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" |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 59 | |
| 60 | #include "buffer_head_io.h" |
| 61 | |
| 62 | static int ocfs2_sync_inode(struct inode *inode) |
| 63 | { |
| 64 | filemap_fdatawrite(inode->i_mapping); |
| 65 | return sync_mapping_buffers(inode->i_mapping); |
| 66 | } |
| 67 | |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 68 | static int ocfs2_init_file_private(struct inode *inode, struct file *file) |
| 69 | { |
| 70 | struct ocfs2_file_private *fp; |
| 71 | |
| 72 | fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL); |
| 73 | if (!fp) |
| 74 | return -ENOMEM; |
| 75 | |
| 76 | fp->fp_file = file; |
| 77 | mutex_init(&fp->fp_mutex); |
| 78 | ocfs2_file_lock_res_init(&fp->fp_flock, fp); |
| 79 | file->private_data = fp; |
| 80 | |
| 81 | return 0; |
| 82 | } |
| 83 | |
| 84 | static void ocfs2_free_file_private(struct inode *inode, struct file *file) |
| 85 | { |
| 86 | struct ocfs2_file_private *fp = file->private_data; |
| 87 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 88 | |
| 89 | if (fp) { |
| 90 | ocfs2_simple_drop_lockres(osb, &fp->fp_flock); |
| 91 | ocfs2_lock_res_free(&fp->fp_flock); |
| 92 | kfree(fp); |
| 93 | file->private_data = NULL; |
| 94 | } |
| 95 | } |
| 96 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 97 | static int ocfs2_file_open(struct inode *inode, struct file *file) |
| 98 | { |
| 99 | int status; |
| 100 | int mode = file->f_flags; |
| 101 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 102 | |
| 103 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 104 | 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] | 105 | |
| 106 | spin_lock(&oi->ip_lock); |
| 107 | |
| 108 | /* Check that the inode hasn't been wiped from disk by another |
| 109 | * node. If it hasn't then we're safe as long as we hold the |
| 110 | * spin lock until our increment of open count. */ |
| 111 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { |
| 112 | spin_unlock(&oi->ip_lock); |
| 113 | |
| 114 | status = -ENOENT; |
| 115 | goto leave; |
| 116 | } |
| 117 | |
| 118 | if (mode & O_DIRECT) |
| 119 | oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT; |
| 120 | |
| 121 | oi->ip_open_count++; |
| 122 | spin_unlock(&oi->ip_lock); |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 123 | |
| 124 | status = ocfs2_init_file_private(inode, file); |
| 125 | if (status) { |
| 126 | /* |
| 127 | * We want to set open count back if we're failing the |
| 128 | * open. |
| 129 | */ |
| 130 | spin_lock(&oi->ip_lock); |
| 131 | oi->ip_open_count--; |
| 132 | spin_unlock(&oi->ip_lock); |
| 133 | } |
| 134 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 135 | leave: |
| 136 | mlog_exit(status); |
| 137 | return status; |
| 138 | } |
| 139 | |
| 140 | static int ocfs2_file_release(struct inode *inode, struct file *file) |
| 141 | { |
| 142 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 143 | |
| 144 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 145 | file->f_path.dentry->d_name.len, |
| 146 | file->f_path.dentry->d_name.name); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 147 | |
| 148 | spin_lock(&oi->ip_lock); |
| 149 | if (!--oi->ip_open_count) |
| 150 | oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT; |
| 151 | spin_unlock(&oi->ip_lock); |
| 152 | |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 153 | ocfs2_free_file_private(inode, file); |
| 154 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 155 | mlog_exit(0); |
| 156 | |
| 157 | return 0; |
| 158 | } |
| 159 | |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 160 | static int ocfs2_dir_open(struct inode *inode, struct file *file) |
| 161 | { |
| 162 | return ocfs2_init_file_private(inode, file); |
| 163 | } |
| 164 | |
| 165 | static int ocfs2_dir_release(struct inode *inode, struct file *file) |
| 166 | { |
| 167 | ocfs2_free_file_private(inode, file); |
| 168 | return 0; |
| 169 | } |
| 170 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 171 | static int ocfs2_sync_file(struct file *file, |
| 172 | struct dentry *dentry, |
| 173 | int datasync) |
| 174 | { |
| 175 | int err = 0; |
| 176 | journal_t *journal; |
| 177 | struct inode *inode = dentry->d_inode; |
| 178 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 179 | |
| 180 | mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync, |
| 181 | dentry->d_name.len, dentry->d_name.name); |
| 182 | |
| 183 | err = ocfs2_sync_inode(dentry->d_inode); |
| 184 | if (err) |
| 185 | goto bail; |
| 186 | |
| 187 | journal = osb->journal->j_journal; |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 188 | err = jbd2_journal_force_commit(journal); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 189 | |
| 190 | bail: |
| 191 | mlog_exit(err); |
| 192 | |
| 193 | return (err < 0) ? -EIO : 0; |
| 194 | } |
| 195 | |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 196 | int ocfs2_should_update_atime(struct inode *inode, |
| 197 | struct vfsmount *vfsmnt) |
| 198 | { |
| 199 | struct timespec now; |
| 200 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 201 | |
| 202 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) |
| 203 | return 0; |
| 204 | |
| 205 | if ((inode->i_flags & S_NOATIME) || |
| 206 | ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))) |
| 207 | return 0; |
| 208 | |
Mark Fasheh | 6c2aad0 | 2006-12-19 15:25:52 -0800 | [diff] [blame] | 209 | /* |
| 210 | * We can be called with no vfsmnt structure - NFSD will |
| 211 | * sometimes do this. |
| 212 | * |
| 213 | * Note that our action here is different than touch_atime() - |
| 214 | * if we can't tell whether this is a noatime mount, then we |
| 215 | * don't know whether to trust the value of s_atime_quantum. |
| 216 | */ |
| 217 | if (vfsmnt == NULL) |
| 218 | return 0; |
| 219 | |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 220 | if ((vfsmnt->mnt_flags & MNT_NOATIME) || |
| 221 | ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) |
| 222 | return 0; |
| 223 | |
Mark Fasheh | 7e913c5 | 2006-12-13 00:34:35 -0800 | [diff] [blame] | 224 | if (vfsmnt->mnt_flags & MNT_RELATIME) { |
| 225 | if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) || |
| 226 | (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0)) |
| 227 | return 1; |
| 228 | |
| 229 | return 0; |
| 230 | } |
| 231 | |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 232 | now = CURRENT_TIME; |
| 233 | if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum)) |
| 234 | return 0; |
| 235 | else |
| 236 | return 1; |
| 237 | } |
| 238 | |
| 239 | int ocfs2_update_inode_atime(struct inode *inode, |
| 240 | struct buffer_head *bh) |
| 241 | { |
| 242 | int ret; |
| 243 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 244 | handle_t *handle; |
Mark Fasheh | c11e9fa | 2007-07-20 11:24:53 -0700 | [diff] [blame] | 245 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data; |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 246 | |
| 247 | mlog_entry_void(); |
| 248 | |
| 249 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
| 250 | if (handle == NULL) { |
| 251 | ret = -ENOMEM; |
| 252 | mlog_errno(ret); |
| 253 | goto out; |
| 254 | } |
| 255 | |
Mark Fasheh | c11e9fa | 2007-07-20 11:24:53 -0700 | [diff] [blame] | 256 | ret = ocfs2_journal_access(handle, inode, bh, |
| 257 | OCFS2_JOURNAL_ACCESS_WRITE); |
| 258 | if (ret) { |
| 259 | mlog_errno(ret); |
| 260 | goto out_commit; |
| 261 | } |
| 262 | |
| 263 | /* |
| 264 | * Don't use ocfs2_mark_inode_dirty() here as we don't always |
| 265 | * have i_mutex to guard against concurrent changes to other |
| 266 | * inode fields. |
| 267 | */ |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 268 | inode->i_atime = CURRENT_TIME; |
Mark Fasheh | c11e9fa | 2007-07-20 11:24:53 -0700 | [diff] [blame] | 269 | di->i_atime = cpu_to_le64(inode->i_atime.tv_sec); |
| 270 | di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec); |
| 271 | |
| 272 | ret = ocfs2_journal_dirty(handle, bh); |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 273 | if (ret < 0) |
| 274 | mlog_errno(ret); |
| 275 | |
Mark Fasheh | c11e9fa | 2007-07-20 11:24:53 -0700 | [diff] [blame] | 276 | out_commit: |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 277 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); |
| 278 | out: |
| 279 | mlog_exit(ret); |
| 280 | return ret; |
| 281 | } |
| 282 | |
Adrian Bunk | 6cb129f | 2007-04-26 00:29:35 -0700 | [diff] [blame] | 283 | static int ocfs2_set_inode_size(handle_t *handle, |
| 284 | struct inode *inode, |
| 285 | struct buffer_head *fe_bh, |
| 286 | u64 new_i_size) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 287 | { |
| 288 | int status; |
| 289 | |
| 290 | mlog_entry_void(); |
| 291 | i_size_write(inode, new_i_size); |
Mark Fasheh | 8110b07 | 2007-03-22 16:53:23 -0700 | [diff] [blame] | 292 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 293 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
| 294 | |
| 295 | status = ocfs2_mark_inode_dirty(handle, inode, fe_bh); |
| 296 | if (status < 0) { |
| 297 | mlog_errno(status); |
| 298 | goto bail; |
| 299 | } |
| 300 | |
| 301 | bail: |
| 302 | mlog_exit(status); |
| 303 | return status; |
| 304 | } |
| 305 | |
| 306 | static int ocfs2_simple_size_update(struct inode *inode, |
| 307 | struct buffer_head *di_bh, |
| 308 | u64 new_i_size) |
| 309 | { |
| 310 | int ret; |
| 311 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 312 | handle_t *handle = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 313 | |
Mark Fasheh | 65eff9c | 2006-10-09 17:26:22 -0700 | [diff] [blame] | 314 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 315 | if (handle == NULL) { |
| 316 | ret = -ENOMEM; |
| 317 | mlog_errno(ret); |
| 318 | goto out; |
| 319 | } |
| 320 | |
| 321 | ret = ocfs2_set_inode_size(handle, inode, di_bh, |
| 322 | new_i_size); |
| 323 | if (ret < 0) |
| 324 | mlog_errno(ret); |
| 325 | |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 326 | ocfs2_commit_trans(osb, handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 327 | out: |
| 328 | return ret; |
| 329 | } |
| 330 | |
| 331 | static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb, |
| 332 | struct inode *inode, |
| 333 | struct buffer_head *fe_bh, |
| 334 | u64 new_i_size) |
| 335 | { |
| 336 | int status; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 337 | handle_t *handle; |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 338 | struct ocfs2_dinode *di; |
Mark Fasheh | 35edec1 | 2007-07-06 14:41:18 -0700 | [diff] [blame] | 339 | u64 cluster_bytes; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 340 | |
| 341 | mlog_entry_void(); |
| 342 | |
| 343 | /* TODO: This needs to actually orphan the inode in this |
| 344 | * transaction. */ |
| 345 | |
Mark Fasheh | 65eff9c | 2006-10-09 17:26:22 -0700 | [diff] [blame] | 346 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 347 | if (IS_ERR(handle)) { |
| 348 | status = PTR_ERR(handle); |
| 349 | mlog_errno(status); |
| 350 | goto out; |
| 351 | } |
| 352 | |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 353 | status = ocfs2_journal_access(handle, inode, fe_bh, |
| 354 | OCFS2_JOURNAL_ACCESS_WRITE); |
| 355 | if (status < 0) { |
| 356 | mlog_errno(status); |
| 357 | goto out_commit; |
| 358 | } |
| 359 | |
| 360 | /* |
| 361 | * Do this before setting i_size. |
| 362 | */ |
