7ea81dSJan Kara * submit fully mapped pages for IO 20254e7ea81dSJan Kara * 20264e7ea81dSJan Kara * @mpd - description of extent to map, on return next extent to map 20274e7ea81dSJan Kara * 20284e7ea81dSJan Kara * Scan buffers corresponding to changed extent (we expect corresponding pages 20294e7ea81dSJan Kara * to be already locked) and update buffer state according to new extent state. 20304e7ea81dSJan Kara * We map delalloc buffers to their physical location, clear unwritten bits, 20314e7ea81dSJan Kara * and mark buffers as uninit when we perform writes to uninitialized extents 20324e7ea81dSJan Kara * and do extent conversion after IO is finished. If the last page is not fully 20334e7ea81dSJan Kara * mapped, we update @map to the next extent in the last page that needs 20344e7ea81dSJan Kara * mapping. Otherwise we submit the page for IO. 20354e7ea81dSJan Kara */ 20364e7ea81dSJan Kara static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd) 20374e7ea81dSJan Kara { 20384e7ea81dSJan Kara struct pagevec pvec; 20394e7ea81dSJan Kara int nr_pages, i; 20404e7ea81dSJan Kara struct inode *inode = mpd->inode; 20414e7ea81dSJan Kara struct buffer_head *head, *bh; 20424e7ea81dSJan Kara int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits; 20434e7ea81dSJan Kara pgoff_t start, end; 20444e7ea81dSJan Kara ext4_lblk_t lblk; 20454e7ea81dSJan Kara sector_t pblock; 20464e7ea81dSJan Kara int err; 20474e7ea81dSJan Kara 20484e7ea81dSJan Kara start = mpd->map.m_lblk >> bpp_bits; 20494e7ea81dSJan Kara end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits; 20504e7ea81dSJan Kara lblk = start << bpp_bits; 20514e7ea81dSJan Kara pblock = mpd->map.m_pblk; 20524e7ea81dSJan Kara 20534e7ea81dSJan Kara pagevec_init(&pvec, 0); 20544e7ea81dSJan Kara while (start <= end) { 20554e7ea81dSJan Kara nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start, 20564e7ea81dSJan Kara PAGEVEC_SIZE); 20574e7ea81dSJan Kara if (nr_pages == 0) 20584e7ea81dSJan Kara break; 20594e7ea81dSJan Kara for (i = 0; i < nr_pages; i++) { 20604e7ea81dSJan Kara struct page *page = pvec.pages[i]; 20614e7ea81dSJan Kara 20624e7ea81dSJan Kara if (page->index > end) 20634e7ea81dSJan Kara break; 20644e7ea81dSJan Kara /* Up to 'end' pages must be contiguous */ 20654e7ea81dSJan Kara BUG_ON(page->index != start); 20664e7ea81dSJan Kara bh = head = page_buffers(page); 20674e7ea81dSJan Kara do { 20684e7ea81dSJan Kara if (lblk < mpd->map.m_lblk) 20694e7ea81dSJan Kara continue; 20704e7ea81dSJan Kara if (lblk >= mpd->map.m_lblk + mpd->map.m_len) { 20714e7ea81dSJan Kara /* 20724e7ea81dSJan Kara * Buffer after end of mapped extent. 20734e7ea81dSJan Kara * Find next buffer in the page to map. 20744e7ea81dSJan Kara */ 20754e7ea81dSJan Kara mpd->map.m_len = 0; 20764e7ea81dSJan Kara mpd->map.m_flags = 0; 20775f1132b2SJan Kara /* 20785f1132b2SJan Kara * FIXME: If dioread_nolock supports 20795f1132b2SJan Kara * blocksize < pagesize, we need to make 20805f1132b2SJan Kara * sure we add size mapped so far to 20815f1132b2SJan Kara * io_end->size as the following call 20825f1132b2SJan Kara * can submit the page for IO. 20835f1132b2SJan Kara */ 20845f1132b2SJan Kara err = mpage_process_page_bufs(mpd, head, 20855f1132b2SJan Kara bh, lblk); 20864e7ea81dSJan Kara pagevec_release(&pvec); 20875f1132b2SJan Kara if (err > 0) 20885f1132b2SJan Kara err = 0; 20895f1132b2SJan Kara return err; 20904e7ea81dSJan Kara } 20914e7ea81dSJan Kara if (buffer_delay(bh)) { 20924e7ea81dSJan Kara clear_buffer_delay(bh); 20934e7ea81dSJan Kara bh->b_blocknr = pblock++; 20944e7ea81dSJan Kara } 20954e7ea81dSJan Kara clear_buffer_unwritten(bh); 20965f1132b2SJan Kara } while (lblk++, (bh = bh->b_this_page) != head); 20974e7ea81dSJan Kara 20984e7ea81dSJan Kara /* 20994e7ea81dSJan Kara * FIXME: This is going to break if dioread_nolock 21004e7ea81dSJan Kara * supports blocksize < pagesize as we will try to 21014e7ea81dSJan Kara * convert potentially unmapped parts of inode. 21024e7ea81dSJan Kara */ 21034e7ea81dSJan Kara mpd->io_submit.io_end->size += PAGE_CACHE_SIZE; 21044e7ea81dSJan Kara /* Page fully mapped - let IO run! */ 21054e7ea81dSJan Kara err = mpage_submit_page(mpd, page); 21064e7ea81dSJan Kara if (err < 0) { 21074e7ea81dSJan Kara pagevec_release(&pvec); 21084e7ea81dSJan Kara return err; 21094e7ea81dSJan Kara } 21104e7ea81dSJan Kara start++; 21114e7ea81dSJan Kara } 21124e7ea81dSJan Kara pagevec_release(&pvec); 21134e7ea81dSJan Kara } 21144e7ea81dSJan Kara /* Extent fully mapped and matches with page boundary. We are done. */ 21154e7ea81dSJan Kara mpd->map.m_len = 0; 21164e7ea81dSJan Kara mpd->map.m_flags = 0; 21174e7ea81dSJan Kara return 0; 21184e7ea81dSJan Kara } 21194e7ea81dSJan Kara 21204e7ea81dSJan Kara static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd) 21214e7ea81dSJan Kara { 21224e7ea81dSJan Kara struct inode *inode = mpd->inode; 21234e7ea81dSJan Kara struct ext4_map_blocks *map = &mpd->map; 21244e7ea81dSJan Kara int get_blocks_flags; 21254e7ea81dSJan Kara int err; 21264e7ea81dSJan Kara 21274e7ea81dSJan Kara trace_ext4_da_write_pages_extent(inode, map); 21284e7ea81dSJan Kara /* 21294e7ea81dSJan Kara * Call ext4_map_blocks() to allocate any delayed allocation blocks, or 21304e7ea81dSJan Kara * to convert an uninitialized extent to be initialized (in the case 21314e7ea81dSJan Kara * where we have written into one or more preallocated blocks). It is 21324e7ea81dSJan Kara * possible that we're going to need more metadata blocks than 21334e7ea81dSJan Kara * previously reserved. However we must not fail because we're in 21344e7ea81dSJan Kara * writeback and there is nothing we can do about it so it might result 21354e7ea81dSJan Kara * in data loss. So use reserved blocks to allocate metadata if 21364e7ea81dSJan Kara * possible. 21374e7ea81dSJan Kara * 21384e7ea81dSJan Kara * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if the blocks 21394e7ea81dSJan Kara * in question are delalloc blocks. This affects functions in many 21404e7ea81dSJan Kara * different parts of the allocation call path. This flag exists 21414e7ea81dSJan Kara * primarily because we don't want to change *many* call functions, so 21424e7ea81dSJan Kara * ext4_map_blocks() will set the EXT4_STATE_DELALLOC_RESERVED flag 21434e7ea81dSJan Kara * once the inode's allocation semaphore is taken. 21444e7ea81dSJan Kara */ 21454e7ea81dSJan Kara get_blocks_flags = EXT4_GET_BLOCKS_CREATE | 21464e7ea81dSJan Kara EXT4_GET_BLOCKS_METADATA_NOFAIL; 21474e7ea81dSJan Kara if (ext4_should_dioread_nolock(inode)) 21484e7ea81dSJan Kara get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; 21494e7ea81dSJan Kara if (map->m_flags & (1 << BH_Delay)) 21504e7ea81dSJan Kara get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; 21514e7ea81dSJan Kara 21524e7ea81dSJan Kara err = ext4_map_blocks(handle, inode, map, get_blocks_flags); 21534e7ea81dSJan Kara if (err < 0) 21544e7ea81dSJan Kara return err; 21556b523df4SJan Kara if (map->m_flags & EXT4_MAP_UNINIT) { 21566b523df4SJan Kara if (!mpd->io_submit.io_end->handle && 21576b523df4SJan Kara ext4_handle_valid(handle)) { 21586b523df4SJan Kara mpd->io_submit.io_end->handle = handle->h_rsv_handle; 21596b523df4SJan Kara handle->h_rsv_handle = NULL; 21606b523df4SJan Kara } 21613613d228SJan Kara ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end); 21626b523df4SJan Kara } 21634e7ea81dSJan Kara 21644e7ea81dSJan Kara BUG_ON(map->m_len == 0); 21654e7ea81dSJan Kara if (map->m_flags & EXT4_MAP_NEW) { 21664e7ea81dSJan Kara struct block_device *bdev = inode->i_sb->s_bdev; 21674e7ea81dSJan Kara int i; 21684e7ea81dSJan Kara 21694e7ea81dSJan Kara for (i = 0; i < map->m_len; i++) 21704e7ea81dSJan Kara unmap_underlying_metadata(bdev, map->m_pblk + i); 21714e7ea81dSJan Kara } 21724e7ea81dSJan Kara return 0; 21734e7ea81dSJan Kara } 21744e7ea81dSJan Kara 21754e7ea81dSJan Kara /* 21764e7ea81dSJan Kara * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length 21774e7ea81dSJan Kara * mpd->len and submit pages underlying it for IO 21784e7ea81dSJan Kara * 21794e7ea81dSJan Kara * @handle - handle for journal operations 21804e7ea81dSJan Kara * @mpd - extent to map 21814e7ea81dSJan Kara * 21824e7ea81dSJan Kara * The function maps extent starting at mpd->lblk of length mpd->len. If it is 21834e7ea81dSJan Kara * delayed, blocks are allocated, if it is unwritten, we may need to convert 21844e7ea81dSJan Kara * them to initialized or split the described range from larger unwritten 21854e7ea81dSJan Kara * extent. Note that we need not map all the described range since allocation 21864e7ea81dSJan Kara * can return less blocks or the range is covered by more unwritten extents. We 21874e7ea81dSJan Kara * cannot map more because we are limited by reserved transaction credits. On 21884e7ea81dSJan Kara * the other hand we always make sure that the last touched page is fully 21894e7ea81dSJan Kara * mapped so that it can be written out (and thus forward progress is 21904e7ea81dSJan Kara * guaranteed). After mapping we submit all mapped pages for IO. 21914e7ea81dSJan Kara */ 21924e7ea81dSJan Kara static int mpage_map_and_submit_extent(handle_t *handle, 2193cb530541STheodore Ts'o struct mpage_da_data *mpd, 2194cb530541STheodore Ts'o bool *give_up_on_write) 21954e7ea81dSJan Kara { 21964e7ea81dSJan Kara struct inode *inode = mpd->inode; 21974e7ea81dSJan Kara struct ext4_map_blocks *map = &mpd->map; 21984e7ea81dSJan Kara int err; 21994e7ea81dSJan Kara loff_t disksize; 22004e7ea81dSJan Kara 22014e7ea81dSJan Kara mpd->io_submit.io_end->offset = 22024e7ea81dSJan Kara ((loff_t)map->m_lblk) << inode->i_blkbits; 220327d7c4edSJan Kara do { 22044e7ea81dSJan Kara err = mpage_map_one_extent(handle, mpd); 22054e7ea81dSJan Kara if (err < 0) { 22064e7ea81dSJan Kara struct super_block *sb = inode->i_sb; 22074e7ea81dSJan Kara 2208cb530541STheodore Ts'o if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED) 2209cb530541STheodore Ts'o goto invalidate_dirty_pages; 22104e7ea81dSJan Kara /* 2211cb530541STheodore Ts'o * Let the uper layers retry transient errors. 2212cb530541STheodore Ts'o * In the case of ENOSPC, if ext4_count_free_blocks() 2213cb530541STheodore Ts'o * is non-zero, a commit should free up blocks. 22144e7ea81dSJan Kara */ 2215cb530541STheodore Ts'o if ((err == -ENOMEM) || 2216cb530541STheodore Ts'o (err == -ENOSPC && ext4_count_free_clusters(sb))) 2217cb530541STheodore Ts'o return err; 22184e7ea81dSJan Kara ext4_msg(sb, KERN_CRIT, 22194e7ea81dSJan Kara "Delayed block allocation failed for " 22204e7ea81dSJan Kara "inode %lu at logical offset %llu with" 22214e7ea81dSJan Kara " max blocks %u with error %d", 22224e7ea81dSJan Kara inode->i_ino, 22234e7ea81dSJan Kara (unsigned long long)map->m_lblk, 2224cb530541STheodore Ts'o (unsigned)map->m_len, -err); 22254e7ea81dSJan Kara ext4_msg(sb, KERN_CRIT, 22264e7ea81dSJan Kara "This should not happen!! Data will " 22274e7ea81dSJan Kara "be lost\n"); 22284e7ea81dSJan Kara if (err == -ENOSPC) 22294e7ea81dSJan Kara ext4_print_free_blocks(inode); 2230cb530541STheodore Ts'o invalidate_dirty_pages: 2231cb530541STheodore Ts'o *give_up_on_write = true; 22324e7ea81dSJan Kara return err; 22334e7ea81dSJan Kara } 22344e7ea81dSJan Kara /* 22354e7ea81dSJan Kara * Update buffer state, submit mapped pages, and get us new 22364e7ea81dSJan Kara * extent to map 22374e7ea81dSJan Kara */ 22384e7ea81dSJan Kara err = mpage_map_and_submit_buffers(mpd); 22394e7ea81dSJan Kara if (err < 0) 22404e7ea81dSJan Kara return err; 224127d7c4edSJan Kara } while (map->m_len); 22424e7ea81dSJan Kara 22434e7ea81dSJan Kara /* Update on-disk size after IO is submitted */ 22444e7ea81dSJan Kara disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT; 22454e7ea81dSJan Kara if (disksize > EXT4_I(inode)->i_disksize) { 22464e7ea81dSJan Kara int err2; 22474e7ea81dSJan Kara 224890e775b7SJan Kara ext4_wb_update_i_disksize(inode, disksize); 22494e7ea81dSJan Kara err2 = ext4_mark_inode_dirty(handle, inode); 22504e7ea81dSJan Kara if (err2) 22514e7ea81dSJan Kara ext4_error(inode->i_sb, 22524e7ea81dSJan Kara "Failed to mark inode %lu dirty", 22534e7ea81dSJan Kara inode->i_ino); 22544e7ea81dSJan Kara if (!err) 22554e7ea81dSJan Kara err = err2; 22564e7ea81dSJan Kara } 22574e7ea81dSJan Kara return err; 22584e7ea81dSJan Kara } 22594e7ea81dSJan Kara 22604e7ea81dSJan Kara /* 2261fffb2739SJan Kara * Calculate the total number of credits to reserve for one writepages 226220970ba6STheodore Ts'o * iteration. This is called from ext4_writepages(). We map an extent of 2263fffb2739SJan Kara * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping 2264fffb2739SJan Kara * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN + 2265fffb2739SJan Kara * bpp - 1 blocks in bpp different extents. 2266525f4ed8SMingming Cao */ 2267fffb2739SJan Kara static int ext4_da_writepages_trans_blocks(struct inode *inode) 2268fffb2739SJan Kara { 2269fffb2739SJan Kara int bpp = ext4_journal_blocks_per_page(inode); 2270525f4ed8SMingming Cao 2271fffb2739SJan Kara return ext4_meta_trans_blocks(inode, 2272fffb2739SJan Kara MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp); 2273525f4ed8SMingming Cao } 227461628a3fSMingming Cao 22758e48dcfbSTheodore Ts'o /* 22764e7ea81dSJan Kara * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages 22774e7ea81dSJan Kara * and underlying extent to map 22784e7ea81dSJan Kara * 22794e7ea81dSJan Kara * @mpd - where to look for pages 22804e7ea81dSJan Kara * 22814e7ea81dSJan Kara * Walk dirty pages in the mapping. If they are fully mapped, submit them for 22824e7ea81dSJan Kara * IO immediately. When we find a page which isn't mapped we start accumulating 22834e7ea81dSJan Kara * extent of buffers underlying these pages that needs mapping (formed by 22844e7ea81dSJan Kara * either delayed or unwritten buffers). We also lock the pages containing 22854e7ea81dSJan Kara * these buffers. The extent found is returned in @mpd structure (starting at 22864e7ea81dSJan Kara * mpd->lblk with length mpd->len blocks). 22874e7ea81dSJan Kara * 22884e7ea81dSJan Kara * Note that this function can attach bios to one io_end structure which are 22894e7ea81dSJan Kara * neither logically nor physically contiguous. Although it may seem as an 22904e7ea81dSJan Kara * unnecessary complication, it is actually inevitable in blocksize < pagesize 22914e7ea81dSJan Kara * case as we need to track IO to all buffers underlying a page in one io_end. 22928e48dcfbSTheodore Ts'o */ 22934e7ea81dSJan Kara static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd) 22948e48dcfbSTheodore Ts'o { 22954e7ea81dSJan Kara struct address_space *mapping = mpd->inode->i_mapping; 22968e48dcfbSTheodore Ts'o struct pagevec pvec; 22974f01b02cSTheodore Ts'o unsigned int nr_pages; 22984e7ea81dSJan Kara pgoff_t index = mpd->first_page; 22994e7ea81dSJan Kara pgoff_t end = mpd->last_page; 23004e7ea81dSJan Kara int tag; 23014e7ea81dSJan Kara int i, err = 0; 23024e7ea81dSJan Kara int blkbits = mpd->inode->i_blkbits; 23034e7ea81dSJan Kara ext4_lblk_t lblk; 23044e7ea81dSJan Kara struct buffer_head *head; 23058e48dcfbSTheodore Ts'o 23064e7ea81dSJan Kara if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages) 23075b41d924SEric Sandeen tag = PAGECACHE_TAG_TOWRITE; 23085b41d924SEric Sandeen else 23095b41d924SEric Sandeen tag = PAGECACHE_TAG_DIRTY; 23105b41d924SEric Sandeen 23114e7ea81dSJan Kara pagevec_init(&pvec, 0); 23124e7ea81dSJan Kara mpd->map.m_len = 0; 23134e7ea81dSJan Kara mpd->next_page = index; 23144f01b02cSTheodore Ts'o while (index <= end) { 23155b41d924SEric Sandeen nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, 23168e48dcfbSTheodore Ts'o min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); 23178e48dcfbSTheodore Ts'o if (nr_pages == 0) 23184e7ea81dSJan Kara goto out; 23198e48dcfbSTheodore Ts'o 23208e48dcfbSTheodore Ts'o for (i = 0; i < nr_pages; i++) { 23218e48dcfbSTheodore Ts'o struct page *page = pvec.pages[i]; 23228e48dcfbSTheodore Ts'o 23238e48dcfbSTheodore Ts'o /* 23248e48dcfbSTheodore Ts'o * At this point, the page may be truncated or 23258e48dcfbSTheodore Ts'o * invalidated (changing page->mapping to NULL), or 23268e48dcfbSTheodore Ts'o * even swizzled back from swapper_space to tmpfs file 23278e48dcfbSTheodore Ts'o * mapping. However, page->index will not change 23288e48dcfbSTheodore Ts'o * because we have a reference on the page. 23298e48dcfbSTheodore Ts'o */ 23304f01b02cSTheodore Ts'o if (page->index > end) 23314f01b02cSTheodore Ts'o goto out; 23328e48dcfbSTheodore Ts'o 23334e7ea81dSJan Kara /* If we can't merge this page, we are done. */ 23344e7ea81dSJan Kara if (mpd->map.m_len > 0 && mpd->next_page != page->index) 23354e7ea81dSJan Kara goto out; 233678aaced3STheodore Ts'o 23378e48dcfbSTheodore Ts'o lock_page(page); 23388e48dcfbSTheodore Ts'o /* 23394e7ea81dSJan Kara * If the page is no longer dirty, or its mapping no 23404e7ea81dSJan Kara * longer corresponds to inode we are writing (which 23414e7ea81dSJan Kara * means it has been truncated or invalidated), or the 23424e7ea81dSJan Kara * page is already under writeback and we are not doing 23434e7ea81dSJan Kara * a data integrity writeback, skip the page 23448e48dcfbSTheodore Ts'o */ 23454f01b02cSTheodore Ts'o if (!PageDirty(page) || 23464f01b02cSTheodore Ts'o (PageWriteback(page) && 23474e7ea81dSJan Kara (mpd->wbc->sync_mode == WB_SYNC_NONE)) || 23484f01b02cSTheodore Ts'o unlikely(page->mapping != mapping)) { 23498e48dcfbSTheodore Ts'o unlock_page(page); 23508e48dcfbSTheodore Ts'o continue; 23518e48dcfbSTheodore Ts'o } 23528e48dcfbSTheodore Ts'o 23538e48dcfbSTheodore Ts'o wait_on_page_writeback(page); 23548e48dcfbSTheodore Ts'o BUG_ON(PageWriteback(page)); 23558e48dcfbSTheodore Ts'o 23564e7ea81dSJan Kara if (mpd->map.m_len == 0) 23578eb9e5ceSTheodore Ts'o mpd->first_page = page->index; 23588eb9e5ceSTheodore Ts'o mpd->next_page = page->index + 1; 2359f8bec370SJan Kara /* Add all dirty buffers to mpd */ 23604e7ea81dSJan Kara lblk = ((ext4_lblk_t)page->index) << 23614e7ea81dSJan Kara (PAGE_CACHE_SHIFT - blkbits); 23628eb9e5ceSTheodore Ts'o head = page_buffers(page); 23635f1132b2SJan Kara err = mpage_process_page_bufs(mpd, head, head, lblk); 23645f1132b2SJan Kara if (err <= 0) 23654e7ea81dSJan Kara goto out; 23665f1132b2SJan Kara err = 0; 23674e7ea81dSJan Kara 23684e7ea81dSJan Kara /* 23694e7ea81dSJan Kara * Accumulated enough dirty pages? This doesn't apply 23704e7ea81dSJan Kara * to WB_SYNC_ALL mode. For integrity sync we have to 23714e7ea81dSJan Kara * keep going because someone may be concurrently 23724e7ea81dSJan Kara * dirtying pages, and we might have synced a lot of 23734e7ea81dSJan Kara * newly appeared dirty pages, but have not synced all 23744e7ea81dSJan Kara * of the old dirty pages. 23754e7ea81dSJan Kara */ 23764e7ea81dSJan Kara if (mpd->wbc->sync_mode == WB_SYNC_NONE && 23774e7ea81dSJan Kara mpd->next_page - mpd->first_page >= 23784e7ea81dSJan Kara mpd->wbc->nr_to_write) 23794e7ea81dSJan Kara goto out; 23808e48dcfbSTheodore Ts'o } 23818e48dcfbSTheodore Ts'o pagevec_release(&pvec); 23828e48dcfbSTheodore Ts'o cond_resched(); 23838e48dcfbSTheodore Ts'o } 23844f01b02cSTheodore Ts'o return 0; 23858eb9e5ceSTheodore Ts'o out: 23868eb9e5ceSTheodore Ts'o pagevec_release(&pvec); 23874e7ea81dSJan Kara return err; 23888e48dcfbSTheodore Ts'o } 23898e48dcfbSTheodore Ts'o 239020970ba6STheodore Ts'o static int __writepage(struct page *page, struct writeback_control *wbc, 239120970ba6STheodore Ts'o void *data) 239220970ba6STheodore Ts'o { 239320970ba6STheodore Ts'o struct address_space *mapping = data; 239420970ba6STheodore Ts'o int ret = ext4_writepage(page, wbc); 239520970ba6STheodore Ts'o mapping_set_error(mapping, ret); 239620970ba6STheodore Ts'o return ret; 239720970ba6STheodore Ts'o } 239820970ba6STheodore Ts'o 239920970ba6STheodore Ts'o static int ext4_writepages(struct address_space *mapping, 240064769240SAlex Tomas struct writeback_control *wbc) 240164769240SAlex Tomas { 24024e7ea81dSJan Kara pgoff_t writeback_index = 0; 24034e7ea81dSJan Kara long nr_to_write = wbc->nr_to_write; 240422208dedSAneesh Kumar K.V int range_whole = 0; 24054e7ea81dSJan Kara int cycled = 1; 240661628a3fSMingming Cao handle_t *handle = NULL; 2407df22291fSAneesh Kumar K.V struct mpage_da_data mpd; 24085e745b04SAneesh Kumar K.V struct inode *inode = mapping->host; 24096b523df4SJan Kara int needed_blocks, rsv_blocks = 0, ret = 0; 24105e745b04SAneesh Kumar K.V struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); 24114e7ea81dSJan Kara bool done; 24121bce63d1SShaohua Li struct blk_plug plug; 2413cb530541STheodore Ts'o bool give_up_on_write = false; 241461628a3fSMingming Cao 241520970ba6STheodore Ts'o trace_ext4_writepages(inode, wbc); 2416ba80b101STheodore Ts'o 241761628a3fSMingming Cao /* 241861628a3fSMingming Cao * No pages to write? This is mainly a kludge to avoid starting 241961628a3fSMingming Cao * a transaction for special inodes like journal inode on last iput() 242061628a3fSMingming Cao * because that could violate lock ordering on umount 242161628a3fSMingming Cao */ 2422a1d6cc56SAneesh Kumar K.V if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 242361628a3fSMingming Cao return 0; 24242a21e37eSTheodore Ts'o 242520970ba6STheodore Ts'o if (ext4_should_journal_data(inode)) { 242620970ba6STheodore Ts'o struct blk_plug plug; 242720970ba6STheodore Ts'o int ret; 242820970ba6STheodore Ts'o 242920970ba6STheodore Ts'o blk_start_plug(&plug); 243020970ba6STheodore Ts'o ret = write_cache_pages(mapping, wbc, __writepage, mapping); 243120970ba6STheodore Ts'o blk_finish_plug(&plug); 243220970ba6STheodore Ts'o return ret; 243320970ba6STheodore Ts'o } 243420970ba6STheodore Ts'o 24352a21e37eSTheodore Ts'o /* 24362a21e37eSTheodore Ts'o * If the filesystem has aborted, it is read-only, so return 24372a21e37eSTheodore Ts'o * right away instead of dumping stack traces later on that 24382a21e37eSTheodore Ts'o * will obscure the real source of the problem. We test 24394ab2f15bSTheodore Ts'o * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because 24402a21e37eSTheodore Ts'o * the latter could be true if the filesystem is mounted 244120970ba6STheodore Ts'o * read-only, and in that case, ext4_writepages should 24422a21e37eSTheodore Ts'o * *never* be called, so if that ever happens, we would want 24432a21e37eSTheodore Ts'o * the stack trace. 