| 1 | /* |
| 2 | * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README |
| 3 | */ |
| 4 | |
| 5 | #include <linux/time.h> |
| 6 | #include "reiserfs.h" |
| 7 | #include "acl.h" |
| 8 | #include "xattr.h" |
| 9 | #include <linux/uaccess.h> |
| 10 | #include <linux/pagemap.h> |
| 11 | #include <linux/swap.h> |
| 12 | #include <linux/writeback.h> |
| 13 | #include <linux/blkdev.h> |
| 14 | #include <linux/buffer_head.h> |
| 15 | #include <linux/quotaops.h> |
| 16 | |
| 17 | /* |
| 18 | * We pack the tails of files on file close, not at the time they are written. |
| 19 | * This implies an unnecessary copy of the tail and an unnecessary indirect item |
| 20 | * insertion/balancing, for files that are written in one write. |
| 21 | * It avoids unnecessary tail packings (balances) for files that are written in |
| 22 | * multiple writes and are small enough to have tails. |
| 23 | * |
| 24 | * file_release is called by the VFS layer when the file is closed. If |
| 25 | * this is the last open file descriptor, and the file |
| 26 | * small enough to have a tail, and the tail is currently in an |
| 27 | * unformatted node, the tail is converted back into a direct item. |
| 28 | * |
| 29 | * We use reiserfs_truncate_file to pack the tail, since it already has |
| 30 | * all the conditions coded. |
| 31 | */ |
| 32 | static int reiserfs_file_release(struct inode *inode, struct file *filp) |
| 33 | { |
| 34 | |
| 35 | struct reiserfs_transaction_handle th; |
| 36 | int err; |
| 37 | int jbegin_failure = 0; |
| 38 | |
| 39 | BUG_ON(!S_ISREG(inode->i_mode)); |
| 40 | |
| 41 | if (atomic_add_unless(&REISERFS_I(inode)->openers, -1, 1)) |
| 42 | return 0; |
| 43 | |
| 44 | mutex_lock(&REISERFS_I(inode)->tailpack); |
| 45 | |
| 46 | if (!atomic_dec_and_test(&REISERFS_I(inode)->openers)) { |
| 47 | mutex_unlock(&REISERFS_I(inode)->tailpack); |
| 48 | return 0; |
| 49 | } |
| 50 | |
| 51 | /* fast out for when nothing needs to be done */ |
| 52 | if ((!(REISERFS_I(inode)->i_flags & i_pack_on_close_mask) || |
| 53 | !tail_has_to_be_packed(inode)) && |
| 54 | REISERFS_I(inode)->i_prealloc_count <= 0) { |
| 55 | mutex_unlock(&REISERFS_I(inode)->tailpack); |
| 56 | return 0; |
| 57 | } |
| 58 | |
| 59 | reiserfs_write_lock(inode->i_sb); |
| 60 | /* |
| 61 | * freeing preallocation only involves relogging blocks that |
| 62 | * are already in the current transaction. preallocation gets |
| 63 | * freed at the end of each transaction, so it is impossible for |
| 64 | * us to log any additional blocks (including quota blocks) |
| 65 | */ |
| 66 | err = journal_begin(&th, inode->i_sb, 1); |
| 67 | if (err) { |
| 68 | /* |
| 69 | * uh oh, we can't allow the inode to go away while there |
| 70 | * is still preallocation blocks pending. Try to join the |
| 71 | * aborted transaction |
| 72 | */ |
| 73 | jbegin_failure = err; |
| 74 | err = journal_join_abort(&th, inode->i_sb); |
| 75 | |
| 76 | if (err) { |
| 77 | /* |
| 78 | * hmpf, our choices here aren't good. We can pin |
| 79 | * the inode which will disallow unmount from ever |
| 80 | * happening, we can do nothing, which will corrupt |
| 81 | * random memory on unmount, or we can forcibly |
| 82 | * remove the file from the preallocation list, which |