Mark Fasheh | 35edec1 | 2007-07-06 14:41:18 -0700 | [diff] [blame] | 363 | cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size); |
| 364 | status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size, |
| 365 | cluster_bytes); |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 366 | if (status) { |
| 367 | mlog_errno(status); |
| 368 | goto out_commit; |
| 369 | } |
| 370 | |
| 371 | i_size_write(inode, new_i_size); |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 372 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
| 373 | |
| 374 | di = (struct ocfs2_dinode *) fe_bh->b_data; |
| 375 | di->i_size = cpu_to_le64(new_i_size); |
| 376 | di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec); |
| 377 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); |
| 378 | |
| 379 | status = ocfs2_journal_dirty(handle, fe_bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 380 | if (status < 0) |
| 381 | mlog_errno(status); |
| 382 | |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 383 | out_commit: |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 384 | ocfs2_commit_trans(osb, handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 385 | out: |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 386 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 387 | mlog_exit(status); |
| 388 | return status; |
| 389 | } |
| 390 | |
| 391 | static int ocfs2_truncate_file(struct inode *inode, |
| 392 | struct buffer_head *di_bh, |
| 393 | u64 new_i_size) |
| 394 | { |
| 395 | int status = 0; |
| 396 | struct ocfs2_dinode *fe = NULL; |
| 397 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 398 | struct ocfs2_truncate_context *tc = NULL; |
| 399 | |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 400 | mlog_entry("(inode = %llu, new_i_size = %llu\n", |
| 401 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
| 402 | (unsigned long long)new_i_size); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 403 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 404 | fe = (struct ocfs2_dinode *) di_bh->b_data; |
| 405 | if (!OCFS2_IS_VALID_DINODE(fe)) { |
| 406 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); |
| 407 | status = -EIO; |
| 408 | goto bail; |
| 409 | } |
| 410 | |
| 411 | 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] | 412 | "Inode %llu, inode i_size = %lld != di " |
| 413 | "i_size = %llu, i_flags = 0x%x\n", |
| 414 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 415 | i_size_read(inode), |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 416 | (unsigned long long)le64_to_cpu(fe->i_size), |
| 417 | le32_to_cpu(fe->i_flags)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 418 | |
| 419 | if (new_i_size > le64_to_cpu(fe->i_size)) { |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 420 | mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n", |
| 421 | (unsigned long long)le64_to_cpu(fe->i_size), |
| 422 | (unsigned long long)new_i_size); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 423 | status = -EINVAL; |
| 424 | mlog_errno(status); |
| 425 | goto bail; |
| 426 | } |
| 427 | |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 428 | mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n", |
| 429 | (unsigned long long)le64_to_cpu(fe->i_blkno), |
| 430 | (unsigned long long)le64_to_cpu(fe->i_size), |
| 431 | (unsigned long long)new_i_size); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 432 | |
| 433 | /* lets handle the simple truncate cases before doing any more |
| 434 | * cluster locking. */ |
| 435 | if (new_i_size == le64_to_cpu(fe->i_size)) |
| 436 | goto bail; |
| 437 | |
Mark Fasheh | 2e89b2e | 2007-05-09 13:40:18 -0700 | [diff] [blame] | 438 | down_write(&OCFS2_I(inode)->ip_alloc_sem); |
| 439 | |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 440 | /* |
| 441 | * The inode lock forced other nodes to sync and drop their |
| 442 | * pages, which (correctly) happens even if we have a truncate |
| 443 | * without allocation change - ocfs2 cluster sizes can be much |
| 444 | * greater than page size, so we have to truncate them |
| 445 | * anyway. |
| 446 | */ |
Mark Fasheh | 2e89b2e | 2007-05-09 13:40:18 -0700 | [diff] [blame] | 447 | unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1); |
| 448 | truncate_inode_pages(inode->i_mapping, new_i_size); |
| 449 | |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 450 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
| 451 | status = ocfs2_truncate_inline(inode, di_bh, new_i_size, |
Mark Fasheh | b1967d0 | 2007-11-20 11:56:39 -0800 | [diff] [blame] | 452 | i_size_read(inode), 1); |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 453 | if (status) |
| 454 | mlog_errno(status); |
| 455 | |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 456 | goto bail_unlock_sem; |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 457 | } |
| 458 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 459 | /* alright, we're going to need to do a full blown alloc size |
| 460 | * change. Orphan the inode so that recovery can complete the |
| 461 | * truncate if necessary. This does the task of marking |
| 462 | * i_size. */ |
| 463 | status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size); |
| 464 | if (status < 0) { |
| 465 | mlog_errno(status); |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 466 | goto bail_unlock_sem; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 467 | } |
| 468 | |
| 469 | status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc); |
| 470 | if (status < 0) { |
| 471 | mlog_errno(status); |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 472 | goto bail_unlock_sem; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 473 | } |
| 474 | |
| 475 | status = ocfs2_commit_truncate(osb, inode, di_bh, tc); |
| 476 | if (status < 0) { |
| 477 | mlog_errno(status); |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 478 | goto bail_unlock_sem; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 479 | } |
| 480 | |
| 481 | /* TODO: orphan dir cleanup here. */ |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 482 | bail_unlock_sem: |
Mark Fasheh | 2e89b2e | 2007-05-09 13:40:18 -0700 | [diff] [blame] | 483 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
| 484 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 485 | bail: |
| 486 | |
| 487 | mlog_exit(status); |
| 488 | return status; |
| 489 | } |
| 490 | |
| 491 | /* |
Tao Ma | 0eb8d47 | 2008-08-18 17:38:45 +0800 | [diff] [blame] | 492 | * extend file allocation only here. |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 493 | * we'll update all the disk stuff, and oip->alloc_size |
| 494 | * |
| 495 | * expect stuff to be locked, a transaction started and enough data / |
| 496 | * metadata reservations in the contexts. |
| 497 | * |
| 498 | * Will return -EAGAIN, and a reason if a restart is needed. |
| 499 | * If passed in, *reason will always be set, even in error. |
| 500 | */ |
Tao Ma | 0eb8d47 | 2008-08-18 17:38:45 +0800 | [diff] [blame] | 501 | int ocfs2_add_inode_data(struct ocfs2_super *osb, |
| 502 | struct inode *inode, |
| 503 | u32 *logical_offset, |
| 504 | u32 clusters_to_add, |
| 505 | int mark_unwritten, |
| 506 | struct buffer_head *fe_bh, |
| 507 | handle_t *handle, |
| 508 | struct ocfs2_alloc_context *data_ac, |
| 509 | struct ocfs2_alloc_context *meta_ac, |
| 510 | enum ocfs2_alloc_restarted *reason_ret) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 511 | { |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 512 | int ret; |
| 513 | struct ocfs2_extent_tree et; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 514 | |
Joel Becker | 8d6220d | 2008-08-22 12:46:09 -0700 | [diff] [blame] | 515 | ocfs2_init_dinode_extent_tree(&et, inode, fe_bh); |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 516 | ret = ocfs2_add_clusters_in_btree(osb, inode, logical_offset, |
Tao Ma | 0eb8d47 | 2008-08-18 17:38:45 +0800 | [diff] [blame] | 517 | clusters_to_add, mark_unwritten, |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 518 | &et, handle, |
| 519 | data_ac, meta_ac, reason_ret); |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 520 | |
| 521 | return ret; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 522 | } |
| 523 | |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 524 | static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start, |
| 525 | u32 clusters_to_add, int mark_unwritten) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 526 | { |
| 527 | int status = 0; |
| 528 | int restart_func = 0; |
Mark Fasheh | abf8b15 | 2007-01-17 13:07:24 -0800 | [diff] [blame] | 529 | int credits; |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 530 | u32 prev_clusters; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 531 | struct buffer_head *bh = NULL; |
| 532 | struct ocfs2_dinode *fe = NULL; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 533 | handle_t *handle = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 534 | struct ocfs2_alloc_context *data_ac = NULL; |
| 535 | struct ocfs2_alloc_context *meta_ac = NULL; |
| 536 | enum ocfs2_alloc_restarted why; |
| 537 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 538 | struct ocfs2_extent_tree et; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 539 | |
| 540 | mlog_entry("(clusters_to_add = %u)\n", clusters_to_add); |
| 541 | |
Mark Fasheh | dcd0538 | 2007-01-16 11:32:23 -0800 | [diff] [blame] | 542 | /* |
| 543 | * This function only exists for file systems which don't |
| 544 | * support holes. |
| 545 | */ |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 546 | BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb)); |
Mark Fasheh | dcd0538 | 2007-01-16 11:32:23 -0800 | [diff] [blame] | 547 | |
Joel Becker | 0fcaa56 | 2008-10-09 17:20:31 -0700 | [diff] [blame] | 548 | status = ocfs2_read_block(inode, OCFS2_I(inode)->ip_blkno, &bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 549 | if (status < 0) { |
| 550 | mlog_errno(status); |
| 551 | goto leave; |
| 552 | } |
| 553 | |
| 554 | fe = (struct ocfs2_dinode *) bh->b_data; |
| 555 | if (!OCFS2_IS_VALID_DINODE(fe)) { |
| 556 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); |
| 557 | status = -EIO; |
| 558 | goto leave; |
| 559 | } |
| 560 | |
| 561 | restart_all: |
| 562 | BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters); |
| 563 | |
Tao Ma | e7d4cb6 | 2008-08-18 17:38:44 +0800 | [diff] [blame] | 564 | mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, " |
| 565 | "clusters_to_add = %u\n", |
| 566 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
| 567 | (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters), |
| 568 | clusters_to_add); |
Joel Becker | 8d6220d | 2008-08-22 12:46:09 -0700 | [diff] [blame] | 569 | ocfs2_init_dinode_extent_tree(&et, inode, bh); |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 570 | status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0, |
| 571 | &data_ac, &meta_ac); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 572 | if (status) { |
| 573 | mlog_errno(status); |
| 574 | goto leave; |
| 575 | } |
| 576 | |
Tao Ma | 811f933 | 2008-08-18 17:38:43 +0800 | [diff] [blame] | 577 | credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list, |
| 578 | clusters_to_add); |