24442a21e37eSTheodore Ts'o */ 24454ab2f15bSTheodore Ts'o if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) 24462a21e37eSTheodore Ts'o return -EROFS; 24472a21e37eSTheodore Ts'o 24486b523df4SJan Kara if (ext4_should_dioread_nolock(inode)) { 24496b523df4SJan Kara /* 24506b523df4SJan Kara * We may need to convert up to one extent per block in 24516b523df4SJan Kara * the page and we may dirty the inode. 24526b523df4SJan Kara */ 24536b523df4SJan Kara rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits); 24546b523df4SJan Kara } 24556b523df4SJan Kara 24564e7ea81dSJan Kara /* 24574e7ea81dSJan Kara * If we have inline data and arrive here, it means that 24584e7ea81dSJan Kara * we will soon create the block for the 1st page, so 24594e7ea81dSJan Kara * we'd better clear the inline data here. 24604e7ea81dSJan Kara */ 24614e7ea81dSJan Kara if (ext4_has_inline_data(inode)) { 24624e7ea81dSJan Kara /* Just inode will be modified... */ 24634e7ea81dSJan Kara handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); 24644e7ea81dSJan Kara if (IS_ERR(handle)) { 24654e7ea81dSJan Kara ret = PTR_ERR(handle); 24664e7ea81dSJan Kara goto out_writepages; 24674e7ea81dSJan Kara } 24684e7ea81dSJan Kara BUG_ON(ext4_test_inode_state(inode, 24694e7ea81dSJan Kara EXT4_STATE_MAY_INLINE_DATA)); 24704e7ea81dSJan Kara ext4_destroy_inline_data(handle, inode); 24714e7ea81dSJan Kara ext4_journal_stop(handle); 24724e7ea81dSJan Kara } 24734e7ea81dSJan Kara 247422208dedSAneesh Kumar K.V if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) 247522208dedSAneesh Kumar K.V range_whole = 1; 247661628a3fSMingming Cao 24772acf2c26SAneesh Kumar K.V if (wbc->range_cyclic) { 24784e7ea81dSJan Kara writeback_index = mapping->writeback_index; 24794e7ea81dSJan Kara if (writeback_index) 24802acf2c26SAneesh Kumar K.V cycled = 0; 24814e7ea81dSJan Kara mpd.first_page = writeback_index; 24824e7ea81dSJan Kara mpd.last_page = -1; 24835b41d924SEric Sandeen } else { 24844e7ea81dSJan Kara mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT; 24854e7ea81dSJan Kara mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT; 24865b41d924SEric Sandeen } 2487a1d6cc56SAneesh Kumar K.V 24884e7ea81dSJan Kara mpd.inode = inode; 24894e7ea81dSJan Kara mpd.wbc = wbc; 24904e7ea81dSJan Kara ext4_io_submit_init(&mpd.io_submit, wbc); 24912acf2c26SAneesh Kumar K.V retry: 24926e6938b6SWu Fengguang if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) 24934e7ea81dSJan Kara tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page); 24944e7ea81dSJan Kara done = false; 24951bce63d1SShaohua Li blk_start_plug(&plug); 24964e7ea81dSJan Kara while (!done && mpd.first_page <= mpd.last_page) { 24974e7ea81dSJan Kara /* For each extent of pages we use new io_end */ 24984e7ea81dSJan Kara mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); 24994e7ea81dSJan Kara if (!mpd.io_submit.io_end) { 25004e7ea81dSJan Kara ret = -ENOMEM; 25014e7ea81dSJan Kara break; 25024e7ea81dSJan Kara } 2503a1d6cc56SAneesh Kumar K.V 2504a1d6cc56SAneesh Kumar K.V /* 25054e7ea81dSJan Kara * We have two constraints: We find one extent to map and we 25064e7ea81dSJan Kara * must always write out whole page (makes a difference when 25074e7ea81dSJan Kara * blocksize < pagesize) so that we don't block on IO when we 25084e7ea81dSJan Kara * try to write out the rest of the page. Journalled mode is 25094e7ea81dSJan Kara * not supported by delalloc. 2510a1d6cc56SAneesh Kumar K.V */ 2511a1d6cc56SAneesh Kumar K.V BUG_ON(ext4_should_journal_data(inode)); 2512525f4ed8SMingming Cao needed_blocks = ext4_da_writepages_trans_blocks(inode); 2513a1d6cc56SAneesh Kumar K.V 251461628a3fSMingming Cao /* start a new transaction */ 25156b523df4SJan Kara handle = ext4_journal_start_with_reserve(inode, 25166b523df4SJan Kara EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks); 251761628a3fSMingming Cao if (IS_ERR(handle)) { 251861628a3fSMingming Cao ret = PTR_ERR(handle); 25191693918eSTheodore Ts'o ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " 2520fbe845ddSCurt Wohlgemuth "%ld pages, ino %lu; err %d", __func__, 2521a1d6cc56SAneesh Kumar K.V wbc->nr_to_write, inode->i_ino, ret); 25224e7ea81dSJan Kara /* Release allocated io_end */ 25234e7ea81dSJan Kara ext4_put_io_end(mpd.io_submit.io_end); 25244e7ea81dSJan Kara break; 252561628a3fSMingming Cao } 2526f63e6005STheodore Ts'o 25274e7ea81dSJan Kara trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc); 25284e7ea81dSJan Kara ret = mpage_prepare_extent_to_map(&mpd); 25294e7ea81dSJan Kara if (!ret) { 25304e7ea81dSJan Kara if (mpd.map.m_len) 2531cb530541STheodore Ts'o ret = mpage_map_and_submit_extent(handle, &mpd, 2532cb530541STheodore Ts'o &give_up_on_write); 25334e7ea81dSJan Kara else { 2534f63e6005STheodore Ts'o /* 25354e7ea81dSJan Kara * We scanned the whole range (or exhausted 25364e7ea81dSJan Kara * nr_to_write), submitted what was mapped and 25374e7ea81dSJan Kara * didn't find anything needing mapping. We are 25384e7ea81dSJan Kara * done. 2539f63e6005STheodore Ts'o */ 25404e7ea81dSJan Kara done = true; 2541f63e6005STheodore Ts'o } 25424e7ea81dSJan Kara } 254361628a3fSMingming Cao ext4_journal_stop(handle); 25444e7ea81dSJan Kara /* Submit prepared bio */ 25454e7ea81dSJan Kara ext4_io_submit(&mpd.io_submit); 25464e7ea81dSJan Kara /* Unlock pages we didn't use */ 2547cb530541STheodore Ts'o mpage_release_unused_pages(&mpd, give_up_on_write); 25484e7ea81dSJan Kara /* Drop our io_end reference we got from init */ 25494e7ea81dSJan Kara ext4_put_io_end(mpd.io_submit.io_end); 2550df22291fSAneesh Kumar K.V 25514e7ea81dSJan Kara if (ret == -ENOSPC && sbi->s_journal) { 25524e7ea81dSJan Kara /* 25534e7ea81dSJan Kara * Commit the transaction which would 255422208dedSAneesh Kumar K.V * free blocks released in the transaction 255522208dedSAneesh Kumar K.V * and try again 255622208dedSAneesh Kumar K.V */ 2557df22291fSAneesh Kumar K.V jbd2_journal_force_commit_nested(sbi->s_journal); 255822208dedSAneesh Kumar K.V ret = 0; 25594e7ea81dSJan Kara continue; 25604e7ea81dSJan Kara } 25614e7ea81dSJan Kara /* Fatal error - ENOMEM, EIO... */ 25624e7ea81dSJan Kara if (ret) 256361628a3fSMingming Cao break; 256461628a3fSMingming Cao } 25651bce63d1SShaohua Li blk_finish_plug(&plug); 25664e7ea81dSJan Kara if (!ret && !cycled) { 25672acf2c26SAneesh Kumar K.V cycled = 1; 25684e7ea81dSJan Kara mpd.last_page = writeback_index - 1; 25694e7ea81dSJan Kara mpd.first_page = 0; 25702acf2c26SAneesh Kumar K.V goto retry; 25712acf2c26SAneesh Kumar K.V } 257261628a3fSMingming Cao 257322208dedSAneesh Kumar K.V /* Update index */ 257422208dedSAneesh Kumar K.V if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 257522208dedSAneesh Kumar K.V /* 25764e7ea81dSJan Kara * Set the writeback_index so that range_cyclic 257722208dedSAneesh Kumar K.V * mode will write it back later 257822208dedSAneesh Kumar K.V */ 25794e7ea81dSJan Kara mapping->writeback_index = mpd.first_page; 2580a1d6cc56SAneesh Kumar K.V 258161628a3fSMingming Cao out_writepages: 258220970ba6STheodore Ts'o trace_ext4_writepages_result(inode, wbc, ret, 25834e7ea81dSJan Kara nr_to_write - wbc->nr_to_write); 258461628a3fSMingming Cao return ret; 258564769240SAlex Tomas } 258664769240SAlex Tomas 258779f0be8dSAneesh Kumar K.V static int ext4_nonda_switch(struct super_block *sb) 258879f0be8dSAneesh Kumar K.V { 25895c1ff336SEric Whitney s64 free_clusters, dirty_clusters; 259079f0be8dSAneesh Kumar K.V struct ext4_sb_info *sbi = EXT4_SB(sb); 259179f0be8dSAneesh Kumar K.V 259279f0be8dSAneesh Kumar K.V /* 259379f0be8dSAneesh Kumar K.V * switch to non delalloc mode if we are running low 259479f0be8dSAneesh Kumar K.V * on free block. The free block accounting via percpu 2595179f7ebfSEric Dumazet * counters can get slightly wrong with percpu_counter_batch getting 259679f0be8dSAneesh Kumar K.V * accumulated on each CPU without updating global counters 259779f0be8dSAneesh Kumar K.V * Delalloc need an accurate free block accounting. So switch 259879f0be8dSAneesh Kumar K.V * to non delalloc when we are near to error range. 259979f0be8dSAneesh Kumar K.V */ 26005c1ff336SEric Whitney free_clusters = 26015c1ff336SEric Whitney percpu_counter_read_positive(&sbi->s_freeclusters_counter); 26025c1ff336SEric Whitney dirty_clusters = 26035c1ff336SEric Whitney percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); 260400d4e736STheodore Ts'o /* 260500d4e736STheodore Ts'o * Start pushing delalloc when 1/2 of free blocks are dirty. 260600d4e736STheodore Ts'o */ 26075c1ff336SEric Whitney if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) 260810ee27a0SMiao Xie try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); 260900d4e736STheodore Ts'o 26105c1ff336SEric Whitney if (2 * free_clusters < 3 * dirty_clusters || 26115c1ff336SEric Whitney free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { 261279f0be8dSAneesh Kumar K.V /* 2613c8afb446SEric Sandeen * free block count is less than 150% of dirty blocks 2614c8afb446SEric Sandeen * or free blocks is less than watermark 261579f0be8dSAneesh Kumar K.V */ 261679f0be8dSAneesh Kumar K.V return 1; 261779f0be8dSAneesh Kumar K.V } 261879f0be8dSAneesh Kumar K.V return 0; 261979f0be8dSAneesh Kumar K.V } 262079f0be8dSAneesh Kumar K.V 262164769240SAlex Tomas static int ext4_da_write_begin(struct file *file, struct address_space *mapping, 262264769240SAlex Tomas loff_t pos, unsigned len, unsigned flags, 262364769240SAlex Tomas struct page **pagep, void **fsdata) 262464769240SAlex Tomas { 262572b8ab9dSEric Sandeen int ret, retries = 0; 262664769240SAlex Tomas struct page *page; 262764769240SAlex Tomas pgoff_t index; 262864769240SAlex Tomas struct inode *inode = mapping->host; 262964769240SAlex Tomas handle_t *handle; 263064769240SAlex Tomas 263164769240SAlex Tomas index = pos >> PAGE_CACHE_SHIFT; 263279f0be8dSAneesh Kumar K.V 263379f0be8dSAneesh Kumar K.V if (ext4_nonda_switch(inode->i_sb)) { 263479f0be8dSAneesh Kumar K.V *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; 263579f0be8dSAneesh Kumar K.V return ext4_write_begin(file, mapping, pos, 263679f0be8dSAneesh Kumar K.V len, flags, pagep, fsdata); 263779f0be8dSAneesh Kumar K.V } 263879f0be8dSAneesh Kumar K.V *fsdata = (void *)0; 26399bffad1eSTheodore Ts'o trace_ext4_da_write_begin(inode, pos, len, flags); 26409c3569b5STao Ma 26419c3569b5STao Ma if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { 26429c3569b5STao Ma ret = ext4_da_write_inline_data_begin(mapping, inode, 26439c3569b5STao Ma pos, len, flags, 26449c3569b5STao Ma pagep, fsdata); 26459c3569b5STao Ma if (ret < 0) 264647564bfbSTheodore Ts'o return ret; 264747564bfbSTheodore Ts'o if (ret == 1) 264847564bfbSTheodore Ts'o return 0; 26499c3569b5STao Ma } 26509c3569b5STao Ma 265147564bfbSTheodore Ts'o /* 265247564bfbSTheodore Ts'o * grab_cache_page_write_begin() can take a long time if the 265347564bfbSTheodore Ts'o * system is thrashing due to memory pressure, or if the page 265447564bfbSTheodore Ts'o * is being written back. So grab it first before we start 265547564bfbSTheodore Ts'o * the transaction handle. This also allows us to allocate 265647564bfbSTheodore Ts'o * the page (if needed) without using GFP_NOFS. 265747564bfbSTheodore Ts'o */ 265847564bfbSTheodore Ts'o retry_grab: 265947564bfbSTheodore Ts'o page = grab_cache_page_write_begin(mapping, index, flags); 266047564bfbSTheodore Ts'o if (!page) 266147564bfbSTheodore Ts'o return -ENOMEM; 266247564bfbSTheodore Ts'o unlock_page(page); 266347564bfbSTheodore Ts'o 266464769240SAlex Tomas /* 266564769240SAlex Tomas * With delayed allocation, we don't log the i_disksize update 266664769240SAlex Tomas * if there is delayed block allocation. But we still need 266764769240SAlex Tomas * to journalling the i_disksize update if writes to the end 266864769240SAlex Tomas * of file which has an already mapped buffer. 266964769240SAlex Tomas */ 267047564bfbSTheodore Ts'o retry_journal: 26719924a92aSTheodore Ts'o handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1); 267264769240SAlex Tomas if (IS_ERR(handle)) { 267347564bfbSTheodore Ts'o page_cache_release(page); 267447564bfbSTheodore Ts'o return PTR_ERR(handle); 267564769240SAlex Tomas } 267664769240SAlex Tomas 267747564bfbSTheodore Ts'o lock_page(page); 267847564bfbSTheodore Ts'o if (page->mapping != mapping) { 267947564bfbSTheodore Ts'o /* The page got truncated from under us */ 268047564bfbSTheodore Ts'o unlock_page(page); 268147564bfbSTheodore Ts'o page_cache_release(page); 2682d5a0d4f7SEric Sandeen ext4_journal_stop(handle); 268347564bfbSTheodore Ts'o goto retry_grab; 2684d5a0d4f7SEric Sandeen } 268547564bfbSTheodore Ts'o /* In case writeback began while the page was unlocked */ 26867afe5aa5SDmitry Monakhov wait_for_stable_page(page); 268764769240SAlex Tomas 26886e1db88dSChristoph Hellwig ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); 268964769240SAlex Tomas if (ret < 0) { 269064769240SAlex Tomas unlock_page(page); 269164769240SAlex Tomas ext4_journal_stop(handle); 2692ae4d5372SAneesh Kumar K.V /* 2693ae4d5372SAneesh Kumar K.V * block_write_begin may have instantiated a few blocks 2694ae4d5372SAneesh Kumar K.V * outside i_size. Trim these off again. Don't need 2695ae4d5372SAneesh Kumar K.V * i_size_read because we hold i_mutex. 2696ae4d5372SAneesh Kumar K.V */ 2697ae4d5372SAneesh Kumar K.V if (pos + len > inode->i_size) 2698b9a4207dSJan Kara ext4_truncate_failed_write(inode); 269947564bfbSTheodore Ts'o 270047564bfbSTheodore Ts'o if (ret == -ENOSPC && 270147564bfbSTheodore Ts'o ext4_should_retry_alloc(inode->i_sb, &retries)) 270247564bfbSTheodore Ts'o goto retry_journal; 270347564bfbSTheodore Ts'o 270447564bfbSTheodore Ts'o page_cache_release(page); 270547564bfbSTheodore Ts'o return ret; 270664769240SAlex Tomas } 270764769240SAlex Tomas 270847564bfbSTheodore Ts'o *pagep = page; 270964769240SAlex Tomas return ret; 271064769240SAlex Tomas } 271164769240SAlex Tomas 2712632eaeabSMingming Cao /* 2713632eaeabSMingming Cao * Check if we should update i_disksize 2714632eaeabSMingming Cao * when write to the end of file but not require block allocation 2715632eaeabSMingming Cao */ 2716632eaeabSMingming Cao static int ext4_da_should_update_i_disksize(struct page *page, 2717632eaeabSMingming Cao unsigned long offset) 2718632eaeabSMingming Cao { 2719632eaeabSMingming Cao struct buffer_head *bh; 2720632eaeabSMingming Cao struct inode *inode = page->mapping->host; 2721632eaeabSMingming Cao unsigned int idx; 2722632eaeabSMingming Cao int i; 2723632eaeabSMingming Cao 2724632eaeabSMingming Cao bh = page_buffers(page); 2725632eaeabSMingming Cao idx = offset >> inode->i_blkbits; 2726632eaeabSMingming Cao 2727632eaeabSMingming Cao for (i = 0; i < idx; i++) 2728632eaeabSMingming Cao bh = bh->b_this_page; 2729632eaeabSMingming Cao 273029fa89d0SAneesh Kumar K.V if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) 2731632eaeabSMingming Cao return 0; 2732632eaeabSMingming Cao return 1; 2733632eaeabSMingming Cao } 2734632eaeabSMingming Cao 273564769240SAlex Tomas static int ext4_da_write_end(struct file *file, 273664769240SAlex Tomas struct address_space *mapping, 273764769240SAlex Tomas loff_t pos, unsigned len, unsigned copied, 273864769240SAlex Tomas struct page *page, void *fsdata) 273964769240SAlex Tomas { 274064769240SAlex Tomas struct inode *inode = mapping->host; 274164769240SAlex Tomas int ret = 0, ret2; 274264769240SAlex Tomas handle_t *handle = ext4_journal_current_handle(); 274364769240SAlex Tomas loff_t new_i_size; 2744632eaeabSMingming Cao unsigned long start, end; 274579f0be8dSAneesh Kumar K.V int write_mode = (int)(unsigned long)fsdata; 274679f0be8dSAneesh Kumar K.V 274774d553aaSTheodore Ts'o if (write_mode == FALL_BACK_TO_NONDELALLOC) 274874d553aaSTheodore Ts'o return ext4_write_end(file, mapping, pos, 274979f0be8dSAneesh Kumar K.V len, copied, page, fsdata); 2750632eaeabSMingming Cao 27519bffad1eSTheodore Ts'o trace_ext4_da_write_end(inode, pos, len, copied); 2752632eaeabSMingming Cao start = pos & (PAGE_CACHE_SIZE - 1); 2753632eaeabSMingming Cao end = start + copied - 1; 275464769240SAlex Tomas 275564769240SAlex Tomas /* 275664769240SAlex Tomas * generic_write_end() will run mark_inode_dirty() if i_size 275764769240SAlex Tomas * changes. So let's piggyback the i_disksize mark_inode_dirty 275864769240SAlex Tomas * into that. 275964769240SAlex Tomas */ 276064769240SAlex Tomas new_i_size = pos + copied; 2761ea51d132SAndrea Arcangeli if (copied && new_i_size > EXT4_I(inode)->i_disksize) { 27629c3569b5STao Ma if (ext4_has_inline_data(inode) || 27639c3569b5STao Ma ext4_da_should_update_i_disksize(page, end)) { 2764632eaeabSMingming Cao down_write(&EXT4_I(inode)->i_data_sem); 2765f3b59291STheodore Ts'o if (new_i_size > EXT4_I(inode)->i_disksize) 276664769240SAlex Tomas EXT4_I(inode)->i_disksize = new_i_size; 2767632eaeabSMingming Cao up_write(&EXT4_I(inode)->i_data_sem); 2768cf17fea6SAneesh Kumar K.V /* We need to mark inode dirty even if 2769cf17fea6SAneesh Kumar K.V * new_i_size is less that inode->i_size 2770cf17fea6SAneesh Kumar K.V * bu greater than i_disksize.(hint delalloc) 2771cf17fea6SAneesh Kumar K.V */ 2772cf17fea6SAneesh Kumar K.V ext4_mark_inode_dirty(handle, inode); 2773632eaeabSMingming Cao } 2774632eaeabSMingming Cao } 27759c3569b5STao Ma 27769c3569b5STao Ma if (write_mode != CONVERT_INLINE_DATA && 27779c3569b5STao Ma ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && 27789c3569b5STao Ma ext4_has_inline_data(inode)) 27799c3569b5STao Ma ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, 27809c3569b5STao Ma page); 27819c3569b5STao Ma else 278264769240SAlex Tomas ret2 = generic_write_end(file, mapping, pos, len, copied, 278364769240SAlex Tomas page, fsdata); 27849c3569b5STao Ma 278564769240SAlex Tomas copied = ret2; 278664769240SAlex Tomas if (ret2 < 0) 278764769240SAlex Tomas ret = ret2; 278864769240SAlex Tomas ret2 = ext4_journal_stop(handle); 278964769240SAlex Tomas if (!ret) 279064769240SAlex Tomas ret = ret2; 279164769240SAlex Tomas 279264769240SAlex Tomas return ret ? ret : copied; 279364769240SAlex Tomas } 279464769240SAlex Tomas 2795d47992f8SLukas Czerner static void ext4_da_invalidatepage(struct page *page, unsigned int offset, 2796d47992f8SLukas Czerner unsigned int length) 279764769240SAlex Tomas { 279864769240SAlex Tomas /* 279964769240SAlex Tomas * Drop reserved blocks 280064769240SAlex Tomas */ 280164769240SAlex Tomas BUG_ON(!PageLocked(page)); 280264769240SAlex Tomas if (!page_has_buffers(page)) 280364769240SAlex Tomas goto out; 280464769240SAlex Tomas 2805ca99fdd2SLukas Czerner ext4_da_page_release_reservation(page, offset, length); 280664769240SAlex Tomas 280764769240SAlex Tomas out: 2808d47992f8SLukas Czerner ext4_invalidatepage(page, offset, length); 280964769240SAlex Tomas 281064769240SAlex Tomas return; 281164769240SAlex Tomas } 281264769240SAlex Tomas 2813ccd2506bSTheodore Ts'o /* 2814ccd2506bSTheodore Ts'o * Force all delayed allocation blocks to be allocated for a given inode. 