Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1 | /* -*- mode: c; c-basic-offset: 8; -*- |
| 2 | * vim: noexpandtab sw=8 ts=8 sts=0: |
| 3 | * |
| 4 | * file.c |
| 5 | * |
| 6 | * File open, close, extend, truncate |
| 7 | * |
| 8 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. |
| 9 | * |
| 10 | * This program is free software; you can redistribute it and/or |
| 11 | * modify it under the terms of the GNU General Public |
| 12 | * License as published by the Free Software Foundation; either |
| 13 | * version 2 of the License, or (at your option) any later version. |
| 14 | * |
| 15 | * This program is distributed in the hope that it will be useful, |
| 16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 18 | * General Public License for more details. |
| 19 | * |
| 20 | * You should have received a copy of the GNU General Public |
| 21 | * License along with this program; if not, write to the |
| 22 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, |
| 23 | * Boston, MA 021110-1307, USA. |
| 24 | */ |
| 25 | |
Randy Dunlap | 16f7e0f | 2006-01-11 12:17:46 -0800 | [diff] [blame] | 26 | #include <linux/capability.h> |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 27 | #include <linux/fs.h> |
| 28 | #include <linux/types.h> |
| 29 | #include <linux/slab.h> |
| 30 | #include <linux/highmem.h> |
| 31 | #include <linux/pagemap.h> |
| 32 | #include <linux/uio.h> |
Mark Fasheh | e2057c5 | 2006-10-03 17:53:05 -0700 | [diff] [blame] | 33 | #include <linux/sched.h> |
Jens Axboe | d6b29d7 | 2007-06-04 09:59:47 +0200 | [diff] [blame] | 34 | #include <linux/splice.h> |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 35 | #include <linux/mount.h> |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 36 | #include <linux/writeback.h> |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 37 | #include <linux/falloc.h> |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 38 | |
| 39 | #define MLOG_MASK_PREFIX ML_INODE |
| 40 | #include <cluster/masklog.h> |
| 41 | |
| 42 | #include "ocfs2.h" |
| 43 | |
| 44 | #include "alloc.h" |
| 45 | #include "aops.h" |
| 46 | #include "dir.h" |
| 47 | #include "dlmglue.h" |
| 48 | #include "extent_map.h" |
| 49 | #include "file.h" |
| 50 | #include "sysfile.h" |
| 51 | #include "inode.h" |
Herbert Poetzl | ca4d147 | 2006-07-03 17:27:12 -0700 | [diff] [blame] | 52 | #include "ioctl.h" |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 53 | #include "journal.h" |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 54 | #include "locks.h" |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 55 | #include "mmap.h" |
| 56 | #include "suballoc.h" |
| 57 | #include "super.h" |
Tiger Yang | cf1d6c7 | 2008-08-18 17:11:00 +0800 | [diff] [blame] | 58 | #include "xattr.h" |
Tiger Yang | 23fc270 | 2008-11-14 11:17:18 +0800 | [diff] [blame^] | 59 | #include "acl.h" |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 60 | |
| 61 | #include "buffer_head_io.h" |
| 62 | |
| 63 | static int ocfs2_sync_inode(struct inode *inode) |
| 64 | { |
| 65 | filemap_fdatawrite(inode->i_mapping); |
| 66 | return sync_mapping_buffers(inode->i_mapping); |
| 67 | } |
| 68 | |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 69 | static int ocfs2_init_file_private(struct inode *inode, struct file *file) |
| 70 | { |
| 71 | struct ocfs2_file_private *fp; |
| 72 | |
| 73 | fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL); |
| 74 | if (!fp) |
| 75 | return -ENOMEM; |
| 76 | |
| 77 | fp->fp_file = file; |
| 78 | mutex_init(&fp->fp_mutex); |
| 79 | ocfs2_file_lock_res_init(&fp->fp_flock, fp); |
| 80 | file->private_data = fp; |
| 81 | |
| 82 | return 0; |
| 83 | } |
| 84 | |
| 85 | static void ocfs2_free_file_private(struct inode *inode, struct file *file) |
| 86 | { |
| 87 | struct ocfs2_file_private *fp = file->private_data; |
| 88 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 89 | |
| 90 | if (fp) { |
| 91 | ocfs2_simple_drop_lockres(osb, &fp->fp_flock); |
| 92 | ocfs2_lock_res_free(&fp->fp_flock); |
| 93 | kfree(fp); |
| 94 | file->private_data = NULL; |
| 95 | } |
| 96 | } |
| 97 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 98 | static int ocfs2_file_open(struct inode *inode, struct file *file) |
| 99 | { |
| 100 | int status; |
| 101 | int mode = file->f_flags; |
| 102 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 103 | |
| 104 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 105 | file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 106 | |
| 107 | spin_lock(&oi->ip_lock); |
| 108 | |
| 109 | /* Check that the inode hasn't been wiped from disk by another |
| 110 | * node. If it hasn't then we're safe as long as we hold the |
| 111 | * spin lock until our increment of open count. */ |
| 112 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { |
| 113 | spin_unlock(&oi->ip_lock); |
| 114 | |
| 115 | status = -ENOENT; |
| 116 | goto leave; |
| 117 | } |
| 118 | |
| 119 | if (mode & O_DIRECT) |
| 120 | oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT; |
| 121 | |
| 122 | oi->ip_open_count++; |
| 123 | spin_unlock(&oi->ip_lock); |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 124 | |
| 125 | status = ocfs2_init_file_private(inode, file); |
| 126 | if (status) { |
| 127 | /* |
| 128 | * We want to set open count back if we're failing the |
| 129 | * open. |
| 130 | */ |
| 131 | spin_lock(&oi->ip_lock); |
| 132 | oi->ip_open_count--; |
| 133 | spin_unlock(&oi->ip_lock); |
| 134 | } |
| 135 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 136 | leave: |
| 137 | mlog_exit(status); |
| 138 | return status; |
| 139 | } |
| 140 | |
| 141 | static int ocfs2_file_release(struct inode *inode, struct file *file) |
| 142 | { |
| 143 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 144 | |
| 145 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 146 | file->f_path.dentry->d_name.len, |
| 147 | file->f_path.dentry->d_name.name); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 148 | |
| 149 | spin_lock(&oi->ip_lock); |
| 150 | if (!--oi->ip_open_count) |
| 151 | oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT; |
| 152 | spin_unlock(&oi->ip_lock); |
| 153 | |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 154 | ocfs2_free_file_private(inode, file); |
| 155 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 156 | mlog_exit(0); |
| 157 | |
| 158 | return 0; |
| 159 | } |
| 160 | |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 161 | static int ocfs2_dir_open(struct inode *inode, struct file *file) |
| 162 | { |
| 163 | return ocfs2_init_file_private(inode, file); |
| 164 | } |
| 165 | |
| 166 | static int ocfs2_dir_release(struct inode *inode, struct file *file) |
| 167 | { |
| 168 | ocfs2_free_file_private(inode, file); |
| 169 | return 0; |
| 170 | } |
| 171 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 172 | static int ocfs2_sync_file(struct file *file, |
| 173 | struct dentry *dentry, |
| 174 | int datasync) |
| 175 | { |
| 176 | int err = 0; |
| 177 | journal_t *journal; |
| 178 | struct inode *inode = dentry->d_inode; |
| 179 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 180 | |
| 181 | mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync, |
| 182 | dentry->d_name.len, dentry->d_name.name); |
| 183 | |
| 184 | err = ocfs2_sync_inode(dentry->d_inode); |
| 185 | if (err) |
| 186 | goto bail; |
| 187 | |
| 188 | journal = osb->journal->j_journal; |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 189 | err = jbd2_journal_force_commit(journal); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 190 | |
| 191 | bail: |
| 192 | mlog_exit(err); |
| 193 | |
| 194 | return (err < 0) ? -EIO : 0; |
| 195 | } |
| 196 | |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 197 | int ocfs2_should_update_atime(struct inode *inode, |
| 198 | struct vfsmount *vfsmnt) |
| 199 | { |
| 200 | struct timespec now; |
| 201 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 202 | |
| 203 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) |
| 204 | return 0; |
| 205 | |
| 206 | if ((inode->i_flags & S_NOATIME) || |
| 207 | ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))) |
| 208 | return 0; |
| 209 | |
Mark Fasheh | 6c2aad0 | 2006-12-19 15:25:52 -0800 | [diff] [blame] | 210 | /* |
| 211 | * We can be called with no vfsmnt structure - NFSD will |
| 212 | * sometimes do this. |
| 213 | * |
| 214 | * Note that our action here is different than touch_atime() - |
| 215 | * if we can't tell whether this is a noatime mount, then we |
| 216 | * don't know whether to trust the value of s_atime_quantum. |
| 217 | */ |
| 218 | if (vfsmnt == NULL) |
| 219 | return 0; |
| 220 | |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 221 | if ((vfsmnt->mnt_flags & MNT_NOATIME) || |
| 222 | ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) |
| 223 | return 0; |
| 224 | |
Mark Fasheh | 7e913c5 | 2006-12-13 00:34:35 -0800 | [diff] [blame] | 225 | if (vfsmnt->mnt_flags & MNT_RELATIME) { |
| 226 | if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) || |
| 227 | (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0)) |
| 228 | return 1; |
| 229 | |
| 230 | return 0; |
| 231 | } |
| 232 | |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 233 | now = CURRENT_TIME; |
| 234 | if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum)) |
| 235 | return 0; |
| 236 | else |
| 237 | return 1; |
| 238 | } |
| 239 | |
| 240 | int ocfs2_update_inode_atime(struct inode *inode, |
| 241 | struct buffer_head *bh) |
| 242 | { |
| 243 | int ret; |
| 244 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 245 | handle_t *handle; |
Mark Fasheh | c11e9fa | 2007-07-20 11:24:53 -0700 | [diff] [blame] | 246 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data; |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 247 | |
| 248 | mlog_entry_void(); |
| 249 | |
| 250 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
Jan Kara | fa38e92 | 2008-10-20 19:23:51 +0200 | [diff] [blame] | 251 | if (IS_ERR(handle)) { |
| 252 | ret = PTR_ERR(handle); |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 253 | mlog_errno(ret); |
| 254 | goto out; |
| 255 | } |
| 256 | |
Mark Fasheh | c11e9fa | 2007-07-20 11:24:53 -0700 | [diff] [blame] | 257 | ret = ocfs2_journal_access(handle, inode, bh, |
| 258 | OCFS2_JOURNAL_ACCESS_WRITE); |
| 259 | if (ret) { |
| 260 | mlog_errno(ret); |
| 261 | goto out_commit; |
| 262 | } |
| 263 | |
| 264 | /* |
| 265 | * Don't use ocfs2_mark_inode_dirty() here as we don't always |
| 266 | * have i_mutex to guard against concurrent changes to other |
| 267 | * inode fields. |
| 268 | */ |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 269 | inode->i_atime = CURRENT_TIME; |
Mark Fasheh | c11e9fa | 2007-07-20 11:24:53 -0700 | [diff] [blame] | 270 | di->i_atime = cpu_to_le64(inode->i_atime.tv_sec); |
| 271 | di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec); |
| 272 | |
| 273 | ret = ocfs2_journal_dirty(handle, bh); |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 274 | if (ret < 0) |
| 275 | mlog_errno(ret); |
| 276 | |
Mark Fasheh | c11e9fa | 2007-07-20 11:24:53 -0700 | [diff] [blame] | 277 | out_commit: |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 278 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); |
| 279 | out: |
| 280 | mlog_exit(ret); |
| 281 | return ret; |
| 282 | } |
| 283 | |
Adrian Bunk | 6cb129f | 2007-04-26 00:29:35 -0700 | [diff] [blame] | 284 | static int ocfs2_set_inode_size(handle_t *handle, |
| 285 | struct inode *inode, |
| 286 | struct buffer_head *fe_bh, |
| 287 | u64 new_i_size) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 288 | { |