Mark Fasheh | 65eff9c | 2006-10-09 17:26:22 -0700 | [diff] [blame] | 579 | handle = ocfs2_start_trans(osb, credits); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 580 | if (IS_ERR(handle)) { |
| 581 | status = PTR_ERR(handle); |
| 582 | handle = NULL; |
| 583 | mlog_errno(status); |
| 584 | goto leave; |
| 585 | } |
| 586 | |
| 587 | restarted_transaction: |
| 588 | /* reserve a write to the file entry early on - that we if we |
| 589 | * run out of credits in the allocation path, we can still |
| 590 | * update i_size. */ |
| 591 | status = ocfs2_journal_access(handle, inode, bh, |
| 592 | OCFS2_JOURNAL_ACCESS_WRITE); |
| 593 | if (status < 0) { |
| 594 | mlog_errno(status); |
| 595 | goto leave; |
| 596 | } |
| 597 | |
| 598 | prev_clusters = OCFS2_I(inode)->ip_clusters; |
| 599 | |
Tao Ma | 0eb8d47 | 2008-08-18 17:38:45 +0800 | [diff] [blame] | 600 | status = ocfs2_add_inode_data(osb, |
| 601 | inode, |
| 602 | &logical_start, |
| 603 | clusters_to_add, |
| 604 | mark_unwritten, |
| 605 | bh, |
| 606 | handle, |
| 607 | data_ac, |
| 608 | meta_ac, |
| 609 | &why); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 610 | if ((status < 0) && (status != -EAGAIN)) { |
| 611 | if (status != -ENOSPC) |
| 612 | mlog_errno(status); |
| 613 | goto leave; |
| 614 | } |
| 615 | |
| 616 | status = ocfs2_journal_dirty(handle, bh); |
| 617 | if (status < 0) { |
| 618 | mlog_errno(status); |
| 619 | goto leave; |
| 620 | } |
| 621 | |
| 622 | spin_lock(&OCFS2_I(inode)->ip_lock); |
| 623 | clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters); |
| 624 | spin_unlock(&OCFS2_I(inode)->ip_lock); |
| 625 | |
| 626 | if (why != RESTART_NONE && clusters_to_add) { |
| 627 | if (why == RESTART_META) { |
| 628 | mlog(0, "restarting function.\n"); |
| 629 | restart_func = 1; |
| 630 | } else { |
| 631 | BUG_ON(why != RESTART_TRANS); |
| 632 | |
| 633 | mlog(0, "restarting transaction.\n"); |
| 634 | /* TODO: This can be more intelligent. */ |
| 635 | credits = ocfs2_calc_extend_credits(osb->sb, |
Tao Ma | 811f933 | 2008-08-18 17:38:43 +0800 | [diff] [blame] | 636 | &fe->id2.i_list, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 637 | clusters_to_add); |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 638 | status = ocfs2_extend_trans(handle, credits); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 639 | if (status < 0) { |
| 640 | /* handle still has to be committed at |
| 641 | * this point. */ |
| 642 | status = -ENOMEM; |
| 643 | mlog_errno(status); |
| 644 | goto leave; |
| 645 | } |
| 646 | goto restarted_transaction; |
| 647 | } |
| 648 | } |
| 649 | |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 650 | mlog(0, "fe: i_clusters = %u, i_size=%llu\n", |
Mark Fasheh | 1ca1a11 | 2007-04-27 16:01:25 -0700 | [diff] [blame] | 651 | le32_to_cpu(fe->i_clusters), |
| 652 | (unsigned long long)le64_to_cpu(fe->i_size)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 653 | mlog(0, "inode: ip_clusters=%u, i_size=%lld\n", |
Jan Kara | 634bf74 | 2007-12-19 15:25:42 +0100 | [diff] [blame] | 654 | OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 655 | |
| 656 | leave: |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 657 | if (handle) { |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 658 | ocfs2_commit_trans(osb, handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 659 | handle = NULL; |
| 660 | } |
| 661 | if (data_ac) { |
| 662 | ocfs2_free_alloc_context(data_ac); |
| 663 | data_ac = NULL; |
| 664 | } |
| 665 | if (meta_ac) { |
| 666 | ocfs2_free_alloc_context(meta_ac); |
| 667 | meta_ac = NULL; |
| 668 | } |
| 669 | if ((!status) && restart_func) { |
| 670 | restart_func = 0; |
| 671 | goto restart_all; |
| 672 | } |
Mark Fasheh | a81cb88 | 2008-10-07 14:25:16 -0700 | [diff] [blame] | 673 | brelse(bh); |
| 674 | bh = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 675 | |
| 676 | mlog_exit(status); |
| 677 | return status; |
| 678 | } |
| 679 | |
| 680 | /* Some parts of this taken from generic_cont_expand, which turned out |
| 681 | * 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^] | 682 | * worry about recursive locking in ->write_begin() and ->write_end(). */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 683 | static int ocfs2_write_zero_page(struct inode *inode, |
| 684 | u64 size) |
| 685 | { |
| 686 | struct address_space *mapping = inode->i_mapping; |
| 687 | struct page *page; |
| 688 | unsigned long index; |
| 689 | unsigned int offset; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 690 | handle_t *handle = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 691 | int ret; |
| 692 | |
| 693 | offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */ |
| 694 | /* ugh. in prepare/commit_write, if from==to==start of block, we |
| 695 | ** skip the prepare. make sure we never send an offset for the start |
| 696 | ** of a block |
| 697 | */ |
| 698 | if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) { |
| 699 | offset++; |
| 700 | } |
| 701 | index = size >> PAGE_CACHE_SHIFT; |
| 702 | |
| 703 | page = grab_cache_page(mapping, index); |
| 704 | if (!page) { |
| 705 | ret = -ENOMEM; |
| 706 | mlog_errno(ret); |
| 707 | goto out; |
| 708 | } |
| 709 | |
Mark Fasheh | 53013cb | 2006-05-05 19:04:03 -0700 | [diff] [blame] | 710 | ret = ocfs2_prepare_write_nolock(inode, page, offset, offset); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 711 | if (ret < 0) { |
| 712 | mlog_errno(ret); |
| 713 | goto out_unlock; |
| 714 | } |
| 715 | |
| 716 | if (ocfs2_should_order_data(inode)) { |
| 717 | handle = ocfs2_start_walk_page_trans(inode, page, offset, |
| 718 | offset); |
| 719 | if (IS_ERR(handle)) { |
| 720 | ret = PTR_ERR(handle); |
| 721 | handle = NULL; |
| 722 | goto out_unlock; |
| 723 | } |
| 724 | } |
| 725 | |
| 726 | /* must not update i_size! */ |
| 727 | ret = block_commit_write(page, offset, offset); |
| 728 | if (ret < 0) |
| 729 | mlog_errno(ret); |
| 730 | else |
| 731 | ret = 0; |
| 732 | |
| 733 | if (handle) |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 734 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 735 | out_unlock: |
| 736 | unlock_page(page); |
| 737 | page_cache_release(page); |
| 738 | out: |
| 739 | return ret; |
| 740 | } |
| 741 | |
| 742 | static int ocfs2_zero_extend(struct inode *inode, |
| 743 | u64 zero_to_size) |
| 744 | { |
| 745 | int ret = 0; |
| 746 | u64 start_off; |
| 747 | struct super_block *sb = inode->i_sb; |
| 748 | |
| 749 | start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode)); |
| 750 | while (start_off < zero_to_size) { |
| 751 | ret = ocfs2_write_zero_page(inode, start_off); |
| 752 | if (ret < 0) { |
| 753 | mlog_errno(ret); |
| 754 | goto out; |
| 755 | } |
| 756 | |
| 757 | start_off += sb->s_blocksize; |
Mark Fasheh | e2057c5 | 2006-10-03 17:53:05 -0700 | [diff] [blame] | 758 | |
| 759 | /* |
| 760 | * Very large extends have the potential to lock up |
| 761 | * the cpu for extended periods of time. |
| 762 | */ |
| 763 | cond_resched(); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 764 | } |
| 765 | |
| 766 | out: |
| 767 | return ret; |
| 768 | } |
| 769 | |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 770 | int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to) |
| 771 | { |
| 772 | int ret; |
| 773 | u32 clusters_to_add; |
| 774 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 775 | |
| 776 | clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size); |
| 777 | if (clusters_to_add < oi->ip_clusters) |
| 778 | clusters_to_add = 0; |
| 779 | else |
| 780 | clusters_to_add -= oi->ip_clusters; |
| 781 | |
| 782 | if (clusters_to_add) { |
| 783 | ret = __ocfs2_extend_allocation(inode, oi->ip_clusters, |
| 784 | clusters_to_add, 0); |
| 785 | if (ret) { |
| 786 | mlog_errno(ret); |
| 787 | goto out; |
| 788 | } |
| 789 | } |
| 790 | |
| 791 | /* |
| 792 | * Call this even if we don't add any clusters to the tree. We |
| 793 | * still need to zero the area between the old i_size and the |
| 794 | * new i_size. |
| 795 | */ |
| 796 | ret = ocfs2_zero_extend(inode, zero_to); |
| 797 | if (ret < 0) |
| 798 | mlog_errno(ret); |
| 799 | |
| 800 | out: |
| 801 | return ret; |
| 802 | } |
| 803 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 804 | static int ocfs2_extend_file(struct inode *inode, |
| 805 | struct buffer_head *di_bh, |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 806 | u64 new_i_size) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 807 | { |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 808 | int ret = 0; |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 809 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 810 | |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 811 | BUG_ON(!di_bh); |
Mark Fasheh | 53013cb | 2006-05-05 19:04:03 -0700 | [diff] [blame] | 812 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 813 | /* setattr sometimes calls us like this. */ |
| 814 | if (new_i_size == 0) |
| 815 | goto out; |
| 816 | |
| 817 | if (i_size_read(inode) == new_i_size) |
| 818 | goto out; |
| 819 | BUG_ON(new_i_size < i_size_read(inode)); |
| 820 | |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 821 | /* |
| 822 | * Fall through for converting inline data, even if the fs |
| 823 | * supports sparse files. |
| 824 | * |
| 825 | * The check for inline data here is legal - nobody can add |
| 826 | * the feature since we have i_mutex. We must check it again |
| 827 | * after acquiring ip_alloc_sem though, as paths like mmap |
| 828 | * might have raced us to converting the inode to extents. |
| 829 | */ |
| 830 | if (!(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) |
| 831 | && ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) |
Mark Fasheh | 3a0782d | 2007-01-17 12:53:31 -0800 | [diff] [blame] | 832 | goto out_update_size; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 833 | |
Mark Fasheh | 0effef7 | 2006-10-03 17:44:42 -0700 | [diff] [blame] | 834 | /* |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 835 | * The alloc sem blocks people in read/write from reading our |
| 836 | * allocation until we're done changing it. We depend on |
| 837 | * i_mutex to block other extend/truncate calls while we're |
| 838 | * here. |
Mark Fasheh | 0effef7 | 2006-10-03 17:44:42 -0700 | [diff] [blame] | 839 | */ |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 840 | down_write(&oi->ip_alloc_sem); |
| 841 | |
| 842 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
| 843 | /* |
| 844 | * We can optimize small extends by keeping the inodes |
| 845 | * inline data. |
| 846 | */ |
| 847 | if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) { |
| 848 | up_write(&oi->ip_alloc_sem); |
| 849 | goto out_update_size; |
| 850 | } |
| 851 | |
| 852 | ret = ocfs2_convert_inline_data_to_extents(inode, di_bh); |
| 853 | if (ret) { |
| 854 | up_write(&oi->ip_alloc_sem); |
| 855 | |
| 856 | mlog_errno(ret); |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 857 | goto out; |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 858 | } |
| 859 | } |
| 860 | |
| 861 | if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) |
| 862 | ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size); |
| 863 | |
| 864 | up_write(&oi->ip_alloc_sem); |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 865 | |
Mark Fasheh | 0effef7 | 2006-10-03 17:44:42 -0700 | [diff] [blame] | 866 | if (ret < 0) { |