2815ccd2506bSTheodore Ts'o */ 2816ccd2506bSTheodore Ts'o int ext4_alloc_da_blocks(struct inode *inode) 2817ccd2506bSTheodore Ts'o { 2818fb40ba0dSTheodore Ts'o trace_ext4_alloc_da_blocks(inode); 2819fb40ba0dSTheodore Ts'o 2820ccd2506bSTheodore Ts'o if (!EXT4_I(inode)->i_reserved_data_blocks && 2821ccd2506bSTheodore Ts'o !EXT4_I(inode)->i_reserved_meta_blocks) 2822ccd2506bSTheodore Ts'o return 0; 2823ccd2506bSTheodore Ts'o 2824ccd2506bSTheodore Ts'o /* 2825ccd2506bSTheodore Ts'o * We do something simple for now. The filemap_flush() will 2826ccd2506bSTheodore Ts'o * also start triggering a write of the data blocks, which is 2827ccd2506bSTheodore Ts'o * not strictly speaking necessary (and for users of 2828ccd2506bSTheodore Ts'o * laptop_mode, not even desirable). However, to do otherwise 2829ccd2506bSTheodore Ts'o * would require replicating code paths in: 2830ccd2506bSTheodore Ts'o * 283120970ba6STheodore Ts'o * ext4_writepages() -> 2832ccd2506bSTheodore Ts'o * write_cache_pages() ---> (via passed in callback function) 2833ccd2506bSTheodore Ts'o * __mpage_da_writepage() --> 2834ccd2506bSTheodore Ts'o * mpage_add_bh_to_extent() 2835ccd2506bSTheodore Ts'o * mpage_da_map_blocks() 2836ccd2506bSTheodore Ts'o * 2837ccd2506bSTheodore Ts'o * The problem is that write_cache_pages(), located in 2838ccd2506bSTheodore Ts'o * mm/page-writeback.c, marks pages clean in preparation for 2839ccd2506bSTheodore Ts'o * doing I/O, which is not desirable if we're not planning on 2840ccd2506bSTheodore Ts'o * doing I/O at all. 2841ccd2506bSTheodore Ts'o * 2842ccd2506bSTheodore Ts'o * We could call write_cache_pages(), and then redirty all of 2843380cf090SWu Fengguang * the pages by calling redirty_page_for_writepage() but that 2844ccd2506bSTheodore Ts'o * would be ugly in the extreme. So instead we would need to 2845ccd2506bSTheodore Ts'o * replicate parts of the code in the above functions, 284625985edcSLucas De Marchi * simplifying them because we wouldn't actually intend to 2847ccd2506bSTheodore Ts'o * write out the pages, but rather only collect contiguous 2848ccd2506bSTheodore Ts'o * logical block extents, call the multi-block allocator, and 2849ccd2506bSTheodore Ts'o * then update the buffer heads with the block allocations. 2850ccd2506bSTheodore Ts'o * 2851ccd2506bSTheodore Ts'o * For now, though, we'll cheat by calling filemap_flush(), 2852ccd2506bSTheodore Ts'o * which will map the blocks, and start the I/O, but not 2853ccd2506bSTheodore Ts'o * actually wait for the I/O to complete. 2854ccd2506bSTheodore Ts'o */ 2855ccd2506bSTheodore Ts'o return filemap_flush(inode->i_mapping); 2856ccd2506bSTheodore Ts'o } 285764769240SAlex Tomas 285864769240SAlex Tomas /* 2859ac27a0ecSDave Kleikamp * bmap() is special. It gets used by applications such as lilo and by 2860ac27a0ecSDave Kleikamp * the swapper to find the on-disk block of a specific piece of data. 2861ac27a0ecSDave Kleikamp * 2862ac27a0ecSDave Kleikamp * Naturally, this is dangerous if the block concerned is still in the 2863617ba13bSMingming Cao * journal. If somebody makes a swapfile on an ext4 data-journaling 2864ac27a0ecSDave Kleikamp * filesystem and enables swap, then they may get a nasty shock when the 2865ac27a0ecSDave Kleikamp * data getting swapped to that swapfile suddenly gets overwritten by 2866ac27a0ecSDave Kleikamp * the original zero's written out previously to the journal and 2867ac27a0ecSDave Kleikamp * awaiting writeback in the kernel's buffer cache. 2868ac27a0ecSDave Kleikamp * 2869ac27a0ecSDave Kleikamp * So, if we see any bmap calls here on a modified, data-journaled file, 2870ac27a0ecSDave Kleikamp * take extra steps to flush any blocks which might be in the cache. 2871ac27a0ecSDave Kleikamp */ 2872617ba13bSMingming Cao static sector_t ext4_bmap(struct address_space *mapping, sector_t block) 2873ac27a0ecSDave Kleikamp { 2874ac27a0ecSDave Kleikamp struct inode *inode = mapping->host; 2875ac27a0ecSDave Kleikamp journal_t *journal; 2876ac27a0ecSDave Kleikamp int err; 2877ac27a0ecSDave Kleikamp 287846c7f254STao Ma /* 287946c7f254STao Ma * We can get here for an inline file via the FIBMAP ioctl 288046c7f254STao Ma */ 288146c7f254STao Ma if (ext4_has_inline_data(inode)) 288246c7f254STao Ma return 0; 288346c7f254STao Ma 288464769240SAlex Tomas if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && 288564769240SAlex Tomas test_opt(inode->i_sb, DELALLOC)) { 288664769240SAlex Tomas /* 288764769240SAlex Tomas * With delalloc we want to sync the file 288864769240SAlex Tomas * so that we can make sure we allocate 288964769240SAlex Tomas * blocks for file 289064769240SAlex Tomas */ 289164769240SAlex Tomas filemap_write_and_wait(mapping); 289264769240SAlex Tomas } 289364769240SAlex Tomas 289419f5fb7aSTheodore Ts'o if (EXT4_JOURNAL(inode) && 289519f5fb7aSTheodore Ts'o ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { 2896ac27a0ecSDave Kleikamp /* 2897ac27a0ecSDave Kleikamp * This is a REALLY heavyweight approach, but the use of 2898ac27a0ecSDave Kleikamp * bmap on dirty files is expected to be extremely rare: 2899ac27a0ecSDave Kleikamp * only if we run lilo or swapon on a freshly made file 2900ac27a0ecSDave Kleikamp * do we expect this to happen. 2901ac27a0ecSDave Kleikamp * 2902ac27a0ecSDave Kleikamp * (bmap requires CAP_SYS_RAWIO so this does not 2903ac27a0ecSDave Kleikamp * represent an unprivileged user DOS attack --- we'd be 2904ac27a0ecSDave Kleikamp * in trouble if mortal users could trigger this path at 2905ac27a0ecSDave Kleikamp * will.) 2906ac27a0ecSDave Kleikamp * 2907617ba13bSMingming Cao * NB. EXT4_STATE_JDATA is not set on files other than 2908ac27a0ecSDave Kleikamp * regular files. If somebody wants to bmap a directory 2909ac27a0ecSDave Kleikamp * or symlink and gets confused because the buffer 2910ac27a0ecSDave Kleikamp * hasn't yet been flushed to disk, they deserve 2911ac27a0ecSDave Kleikamp * everything they get. 2912ac27a0ecSDave Kleikamp */ 2913ac27a0ecSDave Kleikamp 291419f5fb7aSTheodore Ts'o ext4_clear_inode_state(inode, EXT4_STATE_JDATA); 2915617ba13bSMingming Cao journal = EXT4_JOURNAL(inode); 2916dab291afSMingming Cao jbd2_journal_lock_updates(journal); 2917dab291afSMingming Cao err = jbd2_journal_flush(journal); 2918dab291afSMingming Cao jbd2_journal_unlock_updates(journal); 2919ac27a0ecSDave Kleikamp 2920ac27a0ecSDave Kleikamp if (err) 2921ac27a0ecSDave Kleikamp return 0; 2922ac27a0ecSDave Kleikamp } 2923ac27a0ecSDave Kleikamp 2924617ba13bSMingming Cao return generic_block_bmap(mapping, block, ext4_get_block); 2925ac27a0ecSDave Kleikamp } 2926ac27a0ecSDave Kleikamp 2927617ba13bSMingming Cao static int ext4_readpage(struct file *file, struct page *page) 2928ac27a0ecSDave Kleikamp { 292946c7f254STao Ma int ret = -EAGAIN; 293046c7f254STao Ma struct inode *inode = page->mapping->host; 293146c7f254STao Ma 29320562e0baSJiaying Zhang trace_ext4_readpage(page); 293346c7f254STao Ma 293446c7f254STao Ma if (ext4_has_inline_data(inode)) 293546c7f254STao Ma ret = ext4_readpage_inline(inode, page); 293646c7f254STao Ma 293746c7f254STao Ma if (ret == -EAGAIN) 2938617ba13bSMingming Cao return mpage_readpage(page, ext4_get_block); 293946c7f254STao Ma 294046c7f254STao Ma return ret; 2941ac27a0ecSDave Kleikamp } 2942ac27a0ecSDave Kleikamp 2943ac27a0ecSDave Kleikamp static int 2944617ba13bSMingming Cao ext4_readpages(struct file *file, struct address_space *mapping, 2945ac27a0ecSDave Kleikamp struct list_head *pages, unsigned nr_pages) 2946ac27a0ecSDave Kleikamp { 294746c7f254STao Ma struct inode *inode = mapping->host; 294846c7f254STao Ma 294946c7f254STao Ma /* If the file has inline data, no need to do readpages. */ 295046c7f254STao Ma if (ext4_has_inline_data(inode)) 295146c7f254STao Ma return 0; 295246c7f254STao Ma 2953617ba13bSMingming Cao return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); 2954ac27a0ecSDave Kleikamp } 2955ac27a0ecSDave Kleikamp 2956d47992f8SLukas Czerner static void ext4_invalidatepage(struct page *page, unsigned int offset, 2957d47992f8SLukas Czerner unsigned int length) 2958ac27a0ecSDave Kleikamp { 2959ca99fdd2SLukas Czerner trace_ext4_invalidatepage(page, offset, length); 29600562e0baSJiaying Zhang 29614520fb3cSJan Kara /* No journalling happens on data buffers when this function is used */ 29624520fb3cSJan Kara WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); 29634520fb3cSJan Kara 2964ca99fdd2SLukas Czerner block_invalidatepage(page, offset, length); 29654520fb3cSJan Kara } 29664520fb3cSJan Kara 296753e87268SJan Kara static int __ext4_journalled_invalidatepage(struct page *page, 2968ca99fdd2SLukas Czerner unsigned int offset, 2969ca99fdd2SLukas Czerner unsigned int length) 29704520fb3cSJan Kara { 29714520fb3cSJan Kara journal_t *journal = EXT4_JOURNAL(page->mapping->host); 29724520fb3cSJan Kara 2973ca99fdd2SLukas Czerner trace_ext4_journalled_invalidatepage(page, offset, length); 29744520fb3cSJan Kara 2975744692dcSJiaying Zhang /* 2976ac27a0ecSDave Kleikamp * If it's a full truncate we just forget about the pending dirtying 2977ac27a0ecSDave Kleikamp */ 2978ca99fdd2SLukas Czerner if (offset == 0 && length == PAGE_CACHE_SIZE) 2979ac27a0ecSDave Kleikamp ClearPageChecked(page); 2980ac27a0ecSDave Kleikamp 2981ca99fdd2SLukas Czerner return jbd2_journal_invalidatepage(journal, page, offset, length); 298253e87268SJan Kara } 298353e87268SJan Kara 298453e87268SJan Kara /* Wrapper for aops... */ 298553e87268SJan Kara static void ext4_journalled_invalidatepage(struct page *page, 2986d47992f8SLukas Czerner unsigned int offset, 2987d47992f8SLukas Czerner unsigned int length) 298853e87268SJan Kara { 2989ca99fdd2SLukas Czerner WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0); 2990ac27a0ecSDave Kleikamp } 2991ac27a0ecSDave Kleikamp 2992617ba13bSMingming Cao static int ext4_releasepage(struct page *page, gfp_t wait) 2993ac27a0ecSDave Kleikamp { 2994617ba13bSMingming Cao journal_t *journal = EXT4_JOURNAL(page->mapping->host); 2995ac27a0ecSDave Kleikamp 29960562e0baSJiaying Zhang trace_ext4_releasepage(page); 29970562e0baSJiaying Zhang 2998e1c36595SJan Kara /* Page has dirty journalled data -> cannot release */ 2999e1c36595SJan Kara if (PageChecked(page)) 3000ac27a0ecSDave Kleikamp return 0; 30010390131bSFrank Mayhar if (journal) 3002dab291afSMingming Cao return jbd2_journal_try_to_free_buffers(journal, page, wait); 30030390131bSFrank Mayhar else 30040390131bSFrank Mayhar return try_to_free_buffers(page); 3005ac27a0ecSDave Kleikamp } 3006ac27a0ecSDave Kleikamp 3007ac27a0ecSDave Kleikamp /* 30082ed88685STheodore Ts'o * ext4_get_block used when preparing for a DIO write or buffer write. 30092ed88685STheodore Ts'o * We allocate an uinitialized extent if blocks haven't been allocated. 30102ed88685STheodore Ts'o * The extent will be converted to initialized after the IO is complete. 30112ed88685STheodore Ts'o */ 3012f19d5870STao Ma int ext4_get_block_write(struct inode *inode, sector_t iblock, 30134c0425ffSMingming Cao struct buffer_head *bh_result, int create) 30144c0425ffSMingming Cao { 3015c7064ef1SJiaying Zhang ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n", 30168d5d02e6SMingming Cao inode->i_ino, create); 30172ed88685STheodore Ts'o return _ext4_get_block(inode, iblock, bh_result, 30182ed88685STheodore Ts'o EXT4_GET_BLOCKS_IO_CREATE_EXT); 30194c0425ffSMingming Cao } 30204c0425ffSMingming Cao 3021729f52c6SZheng Liu static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, 30228b0f165fSAnatol Pomozov struct buffer_head *bh_result, int create) 3023729f52c6SZheng Liu { 30248b0f165fSAnatol Pomozov ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n", 30258b0f165fSAnatol Pomozov inode->i_ino, create); 30268b0f165fSAnatol Pomozov return _ext4_get_block(inode, iblock, bh_result, 30278b0f165fSAnatol Pomozov EXT4_GET_BLOCKS_NO_LOCK); 3028729f52c6SZheng Liu } 3029729f52c6SZheng Liu 30304c0425ffSMingming Cao static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, 30317b7a8665SChristoph Hellwig ssize_t size, void *private) 30324c0425ffSMingming Cao { 30334c0425ffSMingming Cao ext4_io_end_t *io_end = iocb->private; 30344c0425ffSMingming Cao 303597a851edSJan Kara /* if not async direct IO just return */ 30367b7a8665SChristoph Hellwig if (!io_end) 303797a851edSJan Kara return; 30384b70df18SMingming 30398d5d02e6SMingming Cao ext_debug("ext4_end_io_dio(): io_end 0x%p " 3040ace36ad4SJoe Perches "for inode %lu, iocb 0x%p, offset %llu, size %zd\n", 30418d5d02e6SMingming Cao iocb->private, io_end->inode->i_ino, iocb, offset, 30428d5d02e6SMingming Cao size); 30438d5d02e6SMingming Cao 3044b5a7e970STheodore Ts'o iocb->private = NULL; 30454c0425ffSMingming Cao io_end->offset = offset; 30464c0425ffSMingming Cao io_end->size = size; 30477b7a8665SChristoph Hellwig ext4_put_io_end(io_end); 30484c0425ffSMingming Cao } 3049c7064ef1SJiaying Zhang 30504c0425ffSMingming Cao /* 30514c0425ffSMingming Cao * For ext4 extent files, ext4 will do direct-io write to holes, 30524c0425ffSMingming Cao * preallocated extents, and those write extend the file, no need to 30534c0425ffSMingming Cao * fall back to buffered IO. 30544c0425ffSMingming Cao * 3055b595076aSUwe Kleine-König * For holes, we fallocate those blocks, mark them as uninitialized 305669c499d1STheodore Ts'o * If those blocks were preallocated, we mark sure they are split, but 3057b595076aSUwe Kleine-König * still keep the range to write as uninitialized. 30584c0425ffSMingming Cao * 305969c499d1STheodore Ts'o * The unwritten extents will be converted to written when DIO is completed. 30608d5d02e6SMingming Cao * For async direct IO, since the IO may still pending when return, we 306125985edcSLucas De Marchi * set up an end_io call back function, which will do the conversion 30628d5d02e6SMingming Cao * when async direct IO completed. 30634c0425ffSMingming Cao * 30644c0425ffSMingming Cao * If the O_DIRECT write will extend the file then add this inode to the 30654c0425ffSMingming Cao * orphan list. So recovery will truncate it back to the original size 30664c0425ffSMingming Cao * if the machine crashes during the write. 30674c0425ffSMingming Cao * 30684c0425ffSMingming Cao */ 30694c0425ffSMingming Cao static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb, 30704c0425ffSMingming Cao const struct iovec *iov, loff_t offset, 30714c0425ffSMingming Cao unsigned long nr_segs) 30724c0425ffSMingming Cao { 30734c0425ffSMingming Cao struct file *file = iocb->ki_filp; 30744c0425ffSMingming Cao struct inode *inode = file->f_mapping->host; 30754c0425ffSMingming Cao ssize_t ret; 30764c0425ffSMingming Cao size_t count = iov_length(iov, nr_segs); 3077729f52c6SZheng Liu int overwrite = 0; 30788b0f165fSAnatol Pomozov get_block_t *get_block_func = NULL; 30798b0f165fSAnatol Pomozov int dio_flags = 0; 308069c499d1STheodore Ts'o loff_t final_size = offset + count; 308197a851edSJan Kara ext4_io_end_t *io_end = NULL; 308269c499d1STheodore Ts'o 308369c499d1STheodore Ts'o /* Use the old path for reads and writes beyond i_size. */ 308469c499d1STheodore Ts'o if (rw != WRITE || final_size > inode->i_size) 308569c499d1STheodore Ts'o return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); 3086729f52c6SZheng Liu 30874bd809dbSZheng Liu BUG_ON(iocb->private == NULL); 30884bd809dbSZheng Liu 3089e8340395SJan Kara /* 3090e8340395SJan Kara * Make all waiters for direct IO properly wait also for extent 3091e8340395SJan Kara * conversion. This also disallows race between truncate() and 3092e8340395SJan Kara * overwrite DIO as i_dio_count needs to be incremented under i_mutex. 3093e8340395SJan Kara */ 3094e8340395SJan Kara if (rw == WRITE) 3095e8340395SJan Kara atomic_inc(&inode->i_dio_count); 3096e8340395SJan Kara 30974bd809dbSZheng Liu /* If we do a overwrite dio, i_mutex locking can be released */ 30984bd809dbSZheng Liu overwrite = *((int *)iocb->private); 30994bd809dbSZheng Liu 31004bd809dbSZheng Liu if (overwrite) { 31014bd809dbSZheng Liu down_read(&EXT4_I(inode)->i_data_sem); 31024bd809dbSZheng Liu mutex_unlock(&inode->i_mutex); 31034bd809dbSZheng Liu } 31044bd809dbSZheng Liu 31054c0425ffSMingming Cao /* 31068d5d02e6SMingming Cao * We could direct write to holes and fallocate. 31078d5d02e6SMingming Cao * 310869c499d1STheodore Ts'o * Allocated blocks to fill the hole are marked as 310969c499d1STheodore Ts'o * uninitialized to prevent parallel buffered read to expose 311069c499d1STheodore Ts'o * the stale data before DIO complete the data IO. 31118d5d02e6SMingming Cao * 311269c499d1STheodore Ts'o * As to previously fallocated extents, ext4 get_block will 311369c499d1STheodore Ts'o * just simply mark the buffer mapped but still keep the 311469c499d1STheodore Ts'o * extents uninitialized. 31154c0425ffSMingming Cao * 311669c499d1STheodore Ts'o * For non AIO case, we will convert those unwritten extents 31178d5d02e6SMingming Cao * to written after return back from blockdev_direct_IO. 31184c0425ffSMingming Cao * 311969c499d1STheodore Ts'o * For async DIO, the conversion needs to be deferred when the 312069c499d1STheodore Ts'o * IO is completed. The ext4 end_io callback function will be 312169c499d1STheodore Ts'o * called to take care of the conversion work. Here for async 312269c499d1STheodore Ts'o * case, we allocate an io_end structure to hook to the iocb. 31234c0425ffSMingming Cao */ 31248d5d02e6SMingming Cao iocb->private = NULL; 3125f45ee3a1SDmitry Monakhov ext4_inode_aio_set(inode, NULL); 31268d5d02e6SMingming Cao if (!is_sync_kiocb(iocb)) { 312797a851edSJan Kara io_end = ext4_init_io_end(inode, GFP_NOFS); 31284bd809dbSZheng Liu if (!io_end) { 31294bd809dbSZheng Liu ret = -ENOMEM; 31304bd809dbSZheng Liu goto retake_lock; 31314bd809dbSZheng Liu } 313297a851edSJan Kara /* 313397a851edSJan Kara * Grab reference for DIO. Will be dropped in ext4_end_io_dio() 313497a851edSJan Kara */ 313597a851edSJan Kara iocb->private = ext4_get_io_end(io_end); 31368d5d02e6SMingming Cao /* 313769c499d1STheodore Ts'o * we save the io structure for current async direct 313869c499d1STheodore Ts'o * IO, so that later ext4_map_blocks() could flag the 313969c499d1STheodore Ts'o * io structure whether there is a unwritten extents 314069c499d1STheodore Ts'o * needs to be converted when IO is completed. 31418d5d02e6SMingming Cao */ 3142f45ee3a1SDmitry Monakhov ext4_inode_aio_set(inode, io_end); 31438d5d02e6SMingming Cao } 31448d5d02e6SMingming Cao 31458b0f165fSAnatol Pomozov if (overwrite) { 31468b0f165fSAnatol Pomozov get_block_func = ext4_get_block_write_nolock; 31478b0f165fSAnatol Pomozov } else { 31488b0f165fSAnatol Pomozov get_block_func = ext4_get_block_write; 31498b0f165fSAnatol Pomozov dio_flags = DIO_LOCKING; 31508b0f165fSAnatol Pomozov } 3151729f52c6SZheng Liu ret = __blockdev_direct_IO(rw, iocb, inode, 3152729f52c6SZheng Liu inode->i_sb->s_bdev, iov, 3153729f52c6SZheng Liu offset, nr_segs, 31548b0f165fSAnatol Pomozov get_block_func, 3155729f52c6SZheng Liu ext4_end_io_dio, 3156729f52c6SZheng Liu NULL, 31578b0f165fSAnatol Pomozov dio_flags); 31588b0f165fSAnatol Pomozov 31594eec708dSJan Kara /* 316097a851edSJan Kara * Put our reference to io_end. This can free the io_end structure e.g. 316197a851edSJan Kara * in sync IO case or in case of error. It can even perform extent 316297a851edSJan Kara * conversion if all bios we submitted finished before we got here. 316397a851edSJan Kara * Note that in that case iocb->private can be already set to NULL 316497a851edSJan Kara * here. 31654eec708dSJan Kara */ 316697a851edSJan Kara if (io_end) { 316797a851edSJan Kara ext4_inode_aio_set(inode, NULL); 316897a851edSJan Kara ext4_put_io_end(io_end); 316997a851edSJan Kara /* 317097a851edSJan Kara * When no IO was submitted ext4_end_io_dio() was not 317197a851edSJan Kara * called so we have to put iocb's reference. 317297a851edSJan Kara */ 317397a851edSJan Kara if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) { 317497a851edSJan Kara WARN_ON(iocb->private != io_end); 317597a851edSJan Kara WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN); 317697a851edSJan Kara ext4_put_io_end(io_end); 31778d5d02e6SMingming Cao iocb->private = NULL; 317897a851edSJan Kara } 317997a851edSJan Kara } 318097a851edSJan Kara if (ret > 0 && !overwrite && ext4_test_inode_state(inode, 31815f524950SMingming EXT4_STATE_DIO_UNWRITTEN)) { 3182109f5565SMingming int err; 31838d5d02e6SMingming Cao /* 31848d5d02e6SMingming Cao * for non AIO case, since the IO is already 318525985edcSLucas De Marchi * completed, we could do the conversion right here 31868d5d02e6SMingming Cao */ 31876b523df4SJan Kara err = ext4_convert_unwritten_extents(NULL, inode, 31888d5d02e6SMingming Cao offset, ret); 3189109f5565SMingming if (err < 0) 3190109f5565SMingming ret = err; 319119f5fb7aSTheodore Ts'o ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); 3192109f5565SMingming } 31934bd809dbSZheng Liu 31944bd809dbSZheng Liu retake_lock: 3195e8340395SJan Kara if (rw == WRITE) 3196e8340395SJan Kara inode_dio_done(inode); 31974bd809dbSZheng Liu /* take i_mutex locking again if we do a ovewrite dio */ 31984bd809dbSZheng Liu if (overwrite) { 31994bd809dbSZheng Liu up_read(&EXT4_I(inode)->i_data_sem); 32004bd809dbSZheng Liu mutex_lock(&inode->i_mutex); 32014bd809dbSZheng Liu } 32024bd809dbSZheng Liu 32034c0425ffSMingming Cao return ret; 32044c0425ffSMingming Cao } 32058d5d02e6SMingming Cao 32064c0425ffSMingming Cao static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, 32074c0425ffSMingming Cao const struct iovec *iov, loff_t offset, 32084c0425ffSMingming Cao unsigned long nr_segs) 32094c0425ffSMingming Cao { 32104c0425ffSMingming Cao struct file *file = iocb->ki_filp; 32114c0425ffSMingming Cao struct inode *inode = file->f_mapping->host; 32120562e0baSJiaying Zhang ssize_t ret; 32134c0425ffSMingming Cao 321484ebd795STheodore Ts'o /* 321584ebd795STheodore Ts'o * If we are doing data journalling we don't support O_DIRECT 321684ebd795STheodore Ts'o */ 321784ebd795STheodore Ts'o if (ext4_should_journal_data(inode)) 321884ebd795STheodore Ts'o return 0; 321984ebd795STheodore Ts'o 322046c7f254STao Ma /* Let buffer I/O handle the inline data case. */ 322146c7f254STao Ma if (ext4_has_inline_data(inode)) 322246c7f254STao Ma return 0; 322346c7f254STao Ma 32240562e0baSJiaying Zhang trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw); 322512e9b892SDmitry Monakhov if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 32260562e0baSJiaying Zhang ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs); 32270562e0baSJiaying Zhang else 32280562e0baSJiaying Zhang ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); 32290562e0baSJiaying Zhang trace_ext4_direct_IO_exit(inode, offset, 32300562e0baSJiaying Zhang iov_length(iov, nr_segs), rw, ret); 32310562e0baSJiaying Zhang return ret; 32324c0425ffSMingming Cao } 32334c0425ffSMingming Cao 3234ac27a0ecSDave Kleikamp /* 3235617ba13bSMingming Cao * Pages can be marked dirty completely asynchronously from ext4's journalling 3236ac27a0ecSDave Kleikamp * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do 3237ac27a0ecSDave Kleikamp * much here because ->set_page_dirty is called under VFS locks. The page is 3238ac27a0ecSDave Kleikamp * not necessarily locked. 