| 289 | int status; |
| 290 | |
| 291 | mlog_entry_void(); |
| 292 | i_size_write(inode, new_i_size); |
Mark Fasheh | 8110b07 | 2007-03-22 16:53:23 -0700 | [diff] [blame] | 293 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 294 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
| 295 | |
| 296 | status = ocfs2_mark_inode_dirty(handle, inode, fe_bh); |
| 297 | if (status < 0) { |
| 298 | mlog_errno(status); |
| 299 | goto bail; |
| 300 | } |
| 301 | |
| 302 | bail: |
| 303 | mlog_exit(status); |
| 304 | return status; |
| 305 | } |
| 306 | |
| 307 | static int ocfs2_simple_size_update(struct inode *inode, |
| 308 | struct buffer_head *di_bh, |
| 309 | u64 new_i_size) |
| 310 | { |
| 311 | int ret; |
| 312 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 313 | handle_t *handle = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 314 | |
Mark Fasheh | 65eff9c | 2006-10-09 17:26:22 -0700 | [diff] [blame] | 315 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
Jan Kara | fa38e92 | 2008-10-20 19:23:51 +0200 | [diff] [blame] | 316 | if (IS_ERR(handle)) { |
| 317 | ret = PTR_ERR(handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 318 | mlog_errno(ret); |
| 319 | goto out; |
| 320 | } |
| 321 | |
| 322 | ret = ocfs2_set_inode_size(handle, inode, di_bh, |
| 323 | new_i_size); |
| 324 | if (ret < 0) |
| 325 | mlog_errno(ret); |
| 326 | |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 327 | ocfs2_commit_trans(osb, handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 328 | out: |
| 329 | return ret; |
| 330 | } |
| 331 | |
| 332 | static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb, |
| 333 | struct inode *inode, |
| 334 | struct buffer_head *fe_bh, |
| 335 | u64 new_i_size) |
| 336 | { |
| 337 | int status; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 338 | handle_t *handle; |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 339 | struct ocfs2_dinode *di; |
Mark Fasheh | 35edec1 | 2007-07-06 14:41:18 -0700 | [diff] [blame] | 340 | u64 cluster_bytes; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 341 | |
| 342 | mlog_entry_void(); |
| 343 | |
| 344 | /* TODO: This needs to actually orphan the inode in this |
| 345 | * transaction. */ |
| 346 | |
Mark Fasheh | 65eff9c | 2006-10-09 17:26:22 -0700 | [diff] [blame] | 347 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 348 | if (IS_ERR(handle)) { |
| 349 | status = PTR_ERR(handle); |
| 350 | mlog_errno(status); |
| 351 | goto out; |
| 352 | } |
| 353 | |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 354 | status = ocfs2_journal_access(handle, inode, fe_bh, |
| 355 | OCFS2_JOURNAL_ACCESS_WRITE); |
| 356 | if (status < 0) { |
| 357 | mlog_errno(status); |
| 358 | goto out_commit; |
| 359 | } |
| 360 | |
| 361 | /* |
| 362 | * Do this before setting i_size. |
| 363 | */ |
Mark Fasheh | 35edec1 | 2007-07-06 14:41:18 -0700 | [diff] [blame] | 364 | cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size); |
| 365 | status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size, |
| 366 | cluster_bytes); |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 367 | if (status) { |
| 368 | mlog_errno(status); |
| 369 | goto out_commit; |
| 370 | } |
| 371 | |
| 372 | i_size_write(inode, new_i_size); |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 373 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
| 374 | |
| 375 | di = (struct ocfs2_dinode *) fe_bh->b_data; |
| 376 | di->i_size = cpu_to_le64(new_i_size); |
| 377 | di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec); |
| 378 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); |
| 379 | |
| 380 | status = ocfs2_journal_dirty(handle, fe_bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 381 | if (status < 0) |
| 382 | mlog_errno(status); |
| 383 | |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 384 | out_commit: |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 385 | ocfs2_commit_trans(osb, handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 386 | out: |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 387 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 388 | mlog_exit(status); |
| 389 | return status; |
| 390 | } |
| 391 | |
| 392 | static int ocfs2_truncate_file(struct inode *inode, |
| 393 | struct buffer_head *di_bh, |
| 394 | u64 new_i_size) |
| 395 | { |
| 396 | int status = 0; |
| 397 | struct ocfs2_dinode *fe = NULL; |
| 398 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 399 | struct ocfs2_truncate_context *tc = NULL; |
| 400 | |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 401 | mlog_entry("(inode = %llu, new_i_size = %llu\n", |
| 402 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
| 403 | (unsigned long long)new_i_size); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 404 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 405 | fe = (struct ocfs2_dinode *) di_bh->b_data; |
| 406 | if (!OCFS2_IS_VALID_DINODE(fe)) { |
| 407 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); |
| 408 | status = -EIO; |
| 409 | goto bail; |
| 410 | } |
| 411 | |
| 412 | mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode), |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 413 | "Inode %llu, inode i_size = %lld != di " |
| 414 | "i_size = %llu, i_flags = 0x%x\n", |
| 415 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 416 | i_size_read(inode), |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 417 | (unsigned long long)le64_to_cpu(fe->i_size), |
| 418 | le32_to_cpu(fe->i_flags)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 419 | |
| 420 | if (new_i_size > le64_to_cpu(fe->i_size)) { |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 421 | mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n", |
| 422 | (unsigned long long)le64_to_cpu(fe->i_size), |
| 423 | (unsigned long long)new_i_size); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 424 | status = -EINVAL; |
| 425 | mlog_errno(status); |
| 426 | goto bail; |
| 427 | } |
| 428 | |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 429 | mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n", |
| 430 | (unsigned long long)le64_to_cpu(fe->i_blkno), |
| 431 | (unsigned long long)le64_to_cpu(fe->i_size), |
| 432 | (unsigned long long)new_i_size); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 433 | |
| 434 | /* lets handle the simple truncate cases before doing any more |
| 435 | * cluster locking. */ |
| 436 | if (new_i_size == le64_to_cpu(fe->i_size)) |
| 437 | goto bail; |
| 438 | |
Mark Fasheh | 2e89b2e | 2007-05-09 13:40:18 -0700 | [diff] [blame] | 439 | down_write(&OCFS2_I(inode)->ip_alloc_sem); |
| 440 | |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 441 | /* |
| 442 | * The inode lock forced other nodes to sync and drop their |
| 443 | * pages, which (correctly) happens even if we have a truncate |
| 444 | * without allocation change - ocfs2 cluster sizes can be much |
| 445 | * greater than page size, so we have to truncate them |
| 446 | * anyway. |
| 447 | */ |
Mark Fasheh | 2e89b2e | 2007-05-09 13:40:18 -0700 | [diff] [blame] | 448 | unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1); |
| 449 | truncate_inode_pages(inode->i_mapping, new_i_size); |
| 450 | |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 451 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
| 452 | status = ocfs2_truncate_inline(inode, di_bh, new_i_size, |
Mark Fasheh | b1967d0 | 2007-11-20 11:56:39 -0800 | [diff] [blame] | 453 | i_size_read(inode), 1); |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 454 | if (status) |
| 455 | mlog_errno(status); |
| 456 | |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 457 | goto bail_unlock_sem; |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 458 | } |
| 459 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 460 | /* alright, we're going to need to do a full blown alloc size |
| 461 | * change. Orphan the inode so that recovery can complete the |
| 462 | * truncate if necessary. This does the task of marking |
| 463 | * i_size. */ |
| 464 | status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size); |
| 465 | if (status < 0) { |
| 466 | mlog_errno(status); |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 467 | goto bail_unlock_sem; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 468 | } |
| 469 | |
| 470 | status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc); |
| 471 | if (status < 0) { |
| 472 | mlog_errno(status); |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 473 | goto bail_unlock_sem; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 474 | } |
| 475 | |
| 476 | status = ocfs2_commit_truncate(osb, inode, di_bh, tc); |
| 477 | if (status < 0) { |
| 478 | mlog_errno(status); |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 479 | goto bail_unlock_sem; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 480 | } |
| 481 | |
| 482 | /* TODO: orphan dir cleanup here. */ |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 483 | bail_unlock_sem: |
Mark Fasheh | 2e89b2e | 2007-05-09 13:40:18 -0700 | [diff] [blame] | 484 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
| 485 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 486 | bail: |
| 487 | |
| 488 | mlog_exit(status); |
| 489 | return status; |
| 490 | } |
| 491 | |
| 492 | /* |
Tao Ma | 0eb8d47 | 2008-08-18 17:38:45 +0800 | [diff] [blame] | 493 | * extend file allocation only here. |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 494 | * we'll update all the disk stuff, and oip->alloc_size |
| 495 | * |
| 496 | * expect stuff to be locked, a transaction started and enough data / |
| 497 | * metadata reservations in the contexts. |
| 498 | * |
| 499 | * Will return -EAGAIN, and a reason if a restart is needed. |
| 500 | * If passed in, *reason will always be set, even in error. |
| 501 | */ |
Tao Ma | 0eb8d47 | 2008-08-18 17:38:45 +0800 | [diff] [blame] | 502 | int ocfs2_add_inode_data(struct ocfs2_super *osb, |
| 503 | struct inode *inode, |
| 504 | u32 *logical_offset, |
| 505 | u32 clusters_to_add, |
| 506 | int mark_unwritten, |
| 507 | struct buffer_head *fe_bh, |
| 508 | handle_t *handle, |
| 509 | struct ocfs2_alloc_context *data_ac, |
| 510 | struct ocfs2_alloc_context *meta_ac, |
| 511 | enum ocfs2_alloc_restarted *reason_ret) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 512 | { |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 513 | int ret; |
| 514 | struct ocfs2_extent_tree et; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 515 | |
Joel Becker | 8d6220d | 2008-08-22 12:46:09 -0700 | [diff] [blame] | 516 | ocfs2_init_dinode_extent_tree(&et, inode, fe_bh); |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 517 | ret = ocfs2_add_clusters_in_btree(osb, inode, logical_offset, |
Tao Ma | 0eb8d47 | 2008-08-18 17:38:45 +0800 | [diff] [blame] | 518 | clusters_to_add, mark_unwritten, |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 519 | &et, handle, |
| 520 | data_ac, meta_ac, reason_ret); |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 521 | |
| 522 | return ret; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 523 | } |
| 524 | |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 525 | static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start, |
| 526 | u32 clusters_to_add, int mark_unwritten) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 527 | { |
| 528 | int status = 0; |
| 529 | int restart_func = 0; |
