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> |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 38 | #include <linux/quotaops.h> |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 39 | |
| 40 | #define MLOG_MASK_PREFIX ML_INODE |
| 41 | #include <cluster/masklog.h> |
| 42 | |
| 43 | #include "ocfs2.h" |
| 44 | |
| 45 | #include "alloc.h" |
| 46 | #include "aops.h" |
| 47 | #include "dir.h" |
| 48 | #include "dlmglue.h" |
| 49 | #include "extent_map.h" |
| 50 | #include "file.h" |
| 51 | #include "sysfile.h" |
| 52 | #include "inode.h" |
Herbert Poetzl | ca4d147 | 2006-07-03 17:27:12 -0700 | [diff] [blame] | 53 | #include "ioctl.h" |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 54 | #include "journal.h" |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 55 | #include "locks.h" |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 56 | #include "mmap.h" |
| 57 | #include "suballoc.h" |
| 58 | #include "super.h" |
Tiger Yang | cf1d6c7 | 2008-08-18 17:11:00 +0800 | [diff] [blame] | 59 | #include "xattr.h" |
Tiger Yang | 23fc270 | 2008-11-14 11:17:18 +0800 | [diff] [blame] | 60 | #include "acl.h" |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 61 | #include "quota.h" |
Tao Ma | 293b2f7 | 2009-08-25 08:02:48 +0800 | [diff] [blame^] | 62 | #include "refcounttree.h" |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 63 | |
| 64 | #include "buffer_head_io.h" |
| 65 | |
| 66 | static int ocfs2_sync_inode(struct inode *inode) |
| 67 | { |
| 68 | filemap_fdatawrite(inode->i_mapping); |
| 69 | return sync_mapping_buffers(inode->i_mapping); |
| 70 | } |
| 71 | |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 72 | static int ocfs2_init_file_private(struct inode *inode, struct file *file) |
| 73 | { |
| 74 | struct ocfs2_file_private *fp; |
| 75 | |
| 76 | fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL); |
| 77 | if (!fp) |
| 78 | return -ENOMEM; |
| 79 | |
| 80 | fp->fp_file = file; |
| 81 | mutex_init(&fp->fp_mutex); |
| 82 | ocfs2_file_lock_res_init(&fp->fp_flock, fp); |
| 83 | file->private_data = fp; |
| 84 | |
| 85 | return 0; |
| 86 | } |
| 87 | |
| 88 | static void ocfs2_free_file_private(struct inode *inode, struct file *file) |
| 89 | { |
| 90 | struct ocfs2_file_private *fp = file->private_data; |
| 91 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 92 | |
| 93 | if (fp) { |
| 94 | ocfs2_simple_drop_lockres(osb, &fp->fp_flock); |
| 95 | ocfs2_lock_res_free(&fp->fp_flock); |
| 96 | kfree(fp); |
| 97 | file->private_data = NULL; |
| 98 | } |
| 99 | } |
| 100 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 101 | static int ocfs2_file_open(struct inode *inode, struct file *file) |
| 102 | { |
| 103 | int status; |
| 104 | int mode = file->f_flags; |
| 105 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 106 | |
| 107 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 108 | 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] | 109 | |
| 110 | spin_lock(&oi->ip_lock); |
| 111 | |
| 112 | /* Check that the inode hasn't been wiped from disk by another |
| 113 | * node. If it hasn't then we're safe as long as we hold the |
| 114 | * spin lock until our increment of open count. */ |
| 115 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { |
| 116 | spin_unlock(&oi->ip_lock); |
| 117 | |
| 118 | status = -ENOENT; |
| 119 | goto leave; |
| 120 | } |
| 121 | |
| 122 | if (mode & O_DIRECT) |
| 123 | oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT; |
| 124 | |
| 125 | oi->ip_open_count++; |
| 126 | spin_unlock(&oi->ip_lock); |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 127 | |
| 128 | status = ocfs2_init_file_private(inode, file); |
| 129 | if (status) { |
| 130 | /* |
| 131 | * We want to set open count back if we're failing the |
| 132 | * open. |
| 133 | */ |
| 134 | spin_lock(&oi->ip_lock); |
| 135 | oi->ip_open_count--; |
| 136 | spin_unlock(&oi->ip_lock); |
| 137 | } |
| 138 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 139 | leave: |
| 140 | mlog_exit(status); |
| 141 | return status; |
| 142 | } |
| 143 | |
| 144 | static int ocfs2_file_release(struct inode *inode, struct file *file) |
| 145 | { |
| 146 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 147 | |
| 148 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 149 | file->f_path.dentry->d_name.len, |
| 150 | file->f_path.dentry->d_name.name); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 151 | |
| 152 | spin_lock(&oi->ip_lock); |
| 153 | if (!--oi->ip_open_count) |
| 154 | oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT; |
| 155 | spin_unlock(&oi->ip_lock); |
| 156 | |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 157 | ocfs2_free_file_private(inode, file); |
| 158 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 159 | mlog_exit(0); |
| 160 | |
| 161 | return 0; |
| 162 | } |
| 163 | |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 164 | static int ocfs2_dir_open(struct inode *inode, struct file *file) |
| 165 | { |
| 166 | return ocfs2_init_file_private(inode, file); |
| 167 | } |
| 168 | |
| 169 | static int ocfs2_dir_release(struct inode *inode, struct file *file) |
| 170 | { |
| 171 | ocfs2_free_file_private(inode, file); |
| 172 | return 0; |
| 173 | } |
| 174 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 175 | static int ocfs2_sync_file(struct file *file, |
| 176 | struct dentry *dentry, |
| 177 | int datasync) |
| 178 | { |
| 179 | int err = 0; |
| 180 | journal_t *journal; |
| 181 | struct inode *inode = dentry->d_inode; |
| 182 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 183 | |
| 184 | mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync, |
| 185 | dentry->d_name.len, dentry->d_name.name); |
| 186 | |
| 187 | err = ocfs2_sync_inode(dentry->d_inode); |
| 188 | if (err) |
| 189 | goto bail; |
| 190 | |
Hisashi Hifumi | e04cc15 | 2009-06-09 16:47:45 +0900 | [diff] [blame] | 191 | if (datasync && !(inode->i_state & I_DIRTY_DATASYNC)) |
| 192 | goto bail; |
| 193 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 194 | journal = osb->journal->j_journal; |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 195 | err = jbd2_journal_force_commit(journal); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 196 | |
| 197 | bail: |
| 198 | mlog_exit(err); |
| 199 | |
| 200 | return (err < 0) ? -EIO : 0; |
| 201 | } |
| 202 | |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 203 | int ocfs2_should_update_atime(struct inode *inode, |
| 204 | struct vfsmount *vfsmnt) |
| 205 | { |
| 206 | struct timespec now; |
| 207 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 208 | |
| 209 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) |
| 210 | return 0; |
| 211 | |
| 212 | if ((inode->i_flags & S_NOATIME) || |
| 213 | ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))) |
| 214 | return 0; |
| 215 | |
Mark Fasheh | 6c2aad0 | 2006-12-19 15:25:52 -0800 | [diff] [blame] | 216 | /* |
| 217 | * We can be called with no vfsmnt structure - NFSD will |
| 218 | * sometimes do this. |
| 219 | * |
| 220 | * Note that our action here is different than touch_atime() - |
| 221 | * if we can't tell whether this is a noatime mount, then we |
| 222 | * don't know whether to trust the value of s_atime_quantum. |
| 223 | */ |
| 224 | if (vfsmnt == NULL) |
| 225 | return 0; |
| 226 | |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 227 | if ((vfsmnt->mnt_flags & MNT_NOATIME) || |
| 228 | ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) |
| 229 | return 0; |
| 230 | |
Mark Fasheh | 7e913c5 | 2006-12-13 00:34:35 -0800 | [diff] [blame] | 231 | if (vfsmnt->mnt_flags & MNT_RELATIME) { |
| 232 | if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) || |
| 233 | (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0)) |
| 234 | return 1; |
| 235 | |
| 236 | return 0; |
| 237 | } |
| 238 | |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 239 | now = CURRENT_TIME; |
| 240 | if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum)) |
| 241 | return 0; |
| 242 | else |
| 243 | return 1; |
| 244 | } |
| 245 | |
| 246 | int ocfs2_update_inode_atime(struct inode *inode, |
| 247 | struct buffer_head *bh) |
| 248 | { |
| 249 | int ret; |
| 250 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 251 | handle_t *handle; |
Mark Fasheh | c11e9fa | 2007-07-20 11:24:53 -0700 | [diff] [blame] | 252 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data; |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 253 | |
| 254 | mlog_entry_void(); |
| 255 | |
| 256 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
Jan Kara | fa38e92 | 2008-10-20 19:23:51 +0200 | [diff] [blame] | 257 | if (IS_ERR(handle)) { |
| 258 | ret = PTR_ERR(handle); |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 259 | mlog_errno(ret); |
| 260 | goto out; |
| 261 | } |
| 262 | |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 263 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh, |
Joel Becker | 13723d0 | 2008-10-17 19:25:01 -0700 | [diff] [blame] | 264 | OCFS2_JOURNAL_ACCESS_WRITE); |
Mark Fasheh | c11e9fa | 2007-07-20 11:24:53 -0700 | [diff] [blame] | 265 | if (ret) { |
| 266 | mlog_errno(ret); |
| 267 | goto out_commit; |
| 268 | } |
| 269 | |
| 270 | /* |
| 271 | * Don't use ocfs2_mark_inode_dirty() here as we don't always |
| 272 | * have i_mutex to guard against concurrent changes to other |
| 273 | * inode fields. |
| 274 | */ |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 275 | inode->i_atime = CURRENT_TIME; |
Mark Fasheh | c11e9fa | 2007-07-20 11:24:53 -0700 | [diff] [blame] | 276 | di->i_atime = cpu_to_le64(inode->i_atime.tv_sec); |
| 277 | di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec); |
| 278 | |
| 279 | ret = ocfs2_journal_dirty(handle, bh); |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 280 | if (ret < 0) |
| 281 | mlog_errno(ret); |
| 282 | |
Mark Fasheh | c11e9fa | 2007-07-20 11:24:53 -0700 | [diff] [blame] | 283 | out_commit: |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 284 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); |
| 285 | out: |
| 286 | mlog_exit(ret); |
| 287 | return ret; |
| 288 | } |
| 289 | |
Adrian Bunk | 6cb129f | 2007-04-26 00:29:35 -0700 | [diff] [blame] | 290 | static int ocfs2_set_inode_size(handle_t *handle, |
| 291 | struct inode *inode, |
| 292 | struct buffer_head *fe_bh, |
| 293 | u64 new_i_size) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 294 | { |
| 295 | int status; |
| 296 | |
| 297 | mlog_entry_void(); |
| 298 | i_size_write(inode, new_i_size); |
Mark Fasheh | 8110b07 | 2007-03-22 16:53:23 -0700 | [diff] [blame] | 299 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 300 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
| 301 | |
| 302 | status = ocfs2_mark_inode_dirty(handle, inode, fe_bh); |
| 303 | if (status < 0) { |
| 304 | mlog_errno(status); |
| 305 | goto bail; |
| 306 | } |
| 307 | |
| 308 | bail: |
| 309 | mlog_exit(status); |
| 310 | return status; |
| 311 | } |
| 312 | |
Jan Kara | 9e33d69 | 2008-08-25 19:56:50 +0200 | [diff] [blame] | 313 | int ocfs2_simple_size_update(struct inode *inode, |
| 314 | struct buffer_head *di_bh, |
| 315 | u64 new_i_size) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 316 | { |
| 317 | int ret; |