3239ac27a0ecSDave Kleikamp * 3240ac27a0ecSDave Kleikamp * We cannot just dirty the page and leave attached buffers clean, because the 3241ac27a0ecSDave Kleikamp * buffers' dirty state is "definitive". We cannot just set the buffers dirty 3242ac27a0ecSDave Kleikamp * or jbddirty because all the journalling code will explode. 3243ac27a0ecSDave Kleikamp * 3244ac27a0ecSDave Kleikamp * So what we do is to mark the page "pending dirty" and next time writepage 3245ac27a0ecSDave Kleikamp * is called, propagate that into the buffers appropriately. 3246ac27a0ecSDave Kleikamp */ 3247617ba13bSMingming Cao static int ext4_journalled_set_page_dirty(struct page *page) 3248ac27a0ecSDave Kleikamp { 3249ac27a0ecSDave Kleikamp SetPageChecked(page); 3250ac27a0ecSDave Kleikamp return __set_page_dirty_nobuffers(page); 3251ac27a0ecSDave Kleikamp } 3252ac27a0ecSDave Kleikamp 325374d553aaSTheodore Ts'o static const struct address_space_operations ext4_aops = { 3254617ba13bSMingming Cao .readpage = ext4_readpage, 3255617ba13bSMingming Cao .readpages = ext4_readpages, 325643ce1d23SAneesh Kumar K.V .writepage = ext4_writepage, 325720970ba6STheodore Ts'o .writepages = ext4_writepages, 3258bfc1af65SNick Piggin .write_begin = ext4_write_begin, 325974d553aaSTheodore Ts'o .write_end = ext4_write_end, 3260617ba13bSMingming Cao .bmap = ext4_bmap, 3261617ba13bSMingming Cao .invalidatepage = ext4_invalidatepage, 3262617ba13bSMingming Cao .releasepage = ext4_releasepage, 3263617ba13bSMingming Cao .direct_IO = ext4_direct_IO, 3264ac27a0ecSDave Kleikamp .migratepage = buffer_migrate_page, 32658ab22b9aSHisashi Hifumi .is_partially_uptodate = block_is_partially_uptodate, 3266aa261f54SAndi Kleen .error_remove_page = generic_error_remove_page, 3267ac27a0ecSDave Kleikamp }; 3268ac27a0ecSDave Kleikamp 3269617ba13bSMingming Cao static const struct address_space_operations ext4_journalled_aops = { 3270617ba13bSMingming Cao .readpage = ext4_readpage, 3271617ba13bSMingming Cao .readpages = ext4_readpages, 327243ce1d23SAneesh Kumar K.V .writepage = ext4_writepage, 327320970ba6STheodore Ts'o .writepages = ext4_writepages, 3274bfc1af65SNick Piggin .write_begin = ext4_write_begin, 3275bfc1af65SNick Piggin .write_end = ext4_journalled_write_end, 3276617ba13bSMingming Cao .set_page_dirty = ext4_journalled_set_page_dirty, 3277617ba13bSMingming Cao .bmap = ext4_bmap, 32784520fb3cSJan Kara .invalidatepage = ext4_journalled_invalidatepage, 3279617ba13bSMingming Cao .releasepage = ext4_releasepage, 328084ebd795STheodore Ts'o .direct_IO = ext4_direct_IO, 32818ab22b9aSHisashi Hifumi .is_partially_uptodate = block_is_partially_uptodate, 3282aa261f54SAndi Kleen .error_remove_page = generic_error_remove_page, 3283ac27a0ecSDave Kleikamp }; 3284ac27a0ecSDave Kleikamp 328564769240SAlex Tomas static const struct address_space_operations ext4_da_aops = { 328664769240SAlex Tomas .readpage = ext4_readpage, 328764769240SAlex Tomas .readpages = ext4_readpages, 328843ce1d23SAneesh Kumar K.V .writepage = ext4_writepage, 328920970ba6STheodore Ts'o .writepages = ext4_writepages, 329064769240SAlex Tomas .write_begin = ext4_da_write_begin, 329164769240SAlex Tomas .write_end = ext4_da_write_end, 329264769240SAlex Tomas .bmap = ext4_bmap, 329364769240SAlex Tomas .invalidatepage = ext4_da_invalidatepage, 329464769240SAlex Tomas .releasepage = ext4_releasepage, 329564769240SAlex Tomas .direct_IO = ext4_direct_IO, 329664769240SAlex Tomas .migratepage = buffer_migrate_page, 32978ab22b9aSHisashi Hifumi .is_partially_uptodate = block_is_partially_uptodate, 3298aa261f54SAndi Kleen .error_remove_page = generic_error_remove_page, 329964769240SAlex Tomas }; 330064769240SAlex Tomas 3301617ba13bSMingming Cao void ext4_set_aops(struct inode *inode) 3302ac27a0ecSDave Kleikamp { 33033d2b1582SLukas Czerner switch (ext4_inode_journal_mode(inode)) { 33043d2b1582SLukas Czerner case EXT4_INODE_ORDERED_DATA_MODE: 330574d553aaSTheodore Ts'o ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE); 33063d2b1582SLukas Czerner break; 33073d2b1582SLukas Czerner case EXT4_INODE_WRITEBACK_DATA_MODE: 330874d553aaSTheodore Ts'o ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE); 33093d2b1582SLukas Czerner break; 33103d2b1582SLukas Czerner case EXT4_INODE_JOURNAL_DATA_MODE: 3311617ba13bSMingming Cao inode->i_mapping->a_ops = &ext4_journalled_aops; 331274d553aaSTheodore Ts'o return; 33133d2b1582SLukas Czerner default: 33143d2b1582SLukas Czerner BUG(); 33153d2b1582SLukas Czerner } 331674d553aaSTheodore Ts'o if (test_opt(inode->i_sb, DELALLOC)) 331774d553aaSTheodore Ts'o inode->i_mapping->a_ops = &ext4_da_aops; 331874d553aaSTheodore Ts'o else 331974d553aaSTheodore Ts'o inode->i_mapping->a_ops = &ext4_aops; 3320ac27a0ecSDave Kleikamp } 3321ac27a0ecSDave Kleikamp 3322d863dc36SLukas Czerner /* 3323d863dc36SLukas Czerner * ext4_block_truncate_page() zeroes out a mapping from file offset `from' 3324d863dc36SLukas Czerner * up to the end of the block which corresponds to `from'. 3325d863dc36SLukas Czerner * This required during truncate. We need to physically zero the tail end 3326d863dc36SLukas Czerner * of that block so it doesn't yield old data if the file is later grown. 3327d863dc36SLukas Czerner */ 3328d863dc36SLukas Czerner int ext4_block_truncate_page(handle_t *handle, 3329d863dc36SLukas Czerner struct address_space *mapping, loff_t from) 3330d863dc36SLukas Czerner { 3331d863dc36SLukas Czerner unsigned offset = from & (PAGE_CACHE_SIZE-1); 3332d863dc36SLukas Czerner unsigned length; 3333d863dc36SLukas Czerner unsigned blocksize; 3334d863dc36SLukas Czerner struct inode *inode = mapping->host; 3335d863dc36SLukas Czerner 3336d863dc36SLukas Czerner blocksize = inode->i_sb->s_blocksize; 3337d863dc36SLukas Czerner length = blocksize - (offset & (blocksize - 1)); 3338d863dc36SLukas Czerner 3339d863dc36SLukas Czerner return ext4_block_zero_page_range(handle, mapping, from, length); 3340d863dc36SLukas Czerner } 3341d863dc36SLukas Czerner 3342d863dc36SLukas Czerner /* 3343d863dc36SLukas Czerner * ext4_block_zero_page_range() zeros out a mapping of length 'length' 3344d863dc36SLukas Czerner * starting from file offset 'from'. The range to be zero'd must 3345d863dc36SLukas Czerner * be contained with in one block. If the specified range exceeds 3346d863dc36SLukas Czerner * the end of the block it will be shortened to end of the block 3347d863dc36SLukas Czerner * that cooresponds to 'from' 3348d863dc36SLukas Czerner */ 3349d863dc36SLukas Czerner int ext4_block_zero_page_range(handle_t *handle, 3350d863dc36SLukas Czerner struct address_space *mapping, loff_t from, loff_t length) 3351d863dc36SLukas Czerner { 3352d863dc36SLukas Czerner ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; 3353d863dc36SLukas Czerner unsigned offset = from & (PAGE_CACHE_SIZE-1); 3354d863dc36SLukas Czerner unsigned blocksize, max, pos; 3355d863dc36SLukas Czerner ext4_lblk_t iblock; 3356d863dc36SLukas Czerner struct inode *inode = mapping->host; 3357d863dc36SLukas Czerner struct buffer_head *bh; 3358d863dc36SLukas Czerner struct page *page; 3359d863dc36SLukas Czerner int err = 0; 3360d863dc36SLukas Czerner 3361d863dc36SLukas Czerner page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, 3362d863dc36SLukas Czerner mapping_gfp_mask(mapping) & ~__GFP_FS); 3363d863dc36SLukas Czerner if (!page) 3364d863dc36SLukas Czerner return -ENOMEM; 3365d863dc36SLukas Czerner 3366d863dc36SLukas Czerner blocksize = inode->i_sb->s_blocksize; 3367d863dc36SLukas Czerner max = blocksize - (offset & (blocksize - 1)); 3368d863dc36SLukas Czerner 3369d863dc36SLukas Czerner /* 3370d863dc36SLukas Czerner * correct length if it does not fall between 3371d863dc36SLukas Czerner * 'from' and the end of the block 3372d863dc36SLukas Czerner */ 3373d863dc36SLukas Czerner if (length > max || length < 0) 3374d863dc36SLukas Czerner length = max; 3375d863dc36SLukas Czerner 3376d863dc36SLukas Czerner iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); 3377d863dc36SLukas Czerner 3378d863dc36SLukas Czerner if (!page_has_buffers(page)) 3379d863dc36SLukas Czerner create_empty_buffers(page, blocksize, 0); 3380d863dc36SLukas Czerner 3381d863dc36SLukas Czerner /* Find the buffer that contains "offset" */ 3382d863dc36SLukas Czerner bh = page_buffers(page); 3383d863dc36SLukas Czerner pos = blocksize; 3384d863dc36SLukas Czerner while (offset >= pos) { 3385d863dc36SLukas Czerner bh = bh->b_this_page; 3386d863dc36SLukas Czerner iblock++; 3387d863dc36SLukas Czerner pos += blocksize; 3388d863dc36SLukas Czerner } 3389d863dc36SLukas Czerner if (buffer_freed(bh)) { 3390d863dc36SLukas Czerner BUFFER_TRACE(bh, "freed: skip"); 3391d863dc36SLukas Czerner goto unlock; 3392d863dc36SLukas Czerner } 3393d863dc36SLukas Czerner if (!buffer_mapped(bh)) { 3394d863dc36SLukas Czerner BUFFER_TRACE(bh, "unmapped"); 3395d863dc36SLukas Czerner ext4_get_block(inode, iblock, bh, 0); 3396d863dc36SLukas Czerner /* unmapped? It's a hole - nothing to do */ 3397d863dc36SLukas Czerner if (!buffer_mapped(bh)) { 3398d863dc36SLukas Czerner BUFFER_TRACE(bh, "still unmapped"); 3399d863dc36SLukas Czerner goto unlock; 3400d863dc36SLukas Czerner } 3401d863dc36SLukas Czerner } 3402d863dc36SLukas Czerner 3403d863dc36SLukas Czerner /* Ok, it's mapped. Make sure it's up-to-date */ 3404d863dc36SLukas Czerner if (PageUptodate(page)) 3405d863dc36SLukas Czerner set_buffer_uptodate(bh); 3406d863dc36SLukas Czerner 3407d863dc36SLukas Czerner if (!buffer_uptodate(bh)) { 3408d863dc36SLukas Czerner err = -EIO; 3409d863dc36SLukas Czerner ll_rw_block(READ, 1, &bh); 3410d863dc36SLukas Czerner wait_on_buffer(bh); 3411d863dc36SLukas Czerner /* Uhhuh. Read error. Complain and punt. */ 3412d863dc36SLukas Czerner if (!buffer_uptodate(bh)) 3413d863dc36SLukas Czerner goto unlock; 3414d863dc36SLukas Czerner } 3415d863dc36SLukas Czerner if (ext4_should_journal_data(inode)) { 3416d863dc36SLukas Czerner BUFFER_TRACE(bh, "get write access"); 3417d863dc36SLukas Czerner err = ext4_journal_get_write_access(handle, bh); 3418d863dc36SLukas Czerner if (err) 3419d863dc36SLukas Czerner goto unlock; 3420d863dc36SLukas Czerner } 3421d863dc36SLukas Czerner zero_user(page, offset, length); 3422d863dc36SLukas Czerner BUFFER_TRACE(bh, "zeroed end of block"); 3423d863dc36SLukas Czerner 3424d863dc36SLukas Czerner if (ext4_should_journal_data(inode)) { 3425d863dc36SLukas Czerner err = ext4_handle_dirty_metadata(handle, inode, bh); 34260713ed0cSLukas Czerner } else { 3427353eefd3Sjon ernst err = 0; 3428d863dc36SLukas Czerner mark_buffer_dirty(bh); 34290713ed0cSLukas Czerner if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) 34300713ed0cSLukas Czerner err = ext4_jbd2_file_inode(handle, inode); 34310713ed0cSLukas Czerner } 3432d863dc36SLukas Czerner 3433d863dc36SLukas Czerner unlock: 3434d863dc36SLukas Czerner unlock_page(page); 3435d863dc36SLukas Czerner page_cache_release(page); 3436d863dc36SLukas Czerner return err; 3437d863dc36SLukas Czerner } 3438d863dc36SLukas Czerner 3439a87dd18cSLukas Czerner int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, 3440a87dd18cSLukas Czerner loff_t lstart, loff_t length) 3441a87dd18cSLukas Czerner { 3442a87dd18cSLukas Czerner struct super_block *sb = inode->i_sb; 3443a87dd18cSLukas Czerner struct address_space *mapping = inode->i_mapping; 3444e1be3a92SLukas Czerner unsigned partial_start, partial_end; 3445a87dd18cSLukas Czerner ext4_fsblk_t start, end; 3446a87dd18cSLukas Czerner loff_t byte_end = (lstart + length - 1); 3447a87dd18cSLukas Czerner int err = 0; 3448a87dd18cSLukas Czerner 3449e1be3a92SLukas Czerner partial_start = lstart & (sb->s_blocksize - 1); 3450e1be3a92SLukas Czerner partial_end = byte_end & (sb->s_blocksize - 1); 3451e1be3a92SLukas Czerner 3452a87dd18cSLukas Czerner start = lstart >> sb->s_blocksize_bits; 3453a87dd18cSLukas Czerner end = byte_end >> sb->s_blocksize_bits; 3454a87dd18cSLukas Czerner 3455a87dd18cSLukas Czerner /* Handle partial zero within the single block */ 3456e1be3a92SLukas Czerner if (start == end && 3457e1be3a92SLukas Czerner (partial_start || (partial_end != sb->s_blocksize - 1))) { 3458a87dd18cSLukas Czerner err = ext4_block_zero_page_range(handle, mapping, 3459a87dd18cSLukas Czerner lstart, length); 3460a87dd18cSLukas Czerner return err; 3461a87dd18cSLukas Czerner } 3462a87dd18cSLukas Czerner /* Handle partial zero out on the start of the range */ 3463e1be3a92SLukas Czerner if (partial_start) { 3464a87dd18cSLukas Czerner err = ext4_block_zero_page_range(handle, mapping, 3465a87dd18cSLukas Czerner lstart, sb->s_blocksize); 3466a87dd18cSLukas Czerner if (err) 3467a87dd18cSLukas Czerner return err; 3468a87dd18cSLukas Czerner } 3469a87dd18cSLukas Czerner /* Handle partial zero out on the end of the range */ 3470e1be3a92SLukas Czerner if (partial_end != sb->s_blocksize - 1) 3471a87dd18cSLukas Czerner err = ext4_block_zero_page_range(handle, mapping, 3472e1be3a92SLukas Czerner byte_end - partial_end, 3473e1be3a92SLukas Czerner partial_end + 1); 3474a87dd18cSLukas Czerner return err; 3475a87dd18cSLukas Czerner } 3476a87dd18cSLukas Czerner 347791ef4cafSDuane Griffin int ext4_can_truncate(struct inode *inode) 347891ef4cafSDuane Griffin { 347991ef4cafSDuane Griffin if (S_ISREG(inode->i_mode)) 348091ef4cafSDuane Griffin return 1; 348191ef4cafSDuane Griffin if (S_ISDIR(inode->i_mode)) 348291ef4cafSDuane Griffin return 1; 348391ef4cafSDuane Griffin if (S_ISLNK(inode->i_mode)) 348491ef4cafSDuane Griffin return !ext4_inode_is_fast_symlink(inode); 348591ef4cafSDuane Griffin return 0; 348691ef4cafSDuane Griffin } 348791ef4cafSDuane Griffin 3488ac27a0ecSDave Kleikamp /* 3489a4bb6b64SAllison Henderson * ext4_punch_hole: punches a hole in a file by releaseing the blocks 3490a4bb6b64SAllison Henderson * associated with the given offset and length 3491a4bb6b64SAllison Henderson * 3492a4bb6b64SAllison Henderson * @inode: File inode 3493a4bb6b64SAllison Henderson * @offset: The offset where the hole will begin 3494a4bb6b64SAllison Henderson * @len: The length of the hole 3495a4bb6b64SAllison Henderson * 34964907cb7bSAnatol Pomozov * Returns: 0 on success or negative on failure 3497a4bb6b64SAllison Henderson */ 3498a4bb6b64SAllison Henderson 3499aeb2817aSAshish Sangwan int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length) 3500a4bb6b64SAllison Henderson { 350126a4c0c6STheodore Ts'o struct super_block *sb = inode->i_sb; 350226a4c0c6STheodore Ts'o ext4_lblk_t first_block, stop_block; 350326a4c0c6STheodore Ts'o struct address_space *mapping = inode->i_mapping; 3504a87dd18cSLukas Czerner loff_t first_block_offset, last_block_offset; 350526a4c0c6STheodore Ts'o handle_t *handle; 350626a4c0c6STheodore Ts'o unsigned int credits; 350726a4c0c6STheodore Ts'o int ret = 0; 350826a4c0c6STheodore Ts'o 3509a4bb6b64SAllison Henderson if (!S_ISREG(inode->i_mode)) 351073355192SAllison Henderson return -EOPNOTSUPP; 3511a4bb6b64SAllison Henderson 351226a4c0c6STheodore Ts'o if (EXT4_SB(sb)->s_cluster_ratio > 1) { 3513bab08ab9STheodore Ts'o /* TODO: Add support for bigalloc file systems */ 351473355192SAllison Henderson return -EOPNOTSUPP; 3515bab08ab9STheodore Ts'o } 3516bab08ab9STheodore Ts'o 3517aaddea81SZheng Liu trace_ext4_punch_hole(inode, offset, length); 3518aaddea81SZheng Liu 351926a4c0c6STheodore Ts'o /* 352026a4c0c6STheodore Ts'o * Write out all dirty pages to avoid race conditions 352126a4c0c6STheodore Ts'o * Then release them. 352226a4c0c6STheodore Ts'o */ 352326a4c0c6STheodore Ts'o if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { 352426a4c0c6STheodore Ts'o ret = filemap_write_and_wait_range(mapping, offset, 352526a4c0c6STheodore Ts'o offset + length - 1); 352626a4c0c6STheodore Ts'o if (ret) 352726a4c0c6STheodore Ts'o return ret; 352826a4c0c6STheodore Ts'o } 352926a4c0c6STheodore Ts'o 353026a4c0c6STheodore Ts'o mutex_lock(&inode->i_mutex); 353126a4c0c6STheodore Ts'o /* It's not possible punch hole on append only file */ 353226a4c0c6STheodore Ts'o if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) { 353326a4c0c6STheodore Ts'o ret = -EPERM; 353426a4c0c6STheodore Ts'o goto out_mutex; 353526a4c0c6STheodore Ts'o } 353626a4c0c6STheodore Ts'o if (IS_SWAPFILE(inode)) { 353726a4c0c6STheodore Ts'o ret = -ETXTBSY; 353826a4c0c6STheodore Ts'o goto out_mutex; 353926a4c0c6STheodore Ts'o } 354026a4c0c6STheodore Ts'o 354126a4c0c6STheodore Ts'o /* No need to punch hole beyond i_size */ 354226a4c0c6STheodore Ts'o if (offset >= inode->i_size) 354326a4c0c6STheodore Ts'o goto out_mutex; 354426a4c0c6STheodore Ts'o 354526a4c0c6STheodore Ts'o /* 354626a4c0c6STheodore Ts'o * If the hole extends beyond i_size, set the hole 354726a4c0c6STheodore Ts'o * to end after the page that contains i_size 354826a4c0c6STheodore Ts'o */ 354926a4c0c6STheodore Ts'o if (offset + length > inode->i_size) { 355026a4c0c6STheodore Ts'o length = inode->i_size + 355126a4c0c6STheodore Ts'o PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) - 355226a4c0c6STheodore Ts'o offset; 355326a4c0c6STheodore Ts'o } 355426a4c0c6STheodore Ts'o 3555a361293fSJan Kara if (offset & (sb->s_blocksize - 1) || 3556a361293fSJan Kara (offset + length) & (sb->s_blocksize - 1)) { 3557a361293fSJan Kara /* 3558a361293fSJan Kara * Attach jinode to inode for jbd2 if we do any zeroing of 3559a361293fSJan Kara * partial block 3560a361293fSJan Kara */ 3561a361293fSJan Kara ret = ext4_inode_attach_jinode(inode); 3562a361293fSJan Kara if (ret < 0) 3563a361293fSJan Kara goto out_mutex; 3564a361293fSJan Kara 3565a361293fSJan Kara } 3566a361293fSJan Kara 3567a87dd18cSLukas Czerner first_block_offset = round_up(offset, sb->s_blocksize); 3568a87dd18cSLukas Czerner last_block_offset = round_down((offset + length), sb->s_blocksize) - 1; 356926a4c0c6STheodore Ts'o 3570a87dd18cSLukas Czerner /* Now release the pages and zero block aligned part of pages*/ 3571a87dd18cSLukas Czerner if (last_block_offset > first_block_offset) 3572a87dd18cSLukas Czerner truncate_pagecache_range(inode, first_block_offset, 3573a87dd18cSLukas Czerner last_block_offset); 357426a4c0c6STheodore Ts'o 357526a4c0c6STheodore Ts'o /* Wait all existing dio workers, newcomers will block on i_mutex */ 357626a4c0c6STheodore Ts'o ext4_inode_block_unlocked_dio(inode); 357726a4c0c6STheodore Ts'o inode_dio_wait(inode); 357826a4c0c6STheodore Ts'o 357926a4c0c6STheodore Ts'o if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 358026a4c0c6STheodore Ts'o credits = ext4_writepage_trans_blocks(inode); 358126a4c0c6STheodore Ts'o else 358226a4c0c6STheodore Ts'o credits = ext4_blocks_for_truncate(inode); 358326a4c0c6STheodore Ts'o handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); 358426a4c0c6STheodore Ts'o if (IS_ERR(handle)) { 358526a4c0c6STheodore Ts'o ret = PTR_ERR(handle); 358626a4c0c6STheodore Ts'o ext4_std_error(sb, ret); 358726a4c0c6STheodore Ts'o goto out_dio; 358826a4c0c6STheodore Ts'o } 358926a4c0c6STheodore Ts'o 3590a87dd18cSLukas Czerner ret = ext4_zero_partial_blocks(handle, inode, offset, 3591a87dd18cSLukas Czerner length); 359226a4c0c6STheodore Ts'o if (ret) 359326a4c0c6STheodore Ts'o goto out_stop; 359426a4c0c6STheodore Ts'o 359526a4c0c6STheodore Ts'o first_block = (offset + sb->s_blocksize - 1) >> 359626a4c0c6STheodore Ts'o EXT4_BLOCK_SIZE_BITS(sb); 359726a4c0c6STheodore Ts'o stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); 359826a4c0c6STheodore Ts'o 359926a4c0c6STheodore Ts'o /* If there are no blocks to remove, return now */ 360026a4c0c6STheodore Ts'o if (first_block >= stop_block) 360126a4c0c6STheodore Ts'o goto out_stop; 360226a4c0c6STheodore Ts'o 360326a4c0c6STheodore Ts'o down_write(&EXT4_I(inode)->i_data_sem); 360426a4c0c6STheodore Ts'o ext4_discard_preallocations(inode); 360526a4c0c6STheodore Ts'o 360626a4c0c6STheodore Ts'o ret = ext4_es_remove_extent(inode, first_block, 360726a4c0c6STheodore Ts'o stop_block - first_block); 360826a4c0c6STheodore Ts'o if (ret) { 360926a4c0c6STheodore Ts'o up_write(&EXT4_I(inode)->i_data_sem); 361026a4c0c6STheodore Ts'o goto out_stop; 361126a4c0c6STheodore Ts'o } 361226a4c0c6STheodore Ts'o 361326a4c0c6STheodore Ts'o if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 361426a4c0c6STheodore Ts'o ret = ext4_ext_remove_space(inode, first_block, 361526a4c0c6STheodore Ts'o stop_block - 1); 361626a4c0c6STheodore Ts'o else 361726a4c0c6STheodore Ts'o ret = ext4_free_hole_blocks(handle, inode, first_block, 361826a4c0c6STheodore Ts'o stop_block); 361926a4c0c6STheodore Ts'o 362026a4c0c6STheodore Ts'o ext4_discard_preallocations(inode); 3621819c4920STheodore Ts'o up_write(&EXT4_I(inode)->i_data_sem); 362226a4c0c6STheodore Ts'o if (IS_SYNC(inode)) 362326a4c0c6STheodore Ts'o ext4_handle_sync(handle); 362426a4c0c6STheodore Ts'o inode->i_mtime = inode->i_ctime = ext4_current_time(inode); 362526a4c0c6STheodore Ts'o ext4_mark_inode_dirty(handle, inode); 362626a4c0c6STheodore Ts'o out_stop: 362726a4c0c6STheodore Ts'o ext4_journal_stop(handle); 362826a4c0c6STheodore Ts'o out_dio: 362926a4c0c6STheodore Ts'o ext4_inode_resume_unlocked_dio(inode); 363026a4c0c6STheodore Ts'o out_mutex: 363126a4c0c6STheodore Ts'o mutex_unlock(&inode->i_mutex); 363226a4c0c6STheodore Ts'o return ret; 3633a4bb6b64SAllison Henderson } 3634a4bb6b64SAllison Henderson 3635a361293fSJan Kara int ext4_inode_attach_jinode(struct inode *inode) 3636a361293fSJan Kara { 3637a361293fSJan Kara struct ext4_inode_info *ei = EXT4_I(inode); 3638a361293fSJan Kara struct jbd2_inode *jinode; 3639a361293fSJan Kara 3640a361293fSJan Kara if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal) 3641a361293fSJan Kara return 0; 3642a361293fSJan Kara 3643a361293fSJan Kara jinode = jbd2_alloc_inode(GFP_KERNEL); 3644a361293fSJan Kara spin_lock(&inode->i_lock); 3645a361293fSJan Kara if (!ei->jinode) { 3646a361293fSJan Kara if (!jinode) { 3647a361293fSJan Kara spin_unlock(&inode->i_lock); 3648a361293fSJan Kara return -ENOMEM; 3649a361293fSJan Kara } 3650a361293fSJan Kara ei->jinode = jinode; 3651a361293fSJan Kara jbd2_journal_init_jbd_inode(ei->jinode, inode); 3652a361293fSJan Kara jinode = NULL; 3653a361293fSJan Kara } 3654a361293fSJan Kara spin_unlock(&inode->i_lock); 3655a361293fSJan Kara if (unlikely(jinode != NULL)) 3656a361293fSJan Kara jbd2_free_inode(jinode); 3657a361293fSJan Kara return 0; 3658a361293fSJan Kara } 3659a361293fSJan Kara 3660a4bb6b64SAllison Henderson /* 3661617ba13bSMingming Cao * ext4_truncate() 3662ac27a0ecSDave Kleikamp * 3663617ba13bSMingming Cao * We block out ext4_get_block() block instantiations across the entire 3664617ba13bSMingming Cao * transaction, and VFS/VM ensures that ext4_truncate() cannot run 3665ac27a0ecSDave Kleikamp * simultaneously on behalf of the same inode. 