Mark Fasheh | abf8b15 | 2007-01-17 13:07:24 -0800 | [diff] [blame] | 530 | int credits; |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 531 | u32 prev_clusters; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 532 | struct buffer_head *bh = NULL; |
| 533 | struct ocfs2_dinode *fe = NULL; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 534 | handle_t *handle = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 535 | struct ocfs2_alloc_context *data_ac = NULL; |
| 536 | struct ocfs2_alloc_context *meta_ac = NULL; |
| 537 | enum ocfs2_alloc_restarted why; |
| 538 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 539 | struct ocfs2_extent_tree et; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 540 | |
| 541 | mlog_entry("(clusters_to_add = %u)\n", clusters_to_add); |
| 542 | |
Mark Fasheh | dcd0538 | 2007-01-16 11:32:23 -0800 | [diff] [blame] | 543 | /* |
| 544 | * This function only exists for file systems which don't |
| 545 | * support holes. |
| 546 | */ |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 547 | BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb)); |
Mark Fasheh | dcd0538 | 2007-01-16 11:32:23 -0800 | [diff] [blame] | 548 | |
Joel Becker | 0fcaa56 | 2008-10-09 17:20:31 -0700 | [diff] [blame] | 549 | status = ocfs2_read_block(inode, OCFS2_I(inode)->ip_blkno, &bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 550 | if (status < 0) { |
| 551 | mlog_errno(status); |
| 552 | goto leave; |
| 553 | } |
| 554 | |
| 555 | fe = (struct ocfs2_dinode *) bh->b_data; |
| 556 | if (!OCFS2_IS_VALID_DINODE(fe)) { |
| 557 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); |
| 558 | status = -EIO; |
| 559 | goto leave; |
| 560 | } |
| 561 | |
| 562 | restart_all: |
| 563 | BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters); |
| 564 | |
Tao Ma | e7d4cb6 | 2008-08-18 17:38:44 +0800 | [diff] [blame] | 565 | mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, " |
| 566 | "clusters_to_add = %u\n", |
| 567 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
| 568 | (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters), |
| 569 | clusters_to_add); |
Joel Becker | 8d6220d | 2008-08-22 12:46:09 -0700 | [diff] [blame] | 570 | ocfs2_init_dinode_extent_tree(&et, inode, bh); |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 571 | status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0, |
| 572 | &data_ac, &meta_ac); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 573 | if (status) { |
| 574 | mlog_errno(status); |
| 575 | goto leave; |
| 576 | } |
| 577 | |
Tao Ma | 811f933 | 2008-08-18 17:38:43 +0800 | [diff] [blame] | 578 | credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list, |
| 579 | clusters_to_add); |
Mark Fasheh | 65eff9c | 2006-10-09 17:26:22 -0700 | [diff] [blame] | 580 | handle = ocfs2_start_trans(osb, credits); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 581 | if (IS_ERR(handle)) { |
| 582 | status = PTR_ERR(handle); |
| 583 | handle = NULL; |
| 584 | mlog_errno(status); |
| 585 | goto leave; |
| 586 | } |
| 587 | |
| 588 | restarted_transaction: |
| 589 | /* reserve a write to the file entry early on - that we if we |
| 590 | * run out of credits in the allocation path, we can still |
| 591 | * update i_size. */ |
| 592 | status = ocfs2_journal_access(handle, inode, bh, |
| 593 | OCFS2_JOURNAL_ACCESS_WRITE); |
| 594 | if (status < 0) { |
| 595 | mlog_errno(status); |
| 596 | goto leave; |
| 597 | } |
| 598 | |
| 599 | prev_clusters = OCFS2_I(inode)->ip_clusters; |
| 600 | |
Tao Ma | 0eb8d47 | 2008-08-18 17:38:45 +0800 | [diff] [blame] | 601 | status = ocfs2_add_inode_data(osb, |
| 602 | inode, |
| 603 | &logical_start, |
| 604 | clusters_to_add, |
| 605 | mark_unwritten, |
| 606 | bh, |
| 607 | handle, |
| 608 | data_ac, |
| 609 | meta_ac, |
| 610 | &why); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 611 | if ((status < 0) && (status != -EAGAIN)) { |
| 612 | if (status != -ENOSPC) |
| 613 | mlog_errno(status); |
| 614 | goto leave; |
| 615 | } |
| 616 | |
| 617 | status = ocfs2_journal_dirty(handle, bh); |
| 618 | if (status < 0) { |
| 619 | mlog_errno(status); |
| 620 | goto leave; |
| 621 | } |
| 622 | |
| 623 | spin_lock(&OCFS2_I(inode)->ip_lock); |
| 624 | clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters); |
| 625 | spin_unlock(&OCFS2_I(inode)->ip_lock); |
| 626 | |
| 627 | if (why != RESTART_NONE && clusters_to_add) { |
| 628 | if (why == RESTART_META) { |
| 629 | mlog(0, "restarting function.\n"); |
| 630 | restart_func = 1; |
| 631 | } else { |
| 632 | BUG_ON(why != RESTART_TRANS); |
| 633 | |
| 634 | mlog(0, "restarting transaction.\n"); |
| 635 | /* TODO: This can be more intelligent. */ |
| 636 | credits = ocfs2_calc_extend_credits(osb->sb, |
Tao Ma | 811f933 | 2008-08-18 17:38:43 +0800 | [diff] [blame] | 637 | &fe->id2.i_list, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 638 | clusters_to_add); |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 639 | status = ocfs2_extend_trans(handle, credits); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 640 | if (status < 0) { |
| 641 | /* handle still has to be committed at |
| 642 | * this point. */ |
| 643 | status = -ENOMEM; |
| 644 | mlog_errno(status); |
| 645 | goto leave; |
| 646 | } |
| 647 | goto restarted_transaction; |
| 648 | } |
| 649 | } |
| 650 | |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 651 | mlog(0, "fe: i_clusters = %u, i_size=%llu\n", |
Mark Fasheh | 1ca1a11 | 2007-04-27 16:01:25 -0700 | [diff] [blame] | 652 | le32_to_cpu(fe->i_clusters), |
| 653 | (unsigned long long)le64_to_cpu(fe->i_size)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 654 | mlog(0, "inode: ip_clusters=%u, i_size=%lld\n", |
Jan Kara | 634bf74 | 2007-12-19 15:25:42 +0100 | [diff] [blame] | 655 | OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 656 | |
| 657 | leave: |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 658 | if (handle) { |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 659 | ocfs2_commit_trans(osb, handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 660 | handle = NULL; |
| 661 | } |
| 662 | if (data_ac) { |
| 663 | ocfs2_free_alloc_context(data_ac); |
| 664 | data_ac = NULL; |
| 665 | } |
| 666 | if (meta_ac) { |
| 667 | ocfs2_free_alloc_context(meta_ac); |
| 668 | meta_ac = NULL; |
| 669 | } |
| 670 | if ((!status) && restart_func) { |
| 671 | restart_func = 0; |
| 672 | goto restart_all; |
| 673 | } |
Mark Fasheh | a81cb88 | 2008-10-07 14:25:16 -0700 | [diff] [blame] | 674 | brelse(bh); |
| 675 | bh = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 676 | |
| 677 | mlog_exit(status); |
| 678 | return status; |
| 679 | } |
| 680 | |
| 681 | /* Some parts of this taken from generic_cont_expand, which turned out |
| 682 | * to be too fragile to do exactly what we need without us having to |
Nick Piggin | 4e02ed4 | 2008-10-29 14:00:55 -0700 | [diff] [blame] | 683 | * worry about recursive locking in ->write_begin() and ->write_end(). */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 684 | static int ocfs2_write_zero_page(struct inode *inode, |
| 685 | u64 size) |
| 686 | { |
| 687 | struct address_space *mapping = inode->i_mapping; |
| 688 | struct page *page; |
| 689 | unsigned long index; |
| 690 | unsigned int offset; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 691 | handle_t *handle = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 692 | int ret; |
| 693 | |
| 694 | offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */ |
| 695 | /* ugh. in prepare/commit_write, if from==to==start of block, we |
| 696 | ** skip the prepare. make sure we never send an offset for the start |
| 697 | ** of a block |
| 698 | */ |
| 699 | if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) { |
| 700 | offset++; |
| 701 | } |
| 702 | index = size >> PAGE_CACHE_SHIFT; |
| 703 | |
| 704 | page = grab_cache_page(mapping, index); |
| 705 | if (!page) { |
| 706 | ret = -ENOMEM; |
| 707 | mlog_errno(ret); |
| 708 | goto out; |
| 709 | } |
| 710 | |
Mark Fasheh | 53013cb | 2006-05-05 19:04:03 -0700 | [diff] [blame] | 711 | ret = ocfs2_prepare_write_nolock(inode, page, offset, offset); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 712 | if (ret < 0) { |
| 713 | mlog_errno(ret); |
| 714 | goto out_unlock; |
| 715 | } |
| 716 | |
| 717 | if (ocfs2_should_order_data(inode)) { |
| 718 | handle = ocfs2_start_walk_page_trans(inode, page, offset, |
| 719 | offset); |
| 720 | if (IS_ERR(handle)) { |
| 721 | ret = PTR_ERR(handle); |
| 722 | handle = NULL; |
| 723 | goto out_unlock; |
| 724 | } |
| 725 | } |
| 726 | |
| 727 | /* must not update i_size! */ |
| 728 | ret = block_commit_write(page, offset, offset); |
| 729 | if (ret < 0) |
| 730 | mlog_errno(ret); |
| 731 | else |
| 732 | ret = 0; |
| 733 | |
| 734 | if (handle) |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 735 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 736 | out_unlock: |
| 737 | unlock_page(page); |
| 738 | page_cache_release(page); |
| 739 | out: |
| 740 | return ret; |
| 741 | } |
| 742 | |
| 743 | static int ocfs2_zero_extend(struct inode *inode, |
| 744 | u64 zero_to_size) |
| 745 | { |
| 746 | int ret = 0; |
| 747 | u64 start_off; |
| 748 | struct super_block *sb = inode->i_sb; |
| 749 | |
| 750 | start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode)); |
| 751 | while (start_off < zero_to_size) { |
| 752 | ret = ocfs2_write_zero_page(inode, start_off); |
| 753 | if (ret < 0) { |
| 754 | mlog_errno(ret); |
| 755 | goto out; |
| 756 | } |
| 757 | |
| 758 | start_off += sb->s_blocksize; |
Mark Fasheh | e2057c5 | 2006-10-03 17:53:05 -0700 | [diff] [blame] | 759 | |
| 760 | /* |
| 761 | * Very large extends have the potential to lock up |
| 762 | * the cpu for extended periods of time. |
| 763 | */ |
| 764 | cond_resched(); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 765 | } |
| 766 | |
| 767 | out: |
| 768 | return ret; |
| 769 | } |
| 770 | |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 771 | int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to) |
| 772 | { |
| 773 | int ret; |
| 774 | u32 clusters_to_add; |
| 775 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 776 | |
| 777 | clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size); |
| 778 | if (clusters_to_add < oi->ip_clusters) |
| 779 | clusters_to_add = 0; |
| 780 | else |
| 781 | clusters_to_add -= oi->ip_clusters; |
| 782 | |
| 783 | if (clusters_to_add) { |
| 784 | ret = __ocfs2_extend_allocation(inode, oi->ip_clusters, |
| 785 | clusters_to_add, 0); |
| 786 | if (ret) { |
| 787 | mlog_errno(ret); |
| 788 | goto out; |
| 789 | } |
| 790 | } |
| 791 | |
| 792 | /* |
| 793 | * Call this even if we don't add any clusters to the tree. We |
| 794 | * still need to zero the area between the old i_size and the |
| 795 | * new i_size. |
| 796 | */ |
| 797 | ret = ocfs2_zero_extend(inode, zero_to); |
| 798 | if (ret < 0) |
| 799 | mlog_errno(ret); |
| 800 | |
| 801 | out: |
| 802 | return ret; |
| 803 | } |
| 804 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 805 | static int ocfs2_extend_file(struct inode *inode, |
| 806 | struct buffer_head *di_bh, |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 807 | u64 new_i_size) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 808 | { |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 809 | int ret = 0; |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 810 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 811 | |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 812 | BUG_ON(!di_bh); |
Mark Fasheh | 53013cb | 2006-05-05 19:04:03 -0700 | [diff] [blame] | 813 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 814 | /* setattr sometimes calls us like this. */ |