| 83 | * will leak blocks on disk. Lets pin the inode |
| 84 | * and let the admin know what is going on. |
| 85 | */ |
| 86 | igrab(inode); |
| 87 | reiserfs_warning(inode->i_sb, "clm-9001", |
| 88 | "pinning inode %lu because the " |
| 89 | "preallocation can't be freed", |
| 90 | inode->i_ino); |
| 91 | goto out; |
| 92 | } |
| 93 | } |
| 94 | reiserfs_update_inode_transaction(inode); |
| 95 | |
| 96 | #ifdef REISERFS_PREALLOCATE |
| 97 | reiserfs_discard_prealloc(&th, inode); |
| 98 | #endif |
| 99 | err = journal_end(&th); |
| 100 | |
| 101 | /* copy back the error code from journal_begin */ |
| 102 | if (!err) |
| 103 | err = jbegin_failure; |
| 104 | |
| 105 | if (!err && |
| 106 | (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) && |
| 107 | tail_has_to_be_packed(inode)) { |
| 108 | |
| 109 | /* |
| 110 | * if regular file is released by last holder and it has been |
| 111 | * appended (we append by unformatted node only) or its direct |
| 112 | * item(s) had to be converted, then it may have to be |
| 113 | * indirect2direct converted |
| 114 | */ |
| 115 | err = reiserfs_truncate_file(inode, 0); |
| 116 | } |
| 117 | out: |
| 118 | reiserfs_write_unlock(inode->i_sb); |
| 119 | mutex_unlock(&REISERFS_I(inode)->tailpack); |
| 120 | return err; |
| 121 | } |
| 122 | |
| 123 | static int reiserfs_file_open(struct inode *inode, struct file *file) |
| 124 | { |
| 125 | int err = dquot_file_open(inode, file); |
| 126 | |
| 127 | /* somebody might be tailpacking on final close; wait for it */ |
| 128 | if (!atomic_inc_not_zero(&REISERFS_I(inode)->openers)) { |
| 129 | mutex_lock(&REISERFS_I(inode)->tailpack); |
| 130 | atomic_inc(&REISERFS_I(inode)->openers); |
| 131 | mutex_unlock(&REISERFS_I(inode)->tailpack); |
| 132 | } |
| 133 | return err; |
| 134 | } |
| 135 | |
| 136 | void reiserfs_vfs_truncate_file(struct inode *inode) |
| 137 | { |
| 138 | mutex_lock(&REISERFS_I(inode)->tailpack); |
| 139 | reiserfs_truncate_file(inode, 1); |
| 140 | mutex_unlock(&REISERFS_I(inode)->tailpack); |
| 141 | } |
| 142 | |
| 143 | /* Sync a reiserfs file. */ |
| 144 | |
| 145 | /* |
| 146 | * FIXME: sync_mapping_buffers() never has anything to sync. Can |
| 147 | * be removed... |
| 148 | */ |
| 149 | |
| 150 | static int reiserfs_sync_file(struct file *filp, loff_t start, loff_t end, |
| 151 | int datasync) |
| 152 | { |
| 153 | struct inode *inode = filp->f_mapping->host; |
| 154 | int err; |
| 155 | int barrier_done; |
| 156 | |
| 157 | err = file_write_and_wait_range(filp, start, end); |
| 158 | if (err) |
| 159 | return err; |
| 160 | |
| 161 | inode_lock(inode); |
| 162 | BUG_ON(!S_ISREG(inode->i_mode)); |
| 163 | err = sync_mapping_buffers(inode->i_mapping); |
| 164 | reiserfs_write_lock(inode->i_sb); |
| 165 | barrier_done = reiserfs_commit_for_inode(inode); |
| 166 | reiserfs_write_unlock(inode->i_sb); |
| 167 | if (barrier_done != 1 && reiserfs_barrier_flush(inode->i_sb)) |
| 168 | blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL); |
| 169 | inode_unlock(inode); |
| 170 | if (barrier_done < 0) |
| 171 | return barrier_done; |
| 172 | return (err < 0) ? -EIO : 0; |
| 173 | } |
| 174 | |
| 175 | /* taken fs/buffer.c:__block_commit_write */ |
| 176 | int reiserfs_commit_page(struct inode *inode, struct page *page, |