| 867 | mlog_errno(ret); |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 868 | goto out; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 869 | } |
| 870 | |
Mark Fasheh | 3a0782d | 2007-01-17 12:53:31 -0800 | [diff] [blame] | 871 | out_update_size: |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 872 | ret = ocfs2_simple_size_update(inode, di_bh, new_i_size); |
| 873 | if (ret < 0) |
| 874 | mlog_errno(ret); |
Mark Fasheh | 53013cb | 2006-05-05 19:04:03 -0700 | [diff] [blame] | 875 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 876 | out: |
| 877 | return ret; |
| 878 | } |
| 879 | |
| 880 | int ocfs2_setattr(struct dentry *dentry, struct iattr *attr) |
| 881 | { |
| 882 | int status = 0, size_change; |
| 883 | struct inode *inode = dentry->d_inode; |
| 884 | struct super_block *sb = inode->i_sb; |
| 885 | struct ocfs2_super *osb = OCFS2_SB(sb); |
| 886 | struct buffer_head *bh = NULL; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 887 | handle_t *handle = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 888 | |
| 889 | mlog_entry("(0x%p, '%.*s')\n", dentry, |
| 890 | dentry->d_name.len, dentry->d_name.name); |
| 891 | |
Sunil Mushran | bc53580 | 2008-04-18 10:23:53 -0700 | [diff] [blame] | 892 | /* ensuring we don't even attempt to truncate a symlink */ |
| 893 | if (S_ISLNK(inode->i_mode)) |
| 894 | attr->ia_valid &= ~ATTR_SIZE; |
| 895 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 896 | if (attr->ia_valid & ATTR_MODE) |
| 897 | mlog(0, "mode change: %d\n", attr->ia_mode); |
| 898 | if (attr->ia_valid & ATTR_UID) |
| 899 | mlog(0, "uid change: %d\n", attr->ia_uid); |
| 900 | if (attr->ia_valid & ATTR_GID) |
| 901 | mlog(0, "gid change: %d\n", attr->ia_gid); |
| 902 | if (attr->ia_valid & ATTR_SIZE) |
| 903 | mlog(0, "size change...\n"); |
| 904 | if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME)) |
| 905 | mlog(0, "time change...\n"); |
| 906 | |
| 907 | #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \ |
| 908 | | ATTR_GID | ATTR_UID | ATTR_MODE) |
| 909 | if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) { |
| 910 | mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid); |
| 911 | return 0; |
| 912 | } |
| 913 | |
| 914 | status = inode_change_ok(inode, attr); |
| 915 | if (status) |
| 916 | return status; |
| 917 | |
| 918 | size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE; |
| 919 | if (size_change) { |
| 920 | status = ocfs2_rw_lock(inode, 1); |
| 921 | if (status < 0) { |
| 922 | mlog_errno(status); |
| 923 | goto bail; |
| 924 | } |
| 925 | } |
| 926 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 927 | status = ocfs2_inode_lock(inode, &bh, 1); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 928 | if (status < 0) { |
| 929 | if (status != -ENOENT) |
| 930 | mlog_errno(status); |
| 931 | goto bail_unlock_rw; |
| 932 | } |
| 933 | |
| 934 | if (size_change && attr->ia_size != i_size_read(inode)) { |
Mark Fasheh | ce76fd3 | 2007-07-20 12:02:14 -0700 | [diff] [blame] | 935 | if (attr->ia_size > sb->s_maxbytes) { |
| 936 | status = -EFBIG; |
| 937 | goto bail_unlock; |
| 938 | } |
| 939 | |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 940 | if (i_size_read(inode) > attr->ia_size) { |
| 941 | if (ocfs2_should_order_data(inode)) { |
| 942 | status = ocfs2_begin_ordered_truncate(inode, |
| 943 | attr->ia_size); |
| 944 | if (status) |
| 945 | goto bail_unlock; |
| 946 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 947 | status = ocfs2_truncate_file(inode, bh, attr->ia_size); |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 948 | } else |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 949 | status = ocfs2_extend_file(inode, bh, attr->ia_size); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 950 | if (status < 0) { |
| 951 | if (status != -ENOSPC) |
| 952 | mlog_errno(status); |
| 953 | status = -ENOSPC; |
| 954 | goto bail_unlock; |
| 955 | } |
| 956 | } |
| 957 | |
Mark Fasheh | 65eff9c | 2006-10-09 17:26:22 -0700 | [diff] [blame] | 958 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 959 | if (IS_ERR(handle)) { |
| 960 | status = PTR_ERR(handle); |
| 961 | mlog_errno(status); |
| 962 | goto bail_unlock; |
| 963 | } |
| 964 | |
Mark Fasheh | 7307de8 | 2007-05-09 15:16:19 -0700 | [diff] [blame] | 965 | /* |
| 966 | * This will intentionally not wind up calling vmtruncate(), |
| 967 | * since all the work for a size change has been done above. |
| 968 | * Otherwise, we could get into problems with truncate as |
| 969 | * ip_alloc_sem is used there to protect against i_size |
| 970 | * changes. |
| 971 | */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 972 | status = inode_setattr(inode, attr); |
| 973 | if (status < 0) { |
| 974 | mlog_errno(status); |
| 975 | goto bail_commit; |
| 976 | } |
| 977 | |
| 978 | status = ocfs2_mark_inode_dirty(handle, inode, bh); |
| 979 | if (status < 0) |
| 980 | mlog_errno(status); |
| 981 | |
| 982 | bail_commit: |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 983 | ocfs2_commit_trans(osb, handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 984 | bail_unlock: |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 985 | ocfs2_inode_unlock(inode, 1); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 986 | bail_unlock_rw: |
| 987 | if (size_change) |
| 988 | ocfs2_rw_unlock(inode, 1); |
| 989 | bail: |
Mark Fasheh | a81cb88 | 2008-10-07 14:25:16 -0700 | [diff] [blame] | 990 | brelse(bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 991 | |
| 992 | mlog_exit(status); |
| 993 | return status; |
| 994 | } |
| 995 | |
| 996 | int ocfs2_getattr(struct vfsmount *mnt, |
| 997 | struct dentry *dentry, |
| 998 | struct kstat *stat) |
| 999 | { |
| 1000 | struct inode *inode = dentry->d_inode; |
| 1001 | struct super_block *sb = dentry->d_inode->i_sb; |
| 1002 | struct ocfs2_super *osb = sb->s_fs_info; |
| 1003 | int err; |
| 1004 | |
| 1005 | mlog_entry_void(); |
| 1006 | |
| 1007 | err = ocfs2_inode_revalidate(dentry); |
| 1008 | if (err) { |
| 1009 | if (err != -ENOENT) |
| 1010 | mlog_errno(err); |
| 1011 | goto bail; |
| 1012 | } |
| 1013 | |
| 1014 | generic_fillattr(inode, stat); |
| 1015 | |
| 1016 | /* We set the blksize from the cluster size for performance */ |
| 1017 | stat->blksize = osb->s_clustersize; |
| 1018 | |
| 1019 | bail: |
| 1020 | mlog_exit(err); |
| 1021 | |
| 1022 | return err; |
| 1023 | } |
| 1024 | |
Al Viro | e6305c4 | 2008-07-15 21:03:57 -0400 | [diff] [blame] | 1025 | int ocfs2_permission(struct inode *inode, int mask) |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 1026 | { |
| 1027 | int ret; |
| 1028 | |
| 1029 | mlog_entry_void(); |
| 1030 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1031 | ret = ocfs2_inode_lock(inode, NULL, 0); |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 1032 | if (ret) { |
Mark Fasheh | a9f5f70 | 2007-04-26 11:43:43 -0700 | [diff] [blame] | 1033 | if (ret != -ENOENT) |
| 1034 | mlog_errno(ret); |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 1035 | goto out; |
| 1036 | } |
| 1037 | |
| 1038 | ret = generic_permission(inode, mask, NULL); |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 1039 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1040 | ocfs2_inode_unlock(inode, 0); |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 1041 | out: |
| 1042 | mlog_exit(ret); |
| 1043 | return ret; |
| 1044 | } |
| 1045 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1046 | static int __ocfs2_write_remove_suid(struct inode *inode, |
| 1047 | struct buffer_head *bh) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1048 | { |
| 1049 | int ret; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 1050 | handle_t *handle; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1051 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1052 | struct ocfs2_dinode *di; |
| 1053 | |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 1054 | mlog_entry("(Inode %llu, mode 0%o)\n", |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1055 | (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1056 | |
Mark Fasheh | 65eff9c | 2006-10-09 17:26:22 -0700 | [diff] [blame] | 1057 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1058 | if (handle == NULL) { |
| 1059 | ret = -ENOMEM; |
| 1060 | mlog_errno(ret); |
| 1061 | goto out; |
| 1062 | } |
| 1063 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1064 | ret = ocfs2_journal_access(handle, inode, bh, |
| 1065 | OCFS2_JOURNAL_ACCESS_WRITE); |
| 1066 | if (ret < 0) { |
| 1067 | mlog_errno(ret); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1068 | goto out_trans; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1069 | } |
| 1070 | |
| 1071 | inode->i_mode &= ~S_ISUID; |
| 1072 | if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP)) |
| 1073 | inode->i_mode &= ~S_ISGID; |
| 1074 | |
| 1075 | di = (struct ocfs2_dinode *) bh->b_data; |
| 1076 | di->i_mode = cpu_to_le16(inode->i_mode); |
| 1077 | |
| 1078 | ret = ocfs2_journal_dirty(handle, bh); |
| 1079 | if (ret < 0) |
| 1080 | mlog_errno(ret); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1081 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1082 | out_trans: |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 1083 | ocfs2_commit_trans(osb, handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1084 | out: |
| 1085 | mlog_exit(ret); |
| 1086 | return ret; |
| 1087 | } |
| 1088 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1089 | /* |
| 1090 | * Will look for holes and unwritten extents in the range starting at |
| 1091 | * pos for count bytes (inclusive). |
| 1092 | */ |
| 1093 | static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos, |
| 1094 | size_t count) |
| 1095 | { |
| 1096 | int ret = 0; |
Mark Fasheh | 49cb8d2 | 2007-03-09 16:21:46 -0800 | [diff] [blame] | 1097 | unsigned int extent_flags; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1098 | u32 cpos, clusters, extent_len, phys_cpos; |
| 1099 | struct super_block *sb = inode->i_sb; |
| 1100 | |
| 1101 | cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits; |
| 1102 | clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos; |
| 1103 | |
| 1104 | while (clusters) { |
Mark Fasheh | 49cb8d2 | 2007-03-09 16:21:46 -0800 | [diff] [blame] | 1105 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len, |
| 1106 | &extent_flags); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1107 | if (ret < 0) { |
| 1108 | mlog_errno(ret); |
| 1109 | goto out; |
| 1110 | } |
| 1111 | |
Mark Fasheh | 49cb8d2 | 2007-03-09 16:21:46 -0800 | [diff] [blame] | 1112 | if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) { |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1113 | ret = 1; |
| 1114 | break; |
| 1115 | } |
| 1116 | |
| 1117 | if (extent_len > clusters) |
| 1118 | extent_len = clusters; |
| 1119 | |
| 1120 | clusters -= extent_len; |
| 1121 | cpos += extent_len; |
| 1122 | } |
| 1123 | out: |
| 1124 | return ret; |
| 1125 | } |
| 1126 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1127 | static int ocfs2_write_remove_suid(struct inode *inode) |
| 1128 | { |
| 1129 | int ret; |
| 1130 | struct buffer_head *bh = NULL; |
| 1131 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 1132 | |
Joel Becker | 0fcaa56 | 2008-10-09 17:20:31 -0700 | [diff] [blame] | 1133 | ret = ocfs2_read_block(inode, oi->ip_blkno, &bh); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1134 | if (ret < 0) { |
| 1135 | mlog_errno(ret); |