| 815 | if (new_i_size == 0) |
| 816 | goto out; |
| 817 | |
| 818 | if (i_size_read(inode) == new_i_size) |
| 819 | goto out; |
| 820 | BUG_ON(new_i_size < i_size_read(inode)); |
| 821 | |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 822 | /* |
| 823 | * Fall through for converting inline data, even if the fs |
| 824 | * supports sparse files. |
| 825 | * |
| 826 | * The check for inline data here is legal - nobody can add |
| 827 | * the feature since we have i_mutex. We must check it again |
| 828 | * after acquiring ip_alloc_sem though, as paths like mmap |
| 829 | * might have raced us to converting the inode to extents. |
| 830 | */ |
| 831 | if (!(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) |
| 832 | && ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) |
Mark Fasheh | 3a0782d | 2007-01-17 12:53:31 -0800 | [diff] [blame] | 833 | goto out_update_size; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 834 | |
Mark Fasheh | 0effef7 | 2006-10-03 17:44:42 -0700 | [diff] [blame] | 835 | /* |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 836 | * The alloc sem blocks people in read/write from reading our |
| 837 | * allocation until we're done changing it. We depend on |
| 838 | * i_mutex to block other extend/truncate calls while we're |
| 839 | * here. |
Mark Fasheh | 0effef7 | 2006-10-03 17:44:42 -0700 | [diff] [blame] | 840 | */ |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 841 | down_write(&oi->ip_alloc_sem); |
| 842 | |
| 843 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
| 844 | /* |
| 845 | * We can optimize small extends by keeping the inodes |
| 846 | * inline data. |
| 847 | */ |
| 848 | if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) { |
| 849 | up_write(&oi->ip_alloc_sem); |
| 850 | goto out_update_size; |
| 851 | } |
| 852 | |
| 853 | ret = ocfs2_convert_inline_data_to_extents(inode, di_bh); |
| 854 | if (ret) { |
| 855 | up_write(&oi->ip_alloc_sem); |
| 856 | |
| 857 | mlog_errno(ret); |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 858 | goto out; |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 859 | } |
| 860 | } |
| 861 | |
| 862 | if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) |
| 863 | ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size); |
| 864 | |
| 865 | up_write(&oi->ip_alloc_sem); |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 866 | |
Mark Fasheh | 0effef7 | 2006-10-03 17:44:42 -0700 | [diff] [blame] | 867 | if (ret < 0) { |
| 868 | mlog_errno(ret); |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 869 | goto out; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 870 | } |
| 871 | |
Mark Fasheh | 3a0782d | 2007-01-17 12:53:31 -0800 | [diff] [blame] | 872 | out_update_size: |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 873 | ret = ocfs2_simple_size_update(inode, di_bh, new_i_size); |
| 874 | if (ret < 0) |
| 875 | mlog_errno(ret); |
Mark Fasheh | 53013cb | 2006-05-05 19:04:03 -0700 | [diff] [blame] | 876 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 877 | out: |
| 878 | return ret; |
| 879 | } |
| 880 | |
| 881 | int ocfs2_setattr(struct dentry *dentry, struct iattr *attr) |
| 882 | { |
| 883 | int status = 0, size_change; |
| 884 | struct inode *inode = dentry->d_inode; |
| 885 | struct super_block *sb = inode->i_sb; |
| 886 | struct ocfs2_super *osb = OCFS2_SB(sb); |
| 887 | struct buffer_head *bh = NULL; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 888 | handle_t *handle = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 889 | |
| 890 | mlog_entry("(0x%p, '%.*s')\n", dentry, |
| 891 | dentry->d_name.len, dentry->d_name.name); |
| 892 | |
Sunil Mushran | bc53580 | 2008-04-18 10:23:53 -0700 | [diff] [blame] | 893 | /* ensuring we don't even attempt to truncate a symlink */ |
| 894 | if (S_ISLNK(inode->i_mode)) |
| 895 | attr->ia_valid &= ~ATTR_SIZE; |
| 896 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 897 | if (attr->ia_valid & ATTR_MODE) |
| 898 | mlog(0, "mode change: %d\n", attr->ia_mode); |
| 899 | if (attr->ia_valid & ATTR_UID) |
| 900 | mlog(0, "uid change: %d\n", attr->ia_uid); |
| 901 | if (attr->ia_valid & ATTR_GID) |
| 902 | mlog(0, "gid change: %d\n", attr->ia_gid); |
| 903 | if (attr->ia_valid & ATTR_SIZE) |
| 904 | mlog(0, "size change...\n"); |
| 905 | if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME)) |
| 906 | mlog(0, "time change...\n"); |
| 907 | |
| 908 | #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \ |
| 909 | | ATTR_GID | ATTR_UID | ATTR_MODE) |
| 910 | if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) { |
| 911 | mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid); |
| 912 | return 0; |
| 913 | } |
| 914 | |
| 915 | status = inode_change_ok(inode, attr); |
| 916 | if (status) |
| 917 | return status; |
| 918 | |
| 919 | size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE; |
| 920 | if (size_change) { |
| 921 | status = ocfs2_rw_lock(inode, 1); |
| 922 | if (status < 0) { |
| 923 | mlog_errno(status); |
| 924 | goto bail; |
| 925 | } |
| 926 | } |
| 927 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 928 | status = ocfs2_inode_lock(inode, &bh, 1); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 929 | if (status < 0) { |
| 930 | if (status != -ENOENT) |
| 931 | mlog_errno(status); |
| 932 | goto bail_unlock_rw; |
| 933 | } |
| 934 | |
| 935 | if (size_change && attr->ia_size != i_size_read(inode)) { |
Mark Fasheh | ce76fd3 | 2007-07-20 12:02:14 -0700 | [diff] [blame] | 936 | if (attr->ia_size > sb->s_maxbytes) { |
| 937 | status = -EFBIG; |
| 938 | goto bail_unlock; |
| 939 | } |
| 940 | |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 941 | if (i_size_read(inode) > attr->ia_size) { |
| 942 | if (ocfs2_should_order_data(inode)) { |
| 943 | status = ocfs2_begin_ordered_truncate(inode, |
| 944 | attr->ia_size); |
| 945 | if (status) |
| 946 | goto bail_unlock; |
| 947 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 948 | status = ocfs2_truncate_file(inode, bh, attr->ia_size); |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 949 | } else |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 950 | status = ocfs2_extend_file(inode, bh, attr->ia_size); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 951 | if (status < 0) { |
| 952 | if (status != -ENOSPC) |
| 953 | mlog_errno(status); |
| 954 | status = -ENOSPC; |
| 955 | goto bail_unlock; |
| 956 | } |
| 957 | } |
| 958 | |
Mark Fasheh | 65eff9c | 2006-10-09 17:26:22 -0700 | [diff] [blame] | 959 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 960 | if (IS_ERR(handle)) { |
| 961 | status = PTR_ERR(handle); |
| 962 | mlog_errno(status); |
| 963 | goto bail_unlock; |
| 964 | } |
| 965 | |
Mark Fasheh | 7307de8 | 2007-05-09 15:16:19 -0700 | [diff] [blame] | 966 | /* |
| 967 | * This will intentionally not wind up calling vmtruncate(), |
| 968 | * since all the work for a size change has been done above. |
| 969 | * Otherwise, we could get into problems with truncate as |
| 970 | * ip_alloc_sem is used there to protect against i_size |
| 971 | * changes. |
| 972 | */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 973 | status = inode_setattr(inode, attr); |
| 974 | if (status < 0) { |
| 975 | mlog_errno(status); |
| 976 | goto bail_commit; |
| 977 | } |
| 978 | |
| 979 | status = ocfs2_mark_inode_dirty(handle, inode, bh); |
| 980 | if (status < 0) |
| 981 | mlog_errno(status); |
| 982 | |
| 983 | bail_commit: |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 984 | ocfs2_commit_trans(osb, handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 985 | bail_unlock: |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 986 | ocfs2_inode_unlock(inode, 1); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 987 | bail_unlock_rw: |
| 988 | if (size_change) |
| 989 | ocfs2_rw_unlock(inode, 1); |
| 990 | bail: |
Mark Fasheh | a81cb88 | 2008-10-07 14:25:16 -0700 | [diff] [blame] | 991 | brelse(bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 992 | |
| 993 | mlog_exit(status); |
| 994 | return status; |
| 995 | } |
| 996 | |
| 997 | int ocfs2_getattr(struct vfsmount *mnt, |
| 998 | struct dentry *dentry, |
| 999 | struct kstat *stat) |
| 1000 | { |
| 1001 | struct inode *inode = dentry->d_inode; |
| 1002 | struct super_block *sb = dentry->d_inode->i_sb; |
| 1003 | struct ocfs2_super *osb = sb->s_fs_info; |
| 1004 | int err; |
| 1005 | |
| 1006 | mlog_entry_void(); |
| 1007 | |
| 1008 | err = ocfs2_inode_revalidate(dentry); |
| 1009 | if (err) { |
| 1010 | if (err != -ENOENT) |
| 1011 | mlog_errno(err); |
| 1012 | goto bail; |
| 1013 | } |
| 1014 | |
| 1015 | generic_fillattr(inode, stat); |
| 1016 | |
| 1017 | /* We set the blksize from the cluster size for performance */ |
| 1018 | stat->blksize = osb->s_clustersize; |
| 1019 | |
| 1020 | bail: |
| 1021 | mlog_exit(err); |
| 1022 | |
| 1023 | return err; |
| 1024 | } |
| 1025 | |
Al Viro | e6305c4 | 2008-07-15 21:03:57 -0400 | [diff] [blame] | 1026 | int ocfs2_permission(struct inode *inode, int mask) |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 1027 | { |
| 1028 | int ret; |
| 1029 | |
| 1030 | mlog_entry_void(); |
| 1031 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1032 | ret = ocfs2_inode_lock(inode, NULL, 0); |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 1033 | if (ret) { |
Mark Fasheh | a9f5f70 | 2007-04-26 11:43:43 -0700 | [diff] [blame] | 1034 | if (ret != -ENOENT) |
| 1035 | mlog_errno(ret); |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 1036 | goto out; |
| 1037 | } |
| 1038 | |
Tiger Yang | 23fc270 | 2008-11-14 11:17:18 +0800 | [diff] [blame^] | 1039 | ret = generic_permission(inode, mask, ocfs2_check_acl); |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 1040 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1041 | ocfs2_inode_unlock(inode, 0); |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 1042 | out: |
| 1043 | mlog_exit(ret); |
| 1044 | return ret; |
| 1045 | } |
| 1046 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1047 | static int __ocfs2_write_remove_suid(struct inode *inode, |
| 1048 | struct buffer_head *bh) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1049 | { |
| 1050 | int ret; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 1051 | handle_t *handle; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1052 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1053 | struct ocfs2_dinode *di; |
| 1054 | |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 1055 | mlog_entry("(Inode %llu, mode 0%o)\n", |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1056 | (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1057 | |
Mark Fasheh | 65eff9c | 2006-10-09 17:26:22 -0700 | [diff] [blame] | 1058 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
Jan Kara | fa38e92 | 2008-10-20 19:23:51 +0200 | [diff] [blame] | 1059 | if (IS_ERR(handle)) { |
| 1060 | ret = PTR_ERR(handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1061 | mlog_errno(ret); |
| 1062 | goto out; |
| 1063 | } |
| 1064 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1065 | ret = ocfs2_journal_access(handle, inode, bh, |
| 1066 | OCFS2_JOURNAL_ACCESS_WRITE); |
| 1067 | if (ret < 0) { |
| 1068 | mlog_errno(ret); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1069 | goto out_trans; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1070 | } |
| 1071 | |
| 1072 | inode->i_mode &= ~S_ISUID; |
| 1073 | if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP)) |
| 1074 | inode->i_mode &= ~S_ISGID; |
| 1075 | |
| 1076 | di = (struct ocfs2_dinode *) bh->b_data; |