| 318 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 319 | handle_t *handle = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 320 | |
Mark Fasheh | 65eff9c | 2006-10-09 17:26:22 -0700 | [diff] [blame] | 321 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
Jan Kara | fa38e92 | 2008-10-20 19:23:51 +0200 | [diff] [blame] | 322 | if (IS_ERR(handle)) { |
| 323 | ret = PTR_ERR(handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 324 | mlog_errno(ret); |
| 325 | goto out; |
| 326 | } |
| 327 | |
| 328 | ret = ocfs2_set_inode_size(handle, inode, di_bh, |
| 329 | new_i_size); |
| 330 | if (ret < 0) |
| 331 | mlog_errno(ret); |
| 332 | |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 333 | ocfs2_commit_trans(osb, handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 334 | out: |
| 335 | return ret; |
| 336 | } |
| 337 | |
| 338 | static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb, |
| 339 | struct inode *inode, |
| 340 | struct buffer_head *fe_bh, |
| 341 | u64 new_i_size) |
| 342 | { |
| 343 | int status; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 344 | handle_t *handle; |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 345 | struct ocfs2_dinode *di; |
Mark Fasheh | 35edec1 | 2007-07-06 14:41:18 -0700 | [diff] [blame] | 346 | u64 cluster_bytes; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 347 | |
| 348 | mlog_entry_void(); |
| 349 | |
| 350 | /* TODO: This needs to actually orphan the inode in this |
| 351 | * transaction. */ |
| 352 | |
Mark Fasheh | 65eff9c | 2006-10-09 17:26:22 -0700 | [diff] [blame] | 353 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 354 | if (IS_ERR(handle)) { |
| 355 | status = PTR_ERR(handle); |
| 356 | mlog_errno(status); |
| 357 | goto out; |
| 358 | } |
| 359 | |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 360 | status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh, |
Joel Becker | 13723d0 | 2008-10-17 19:25:01 -0700 | [diff] [blame] | 361 | OCFS2_JOURNAL_ACCESS_WRITE); |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 362 | if (status < 0) { |
| 363 | mlog_errno(status); |
| 364 | goto out_commit; |
| 365 | } |
| 366 | |
| 367 | /* |
| 368 | * Do this before setting i_size. |
| 369 | */ |
Mark Fasheh | 35edec1 | 2007-07-06 14:41:18 -0700 | [diff] [blame] | 370 | cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size); |
| 371 | status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size, |
| 372 | cluster_bytes); |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 373 | if (status) { |
| 374 | mlog_errno(status); |
| 375 | goto out_commit; |
| 376 | } |
| 377 | |
| 378 | i_size_write(inode, new_i_size); |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 379 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
| 380 | |
| 381 | di = (struct ocfs2_dinode *) fe_bh->b_data; |
| 382 | di->i_size = cpu_to_le64(new_i_size); |
| 383 | di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec); |
| 384 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); |
| 385 | |
| 386 | status = ocfs2_journal_dirty(handle, fe_bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 387 | if (status < 0) |
| 388 | mlog_errno(status); |
| 389 | |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 390 | out_commit: |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 391 | ocfs2_commit_trans(osb, handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 392 | out: |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 393 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 394 | mlog_exit(status); |
| 395 | return status; |
| 396 | } |
| 397 | |
| 398 | static int ocfs2_truncate_file(struct inode *inode, |
| 399 | struct buffer_head *di_bh, |
| 400 | u64 new_i_size) |
| 401 | { |
| 402 | int status = 0; |
| 403 | struct ocfs2_dinode *fe = NULL; |
| 404 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 405 | struct ocfs2_truncate_context *tc = NULL; |
| 406 | |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 407 | mlog_entry("(inode = %llu, new_i_size = %llu\n", |
| 408 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
| 409 | (unsigned long long)new_i_size); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 410 | |
Joel Becker | b657c95 | 2008-11-13 14:49:11 -0800 | [diff] [blame] | 411 | /* We trust di_bh because it comes from ocfs2_inode_lock(), which |
| 412 | * already validated it */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 413 | fe = (struct ocfs2_dinode *) di_bh->b_data; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 414 | |
| 415 | 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] | 416 | "Inode %llu, inode i_size = %lld != di " |
| 417 | "i_size = %llu, i_flags = 0x%x\n", |
| 418 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 419 | i_size_read(inode), |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 420 | (unsigned long long)le64_to_cpu(fe->i_size), |
| 421 | le32_to_cpu(fe->i_flags)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 422 | |
| 423 | if (new_i_size > le64_to_cpu(fe->i_size)) { |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 424 | mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n", |
| 425 | (unsigned long long)le64_to_cpu(fe->i_size), |
| 426 | (unsigned long long)new_i_size); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 427 | status = -EINVAL; |
| 428 | mlog_errno(status); |
| 429 | goto bail; |
| 430 | } |
| 431 | |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 432 | mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n", |
| 433 | (unsigned long long)le64_to_cpu(fe->i_blkno), |
| 434 | (unsigned long long)le64_to_cpu(fe->i_size), |
| 435 | (unsigned long long)new_i_size); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 436 | |
| 437 | /* lets handle the simple truncate cases before doing any more |
| 438 | * cluster locking. */ |
| 439 | if (new_i_size == le64_to_cpu(fe->i_size)) |
| 440 | goto bail; |
| 441 | |
Mark Fasheh | 2e89b2e | 2007-05-09 13:40:18 -0700 | [diff] [blame] | 442 | down_write(&OCFS2_I(inode)->ip_alloc_sem); |
| 443 | |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 444 | /* |
| 445 | * The inode lock forced other nodes to sync and drop their |
| 446 | * pages, which (correctly) happens even if we have a truncate |
| 447 | * without allocation change - ocfs2 cluster sizes can be much |
| 448 | * greater than page size, so we have to truncate them |
| 449 | * anyway. |
| 450 | */ |
Mark Fasheh | 2e89b2e | 2007-05-09 13:40:18 -0700 | [diff] [blame] | 451 | unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1); |
| 452 | truncate_inode_pages(inode->i_mapping, new_i_size); |
| 453 | |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 454 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
| 455 | status = ocfs2_truncate_inline(inode, di_bh, new_i_size, |
Mark Fasheh | b1967d0 | 2007-11-20 11:56:39 -0800 | [diff] [blame] | 456 | i_size_read(inode), 1); |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 457 | if (status) |
| 458 | mlog_errno(status); |
| 459 | |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 460 | goto bail_unlock_sem; |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 461 | } |
| 462 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 463 | /* alright, we're going to need to do a full blown alloc size |
| 464 | * change. Orphan the inode so that recovery can complete the |
| 465 | * truncate if necessary. This does the task of marking |
| 466 | * i_size. */ |
| 467 | status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size); |
| 468 | if (status < 0) { |
| 469 | mlog_errno(status); |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 470 | goto bail_unlock_sem; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 471 | } |
| 472 | |
| 473 | status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc); |
| 474 | if (status < 0) { |
| 475 | mlog_errno(status); |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 476 | goto bail_unlock_sem; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 477 | } |
| 478 | |
| 479 | status = ocfs2_commit_truncate(osb, inode, di_bh, tc); |
| 480 | if (status < 0) { |
| 481 | mlog_errno(status); |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 482 | goto bail_unlock_sem; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 483 | } |
| 484 | |
| 485 | /* TODO: orphan dir cleanup here. */ |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 486 | bail_unlock_sem: |
Mark Fasheh | 2e89b2e | 2007-05-09 13:40:18 -0700 | [diff] [blame] | 487 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
| 488 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 489 | bail: |
| 490 | |
| 491 | mlog_exit(status); |
| 492 | return status; |
| 493 | } |
| 494 | |
| 495 | /* |
Tao Ma | 0eb8d47 | 2008-08-18 17:38:45 +0800 | [diff] [blame] | 496 | * extend file allocation only here. |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 497 | * we'll update all the disk stuff, and oip->alloc_size |
| 498 | * |
| 499 | * expect stuff to be locked, a transaction started and enough data / |
| 500 | * metadata reservations in the contexts. |
| 501 | * |
| 502 | * Will return -EAGAIN, and a reason if a restart is needed. |
| 503 | * If passed in, *reason will always be set, even in error. |
| 504 | */ |
Tao Ma | 0eb8d47 | 2008-08-18 17:38:45 +0800 | [diff] [blame] | 505 | int ocfs2_add_inode_data(struct ocfs2_super *osb, |
| 506 | struct inode *inode, |
| 507 | u32 *logical_offset, |
| 508 | u32 clusters_to_add, |
| 509 | int mark_unwritten, |
| 510 | struct buffer_head *fe_bh, |
| 511 | handle_t *handle, |
| 512 | struct ocfs2_alloc_context *data_ac, |
| 513 | struct ocfs2_alloc_context *meta_ac, |
| 514 | enum ocfs2_alloc_restarted *reason_ret) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 515 | { |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 516 | int ret; |
| 517 | struct ocfs2_extent_tree et; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 518 | |
Joel Becker | 5e404e9 | 2009-02-13 03:54:22 -0800 | [diff] [blame] | 519 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh); |
Joel Becker | cbee7e1 | 2009-02-13 03:34:15 -0800 | [diff] [blame] | 520 | ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset, |
| 521 | clusters_to_add, mark_unwritten, |
| 522 | data_ac, meta_ac, reason_ret); |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 523 | |
| 524 | return ret; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 525 | } |
| 526 | |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 527 | static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start, |
| 528 | u32 clusters_to_add, int mark_unwritten) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 529 | { |
| 530 | int status = 0; |
| 531 | int restart_func = 0; |
Mark Fasheh | abf8b15 | 2007-01-17 13:07:24 -0800 | [diff] [blame] | 532 | int credits; |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 533 | u32 prev_clusters; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 534 | struct buffer_head *bh = NULL; |
| 535 | struct ocfs2_dinode *fe = NULL; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 536 | handle_t *handle = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 537 | struct ocfs2_alloc_context *data_ac = NULL; |