| 1136 | goto out; |
| 1137 | } |
| 1138 | |
| 1139 | ret = __ocfs2_write_remove_suid(inode, bh); |
| 1140 | out: |
| 1141 | brelse(bh); |
| 1142 | return ret; |
| 1143 | } |
| 1144 | |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 1145 | /* |
| 1146 | * Allocate enough extents to cover the region starting at byte offset |
| 1147 | * start for len bytes. Existing extents are skipped, any extents |
| 1148 | * added are marked as "unwritten". |
| 1149 | */ |
| 1150 | static int ocfs2_allocate_unwritten_extents(struct inode *inode, |
| 1151 | u64 start, u64 len) |
| 1152 | { |
| 1153 | int ret; |
| 1154 | u32 cpos, phys_cpos, clusters, alloc_size; |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1155 | u64 end = start + len; |
| 1156 | struct buffer_head *di_bh = NULL; |
| 1157 | |
| 1158 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
Joel Becker | 31d3307 | 2008-10-09 17:20:30 -0700 | [diff] [blame] | 1159 | ret = ocfs2_read_block(inode, OCFS2_I(inode)->ip_blkno, |
Joel Becker | 0fcaa56 | 2008-10-09 17:20:31 -0700 | [diff] [blame] | 1160 | &di_bh); |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1161 | if (ret) { |
| 1162 | mlog_errno(ret); |
| 1163 | goto out; |
| 1164 | } |
| 1165 | |
| 1166 | /* |
| 1167 | * Nothing to do if the requested reservation range |
| 1168 | * fits within the inode. |
| 1169 | */ |
| 1170 | if (ocfs2_size_fits_inline_data(di_bh, end)) |
| 1171 | goto out; |
| 1172 | |
| 1173 | ret = ocfs2_convert_inline_data_to_extents(inode, di_bh); |
| 1174 | if (ret) { |
| 1175 | mlog_errno(ret); |
| 1176 | goto out; |
| 1177 | } |
| 1178 | } |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 1179 | |
| 1180 | /* |
| 1181 | * We consider both start and len to be inclusive. |
| 1182 | */ |
| 1183 | cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits; |
| 1184 | clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len); |
| 1185 | clusters -= cpos; |
| 1186 | |
| 1187 | while (clusters) { |
| 1188 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, |
| 1189 | &alloc_size, NULL); |
| 1190 | if (ret) { |
| 1191 | mlog_errno(ret); |
| 1192 | goto out; |
| 1193 | } |
| 1194 | |
| 1195 | /* |
| 1196 | * Hole or existing extent len can be arbitrary, so |
| 1197 | * cap it to our own allocation request. |
| 1198 | */ |
| 1199 | if (alloc_size > clusters) |
| 1200 | alloc_size = clusters; |
| 1201 | |
| 1202 | if (phys_cpos) { |
| 1203 | /* |
| 1204 | * We already have an allocation at this |
| 1205 | * region so we can safely skip it. |
| 1206 | */ |
| 1207 | goto next; |
| 1208 | } |
| 1209 | |
| 1210 | ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1); |
| 1211 | if (ret) { |
| 1212 | if (ret != -ENOSPC) |
| 1213 | mlog_errno(ret); |
| 1214 | goto out; |
| 1215 | } |
| 1216 | |
| 1217 | next: |
| 1218 | cpos += alloc_size; |
| 1219 | clusters -= alloc_size; |
| 1220 | } |
| 1221 | |
| 1222 | ret = 0; |
| 1223 | out: |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1224 | |
| 1225 | brelse(di_bh); |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 1226 | return ret; |
| 1227 | } |
| 1228 | |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1229 | static int __ocfs2_remove_inode_range(struct inode *inode, |
| 1230 | struct buffer_head *di_bh, |
| 1231 | u32 cpos, u32 phys_cpos, u32 len, |
| 1232 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
| 1233 | { |
| 1234 | int ret; |
| 1235 | u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos); |
| 1236 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1237 | struct inode *tl_inode = osb->osb_tl_inode; |
| 1238 | handle_t *handle; |
| 1239 | struct ocfs2_alloc_context *meta_ac = NULL; |
| 1240 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 1241 | struct ocfs2_extent_tree et; |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1242 | |
Joel Becker | 8d6220d | 2008-08-22 12:46:09 -0700 | [diff] [blame] | 1243 | ocfs2_init_dinode_extent_tree(&et, inode, di_bh); |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 1244 | |
| 1245 | ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac); |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1246 | if (ret) { |
| 1247 | mlog_errno(ret); |
| 1248 | return ret; |
| 1249 | } |
| 1250 | |
| 1251 | mutex_lock(&tl_inode->i_mutex); |
| 1252 | |
| 1253 | if (ocfs2_truncate_log_needs_flush(osb)) { |
| 1254 | ret = __ocfs2_flush_truncate_log(osb); |
| 1255 | if (ret < 0) { |
| 1256 | mlog_errno(ret); |
| 1257 | goto out; |
| 1258 | } |
| 1259 | } |
| 1260 | |
| 1261 | handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS); |
| 1262 | if (handle == NULL) { |
| 1263 | ret = -ENOMEM; |
| 1264 | mlog_errno(ret); |
| 1265 | goto out; |
| 1266 | } |
| 1267 | |
| 1268 | ret = ocfs2_journal_access(handle, inode, di_bh, |
| 1269 | OCFS2_JOURNAL_ACCESS_WRITE); |
| 1270 | if (ret) { |
| 1271 | mlog_errno(ret); |
| 1272 | goto out; |
| 1273 | } |
| 1274 | |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 1275 | ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac, |
| 1276 | dealloc); |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1277 | if (ret) { |
| 1278 | mlog_errno(ret); |
| 1279 | goto out_commit; |
| 1280 | } |
| 1281 | |
| 1282 | OCFS2_I(inode)->ip_clusters -= len; |
| 1283 | di->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters); |
| 1284 | |
| 1285 | ret = ocfs2_journal_dirty(handle, di_bh); |
| 1286 | if (ret) { |
| 1287 | mlog_errno(ret); |
| 1288 | goto out_commit; |
| 1289 | } |
| 1290 | |
| 1291 | ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len); |
| 1292 | if (ret) |
| 1293 | mlog_errno(ret); |
| 1294 | |
| 1295 | out_commit: |
| 1296 | ocfs2_commit_trans(osb, handle); |
| 1297 | out: |
| 1298 | mutex_unlock(&tl_inode->i_mutex); |
| 1299 | |
| 1300 | if (meta_ac) |
| 1301 | ocfs2_free_alloc_context(meta_ac); |
| 1302 | |
| 1303 | return ret; |
| 1304 | } |
| 1305 | |
| 1306 | /* |
| 1307 | * Truncate a byte range, avoiding pages within partial clusters. This |
| 1308 | * preserves those pages for the zeroing code to write to. |
| 1309 | */ |
| 1310 | static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start, |
| 1311 | u64 byte_len) |
| 1312 | { |
| 1313 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1314 | loff_t start, end; |
| 1315 | struct address_space *mapping = inode->i_mapping; |
| 1316 | |
| 1317 | start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start); |
| 1318 | end = byte_start + byte_len; |
| 1319 | end = end & ~(osb->s_clustersize - 1); |
| 1320 | |
| 1321 | if (start < end) { |
| 1322 | unmap_mapping_range(mapping, start, end - start, 0); |
| 1323 | truncate_inode_pages_range(mapping, start, end - 1); |
| 1324 | } |
| 1325 | } |
| 1326 | |
| 1327 | static int ocfs2_zero_partial_clusters(struct inode *inode, |
| 1328 | u64 start, u64 len) |
| 1329 | { |
| 1330 | int ret = 0; |
| 1331 | u64 tmpend, end = start + len; |
| 1332 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1333 | unsigned int csize = osb->s_clustersize; |
| 1334 | handle_t *handle; |
| 1335 | |
| 1336 | /* |
| 1337 | * The "start" and "end" values are NOT necessarily part of |
| 1338 | * the range whose allocation is being deleted. Rather, this |
| 1339 | * is what the user passed in with the request. We must zero |
| 1340 | * partial clusters here. There's no need to worry about |
| 1341 | * physical allocation - the zeroing code knows to skip holes. |
| 1342 | */ |
| 1343 | mlog(0, "byte start: %llu, end: %llu\n", |
| 1344 | (unsigned long long)start, (unsigned long long)end); |
| 1345 | |
| 1346 | /* |
| 1347 | * If both edges are on a cluster boundary then there's no |
| 1348 | * zeroing required as the region is part of the allocation to |
| 1349 | * be truncated. |
| 1350 | */ |
| 1351 | if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0) |
| 1352 | goto out; |
| 1353 | |
| 1354 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
| 1355 | if (handle == NULL) { |
| 1356 | ret = -ENOMEM; |
| 1357 | mlog_errno(ret); |
| 1358 | goto out; |
| 1359 | } |
| 1360 | |
| 1361 | /* |
| 1362 | * We want to get the byte offset of the end of the 1st cluster. |
| 1363 | */ |
| 1364 | tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1)); |
| 1365 | if (tmpend > end) |
| 1366 | tmpend = end; |
| 1367 | |
| 1368 | mlog(0, "1st range: start: %llu, tmpend: %llu\n", |
| 1369 | (unsigned long long)start, (unsigned long long)tmpend); |
| 1370 | |
| 1371 | ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend); |
| 1372 | if (ret) |
| 1373 | mlog_errno(ret); |
| 1374 | |
| 1375 | if (tmpend < end) { |
| 1376 | /* |
| 1377 | * This may make start and end equal, but the zeroing |
| 1378 | * code will skip any work in that case so there's no |
| 1379 | * need to catch it up here. |
| 1380 | */ |
| 1381 | start = end & ~(osb->s_clustersize - 1); |
| 1382 | |
| 1383 | mlog(0, "2nd range: start: %llu, end: %llu\n", |
| 1384 | (unsigned long long)start, (unsigned long long)end); |
| 1385 | |
| 1386 | ret = ocfs2_zero_range_for_truncate(inode, handle, start, end); |
| 1387 | if (ret) |
| 1388 | mlog_errno(ret); |
| 1389 | } |
| 1390 | |
| 1391 | ocfs2_commit_trans(osb, handle); |
| 1392 | out: |
| 1393 | return ret; |
| 1394 | } |
| 1395 | |
| 1396 | static int ocfs2_remove_inode_range(struct inode *inode, |
| 1397 | struct buffer_head *di_bh, u64 byte_start, |
| 1398 | u64 byte_len) |
| 1399 | { |
| 1400 | int ret = 0; |
| 1401 | u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size; |
| 1402 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1403 | struct ocfs2_cached_dealloc_ctxt dealloc; |
Mark Fasheh | b1967d0 | 2007-11-20 11:56:39 -0800 | [diff] [blame] | 1404 | struct address_space *mapping = inode->i_mapping; |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1405 | |
| 1406 | ocfs2_init_dealloc_ctxt(&dealloc); |
| 1407 | |
| 1408 | if (byte_len == 0) |
| 1409 | return 0; |
| 1410 | |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1411 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
| 1412 | ret = ocfs2_truncate_inline(inode, di_bh, byte_start, |
Mark Fasheh | b1967d0 | 2007-11-20 11:56:39 -0800 | [diff] [blame] | 1413 | byte_start + byte_len, 0); |
| 1414 | if (ret) { |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1415 | mlog_errno(ret); |
Mark Fasheh | b1967d0 | 2007-11-20 11:56:39 -0800 | [diff] [blame] | 1416 | goto out; |
| 1417 | } |
| 1418 | /* |
| 1419 | * There's no need to get fancy with the page cache |
| 1420 | * truncate of an inline-data inode. We're talking |
| 1421 | * about less than a page here, which will be cached |
| 1422 | * in the dinode buffer anyway. |
| 1423 | */ |
| 1424 | unmap_mapping_range(mapping, 0, 0, 0); |
| 1425 | truncate_inode_pages(mapping, 0); |
| 1426 | goto out; |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1427 | } |
| 1428 | |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1429 | trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start); |
| 1430 | trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits; |
| 1431 | if (trunc_len >= trunc_start) |
| 1432 | trunc_len -= trunc_start; |
| 1433 | else |
| 1434 | trunc_len = 0; |
| 1435 | |
| 1436 | mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n", |
| 1437 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
| 1438 | (unsigned long long)byte_start, |
| 1439 | (unsigned long long)byte_len, trunc_start, trunc_len); |
| 1440 | |