| 1077 | di->i_mode = cpu_to_le16(inode->i_mode); |
| 1078 | |
| 1079 | ret = ocfs2_journal_dirty(handle, bh); |
| 1080 | if (ret < 0) |
| 1081 | mlog_errno(ret); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1082 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1083 | out_trans: |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 1084 | ocfs2_commit_trans(osb, handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1085 | out: |
| 1086 | mlog_exit(ret); |
| 1087 | return ret; |
| 1088 | } |
| 1089 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1090 | /* |
| 1091 | * Will look for holes and unwritten extents in the range starting at |
| 1092 | * pos for count bytes (inclusive). |
| 1093 | */ |
| 1094 | static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos, |
| 1095 | size_t count) |
| 1096 | { |
| 1097 | int ret = 0; |
Mark Fasheh | 49cb8d2 | 2007-03-09 16:21:46 -0800 | [diff] [blame] | 1098 | unsigned int extent_flags; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1099 | u32 cpos, clusters, extent_len, phys_cpos; |
| 1100 | struct super_block *sb = inode->i_sb; |
| 1101 | |
| 1102 | cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits; |
| 1103 | clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos; |
| 1104 | |
| 1105 | while (clusters) { |
Mark Fasheh | 49cb8d2 | 2007-03-09 16:21:46 -0800 | [diff] [blame] | 1106 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len, |
| 1107 | &extent_flags); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1108 | if (ret < 0) { |
| 1109 | mlog_errno(ret); |
| 1110 | goto out; |
| 1111 | } |
| 1112 | |
Mark Fasheh | 49cb8d2 | 2007-03-09 16:21:46 -0800 | [diff] [blame] | 1113 | if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) { |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1114 | ret = 1; |
| 1115 | break; |
| 1116 | } |
| 1117 | |
| 1118 | if (extent_len > clusters) |
| 1119 | extent_len = clusters; |
| 1120 | |
| 1121 | clusters -= extent_len; |
| 1122 | cpos += extent_len; |
| 1123 | } |
| 1124 | out: |
| 1125 | return ret; |
| 1126 | } |
| 1127 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1128 | static int ocfs2_write_remove_suid(struct inode *inode) |
| 1129 | { |
| 1130 | int ret; |
| 1131 | struct buffer_head *bh = NULL; |
| 1132 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 1133 | |
Joel Becker | 0fcaa56 | 2008-10-09 17:20:31 -0700 | [diff] [blame] | 1134 | ret = ocfs2_read_block(inode, oi->ip_blkno, &bh); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1135 | if (ret < 0) { |
| 1136 | mlog_errno(ret); |
| 1137 | goto out; |
| 1138 | } |
| 1139 | |
| 1140 | ret = __ocfs2_write_remove_suid(inode, bh); |
| 1141 | out: |
| 1142 | brelse(bh); |
| 1143 | return ret; |
| 1144 | } |
| 1145 | |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 1146 | /* |
| 1147 | * Allocate enough extents to cover the region starting at byte offset |
| 1148 | * start for len bytes. Existing extents are skipped, any extents |
| 1149 | * added are marked as "unwritten". |
| 1150 | */ |
| 1151 | static int ocfs2_allocate_unwritten_extents(struct inode *inode, |
| 1152 | u64 start, u64 len) |
| 1153 | { |
| 1154 | int ret; |
| 1155 | u32 cpos, phys_cpos, clusters, alloc_size; |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1156 | u64 end = start + len; |
| 1157 | struct buffer_head *di_bh = NULL; |
| 1158 | |
| 1159 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
Joel Becker | 31d3307 | 2008-10-09 17:20:30 -0700 | [diff] [blame] | 1160 | ret = ocfs2_read_block(inode, OCFS2_I(inode)->ip_blkno, |
Joel Becker | 0fcaa56 | 2008-10-09 17:20:31 -0700 | [diff] [blame] | 1161 | &di_bh); |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1162 | if (ret) { |
| 1163 | mlog_errno(ret); |
| 1164 | goto out; |
| 1165 | } |
| 1166 | |
| 1167 | /* |
| 1168 | * Nothing to do if the requested reservation range |
| 1169 | * fits within the inode. |
| 1170 | */ |
| 1171 | if (ocfs2_size_fits_inline_data(di_bh, end)) |
| 1172 | goto out; |
| 1173 | |
| 1174 | ret = ocfs2_convert_inline_data_to_extents(inode, di_bh); |
| 1175 | if (ret) { |
| 1176 | mlog_errno(ret); |
| 1177 | goto out; |
| 1178 | } |
| 1179 | } |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 1180 | |
| 1181 | /* |
| 1182 | * We consider both start and len to be inclusive. |
| 1183 | */ |
| 1184 | cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits; |
| 1185 | clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len); |
| 1186 | clusters -= cpos; |
| 1187 | |
| 1188 | while (clusters) { |
| 1189 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, |
| 1190 | &alloc_size, NULL); |
| 1191 | if (ret) { |
| 1192 | mlog_errno(ret); |
| 1193 | goto out; |
| 1194 | } |
| 1195 | |
| 1196 | /* |
| 1197 | * Hole or existing extent len can be arbitrary, so |
| 1198 | * cap it to our own allocation request. |
| 1199 | */ |
| 1200 | if (alloc_size > clusters) |
| 1201 | alloc_size = clusters; |
| 1202 | |
| 1203 | if (phys_cpos) { |
| 1204 | /* |
| 1205 | * We already have an allocation at this |
| 1206 | * region so we can safely skip it. |
| 1207 | */ |
| 1208 | goto next; |
| 1209 | } |
| 1210 | |
| 1211 | ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1); |
| 1212 | if (ret) { |
| 1213 | if (ret != -ENOSPC) |
| 1214 | mlog_errno(ret); |
| 1215 | goto out; |
| 1216 | } |
| 1217 | |
| 1218 | next: |
| 1219 | cpos += alloc_size; |
| 1220 | clusters -= alloc_size; |
| 1221 | } |
| 1222 | |
| 1223 | ret = 0; |
| 1224 | out: |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1225 | |
| 1226 | brelse(di_bh); |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 1227 | return ret; |
| 1228 | } |
| 1229 | |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1230 | /* |
| 1231 | * Truncate a byte range, avoiding pages within partial clusters. This |
| 1232 | * preserves those pages for the zeroing code to write to. |
| 1233 | */ |
| 1234 | static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start, |
| 1235 | u64 byte_len) |
| 1236 | { |
| 1237 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1238 | loff_t start, end; |
| 1239 | struct address_space *mapping = inode->i_mapping; |
| 1240 | |
| 1241 | start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start); |
| 1242 | end = byte_start + byte_len; |
| 1243 | end = end & ~(osb->s_clustersize - 1); |
| 1244 | |
| 1245 | if (start < end) { |
| 1246 | unmap_mapping_range(mapping, start, end - start, 0); |
| 1247 | truncate_inode_pages_range(mapping, start, end - 1); |
| 1248 | } |
| 1249 | } |
| 1250 | |
| 1251 | static int ocfs2_zero_partial_clusters(struct inode *inode, |
| 1252 | u64 start, u64 len) |
| 1253 | { |
| 1254 | int ret = 0; |
| 1255 | u64 tmpend, end = start + len; |
| 1256 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1257 | unsigned int csize = osb->s_clustersize; |
| 1258 | handle_t *handle; |
| 1259 | |
| 1260 | /* |
| 1261 | * The "start" and "end" values are NOT necessarily part of |
| 1262 | * the range whose allocation is being deleted. Rather, this |
| 1263 | * is what the user passed in with the request. We must zero |
| 1264 | * partial clusters here. There's no need to worry about |
| 1265 | * physical allocation - the zeroing code knows to skip holes. |
| 1266 | */ |
| 1267 | mlog(0, "byte start: %llu, end: %llu\n", |
| 1268 | (unsigned long long)start, (unsigned long long)end); |
| 1269 | |
| 1270 | /* |
| 1271 | * If both edges are on a cluster boundary then there's no |
| 1272 | * zeroing required as the region is part of the allocation to |
| 1273 | * be truncated. |
| 1274 | */ |
| 1275 | if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0) |
| 1276 | goto out; |
| 1277 | |
| 1278 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
Jan Kara | fa38e92 | 2008-10-20 19:23:51 +0200 | [diff] [blame] | 1279 | if (IS_ERR(handle)) { |
| 1280 | ret = PTR_ERR(handle); |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1281 | mlog_errno(ret); |
| 1282 | goto out; |
| 1283 | } |
| 1284 | |
| 1285 | /* |
| 1286 | * We want to get the byte offset of the end of the 1st cluster. |
| 1287 | */ |
| 1288 | tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1)); |
| 1289 | if (tmpend > end) |
| 1290 | tmpend = end; |
| 1291 | |
| 1292 | mlog(0, "1st range: start: %llu, tmpend: %llu\n", |
| 1293 | (unsigned long long)start, (unsigned long long)tmpend); |
| 1294 | |
| 1295 | ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend); |
| 1296 | if (ret) |
| 1297 | mlog_errno(ret); |
| 1298 | |
| 1299 | if (tmpend < end) { |
| 1300 | /* |
| 1301 | * This may make start and end equal, but the zeroing |
| 1302 | * code will skip any work in that case so there's no |
| 1303 | * need to catch it up here. |
| 1304 | */ |
| 1305 | start = end & ~(osb->s_clustersize - 1); |
| 1306 | |
| 1307 | mlog(0, "2nd range: start: %llu, end: %llu\n", |
| 1308 | (unsigned long long)start, (unsigned long long)end); |
| 1309 | |
| 1310 | ret = ocfs2_zero_range_for_truncate(inode, handle, start, end); |
| 1311 | if (ret) |
| 1312 | mlog_errno(ret); |
| 1313 | } |
| 1314 | |
| 1315 | ocfs2_commit_trans(osb, handle); |
| 1316 | out: |
| 1317 | return ret; |
| 1318 | } |
| 1319 | |
| 1320 | static int ocfs2_remove_inode_range(struct inode *inode, |
| 1321 | struct buffer_head *di_bh, u64 byte_start, |
| 1322 | u64 byte_len) |
| 1323 | { |
| 1324 | int ret = 0; |
| 1325 | u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size; |
| 1326 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1327 | struct ocfs2_cached_dealloc_ctxt dealloc; |
Mark Fasheh | b1967d0 | 2007-11-20 11:56:39 -0800 | [diff] [blame] | 1328 | struct address_space *mapping = inode->i_mapping; |
Mark Fasheh | fecc011 | 2008-11-12 15:16:38 -0800 | [diff] [blame] | 1329 | struct ocfs2_extent_tree et; |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1330 | |
Mark Fasheh | fecc011 | 2008-11-12 15:16:38 -0800 | [diff] [blame] | 1331 | ocfs2_init_dinode_extent_tree(&et, inode, di_bh); |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1332 | ocfs2_init_dealloc_ctxt(&dealloc); |
| 1333 | |
| 1334 | if (byte_len == 0) |
| 1335 | return 0; |
| 1336 | |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1337 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
| 1338 | ret = ocfs2_truncate_inline(inode, di_bh, byte_start, |
Mark Fasheh | b1967d0 | 2007-11-20 11:56:39 -0800 | [diff] [blame] | 1339 | byte_start + byte_len, 0); |
| 1340 | if (ret) { |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1341 | mlog_errno(ret); |
Mark Fasheh | b1967d0 | 2007-11-20 11:56:39 -0800 | [diff] [blame] | 1342 | goto out; |
| 1343 | } |
| 1344 | /* |
| 1345 | * There's no need to get fancy with the page cache |
| 1346 | * truncate of an inline-data inode. We're talking |
| 1347 | * about less than a page here, which will be cached |
| 1348 | * in the dinode buffer anyway. |
| 1349 | */ |
| 1350 | unmap_mapping_range(mapping, 0, 0, 0); |
| 1351 | truncate_inode_pages(mapping, 0); |
| 1352 | goto out; |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1353 | } |
| 1354 | |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1355 | trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start); |
| 1356 | trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits; |
| 1357 | if (trunc_len >= trunc_start) |
| 1358 | trunc_len -= trunc_start; |
| 1359 | else |
| 1360 | trunc_len = 0; |
| 1361 | |
| 1362 | mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n", |
| 1363 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
| 1364 | (unsigned long long)byte_start, |