| 538 | struct ocfs2_alloc_context *meta_ac = NULL; |
| 539 | enum ocfs2_alloc_restarted why; |
| 540 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 541 | struct ocfs2_extent_tree et; |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 542 | int did_quota = 0; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 543 | |
| 544 | mlog_entry("(clusters_to_add = %u)\n", clusters_to_add); |
| 545 | |
Mark Fasheh | dcd0538 | 2007-01-16 11:32:23 -0800 | [diff] [blame] | 546 | /* |
| 547 | * This function only exists for file systems which don't |
| 548 | * support holes. |
| 549 | */ |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 550 | BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb)); |
Mark Fasheh | dcd0538 | 2007-01-16 11:32:23 -0800 | [diff] [blame] | 551 | |
Joel Becker | b657c95 | 2008-11-13 14:49:11 -0800 | [diff] [blame] | 552 | status = ocfs2_read_inode_block(inode, &bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 553 | if (status < 0) { |
| 554 | mlog_errno(status); |
| 555 | goto leave; |
| 556 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 557 | fe = (struct ocfs2_dinode *) bh->b_data; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 558 | |
| 559 | restart_all: |
| 560 | BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters); |
| 561 | |
Tao Ma | e7d4cb6 | 2008-08-18 17:38:44 +0800 | [diff] [blame] | 562 | mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, " |
| 563 | "clusters_to_add = %u\n", |
| 564 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
| 565 | (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters), |
| 566 | clusters_to_add); |
Joel Becker | 5e404e9 | 2009-02-13 03:54:22 -0800 | [diff] [blame] | 567 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh); |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 568 | status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0, |
| 569 | &data_ac, &meta_ac); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 570 | if (status) { |
| 571 | mlog_errno(status); |
| 572 | goto leave; |
| 573 | } |
| 574 | |
Tao Ma | 811f933 | 2008-08-18 17:38:43 +0800 | [diff] [blame] | 575 | credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list, |
| 576 | clusters_to_add); |
Mark Fasheh | 65eff9c | 2006-10-09 17:26:22 -0700 | [diff] [blame] | 577 | handle = ocfs2_start_trans(osb, credits); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 578 | if (IS_ERR(handle)) { |
| 579 | status = PTR_ERR(handle); |
| 580 | handle = NULL; |
| 581 | mlog_errno(status); |
| 582 | goto leave; |
| 583 | } |
| 584 | |
| 585 | restarted_transaction: |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 586 | if (vfs_dq_alloc_space_nodirty(inode, ocfs2_clusters_to_bytes(osb->sb, |
| 587 | clusters_to_add))) { |
| 588 | status = -EDQUOT; |
| 589 | goto leave; |
| 590 | } |
| 591 | did_quota = 1; |
| 592 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 593 | /* reserve a write to the file entry early on - that we if we |
| 594 | * run out of credits in the allocation path, we can still |
| 595 | * update i_size. */ |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 596 | status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh, |
Joel Becker | 13723d0 | 2008-10-17 19:25:01 -0700 | [diff] [blame] | 597 | OCFS2_JOURNAL_ACCESS_WRITE); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 598 | if (status < 0) { |
| 599 | mlog_errno(status); |
| 600 | goto leave; |
| 601 | } |
| 602 | |
| 603 | prev_clusters = OCFS2_I(inode)->ip_clusters; |
| 604 | |
Tao Ma | 0eb8d47 | 2008-08-18 17:38:45 +0800 | [diff] [blame] | 605 | status = ocfs2_add_inode_data(osb, |
| 606 | inode, |
| 607 | &logical_start, |
| 608 | clusters_to_add, |
| 609 | mark_unwritten, |
| 610 | bh, |
| 611 | handle, |
| 612 | data_ac, |
| 613 | meta_ac, |
| 614 | &why); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 615 | if ((status < 0) && (status != -EAGAIN)) { |
| 616 | if (status != -ENOSPC) |
| 617 | mlog_errno(status); |
| 618 | goto leave; |
| 619 | } |
| 620 | |
| 621 | status = ocfs2_journal_dirty(handle, bh); |
| 622 | if (status < 0) { |
| 623 | mlog_errno(status); |
| 624 | goto leave; |
| 625 | } |
| 626 | |
| 627 | spin_lock(&OCFS2_I(inode)->ip_lock); |
| 628 | clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters); |
| 629 | spin_unlock(&OCFS2_I(inode)->ip_lock); |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 630 | /* Release unused quota reservation */ |
| 631 | vfs_dq_free_space(inode, |
| 632 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_add)); |
| 633 | did_quota = 0; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 634 | |
| 635 | if (why != RESTART_NONE && clusters_to_add) { |
| 636 | if (why == RESTART_META) { |
| 637 | mlog(0, "restarting function.\n"); |
| 638 | restart_func = 1; |
| 639 | } else { |
| 640 | BUG_ON(why != RESTART_TRANS); |
| 641 | |
| 642 | mlog(0, "restarting transaction.\n"); |
| 643 | /* TODO: This can be more intelligent. */ |
| 644 | credits = ocfs2_calc_extend_credits(osb->sb, |
Tao Ma | 811f933 | 2008-08-18 17:38:43 +0800 | [diff] [blame] | 645 | &fe->id2.i_list, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 646 | clusters_to_add); |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 647 | status = ocfs2_extend_trans(handle, credits); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 648 | if (status < 0) { |
| 649 | /* handle still has to be committed at |
| 650 | * this point. */ |
| 651 | status = -ENOMEM; |
| 652 | mlog_errno(status); |
| 653 | goto leave; |
| 654 | } |
| 655 | goto restarted_transaction; |
| 656 | } |
| 657 | } |
| 658 | |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 659 | mlog(0, "fe: i_clusters = %u, i_size=%llu\n", |
Mark Fasheh | 1ca1a11 | 2007-04-27 16:01:25 -0700 | [diff] [blame] | 660 | le32_to_cpu(fe->i_clusters), |
| 661 | (unsigned long long)le64_to_cpu(fe->i_size)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 662 | mlog(0, "inode: ip_clusters=%u, i_size=%lld\n", |
Jan Kara | 634bf74 | 2007-12-19 15:25:42 +0100 | [diff] [blame] | 663 | OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 664 | |
| 665 | leave: |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 666 | if (status < 0 && did_quota) |
| 667 | vfs_dq_free_space(inode, |
| 668 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_add)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 669 | if (handle) { |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 670 | ocfs2_commit_trans(osb, handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 671 | handle = NULL; |
| 672 | } |
| 673 | if (data_ac) { |
| 674 | ocfs2_free_alloc_context(data_ac); |
| 675 | data_ac = NULL; |
| 676 | } |
| 677 | if (meta_ac) { |
| 678 | ocfs2_free_alloc_context(meta_ac); |
| 679 | meta_ac = NULL; |
| 680 | } |
| 681 | if ((!status) && restart_func) { |
| 682 | restart_func = 0; |
| 683 | goto restart_all; |
| 684 | } |
Mark Fasheh | a81cb88 | 2008-10-07 14:25:16 -0700 | [diff] [blame] | 685 | brelse(bh); |
| 686 | bh = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 687 | |
| 688 | mlog_exit(status); |
| 689 | return status; |
| 690 | } |
| 691 | |
| 692 | /* Some parts of this taken from generic_cont_expand, which turned out |
| 693 | * 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] | 694 | * worry about recursive locking in ->write_begin() and ->write_end(). */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 695 | static int ocfs2_write_zero_page(struct inode *inode, |
| 696 | u64 size) |
| 697 | { |
| 698 | struct address_space *mapping = inode->i_mapping; |
| 699 | struct page *page; |
| 700 | unsigned long index; |
| 701 | unsigned int offset; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 702 | handle_t *handle = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 703 | int ret; |
| 704 | |
| 705 | offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */ |
| 706 | /* ugh. in prepare/commit_write, if from==to==start of block, we |
| 707 | ** skip the prepare. make sure we never send an offset for the start |
| 708 | ** of a block |
| 709 | */ |
| 710 | if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) { |
| 711 | offset++; |
| 712 | } |
| 713 | index = size >> PAGE_CACHE_SHIFT; |
| 714 | |
| 715 | page = grab_cache_page(mapping, index); |
| 716 | if (!page) { |
| 717 | ret = -ENOMEM; |
| 718 | mlog_errno(ret); |
| 719 | goto out; |
| 720 | } |
| 721 | |
Mark Fasheh | 53013cb | 2006-05-05 19:04:03 -0700 | [diff] [blame] | 722 | ret = ocfs2_prepare_write_nolock(inode, page, offset, offset); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 723 | if (ret < 0) { |
| 724 | mlog_errno(ret); |
| 725 | goto out_unlock; |
| 726 | } |
| 727 | |
| 728 | if (ocfs2_should_order_data(inode)) { |
| 729 | handle = ocfs2_start_walk_page_trans(inode, page, offset, |
| 730 | offset); |
| 731 | if (IS_ERR(handle)) { |
| 732 | ret = PTR_ERR(handle); |
| 733 | handle = NULL; |
| 734 | goto out_unlock; |
| 735 | } |
| 736 | } |
| 737 | |
| 738 | /* must not update i_size! */ |
| 739 | ret = block_commit_write(page, offset, offset); |
| 740 | if (ret < 0) |
| 741 | mlog_errno(ret); |
| 742 | else |
| 743 | ret = 0; |
| 744 | |
| 745 | if (handle) |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 746 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 747 | out_unlock: |
| 748 | unlock_page(page); |
| 749 | page_cache_release(page); |
| 750 | out: |
| 751 | return ret; |
| 752 | } |
| 753 | |
| 754 | static int ocfs2_zero_extend(struct inode *inode, |
| 755 | u64 zero_to_size) |
| 756 | { |
| 757 | int ret = 0; |
| 758 | u64 start_off; |
| 759 | struct super_block *sb = inode->i_sb; |
| 760 | |
| 761 | start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode)); |
| 762 | while (start_off < zero_to_size) { |
| 763 | ret = ocfs2_write_zero_page(inode, start_off); |
| 764 | if (ret < 0) { |
| 765 | mlog_errno(ret); |
| 766 | goto out; |
| 767 | } |
| 768 | |
| 769 | start_off += sb->s_blocksize; |
Mark Fasheh | e2057c5 | 2006-10-03 17:53:05 -0700 | [diff] [blame] | 770 | |
| 771 | /* |
| 772 | * Very large extends have the potential to lock up |
| 773 | * the cpu for extended periods of time. |
| 774 | */ |
| 775 | cond_resched(); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 776 | } |
| 777 | |
| 778 | out: |
| 779 | return ret; |
| 780 | } |
| 781 | |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 782 | int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to) |
| 783 | { |
| 784 | int ret; |
| 785 | u32 clusters_to_add; |
| 786 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 787 | |
| 788 | clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size); |
| 789 | if (clusters_to_add < oi->ip_clusters) |
| 790 | clusters_to_add = 0; |
| 791 | else |
| 792 | clusters_to_add -= oi->ip_clusters; |
| 793 | |
| 794 | if (clusters_to_add) { |
| 795 | ret = __ocfs2_extend_allocation(inode, oi->ip_clusters, |
| 796 | clusters_to_add, 0); |
| 797 | if (ret) { |
| 798 | mlog_errno(ret); |
| 799 | goto out; |
| 800 | } |
| 801 | } |
| 802 | |
| 803 | /* |
| 804 | * Call this even if we don't add any clusters to the tree. We |
| 805 | * still need to zero the area between the old i_size and the |
| 806 | * new i_size. |
| 807 | */ |