3666ac27a0ecSDave Kleikamp * 366742b2aa86SJustin P. Mattock * As we work through the truncate and commit bits of it to the journal there 3668ac27a0ecSDave Kleikamp * is one core, guiding principle: the file's tree must always be consistent on 3669ac27a0ecSDave Kleikamp * disk. We must be able to restart the truncate after a crash. 3670ac27a0ecSDave Kleikamp * 3671ac27a0ecSDave Kleikamp * The file's tree may be transiently inconsistent in memory (although it 3672ac27a0ecSDave Kleikamp * probably isn't), but whenever we close off and commit a journal transaction, 3673ac27a0ecSDave Kleikamp * the contents of (the filesystem + the journal) must be consistent and 3674ac27a0ecSDave Kleikamp * restartable. It's pretty simple, really: bottom up, right to left (although 3675ac27a0ecSDave Kleikamp * left-to-right works OK too). 3676ac27a0ecSDave Kleikamp * 3677ac27a0ecSDave Kleikamp * Note that at recovery time, journal replay occurs *before* the restart of 3678ac27a0ecSDave Kleikamp * truncate against the orphan inode list. 3679ac27a0ecSDave Kleikamp * 3680ac27a0ecSDave Kleikamp * The committed inode has the new, desired i_size (which is the same as 3681617ba13bSMingming Cao * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see 3682ac27a0ecSDave Kleikamp * that this inode's truncate did not complete and it will again call 3683617ba13bSMingming Cao * ext4_truncate() to have another go. So there will be instantiated blocks 3684617ba13bSMingming Cao * to the right of the truncation point in a crashed ext4 filesystem. But 3685ac27a0ecSDave Kleikamp * that's fine - as long as they are linked from the inode, the post-crash 3686617ba13bSMingming Cao * ext4_truncate() run will find them and release them. 3687ac27a0ecSDave Kleikamp */ 3688617ba13bSMingming Cao void ext4_truncate(struct inode *inode) 3689ac27a0ecSDave Kleikamp { 3690819c4920STheodore Ts'o struct ext4_inode_info *ei = EXT4_I(inode); 3691819c4920STheodore Ts'o unsigned int credits; 3692819c4920STheodore Ts'o handle_t *handle; 3693819c4920STheodore Ts'o struct address_space *mapping = inode->i_mapping; 3694819c4920STheodore Ts'o 369519b5ef61STheodore Ts'o /* 369619b5ef61STheodore Ts'o * There is a possibility that we're either freeing the inode 369719b5ef61STheodore Ts'o * or it completely new indode. In those cases we might not 369819b5ef61STheodore Ts'o * have i_mutex locked because it's not necessary. 369919b5ef61STheodore Ts'o */ 370019b5ef61STheodore Ts'o if (!(inode->i_state & (I_NEW|I_FREEING))) 370119b5ef61STheodore Ts'o WARN_ON(!mutex_is_locked(&inode->i_mutex)); 37020562e0baSJiaying Zhang trace_ext4_truncate_enter(inode); 37030562e0baSJiaying Zhang 370491ef4cafSDuane Griffin if (!ext4_can_truncate(inode)) 3705ac27a0ecSDave Kleikamp return; 3706ac27a0ecSDave Kleikamp 370712e9b892SDmitry Monakhov ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); 3708c8d46e41SJiaying Zhang 37095534fb5bSTheodore Ts'o if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) 371019f5fb7aSTheodore Ts'o ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); 37117d8f9f7dSTheodore Ts'o 3712aef1c851STao Ma if (ext4_has_inline_data(inode)) { 3713aef1c851STao Ma int has_inline = 1; 3714aef1c851STao Ma 3715aef1c851STao Ma ext4_inline_data_truncate(inode, &has_inline); 3716aef1c851STao Ma if (has_inline) 3717aef1c851STao Ma return; 3718aef1c851STao Ma } 3719aef1c851STao Ma 3720a361293fSJan Kara /* If we zero-out tail of the page, we have to create jinode for jbd2 */ 3721a361293fSJan Kara if (inode->i_size & (inode->i_sb->s_blocksize - 1)) { 3722a361293fSJan Kara if (ext4_inode_attach_jinode(inode) < 0) 3723a361293fSJan Kara return; 3724a361293fSJan Kara } 3725a361293fSJan Kara 3726ff9893dcSAmir Goldstein if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 3727819c4920STheodore Ts'o credits = ext4_writepage_trans_blocks(inode); 3728ff9893dcSAmir Goldstein else 3729819c4920STheodore Ts'o credits = ext4_blocks_for_truncate(inode); 3730819c4920STheodore Ts'o 3731819c4920STheodore Ts'o handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); 3732819c4920STheodore Ts'o if (IS_ERR(handle)) { 3733819c4920STheodore Ts'o ext4_std_error(inode->i_sb, PTR_ERR(handle)); 3734819c4920STheodore Ts'o return; 3735819c4920STheodore Ts'o } 3736819c4920STheodore Ts'o 3737eb3544c6SLukas Czerner if (inode->i_size & (inode->i_sb->s_blocksize - 1)) 3738eb3544c6SLukas Czerner ext4_block_truncate_page(handle, mapping, inode->i_size); 3739819c4920STheodore Ts'o 3740819c4920STheodore Ts'o /* 3741819c4920STheodore Ts'o * We add the inode to the orphan list, so that if this 3742819c4920STheodore Ts'o * truncate spans multiple transactions, and we crash, we will 3743819c4920STheodore Ts'o * resume the truncate when the filesystem recovers. It also 3744819c4920STheodore Ts'o * marks the inode dirty, to catch the new size. 3745819c4920STheodore Ts'o * 3746819c4920STheodore Ts'o * Implication: the file must always be in a sane, consistent 3747819c4920STheodore Ts'o * truncatable state while each transaction commits. 3748819c4920STheodore Ts'o */ 3749819c4920STheodore Ts'o if (ext4_orphan_add(handle, inode)) 3750819c4920STheodore Ts'o goto out_stop; 3751819c4920STheodore Ts'o 3752819c4920STheodore Ts'o down_write(&EXT4_I(inode)->i_data_sem); 3753819c4920STheodore Ts'o 3754819c4920STheodore Ts'o ext4_discard_preallocations(inode); 3755819c4920STheodore Ts'o 3756819c4920STheodore Ts'o if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 3757819c4920STheodore Ts'o ext4_ext_truncate(handle, inode); 3758819c4920STheodore Ts'o else 3759819c4920STheodore Ts'o ext4_ind_truncate(handle, inode); 3760819c4920STheodore Ts'o 3761819c4920STheodore Ts'o up_write(&ei->i_data_sem); 3762819c4920STheodore Ts'o 3763819c4920STheodore Ts'o if (IS_SYNC(inode)) 3764819c4920STheodore Ts'o ext4_handle_sync(handle); 3765819c4920STheodore Ts'o 3766819c4920STheodore Ts'o out_stop: 3767819c4920STheodore Ts'o /* 3768819c4920STheodore Ts'o * If this was a simple ftruncate() and the file will remain alive, 3769819c4920STheodore Ts'o * then we need to clear up the orphan record which we created above. 3770819c4920STheodore Ts'o * However, if this was a real unlink then we were called by 3771819c4920STheodore Ts'o * ext4_delete_inode(), and we allow that function to clean up the 3772819c4920STheodore Ts'o * orphan info for us. 3773819c4920STheodore Ts'o */ 3774819c4920STheodore Ts'o if (inode->i_nlink) 3775819c4920STheodore Ts'o ext4_orphan_del(handle, inode); 3776819c4920STheodore Ts'o 3777819c4920STheodore Ts'o inode->i_mtime = inode->i_ctime = ext4_current_time(inode); 3778819c4920STheodore Ts'o ext4_mark_inode_dirty(handle, inode); 3779819c4920STheodore Ts'o ext4_journal_stop(handle); 3780a86c6181SAlex Tomas 37810562e0baSJiaying Zhang trace_ext4_truncate_exit(inode); 3782ac27a0ecSDave Kleikamp } 3783ac27a0ecSDave Kleikamp 3784ac27a0ecSDave Kleikamp /* 3785617ba13bSMingming Cao * ext4_get_inode_loc returns with an extra refcount against the inode's 3786ac27a0ecSDave Kleikamp * underlying buffer_head on success. If 'in_mem' is true, we have all 3787ac27a0ecSDave Kleikamp * data in memory that is needed to recreate the on-disk version of this 3788ac27a0ecSDave Kleikamp * inode. 3789ac27a0ecSDave Kleikamp */ 3790617ba13bSMingming Cao static int __ext4_get_inode_loc(struct inode *inode, 3791617ba13bSMingming Cao struct ext4_iloc *iloc, int in_mem) 3792ac27a0ecSDave Kleikamp { 3793240799cdSTheodore Ts'o struct ext4_group_desc *gdp; 3794ac27a0ecSDave Kleikamp struct buffer_head *bh; 3795240799cdSTheodore Ts'o struct super_block *sb = inode->i_sb; 3796240799cdSTheodore Ts'o ext4_fsblk_t block; 3797240799cdSTheodore Ts'o int inodes_per_block, inode_offset; 3798ac27a0ecSDave Kleikamp 37993a06d778SAneesh Kumar K.V iloc->bh = NULL; 3800240799cdSTheodore Ts'o if (!ext4_valid_inum(sb, inode->i_ino)) 3801ac27a0ecSDave Kleikamp return -EIO; 3802ac27a0ecSDave Kleikamp 3803240799cdSTheodore Ts'o iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); 3804240799cdSTheodore Ts'o gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); 3805240799cdSTheodore Ts'o if (!gdp) 3806240799cdSTheodore Ts'o return -EIO; 3807240799cdSTheodore Ts'o 3808240799cdSTheodore Ts'o /* 3809240799cdSTheodore Ts'o * Figure out the offset within the block group inode table 3810240799cdSTheodore Ts'o */ 381100d09882STao Ma inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; 3812240799cdSTheodore Ts'o inode_offset = ((inode->i_ino - 1) % 3813240799cdSTheodore Ts'o EXT4_INODES_PER_GROUP(sb)); 3814240799cdSTheodore Ts'o block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); 3815240799cdSTheodore Ts'o iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); 3816240799cdSTheodore Ts'o 3817240799cdSTheodore Ts'o bh = sb_getblk(sb, block); 3818aebf0243SWang Shilong if (unlikely(!bh)) 3819860d21e2STheodore Ts'o return -ENOMEM; 3820ac27a0ecSDave Kleikamp if (!buffer_uptodate(bh)) { 3821ac27a0ecSDave Kleikamp lock_buffer(bh); 38229c83a923SHidehiro Kawai 38239c83a923SHidehiro Kawai /* 38249c83a923SHidehiro Kawai * If the buffer has the write error flag, we have failed 38259c83a923SHidehiro Kawai * to write out another inode in the same block. In this 38269c83a923SHidehiro Kawai * case, we don't have to read the block because we may 38279c83a923SHidehiro Kawai * read the old inode data successfully. 38289c83a923SHidehiro Kawai */ 38299c83a923SHidehiro Kawai if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) 38309c83a923SHidehiro Kawai set_buffer_uptodate(bh); 38319c83a923SHidehiro Kawai 3832ac27a0ecSDave Kleikamp if (buffer_uptodate(bh)) { 3833ac27a0ecSDave Kleikamp /* someone brought it uptodate while we waited */ 3834ac27a0ecSDave Kleikamp unlock_buffer(bh); 3835ac27a0ecSDave Kleikamp goto has_buffer; 3836ac27a0ecSDave Kleikamp } 3837ac27a0ecSDave Kleikamp 3838ac27a0ecSDave Kleikamp /* 3839ac27a0ecSDave Kleikamp * If we have all information of the inode in memory and this 3840ac27a0ecSDave Kleikamp * is the only valid inode in the block, we need not read the 3841ac27a0ecSDave Kleikamp * block. 3842ac27a0ecSDave Kleikamp */ 3843ac27a0ecSDave Kleikamp if (in_mem) { 3844ac27a0ecSDave Kleikamp struct buffer_head *bitmap_bh; 3845240799cdSTheodore Ts'o int i, start; 3846ac27a0ecSDave Kleikamp 3847240799cdSTheodore Ts'o start = inode_offset & ~(inodes_per_block - 1); 3848ac27a0ecSDave Kleikamp 3849ac27a0ecSDave Kleikamp /* Is the inode bitmap in cache? */ 3850240799cdSTheodore Ts'o bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); 3851aebf0243SWang Shilong if (unlikely(!bitmap_bh)) 3852ac27a0ecSDave Kleikamp goto make_io; 3853ac27a0ecSDave Kleikamp 3854ac27a0ecSDave Kleikamp /* 3855ac27a0ecSDave Kleikamp * If the inode bitmap isn't in cache then the 3856ac27a0ecSDave Kleikamp * optimisation may end up performing two reads instead 3857ac27a0ecSDave Kleikamp * of one, so skip it. 3858ac27a0ecSDave Kleikamp */ 3859ac27a0ecSDave Kleikamp if (!buffer_uptodate(bitmap_bh)) { 3860ac27a0ecSDave Kleikamp brelse(bitmap_bh); 3861ac27a0ecSDave Kleikamp goto make_io; 3862ac27a0ecSDave Kleikamp } 3863240799cdSTheodore Ts'o for (i = start; i < start + inodes_per_block; i++) { 3864ac27a0ecSDave Kleikamp if (i == inode_offset) 3865ac27a0ecSDave Kleikamp continue; 3866617ba13bSMingming Cao if (ext4_test_bit(i, bitmap_bh->b_data)) 3867ac27a0ecSDave Kleikamp break; 3868ac27a0ecSDave Kleikamp } 3869ac27a0ecSDave Kleikamp brelse(bitmap_bh); 3870240799cdSTheodore Ts'o if (i == start + inodes_per_block) { 3871ac27a0ecSDave Kleikamp /* all other inodes are free, so skip I/O */ 3872ac27a0ecSDave Kleikamp memset(bh->b_data, 0, bh->b_size); 3873ac27a0ecSDave Kleikamp set_buffer_uptodate(bh); 3874ac27a0ecSDave Kleikamp unlock_buffer(bh); 3875ac27a0ecSDave Kleikamp goto has_buffer; 3876ac27a0ecSDave Kleikamp } 3877ac27a0ecSDave Kleikamp } 3878ac27a0ecSDave Kleikamp 3879ac27a0ecSDave Kleikamp make_io: 3880ac27a0ecSDave Kleikamp /* 3881240799cdSTheodore Ts'o * If we need to do any I/O, try to pre-readahead extra 3882240799cdSTheodore Ts'o * blocks from the inode table. 3883240799cdSTheodore Ts'o */ 3884240799cdSTheodore Ts'o if (EXT4_SB(sb)->s_inode_readahead_blks) { 3885240799cdSTheodore Ts'o ext4_fsblk_t b, end, table; 3886240799cdSTheodore Ts'o unsigned num; 38870d606e2cSTheodore Ts'o __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; 3888240799cdSTheodore Ts'o 3889240799cdSTheodore Ts'o table = ext4_inode_table(sb, gdp); 3890b713a5ecSTheodore Ts'o /* s_inode_readahead_blks is always a power of 2 */ 38910d606e2cSTheodore Ts'o b = block & ~((ext4_fsblk_t) ra_blks - 1); 3892240799cdSTheodore Ts'o if (table > b) 3893240799cdSTheodore Ts'o b = table; 38940d606e2cSTheodore Ts'o end = b + ra_blks; 3895240799cdSTheodore Ts'o num = EXT4_INODES_PER_GROUP(sb); 3896feb0ab32SDarrick J. Wong if (ext4_has_group_desc_csum(sb)) 3897560671a0SAneesh Kumar K.V num -= ext4_itable_unused_count(sb, gdp); 3898240799cdSTheodore Ts'o table += num / inodes_per_block; 3899240799cdSTheodore Ts'o if (end > table) 3900240799cdSTheodore Ts'o end = table; 3901240799cdSTheodore Ts'o while (b <= end) 3902240799cdSTheodore Ts'o sb_breadahead(sb, b++); 3903240799cdSTheodore Ts'o } 3904240799cdSTheodore Ts'o 3905240799cdSTheodore Ts'o /* 3906ac27a0ecSDave Kleikamp * There are other valid inodes in the buffer, this inode 3907ac27a0ecSDave Kleikamp * has in-inode xattrs, or we don't have this inode in memory. 3908ac27a0ecSDave Kleikamp * Read the block from disk. 3909ac27a0ecSDave Kleikamp */ 39100562e0baSJiaying Zhang trace_ext4_load_inode(inode); 3911ac27a0ecSDave Kleikamp get_bh(bh); 3912ac27a0ecSDave Kleikamp bh->b_end_io = end_buffer_read_sync; 391365299a3bSChristoph Hellwig submit_bh(READ | REQ_META | REQ_PRIO, bh); 3914ac27a0ecSDave Kleikamp wait_on_buffer(bh); 3915ac27a0ecSDave Kleikamp if (!buffer_uptodate(bh)) { 3916c398eda0STheodore Ts'o EXT4_ERROR_INODE_BLOCK(inode, block, 3917c398eda0STheodore Ts'o "unable to read itable block"); 3918ac27a0ecSDave Kleikamp brelse(bh); 3919ac27a0ecSDave Kleikamp return -EIO; 3920ac27a0ecSDave Kleikamp } 3921ac27a0ecSDave Kleikamp } 3922ac27a0ecSDave Kleikamp has_buffer: 3923ac27a0ecSDave Kleikamp iloc->bh = bh; 3924ac27a0ecSDave Kleikamp return 0; 3925ac27a0ecSDave Kleikamp } 3926ac27a0ecSDave Kleikamp 3927617ba13bSMingming Cao int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) 3928ac27a0ecSDave Kleikamp { 3929ac27a0ecSDave Kleikamp /* We have all inode data except xattrs in memory here. */ 3930617ba13bSMingming Cao return __ext4_get_inode_loc(inode, iloc, 393119f5fb7aSTheodore Ts'o !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); 3932ac27a0ecSDave Kleikamp } 3933ac27a0ecSDave Kleikamp 3934617ba13bSMingming Cao void ext4_set_inode_flags(struct inode *inode) 3935ac27a0ecSDave Kleikamp { 3936617ba13bSMingming Cao unsigned int flags = EXT4_I(inode)->i_flags; 3937ac27a0ecSDave Kleikamp 3938ac27a0ecSDave Kleikamp inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); 3939617ba13bSMingming Cao if (flags & EXT4_SYNC_FL) 3940ac27a0ecSDave Kleikamp inode->i_flags |= S_SYNC; 3941617ba13bSMingming Cao if (flags & EXT4_APPEND_FL) 3942ac27a0ecSDave Kleikamp inode->i_flags |= S_APPEND; 3943617ba13bSMingming Cao if (flags & EXT4_IMMUTABLE_FL) 3944ac27a0ecSDave Kleikamp inode->i_flags |= S_IMMUTABLE; 3945617ba13bSMingming Cao if (flags & EXT4_NOATIME_FL) 3946ac27a0ecSDave Kleikamp inode->i_flags |= S_NOATIME; 3947617ba13bSMingming Cao if (flags & EXT4_DIRSYNC_FL) 3948ac27a0ecSDave Kleikamp inode->i_flags |= S_DIRSYNC; 3949ac27a0ecSDave Kleikamp } 3950ac27a0ecSDave Kleikamp 3951ff9ddf7eSJan Kara /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ 3952ff9ddf7eSJan Kara void ext4_get_inode_flags(struct ext4_inode_info *ei) 3953ff9ddf7eSJan Kara { 395484a8dce2SDmitry Monakhov unsigned int vfs_fl; 395584a8dce2SDmitry Monakhov unsigned long old_fl, new_fl; 3956ff9ddf7eSJan Kara 395784a8dce2SDmitry Monakhov do { 395884a8dce2SDmitry Monakhov vfs_fl = ei->vfs_inode.i_flags; 395984a8dce2SDmitry Monakhov old_fl = ei->i_flags; 396084a8dce2SDmitry Monakhov new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL| 396184a8dce2SDmitry Monakhov EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL| 396284a8dce2SDmitry Monakhov EXT4_DIRSYNC_FL); 396384a8dce2SDmitry Monakhov if (vfs_fl & S_SYNC) 396484a8dce2SDmitry Monakhov new_fl |= EXT4_SYNC_FL; 396584a8dce2SDmitry Monakhov if (vfs_fl & S_APPEND) 396684a8dce2SDmitry Monakhov new_fl |= EXT4_APPEND_FL; 396784a8dce2SDmitry Monakhov if (vfs_fl & S_IMMUTABLE) 396884a8dce2SDmitry Monakhov new_fl |= EXT4_IMMUTABLE_FL; 396984a8dce2SDmitry Monakhov if (vfs_fl & S_NOATIME) 397084a8dce2SDmitry Monakhov new_fl |= EXT4_NOATIME_FL; 397184a8dce2SDmitry Monakhov if (vfs_fl & S_DIRSYNC) 397284a8dce2SDmitry Monakhov new_fl |= EXT4_DIRSYNC_FL; 397384a8dce2SDmitry Monakhov } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl); 3974ff9ddf7eSJan Kara } 3975de9a55b8STheodore Ts'o 39760fc1b451SAneesh Kumar K.V static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, 39770fc1b451SAneesh Kumar K.V struct ext4_inode_info *ei) 39780fc1b451SAneesh Kumar K.V { 39790fc1b451SAneesh Kumar K.V blkcnt_t i_blocks ; 39808180a562SAneesh Kumar K.V struct inode *inode = &(ei->vfs_inode); 39818180a562SAneesh Kumar K.V struct super_block *sb = inode->i_sb; 39820fc1b451SAneesh Kumar K.V 39830fc1b451SAneesh Kumar K.V if (EXT4_HAS_RO_COMPAT_FEATURE(sb, 39840fc1b451SAneesh Kumar K.V EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { 39850fc1b451SAneesh Kumar K.V /* we are using combined 48 bit field */ 39860fc1b451SAneesh Kumar K.V i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | 39870fc1b451SAneesh Kumar K.V le32_to_cpu(raw_inode->i_blocks_lo); 398807a03824STheodore Ts'o if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { 39898180a562SAneesh Kumar K.V /* i_blocks represent file system block size */ 39908180a562SAneesh Kumar K.V return i_blocks << (inode->i_blkbits - 9); 39918180a562SAneesh Kumar K.V } else { 39920fc1b451SAneesh Kumar K.V return i_blocks; 39938180a562SAneesh Kumar K.V } 39940fc1b451SAneesh Kumar K.V } else { 39950fc1b451SAneesh Kumar K.V return le32_to_cpu(raw_inode->i_blocks_lo); 39960fc1b451SAneesh Kumar K.V } 39970fc1b451SAneesh Kumar K.V } 3998ff9ddf7eSJan Kara 3999152a7b0aSTao Ma static inline void ext4_iget_extra_inode(struct inode *inode, 4000152a7b0aSTao Ma struct ext4_inode *raw_inode, 4001152a7b0aSTao Ma struct ext4_inode_info *ei) 4002152a7b0aSTao Ma { 4003152a7b0aSTao Ma __le32 *magic = (void *)raw_inode + 4004152a7b0aSTao Ma EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; 400567cf5b09STao Ma if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { 4006152a7b0aSTao Ma ext4_set_inode_state(inode, EXT4_STATE_XATTR); 400767cf5b09STao Ma ext4_find_inline_data_nolock(inode); 4008f19d5870STao Ma } else 4009f19d5870STao Ma EXT4_I(inode)->i_inline_off = 0; 4010152a7b0aSTao Ma } 4011152a7b0aSTao Ma 40121d1fe1eeSDavid Howells struct inode *ext4_iget(struct super_block *sb, unsigned long ino) 4013ac27a0ecSDave Kleikamp { 4014617ba13bSMingming Cao struct ext4_iloc iloc; 4015617ba13bSMingming Cao struct ext4_inode *raw_inode; 40161d1fe1eeSDavid Howells struct ext4_inode_info *ei; 40171d1fe1eeSDavid Howells struct inode *inode; 4018b436b9beSJan Kara journal_t *journal = EXT4_SB(sb)->s_journal; 40191d1fe1eeSDavid Howells long ret; 4020ac27a0ecSDave Kleikamp int block; 402108cefc7aSEric W. Biederman uid_t i_uid; 402208cefc7aSEric W. Biederman gid_t i_gid; 4023ac27a0ecSDave Kleikamp 40241d1fe1eeSDavid Howells inode = iget_locked(sb, ino); 40251d1fe1eeSDavid Howells if (!inode) 40261d1fe1eeSDavid Howells return ERR_PTR(-ENOMEM); 40271d1fe1eeSDavid Howells if (!