| 1365 | (unsigned long long)byte_len, trunc_start, trunc_len); |
| 1366 | |
| 1367 | ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len); |
| 1368 | if (ret) { |
| 1369 | mlog_errno(ret); |
| 1370 | goto out; |
| 1371 | } |
| 1372 | |
| 1373 | cpos = trunc_start; |
| 1374 | while (trunc_len) { |
| 1375 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, |
| 1376 | &alloc_size, NULL); |
| 1377 | if (ret) { |
| 1378 | mlog_errno(ret); |
| 1379 | goto out; |
| 1380 | } |
| 1381 | |
| 1382 | if (alloc_size > trunc_len) |
| 1383 | alloc_size = trunc_len; |
| 1384 | |
| 1385 | /* Only do work for non-holes */ |
| 1386 | if (phys_cpos != 0) { |
Mark Fasheh | fecc011 | 2008-11-12 15:16:38 -0800 | [diff] [blame] | 1387 | ret = ocfs2_remove_btree_range(inode, &et, cpos, |
| 1388 | phys_cpos, alloc_size, |
| 1389 | &dealloc); |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1390 | if (ret) { |
| 1391 | mlog_errno(ret); |
| 1392 | goto out; |
| 1393 | } |
| 1394 | } |
| 1395 | |
| 1396 | cpos += alloc_size; |
| 1397 | trunc_len -= alloc_size; |
| 1398 | } |
| 1399 | |
| 1400 | ocfs2_truncate_cluster_pages(inode, byte_start, byte_len); |
| 1401 | |
| 1402 | out: |
| 1403 | ocfs2_schedule_truncate_log_flush(osb, 1); |
| 1404 | ocfs2_run_deallocs(osb, &dealloc); |
| 1405 | |
| 1406 | return ret; |
| 1407 | } |
| 1408 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1409 | /* |
| 1410 | * Parts of this function taken from xfs_change_file_space() |
| 1411 | */ |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1412 | static int __ocfs2_change_file_space(struct file *file, struct inode *inode, |
| 1413 | loff_t f_pos, unsigned int cmd, |
| 1414 | struct ocfs2_space_resv *sr, |
| 1415 | int change_size) |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1416 | { |
| 1417 | int ret; |
| 1418 | s64 llen; |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1419 | loff_t size; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1420 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1421 | struct buffer_head *di_bh = NULL; |
| 1422 | handle_t *handle; |
Mark Fasheh | a00cce3 | 2007-07-20 11:28:30 -0700 | [diff] [blame] | 1423 | unsigned long long max_off = inode->i_sb->s_maxbytes; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1424 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1425 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) |
| 1426 | return -EROFS; |
| 1427 | |
| 1428 | mutex_lock(&inode->i_mutex); |
| 1429 | |
| 1430 | /* |
| 1431 | * This prevents concurrent writes on other nodes |
| 1432 | */ |
| 1433 | ret = ocfs2_rw_lock(inode, 1); |
| 1434 | if (ret) { |
| 1435 | mlog_errno(ret); |
| 1436 | goto out; |
| 1437 | } |
| 1438 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1439 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1440 | if (ret) { |
| 1441 | mlog_errno(ret); |
| 1442 | goto out_rw_unlock; |
| 1443 | } |
| 1444 | |
| 1445 | if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) { |
| 1446 | ret = -EPERM; |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1447 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1448 | } |
| 1449 | |
| 1450 | switch (sr->l_whence) { |
| 1451 | case 0: /*SEEK_SET*/ |
| 1452 | break; |
| 1453 | case 1: /*SEEK_CUR*/ |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1454 | sr->l_start += f_pos; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1455 | break; |
| 1456 | case 2: /*SEEK_END*/ |
| 1457 | sr->l_start += i_size_read(inode); |
| 1458 | break; |
| 1459 | default: |
| 1460 | ret = -EINVAL; |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1461 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1462 | } |
| 1463 | sr->l_whence = 0; |
| 1464 | |
| 1465 | llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len; |
| 1466 | |
| 1467 | if (sr->l_start < 0 |
| 1468 | || sr->l_start > max_off |
| 1469 | || (sr->l_start + llen) < 0 |
| 1470 | || (sr->l_start + llen) > max_off) { |
| 1471 | ret = -EINVAL; |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1472 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1473 | } |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1474 | size = sr->l_start + sr->l_len; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1475 | |
| 1476 | if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) { |
| 1477 | if (sr->l_len <= 0) { |
| 1478 | ret = -EINVAL; |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1479 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1480 | } |
| 1481 | } |
| 1482 | |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1483 | if (file && should_remove_suid(file->f_path.dentry)) { |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1484 | ret = __ocfs2_write_remove_suid(inode, di_bh); |
| 1485 | if (ret) { |
| 1486 | mlog_errno(ret); |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1487 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1488 | } |
| 1489 | } |
| 1490 | |
| 1491 | down_write(&OCFS2_I(inode)->ip_alloc_sem); |
| 1492 | switch (cmd) { |
| 1493 | case OCFS2_IOC_RESVSP: |
| 1494 | case OCFS2_IOC_RESVSP64: |
| 1495 | /* |
| 1496 | * This takes unsigned offsets, but the signed ones we |
| 1497 | * pass have been checked against overflow above. |
| 1498 | */ |
| 1499 | ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start, |
| 1500 | sr->l_len); |
| 1501 | break; |
| 1502 | case OCFS2_IOC_UNRESVSP: |
| 1503 | case OCFS2_IOC_UNRESVSP64: |
| 1504 | ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start, |
| 1505 | sr->l_len); |
| 1506 | break; |
| 1507 | default: |
| 1508 | ret = -EINVAL; |
| 1509 | } |
| 1510 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
| 1511 | if (ret) { |
| 1512 | mlog_errno(ret); |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1513 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1514 | } |
| 1515 | |
| 1516 | /* |
| 1517 | * We update c/mtime for these changes |
| 1518 | */ |
| 1519 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
| 1520 | if (IS_ERR(handle)) { |
| 1521 | ret = PTR_ERR(handle); |
| 1522 | mlog_errno(ret); |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1523 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1524 | } |
| 1525 | |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1526 | if (change_size && i_size_read(inode) < size) |
| 1527 | i_size_write(inode, size); |
| 1528 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1529 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
| 1530 | ret = ocfs2_mark_inode_dirty(handle, inode, di_bh); |
| 1531 | if (ret < 0) |
| 1532 | mlog_errno(ret); |
| 1533 | |
| 1534 | ocfs2_commit_trans(osb, handle); |
| 1535 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1536 | out_inode_unlock: |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1537 | brelse(di_bh); |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1538 | ocfs2_inode_unlock(inode, 1); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1539 | out_rw_unlock: |
| 1540 | ocfs2_rw_unlock(inode, 1); |
| 1541 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1542 | out: |
Julia Lawall | c259ae5 | 2008-07-21 09:59:15 +0200 | [diff] [blame] | 1543 | mutex_unlock(&inode->i_mutex); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1544 | return ret; |
| 1545 | } |
| 1546 | |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1547 | int ocfs2_change_file_space(struct file *file, unsigned int cmd, |
| 1548 | struct ocfs2_space_resv *sr) |
| 1549 | { |
| 1550 | struct inode *inode = file->f_path.dentry->d_inode; |
| 1551 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);; |
| 1552 | |
| 1553 | if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) && |
| 1554 | !ocfs2_writes_unwritten_extents(osb)) |
| 1555 | return -ENOTTY; |
| 1556 | else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) && |
| 1557 | !ocfs2_sparse_alloc(osb)) |
| 1558 | return -ENOTTY; |
| 1559 | |
| 1560 | if (!S_ISREG(inode->i_mode)) |
| 1561 | return -EINVAL; |
| 1562 | |
| 1563 | if (!(file->f_mode & FMODE_WRITE)) |
| 1564 | return -EBADF; |
| 1565 | |
| 1566 | return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0); |
| 1567 | } |
| 1568 | |
| 1569 | static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset, |
| 1570 | loff_t len) |
| 1571 | { |
| 1572 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1573 | struct ocfs2_space_resv sr; |
| 1574 | int change_size = 1; |
| 1575 | |
| 1576 | if (!ocfs2_writes_unwritten_extents(osb)) |
| 1577 | return -EOPNOTSUPP; |
| 1578 | |
| 1579 | if (S_ISDIR(inode->i_mode)) |
| 1580 | return -ENODEV; |
| 1581 | |
| 1582 | if (mode & FALLOC_FL_KEEP_SIZE) |
| 1583 | change_size = 0; |
| 1584 | |
| 1585 | sr.l_whence = 0; |
| 1586 | sr.l_start = (s64)offset; |
| 1587 | sr.l_len = (s64)len; |
| 1588 | |
| 1589 | return __ocfs2_change_file_space(NULL, inode, offset, |
| 1590 | OCFS2_IOC_RESVSP64, &sr, change_size); |
| 1591 | } |
| 1592 | |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1593 | static int ocfs2_prepare_inode_for_write(struct dentry *dentry, |
| 1594 | loff_t *ppos, |
| 1595 | size_t count, |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1596 | int appending, |
| 1597 | int *direct_io) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1598 | { |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1599 | int ret = 0, meta_level = 0; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1600 | struct inode *inode = dentry->d_inode; |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1601 | loff_t saved_pos, end; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1602 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1603 | /* |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1604 | * We start with a read level meta lock and only jump to an ex |
| 1605 | * if we need to make modifications here. |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1606 | */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1607 | for(;;) { |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1608 | ret = ocfs2_inode_lock(inode, NULL, meta_level); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1609 | if (ret < 0) { |
| 1610 | meta_level = -1; |
| 1611 | mlog_errno(ret); |
| 1612 | goto out; |
| 1613 | } |
| 1614 | |
| 1615 | /* Clear suid / sgid if necessary. We do this here |
| 1616 | * instead of later in the write path because |
| 1617 | * remove_suid() calls ->setattr without any hint that |
| 1618 | * we may have already done our cluster locking. Since |
| 1619 | * ocfs2_setattr() *must* take cluster locks to |
| 1620 | * proceeed, this will lead us to recursively lock the |
| 1621 | * inode. There's also the dinode i_size state which |
| 1622 | * can be lost via setattr during extending writes (we |
| 1623 | * set inode->i_size at the end of a write. */ |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1624 | if (should_remove_suid(dentry)) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1625 | if (meta_level == 0) { |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1626 | ocfs2_inode_unlock(inode, meta_level); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1627 | meta_level = 1; |
| 1628 | continue; |
| 1629 | } |
| 1630 | |
| 1631 | ret = ocfs2_write_remove_suid(inode); |
| 1632 | if (ret < 0) { |
| 1633 | mlog_errno(ret); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1634 | goto out_unlock; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1635 | } |
| 1636 | } |
| 1637 | |
| 1638 | /* work on a copy of ppos until we're sure that we won't have |
| 1639 | * to recalculate it due to relocking. */ |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1640 | if (appending) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1641 | saved_pos = i_size_read(inode); |