| 808 | ret = ocfs2_zero_extend(inode, zero_to); |
| 809 | if (ret < 0) |
| 810 | mlog_errno(ret); |
| 811 | |
| 812 | out: |
| 813 | return ret; |
| 814 | } |
| 815 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 816 | static int ocfs2_extend_file(struct inode *inode, |
| 817 | struct buffer_head *di_bh, |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 818 | u64 new_i_size) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 819 | { |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 820 | int ret = 0; |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 821 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 822 | |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 823 | BUG_ON(!di_bh); |
Mark Fasheh | 53013cb | 2006-05-05 19:04:03 -0700 | [diff] [blame] | 824 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 825 | /* setattr sometimes calls us like this. */ |
| 826 | if (new_i_size == 0) |
| 827 | goto out; |
| 828 | |
| 829 | if (i_size_read(inode) == new_i_size) |
| 830 | goto out; |
| 831 | BUG_ON(new_i_size < i_size_read(inode)); |
| 832 | |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 833 | /* |
| 834 | * Fall through for converting inline data, even if the fs |
| 835 | * supports sparse files. |
| 836 | * |
| 837 | * The check for inline data here is legal - nobody can add |
| 838 | * the feature since we have i_mutex. We must check it again |
| 839 | * after acquiring ip_alloc_sem though, as paths like mmap |
| 840 | * might have raced us to converting the inode to extents. |
| 841 | */ |
| 842 | if (!(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) |
| 843 | && ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) |
Mark Fasheh | 3a0782d | 2007-01-17 12:53:31 -0800 | [diff] [blame] | 844 | goto out_update_size; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 845 | |
Mark Fasheh | 0effef7 | 2006-10-03 17:44:42 -0700 | [diff] [blame] | 846 | /* |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 847 | * The alloc sem blocks people in read/write from reading our |
| 848 | * allocation until we're done changing it. We depend on |
| 849 | * i_mutex to block other extend/truncate calls while we're |
| 850 | * here. |
Mark Fasheh | 0effef7 | 2006-10-03 17:44:42 -0700 | [diff] [blame] | 851 | */ |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 852 | down_write(&oi->ip_alloc_sem); |
| 853 | |
| 854 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
| 855 | /* |
| 856 | * We can optimize small extends by keeping the inodes |
| 857 | * inline data. |
| 858 | */ |
| 859 | if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) { |
| 860 | up_write(&oi->ip_alloc_sem); |
| 861 | goto out_update_size; |
| 862 | } |
| 863 | |
| 864 | ret = ocfs2_convert_inline_data_to_extents(inode, di_bh); |
| 865 | if (ret) { |
| 866 | up_write(&oi->ip_alloc_sem); |
| 867 | |
| 868 | mlog_errno(ret); |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 869 | goto out; |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 870 | } |
| 871 | } |
| 872 | |
| 873 | if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) |
| 874 | ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size); |
| 875 | |
| 876 | up_write(&oi->ip_alloc_sem); |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 877 | |
Mark Fasheh | 0effef7 | 2006-10-03 17:44:42 -0700 | [diff] [blame] | 878 | if (ret < 0) { |
| 879 | mlog_errno(ret); |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 880 | goto out; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 881 | } |
| 882 | |
Mark Fasheh | 3a0782d | 2007-01-17 12:53:31 -0800 | [diff] [blame] | 883 | out_update_size: |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 884 | ret = ocfs2_simple_size_update(inode, di_bh, new_i_size); |
| 885 | if (ret < 0) |
| 886 | mlog_errno(ret); |
Mark Fasheh | 53013cb | 2006-05-05 19:04:03 -0700 | [diff] [blame] | 887 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 888 | out: |
| 889 | return ret; |
| 890 | } |
| 891 | |
| 892 | int ocfs2_setattr(struct dentry *dentry, struct iattr *attr) |
| 893 | { |
| 894 | int status = 0, size_change; |
| 895 | struct inode *inode = dentry->d_inode; |
| 896 | struct super_block *sb = inode->i_sb; |
| 897 | struct ocfs2_super *osb = OCFS2_SB(sb); |
| 898 | struct buffer_head *bh = NULL; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 899 | handle_t *handle = NULL; |
Jan Kara | 65bac57 | 2009-06-02 14:24:01 +0200 | [diff] [blame] | 900 | int qtype; |
| 901 | struct dquot *transfer_from[MAXQUOTAS] = { }; |
| 902 | struct dquot *transfer_to[MAXQUOTAS] = { }; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 903 | |
| 904 | mlog_entry("(0x%p, '%.*s')\n", dentry, |
| 905 | dentry->d_name.len, dentry->d_name.name); |
| 906 | |
Sunil Mushran | bc53580 | 2008-04-18 10:23:53 -0700 | [diff] [blame] | 907 | /* ensuring we don't even attempt to truncate a symlink */ |
| 908 | if (S_ISLNK(inode->i_mode)) |
| 909 | attr->ia_valid &= ~ATTR_SIZE; |
| 910 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 911 | if (attr->ia_valid & ATTR_MODE) |
| 912 | mlog(0, "mode change: %d\n", attr->ia_mode); |
| 913 | if (attr->ia_valid & ATTR_UID) |
| 914 | mlog(0, "uid change: %d\n", attr->ia_uid); |
| 915 | if (attr->ia_valid & ATTR_GID) |
| 916 | mlog(0, "gid change: %d\n", attr->ia_gid); |
| 917 | if (attr->ia_valid & ATTR_SIZE) |
| 918 | mlog(0, "size change...\n"); |
| 919 | if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME)) |
| 920 | mlog(0, "time change...\n"); |
| 921 | |
| 922 | #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \ |
| 923 | | ATTR_GID | ATTR_UID | ATTR_MODE) |
| 924 | if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) { |
| 925 | mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid); |
| 926 | return 0; |
| 927 | } |
| 928 | |
| 929 | status = inode_change_ok(inode, attr); |
| 930 | if (status) |
| 931 | return status; |
| 932 | |
| 933 | size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE; |
| 934 | if (size_change) { |
| 935 | status = ocfs2_rw_lock(inode, 1); |
| 936 | if (status < 0) { |
| 937 | mlog_errno(status); |
| 938 | goto bail; |
| 939 | } |
| 940 | } |
| 941 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 942 | status = ocfs2_inode_lock(inode, &bh, 1); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 943 | if (status < 0) { |
| 944 | if (status != -ENOENT) |
| 945 | mlog_errno(status); |
| 946 | goto bail_unlock_rw; |
| 947 | } |
| 948 | |
| 949 | if (size_change && attr->ia_size != i_size_read(inode)) { |
Mark Fasheh | ce76fd3 | 2007-07-20 12:02:14 -0700 | [diff] [blame] | 950 | if (attr->ia_size > sb->s_maxbytes) { |
| 951 | status = -EFBIG; |
| 952 | goto bail_unlock; |
| 953 | } |
| 954 | |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 955 | if (i_size_read(inode) > attr->ia_size) { |
| 956 | if (ocfs2_should_order_data(inode)) { |
| 957 | status = ocfs2_begin_ordered_truncate(inode, |
| 958 | attr->ia_size); |
| 959 | if (status) |
| 960 | goto bail_unlock; |
| 961 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 962 | status = ocfs2_truncate_file(inode, bh, attr->ia_size); |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 963 | } else |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 964 | status = ocfs2_extend_file(inode, bh, attr->ia_size); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 965 | if (status < 0) { |
| 966 | if (status != -ENOSPC) |
| 967 | mlog_errno(status); |
| 968 | status = -ENOSPC; |
| 969 | goto bail_unlock; |
| 970 | } |
| 971 | } |
| 972 | |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 973 | if ((attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) || |
| 974 | (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) { |
Jan Kara | 65bac57 | 2009-06-02 14:24:01 +0200 | [diff] [blame] | 975 | /* |
| 976 | * Gather pointers to quota structures so that allocation / |
| 977 | * freeing of quota structures happens here and not inside |
| 978 | * vfs_dq_transfer() where we have problems with lock ordering |
| 979 | */ |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 980 | if (attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid |
| 981 | && OCFS2_HAS_RO_COMPAT_FEATURE(sb, |
| 982 | OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) { |
Jan Kara | 65bac57 | 2009-06-02 14:24:01 +0200 | [diff] [blame] | 983 | transfer_to[USRQUOTA] = dqget(sb, attr->ia_uid, |
| 984 | USRQUOTA); |
| 985 | transfer_from[USRQUOTA] = dqget(sb, inode->i_uid, |
| 986 | USRQUOTA); |
| 987 | if (!transfer_to[USRQUOTA] || !transfer_from[USRQUOTA]) { |
| 988 | status = -ESRCH; |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 989 | goto bail_unlock; |
Jan Kara | 65bac57 | 2009-06-02 14:24:01 +0200 | [diff] [blame] | 990 | } |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 991 | } |
| 992 | if (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid |
| 993 | && OCFS2_HAS_RO_COMPAT_FEATURE(sb, |
| 994 | OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) { |
Jan Kara | 65bac57 | 2009-06-02 14:24:01 +0200 | [diff] [blame] | 995 | transfer_to[GRPQUOTA] = dqget(sb, attr->ia_gid, |
| 996 | GRPQUOTA); |
| 997 | transfer_from[GRPQUOTA] = dqget(sb, inode->i_gid, |
| 998 | GRPQUOTA); |
| 999 | if (!transfer_to[GRPQUOTA] || !transfer_from[GRPQUOTA]) { |
| 1000 | status = -ESRCH; |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 1001 | goto bail_unlock; |
Jan Kara | 65bac57 | 2009-06-02 14:24:01 +0200 | [diff] [blame] | 1002 | } |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 1003 | } |
Jan Kara | 65bac57 | 2009-06-02 14:24:01 +0200 | [diff] [blame] | 1004 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS + |
| 1005 | 2 * ocfs2_quota_trans_credits(sb)); |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 1006 | if (IS_ERR(handle)) { |
| 1007 | status = PTR_ERR(handle); |
| 1008 | mlog_errno(status); |
| 1009 | goto bail_unlock; |
| 1010 | } |
| 1011 | status = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0; |
| 1012 | if (status < 0) |
| 1013 | goto bail_commit; |
| 1014 | } else { |
| 1015 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
| 1016 | if (IS_ERR(handle)) { |
| 1017 | status = PTR_ERR(handle); |
| 1018 | mlog_errno(status); |
| 1019 | goto bail_unlock; |
| 1020 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1021 | } |
| 1022 | |
Mark Fasheh | 7307de8 | 2007-05-09 15:16:19 -0700 | [diff] [blame] | 1023 | /* |
| 1024 | * This will intentionally not wind up calling vmtruncate(), |
| 1025 | * since all the work for a size change has been done above. |
| 1026 | * Otherwise, we could get into problems with truncate as |
| 1027 | * ip_alloc_sem is used there to protect against i_size |
| 1028 | * changes. |
| 1029 | */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1030 | status = inode_setattr(inode, attr); |
| 1031 | if (status < 0) { |
| 1032 | mlog_errno(status); |
| 1033 | goto bail_commit; |
| 1034 | } |
| 1035 | |
| 1036 | status = ocfs2_mark_inode_dirty(handle, inode, bh); |
| 1037 | if (status < 0) |
| 1038 | mlog_errno(status); |
| 1039 | |
| 1040 | bail_commit: |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 1041 | ocfs2_commit_trans(osb, handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1042 | bail_unlock: |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1043 | ocfs2_inode_unlock(inode, 1); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1044 | bail_unlock_rw: |