| 177 | unsigned from, unsigned to) |
| 178 | { |
| 179 | unsigned block_start, block_end; |
| 180 | int partial = 0; |
| 181 | unsigned blocksize; |
| 182 | struct buffer_head *bh, *head; |
| 183 | unsigned long i_size_index = inode->i_size >> PAGE_SHIFT; |
| 184 | int new; |
| 185 | int logit = reiserfs_file_data_log(inode); |
| 186 | struct super_block *s = inode->i_sb; |
| 187 | int bh_per_page = PAGE_SIZE / s->s_blocksize; |
| 188 | struct reiserfs_transaction_handle th; |
| 189 | int ret = 0; |
| 190 | |
| 191 | th.t_trans_id = 0; |
| 192 | blocksize = i_blocksize(inode); |
| 193 | |
| 194 | if (logit) { |
| 195 | reiserfs_write_lock(s); |
| 196 | ret = journal_begin(&th, s, bh_per_page + 1); |
| 197 | if (ret) |
| 198 | goto drop_write_lock; |
| 199 | reiserfs_update_inode_transaction(inode); |
| 200 | } |
| 201 | for (bh = head = page_buffers(page), block_start = 0; |
| 202 | bh != head || !block_start; |
| 203 | block_start = block_end, bh = bh->b_this_page) { |
| 204 | |
| 205 | new = buffer_new(bh); |
| 206 | clear_buffer_new(bh); |
| 207 | block_end = block_start + blocksize; |
| 208 | if (block_end <= from || block_start >= to) { |
| 209 | if (!buffer_uptodate(bh)) |
| 210 | partial = 1; |
| 211 | } else { |
| 212 | set_buffer_uptodate(bh); |
| 213 | if (logit) { |
| 214 | reiserfs_prepare_for_journal(s, bh, 1); |
| 215 | journal_mark_dirty(&th, bh); |
| 216 | } else if (!buffer_dirty(bh)) { |
| 217 | mark_buffer_dirty(bh); |
| 218 | /* |
| 219 | * do data=ordered on any page past the end |
| 220 | * of file and any buffer marked BH_New. |
| 221 | */ |
| 222 | if (reiserfs_data_ordered(inode->i_sb) && |
| 223 | (new || page->index >= i_size_index)) { |
| 224 | reiserfs_add_ordered_list(inode, bh); |
| 225 | } |
| 226 | } |
| 227 | } |
| 228 | } |
| 229 | if (logit) { |
| 230 | ret = journal_end(&th); |
| 231 | drop_write_lock: |
| 232 | reiserfs_write_unlock(s); |
| 233 | } |
| 234 | /* |
| 235 | * If this is a partial write which happened to make all buffers |
| 236 | * uptodate then we can optimize away a bogus readpage() for |
| 237 | * the next read(). Here we 'discover' whether the page went |
| 238 | * uptodate as a result of this (potentially partial) write. |
| 239 | */ |
| 240 | if (!partial) |
| 241 | SetPageUptodate(page); |
| 242 | return ret; |
| 243 | } |
| 244 | |
| 245 | const struct file_operations reiserfs_file_operations = { |
| 246 | .unlocked_ioctl = reiserfs_ioctl, |
| 247 | #ifdef CONFIG_COMPAT |
| 248 | .compat_ioctl = reiserfs_compat_ioctl, |
| 249 | #endif |
| 250 | .mmap = generic_file_mmap, |
| 251 | .open = reiserfs_file_open, |
| 252 | .release = reiserfs_file_release, |
| 253 | .fsync = reiserfs_sync_file, |
| 254 | .read_iter = generic_file_read_iter, |
| 255 | .write_iter = generic_file_write_iter, |
| 256 | .splice_read = generic_file_splice_read, |
| 257 | .splice_write = iter_file_splice_write, |
| 258 | .llseek = generic_file_llseek, |
| 259 | }; |
| 260 | |
| 261 | const struct inode_operations reiserfs_file_inode_operations = { |
| 262 | .setattr = reiserfs_setattr, |
| 263 | .listxattr = reiserfs_listxattr, |
| 264 | .permission = reiserfs_permission, |
| 265 | .get_acl = reiserfs_get_acl, |
| 266 | .set_acl = reiserfs_set_acl, |
| 267 | }; |