(inode->i_state & I_NEW)) 40281d1fe1eeSDavid Howells return inode; 40291d1fe1eeSDavid Howells 40301d1fe1eeSDavid Howells ei = EXT4_I(inode); 40317dc57615SPeter Huewe iloc.bh = NULL; 4032ac27a0ecSDave Kleikamp 40331d1fe1eeSDavid Howells ret = __ext4_get_inode_loc(inode, &iloc, 0); 40341d1fe1eeSDavid Howells if (ret < 0) 4035ac27a0ecSDave Kleikamp goto bad_inode; 4036617ba13bSMingming Cao raw_inode = ext4_raw_inode(&iloc); 4037814525f4SDarrick J. Wong 4038814525f4SDarrick J. Wong if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { 4039814525f4SDarrick J. Wong ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); 4040814525f4SDarrick J. Wong if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > 4041814525f4SDarrick J. Wong EXT4_INODE_SIZE(inode->i_sb)) { 4042814525f4SDarrick J. Wong EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)", 4043814525f4SDarrick J. Wong EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize, 4044814525f4SDarrick J. Wong EXT4_INODE_SIZE(inode->i_sb)); 4045814525f4SDarrick J. Wong ret = -EIO; 4046814525f4SDarrick J. Wong goto bad_inode; 4047814525f4SDarrick J. Wong } 4048814525f4SDarrick J. Wong } else 4049814525f4SDarrick J. Wong ei->i_extra_isize = 0; 4050814525f4SDarrick J. Wong 4051814525f4SDarrick J. Wong /* Precompute checksum seed for inode metadata */ 4052814525f4SDarrick J. Wong if (EXT4_HAS_RO_COMPAT_FEATURE(sb, 4053814525f4SDarrick J. Wong EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) { 4054814525f4SDarrick J. Wong struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 4055814525f4SDarrick J. Wong __u32 csum; 4056814525f4SDarrick J. Wong __le32 inum = cpu_to_le32(inode->i_ino); 4057814525f4SDarrick J. Wong __le32 gen = raw_inode->i_generation; 4058814525f4SDarrick J. Wong csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, 4059814525f4SDarrick J. Wong sizeof(inum)); 4060814525f4SDarrick J. Wong ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, 4061814525f4SDarrick J. Wong sizeof(gen)); 4062814525f4SDarrick J. Wong } 4063814525f4SDarrick J. Wong 4064814525f4SDarrick J. Wong if (!ext4_inode_csum_verify(inode, raw_inode, ei)) { 4065814525f4SDarrick J. Wong EXT4_ERROR_INODE(inode, "checksum invalid"); 4066814525f4SDarrick J. Wong ret = -EIO; 4067814525f4SDarrick J. Wong goto bad_inode; 4068814525f4SDarrick J. Wong } 4069814525f4SDarrick J. Wong 4070ac27a0ecSDave Kleikamp inode->i_mode = le16_to_cpu(raw_inode->i_mode); 407108cefc7aSEric W. Biederman i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); 407208cefc7aSEric W. Biederman i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); 4073ac27a0ecSDave Kleikamp if (!(test_opt(inode->i_sb, NO_UID32))) { 407408cefc7aSEric W. Biederman i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; 407508cefc7aSEric W. Biederman i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; 4076ac27a0ecSDave Kleikamp } 407708cefc7aSEric W. Biederman i_uid_write(inode, i_uid); 407808cefc7aSEric W. Biederman i_gid_write(inode, i_gid); 4079bfe86848SMiklos Szeredi set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); 4080ac27a0ecSDave Kleikamp 4081353eb83cSTheodore Ts'o ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ 408267cf5b09STao Ma ei->i_inline_off = 0; 4083ac27a0ecSDave Kleikamp ei->i_dir_start_lookup = 0; 4084ac27a0ecSDave Kleikamp ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); 4085ac27a0ecSDave Kleikamp /* We now have enough fields to check if the inode was active or not. 4086ac27a0ecSDave Kleikamp * This is needed because nfsd might try to access dead inodes 4087ac27a0ecSDave Kleikamp * the test is that same one that e2fsck uses 4088ac27a0ecSDave Kleikamp * NeilBrown 1999oct15 4089ac27a0ecSDave Kleikamp */ 4090ac27a0ecSDave Kleikamp if (inode->i_nlink == 0) { 4091393d1d1dSDr. Tilmann Bubeck if ((inode->i_mode == 0 || 4092393d1d1dSDr. Tilmann Bubeck !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && 4093393d1d1dSDr. Tilmann Bubeck ino != EXT4_BOOT_LOADER_INO) { 4094ac27a0ecSDave Kleikamp /* this inode is deleted */ 40951d1fe1eeSDavid Howells ret = -ESTALE; 4096ac27a0ecSDave Kleikamp goto bad_inode; 4097ac27a0ecSDave Kleikamp } 4098ac27a0ecSDave Kleikamp /* The only unlinked inodes we let through here have 4099ac27a0ecSDave Kleikamp * valid i_mode and are being read by the orphan 4100ac27a0ecSDave Kleikamp * recovery code: that's fine, we're about to complete 4101393d1d1dSDr. Tilmann Bubeck * the process of deleting those. 4102393d1d1dSDr. Tilmann Bubeck * OR it is the EXT4_BOOT_LOADER_INO which is 4103393d1d1dSDr. Tilmann Bubeck * not initialized on a new filesystem. */ 4104ac27a0ecSDave Kleikamp } 4105ac27a0ecSDave Kleikamp ei->i_flags = le32_to_cpu(raw_inode->i_flags); 41060fc1b451SAneesh Kumar K.V inode->i_blocks = ext4_inode_blocks(raw_inode, ei); 41077973c0c1SAneesh Kumar K.V ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); 4108a9e81742STheodore Ts'o if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) 4109a1ddeb7eSBadari Pulavarty ei->i_file_acl |= 4110a1ddeb7eSBadari Pulavarty ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; 4111a48380f7SAneesh Kumar K.V inode->i_size = ext4_isize(raw_inode); 4112ac27a0ecSDave Kleikamp ei->i_disksize = inode->i_size; 4113a9e7f447SDmitry Monakhov #ifdef CONFIG_QUOTA 4114a9e7f447SDmitry Monakhov ei->i_reserved_quota = 0; 4115a9e7f447SDmitry Monakhov #endif 4116ac27a0ecSDave Kleikamp inode->i_generation = le32_to_cpu(raw_inode->i_generation); 4117ac27a0ecSDave Kleikamp ei->i_block_group = iloc.block_group; 4118a4912123STheodore Ts'o ei->i_last_alloc_group = ~0; 4119ac27a0ecSDave Kleikamp /* 4120ac27a0ecSDave Kleikamp * NOTE! The in-memory inode i_data array is in little-endian order 4121ac27a0ecSDave Kleikamp * even on big-endian machines: we do NOT byteswap the block numbers! 4122ac27a0ecSDave Kleikamp */ 4123617ba13bSMingming Cao for (block = 0; block < EXT4_N_BLOCKS; block++) 4124ac27a0ecSDave Kleikamp ei->i_data[block] = raw_inode->i_block[block]; 4125ac27a0ecSDave Kleikamp INIT_LIST_HEAD(&ei->i_orphan); 4126ac27a0ecSDave Kleikamp 4127b436b9beSJan Kara /* 4128b436b9beSJan Kara * Set transaction id's of transactions that have to be committed 4129b436b9beSJan Kara * to finish f[data]sync. We set them to currently running transaction 4130b436b9beSJan Kara * as we cannot be sure that the inode or some of its metadata isn't 4131b436b9beSJan Kara * part of the transaction - the inode could have been reclaimed and 4132b436b9beSJan Kara * now it is reread from disk. 4133b436b9beSJan Kara */ 4134b436b9beSJan Kara if (journal) { 4135b436b9beSJan Kara transaction_t *transaction; 4136b436b9beSJan Kara tid_t tid; 4137b436b9beSJan Kara 4138a931da6aSTheodore Ts'o read_lock(&journal->j_state_lock); 4139b436b9beSJan Kara if (journal->j_running_transaction) 4140b436b9beSJan Kara transaction = journal->j_running_transaction; 4141b436b9beSJan Kara else 4142b436b9beSJan Kara transaction = journal->j_committing_transaction; 4143b436b9beSJan Kara if (transaction) 4144b436b9beSJan Kara tid = transaction->t_tid; 4145b436b9beSJan Kara else 4146b436b9beSJan Kara tid = journal->j_commit_sequence; 4147a931da6aSTheodore Ts'o read_unlock(&journal->j_state_lock); 4148b436b9beSJan Kara ei->i_sync_tid = tid; 4149b436b9beSJan Kara ei->i_datasync_tid = tid; 4150b436b9beSJan Kara } 4151b436b9beSJan Kara 41520040d987SEric Sandeen if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { 4153ac27a0ecSDave Kleikamp if (ei->i_extra_isize == 0) { 4154ac27a0ecSDave Kleikamp /* The extra space is currently unused. Use it. */ 4155617ba13bSMingming Cao ei->i_extra_isize = sizeof(struct ext4_inode) - 4156617ba13bSMingming Cao EXT4_GOOD_OLD_INODE_SIZE; 4157ac27a0ecSDave Kleikamp } else { 4158152a7b0aSTao Ma ext4_iget_extra_inode(inode, raw_inode, ei); 4159ac27a0ecSDave Kleikamp } 4160814525f4SDarrick J. Wong } 4161ac27a0ecSDave Kleikamp 4162ef7f3835SKalpak Shah EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); 4163ef7f3835SKalpak Shah EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); 4164ef7f3835SKalpak Shah EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); 4165ef7f3835SKalpak Shah EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); 4166ef7f3835SKalpak Shah 416725ec56b5SJean Noel Cordenner inode->i_version = le32_to_cpu(raw_inode->i_disk_version); 416825ec56b5SJean Noel Cordenner if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { 416925ec56b5SJean Noel Cordenner if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) 417025ec56b5SJean Noel Cordenner inode->i_version |= 417125ec56b5SJean Noel Cordenner (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; 417225ec56b5SJean Noel Cordenner } 417325ec56b5SJean Noel Cordenner 4174c4b5a614STheodore Ts'o ret = 0; 4175485c26ecSTheodore Ts'o if (ei->i_file_acl && 41761032988cSTheodore Ts'o !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { 417724676da4STheodore Ts'o EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", 417824676da4STheodore Ts'o ei->i_file_acl); 4179485c26ecSTheodore Ts'o ret = -EIO; 4180485c26ecSTheodore Ts'o goto bad_inode; 4181f19d5870STao Ma } else if (!ext4_has_inline_data(inode)) { 4182f19d5870STao Ma if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { 4183f19d5870STao Ma if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 4184c4b5a614STheodore Ts'o (S_ISLNK(inode->i_mode) && 4185f19d5870STao Ma !ext4_inode_is_fast_symlink(inode)))) 41867a262f7cSAneesh Kumar K.V /* Validate extent which is part of inode */ 41877a262f7cSAneesh Kumar K.V ret = ext4_ext_check_inode(inode); 4188fe2c8191SThiemo Nagel } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 4189fe2c8191SThiemo Nagel (S_ISLNK(inode->i_mode) && 4190fe2c8191SThiemo Nagel !ext4_inode_is_fast_symlink(inode))) { 4191fe2c8191SThiemo Nagel /* Validate block references which are part of inode */ 41921f7d1e77STheodore Ts'o ret = ext4_ind_check_inode(inode); 4193fe2c8191SThiemo Nagel } 4194f19d5870STao Ma } 4195567f3e9aSTheodore Ts'o if (ret) 41967a262f7cSAneesh Kumar K.V goto bad_inode; 41977a262f7cSAneesh Kumar K.V 4198ac27a0ecSDave Kleikamp if (S_ISREG(inode->i_mode)) { 4199617ba13bSMingming Cao inode->i_op = &ext4_file_inode_operations; 4200617ba13bSMingming Cao inode->i_fop = &ext4_file_operations; 4201617ba13bSMingming Cao ext4_set_aops(inode); 4202ac27a0ecSDave Kleikamp } else if (S_ISDIR(inode->i_mode)) { 4203617ba13bSMingming Cao inode->i_op = &ext4_dir_inode_operations; 4204617ba13bSMingming Cao inode->i_fop = &ext4_dir_operations; 4205ac27a0ecSDave Kleikamp } else if (S_ISLNK(inode->i_mode)) { 4206e83c1397SDuane Griffin if (ext4_inode_is_fast_symlink(inode)) { 4207617ba13bSMingming Cao inode->i_op = &ext4_fast_symlink_inode_operations; 4208e83c1397SDuane Griffin nd_terminate_link(ei->i_data, inode->i_size, 4209e83c1397SDuane Griffin sizeof(ei->i_data) - 1); 4210e83c1397SDuane Griffin } else { 4211617ba13bSMingming Cao inode->i_op = &ext4_symlink_inode_operations; 4212617ba13bSMingming Cao ext4_set_aops(inode); 4213ac27a0ecSDave Kleikamp } 4214563bdd61STheodore Ts'o } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || 4215563bdd61STheodore Ts'o S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { 4216617ba13bSMingming Cao inode->i_op = &ext4_special_inode_operations; 4217ac27a0ecSDave Kleikamp if (raw_inode->i_block[0]) 4218ac27a0ecSDave Kleikamp init_special_inode(inode, inode->i_mode, 4219ac27a0ecSDave Kleikamp old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); 4220ac27a0ecSDave Kleikamp else 4221ac27a0ecSDave Kleikamp init_special_inode(inode, inode->i_mode, 4222ac27a0ecSDave Kleikamp new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); 4223393d1d1dSDr. Tilmann Bubeck } else if (ino == EXT4_BOOT_LOADER_INO) { 4224393d1d1dSDr. Tilmann Bubeck make_bad_inode(inode); 4225563bdd61STheodore Ts'o } else { 4226563bdd61STheodore Ts'o ret = -EIO; 422724676da4STheodore Ts'o EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); 4228563bdd61STheodore Ts'o goto bad_inode; 4229ac27a0ecSDave Kleikamp } 4230ac27a0ecSDave Kleikamp brelse(iloc.bh); 4231617ba13bSMingming Cao ext4_set_inode_flags(inode); 42321d1fe1eeSDavid Howells unlock_new_inode(inode); 42331d1fe1eeSDavid Howells return inode; 4234ac27a0ecSDave Kleikamp 4235ac27a0ecSDave Kleikamp bad_inode: 4236567f3e9aSTheodore Ts'o brelse(iloc.bh); 42371d1fe1eeSDavid Howells iget_failed(inode); 42381d1fe1eeSDavid Howells return ERR_PTR(ret); 4239ac27a0ecSDave Kleikamp } 4240ac27a0ecSDave Kleikamp 42410fc1b451SAneesh Kumar K.V static int ext4_inode_blocks_set(handle_t *handle, 42420fc1b451SAneesh Kumar K.V struct ext4_inode *raw_inode, 42430fc1b451SAneesh Kumar K.V struct ext4_inode_info *ei) 42440fc1b451SAneesh Kumar K.V { 42450fc1b451SAneesh Kumar K.V struct inode *inode = &(ei->vfs_inode); 42460fc1b451SAneesh Kumar K.V u64 i_blocks = inode->i_blocks; 42470fc1b451SAneesh Kumar K.V struct super_block *sb = inode->i_sb; 42480fc1b451SAneesh Kumar K.V 42490fc1b451SAneesh Kumar K.V if (i_blocks <= ~0U) { 42500fc1b451SAneesh Kumar K.V /* 42514907cb7bSAnatol Pomozov * i_blocks can be represented in a 32 bit variable 42520fc1b451SAneesh Kumar K.V * as multiple of 512 bytes 42530fc1b451SAneesh Kumar K.V */ 42548180a562SAneesh Kumar K.V raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); 42550fc1b451SAneesh Kumar K.V raw_inode->i_blocks_high = 0; 425684a8dce2SDmitry Monakhov ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); 4257f287a1a5STheodore Ts'o return 0; 4258f287a1a5STheodore Ts'o } 4259f287a1a5STheodore Ts'o if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) 4260f287a1a5STheodore Ts'o return -EFBIG; 4261f287a1a5STheodore Ts'o 4262f287a1a5STheodore Ts'o if (i_blocks <= 0xffffffffffffULL) { 42630fc1b451SAneesh Kumar K.V /* 42640fc1b451SAneesh Kumar K.V * i_blocks can be represented in a 48 bit variable 42650fc1b451SAneesh Kumar K.V * as multiple of 512 bytes 42660fc1b451SAneesh Kumar K.V */ 42678180a562SAneesh Kumar K.V raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); 42680fc1b451SAneesh Kumar K.V raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); 426984a8dce2SDmitry Monakhov ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); 42700fc1b451SAneesh Kumar K.V } else { 427184a8dce2SDmitry Monakhov ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); 42728180a562SAneesh Kumar K.V /* i_block is stored in file system block size */ 42738180a562SAneesh Kumar K.V i_blocks = i_blocks >> (inode->i_blkbits - 9); 42748180a562SAneesh Kumar K.V raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); 42758180a562SAneesh Kumar K.V raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); 42760fc1b451SAneesh Kumar K.V } 4277f287a1a5STheodore Ts'o return 0; 42780fc1b451SAneesh Kumar K.V } 42790fc1b451SAneesh Kumar K.V 4280ac27a0ecSDave Kleikamp /* 4281ac27a0ecSDave Kleikamp * Post the struct inode info into an on-disk inode location in the 4282ac27a0ecSDave Kleikamp * buffer-cache. This gobbles the caller's reference to the 4283ac27a0ecSDave Kleikamp * buffer_head in the inode location struct. 4284ac27a0ecSDave Kleikamp * 4285ac27a0ecSDave Kleikamp * The caller must have write access to iloc->bh. 4286ac27a0ecSDave Kleikamp */ 4287617ba13bSMingming Cao static int ext4_do_update_inode(handle_t *handle, 4288ac27a0ecSDave Kleikamp struct inode *inode, 4289830156c7SFrank Mayhar struct ext4_iloc *iloc) 4290ac27a0ecSDave Kleikamp { 4291617ba13bSMingming Cao struct ext4_inode *raw_inode = ext4_raw_inode(iloc); 4292617ba13bSMingming Cao struct ext4_inode_info *ei = EXT4_I(inode); 4293ac27a0ecSDave Kleikamp struct buffer_head *bh = iloc->bh; 4294ac27a0ecSDave Kleikamp int err = 0, rc, block; 4295b71fc079SJan Kara int need_datasync = 0; 429608cefc7aSEric W. Biederman uid_t i_uid; 429708cefc7aSEric W. Biederman gid_t i_gid; 4298ac27a0ecSDave Kleikamp 4299ac27a0ecSDave Kleikamp /* For fields not not tracking in the in-memory inode, 4300ac27a0ecSDave Kleikamp * initialise them to zero for new inodes. */ 430119f5fb7aSTheodore Ts'o if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) 4302617ba13bSMingming Cao memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); 4303ac27a0ecSDave Kleikamp 4304ff9ddf7eSJan Kara ext4_get_inode_flags(ei); 4305ac27a0ecSDave Kleikamp raw_inode->i_mode = cpu_to_le16(inode->i_mode); 430608cefc7aSEric W. Biederman i_uid = i_uid_read(inode); 430708cefc7aSEric W. Biederman i_gid = i_gid_read(inode); 4308ac27a0ecSDave Kleikamp if (!(test_opt(inode->i_sb, NO_UID32))) { 430908cefc7aSEric W. Biederman raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); 431008cefc7aSEric W. Biederman raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); 4311ac27a0ecSDave Kleikamp /* 4312ac27a0ecSDave Kleikamp * Fix up interoperability with old kernels. Otherwise, old inodes get 4313ac27a0ecSDave Kleikamp * re-used with the upper 16 bits of the uid/gid intact 4314ac27a0ecSDave Kleikamp */ 4315ac27a0ecSDave Kleikamp if (!ei->i_dtime) { 4316ac27a0ecSDave Kleikamp raw_inode->i_uid_high = 431708cefc7aSEric W. Biederman cpu_to_le16(high_16_bits(i_uid)); 4318ac27a0ecSDave Kleikamp raw_inode->i_gid_high = 431908cefc7aSEric W. Biederman cpu_to_le16(high_16_bits(i_gid)); 4320ac27a0ecSDave Kleikamp } else { 4321ac27a0ecSDave Kleikamp raw_inode->i_uid_high = 0; 4322ac27a0ecSDave Kleikamp raw_inode->i_gid_high = 0; 4323ac27a0ecSDave Kleikamp } 4324ac27a0ecSDave Kleikamp } else { 432508cefc7aSEric W. Biederman raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); 432608cefc7aSEric W. Biederman raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); 4327ac27a0ecSDave Kleikamp raw_inode->i_uid_high = 0; 4328ac27a0ecSDave Kleikamp raw_inode->i_gid_high = 0; 4329ac27a0ecSDave Kleikamp } 4330ac27a0ecSDave Kleikamp raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); 4331ef7f3835SKalpak Shah 4332ef7f3835SKalpak Shah EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); 4333ef7f3835SKalpak Shah EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); 4334ef7f3835SKalpak Shah EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); 4335ef7f3835SKalpak Shah EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); 4336ef7f3835SKalpak Shah 43370fc1b451SAneesh Kumar K.V if (ext4_inode_blocks_set(handle, raw_inode, ei)) 43380fc1b451SAneesh Kumar K.V goto out_brelse; 4339ac27a0ecSDave Kleikamp raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); 4340353eb83cSTheodore Ts'o raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); 43419b8f1f01SMingming Cao if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != 43429b8f1f01SMingming Cao cpu_to_le32(EXT4_OS_HURD)) 4343a1ddeb7eSBadari Pulavarty raw_inode->i_file_acl_high = 4344a1ddeb7eSBadari Pulavarty cpu_to_le16(ei->i_file_acl >> 32); 43457973c0c1SAneesh Kumar K.V raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); 4346b71fc079SJan Kara if (ei->i_disksize != ext4_isize(raw_inode)) { 4347a48380f7SAneesh Kumar K.V ext4_isize_set(raw_inode, ei->i_disksize); 4348b71fc079SJan Kara need_datasync = 1; 4349b71fc079SJan Kara } 4350ac27a0ecSDave Kleikamp if (ei->i_disksize > 0x7fffffffULL) { 4351ac27a0ecSDave Kleikamp struct super_block *sb = inode->i_sb; 4352617ba13bSMingming Cao if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, 4353617ba13bSMingming Cao EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || 4354617ba13bSMingming Cao EXT4_SB(sb)->s_es->s_rev_level == 4355617ba13bSMingming Cao cpu_to_le32(EXT4_GOOD_OLD_REV)) { 4356ac27a0ecSDave Kleikamp /* If this is the first large file 4357ac27a0ecSDave Kleikamp * created, add a flag to the superblock. 