| 1642 | mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos); |
| 1643 | } else { |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1644 | saved_pos = *ppos; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1645 | } |
Mark Fasheh | 3a0782d | 2007-01-17 12:53:31 -0800 | [diff] [blame] | 1646 | |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1647 | end = saved_pos + count; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1648 | |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1649 | /* |
| 1650 | * Skip the O_DIRECT checks if we don't need |
| 1651 | * them. |
| 1652 | */ |
| 1653 | if (!direct_io || !(*direct_io)) |
| 1654 | break; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1655 | |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1656 | /* |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1657 | * There's no sane way to do direct writes to an inode |
| 1658 | * with inline data. |
| 1659 | */ |
| 1660 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
| 1661 | *direct_io = 0; |
| 1662 | break; |
| 1663 | } |
| 1664 | |
| 1665 | /* |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1666 | * Allowing concurrent direct writes means |
| 1667 | * i_size changes wouldn't be synchronized, so |
| 1668 | * one node could wind up truncating another |
| 1669 | * nodes writes. |
| 1670 | */ |
| 1671 | if (end > i_size_read(inode)) { |
| 1672 | *direct_io = 0; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1673 | break; |
| 1674 | } |
| 1675 | |
Mark Fasheh | 3a0782d | 2007-01-17 12:53:31 -0800 | [diff] [blame] | 1676 | /* |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1677 | * We don't fill holes during direct io, so |
| 1678 | * check for them here. If any are found, the |
| 1679 | * caller will have to retake some cluster |
| 1680 | * locks and initiate the io as buffered. |
Mark Fasheh | 3a0782d | 2007-01-17 12:53:31 -0800 | [diff] [blame] | 1681 | */ |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1682 | ret = ocfs2_check_range_for_holes(inode, saved_pos, count); |
| 1683 | if (ret == 1) { |
| 1684 | *direct_io = 0; |
| 1685 | ret = 0; |
| 1686 | } else if (ret < 0) |
| 1687 | mlog_errno(ret); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1688 | break; |
| 1689 | } |
| 1690 | |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1691 | if (appending) |
| 1692 | *ppos = saved_pos; |
| 1693 | |
| 1694 | out_unlock: |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1695 | ocfs2_inode_unlock(inode, meta_level); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1696 | |
| 1697 | out: |
| 1698 | return ret; |
| 1699 | } |
| 1700 | |
| 1701 | static ssize_t ocfs2_file_aio_write(struct kiocb *iocb, |
| 1702 | const struct iovec *iov, |
| 1703 | unsigned long nr_segs, |
| 1704 | loff_t pos) |
| 1705 | { |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1706 | int ret, direct_io, appending, rw_level, have_alloc_sem = 0; |
Nick Piggin | b6af1bc | 2007-10-16 01:25:24 -0700 | [diff] [blame] | 1707 | int can_do_direct; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1708 | ssize_t written = 0; |
| 1709 | size_t ocount; /* original count */ |
| 1710 | size_t count; /* after file limit checks */ |
Mark Fasheh | 9ea2d32 | 2007-10-18 14:14:45 -0700 | [diff] [blame] | 1711 | loff_t old_size, *ppos = &iocb->ki_pos; |
| 1712 | u32 old_clusters; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1713 | struct file *file = iocb->ki_filp; |
| 1714 | struct inode *inode = file->f_path.dentry->d_inode; |
Mark Fasheh | 9ea2d32 | 2007-10-18 14:14:45 -0700 | [diff] [blame] | 1715 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1716 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1717 | mlog_entry("(0x%p, %u, '%.*s')\n", file, |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1718 | (unsigned int)nr_segs, |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1719 | file->f_path.dentry->d_name.len, |
| 1720 | file->f_path.dentry->d_name.name); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1721 | |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1722 | if (iocb->ki_left == 0) |
| 1723 | return 0; |
| 1724 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1725 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); |
| 1726 | |
| 1727 | appending = file->f_flags & O_APPEND ? 1 : 0; |
| 1728 | direct_io = file->f_flags & O_DIRECT ? 1 : 0; |
| 1729 | |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1730 | mutex_lock(&inode->i_mutex); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1731 | |
| 1732 | relock: |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1733 | /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */ |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1734 | if (direct_io) { |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1735 | down_read(&inode->i_alloc_sem); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1736 | have_alloc_sem = 1; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1737 | } |
| 1738 | |
| 1739 | /* concurrent O_DIRECT writes are allowed */ |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1740 | rw_level = !direct_io; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1741 | ret = ocfs2_rw_lock(inode, rw_level); |
| 1742 | if (ret < 0) { |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1743 | mlog_errno(ret); |
| 1744 | goto out_sems; |
| 1745 | } |
| 1746 | |
| 1747 | can_do_direct = direct_io; |
| 1748 | ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos, |
| 1749 | iocb->ki_left, appending, |
| 1750 | &can_do_direct); |
| 1751 | if (ret < 0) { |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1752 | mlog_errno(ret); |
| 1753 | goto out; |
| 1754 | } |
| 1755 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1756 | /* |
| 1757 | * We can't complete the direct I/O as requested, fall back to |
| 1758 | * buffered I/O. |
| 1759 | */ |
| 1760 | if (direct_io && !can_do_direct) { |
| 1761 | ocfs2_rw_unlock(inode, rw_level); |
| 1762 | up_read(&inode->i_alloc_sem); |
| 1763 | |
| 1764 | have_alloc_sem = 0; |
| 1765 | rw_level = -1; |
| 1766 | |
| 1767 | direct_io = 0; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1768 | goto relock; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1769 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1770 | |
Mark Fasheh | 9ea2d32 | 2007-10-18 14:14:45 -0700 | [diff] [blame] | 1771 | /* |
| 1772 | * To later detect whether a journal commit for sync writes is |
| 1773 | * necessary, we sample i_size, and cluster count here. |
| 1774 | */ |
| 1775 | old_size = i_size_read(inode); |
| 1776 | old_clusters = OCFS2_I(inode)->ip_clusters; |
| 1777 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1778 | /* communicate with ocfs2_dio_end_io */ |
Mark Fasheh | 7cdfc3a | 2007-04-16 17:28:51 -0700 | [diff] [blame] | 1779 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1780 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1781 | if (direct_io) { |
Nick Piggin | b6af1bc | 2007-10-16 01:25:24 -0700 | [diff] [blame] | 1782 | ret = generic_segment_checks(iov, &nr_segs, &ocount, |
| 1783 | VERIFY_READ); |
| 1784 | if (ret) |
| 1785 | goto out_dio; |
| 1786 | |
| 1787 | ret = generic_write_checks(file, ppos, &count, |
| 1788 | S_ISBLK(inode->i_mode)); |
| 1789 | if (ret) |
| 1790 | goto out_dio; |
| 1791 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1792 | written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos, |
| 1793 | ppos, count, ocount); |
| 1794 | if (written < 0) { |
Dmitri Monakhov | c435400 | 2008-10-27 13:01:49 -0700 | [diff] [blame] | 1795 | /* |
| 1796 | * direct write may have instantiated a few |
| 1797 | * blocks outside i_size. Trim these off again. |
| 1798 | * Don't need i_size_read because we hold i_mutex. |
| 1799 | */ |
| 1800 | if (*ppos + count > inode->i_size) |
| 1801 | vmtruncate(inode, inode->i_size); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1802 | ret = written; |
| 1803 | goto out_dio; |
| 1804 | } |
| 1805 | } else { |
Nick Piggin | b6af1bc | 2007-10-16 01:25:24 -0700 | [diff] [blame] | 1806 | written = generic_file_aio_write_nolock(iocb, iov, nr_segs, |
| 1807 | *ppos); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1808 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1809 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1810 | out_dio: |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1811 | /* buffered aio wouldn't have proper lock coverage today */ |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1812 | BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1813 | |
Mark Fasheh | 9ea2d32 | 2007-10-18 14:14:45 -0700 | [diff] [blame] | 1814 | if ((file->f_flags & O_SYNC && !direct_io) || IS_SYNC(inode)) { |
| 1815 | /* |
| 1816 | * The generic write paths have handled getting data |
| 1817 | * to disk, but since we don't make use of the dirty |
| 1818 | * inode list, a manual journal commit is necessary |
| 1819 | * here. |
| 1820 | */ |
| 1821 | if (old_size != i_size_read(inode) || |
| 1822 | old_clusters != OCFS2_I(inode)->ip_clusters) { |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 1823 | ret = jbd2_journal_force_commit(osb->journal->j_journal); |
Mark Fasheh | 9ea2d32 | 2007-10-18 14:14:45 -0700 | [diff] [blame] | 1824 | if (ret < 0) |
| 1825 | written = ret; |
| 1826 | } |
| 1827 | } |
| 1828 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1829 | /* |
| 1830 | * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io |
| 1831 | * function pointer which is called when o_direct io completes so that |
| 1832 | * it can unlock our rw lock. (it's the clustered equivalent of |
| 1833 | * i_alloc_sem; protects truncate from racing with pending ios). |
| 1834 | * Unfortunately there are error cases which call end_io and others |
| 1835 | * that don't. so we don't have to unlock the rw_lock if either an |
| 1836 | * async dio is going to do it in the future or an end_io after an |
| 1837 | * error has already done it. |
| 1838 | */ |
| 1839 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { |
| 1840 | rw_level = -1; |
| 1841 | have_alloc_sem = 0; |
| 1842 | } |
| 1843 | |
| 1844 | out: |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1845 | if (rw_level != -1) |
| 1846 | ocfs2_rw_unlock(inode, rw_level); |
| 1847 | |
| 1848 | out_sems: |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1849 | if (have_alloc_sem) |
| 1850 | up_read(&inode->i_alloc_sem); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1851 | |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 1852 | mutex_unlock(&inode->i_mutex); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1853 | |
| 1854 | mlog_exit(ret); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1855 | return written ? written : ret; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1856 | } |
| 1857 | |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1858 | static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe, |
| 1859 | struct file *out, |
| 1860 | loff_t *ppos, |
| 1861 | size_t len, |
| 1862 | unsigned int flags) |
| 1863 | { |
| 1864 | int ret; |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 1865 | struct inode *inode = out->f_path.dentry->d_inode; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1866 | |
| 1867 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe, |
| 1868 | (unsigned int)len, |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 1869 | out->f_path.dentry->d_name.len, |
| 1870 | out->f_path.dentry->d_name.name); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1871 | |