| 1441 | ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len); |
| 1442 | if (ret) { |
| 1443 | mlog_errno(ret); |
| 1444 | goto out; |
| 1445 | } |
| 1446 | |
| 1447 | cpos = trunc_start; |
| 1448 | while (trunc_len) { |
| 1449 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, |
| 1450 | &alloc_size, NULL); |
| 1451 | if (ret) { |
| 1452 | mlog_errno(ret); |
| 1453 | goto out; |
| 1454 | } |
| 1455 | |
| 1456 | if (alloc_size > trunc_len) |
| 1457 | alloc_size = trunc_len; |
| 1458 | |
| 1459 | /* Only do work for non-holes */ |
| 1460 | if (phys_cpos != 0) { |
| 1461 | ret = __ocfs2_remove_inode_range(inode, di_bh, cpos, |
| 1462 | phys_cpos, alloc_size, |
| 1463 | &dealloc); |
| 1464 | if (ret) { |
| 1465 | mlog_errno(ret); |
| 1466 | goto out; |
| 1467 | } |
| 1468 | } |
| 1469 | |
| 1470 | cpos += alloc_size; |
| 1471 | trunc_len -= alloc_size; |
| 1472 | } |
| 1473 | |
| 1474 | ocfs2_truncate_cluster_pages(inode, byte_start, byte_len); |
| 1475 | |
| 1476 | out: |
| 1477 | ocfs2_schedule_truncate_log_flush(osb, 1); |
| 1478 | ocfs2_run_deallocs(osb, &dealloc); |
| 1479 | |
| 1480 | return ret; |
| 1481 | } |
| 1482 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1483 | /* |
| 1484 | * Parts of this function taken from xfs_change_file_space() |
| 1485 | */ |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1486 | static int __ocfs2_change_file_space(struct file *file, struct inode *inode, |
| 1487 | loff_t f_pos, unsigned int cmd, |
| 1488 | struct ocfs2_space_resv *sr, |
| 1489 | int change_size) |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1490 | { |
| 1491 | int ret; |
| 1492 | s64 llen; |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1493 | loff_t size; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1494 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1495 | struct buffer_head *di_bh = NULL; |
| 1496 | handle_t *handle; |
Mark Fasheh | a00cce3 | 2007-07-20 11:28:30 -0700 | [diff] [blame] | 1497 | unsigned long long max_off = inode->i_sb->s_maxbytes; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1498 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1499 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) |
| 1500 | return -EROFS; |
| 1501 | |
| 1502 | mutex_lock(&inode->i_mutex); |
| 1503 | |
| 1504 | /* |
| 1505 | * This prevents concurrent writes on other nodes |
| 1506 | */ |
| 1507 | ret = ocfs2_rw_lock(inode, 1); |
| 1508 | if (ret) { |
| 1509 | mlog_errno(ret); |
| 1510 | goto out; |
| 1511 | } |
| 1512 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1513 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1514 | if (ret) { |
| 1515 | mlog_errno(ret); |
| 1516 | goto out_rw_unlock; |
| 1517 | } |
| 1518 | |
| 1519 | if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) { |
| 1520 | ret = -EPERM; |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1521 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1522 | } |
| 1523 | |
| 1524 | switch (sr->l_whence) { |
| 1525 | case 0: /*SEEK_SET*/ |
| 1526 | break; |
| 1527 | case 1: /*SEEK_CUR*/ |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1528 | sr->l_start += f_pos; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1529 | break; |
| 1530 | case 2: /*SEEK_END*/ |
| 1531 | sr->l_start += i_size_read(inode); |
| 1532 | break; |
| 1533 | default: |
| 1534 | ret = -EINVAL; |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1535 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1536 | } |
| 1537 | sr->l_whence = 0; |
| 1538 | |
| 1539 | llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len; |
| 1540 | |
| 1541 | if (sr->l_start < 0 |
| 1542 | || sr->l_start > max_off |
| 1543 | || (sr->l_start + llen) < 0 |
| 1544 | || (sr->l_start + llen) > max_off) { |
| 1545 | ret = -EINVAL; |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1546 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1547 | } |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1548 | size = sr->l_start + sr->l_len; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1549 | |
| 1550 | if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) { |
| 1551 | if (sr->l_len <= 0) { |
| 1552 | ret = -EINVAL; |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1553 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1554 | } |
| 1555 | } |
| 1556 | |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1557 | if (file && should_remove_suid(file->f_path.dentry)) { |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1558 | ret = __ocfs2_write_remove_suid(inode, di_bh); |
| 1559 | if (ret) { |
| 1560 | mlog_errno(ret); |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1561 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1562 | } |
| 1563 | } |
| 1564 | |
| 1565 | down_write(&OCFS2_I(inode)->ip_alloc_sem); |
| 1566 | switch (cmd) { |
| 1567 | case OCFS2_IOC_RESVSP: |
| 1568 | case OCFS2_IOC_RESVSP64: |
| 1569 | /* |
| 1570 | * This takes unsigned offsets, but the signed ones we |
| 1571 | * pass have been checked against overflow above. |
| 1572 | */ |
| 1573 | ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start, |
| 1574 | sr->l_len); |
| 1575 | break; |
| 1576 | case OCFS2_IOC_UNRESVSP: |
| 1577 | case OCFS2_IOC_UNRESVSP64: |
| 1578 | ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start, |
| 1579 | sr->l_len); |
| 1580 | break; |
| 1581 | default: |
| 1582 | ret = -EINVAL; |
| 1583 | } |
| 1584 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
| 1585 | if (ret) { |
| 1586 | mlog_errno(ret); |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1587 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1588 | } |
| 1589 | |
| 1590 | /* |
| 1591 | * We update c/mtime for these changes |
| 1592 | */ |
| 1593 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
| 1594 | if (IS_ERR(handle)) { |
| 1595 | ret = PTR_ERR(handle); |
| 1596 | mlog_errno(ret); |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1597 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1598 | } |
| 1599 | |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1600 | if (change_size && i_size_read(inode) < size) |
| 1601 | i_size_write(inode, size); |
| 1602 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1603 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
| 1604 | ret = ocfs2_mark_inode_dirty(handle, inode, di_bh); |
| 1605 | if (ret < 0) |
| 1606 | mlog_errno(ret); |
| 1607 | |
| 1608 | ocfs2_commit_trans(osb, handle); |
| 1609 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1610 | out_inode_unlock: |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1611 | brelse(di_bh); |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1612 | ocfs2_inode_unlock(inode, 1); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1613 | out_rw_unlock: |
| 1614 | ocfs2_rw_unlock(inode, 1); |
| 1615 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1616 | out: |
Julia Lawall | c259ae5 | 2008-07-21 09:59:15 +0200 | [diff] [blame] | 1617 | mutex_unlock(&inode->i_mutex); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1618 | return ret; |
| 1619 | } |
| 1620 | |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1621 | int ocfs2_change_file_space(struct file *file, unsigned int cmd, |
| 1622 | struct ocfs2_space_resv *sr) |
| 1623 | { |
| 1624 | struct inode *inode = file->f_path.dentry->d_inode; |
| 1625 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);; |
| 1626 | |
| 1627 | if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) && |
| 1628 | !ocfs2_writes_unwritten_extents(osb)) |
| 1629 | return -ENOTTY; |
| 1630 | else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) && |
| 1631 | !ocfs2_sparse_alloc(osb)) |
| 1632 | return -ENOTTY; |
| 1633 | |
| 1634 | if (!S_ISREG(inode->i_mode)) |
| 1635 | return -EINVAL; |
| 1636 | |
| 1637 | if (!(file->f_mode & FMODE_WRITE)) |
| 1638 | return -EBADF; |
| 1639 | |
| 1640 | return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0); |
| 1641 | } |
| 1642 | |
| 1643 | static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset, |
| 1644 | loff_t len) |
| 1645 | { |
| 1646 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1647 | struct ocfs2_space_resv sr; |
| 1648 | int change_size = 1; |
| 1649 | |
| 1650 | if (!ocfs2_writes_unwritten_extents(osb)) |
| 1651 | return -EOPNOTSUPP; |
| 1652 | |
| 1653 | if (S_ISDIR(inode->i_mode)) |
| 1654 | return -ENODEV; |
| 1655 | |
| 1656 | if (mode & FALLOC_FL_KEEP_SIZE) |
| 1657 | change_size = 0; |
| 1658 | |
| 1659 | sr.l_whence = 0; |
| 1660 | sr.l_start = (s64)offset; |
| 1661 | sr.l_len = (s64)len; |
| 1662 | |
| 1663 | return __ocfs2_change_file_space(NULL, inode, offset, |
| 1664 | OCFS2_IOC_RESVSP64, &sr, change_size); |
| 1665 | } |
| 1666 | |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1667 | static int ocfs2_prepare_inode_for_write(struct dentry *dentry, |
| 1668 | loff_t *ppos, |
| 1669 | size_t count, |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1670 | int appending, |
| 1671 | int *direct_io) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1672 | { |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1673 | int ret = 0, meta_level = 0; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1674 | struct inode *inode = dentry->d_inode; |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1675 | loff_t saved_pos, end; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1676 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1677 | /* |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1678 | * We start with a read level meta lock and only jump to an ex |
| 1679 | * if we need to make modifications here. |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1680 | */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1681 | for(;;) { |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1682 | ret = ocfs2_inode_lock(inode, NULL, meta_level); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1683 | if (ret < 0) { |
| 1684 | meta_level = -1; |
| 1685 | mlog_errno(ret); |
| 1686 | goto out; |
| 1687 | } |
| 1688 | |
| 1689 | /* Clear suid / sgid if necessary. We do this here |
| 1690 | * instead of later in the write path because |
| 1691 | * remove_suid() calls ->setattr without any hint that |
| 1692 | * we may have already done our cluster locking. Since |
| 1693 | * ocfs2_setattr() *must* take cluster locks to |
| 1694 | * proceeed, this will lead us to recursively lock the |
| 1695 | * inode. There's also the dinode i_size state which |
| 1696 | * can be lost via setattr during extending writes (we |
| 1697 | * set inode->i_size at the end of a write. */ |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1698 | if (should_remove_suid(dentry)) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1699 | if (meta_level == 0) { |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1700 | ocfs2_inode_unlock(inode, meta_level); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1701 | meta_level = 1; |
| 1702 | continue; |
| 1703 | } |
| 1704 | |
| 1705 | ret = ocfs2_write_remove_suid(inode); |
| 1706 | if (ret < 0) { |
| 1707 | mlog_errno(ret); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1708 | goto out_unlock; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1709 | } |
| 1710 | } |
| 1711 | |
| 1712 | /* work on a copy of ppos until we're sure that we won't have |
| 1713 | * to recalculate it due to relocking. */ |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1714 | if (appending) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1715 | saved_pos = i_size_read(inode); |