| 1045 | if (size_change) |
| 1046 | ocfs2_rw_unlock(inode, 1); |
| 1047 | bail: |
Mark Fasheh | a81cb88 | 2008-10-07 14:25:16 -0700 | [diff] [blame] | 1048 | brelse(bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1049 | |
Jan Kara | 65bac57 | 2009-06-02 14:24:01 +0200 | [diff] [blame] | 1050 | /* Release quota pointers in case we acquired them */ |
| 1051 | for (qtype = 0; qtype < MAXQUOTAS; qtype++) { |
| 1052 | dqput(transfer_to[qtype]); |
| 1053 | dqput(transfer_from[qtype]); |
| 1054 | } |
| 1055 | |
Tiger Yang | 060bc66 | 2008-11-14 11:17:29 +0800 | [diff] [blame] | 1056 | if (!status && attr->ia_valid & ATTR_MODE) { |
| 1057 | status = ocfs2_acl_chmod(inode); |
| 1058 | if (status < 0) |
| 1059 | mlog_errno(status); |
| 1060 | } |
| 1061 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1062 | mlog_exit(status); |
| 1063 | return status; |
| 1064 | } |
| 1065 | |
| 1066 | int ocfs2_getattr(struct vfsmount *mnt, |
| 1067 | struct dentry *dentry, |
| 1068 | struct kstat *stat) |
| 1069 | { |
| 1070 | struct inode *inode = dentry->d_inode; |
| 1071 | struct super_block *sb = dentry->d_inode->i_sb; |
| 1072 | struct ocfs2_super *osb = sb->s_fs_info; |
| 1073 | int err; |
| 1074 | |
| 1075 | mlog_entry_void(); |
| 1076 | |
| 1077 | err = ocfs2_inode_revalidate(dentry); |
| 1078 | if (err) { |
| 1079 | if (err != -ENOENT) |
| 1080 | mlog_errno(err); |
| 1081 | goto bail; |
| 1082 | } |
| 1083 | |
| 1084 | generic_fillattr(inode, stat); |
| 1085 | |
| 1086 | /* We set the blksize from the cluster size for performance */ |
| 1087 | stat->blksize = osb->s_clustersize; |
| 1088 | |
| 1089 | bail: |
| 1090 | mlog_exit(err); |
| 1091 | |
| 1092 | return err; |
| 1093 | } |
| 1094 | |
Al Viro | e6305c4 | 2008-07-15 21:03:57 -0400 | [diff] [blame] | 1095 | int ocfs2_permission(struct inode *inode, int mask) |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 1096 | { |
| 1097 | int ret; |
| 1098 | |
| 1099 | mlog_entry_void(); |
| 1100 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1101 | ret = ocfs2_inode_lock(inode, NULL, 0); |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 1102 | if (ret) { |
Mark Fasheh | a9f5f70 | 2007-04-26 11:43:43 -0700 | [diff] [blame] | 1103 | if (ret != -ENOENT) |
| 1104 | mlog_errno(ret); |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 1105 | goto out; |
| 1106 | } |
| 1107 | |
Tiger Yang | 23fc270 | 2008-11-14 11:17:18 +0800 | [diff] [blame] | 1108 | ret = generic_permission(inode, mask, ocfs2_check_acl); |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 1109 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1110 | ocfs2_inode_unlock(inode, 0); |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 1111 | out: |
| 1112 | mlog_exit(ret); |
| 1113 | return ret; |
| 1114 | } |
| 1115 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1116 | static int __ocfs2_write_remove_suid(struct inode *inode, |
| 1117 | struct buffer_head *bh) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1118 | { |
| 1119 | int ret; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 1120 | handle_t *handle; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1121 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1122 | struct ocfs2_dinode *di; |
| 1123 | |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 1124 | mlog_entry("(Inode %llu, mode 0%o)\n", |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1125 | (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1126 | |
Mark Fasheh | 65eff9c | 2006-10-09 17:26:22 -0700 | [diff] [blame] | 1127 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
Jan Kara | fa38e92 | 2008-10-20 19:23:51 +0200 | [diff] [blame] | 1128 | if (IS_ERR(handle)) { |
| 1129 | ret = PTR_ERR(handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1130 | mlog_errno(ret); |
| 1131 | goto out; |
| 1132 | } |
| 1133 | |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 1134 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh, |
Joel Becker | 13723d0 | 2008-10-17 19:25:01 -0700 | [diff] [blame] | 1135 | OCFS2_JOURNAL_ACCESS_WRITE); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1136 | if (ret < 0) { |
| 1137 | mlog_errno(ret); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1138 | goto out_trans; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1139 | } |
| 1140 | |
| 1141 | inode->i_mode &= ~S_ISUID; |
| 1142 | if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP)) |
| 1143 | inode->i_mode &= ~S_ISGID; |
| 1144 | |
| 1145 | di = (struct ocfs2_dinode *) bh->b_data; |
| 1146 | di->i_mode = cpu_to_le16(inode->i_mode); |
| 1147 | |
| 1148 | ret = ocfs2_journal_dirty(handle, bh); |
| 1149 | if (ret < 0) |
| 1150 | mlog_errno(ret); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1151 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1152 | out_trans: |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 1153 | ocfs2_commit_trans(osb, handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1154 | out: |
| 1155 | mlog_exit(ret); |
| 1156 | return ret; |
| 1157 | } |
| 1158 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1159 | /* |
| 1160 | * Will look for holes and unwritten extents in the range starting at |
| 1161 | * pos for count bytes (inclusive). |
| 1162 | */ |
| 1163 | static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos, |
| 1164 | size_t count) |
| 1165 | { |
| 1166 | int ret = 0; |
Mark Fasheh | 49cb8d2 | 2007-03-09 16:21:46 -0800 | [diff] [blame] | 1167 | unsigned int extent_flags; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1168 | u32 cpos, clusters, extent_len, phys_cpos; |
| 1169 | struct super_block *sb = inode->i_sb; |
| 1170 | |
| 1171 | cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits; |
| 1172 | clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos; |
| 1173 | |
| 1174 | while (clusters) { |
Mark Fasheh | 49cb8d2 | 2007-03-09 16:21:46 -0800 | [diff] [blame] | 1175 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len, |
| 1176 | &extent_flags); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1177 | if (ret < 0) { |
| 1178 | mlog_errno(ret); |
| 1179 | goto out; |
| 1180 | } |
| 1181 | |
Mark Fasheh | 49cb8d2 | 2007-03-09 16:21:46 -0800 | [diff] [blame] | 1182 | if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) { |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1183 | ret = 1; |
| 1184 | break; |
| 1185 | } |
| 1186 | |
| 1187 | if (extent_len > clusters) |
| 1188 | extent_len = clusters; |
| 1189 | |
| 1190 | clusters -= extent_len; |
| 1191 | cpos += extent_len; |
| 1192 | } |
| 1193 | out: |
| 1194 | return ret; |
| 1195 | } |
| 1196 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1197 | static int ocfs2_write_remove_suid(struct inode *inode) |
| 1198 | { |
| 1199 | int ret; |
| 1200 | struct buffer_head *bh = NULL; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1201 | |
Joel Becker | b657c95 | 2008-11-13 14:49:11 -0800 | [diff] [blame] | 1202 | ret = ocfs2_read_inode_block(inode, &bh); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1203 | if (ret < 0) { |
| 1204 | mlog_errno(ret); |
| 1205 | goto out; |
| 1206 | } |
| 1207 | |
| 1208 | ret = __ocfs2_write_remove_suid(inode, bh); |
| 1209 | out: |
| 1210 | brelse(bh); |
| 1211 | return ret; |
| 1212 | } |
| 1213 | |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 1214 | /* |
| 1215 | * Allocate enough extents to cover the region starting at byte offset |
| 1216 | * start for len bytes. Existing extents are skipped, any extents |
| 1217 | * added are marked as "unwritten". |
| 1218 | */ |
| 1219 | static int ocfs2_allocate_unwritten_extents(struct inode *inode, |
| 1220 | u64 start, u64 len) |
| 1221 | { |
| 1222 | int ret; |
| 1223 | u32 cpos, phys_cpos, clusters, alloc_size; |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1224 | u64 end = start + len; |
| 1225 | struct buffer_head *di_bh = NULL; |
| 1226 | |
| 1227 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
Joel Becker | b657c95 | 2008-11-13 14:49:11 -0800 | [diff] [blame] | 1228 | ret = ocfs2_read_inode_block(inode, &di_bh); |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1229 | if (ret) { |
| 1230 | mlog_errno(ret); |
| 1231 | goto out; |
| 1232 | } |
| 1233 | |
| 1234 | /* |
| 1235 | * Nothing to do if the requested reservation range |
| 1236 | * fits within the inode. |
| 1237 | */ |
| 1238 | if (ocfs2_size_fits_inline_data(di_bh, end)) |
| 1239 | goto out; |
| 1240 | |
| 1241 | ret = ocfs2_convert_inline_data_to_extents(inode, di_bh); |
| 1242 | if (ret) { |
| 1243 | mlog_errno(ret); |
| 1244 | goto out; |
| 1245 | } |
| 1246 | } |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 1247 | |
| 1248 | /* |
| 1249 | * We consider both start and len to be inclusive. |
| 1250 | */ |
| 1251 | cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits; |
| 1252 | clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len); |
| 1253 | clusters -= cpos; |
| 1254 | |
| 1255 | while (clusters) { |
| 1256 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, |
| 1257 | &alloc_size, NULL); |
| 1258 | if (ret) { |
| 1259 | mlog_errno(ret); |
| 1260 | goto out; |
| 1261 | } |
| 1262 | |
| 1263 | /* |
| 1264 | * Hole or existing extent len can be arbitrary, so |
| 1265 | * cap it to our own allocation request. |
| 1266 | */ |
| 1267 | if (alloc_size > clusters) |
| 1268 | alloc_size = clusters; |
| 1269 | |
| 1270 | if (phys_cpos) { |
| 1271 | /* |
| 1272 | * We already have an allocation at this |
| 1273 | * region so we can safely skip it. |
| 1274 | */ |
| 1275 | goto next; |
| 1276 | } |
| 1277 | |
| 1278 | ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1); |
| 1279 | if (ret) { |
| 1280 | if (ret != -ENOSPC) |
| 1281 | mlog_errno(ret); |
| 1282 | goto out; |
| 1283 | } |
| 1284 | |
| 1285 | next: |
| 1286 | cpos += alloc_size; |
| 1287 | clusters -= alloc_size; |
| 1288 | } |
| 1289 | |
| 1290 | ret = 0; |
| 1291 | out: |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1292 | |
| 1293 | brelse(di_bh); |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 1294 | return ret; |
| 1295 | } |
| 1296 | |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1297 | /* |
| 1298 | * Truncate a byte range, avoiding pages within partial clusters. This |
| 1299 | * preserves those pages for the zeroing code to write to. |
| 1300 | */ |
| 1301 | static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start, |
| 1302 | u64 byte_len) |
| 1303 | { |
| 1304 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1305 | loff_t start, end; |
| 1306 | struct address_space *mapping = inode->i_mapping; |
| 1307 | |
| 1308 | start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start); |
| 1309 | end = byte_start + byte_len; |
| 1310 | end = end & ~(osb->s_clustersize - 1); |
| 1311 | |
| 1312 | if (start < end) { |
| 1313 | unmap_mapping_range(mapping, start, end - start, 0); |
| 1314 | truncate_inode_pages_range(mapping, start, end - 1); |
| 1315 | } |
| 1316 | } |
| 1317 | |
| 1318 | static int ocfs2_zero_partial_clusters(struct inode *inode, |
| 1319 | u64 start, u64 len) |
| 1320 | { |
| 1321 | int ret = 0; |