4358ac27a0ecSDave Kleikamp */ 4359617ba13bSMingming Cao err = ext4_journal_get_write_access(handle, 4360617ba13bSMingming Cao EXT4_SB(sb)->s_sbh); 4361ac27a0ecSDave Kleikamp if (err) 4362ac27a0ecSDave Kleikamp goto out_brelse; 4363617ba13bSMingming Cao ext4_update_dynamic_rev(sb); 4364617ba13bSMingming Cao EXT4_SET_RO_COMPAT_FEATURE(sb, 4365617ba13bSMingming Cao EXT4_FEATURE_RO_COMPAT_LARGE_FILE); 43660390131bSFrank Mayhar ext4_handle_sync(handle); 4367b50924c2SArtem Bityutskiy err = ext4_handle_dirty_super(handle, sb); 4368ac27a0ecSDave Kleikamp } 4369ac27a0ecSDave Kleikamp } 4370ac27a0ecSDave Kleikamp raw_inode->i_generation = cpu_to_le32(inode->i_generation); 4371ac27a0ecSDave Kleikamp if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { 4372ac27a0ecSDave Kleikamp if (old_valid_dev(inode->i_rdev)) { 4373ac27a0ecSDave Kleikamp raw_inode->i_block[0] = 4374ac27a0ecSDave Kleikamp cpu_to_le32(old_encode_dev(inode->i_rdev)); 4375ac27a0ecSDave Kleikamp raw_inode->i_block[1] = 0; 4376ac27a0ecSDave Kleikamp } else { 4377ac27a0ecSDave Kleikamp raw_inode->i_block[0] = 0; 4378ac27a0ecSDave Kleikamp raw_inode->i_block[1] = 4379ac27a0ecSDave Kleikamp cpu_to_le32(new_encode_dev(inode->i_rdev)); 4380ac27a0ecSDave Kleikamp raw_inode->i_block[2] = 0; 4381ac27a0ecSDave Kleikamp } 4382f19d5870STao Ma } else if (!ext4_has_inline_data(inode)) { 4383de9a55b8STheodore Ts'o for (block = 0; block < EXT4_N_BLOCKS; block++) 4384ac27a0ecSDave Kleikamp raw_inode->i_block[block] = ei->i_data[block]; 4385f19d5870STao Ma } 4386ac27a0ecSDave Kleikamp 438725ec56b5SJean Noel Cordenner raw_inode->i_disk_version = cpu_to_le32(inode->i_version); 438825ec56b5SJean Noel Cordenner if (ei->i_extra_isize) { 438925ec56b5SJean Noel Cordenner if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) 439025ec56b5SJean Noel Cordenner raw_inode->i_version_hi = 439125ec56b5SJean Noel Cordenner cpu_to_le32(inode->i_version >> 32); 4392ac27a0ecSDave Kleikamp raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); 439325ec56b5SJean Noel Cordenner } 439425ec56b5SJean Noel Cordenner 4395814525f4SDarrick J. Wong ext4_inode_csum_set(inode, raw_inode, ei); 4396814525f4SDarrick J. Wong 43970390131bSFrank Mayhar BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); 439873b50c1cSCurt Wohlgemuth rc = ext4_handle_dirty_metadata(handle, NULL, bh); 4399ac27a0ecSDave Kleikamp if (!err) 4400ac27a0ecSDave Kleikamp err = rc; 440119f5fb7aSTheodore Ts'o ext4_clear_inode_state(inode, EXT4_STATE_NEW); 4402ac27a0ecSDave Kleikamp 4403b71fc079SJan Kara ext4_update_inode_fsync_trans(handle, inode, need_datasync); 4404ac27a0ecSDave Kleikamp out_brelse: 4405ac27a0ecSDave Kleikamp brelse(bh); 4406617ba13bSMingming Cao ext4_std_error(inode->i_sb, err); 4407ac27a0ecSDave Kleikamp return err; 4408ac27a0ecSDave Kleikamp } 4409ac27a0ecSDave Kleikamp 4410ac27a0ecSDave Kleikamp /* 4411617ba13bSMingming Cao * ext4_write_inode() 4412ac27a0ecSDave Kleikamp * 4413ac27a0ecSDave Kleikamp * We are called from a few places: 4414ac27a0ecSDave Kleikamp * 4415ac27a0ecSDave Kleikamp * - Within generic_file_write() for O_SYNC files. 4416ac27a0ecSDave Kleikamp * Here, there will be no transaction running. We wait for any running 44174907cb7bSAnatol Pomozov * transaction to commit. 4418ac27a0ecSDave Kleikamp * 4419ac27a0ecSDave Kleikamp * - Within sys_sync(), kupdate and such. 4420ac27a0ecSDave Kleikamp * We wait on commit, if tol to. 4421ac27a0ecSDave Kleikamp * 4422ac27a0ecSDave Kleikamp * - Within prune_icache() (PF_MEMALLOC == true) 4423ac27a0ecSDave Kleikamp * Here we simply return. We can't afford to block kswapd on the 4424ac27a0ecSDave Kleikamp * journal commit. 4425ac27a0ecSDave Kleikamp * 4426ac27a0ecSDave Kleikamp * In all cases it is actually safe for us to return without doing anything, 4427ac27a0ecSDave Kleikamp * because the inode has been copied into a raw inode buffer in 4428617ba13bSMingming Cao * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for 4429ac27a0ecSDave Kleikamp * knfsd. 4430ac27a0ecSDave Kleikamp * 4431ac27a0ecSDave Kleikamp * Note that we are absolutely dependent upon all inode dirtiers doing the 4432ac27a0ecSDave Kleikamp * right thing: they *must* call mark_inode_dirty() after dirtying info in 4433ac27a0ecSDave Kleikamp * which we are interested. 4434ac27a0ecSDave Kleikamp * 4435ac27a0ecSDave Kleikamp * It would be a bug for them to not do this. The code: 4436ac27a0ecSDave Kleikamp * 4437ac27a0ecSDave Kleikamp * mark_inode_dirty(inode) 4438ac27a0ecSDave Kleikamp * stuff(); 4439ac27a0ecSDave Kleikamp * inode->i_size = expr; 4440ac27a0ecSDave Kleikamp * 4441ac27a0ecSDave Kleikamp * is in error because a kswapd-driven write_inode() could occur while 4442ac27a0ecSDave Kleikamp * `stuff()' is running, and the new i_size will be lost. Plus the inode 4443ac27a0ecSDave Kleikamp * will no longer be on the superblock's dirty inode list. 4444ac27a0ecSDave Kleikamp */ 4445a9185b41SChristoph Hellwig int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) 4446ac27a0ecSDave Kleikamp { 444791ac6f43SFrank Mayhar int err; 444891ac6f43SFrank Mayhar 4449ac27a0ecSDave Kleikamp if (current->flags & PF_MEMALLOC) 4450ac27a0ecSDave Kleikamp return 0; 4451ac27a0ecSDave Kleikamp 445291ac6f43SFrank Mayhar if (EXT4_SB(inode->i_sb)->s_journal) { 4453617ba13bSMingming Cao if (ext4_journal_current_handle()) { 4454b38bd33aSMingming Cao jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); 4455ac27a0ecSDave Kleikamp dump_stack(); 4456ac27a0ecSDave Kleikamp return -EIO; 4457ac27a0ecSDave Kleikamp } 4458ac27a0ecSDave Kleikamp 4459a9185b41SChristoph Hellwig if (wbc->sync_mode != WB_SYNC_ALL) 4460ac27a0ecSDave Kleikamp return 0; 4461ac27a0ecSDave Kleikamp 446291ac6f43SFrank Mayhar err = ext4_force_commit(inode->i_sb); 446391ac6f43SFrank Mayhar } else { 446491ac6f43SFrank Mayhar struct ext4_iloc iloc; 446591ac6f43SFrank Mayhar 44668b472d73SCurt Wohlgemuth err = __ext4_get_inode_loc(inode, &iloc, 0); 446791ac6f43SFrank Mayhar if (err) 446891ac6f43SFrank Mayhar return err; 4469a9185b41SChristoph Hellwig if (wbc->sync_mode == WB_SYNC_ALL) 4470830156c7SFrank Mayhar sync_dirty_buffer(iloc.bh); 4471830156c7SFrank Mayhar if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { 4472c398eda0STheodore Ts'o EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, 4473c398eda0STheodore Ts'o "IO error syncing inode"); 4474830156c7SFrank Mayhar err = -EIO; 4475830156c7SFrank Mayhar } 4476fd2dd9fbSCurt Wohlgemuth brelse(iloc.bh); 447791ac6f43SFrank Mayhar } 447891ac6f43SFrank Mayhar return err; 4479ac27a0ecSDave Kleikamp } 4480ac27a0ecSDave Kleikamp 4481ac27a0ecSDave Kleikamp /* 448253e87268SJan Kara * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate 448353e87268SJan Kara * buffers that are attached to a page stradding i_size and are undergoing 448453e87268SJan Kara * commit. In that case we have to wait for commit to finish and try again. 448553e87268SJan Kara */ 448653e87268SJan Kara static void ext4_wait_for_tail_page_commit(struct inode *inode) 448753e87268SJan Kara { 448853e87268SJan Kara struct page *page; 448953e87268SJan Kara unsigned offset; 449053e87268SJan Kara journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; 449153e87268SJan Kara tid_t commit_tid = 0; 449253e87268SJan Kara int ret; 449353e87268SJan Kara 449453e87268SJan Kara offset = inode->i_size & (PAGE_CACHE_SIZE - 1); 449553e87268SJan Kara /* 449653e87268SJan Kara * All buffers in the last page remain valid? Then there's nothing to 449753e87268SJan Kara * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE == 449853e87268SJan Kara * blocksize case 449953e87268SJan Kara */ 450053e87268SJan Kara if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits)) 450153e87268SJan Kara return; 450253e87268SJan Kara while (1) { 450353e87268SJan Kara page = find_lock_page(inode->i_mapping, 450453e87268SJan Kara inode->i_size >> PAGE_CACHE_SHIFT); 450553e87268SJan Kara if (!page) 450653e87268SJan Kara return; 4507ca99fdd2SLukas Czerner ret = __ext4_journalled_invalidatepage(page, offset, 4508ca99fdd2SLukas Czerner PAGE_CACHE_SIZE - offset); 450953e87268SJan Kara unlock_page(page); 451053e87268SJan Kara page_cache_release(page); 451153e87268SJan Kara if (ret != -EBUSY) 451253e87268SJan Kara return; 451353e87268SJan Kara commit_tid = 0; 451453e87268SJan Kara read_lock(&journal->j_state_lock); 451553e87268SJan Kara if (journal->j_committing_transaction) 451653e87268SJan Kara commit_tid = journal->j_committing_transaction->t_tid; 451753e87268SJan Kara read_unlock(&journal->j_state_lock); 451853e87268SJan Kara if (commit_tid) 451953e87268SJan Kara jbd2_log_wait_commit(journal, commit_tid); 452053e87268SJan Kara } 452153e87268SJan Kara } 452253e87268SJan Kara 452353e87268SJan Kara /* 4524617ba13bSMingming Cao * ext4_setattr() 4525ac27a0ecSDave Kleikamp * 4526ac27a0ecSDave Kleikamp * Called from notify_change. 4527ac27a0ecSDave Kleikamp * 4528ac27a0ecSDave Kleikamp * We want to trap VFS attempts to truncate the file as soon as 4529ac27a0ecSDave Kleikamp * possible. In particular, we want to make sure that when the VFS 4530ac27a0ecSDave Kleikamp * shrinks i_size, we put the inode on the orphan list and modify 4531ac27a0ecSDave Kleikamp * i_disksize immediately, so that during the subsequent flushing of 4532ac27a0ecSDave Kleikamp * dirty pages and freeing of disk blocks, we can guarantee that any 4533ac27a0ecSDave Kleikamp * commit will leave the blocks being flushed in an unused state on 4534ac27a0ecSDave Kleikamp * disk. (On recovery, the inode will get truncated and the blocks will 4535ac27a0ecSDave Kleikamp * be freed, so we have a strong guarantee that no future commit will 4536ac27a0ecSDave Kleikamp * leave these blocks visible to the user.) 4537ac27a0ecSDave Kleikamp * 4538678aaf48SJan Kara * Another thing we have to assure is that if we are in ordered mode 4539678aaf48SJan Kara * and inode is still attached to the committing transaction, we must 4540678aaf48SJan Kara * we start writeout of all the dirty pages which are being truncated. 4541678aaf48SJan Kara * This way we are sure that all the data written in the previous 4542678aaf48SJan Kara * transaction are already on disk (truncate waits for pages under 4543678aaf48SJan Kara * writeback). 4544678aaf48SJan Kara * 4545678aaf48SJan Kara * Called with inode->i_mutex down. 4546ac27a0ecSDave Kleikamp */ 4547617ba13bSMingming Cao int ext4_setattr(struct dentry *dentry, struct iattr *attr) 4548ac27a0ecSDave Kleikamp { 4549ac27a0ecSDave Kleikamp struct inode *inode = dentry->d_inode; 4550ac27a0ecSDave Kleikamp int error, rc = 0; 45513d287de3SDmitry Monakhov int orphan = 0; 4552ac27a0ecSDave Kleikamp const unsigned int ia_valid = attr->ia_valid; 4553ac27a0ecSDave Kleikamp 4554ac27a0ecSDave Kleikamp error = inode_change_ok(inode, attr); 4555ac27a0ecSDave Kleikamp if (error) 4556ac27a0ecSDave Kleikamp return error; 4557ac27a0ecSDave Kleikamp 455812755627SDmitry Monakhov if (is_quota_modification(inode, attr)) 4559871a2931SChristoph Hellwig dquot_initialize(inode); 456008cefc7aSEric W. Biederman if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || 456108cefc7aSEric W. Biederman (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { 4562ac27a0ecSDave Kleikamp handle_t *handle; 4563ac27a0ecSDave Kleikamp 4564ac27a0ecSDave Kleikamp /* (user+group)*(old+new) structure, inode write (sb, 4565ac27a0ecSDave Kleikamp * inode block, ? - but truncate inode update has it) */ 45669924a92aSTheodore Ts'o handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 45679924a92aSTheodore Ts'o (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + 4568194074acSDmitry Monakhov EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); 4569ac27a0ecSDave Kleikamp if (IS_ERR(handle)) { 4570ac27a0ecSDave Kleikamp error = PTR_ERR(handle); 4571ac27a0ecSDave Kleikamp goto err_out; 4572ac27a0ecSDave Kleikamp } 4573b43fa828SChristoph Hellwig error = dquot_transfer(inode, attr); 4574ac27a0ecSDave Kleikamp if (error) { 4575617ba13bSMingming Cao ext4_journal_stop(handle); 4576ac27a0ecSDave Kleikamp return error; 4577ac27a0ecSDave Kleikamp } 4578ac27a0ecSDave Kleikamp /* Update corresponding info in inode so that everything is in 4579ac27a0ecSDave Kleikamp * one transaction */ 4580ac27a0ecSDave Kleikamp if (attr->ia_valid & ATTR_UID) 4581ac27a0ecSDave Kleikamp inode->i_uid = attr->ia_uid; 4582ac27a0ecSDave Kleikamp if (attr->ia_valid & ATTR_GID) 4583ac27a0ecSDave Kleikamp inode->i_gid = attr->ia_gid; 4584617ba13bSMingming Cao error = ext4_mark_inode_dirty(handle, inode); 4585617ba13bSMingming Cao ext4_journal_stop(handle); 4586ac27a0ecSDave Kleikamp } 4587ac27a0ecSDave Kleikamp 45885208386cSJan Kara if (attr->ia_valid & ATTR_SIZE && attr->ia_size != inode->i_size) { 45895208386cSJan Kara handle_t *handle; 4590562c72aaSChristoph Hellwig 459112e9b892SDmitry Monakhov if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { 4592e2b46574SEric Sandeen struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 4593e2b46574SEric Sandeen 45940c095c7fSTheodore Ts'o if (attr->ia_size > sbi->s_bitmap_maxbytes) 45950c095c7fSTheodore Ts'o return -EFBIG; 4596e2b46574SEric Sandeen } 4597ac27a0ecSDave Kleikamp if (S_ISREG(inode->i_mode) && 4598072bd7eaSTheodore Ts'o (attr->ia_size < inode->i_size)) { 45995208386cSJan Kara if (ext4_should_order_data(inode)) { 46005208386cSJan Kara error = ext4_begin_ordered_truncate(inode, 46015208386cSJan Kara attr->ia_size); 46025208386cSJan Kara if (error) 46035208386cSJan Kara goto err_out; 46045208386cSJan Kara } 46059924a92aSTheodore Ts'o handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); 4606ac27a0ecSDave Kleikamp if (IS_ERR(handle)) { 4607ac27a0ecSDave Kleikamp error = PTR_ERR(handle); 4608ac27a0ecSDave Kleikamp goto err_out; 4609ac27a0ecSDave Kleikamp } 46103d287de3SDmitry Monakhov if (ext4_handle_valid(handle)) { 4611617ba13bSMingming Cao error = ext4_orphan_add(handle, inode); 46123d287de3SDmitry Monakhov orphan = 1; 46133d287de3SDmitry Monakhov } 461490e775b7SJan Kara down_write(&EXT4_I(inode)->i_data_sem); 4615617ba13bSMingming Cao EXT4_I(inode)->i_disksize = attr->ia_size; 4616617ba13bSMingming Cao rc = ext4_mark_inode_dirty(handle, inode); 4617ac27a0ecSDave Kleikamp if (!error) 4618ac27a0ecSDave Kleikamp error = rc; 461990e775b7SJan Kara /* 462090e775b7SJan Kara * We have to update i_size under i_data_sem together 462190e775b7SJan Kara * with i_disksize to avoid races with writeback code 462290e775b7SJan Kara * running ext4_wb_update_i_disksize(). 462390e775b7SJan Kara */ 462490e775b7SJan Kara if (!error) 462590e775b7SJan Kara i_size_write(inode, attr->ia_size); 462690e775b7SJan Kara up_write(&EXT4_I(inode)->i_data_sem); 4627617ba13bSMingming Cao ext4_journal_stop(handle); 4628678aaf48SJan Kara if (error) { 4629678aaf48SJan Kara ext4_orphan_del(NULL, inode); 4630678aaf48SJan Kara goto err_out; 4631678aaf48SJan Kara } 463290e775b7SJan Kara } else 463353e87268SJan Kara i_size_write(inode, attr->ia_size); 463490e775b7SJan Kara 463553e87268SJan Kara /* 463653e87268SJan Kara * Blocks are going to be removed from the inode. Wait 463753e87268SJan Kara * for dio in flight. Temporarily disable 463853e87268SJan Kara * dioread_nolock to prevent livelock. 463953e87268SJan Kara */ 46401b65007eSDmitry Monakhov if (orphan) { 464153e87268SJan Kara if (!ext4_should_journal_data(inode)) { 46421b65007eSDmitry Monakhov ext4_inode_block_unlocked_dio(inode); 46431c9114f9SDmitry Monakhov inode_dio_wait(inode); 46441b65007eSDmitry Monakhov ext4_inode_resume_unlocked_dio(inode); 464553e87268SJan Kara } else 464653e87268SJan Kara ext4_wait_for_tail_page_commit(inode); 46471b65007eSDmitry Monakhov } 464853e87268SJan Kara /* 464953e87268SJan Kara * Truncate pagecache after we've waited for commit 465053e87268SJan Kara * in data=journal mode to make pages freeable. 465153e87268SJan Kara */ 4652*7caef267SKirill A. Shutemov truncate_pagecache(inode, inode->i_size); 46531c9114f9SDmitry Monakhov } 46545208386cSJan Kara /* 46555208386cSJan Kara * We want to call ext4_truncate() even if attr->ia_size == 46565208386cSJan Kara * inode->i_size for cases like truncation of fallocated space 46575208386cSJan Kara */ 46585208386cSJan Kara if (attr->ia_valid & ATTR_SIZE) 4659072bd7eaSTheodore Ts'o ext4_truncate(inode); 4660ac27a0ecSDave Kleikamp 46611025774cSChristoph Hellwig if (!rc) { 46621025774cSChristoph Hellwig setattr_copy(inode, attr); 46631025774cSChristoph Hellwig mark_inode_dirty(inode); 46641025774cSChristoph Hellwig } 46651025774cSChristoph Hellwig 46661025774cSChristoph Hellwig /* 46671025774cSChristoph Hellwig * If the call to ext4_truncate failed to get a transaction handle at 46681025774cSChristoph Hellwig * all, we need to clean up the in-core orphan list manually. 46691025774cSChristoph Hellwig */ 46703d287de3SDmitry Monakhov if (orphan && inode->i_nlink) 4671617ba13bSMingming Cao ext4_orphan_del(NULL, inode); 4672ac27a0ecSDave Kleikamp 4673ac27a0ecSDave Kleikamp if (!rc && (ia_valid & ATTR_MODE)) 4674617ba13bSMingming Cao rc = ext4_acl_chmod(inode); 4675ac27a0ecSDave Kleikamp 4676ac27a0ecSDave Kleikamp err_out: 4677617ba13bSMingming Cao ext4_std_error(inode->i_sb, error); 4678ac27a0ecSDave Kleikamp if (!error) 4679ac27a0ecSDave Kleikamp error = rc; 4680ac27a0ecSDave Kleikamp return error; 4681ac27a0ecSDave Kleikamp } 4682ac27a0ecSDave Kleikamp 46833e3398a0SMingming Cao int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, 46843e3398a0SMingming Cao struct kstat *stat) 46853e3398a0SMingming Cao { 46863e3398a0SMingming Cao struct inode *inode; 46878af8eeccSJan Kara unsigned long long delalloc_blocks; 46883e3398a0SMingming Cao 46893e3398a0SMingming Cao inode = dentry->d_inode; 46903e3398a0SMingming Cao generic_fillattr(inode, stat); 46913e3398a0SMingming Cao 46923e3398a0SMingming Cao /* 46933e3398a0SMingming Cao * We can't update i_blocks if the block allocation is delayed 46943e3398a0SMingming Cao * otherwise in the case of system crash before the real block 46953e3398a0SMingming Cao * allocation is done, we will have i_blocks inconsistent with 46963e3398a0SMingming Cao * on-disk file blocks. 46973e3398a0SMingming Cao * We always keep i_blocks updated together with real 46983e3398a0SMingming Cao * allocation. But to not confuse with user, stat 46993e3398a0SMingming Cao * will return the blocks that include the delayed allocation 47003e3398a0SMingming Cao * blocks for this file. 47013e3398a0SMingming Cao */ 470296607551STao Ma delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), 470396607551STao Ma EXT4_I(inode)->i_reserved_data_blocks); 47043e3398a0SMingming Cao 47058af8eeccSJan Kara stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits-9); 47063e3398a0SMingming Cao return 0; 47073e3398a0SMingming Cao } 4708ac27a0ecSDave Kleikamp 4709fffb2739SJan Kara static int ext4_index_trans_blocks(struct inode *inode, int lblocks, 4710fffb2739SJan Kara int pextents) 4711a02908f1SMingming Cao { 471212e9b892SDmitry Monakhov if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) 4713fffb2739SJan Kara return ext4_ind_trans_blocks(inode, lblocks); 4714fffb2739SJan Kara return ext4_ext_index_trans_blocks(inode, pextents); 4715a02908f1SMingming Cao } 4716ac51d837STheodore Ts'o 4717a02908f1SMingming Cao /* 4718a02908f1SMingming Cao * Account for index blocks, block groups bitmaps and block group 4719a02908f1SMingming Cao * descriptor blocks if modify datablocks and index blocks 4720a02908f1SMingming Cao * worse case, the indexs blocks spread over different block groups 4721a02908f1SMingming Cao * 4722a02908f1SMingming Cao * If datablocks are discontiguous, they are possible to spread over 47234907cb7bSAnatol Pomozov * different block groups too. If they are contiguous, with flexbg, 4724a02908f1SMingming Cao * they could still across block group boundary. 4725a02908f1SMingming Cao * 4726a02908f1SMingming Cao * Also account for superblock, inode, quota and xattr blocks 4727a02908f1SMingming Cao */ 4728fffb2739SJan Kara static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, 4729fffb2739SJan Kara int pextents) 4730a02908f1SMingming Cao { 47318df9675fSTheodore Ts'o ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); 47328df9675fSTheodore Ts'o int gdpblocks; 4733a02908f1SMingming Cao int idxblocks; 4734a02908f1SMingming Cao int ret = 0; 4735a02908f1SMingming Cao 4736a02908f1SMingming Cao /* 4737fffb2739SJan Kara * How many index blocks need to touch to map @lblocks logical blocks 4738fffb2739SJan Kara * to @pextents physical extents? 4739a02908f1SMingming Cao */ 4740fffb2739SJan Kara idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents); 4741a02908f1SMingming Cao 4742a02908f1SMingming Cao ret = idxblocks; 4743a02908f1SMingming Cao 4744a02908f1SMingming Cao /* 4745a02908f1SMingming Cao * Now let's see how many group bitmaps and group descriptors need 4746a02908f1SMingming Cao * to account 4747a02908f1SMingming Cao */ 4748fffb2739SJan Kara groups = idxblocks + pextents; 4749a02908f1SMingming Cao gdpblocks = groups; 47508df9675fSTheodore Ts'o if (groups > ngroups) 47518df9675fSTheodore Ts'o groups = ngroups; 4752a02908f1SMingming Cao if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) 4753a02908f1SMingming Cao gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; 4754a02908f1SMingming Cao 4755a02908f1SMingming Cao /* bitmaps and block group descriptor blocks */ 4756a02908f1SMingming Cao ret += groups + gdpblocks; 4757a02908f1SMingming Cao 4758a02908f1SMingming Cao /* Blocks for super block, inode, quota and xattr blocks */ 4759a02908f1SMingming Cao ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); 4760ac27a0ecSDave Kleikamp 4761ac27a0ecSDave Kleikamp return ret; 4762ac27a0ecSDave Kleikamp } 4763ac27a0ecSDave Kleikamp 4764ac27a0ecSDave Kleikamp /* 476525985edcSLucas De Marchi * Calculate the total number of credits to reserve to fit 4766f3bd1f3fSMingming Cao * the modification of a single pages into a single transaction, 4767f3bd1f3fSMingming Cao * which may include multiple chunks of block allocations. 