1/* 2 * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README 3 */ 4 5#include <linux/time.h> 6#include <linux/reiserfs_fs.h> 7#include <linux/reiserfs_acl.h> 8#include <linux/reiserfs_xattr.h> 9#include <asm/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*/ 32static 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 /* freeing preallocation only involves relogging blocks that 61 * are already in the current transaction. preallocation gets 62 * freed at the end of each transaction, so it is impossible for 63 * us to log any additional blocks (including quota blocks) 64 */ 65 err = journal_begin(&th, inode->i_sb, 1); 66 if (err) { 67 /* uh oh, we can't allow the inode to go away while there 68 * is still preallocation blocks pending. Try to join the 69 * aborted transaction 70 */ 71 jbegin_failure = err; 72 err = journal_join_abort(&th, inode->i_sb, 1); 73 74 if (err) { 75 /* hmpf, our choices here aren't good. We can pin the inode 76 * which will disallow unmount from every happening, we can 77 * do nothing, which will corrupt random memory on unmount, 78 * or we can forcibly remove the file from the preallocation 79 * list, which will leak blocks on disk. Lets pin the inode 80 * and let the admin know what is going on. 81 */ 82 igrab(inode); 83 reiserfs_warning(inode->i_sb, "clm-9001", 84 "pinning inode %lu because the " 85 "preallocation can't be freed", 86 inode->i_ino); 87 goto out; 88 } 89 } 90 reiserfs_update_inode_transaction(inode); 91 92#ifdef REISERFS_PREALLOCATE 93 reiserfs_discard_prealloc(&th, inode); 94#endif 95 err = journal_end(&th, inode->i_sb, 1); 96 97 /* copy back the error code from journal_begin */ 98 if (!err) 99 err = jbegin_failure; 100 101 if (!err && 102 (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) && 103 tail_has_to_be_packed(inode)) { 104 105 /* if regular file is released by last holder and it has been 106 appended (we append by unformatted node only) or its direct 107 item(s) had to be converted, then it may have to be 108 indirect2direct converted */ 109 err = reiserfs_truncate_file(inode, 0); 110 } 111 out: 112 reiserfs_write_unlock(inode->i_sb); 113 mutex_unlock(&(REISERFS_I(inode)->tailpack)); 114 return err; 115} 116 117static int reiserfs_file_open(struct inode *inode, struct file *file) 118{ 119 int err = dquot_file_open(inode, file); 120 if (!atomic_inc_not_zero(&REISERFS_I(inode)->openers)) { 121 /* somebody might be tailpacking on final close; wait for it */ 122 mutex_lock(&(REISERFS_I(inode)->tailpack)); 123 atomic_inc(&REISERFS_I(inode)->openers); 124 mutex_unlock(&(REISERFS_I(inode)->tailpack)); 125 } 126 return err; 127} 128 129static void reiserfs_vfs_truncate_file(struct inode *inode) 130{ 131 mutex_lock(&(REISERFS_I(inode)->tailpack)); 132 reiserfs_truncate_file(inode, 1); 133 mutex_unlock(&(REISERFS_I(inode)->tailpack)); 134} 135 136/* Sync a reiserfs file. */ 137 138 139static int reiserfs_sync_file(struct file *filp, int datasync) 140{ 141 struct inode *inode = filp->f_mapping->host; 142 int err; 143 int barrier_done; 144 145 BUG_ON(!S_ISREG(inode->i_mode)); 146 err = sync_mapping_buffers(inode->i_mapping); 147 reiserfs_write_lock(inode->i_sb); 148 barrier_done = reiserfs_commit_for_inode(inode); 149 reiserfs_write_unlock(inode->i_sb); 150 if (barrier_done != 1 && reiserfs_barrier_flush(inode->i_sb)) 151 blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL, 152 BLKDEV_IFL_WAIT); 153 if (barrier_done < 0) 154 return barrier_done; 155 return (err < 0) ? -EIO : 0; 156} 157 158/* taken fs/buffer.c:__block_commit_write */ 159int reiserfs_commit_page(struct inode *inode, struct page *page, 160 unsigned from, unsigned to) 161{ 162 unsigned block_start, block_end; 163 int partial = 0; 164 unsigned blocksize; 165 struct buffer_head *bh, *head; 166 unsigned long i_size_index = inode->i_size >> PAGE_CACHE_SHIFT; 167 int new; 168 int logit = reiserfs_file_data_log(inode); 169 struct super_block *s = inode->i_sb; 170 int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize; 171 struct reiserfs_transaction_handle th; 172 int ret = 0; 173 174 th.t_trans_id = 0; 175 blocksize = 1 << inode->i_blkbits; 176 177 if (logit) { 178 reiserfs_write_lock(s); 179 ret = journal_begin(&th, s, bh_per_page + 1); 180 if (ret) 181 goto drop_write_lock; 182 reiserfs_update_inode_transaction(inode); 183 } 184 for (bh = head = page_buffers(page), block_start = 0; 185 bh != head || !block_start; 186 block_start = block_end, bh = bh->b_this_page) { 187 188 new = buffer_new(bh); 189 clear_buffer_new(bh); 190 block_end = block_start + blocksize; 191 if (block_end <= from || block_start >= to) { 192 if (!buffer_uptodate(bh)) 193 partial = 1; 194 } else { 195 set_buffer_uptodate(bh); 196 if (logit) { 197 reiserfs_prepare_for_journal(s, bh, 1); 198 journal_mark_dirty(&th, s, bh); 199 } else if (!buffer_dirty(bh)) { 200 mark_buffer_dirty(bh); 201 /* do data=ordered on any page past the end 202 * of file and any buffer marked BH_New. 203 */ 204 if (reiserfs_data_ordered(inode->i_sb) && 205 (new || page->index >= i_size_index)) { 206 reiserfs_add_ordered_list(inode, bh); 207 } 208 } 209 } 210 } 211 if (logit) { 212 ret = journal_end(&th, s, bh_per_page + 1); 213 drop_write_lock: 214 reiserfs_write_unlock(s); 215 } 216 /* 217 * If this is a partial write which happened to make all buffers 218 * uptodate then we can optimize away a bogus readpage() for 219 * the next read(). Here we 'discover' whether the page went 220 * uptodate as a result of this (potentially partial) write. 221 */ 222 if (!partial) 223 SetPageUptodate(page); 224 return ret; 225} 226 227/* Write @count bytes at position @ppos in a file indicated by @file 228 from the buffer @buf. 229 230 generic_file_write() is only appropriate for filesystems that are not seeking to optimize performance and want 231 something simple that works. It is not for serious use by general purpose filesystems, excepting the one that it was 232 written for (ext2/3). This is for several reasons: 233 234 * It has no understanding of any filesystem specific optimizations. 235 236 * It enters the filesystem repeatedly for each page that is written. 237 238 * It depends on reiserfs_get_block() function which if implemented by reiserfs performs costly search_by_key 239 * operation for each page it is supplied with. By contrast reiserfs_file_write() feeds as much as possible at a time 240 * to reiserfs which allows for fewer tree traversals. 241 242 * Each indirect pointer insertion takes a lot of cpu, because it involves memory moves inside of blocks. 243 244 * Asking the block allocation code for blocks one at a time is slightly less efficient. 245 246 All of these reasons for not using only generic file write were understood back when reiserfs was first miscoded to 247 use it, but we were in a hurry to make code freeze, and so it couldn't be revised then. This new code should make 248 things right finally. 249 250 Future Features: providing search_by_key with hints. 251 252*/ 253static ssize_t reiserfs_file_write(struct file *file, /* the file we are going to write into */ 254 const char __user * buf, /* pointer to user supplied data 255 (in userspace) */ 256 size_t count, /* amount of bytes to write */ 257 loff_t * ppos /* pointer to position in file that we start writing at. Should be updated to 258 * new current position before returning. */ 259 ) 260{ 261 struct inode *inode = file->f_path.dentry->d_inode; // Inode of the file that we are writing to. 262 /* To simplify coding at this time, we store 263 locked pages in array for now */ 264 struct reiserfs_transaction_handle th; 265 th.t_trans_id = 0; 266 267 /* If a filesystem is converted from 3.5 to 3.6, we'll have v3.5 items 268 * lying around (most of the disk, in fact). Despite the filesystem 269 * now being a v3.6 format, the old items still can't support large 270 * file sizes. Catch this case here, as the rest of the VFS layer is 271 * oblivious to the different limitations between old and new items. 272 * reiserfs_setattr catches this for truncates. This chunk is lifted 273 * from generic_write_checks. */ 274 if (get_inode_item_key_version (inode) == KEY_FORMAT_3_5 && 275 *ppos + count > MAX_NON_LFS) { 276 if (*ppos >= MAX_NON_LFS) { 277 return -EFBIG; 278 } 279 if (count > MAX_NON_LFS - (unsigned long)*ppos) 280 count = MAX_NON_LFS - (unsigned long)*ppos; 281 } 282 283 return do_sync_write(file, buf, count, ppos); 284} 285 286const struct file_operations reiserfs_file_operations = { 287 .read = do_sync_read, 288 .write = reiserfs_file_write, 289 .unlocked_ioctl = reiserfs_ioctl, 290#ifdef CONFIG_COMPAT 291 .compat_ioctl = reiserfs_compat_ioctl, 292#endif 293 .mmap = generic_file_mmap, 294 .open = reiserfs_file_open, 295 .release = reiserfs_file_release, 296 .fsync = reiserfs_sync_file, 297 .aio_read = generic_file_aio_read, 298 .aio_write = generic_file_aio_write, 299 .splice_read = generic_file_splice_read, 300 .splice_write = generic_file_splice_write, 301 .llseek = generic_file_llseek, 302}; 303 304const struct inode_operations reiserfs_file_inode_operations = { 305 .truncate = reiserfs_vfs_truncate_file, 306 .setattr = reiserfs_setattr, 307 .setxattr = reiserfs_setxattr, 308 .getxattr = reiserfs_getxattr, 309 .listxattr = reiserfs_listxattr, 310 .removexattr = reiserfs_removexattr, 311 .permission = reiserfs_permission, 312}; 313