| 1322 | u64 tmpend, end = start + len; |
| 1323 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1324 | unsigned int csize = osb->s_clustersize; |
| 1325 | handle_t *handle; |
| 1326 | |
| 1327 | /* |
| 1328 | * The "start" and "end" values are NOT necessarily part of |
| 1329 | * the range whose allocation is being deleted. Rather, this |
| 1330 | * is what the user passed in with the request. We must zero |
| 1331 | * partial clusters here. There's no need to worry about |
| 1332 | * physical allocation - the zeroing code knows to skip holes. |
| 1333 | */ |
| 1334 | mlog(0, "byte start: %llu, end: %llu\n", |
| 1335 | (unsigned long long)start, (unsigned long long)end); |
| 1336 | |
| 1337 | /* |
| 1338 | * If both edges are on a cluster boundary then there's no |
| 1339 | * zeroing required as the region is part of the allocation to |
| 1340 | * be truncated. |
| 1341 | */ |
| 1342 | if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0) |
| 1343 | goto out; |
| 1344 | |
| 1345 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
Jan Kara | fa38e92 | 2008-10-20 19:23:51 +0200 | [diff] [blame] | 1346 | if (IS_ERR(handle)) { |
| 1347 | ret = PTR_ERR(handle); |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1348 | mlog_errno(ret); |
| 1349 | goto out; |
| 1350 | } |
| 1351 | |
| 1352 | /* |
| 1353 | * We want to get the byte offset of the end of the 1st cluster. |
| 1354 | */ |
| 1355 | tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1)); |
| 1356 | if (tmpend > end) |
| 1357 | tmpend = end; |
| 1358 | |
| 1359 | mlog(0, "1st range: start: %llu, tmpend: %llu\n", |
| 1360 | (unsigned long long)start, (unsigned long long)tmpend); |
| 1361 | |
| 1362 | ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend); |
| 1363 | if (ret) |
| 1364 | mlog_errno(ret); |
| 1365 | |
| 1366 | if (tmpend < end) { |
| 1367 | /* |
| 1368 | * This may make start and end equal, but the zeroing |
| 1369 | * code will skip any work in that case so there's no |
| 1370 | * need to catch it up here. |
| 1371 | */ |
| 1372 | start = end & ~(osb->s_clustersize - 1); |
| 1373 | |
| 1374 | mlog(0, "2nd range: start: %llu, end: %llu\n", |
| 1375 | (unsigned long long)start, (unsigned long long)end); |
| 1376 | |
| 1377 | ret = ocfs2_zero_range_for_truncate(inode, handle, start, end); |
| 1378 | if (ret) |
| 1379 | mlog_errno(ret); |
| 1380 | } |
| 1381 | |
| 1382 | ocfs2_commit_trans(osb, handle); |
| 1383 | out: |
| 1384 | return ret; |
| 1385 | } |
| 1386 | |
| 1387 | static int ocfs2_remove_inode_range(struct inode *inode, |
| 1388 | struct buffer_head *di_bh, u64 byte_start, |
| 1389 | u64 byte_len) |
| 1390 | { |
| 1391 | int ret = 0; |
| 1392 | u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size; |
| 1393 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1394 | struct ocfs2_cached_dealloc_ctxt dealloc; |
Mark Fasheh | b1967d0 | 2007-11-20 11:56:39 -0800 | [diff] [blame] | 1395 | struct address_space *mapping = inode->i_mapping; |
Mark Fasheh | fecc011 | 2008-11-12 15:16:38 -0800 | [diff] [blame] | 1396 | struct ocfs2_extent_tree et; |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1397 | |
Joel Becker | 5e404e9 | 2009-02-13 03:54:22 -0800 | [diff] [blame] | 1398 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh); |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1399 | ocfs2_init_dealloc_ctxt(&dealloc); |
| 1400 | |
| 1401 | if (byte_len == 0) |
| 1402 | return 0; |
| 1403 | |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1404 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
| 1405 | ret = ocfs2_truncate_inline(inode, di_bh, byte_start, |
Mark Fasheh | b1967d0 | 2007-11-20 11:56:39 -0800 | [diff] [blame] | 1406 | byte_start + byte_len, 0); |
| 1407 | if (ret) { |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1408 | mlog_errno(ret); |
Mark Fasheh | b1967d0 | 2007-11-20 11:56:39 -0800 | [diff] [blame] | 1409 | goto out; |
| 1410 | } |
| 1411 | /* |
| 1412 | * There's no need to get fancy with the page cache |
| 1413 | * truncate of an inline-data inode. We're talking |
| 1414 | * about less than a page here, which will be cached |
| 1415 | * in the dinode buffer anyway. |
| 1416 | */ |
| 1417 | unmap_mapping_range(mapping, 0, 0, 0); |
| 1418 | truncate_inode_pages(mapping, 0); |
| 1419 | goto out; |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1420 | } |
| 1421 | |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1422 | trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start); |
| 1423 | trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits; |
| 1424 | if (trunc_len >= trunc_start) |
| 1425 | trunc_len -= trunc_start; |
| 1426 | else |
| 1427 | trunc_len = 0; |
| 1428 | |
| 1429 | mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n", |
| 1430 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
| 1431 | (unsigned long long)byte_start, |
| 1432 | (unsigned long long)byte_len, trunc_start, trunc_len); |
| 1433 | |
| 1434 | ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len); |
| 1435 | if (ret) { |
| 1436 | mlog_errno(ret); |
| 1437 | goto out; |
| 1438 | } |
| 1439 | |
| 1440 | cpos = trunc_start; |
| 1441 | while (trunc_len) { |
| 1442 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, |
| 1443 | &alloc_size, NULL); |
| 1444 | if (ret) { |
| 1445 | mlog_errno(ret); |
| 1446 | goto out; |
| 1447 | } |
| 1448 | |
| 1449 | if (alloc_size > trunc_len) |
| 1450 | alloc_size = trunc_len; |
| 1451 | |
| 1452 | /* Only do work for non-holes */ |
| 1453 | if (phys_cpos != 0) { |
Mark Fasheh | fecc011 | 2008-11-12 15:16:38 -0800 | [diff] [blame] | 1454 | ret = ocfs2_remove_btree_range(inode, &et, cpos, |
| 1455 | phys_cpos, alloc_size, |
| 1456 | &dealloc); |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1457 | if (ret) { |
| 1458 | mlog_errno(ret); |
| 1459 | goto out; |
| 1460 | } |
| 1461 | } |
| 1462 | |
| 1463 | cpos += alloc_size; |
| 1464 | trunc_len -= alloc_size; |
| 1465 | } |
| 1466 | |
| 1467 | ocfs2_truncate_cluster_pages(inode, byte_start, byte_len); |
| 1468 | |
| 1469 | out: |
| 1470 | ocfs2_schedule_truncate_log_flush(osb, 1); |
| 1471 | ocfs2_run_deallocs(osb, &dealloc); |
| 1472 | |
| 1473 | return ret; |
| 1474 | } |
| 1475 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1476 | /* |
| 1477 | * Parts of this function taken from xfs_change_file_space() |
| 1478 | */ |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1479 | static int __ocfs2_change_file_space(struct file *file, struct inode *inode, |
| 1480 | loff_t f_pos, unsigned int cmd, |
| 1481 | struct ocfs2_space_resv *sr, |
| 1482 | int change_size) |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1483 | { |
| 1484 | int ret; |
| 1485 | s64 llen; |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1486 | loff_t size; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1487 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1488 | struct buffer_head *di_bh = NULL; |
| 1489 | handle_t *handle; |
Mark Fasheh | a00cce3 | 2007-07-20 11:28:30 -0700 | [diff] [blame] | 1490 | unsigned long long max_off = inode->i_sb->s_maxbytes; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1491 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1492 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) |
| 1493 | return -EROFS; |
| 1494 | |
| 1495 | mutex_lock(&inode->i_mutex); |
| 1496 | |
| 1497 | /* |
| 1498 | * This prevents concurrent writes on other nodes |
| 1499 | */ |
| 1500 | ret = ocfs2_rw_lock(inode, 1); |
| 1501 | if (ret) { |
| 1502 | mlog_errno(ret); |
| 1503 | goto out; |
| 1504 | } |
| 1505 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1506 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1507 | if (ret) { |
| 1508 | mlog_errno(ret); |
| 1509 | goto out_rw_unlock; |
| 1510 | } |
| 1511 | |
| 1512 | if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) { |
| 1513 | ret = -EPERM; |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1514 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1515 | } |
| 1516 | |
| 1517 | switch (sr->l_whence) { |
| 1518 | case 0: /*SEEK_SET*/ |
| 1519 | break; |
| 1520 | case 1: /*SEEK_CUR*/ |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1521 | sr->l_start += f_pos; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1522 | break; |
| 1523 | case 2: /*SEEK_END*/ |
| 1524 | sr->l_start += i_size_read(inode); |
| 1525 | break; |
| 1526 | default: |
| 1527 | ret = -EINVAL; |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1528 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1529 | } |
| 1530 | sr->l_whence = 0; |
| 1531 | |
| 1532 | llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len; |
| 1533 | |
| 1534 | if (sr->l_start < 0 |
| 1535 | || sr->l_start > max_off |
| 1536 | || (sr->l_start + llen) < 0 |
| 1537 | || (sr->l_start + llen) > max_off) { |
| 1538 | ret = -EINVAL; |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1539 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1540 | } |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1541 | size = sr->l_start + sr->l_len; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1542 | |
| 1543 | if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) { |
| 1544 | if (sr->l_len <= 0) { |
| 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 | } |
| 1548 | } |
| 1549 | |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1550 | if (file && should_remove_suid(file->f_path.dentry)) { |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1551 | ret = __ocfs2_write_remove_suid(inode, di_bh); |
| 1552 | if (ret) { |
| 1553 | mlog_errno(ret); |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1554 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1555 | } |
| 1556 | } |
| 1557 | |
| 1558 | down_write(&OCFS2_I(inode)->ip_alloc_sem); |
| 1559 | switch (cmd) { |
| 1560 | case OCFS2_IOC_RESVSP: |
| 1561 | case OCFS2_IOC_RESVSP64: |
| 1562 | /* |
| 1563 | * This takes unsigned offsets, but the signed ones we |
| 1564 | * pass have been checked against overflow above. |
| 1565 | */ |
| 1566 | ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start, |
| 1567 | sr->l_len); |
| 1568 | break; |
| 1569 | case OCFS2_IOC_UNRESVSP: |
| 1570 | case OCFS2_IOC_UNRESVSP64: |
| 1571 | ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start, |
| 1572 | sr->l_len); |
| 1573 | break; |
| 1574 | default: |
| 1575 | ret = -EINVAL; |
| 1576 | } |
| 1577 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
| 1578 | if (ret) { |
| 1579 | mlog_errno(ret); |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1580 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1581 | } |
| 1582 | |
| 1583 | /* |
| 1584 | * We update c/mtime for these changes |
| 1585 | */ |
| 1586 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
| 1587 | if (IS_ERR(handle)) { |
| 1588 | ret = PTR_ERR(handle); |
| 1589 | mlog_errno(ret); |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1590 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1591 | } |
| 1592 | |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1593 | if (change_size && i_size_read(inode) < size) |
| 1594 | i_size_write(inode, size); |
| 1595 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1596 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