4768a02908f1SMingming Cao * 4769525f4ed8SMingming Cao * This could be called via ext4_write_begin() 4770a02908f1SMingming Cao * 4771525f4ed8SMingming Cao * We need to consider the worse case, when 4772a02908f1SMingming Cao * one new block per extent. 4773a02908f1SMingming Cao */ 4774a02908f1SMingming Cao int ext4_writepage_trans_blocks(struct inode *inode) 4775a02908f1SMingming Cao { 4776a02908f1SMingming Cao int bpp = ext4_journal_blocks_per_page(inode); 4777a02908f1SMingming Cao int ret; 4778a02908f1SMingming Cao 4779fffb2739SJan Kara ret = ext4_meta_trans_blocks(inode, bpp, bpp); 4780a02908f1SMingming Cao 4781a02908f1SMingming Cao /* Account for data blocks for journalled mode */ 4782a02908f1SMingming Cao if (ext4_should_journal_data(inode)) 4783a02908f1SMingming Cao ret += bpp; 4784a02908f1SMingming Cao return ret; 4785a02908f1SMingming Cao } 4786f3bd1f3fSMingming Cao 4787f3bd1f3fSMingming Cao /* 4788f3bd1f3fSMingming Cao * Calculate the journal credits for a chunk of data modification. 4789f3bd1f3fSMingming Cao * 4790f3bd1f3fSMingming Cao * This is called from DIO, fallocate or whoever calling 479179e83036SEric Sandeen * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. 4792f3bd1f3fSMingming Cao * 4793f3bd1f3fSMingming Cao * journal buffers for data blocks are not included here, as DIO 4794f3bd1f3fSMingming Cao * and fallocate do no need to journal data buffers. 4795f3bd1f3fSMingming Cao */ 4796f3bd1f3fSMingming Cao int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) 4797f3bd1f3fSMingming Cao { 4798f3bd1f3fSMingming Cao return ext4_meta_trans_blocks(inode, nrblocks, 1); 4799f3bd1f3fSMingming Cao } 4800f3bd1f3fSMingming Cao 4801a02908f1SMingming Cao /* 4802617ba13bSMingming Cao * The caller must have previously called ext4_reserve_inode_write(). 4803ac27a0ecSDave Kleikamp * Give this, we know that the caller already has write access to iloc->bh. 4804ac27a0ecSDave Kleikamp */ 4805617ba13bSMingming Cao int ext4_mark_iloc_dirty(handle_t *handle, 4806617ba13bSMingming Cao struct inode *inode, struct ext4_iloc *iloc) 4807ac27a0ecSDave Kleikamp { 4808ac27a0ecSDave Kleikamp int err = 0; 4809ac27a0ecSDave Kleikamp 4810c64db50eSTheodore Ts'o if (IS_I_VERSION(inode)) 481125ec56b5SJean Noel Cordenner inode_inc_iversion(inode); 481225ec56b5SJean Noel Cordenner 4813ac27a0ecSDave Kleikamp /* the do_update_inode consumes one bh->b_count */ 4814ac27a0ecSDave Kleikamp get_bh(iloc->bh); 4815ac27a0ecSDave Kleikamp 4816dab291afSMingming Cao /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ 4817830156c7SFrank Mayhar err = ext4_do_update_inode(handle, inode, iloc); 4818ac27a0ecSDave Kleikamp put_bh(iloc->bh); 4819ac27a0ecSDave Kleikamp return err; 4820ac27a0ecSDave Kleikamp } 4821ac27a0ecSDave Kleikamp 4822ac27a0ecSDave Kleikamp /* 4823ac27a0ecSDave Kleikamp * On success, We end up with an outstanding reference count against 4824ac27a0ecSDave Kleikamp * iloc->bh. This _must_ be cleaned up later. 4825ac27a0ecSDave Kleikamp */ 4826ac27a0ecSDave Kleikamp 4827ac27a0ecSDave Kleikamp int 4828617ba13bSMingming Cao ext4_reserve_inode_write(handle_t *handle, struct inode *inode, 4829617ba13bSMingming Cao struct ext4_iloc *iloc) 4830ac27a0ecSDave Kleikamp { 48310390131bSFrank Mayhar int err; 48320390131bSFrank Mayhar 4833617ba13bSMingming Cao err = ext4_get_inode_loc(inode, iloc); 4834ac27a0ecSDave Kleikamp if (!err) { 4835ac27a0ecSDave Kleikamp BUFFER_TRACE(iloc->bh, "get_write_access"); 4836617ba13bSMingming Cao err = ext4_journal_get_write_access(handle, iloc->bh); 4837ac27a0ecSDave Kleikamp if (err) { 4838ac27a0ecSDave Kleikamp brelse(iloc->bh); 4839ac27a0ecSDave Kleikamp iloc->bh = NULL; 4840ac27a0ecSDave Kleikamp } 4841ac27a0ecSDave Kleikamp } 4842617ba13bSMingming Cao ext4_std_error(inode->i_sb, err); 4843ac27a0ecSDave Kleikamp return err; 4844ac27a0ecSDave Kleikamp } 4845ac27a0ecSDave Kleikamp 4846ac27a0ecSDave Kleikamp /* 48476dd4ee7cSKalpak Shah * Expand an inode by new_extra_isize bytes. 48486dd4ee7cSKalpak Shah * Returns 0 on success or negative error number on failure. 48496dd4ee7cSKalpak Shah */ 48501d03ec98SAneesh Kumar K.V static int ext4_expand_extra_isize(struct inode *inode, 48511d03ec98SAneesh Kumar K.V unsigned int new_extra_isize, 48521d03ec98SAneesh Kumar K.V struct ext4_iloc iloc, 48531d03ec98SAneesh Kumar K.V handle_t *handle) 48546dd4ee7cSKalpak Shah { 48556dd4ee7cSKalpak Shah struct ext4_inode *raw_inode; 48566dd4ee7cSKalpak Shah struct ext4_xattr_ibody_header *header; 48576dd4ee7cSKalpak Shah 48586dd4ee7cSKalpak Shah if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) 48596dd4ee7cSKalpak Shah return 0; 48606dd4ee7cSKalpak Shah 48616dd4ee7cSKalpak Shah raw_inode = ext4_raw_inode(&iloc); 48626dd4ee7cSKalpak Shah 48636dd4ee7cSKalpak Shah header = IHDR(inode, raw_inode); 48646dd4ee7cSKalpak Shah 48656dd4ee7cSKalpak Shah /* No extended attributes present */ 486619f5fb7aSTheodore Ts'o if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || 48676dd4ee7cSKalpak Shah header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { 48686dd4ee7cSKalpak Shah memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, 48696dd4ee7cSKalpak Shah new_extra_isize); 48706dd4ee7cSKalpak Shah EXT4_I(inode)->i_extra_isize = new_extra_isize; 48716dd4ee7cSKalpak Shah return 0; 48726dd4ee7cSKalpak Shah } 48736dd4ee7cSKalpak Shah 48746dd4ee7cSKalpak Shah /* try to expand with EAs present */ 48756dd4ee7cSKalpak Shah return ext4_expand_extra_isize_ea(inode, new_extra_isize, 48766dd4ee7cSKalpak Shah raw_inode, handle); 48776dd4ee7cSKalpak Shah } 48786dd4ee7cSKalpak Shah 48796dd4ee7cSKalpak Shah /* 4880ac27a0ecSDave Kleikamp * What we do here is to mark the in-core inode as clean with respect to inode 4881ac27a0ecSDave Kleikamp * dirtiness (it may still be data-dirty). 4882ac27a0ecSDave Kleikamp * This means that the in-core inode may be reaped by prune_icache 4883ac27a0ecSDave Kleikamp * without having to perform any I/O. This is a very good thing, 4884ac27a0ecSDave Kleikamp * because *any* task may call prune_icache - even ones which 4885ac27a0ecSDave Kleikamp * have a transaction open against a different journal. 4886ac27a0ecSDave Kleikamp * 4887ac27a0ecSDave Kleikamp * Is this cheating? Not really. Sure, we haven't written the 4888ac27a0ecSDave Kleikamp * inode out, but prune_icache isn't a user-visible syncing function. 4889ac27a0ecSDave Kleikamp * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) 4890ac27a0ecSDave Kleikamp * we start and wait on commits. 4891ac27a0ecSDave Kleikamp */ 4892617ba13bSMingming Cao int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) 4893ac27a0ecSDave Kleikamp { 4894617ba13bSMingming Cao struct ext4_iloc iloc; 48956dd4ee7cSKalpak Shah struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 48966dd4ee7cSKalpak Shah static unsigned int mnt_count; 48976dd4ee7cSKalpak Shah int err, ret; 4898ac27a0ecSDave Kleikamp 4899ac27a0ecSDave Kleikamp might_sleep(); 49007ff9c073STheodore Ts'o trace_ext4_mark_inode_dirty(inode, _RET_IP_); 4901617ba13bSMingming Cao err = ext4_reserve_inode_write(handle, inode, &iloc); 49020390131bSFrank Mayhar if (ext4_handle_valid(handle) && 49030390131bSFrank Mayhar EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && 490419f5fb7aSTheodore Ts'o !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { 49056dd4ee7cSKalpak Shah /* 49066dd4ee7cSKalpak Shah * We need extra buffer credits since we may write into EA block 49076dd4ee7cSKalpak Shah * with this same handle. If journal_extend fails, then it will 49086dd4ee7cSKalpak Shah * only result in a minor loss of functionality for that inode. 49096dd4ee7cSKalpak Shah * If this is felt to be critical, then e2fsck should be run to 49106dd4ee7cSKalpak Shah * force a large enough s_min_extra_isize. 49116dd4ee7cSKalpak Shah */ 49126dd4ee7cSKalpak Shah if ((jbd2_journal_extend(handle, 49136dd4ee7cSKalpak Shah EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { 49146dd4ee7cSKalpak Shah ret = ext4_expand_extra_isize(inode, 49156dd4ee7cSKalpak Shah sbi->s_want_extra_isize, 49166dd4ee7cSKalpak Shah iloc, handle); 49176dd4ee7cSKalpak Shah if (ret) { 491819f5fb7aSTheodore Ts'o ext4_set_inode_state(inode, 491919f5fb7aSTheodore Ts'o EXT4_STATE_NO_EXPAND); 4920c1bddad9SAneesh Kumar K.V if (mnt_count != 4921c1bddad9SAneesh Kumar K.V le16_to_cpu(sbi->s_es->s_mnt_count)) { 492212062dddSEric Sandeen ext4_warning(inode->i_sb, 49236dd4ee7cSKalpak Shah "Unable to expand inode %lu. Delete" 49246dd4ee7cSKalpak Shah " some EAs or run e2fsck.", 49256dd4ee7cSKalpak Shah inode->i_ino); 4926c1bddad9SAneesh Kumar K.V mnt_count = 4927c1bddad9SAneesh Kumar K.V le16_to_cpu(sbi->s_es->s_mnt_count); 49286dd4ee7cSKalpak Shah } 49296dd4ee7cSKalpak Shah } 49306dd4ee7cSKalpak Shah } 49316dd4ee7cSKalpak Shah } 4932ac27a0ecSDave Kleikamp if (!err) 4933617ba13bSMingming Cao err = ext4_mark_iloc_dirty(handle, inode, &iloc); 4934ac27a0ecSDave Kleikamp return err; 4935ac27a0ecSDave Kleikamp } 4936ac27a0ecSDave Kleikamp 4937ac27a0ecSDave Kleikamp /* 4938617ba13bSMingming Cao * ext4_dirty_inode() is called from __mark_inode_dirty() 4939ac27a0ecSDave Kleikamp * 4940ac27a0ecSDave Kleikamp * We're really interested in the case where a file is being extended. 4941ac27a0ecSDave Kleikamp * i_size has been changed by generic_commit_write() and we thus need 4942ac27a0ecSDave Kleikamp * to include the updated inode in the current transaction. 4943ac27a0ecSDave Kleikamp * 49445dd4056dSChristoph Hellwig * Also, dquot_alloc_block() will always dirty the inode when blocks 4945ac27a0ecSDave Kleikamp * are allocated to the file. 4946ac27a0ecSDave Kleikamp * 4947ac27a0ecSDave Kleikamp * If the inode is marked synchronous, we don't honour that here - doing 4948ac27a0ecSDave Kleikamp * so would cause a commit on atime updates, which we don't bother doing. 4949ac27a0ecSDave Kleikamp * We handle synchronous inodes at the highest possible level. 4950ac27a0ecSDave Kleikamp */ 4951aa385729SChristoph Hellwig void ext4_dirty_inode(struct inode *inode, int flags) 4952ac27a0ecSDave Kleikamp { 4953ac27a0ecSDave Kleikamp handle_t *handle; 4954ac27a0ecSDave Kleikamp 49559924a92aSTheodore Ts'o handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); 4956ac27a0ecSDave Kleikamp if (IS_ERR(handle)) 4957ac27a0ecSDave Kleikamp goto out; 4958f3dc272fSCurt Wohlgemuth 4959617ba13bSMingming Cao ext4_mark_inode_dirty(handle, inode); 4960f3dc272fSCurt Wohlgemuth 4961617ba13bSMingming Cao ext4_journal_stop(handle); 4962ac27a0ecSDave Kleikamp out: 4963ac27a0ecSDave Kleikamp return; 4964ac27a0ecSDave Kleikamp } 4965ac27a0ecSDave Kleikamp 4966ac27a0ecSDave Kleikamp #if 0 4967ac27a0ecSDave Kleikamp /* 4968ac27a0ecSDave Kleikamp * Bind an inode's backing buffer_head into this transaction, to prevent 4969ac27a0ecSDave Kleikamp * it from being flushed to disk early. Unlike 4970617ba13bSMingming Cao * ext4_reserve_inode_write, this leaves behind no bh reference and 4971ac27a0ecSDave Kleikamp * returns no iloc structure, so the caller needs to repeat the iloc 4972ac27a0ecSDave Kleikamp * lookup to mark the inode dirty later. 4973ac27a0ecSDave Kleikamp */ 4974617ba13bSMingming Cao static int ext4_pin_inode(handle_t *handle, struct inode *inode) 4975ac27a0ecSDave Kleikamp { 4976617ba13bSMingming Cao struct ext4_iloc iloc; 4977ac27a0ecSDave Kleikamp 4978ac27a0ecSDave Kleikamp int err = 0; 4979ac27a0ecSDave Kleikamp if (handle) { 4980617ba13bSMingming Cao err = ext4_get_inode_loc(inode, &iloc); 4981ac27a0ecSDave Kleikamp if (!err) { 4982ac27a0ecSDave Kleikamp BUFFER_TRACE(iloc.bh, "get_write_access"); 4983dab291afSMingming Cao err = jbd2_journal_get_write_access(handle, iloc.bh); 4984ac27a0ecSDave Kleikamp if (!err) 49850390131bSFrank Mayhar err = ext4_handle_dirty_metadata(handle, 498673b50c1cSCurt Wohlgemuth NULL, 4987ac27a0ecSDave Kleikamp iloc.bh); 4988ac27a0ecSDave Kleikamp brelse(iloc.bh); 4989ac27a0ecSDave Kleikamp } 4990ac27a0ecSDave Kleikamp } 4991617ba13bSMingming Cao ext4_std_error(inode->i_sb, err); 4992ac27a0ecSDave Kleikamp return err; 4993ac27a0ecSDave Kleikamp } 4994ac27a0ecSDave Kleikamp #endif 4995ac27a0ecSDave Kleikamp 4996617ba13bSMingming Cao int ext4_change_inode_journal_flag(struct inode *inode, int val) 4997ac27a0ecSDave Kleikamp { 4998ac27a0ecSDave Kleikamp journal_t *journal; 4999ac27a0ecSDave Kleikamp handle_t *handle; 5000ac27a0ecSDave Kleikamp int err; 5001ac27a0ecSDave Kleikamp 5002ac27a0ecSDave Kleikamp /* 5003ac27a0ecSDave Kleikamp * We have to be very careful here: changing a data block's 5004ac27a0ecSDave Kleikamp * journaling status dynamically is dangerous. If we write a 5005ac27a0ecSDave Kleikamp * data block to the journal, change the status and then delete 5006ac27a0ecSDave Kleikamp * that block, we risk forgetting to revoke the old log record 5007ac27a0ecSDave Kleikamp * from the journal and so a subsequent replay can corrupt data. 5008ac27a0ecSDave Kleikamp * So, first we make sure that the journal is empty and that 5009ac27a0ecSDave Kleikamp * nobody is changing anything. 5010ac27a0ecSDave Kleikamp */ 5011ac27a0ecSDave Kleikamp 5012617ba13bSMingming Cao journal = EXT4_JOURNAL(inode); 50130390131bSFrank Mayhar if (!journal) 50140390131bSFrank Mayhar return 0; 5015d699594dSDave Hansen if (is_journal_aborted(journal)) 5016ac27a0ecSDave Kleikamp return -EROFS; 50172aff57b0SYongqiang Yang /* We have to allocate physical blocks for delalloc blocks 50182aff57b0SYongqiang Yang * before flushing journal. otherwise delalloc blocks can not 50192aff57b0SYongqiang Yang * be allocated any more. even more truncate on delalloc blocks 50202aff57b0SYongqiang Yang * could trigger BUG by flushing delalloc blocks in journal. 50212aff57b0SYongqiang Yang * There is no delalloc block in non-journal data mode. 50222aff57b0SYongqiang Yang */ 50232aff57b0SYongqiang Yang if (val && test_opt(inode->i_sb, DELALLOC)) { 50242aff57b0SYongqiang Yang err = ext4_alloc_da_blocks(inode); 50252aff57b0SYongqiang Yang if (err < 0) 50262aff57b0SYongqiang Yang return err; 50272aff57b0SYongqiang Yang } 5028ac27a0ecSDave Kleikamp 502917335dccSDmitry Monakhov /* Wait for all existing dio workers */ 503017335dccSDmitry Monakhov ext4_inode_block_unlocked_dio(inode); 503117335dccSDmitry Monakhov inode_dio_wait(inode); 503217335dccSDmitry Monakhov 5033dab291afSMingming Cao jbd2_journal_lock_updates(journal); 5034ac27a0ecSDave Kleikamp 5035ac27a0ecSDave Kleikamp /* 5036ac27a0ecSDave Kleikamp * OK, there are no updates running now, and all cached data is 5037ac27a0ecSDave Kleikamp * synced to disk. We are now in a completely consistent state 5038ac27a0ecSDave Kleikamp * which doesn't have anything in the journal, and we know that 5039ac27a0ecSDave Kleikamp * no filesystem updates are running, so it is safe to modify 5040ac27a0ecSDave Kleikamp * the inode's in-core data-journaling state flag now. 5041ac27a0ecSDave Kleikamp */ 5042ac27a0ecSDave Kleikamp 5043ac27a0ecSDave Kleikamp if (val) 504412e9b892SDmitry Monakhov ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); 50455872ddaaSYongqiang Yang else { 50465872ddaaSYongqiang Yang jbd2_journal_flush(journal); 504712e9b892SDmitry Monakhov ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); 50485872ddaaSYongqiang Yang } 5049617ba13bSMingming Cao ext4_set_aops(inode); 5050ac27a0ecSDave Kleikamp 5051dab291afSMingming Cao jbd2_journal_unlock_updates(journal); 505217335dccSDmitry Monakhov ext4_inode_resume_unlocked_dio(inode); 5053ac27a0ecSDave Kleikamp 5054ac27a0ecSDave Kleikamp /* Finally we can mark the inode as dirty. */ 5055ac27a0ecSDave Kleikamp 50569924a92aSTheodore Ts'o handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); 5057ac27a0ecSDave Kleikamp if (IS_ERR(handle)) 5058ac27a0ecSDave Kleikamp return PTR_ERR(handle); 5059ac27a0ecSDave Kleikamp 5060617ba13bSMingming Cao err = ext4_mark_inode_dirty(handle, inode); 50610390131bSFrank Mayhar ext4_handle_sync(handle); 5062617ba13bSMingming Cao ext4_journal_stop(handle); 5063617ba13bSMingming Cao ext4_std_error(inode->i_sb, err); 5064ac27a0ecSDave Kleikamp 5065ac27a0ecSDave Kleikamp return err; 5066ac27a0ecSDave Kleikamp } 50672e9ee850SAneesh Kumar K.V 50682e9ee850SAneesh Kumar K.V static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) 50692e9ee850SAneesh Kumar K.V { 50702e9ee850SAneesh Kumar K.V return !buffer_mapped(bh); 50712e9ee850SAneesh Kumar K.V } 50722e9ee850SAneesh Kumar K.V 5073c2ec175cSNick Piggin int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) 50742e9ee850SAneesh Kumar K.V { 5075c2ec175cSNick Piggin struct page *page = vmf->page; 50762e9ee850SAneesh Kumar K.V loff_t size; 50772e9ee850SAneesh Kumar K.V unsigned long len; 50789ea7df53SJan Kara int ret; 50792e9ee850SAneesh Kumar K.V struct file *file = vma->vm_file; 5080496ad9aaSAl Viro struct inode *inode = file_inode(file); 50812e9ee850SAneesh Kumar K.V struct address_space *mapping = inode->i_mapping; 50829ea7df53SJan Kara handle_t *handle; 50839ea7df53SJan Kara get_block_t *get_block; 50849ea7df53SJan Kara int retries = 0; 50852e9ee850SAneesh Kumar K.V 50868e8ad8a5SJan Kara sb_start_pagefault(inode->i_sb); 5087041bbb6dSTheodore Ts'o file_update_time(vma->vm_file); 50889ea7df53SJan Kara /* Delalloc case is easy... */ 50899ea7df53SJan Kara if (test_opt(inode->i_sb, DELALLOC) && 50909ea7df53SJan Kara !ext4_should_journal_data(inode) && 50919ea7df53SJan Kara !ext4_nonda_switch(inode->i_sb)) { 50929ea7df53SJan Kara do { 50939ea7df53SJan Kara ret = __block_page_mkwrite(vma, vmf, 50949ea7df53SJan Kara ext4_da_get_block_prep); 50959ea7df53SJan Kara } while (ret == -ENOSPC && 50969ea7df53SJan Kara ext4_should_retry_alloc(inode->i_sb, &retries)); 50979ea7df53SJan Kara goto out_ret; 50982e9ee850SAneesh Kumar K.V } 50990e499890SDarrick J. Wong 51000e499890SDarrick J. Wong lock_page(page); 51019ea7df53SJan Kara size = i_size_read(inode); 51029ea7df53SJan Kara /* Page got truncated from under us? */ 51039ea7df53SJan Kara if (page->mapping != mapping || page_offset(page) > size) { 51049ea7df53SJan Kara unlock_page(page); 51059ea7df53SJan Kara ret = VM_FAULT_NOPAGE; 51069ea7df53SJan Kara goto out; 51070e499890SDarrick J. Wong } 51082e9ee850SAneesh Kumar K.V 51092e9ee850SAneesh Kumar K.V if (page->index == size >> PAGE_CACHE_SHIFT) 51102e9ee850SAneesh Kumar K.V len = size & ~PAGE_CACHE_MASK; 51112e9ee850SAneesh Kumar K.V else 51122e9ee850SAneesh Kumar K.V len = PAGE_CACHE_SIZE; 5113a827eaffSAneesh Kumar K.V /* 51149ea7df53SJan Kara * Return if we have all the buffers mapped. This avoids the need to do 51159ea7df53SJan Kara * journal_start/journal_stop which can block and take a long time 5116a827eaffSAneesh Kumar K.V */ 51172e9ee850SAneesh Kumar K.V if (page_has_buffers(page)) { 5118f19d5870STao Ma if (!ext4_walk_page_buffers(NULL, page_buffers(page), 5119f19d5870STao Ma 0, len, NULL, 5120a827eaffSAneesh Kumar K.V ext4_bh_unmapped)) { 51219ea7df53SJan Kara /* Wait so that we don't change page under IO */ 51221d1d1a76SDarrick J. Wong wait_for_stable_page(page); 51239ea7df53SJan Kara ret = VM_FAULT_LOCKED; 51249ea7df53SJan Kara goto out; 51252e9ee850SAneesh Kumar K.V } 5126a827eaffSAneesh Kumar K.V } 5127a827eaffSAneesh Kumar K.V unlock_page(page); 51289ea7df53SJan Kara /* OK, we need to fill the hole... */ 51299ea7df53SJan Kara if (ext4_should_dioread_nolock(inode)) 51309ea7df53SJan Kara get_block = ext4_get_block_write; 51319ea7df53SJan Kara else 51329ea7df53SJan Kara get_block = ext4_get_block; 51339ea7df53SJan Kara retry_alloc: 51349924a92aSTheodore Ts'o handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 51359924a92aSTheodore Ts'o ext4_writepage_trans_blocks(inode)); 51369ea7df53SJan Kara if (IS_ERR(handle)) { 5137c2ec175cSNick Piggin ret = VM_FAULT_SIGBUS; 51389ea7df53SJan Kara goto out; 51399ea7df53SJan Kara } 51409ea7df53SJan Kara ret = __block_page_mkwrite(vma, vmf, get_block); 51419ea7df53SJan Kara if (!ret && ext4_should_journal_data(inode)) { 5142f19d5870STao Ma if (ext4_walk_page_buffers(handle, page_buffers(page), 0, 51439ea7df53SJan Kara PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) { 51449ea7df53SJan Kara unlock_page(page); 51459ea7df53SJan Kara ret = VM_FAULT_SIGBUS; 5146fcbb5515SYongqiang Yang ext4_journal_stop(handle); 51479ea7df53SJan Kara goto out; 51489ea7df53SJan Kara } 51499ea7df53SJan Kara ext4_set_inode_state(inode, EXT4_STATE_JDATA); 51509ea7df53SJan Kara } 51519ea7df53SJan Kara ext4_journal_stop(handle); 51529ea7df53SJan Kara if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) 51539ea7df53SJan Kara goto retry_alloc; 51549ea7df53SJan Kara out_ret: 51559ea7df53SJan Kara ret = block_page_mkwrite_return(ret); 51569ea7df53SJan Kara out: 51578e8ad8a5SJan Kara sb_end_pagefault(inode->i_sb); 51582e9ee850SAneesh Kumar K.V return ret; 51592e9ee850SAneesh Kumar K.V } 5160