| 1716 | mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos); |
| 1717 | } else { |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1718 | saved_pos = *ppos; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1719 | } |
Mark Fasheh | 3a0782d | 2007-01-17 12:53:31 -0800 | [diff] [blame] | 1720 | |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1721 | end = saved_pos + count; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1722 | |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1723 | /* |
| 1724 | * Skip the O_DIRECT checks if we don't need |
| 1725 | * them. |
| 1726 | */ |
| 1727 | if (!direct_io || !(*direct_io)) |
| 1728 | break; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1729 | |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1730 | /* |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1731 | * There's no sane way to do direct writes to an inode |
| 1732 | * with inline data. |
| 1733 | */ |
| 1734 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
| 1735 | *direct_io = 0; |
| 1736 | break; |
| 1737 | } |
| 1738 | |
| 1739 | /* |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1740 | * Allowing concurrent direct writes means |
| 1741 | * i_size changes wouldn't be synchronized, so |
| 1742 | * one node could wind up truncating another |
| 1743 | * nodes writes. |
| 1744 | */ |
| 1745 | if (end > i_size_read(inode)) { |
| 1746 | *direct_io = 0; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1747 | break; |
| 1748 | } |
| 1749 | |
Mark Fasheh | 3a0782d | 2007-01-17 12:53:31 -0800 | [diff] [blame] | 1750 | /* |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1751 | * We don't fill holes during direct io, so |
| 1752 | * check for them here. If any are found, the |
| 1753 | * caller will have to retake some cluster |
| 1754 | * locks and initiate the io as buffered. |
Mark Fasheh | 3a0782d | 2007-01-17 12:53:31 -0800 | [diff] [blame] | 1755 | */ |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1756 | ret = ocfs2_check_range_for_holes(inode, saved_pos, count); |
| 1757 | if (ret == 1) { |
| 1758 | *direct_io = 0; |
| 1759 | ret = 0; |
| 1760 | } else if (ret < 0) |
| 1761 | mlog_errno(ret); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1762 | break; |
| 1763 | } |
| 1764 | |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1765 | if (appending) |
| 1766 | *ppos = saved_pos; |
| 1767 | |
| 1768 | out_unlock: |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1769 | ocfs2_inode_unlock(inode, meta_level); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1770 | |
| 1771 | out: |
| 1772 | return ret; |
| 1773 | } |
| 1774 | |
| 1775 | static ssize_t ocfs2_file_aio_write(struct kiocb *iocb, |
| 1776 | const struct iovec *iov, |
| 1777 | unsigned long nr_segs, |
| 1778 | loff_t pos) |
| 1779 | { |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1780 | int ret, direct_io, appending, rw_level, have_alloc_sem = 0; |
Nick Piggin | b6af1bc | 2007-10-16 01:25:24 -0700 | [diff] [blame] | 1781 | int can_do_direct; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1782 | ssize_t written = 0; |
| 1783 | size_t ocount; /* original count */ |
| 1784 | size_t count; /* after file limit checks */ |
Mark Fasheh | 9ea2d32 | 2007-10-18 14:14:45 -0700 | [diff] [blame] | 1785 | loff_t old_size, *ppos = &iocb->ki_pos; |
| 1786 | u32 old_clusters; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1787 | struct file *file = iocb->ki_filp; |
| 1788 | struct inode *inode = file->f_path.dentry->d_inode; |
Mark Fasheh | 9ea2d32 | 2007-10-18 14:14:45 -0700 | [diff] [blame] | 1789 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1790 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1791 | mlog_entry("(0x%p, %u, '%.*s')\n", file, |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1792 | (unsigned int)nr_segs, |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1793 | file->f_path.dentry->d_name.len, |
| 1794 | file->f_path.dentry->d_name.name); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1795 | |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1796 | if (iocb->ki_left == 0) |
| 1797 | return 0; |
| 1798 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1799 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); |
| 1800 | |
| 1801 | appending = file->f_flags & O_APPEND ? 1 : 0; |
| 1802 | direct_io = file->f_flags & O_DIRECT ? 1 : 0; |
| 1803 | |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1804 | mutex_lock(&inode->i_mutex); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1805 | |
| 1806 | relock: |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1807 | /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */ |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1808 | if (direct_io) { |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1809 | down_read(&inode->i_alloc_sem); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1810 | have_alloc_sem = 1; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1811 | } |
| 1812 | |
| 1813 | /* concurrent O_DIRECT writes are allowed */ |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1814 | rw_level = !direct_io; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1815 | ret = ocfs2_rw_lock(inode, rw_level); |
| 1816 | if (ret < 0) { |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1817 | mlog_errno(ret); |
| 1818 | goto out_sems; |
| 1819 | } |
| 1820 | |
| 1821 | can_do_direct = direct_io; |
| 1822 | ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos, |
| 1823 | iocb->ki_left, appending, |
| 1824 | &can_do_direct); |
| 1825 | if (ret < 0) { |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1826 | mlog_errno(ret); |
| 1827 | goto out; |
| 1828 | } |
| 1829 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1830 | /* |
| 1831 | * We can't complete the direct I/O as requested, fall back to |
| 1832 | * buffered I/O. |
| 1833 | */ |
| 1834 | if (direct_io && !can_do_direct) { |
| 1835 | ocfs2_rw_unlock(inode, rw_level); |
| 1836 | up_read(&inode->i_alloc_sem); |
| 1837 | |
| 1838 | have_alloc_sem = 0; |
| 1839 | rw_level = -1; |
| 1840 | |
| 1841 | direct_io = 0; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1842 | goto relock; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1843 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1844 | |
Mark Fasheh | 9ea2d32 | 2007-10-18 14:14:45 -0700 | [diff] [blame] | 1845 | /* |
| 1846 | * To later detect whether a journal commit for sync writes is |
| 1847 | * necessary, we sample i_size, and cluster count here. |
| 1848 | */ |
| 1849 | old_size = i_size_read(inode); |
| 1850 | old_clusters = OCFS2_I(inode)->ip_clusters; |
| 1851 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1852 | /* communicate with ocfs2_dio_end_io */ |
Mark Fasheh | 7cdfc3a | 2007-04-16 17:28:51 -0700 | [diff] [blame] | 1853 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1854 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1855 | if (direct_io) { |
Nick Piggin | b6af1bc | 2007-10-16 01:25:24 -0700 | [diff] [blame] | 1856 | ret = generic_segment_checks(iov, &nr_segs, &ocount, |
| 1857 | VERIFY_READ); |
| 1858 | if (ret) |
| 1859 | goto out_dio; |
| 1860 | |
| 1861 | ret = generic_write_checks(file, ppos, &count, |
| 1862 | S_ISBLK(inode->i_mode)); |
| 1863 | if (ret) |
| 1864 | goto out_dio; |
| 1865 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1866 | written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos, |
| 1867 | ppos, count, ocount); |
| 1868 | if (written < 0) { |
| 1869 | ret = written; |
| 1870 | goto out_dio; |
| 1871 | } |
| 1872 | } else { |
Nick Piggin | b6af1bc | 2007-10-16 01:25:24 -0700 | [diff] [blame] | 1873 | written = generic_file_aio_write_nolock(iocb, iov, nr_segs, |
| 1874 | *ppos); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1875 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1876 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1877 | out_dio: |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1878 | /* buffered aio wouldn't have proper lock coverage today */ |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1879 | BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1880 | |
Mark Fasheh | 9ea2d32 | 2007-10-18 14:14:45 -0700 | [diff] [blame] | 1881 | if ((file->f_flags & O_SYNC && !direct_io) || IS_SYNC(inode)) { |
| 1882 | /* |
| 1883 | * The generic write paths have handled getting data |
| 1884 | * to disk, but since we don't make use of the dirty |
| 1885 | * inode list, a manual journal commit is necessary |
| 1886 | * here. |
| 1887 | */ |
| 1888 | if (old_size != i_size_read(inode) || |
| 1889 | old_clusters != OCFS2_I(inode)->ip_clusters) { |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 1890 | ret = jbd2_journal_force_commit(osb->journal->j_journal); |
Mark Fasheh | 9ea2d32 | 2007-10-18 14:14:45 -0700 | [diff] [blame] | 1891 | if (ret < 0) |
| 1892 | written = ret; |
| 1893 | } |
| 1894 | } |
| 1895 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1896 | /* |
| 1897 | * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io |
| 1898 | * function pointer which is called when o_direct io completes so that |
| 1899 | * it can unlock our rw lock. (it's the clustered equivalent of |
| 1900 | * i_alloc_sem; protects truncate from racing with pending ios). |
| 1901 | * Unfortunately there are error cases which call end_io and others |
| 1902 | * that don't. so we don't have to unlock the rw_lock if either an |
| 1903 | * async dio is going to do it in the future or an end_io after an |
| 1904 | * error has already done it. |
| 1905 | */ |
| 1906 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { |
| 1907 | rw_level = -1; |
| 1908 | have_alloc_sem = 0; |
| 1909 | } |
| 1910 | |
| 1911 | out: |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1912 | if (rw_level != -1) |
| 1913 | ocfs2_rw_unlock(inode, rw_level); |
| 1914 | |
| 1915 | out_sems: |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1916 | if (have_alloc_sem) |
| 1917 | up_read(&inode->i_alloc_sem); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1918 | |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 1919 | mutex_unlock(&inode->i_mutex); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1920 | |
| 1921 | mlog_exit(ret); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1922 | return written ? written : ret; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1923 | } |
| 1924 | |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1925 | static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe, |
| 1926 | struct file *out, |
| 1927 | loff_t *ppos, |
| 1928 | size_t len, |
| 1929 | unsigned int flags) |
| 1930 | { |
| 1931 | int ret; |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 1932 | struct inode *inode = out->f_path.dentry->d_inode; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1933 | |
| 1934 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe, |
| 1935 | (unsigned int)len, |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 1936 | out->f_path.dentry->d_name.len, |
| 1937 | out->f_path.dentry->d_name.name); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1938 | |
| 1939 | inode_double_lock(inode, pipe->inode); |
| 1940 | |
| 1941 | ret = ocfs2_rw_lock(inode, 1); |
| 1942 | if (ret < 0) { |
| 1943 | mlog_errno(ret); |