| 1597 | ret = ocfs2_mark_inode_dirty(handle, inode, di_bh); |
| 1598 | if (ret < 0) |
| 1599 | mlog_errno(ret); |
| 1600 | |
| 1601 | ocfs2_commit_trans(osb, handle); |
| 1602 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1603 | out_inode_unlock: |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1604 | brelse(di_bh); |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1605 | ocfs2_inode_unlock(inode, 1); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1606 | out_rw_unlock: |
| 1607 | ocfs2_rw_unlock(inode, 1); |
| 1608 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1609 | out: |
Julia Lawall | c259ae5 | 2008-07-21 09:59:15 +0200 | [diff] [blame] | 1610 | mutex_unlock(&inode->i_mutex); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1611 | return ret; |
| 1612 | } |
| 1613 | |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1614 | int ocfs2_change_file_space(struct file *file, unsigned int cmd, |
| 1615 | struct ocfs2_space_resv *sr) |
| 1616 | { |
| 1617 | struct inode *inode = file->f_path.dentry->d_inode; |
Fernando Carrijo | c19a28e | 2009-01-07 18:09:08 -0800 | [diff] [blame] | 1618 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1619 | |
| 1620 | if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) && |
| 1621 | !ocfs2_writes_unwritten_extents(osb)) |
| 1622 | return -ENOTTY; |
| 1623 | else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) && |
| 1624 | !ocfs2_sparse_alloc(osb)) |
| 1625 | return -ENOTTY; |
| 1626 | |
| 1627 | if (!S_ISREG(inode->i_mode)) |
| 1628 | return -EINVAL; |
| 1629 | |
| 1630 | if (!(file->f_mode & FMODE_WRITE)) |
| 1631 | return -EBADF; |
| 1632 | |
| 1633 | return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0); |
| 1634 | } |
| 1635 | |
| 1636 | static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset, |
| 1637 | loff_t len) |
| 1638 | { |
| 1639 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1640 | struct ocfs2_space_resv sr; |
| 1641 | int change_size = 1; |
| 1642 | |
| 1643 | if (!ocfs2_writes_unwritten_extents(osb)) |
| 1644 | return -EOPNOTSUPP; |
| 1645 | |
| 1646 | if (S_ISDIR(inode->i_mode)) |
| 1647 | return -ENODEV; |
| 1648 | |
| 1649 | if (mode & FALLOC_FL_KEEP_SIZE) |
| 1650 | change_size = 0; |
| 1651 | |
| 1652 | sr.l_whence = 0; |
| 1653 | sr.l_start = (s64)offset; |
| 1654 | sr.l_len = (s64)len; |
| 1655 | |
| 1656 | return __ocfs2_change_file_space(NULL, inode, offset, |
| 1657 | OCFS2_IOC_RESVSP64, &sr, change_size); |
| 1658 | } |
| 1659 | |
Tao Ma | 293b2f7 | 2009-08-25 08:02:48 +0800 | [diff] [blame^] | 1660 | int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos, |
| 1661 | size_t count) |
| 1662 | { |
| 1663 | int ret = 0; |
| 1664 | unsigned int extent_flags; |
| 1665 | u32 cpos, clusters, extent_len, phys_cpos; |
| 1666 | struct super_block *sb = inode->i_sb; |
| 1667 | |
| 1668 | if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) || |
| 1669 | !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL)) |
| 1670 | return 0; |
| 1671 | |
| 1672 | cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits; |
| 1673 | clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos; |
| 1674 | |
| 1675 | while (clusters) { |
| 1676 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len, |
| 1677 | &extent_flags); |
| 1678 | if (ret < 0) { |
| 1679 | mlog_errno(ret); |
| 1680 | goto out; |
| 1681 | } |
| 1682 | |
| 1683 | if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) { |
| 1684 | ret = 1; |
| 1685 | break; |
| 1686 | } |
| 1687 | |
| 1688 | if (extent_len > clusters) |
| 1689 | extent_len = clusters; |
| 1690 | |
| 1691 | clusters -= extent_len; |
| 1692 | cpos += extent_len; |
| 1693 | } |
| 1694 | out: |
| 1695 | return ret; |
| 1696 | } |
| 1697 | |
| 1698 | static int ocfs2_prepare_inode_for_refcount(struct inode *inode, |
| 1699 | loff_t pos, size_t count, |
| 1700 | int *meta_level) |
| 1701 | { |
| 1702 | int ret; |
| 1703 | struct buffer_head *di_bh = NULL; |
| 1704 | u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits; |
| 1705 | u32 clusters = |
| 1706 | ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos; |
| 1707 | |
| 1708 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
| 1709 | if (ret) { |
| 1710 | mlog_errno(ret); |
| 1711 | goto out; |
| 1712 | } |
| 1713 | |
| 1714 | *meta_level = 1; |
| 1715 | |
| 1716 | ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters); |
| 1717 | if (ret) |
| 1718 | mlog_errno(ret); |
| 1719 | out: |
| 1720 | brelse(di_bh); |
| 1721 | return ret; |
| 1722 | } |
| 1723 | |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1724 | static int ocfs2_prepare_inode_for_write(struct dentry *dentry, |
| 1725 | loff_t *ppos, |
| 1726 | size_t count, |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1727 | int appending, |
| 1728 | int *direct_io) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1729 | { |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1730 | int ret = 0, meta_level = 0; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1731 | struct inode *inode = dentry->d_inode; |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1732 | loff_t saved_pos, end; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1733 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1734 | /* |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1735 | * We start with a read level meta lock and only jump to an ex |
| 1736 | * if we need to make modifications here. |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1737 | */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1738 | for(;;) { |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1739 | ret = ocfs2_inode_lock(inode, NULL, meta_level); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1740 | if (ret < 0) { |
| 1741 | meta_level = -1; |
| 1742 | mlog_errno(ret); |
| 1743 | goto out; |
| 1744 | } |
| 1745 | |
| 1746 | /* Clear suid / sgid if necessary. We do this here |
| 1747 | * instead of later in the write path because |
| 1748 | * remove_suid() calls ->setattr without any hint that |
| 1749 | * we may have already done our cluster locking. Since |
| 1750 | * ocfs2_setattr() *must* take cluster locks to |
| 1751 | * proceeed, this will lead us to recursively lock the |
| 1752 | * inode. There's also the dinode i_size state which |
| 1753 | * can be lost via setattr during extending writes (we |
| 1754 | * set inode->i_size at the end of a write. */ |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1755 | if (should_remove_suid(dentry)) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1756 | if (meta_level == 0) { |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1757 | ocfs2_inode_unlock(inode, meta_level); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1758 | meta_level = 1; |
| 1759 | continue; |
| 1760 | } |
| 1761 | |
| 1762 | ret = ocfs2_write_remove_suid(inode); |
| 1763 | if (ret < 0) { |
| 1764 | mlog_errno(ret); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1765 | goto out_unlock; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1766 | } |
| 1767 | } |
| 1768 | |
| 1769 | /* work on a copy of ppos until we're sure that we won't have |
| 1770 | * to recalculate it due to relocking. */ |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1771 | if (appending) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1772 | saved_pos = i_size_read(inode); |
| 1773 | mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos); |
| 1774 | } else { |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1775 | saved_pos = *ppos; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1776 | } |
Mark Fasheh | 3a0782d | 2007-01-17 12:53:31 -0800 | [diff] [blame] | 1777 | |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1778 | end = saved_pos + count; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1779 | |
Tao Ma | 293b2f7 | 2009-08-25 08:02:48 +0800 | [diff] [blame^] | 1780 | ret = ocfs2_check_range_for_refcount(inode, saved_pos, count); |
| 1781 | if (ret == 1) { |
| 1782 | ocfs2_inode_unlock(inode, meta_level); |
| 1783 | meta_level = -1; |
| 1784 | |
| 1785 | ret = ocfs2_prepare_inode_for_refcount(inode, |
| 1786 | saved_pos, |
| 1787 | count, |
| 1788 | &meta_level); |
| 1789 | } |
| 1790 | |
| 1791 | if (ret < 0) { |
| 1792 | mlog_errno(ret); |
| 1793 | goto out_unlock; |
| 1794 | } |
| 1795 | |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1796 | /* |
| 1797 | * Skip the O_DIRECT checks if we don't need |
| 1798 | * them. |
| 1799 | */ |
| 1800 | if (!direct_io || !(*direct_io)) |
| 1801 | break; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1802 | |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1803 | /* |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1804 | * There's no sane way to do direct writes to an inode |
| 1805 | * with inline data. |
| 1806 | */ |
| 1807 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
| 1808 | *direct_io = 0; |
| 1809 | break; |
| 1810 | } |
| 1811 | |
| 1812 | /* |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1813 | * Allowing concurrent direct writes means |
| 1814 | * i_size changes wouldn't be synchronized, so |
| 1815 | * one node could wind up truncating another |
| 1816 | * nodes writes. |
| 1817 | */ |
| 1818 | if (end > i_size_read(inode)) { |
| 1819 | *direct_io = 0; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1820 | break; |
| 1821 | } |
| 1822 | |
Mark Fasheh | 3a0782d | 2007-01-17 12:53:31 -0800 | [diff] [blame] | 1823 | /* |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1824 | * We don't fill holes during direct io, so |
| 1825 | * check for them here. If any are found, the |
| 1826 | * caller will have to retake some cluster |
| 1827 | * locks and initiate the io as buffered. |
Mark Fasheh | 3a0782d | 2007-01-17 12:53:31 -0800 | [diff] [blame] | 1828 | */ |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1829 | ret = ocfs2_check_range_for_holes(inode, saved_pos, count); |
| 1830 | if (ret == 1) { |
| 1831 | *direct_io = 0; |
| 1832 | ret = 0; |
| 1833 | } else if (ret < 0) |
| 1834 | mlog_errno(ret); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1835 | break; |
| 1836 | } |
| 1837 | |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1838 | if (appending) |
| 1839 | *ppos = saved_pos; |
| 1840 | |
| 1841 | out_unlock: |
Tao Ma | 293b2f7 | 2009-08-25 08:02:48 +0800 | [diff] [blame^] | 1842 | if (meta_level >= 0) |
| 1843 | ocfs2_inode_unlock(inode, meta_level); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1844 | |
| 1845 | out: |
| 1846 | return ret; |
| 1847 | } |
| 1848 | |
| 1849 | static ssize_t ocfs2_file_aio_write(struct kiocb *iocb, |
| 1850 | const struct iovec *iov, |
| 1851 | unsigned long nr_segs, |
| 1852 | loff_t pos) |
| 1853 | { |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1854 | int ret, direct_io, appending, rw_level, have_alloc_sem = 0; |
Nick Piggin | b6af1bc | 2007-10-16 01:25:24 -0700 | [diff] [blame] | 1855 | int can_do_direct; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1856 | ssize_t written = 0; |
| 1857 | size_t ocount; /* original count */ |
| 1858 | size_t count; /* after file limit checks */ |
Mark Fasheh | 9ea2d32 | 2007-10-18 14:14:45 -0700 | [diff] [blame] | 1859 | loff_t old_size, *ppos = &iocb->ki_pos; |