| 1872 | inode_double_lock(inode, pipe->inode); |
| 1873 | |
| 1874 | ret = ocfs2_rw_lock(inode, 1); |
| 1875 | if (ret < 0) { |
| 1876 | mlog_errno(ret); |
| 1877 | goto out; |
| 1878 | } |
| 1879 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1880 | ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0, |
| 1881 | NULL); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1882 | if (ret < 0) { |
| 1883 | mlog_errno(ret); |
| 1884 | goto out_unlock; |
| 1885 | } |
| 1886 | |
Nick Piggin | b6af1bc | 2007-10-16 01:25:24 -0700 | [diff] [blame] | 1887 | ret = generic_file_splice_write_nolock(pipe, out, ppos, len, flags); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1888 | |
| 1889 | out_unlock: |
| 1890 | ocfs2_rw_unlock(inode, 1); |
| 1891 | out: |
| 1892 | inode_double_unlock(inode, pipe->inode); |
| 1893 | |
| 1894 | mlog_exit(ret); |
| 1895 | return ret; |
| 1896 | } |
| 1897 | |
| 1898 | static ssize_t ocfs2_file_splice_read(struct file *in, |
| 1899 | loff_t *ppos, |
| 1900 | struct pipe_inode_info *pipe, |
| 1901 | size_t len, |
| 1902 | unsigned int flags) |
| 1903 | { |
| 1904 | int ret = 0; |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 1905 | struct inode *inode = in->f_path.dentry->d_inode; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1906 | |
| 1907 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe, |
| 1908 | (unsigned int)len, |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 1909 | in->f_path.dentry->d_name.len, |
| 1910 | in->f_path.dentry->d_name.name); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1911 | |
| 1912 | /* |
| 1913 | * See the comment in ocfs2_file_aio_read() |
| 1914 | */ |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1915 | ret = ocfs2_inode_lock(inode, NULL, 0); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1916 | if (ret < 0) { |
| 1917 | mlog_errno(ret); |
| 1918 | goto bail; |
| 1919 | } |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1920 | ocfs2_inode_unlock(inode, 0); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1921 | |
| 1922 | ret = generic_file_splice_read(in, ppos, pipe, len, flags); |
| 1923 | |
| 1924 | bail: |
| 1925 | mlog_exit(ret); |
| 1926 | return ret; |
| 1927 | } |
| 1928 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1929 | static ssize_t ocfs2_file_aio_read(struct kiocb *iocb, |
Badari Pulavarty | 027445c | 2006-09-30 23:28:46 -0700 | [diff] [blame] | 1930 | const struct iovec *iov, |
| 1931 | unsigned long nr_segs, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1932 | loff_t pos) |
| 1933 | { |
Tiger Yang | 25899de | 2006-11-15 15:49:02 +0800 | [diff] [blame] | 1934 | int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1935 | struct file *filp = iocb->ki_filp; |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 1936 | struct inode *inode = filp->f_path.dentry->d_inode; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1937 | |
Badari Pulavarty | 027445c | 2006-09-30 23:28:46 -0700 | [diff] [blame] | 1938 | mlog_entry("(0x%p, %u, '%.*s')\n", filp, |
| 1939 | (unsigned int)nr_segs, |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 1940 | filp->f_path.dentry->d_name.len, |
| 1941 | filp->f_path.dentry->d_name.name); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1942 | |
| 1943 | if (!inode) { |
| 1944 | ret = -EINVAL; |
| 1945 | mlog_errno(ret); |
| 1946 | goto bail; |
| 1947 | } |
| 1948 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1949 | /* |
| 1950 | * buffered reads protect themselves in ->readpage(). O_DIRECT reads |
| 1951 | * need locks to protect pending reads from racing with truncate. |
| 1952 | */ |
| 1953 | if (filp->f_flags & O_DIRECT) { |
| 1954 | down_read(&inode->i_alloc_sem); |
| 1955 | have_alloc_sem = 1; |
| 1956 | |
| 1957 | ret = ocfs2_rw_lock(inode, 0); |
| 1958 | if (ret < 0) { |
| 1959 | mlog_errno(ret); |
| 1960 | goto bail; |
| 1961 | } |
| 1962 | rw_level = 0; |
| 1963 | /* communicate with ocfs2_dio_end_io */ |
Mark Fasheh | 7cdfc3a | 2007-04-16 17:28:51 -0700 | [diff] [blame] | 1964 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1965 | } |
| 1966 | |
Mark Fasheh | c4374f8 | 2006-05-05 19:04:35 -0700 | [diff] [blame] | 1967 | /* |
| 1968 | * We're fine letting folks race truncates and extending |
| 1969 | * writes with read across the cluster, just like they can |
| 1970 | * locally. Hence no rw_lock during read. |
| 1971 | * |
| 1972 | * Take and drop the meta data lock to update inode fields |
| 1973 | * like i_size. This allows the checks down below |
| 1974 | * generic_file_aio_read() a chance of actually working. |
| 1975 | */ |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1976 | ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level); |
Mark Fasheh | c4374f8 | 2006-05-05 19:04:35 -0700 | [diff] [blame] | 1977 | if (ret < 0) { |
| 1978 | mlog_errno(ret); |
| 1979 | goto bail; |
| 1980 | } |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1981 | ocfs2_inode_unlock(inode, lock_level); |
Mark Fasheh | c4374f8 | 2006-05-05 19:04:35 -0700 | [diff] [blame] | 1982 | |
Badari Pulavarty | 027445c | 2006-09-30 23:28:46 -0700 | [diff] [blame] | 1983 | ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1984 | if (ret == -EINVAL) |
Sunil Mushran | 56753bd | 2008-06-09 11:24:41 -0700 | [diff] [blame] | 1985 | mlog(0, "generic_file_aio_read returned -EINVAL\n"); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1986 | |
| 1987 | /* buffered aio wouldn't have proper lock coverage today */ |
| 1988 | BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT)); |
| 1989 | |
| 1990 | /* see ocfs2_file_aio_write */ |
| 1991 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { |
| 1992 | rw_level = -1; |
| 1993 | have_alloc_sem = 0; |
| 1994 | } |
| 1995 | |
| 1996 | bail: |
| 1997 | if (have_alloc_sem) |
| 1998 | up_read(&inode->i_alloc_sem); |
| 1999 | if (rw_level != -1) |
| 2000 | ocfs2_rw_unlock(inode, rw_level); |
| 2001 | mlog_exit(ret); |
| 2002 | |
| 2003 | return ret; |
| 2004 | } |
| 2005 | |
Arjan van de Ven | 92e1d5b | 2007-02-12 00:55:39 -0800 | [diff] [blame] | 2006 | const struct inode_operations ocfs2_file_iops = { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2007 | .setattr = ocfs2_setattr, |
| 2008 | .getattr = ocfs2_getattr, |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 2009 | .permission = ocfs2_permission, |
Tiger Yang | cf1d6c7 | 2008-08-18 17:11:00 +0800 | [diff] [blame] | 2010 | .setxattr = generic_setxattr, |
| 2011 | .getxattr = generic_getxattr, |
| 2012 | .listxattr = ocfs2_listxattr, |
| 2013 | .removexattr = generic_removexattr, |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 2014 | .fallocate = ocfs2_fallocate, |
Mark Fasheh | 00dc417 | 2008-10-03 17:32:11 -0400 | [diff] [blame] | 2015 | .fiemap = ocfs2_fiemap, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2016 | }; |
| 2017 | |
Arjan van de Ven | 92e1d5b | 2007-02-12 00:55:39 -0800 | [diff] [blame] | 2018 | const struct inode_operations ocfs2_special_file_iops = { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2019 | .setattr = ocfs2_setattr, |
| 2020 | .getattr = ocfs2_getattr, |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 2021 | .permission = ocfs2_permission, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2022 | }; |
| 2023 | |
Mark Fasheh | 53da493 | 2008-07-21 14:29:16 -0700 | [diff] [blame] | 2024 | /* |
| 2025 | * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with |
| 2026 | * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks! |
| 2027 | */ |
Arjan van de Ven | 4b6f5d2 | 2006-03-28 01:56:42 -0800 | [diff] [blame] | 2028 | const struct file_operations ocfs2_fops = { |
Jan Kara | 32c3c0e | 2007-12-19 15:24:52 +0100 | [diff] [blame] | 2029 | .llseek = generic_file_llseek, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2030 | .read = do_sync_read, |
| 2031 | .write = do_sync_write, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2032 | .mmap = ocfs2_mmap, |
| 2033 | .fsync = ocfs2_sync_file, |
| 2034 | .release = ocfs2_file_release, |
| 2035 | .open = ocfs2_file_open, |
| 2036 | .aio_read = ocfs2_file_aio_read, |
| 2037 | .aio_write = ocfs2_file_aio_write, |
Andi Kleen | c9ec148 | 2008-01-27 03:17:17 +0100 | [diff] [blame] | 2038 | .unlocked_ioctl = ocfs2_ioctl, |
Mark Fasheh | 586d232 | 2007-03-09 15:56:28 -0800 | [diff] [blame] | 2039 | #ifdef CONFIG_COMPAT |
| 2040 | .compat_ioctl = ocfs2_compat_ioctl, |
| 2041 | #endif |
Mark Fasheh | 53da493 | 2008-07-21 14:29:16 -0700 | [diff] [blame] | 2042 | .lock = ocfs2_lock, |
| 2043 | .flock = ocfs2_flock, |
| 2044 | .splice_read = ocfs2_file_splice_read, |
| 2045 | .splice_write = ocfs2_file_splice_write, |
| 2046 | }; |
| 2047 | |
| 2048 | const struct file_operations ocfs2_dops = { |
| 2049 | .llseek = generic_file_llseek, |
| 2050 | .read = generic_read_dir, |
| 2051 | .readdir = ocfs2_readdir, |
| 2052 | .fsync = ocfs2_sync_file, |
| 2053 | .release = ocfs2_dir_release, |
| 2054 | .open = ocfs2_dir_open, |
| 2055 | .unlocked_ioctl = ocfs2_ioctl, |
| 2056 | #ifdef CONFIG_COMPAT |
| 2057 | .compat_ioctl = ocfs2_compat_ioctl, |
| 2058 | #endif |
| 2059 | .lock = ocfs2_lock, |
| 2060 | .flock = ocfs2_flock, |
| 2061 | }; |
| 2062 | |
| 2063 | /* |
| 2064 | * POSIX-lockless variants of our file_operations. |
| 2065 | * |
| 2066 | * These will be used if the underlying cluster stack does not support |
| 2067 | * posix file locking, if the user passes the "localflocks" mount |
| 2068 | * option, or if we have a local-only fs. |
| 2069 | * |
| 2070 | * ocfs2_flock is in here because all stacks handle UNIX file locks, |
| 2071 | * so we still want it in the case of no stack support for |
| 2072 | * plocks. Internally, it will do the right thing when asked to ignore |
| 2073 | * the cluster. |
| 2074 | */ |
| 2075 | const struct file_operations ocfs2_fops_no_plocks = { |
| 2076 | .llseek = generic_file_llseek, |
| 2077 | .read = do_sync_read, |
| 2078 | .write = do_sync_write, |
| 2079 | .mmap = ocfs2_mmap, |
| 2080 | .fsync = ocfs2_sync_file, |
| 2081 | .release = ocfs2_file_release, |
| 2082 | .open = ocfs2_file_open, |
| 2083 | .aio_read = ocfs2_file_aio_read, |
| 2084 | .aio_write = ocfs2_file_aio_write, |
| 2085 | .unlocked_ioctl = ocfs2_ioctl, |
| 2086 | #ifdef CONFIG_COMPAT |
| 2087 | .compat_ioctl = ocfs2_compat_ioctl, |
| 2088 | #endif |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 2089 | .flock = ocfs2_flock, |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 2090 | .splice_read = ocfs2_file_splice_read, |
| 2091 | .splice_write = ocfs2_file_splice_write, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2092 | }; |
| 2093 | |
Mark Fasheh | 53da493 | 2008-07-21 14:29:16 -0700 | [diff] [blame] | 2094 | const struct file_operations ocfs2_dops_no_plocks = { |
Jan Kara | 32c3c0e | 2007-12-19 15:24:52 +0100 | [diff] [blame] | 2095 | .llseek = generic_file_llseek, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2096 | .read = generic_read_dir, |
| 2097 | .readdir = ocfs2_readdir, |
| 2098 | .fsync = ocfs2_sync_file, |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 2099 | .release = ocfs2_dir_release, |
| 2100 | .open = ocfs2_dir_open, |
Andi Kleen | c9ec148 | 2008-01-27 03:17:17 +0100 | [diff] [blame] | 2101 | .unlocked_ioctl = ocfs2_ioctl, |
Mark Fasheh | 586d232 | 2007-03-09 15:56:28 -0800 | [diff] [blame] | 2102 | #ifdef CONFIG_COMPAT |
| 2103 | .compat_ioctl = ocfs2_compat_ioctl, |
| 2104 | #endif |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 2105 | .flock = ocfs2_flock, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2106 | }; |