| 1944 | goto out; |
| 1945 | } |
| 1946 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1947 | ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0, |
| 1948 | NULL); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1949 | if (ret < 0) { |
| 1950 | mlog_errno(ret); |
| 1951 | goto out_unlock; |
| 1952 | } |
| 1953 | |
Nick Piggin | b6af1bc | 2007-10-16 01:25:24 -0700 | [diff] [blame] | 1954 | ret = generic_file_splice_write_nolock(pipe, out, ppos, len, flags); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1955 | |
| 1956 | out_unlock: |
| 1957 | ocfs2_rw_unlock(inode, 1); |
| 1958 | out: |
| 1959 | inode_double_unlock(inode, pipe->inode); |
| 1960 | |
| 1961 | mlog_exit(ret); |
| 1962 | return ret; |
| 1963 | } |
| 1964 | |
| 1965 | static ssize_t ocfs2_file_splice_read(struct file *in, |
| 1966 | loff_t *ppos, |
| 1967 | struct pipe_inode_info *pipe, |
| 1968 | size_t len, |
| 1969 | unsigned int flags) |
| 1970 | { |
| 1971 | int ret = 0; |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 1972 | struct inode *inode = in->f_path.dentry->d_inode; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1973 | |
| 1974 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe, |
| 1975 | (unsigned int)len, |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 1976 | in->f_path.dentry->d_name.len, |
| 1977 | in->f_path.dentry->d_name.name); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1978 | |
| 1979 | /* |
| 1980 | * See the comment in ocfs2_file_aio_read() |
| 1981 | */ |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1982 | ret = ocfs2_inode_lock(inode, NULL, 0); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1983 | if (ret < 0) { |
| 1984 | mlog_errno(ret); |
| 1985 | goto bail; |
| 1986 | } |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1987 | ocfs2_inode_unlock(inode, 0); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1988 | |
| 1989 | ret = generic_file_splice_read(in, ppos, pipe, len, flags); |
| 1990 | |
| 1991 | bail: |
| 1992 | mlog_exit(ret); |
| 1993 | return ret; |
| 1994 | } |
| 1995 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1996 | static ssize_t ocfs2_file_aio_read(struct kiocb *iocb, |
Badari Pulavarty | 027445c | 2006-09-30 23:28:46 -0700 | [diff] [blame] | 1997 | const struct iovec *iov, |
| 1998 | unsigned long nr_segs, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1999 | loff_t pos) |
| 2000 | { |
Tiger Yang | 25899de | 2006-11-15 15:49:02 +0800 | [diff] [blame] | 2001 | 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] | 2002 | struct file *filp = iocb->ki_filp; |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 2003 | struct inode *inode = filp->f_path.dentry->d_inode; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2004 | |
Badari Pulavarty | 027445c | 2006-09-30 23:28:46 -0700 | [diff] [blame] | 2005 | mlog_entry("(0x%p, %u, '%.*s')\n", filp, |
| 2006 | (unsigned int)nr_segs, |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 2007 | filp->f_path.dentry->d_name.len, |
| 2008 | filp->f_path.dentry->d_name.name); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2009 | |
| 2010 | if (!inode) { |
| 2011 | ret = -EINVAL; |
| 2012 | mlog_errno(ret); |
| 2013 | goto bail; |
| 2014 | } |
| 2015 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2016 | /* |
| 2017 | * buffered reads protect themselves in ->readpage(). O_DIRECT reads |
| 2018 | * need locks to protect pending reads from racing with truncate. |
| 2019 | */ |
| 2020 | if (filp->f_flags & O_DIRECT) { |
| 2021 | down_read(&inode->i_alloc_sem); |
| 2022 | have_alloc_sem = 1; |
| 2023 | |
| 2024 | ret = ocfs2_rw_lock(inode, 0); |
| 2025 | if (ret < 0) { |
| 2026 | mlog_errno(ret); |
| 2027 | goto bail; |
| 2028 | } |
| 2029 | rw_level = 0; |
| 2030 | /* communicate with ocfs2_dio_end_io */ |
Mark Fasheh | 7cdfc3a | 2007-04-16 17:28:51 -0700 | [diff] [blame] | 2031 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2032 | } |
| 2033 | |
Mark Fasheh | c4374f8 | 2006-05-05 19:04:35 -0700 | [diff] [blame] | 2034 | /* |
| 2035 | * We're fine letting folks race truncates and extending |
| 2036 | * writes with read across the cluster, just like they can |
| 2037 | * locally. Hence no rw_lock during read. |
| 2038 | * |
| 2039 | * Take and drop the meta data lock to update inode fields |
| 2040 | * like i_size. This allows the checks down below |
| 2041 | * generic_file_aio_read() a chance of actually working. |
| 2042 | */ |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 2043 | ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level); |
Mark Fasheh | c4374f8 | 2006-05-05 19:04:35 -0700 | [diff] [blame] | 2044 | if (ret < 0) { |
| 2045 | mlog_errno(ret); |
| 2046 | goto bail; |
| 2047 | } |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 2048 | ocfs2_inode_unlock(inode, lock_level); |
Mark Fasheh | c4374f8 | 2006-05-05 19:04:35 -0700 | [diff] [blame] | 2049 | |
Badari Pulavarty | 027445c | 2006-09-30 23:28:46 -0700 | [diff] [blame] | 2050 | ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2051 | if (ret == -EINVAL) |
Sunil Mushran | 56753bd | 2008-06-09 11:24:41 -0700 | [diff] [blame] | 2052 | mlog(0, "generic_file_aio_read returned -EINVAL\n"); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2053 | |
| 2054 | /* buffered aio wouldn't have proper lock coverage today */ |
| 2055 | BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT)); |
| 2056 | |
| 2057 | /* see ocfs2_file_aio_write */ |
| 2058 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { |
| 2059 | rw_level = -1; |
| 2060 | have_alloc_sem = 0; |
| 2061 | } |
| 2062 | |
| 2063 | bail: |
| 2064 | if (have_alloc_sem) |
| 2065 | up_read(&inode->i_alloc_sem); |
| 2066 | if (rw_level != -1) |
| 2067 | ocfs2_rw_unlock(inode, rw_level); |
| 2068 | mlog_exit(ret); |
| 2069 | |
| 2070 | return ret; |
| 2071 | } |
| 2072 | |
Arjan van de Ven | 92e1d5b | 2007-02-12 00:55:39 -0800 | [diff] [blame] | 2073 | const struct inode_operations ocfs2_file_iops = { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2074 | .setattr = ocfs2_setattr, |
| 2075 | .getattr = ocfs2_getattr, |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 2076 | .permission = ocfs2_permission, |
Tiger Yang | cf1d6c7 | 2008-08-18 17:11:00 +0800 | [diff] [blame] | 2077 | .setxattr = generic_setxattr, |
| 2078 | .getxattr = generic_getxattr, |
| 2079 | .listxattr = ocfs2_listxattr, |
| 2080 | .removexattr = generic_removexattr, |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 2081 | .fallocate = ocfs2_fallocate, |
Mark Fasheh | 00dc417 | 2008-10-03 17:32:11 -0400 | [diff] [blame] | 2082 | .fiemap = ocfs2_fiemap, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2083 | }; |
| 2084 | |
Arjan van de Ven | 92e1d5b | 2007-02-12 00:55:39 -0800 | [diff] [blame] | 2085 | const struct inode_operations ocfs2_special_file_iops = { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2086 | .setattr = ocfs2_setattr, |
| 2087 | .getattr = ocfs2_getattr, |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 2088 | .permission = ocfs2_permission, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2089 | }; |
| 2090 | |
Mark Fasheh | 53da493 | 2008-07-21 14:29:16 -0700 | [diff] [blame] | 2091 | /* |
| 2092 | * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with |
| 2093 | * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks! |
| 2094 | */ |
Arjan van de Ven | 4b6f5d2 | 2006-03-28 01:56:42 -0800 | [diff] [blame] | 2095 | const struct file_operations ocfs2_fops = { |
Jan Kara | 32c3c0e | 2007-12-19 15:24:52 +0100 | [diff] [blame] | 2096 | .llseek = generic_file_llseek, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2097 | .read = do_sync_read, |
| 2098 | .write = do_sync_write, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2099 | .mmap = ocfs2_mmap, |
| 2100 | .fsync = ocfs2_sync_file, |
| 2101 | .release = ocfs2_file_release, |
| 2102 | .open = ocfs2_file_open, |
| 2103 | .aio_read = ocfs2_file_aio_read, |
| 2104 | .aio_write = ocfs2_file_aio_write, |
Andi Kleen | c9ec148 | 2008-01-27 03:17:17 +0100 | [diff] [blame] | 2105 | .unlocked_ioctl = ocfs2_ioctl, |
Mark Fasheh | 586d232 | 2007-03-09 15:56:28 -0800 | [diff] [blame] | 2106 | #ifdef CONFIG_COMPAT |
| 2107 | .compat_ioctl = ocfs2_compat_ioctl, |
| 2108 | #endif |
Mark Fasheh | 53da493 | 2008-07-21 14:29:16 -0700 | [diff] [blame] | 2109 | .lock = ocfs2_lock, |
| 2110 | .flock = ocfs2_flock, |
| 2111 | .splice_read = ocfs2_file_splice_read, |
| 2112 | .splice_write = ocfs2_file_splice_write, |
| 2113 | }; |
| 2114 | |
| 2115 | const struct file_operations ocfs2_dops = { |
| 2116 | .llseek = generic_file_llseek, |
| 2117 | .read = generic_read_dir, |
| 2118 | .readdir = ocfs2_readdir, |
| 2119 | .fsync = ocfs2_sync_file, |
| 2120 | .release = ocfs2_dir_release, |
| 2121 | .open = ocfs2_dir_open, |
| 2122 | .unlocked_ioctl = ocfs2_ioctl, |
| 2123 | #ifdef CONFIG_COMPAT |
| 2124 | .compat_ioctl = ocfs2_compat_ioctl, |
| 2125 | #endif |
| 2126 | .lock = ocfs2_lock, |
| 2127 | .flock = ocfs2_flock, |
| 2128 | }; |
| 2129 | |
| 2130 | /* |
| 2131 | * POSIX-lockless variants of our file_operations. |
| 2132 | * |
| 2133 | * These will be used if the underlying cluster stack does not support |
| 2134 | * posix file locking, if the user passes the "localflocks" mount |
| 2135 | * option, or if we have a local-only fs. |
| 2136 | * |
| 2137 | * ocfs2_flock is in here because all stacks handle UNIX file locks, |
| 2138 | * so we still want it in the case of no stack support for |
| 2139 | * plocks. Internally, it will do the right thing when asked to ignore |
| 2140 | * the cluster. |
| 2141 | */ |
| 2142 | const struct file_operations ocfs2_fops_no_plocks = { |
| 2143 | .llseek = generic_file_llseek, |
| 2144 | .read = do_sync_read, |
| 2145 | .write = do_sync_write, |
| 2146 | .mmap = ocfs2_mmap, |
| 2147 | .fsync = ocfs2_sync_file, |
| 2148 | .release = ocfs2_file_release, |
| 2149 | .open = ocfs2_file_open, |
| 2150 | .aio_read = ocfs2_file_aio_read, |
| 2151 | .aio_write = ocfs2_file_aio_write, |
| 2152 | .unlocked_ioctl = ocfs2_ioctl, |
| 2153 | #ifdef CONFIG_COMPAT |
| 2154 | .compat_ioctl = ocfs2_compat_ioctl, |
| 2155 | #endif |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 2156 | .flock = ocfs2_flock, |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 2157 | .splice_read = ocfs2_file_splice_read, |
| 2158 | .splice_write = ocfs2_file_splice_write, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2159 | }; |
| 2160 | |
Mark Fasheh | 53da493 | 2008-07-21 14:29:16 -0700 | [diff] [blame] | 2161 | const struct file_operations ocfs2_dops_no_plocks = { |
Jan Kara | 32c3c0e | 2007-12-19 15:24:52 +0100 | [diff] [blame] | 2162 | .llseek = generic_file_llseek, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2163 | .read = generic_read_dir, |
| 2164 | .readdir = ocfs2_readdir, |
| 2165 | .fsync = ocfs2_sync_file, |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 2166 | .release = ocfs2_dir_release, |
| 2167 | .open = ocfs2_dir_open, |
Andi Kleen | c9ec148 | 2008-01-27 03:17:17 +0100 | [diff] [blame] | 2168 | .unlocked_ioctl = ocfs2_ioctl, |
Mark Fasheh | 586d232 | 2007-03-09 15:56:28 -0800 | [diff] [blame] | 2169 | #ifdef CONFIG_COMPAT |
| 2170 | .compat_ioctl = ocfs2_compat_ioctl, |
| 2171 | #endif |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 2172 | .flock = ocfs2_flock, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2173 | }; |