| 1860 | u32 old_clusters; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1861 | struct file *file = iocb->ki_filp; |
| 1862 | struct inode *inode = file->f_path.dentry->d_inode; |
Mark Fasheh | 9ea2d32 | 2007-10-18 14:14:45 -0700 | [diff] [blame] | 1863 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1864 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1865 | mlog_entry("(0x%p, %u, '%.*s')\n", file, |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1866 | (unsigned int)nr_segs, |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1867 | file->f_path.dentry->d_name.len, |
| 1868 | file->f_path.dentry->d_name.name); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1869 | |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1870 | if (iocb->ki_left == 0) |
| 1871 | return 0; |
| 1872 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1873 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); |
| 1874 | |
| 1875 | appending = file->f_flags & O_APPEND ? 1 : 0; |
| 1876 | direct_io = file->f_flags & O_DIRECT ? 1 : 0; |
| 1877 | |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1878 | mutex_lock(&inode->i_mutex); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1879 | |
| 1880 | relock: |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1881 | /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */ |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1882 | if (direct_io) { |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1883 | down_read(&inode->i_alloc_sem); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1884 | have_alloc_sem = 1; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1885 | } |
| 1886 | |
| 1887 | /* concurrent O_DIRECT writes are allowed */ |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1888 | rw_level = !direct_io; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1889 | ret = ocfs2_rw_lock(inode, rw_level); |
| 1890 | if (ret < 0) { |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1891 | mlog_errno(ret); |
| 1892 | goto out_sems; |
| 1893 | } |
| 1894 | |
| 1895 | can_do_direct = direct_io; |
| 1896 | ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos, |
| 1897 | iocb->ki_left, appending, |
| 1898 | &can_do_direct); |
| 1899 | if (ret < 0) { |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1900 | mlog_errno(ret); |
| 1901 | goto out; |
| 1902 | } |
| 1903 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1904 | /* |
| 1905 | * We can't complete the direct I/O as requested, fall back to |
| 1906 | * buffered I/O. |
| 1907 | */ |
| 1908 | if (direct_io && !can_do_direct) { |
| 1909 | ocfs2_rw_unlock(inode, rw_level); |
| 1910 | up_read(&inode->i_alloc_sem); |
| 1911 | |
| 1912 | have_alloc_sem = 0; |
| 1913 | rw_level = -1; |
| 1914 | |
| 1915 | direct_io = 0; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1916 | goto relock; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1917 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1918 | |
Mark Fasheh | 9ea2d32 | 2007-10-18 14:14:45 -0700 | [diff] [blame] | 1919 | /* |
| 1920 | * To later detect whether a journal commit for sync writes is |
| 1921 | * necessary, we sample i_size, and cluster count here. |
| 1922 | */ |
| 1923 | old_size = i_size_read(inode); |
| 1924 | old_clusters = OCFS2_I(inode)->ip_clusters; |
| 1925 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1926 | /* communicate with ocfs2_dio_end_io */ |
Mark Fasheh | 7cdfc3a | 2007-04-16 17:28:51 -0700 | [diff] [blame] | 1927 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1928 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1929 | if (direct_io) { |
Nick Piggin | b6af1bc | 2007-10-16 01:25:24 -0700 | [diff] [blame] | 1930 | ret = generic_segment_checks(iov, &nr_segs, &ocount, |
| 1931 | VERIFY_READ); |
| 1932 | if (ret) |
| 1933 | goto out_dio; |
| 1934 | |
Goldwyn Rodrigues | cefcb80 | 2009-07-11 10:57:27 -0500 | [diff] [blame] | 1935 | count = ocount; |
Nick Piggin | b6af1bc | 2007-10-16 01:25:24 -0700 | [diff] [blame] | 1936 | ret = generic_write_checks(file, ppos, &count, |
| 1937 | S_ISBLK(inode->i_mode)); |
| 1938 | if (ret) |
| 1939 | goto out_dio; |
| 1940 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1941 | written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos, |
| 1942 | ppos, count, ocount); |
| 1943 | if (written < 0) { |
Dmitri Monakhov | c435400 | 2008-10-27 13:01:49 -0700 | [diff] [blame] | 1944 | /* |
| 1945 | * direct write may have instantiated a few |
| 1946 | * blocks outside i_size. Trim these off again. |
| 1947 | * Don't need i_size_read because we hold i_mutex. |
| 1948 | */ |
| 1949 | if (*ppos + count > inode->i_size) |
| 1950 | vmtruncate(inode, inode->i_size); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1951 | ret = written; |
| 1952 | goto out_dio; |
| 1953 | } |
| 1954 | } else { |
Nick Piggin | b6af1bc | 2007-10-16 01:25:24 -0700 | [diff] [blame] | 1955 | written = generic_file_aio_write_nolock(iocb, iov, nr_segs, |
| 1956 | *ppos); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1957 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1958 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1959 | out_dio: |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1960 | /* buffered aio wouldn't have proper lock coverage today */ |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1961 | BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1962 | |
Mark Fasheh | 9ea2d32 | 2007-10-18 14:14:45 -0700 | [diff] [blame] | 1963 | if ((file->f_flags & O_SYNC && !direct_io) || IS_SYNC(inode)) { |
| 1964 | /* |
| 1965 | * The generic write paths have handled getting data |
| 1966 | * to disk, but since we don't make use of the dirty |
| 1967 | * inode list, a manual journal commit is necessary |
| 1968 | * here. |
| 1969 | */ |
| 1970 | if (old_size != i_size_read(inode) || |
| 1971 | old_clusters != OCFS2_I(inode)->ip_clusters) { |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 1972 | ret = jbd2_journal_force_commit(osb->journal->j_journal); |
Mark Fasheh | 9ea2d32 | 2007-10-18 14:14:45 -0700 | [diff] [blame] | 1973 | if (ret < 0) |
| 1974 | written = ret; |
| 1975 | } |
| 1976 | } |
| 1977 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1978 | /* |
| 1979 | * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io |
| 1980 | * function pointer which is called when o_direct io completes so that |
| 1981 | * it can unlock our rw lock. (it's the clustered equivalent of |
| 1982 | * i_alloc_sem; protects truncate from racing with pending ios). |
| 1983 | * Unfortunately there are error cases which call end_io and others |
| 1984 | * that don't. so we don't have to unlock the rw_lock if either an |
| 1985 | * async dio is going to do it in the future or an end_io after an |
| 1986 | * error has already done it. |
| 1987 | */ |
| 1988 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { |
| 1989 | rw_level = -1; |
| 1990 | have_alloc_sem = 0; |
| 1991 | } |
| 1992 | |
| 1993 | out: |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1994 | if (rw_level != -1) |
| 1995 | ocfs2_rw_unlock(inode, rw_level); |
| 1996 | |
| 1997 | out_sems: |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1998 | if (have_alloc_sem) |
| 1999 | up_read(&inode->i_alloc_sem); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 2000 | |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 2001 | mutex_unlock(&inode->i_mutex); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2002 | |
Wengang Wang | 812e7a6 | 2009-07-10 13:26:04 +0800 | [diff] [blame] | 2003 | if (written) |
| 2004 | ret = written; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2005 | mlog_exit(ret); |
Wengang Wang | 812e7a6 | 2009-07-10 13:26:04 +0800 | [diff] [blame] | 2006 | return ret; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2007 | } |
| 2008 | |
Miklos Szeredi | 328eaab | 2009-04-14 19:48:39 +0200 | [diff] [blame] | 2009 | static int ocfs2_splice_to_file(struct pipe_inode_info *pipe, |
| 2010 | struct file *out, |
| 2011 | struct splice_desc *sd) |
| 2012 | { |
| 2013 | int ret; |
| 2014 | |
| 2015 | ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, &sd->pos, |
| 2016 | sd->total_len, 0, NULL); |
| 2017 | if (ret < 0) { |
| 2018 | mlog_errno(ret); |
| 2019 | return ret; |
| 2020 | } |
| 2021 | |
| 2022 | return splice_from_pipe_feed(pipe, sd, pipe_to_file); |
| 2023 | } |
| 2024 | |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 2025 | static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe, |
| 2026 | struct file *out, |
| 2027 | loff_t *ppos, |
| 2028 | size_t len, |
| 2029 | unsigned int flags) |
| 2030 | { |
| 2031 | int ret; |
Miklos Szeredi | 328eaab | 2009-04-14 19:48:39 +0200 | [diff] [blame] | 2032 | struct address_space *mapping = out->f_mapping; |
| 2033 | struct inode *inode = mapping->host; |
| 2034 | struct splice_desc sd = { |
| 2035 | .total_len = len, |
| 2036 | .flags = flags, |
| 2037 | .pos = *ppos, |
| 2038 | .u.file = out, |
| 2039 | }; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 2040 | |
| 2041 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe, |
| 2042 | (unsigned int)len, |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 2043 | out->f_path.dentry->d_name.len, |
| 2044 | out->f_path.dentry->d_name.name); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 2045 | |
Miklos Szeredi | 7bfac9e | 2009-04-06 17:41:00 +0200 | [diff] [blame] | 2046 | if (pipe->inode) |
Miklos Szeredi | 328eaab | 2009-04-14 19:48:39 +0200 | [diff] [blame] | 2047 | mutex_lock_nested(&pipe->inode->i_mutex, I_MUTEX_PARENT); |
| 2048 | |
| 2049 | splice_from_pipe_begin(&sd); |
| 2050 | do { |
| 2051 | ret = splice_from_pipe_next(pipe, &sd); |
| 2052 | if (ret <= 0) |
| 2053 | break; |
| 2054 | |
| 2055 | mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD); |
| 2056 | ret = ocfs2_rw_lock(inode, 1); |
| 2057 | if (ret < 0) |
| 2058 | mlog_errno(ret); |
| 2059 | else { |
| 2060 | ret = ocfs2_splice_to_file(pipe, out, &sd); |
| 2061 | ocfs2_rw_unlock(inode, 1); |
| 2062 | } |
| 2063 | mutex_unlock(&inode->i_mutex); |
| 2064 | } while (ret > 0); |
| 2065 | splice_from_pipe_end(pipe, &sd); |
| 2066 | |
Miklos Szeredi | 7bfac9e | 2009-04-06 17:41:00 +0200 | [diff] [blame] | 2067 | if (pipe->inode) |
| 2068 | mutex_unlock(&pipe->inode->i_mutex); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 2069 | |
Miklos Szeredi | 328eaab | 2009-04-14 19:48:39 +0200 | [diff] [blame] | 2070 | if (sd.num_spliced) |
| 2071 | ret = sd.num_spliced; |
| 2072 | |
| 2073 | if (ret > 0) { |
| 2074 | unsigned long nr_pages; |
| 2075 | |
| 2076 | *ppos += ret; |
| 2077 | nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
| 2078 | |
| 2079 | /* |
| 2080 | * If file or inode is SYNC and we actually wrote some data, |
| 2081 | * sync it. |
| 2082 | */ |
| 2083 | if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) { |
| 2084 | int err; |
| 2085 | |
| 2086 | mutex_lock(&inode->i_mutex); |
| 2087 | err = ocfs2_rw_lock(inode, 1); |
| 2088 | if (err < 0) { |
| 2089 | mlog_errno(err); |
| 2090 | } else { |
| 2091 | err = generic_osync_inode(inode, mapping, |
| 2092 | OSYNC_METADATA|OSYNC_DATA); |
| 2093 | ocfs2_rw_unlock(inode, 1); |
| 2094 | } |
| 2095 | mutex_unlock(&inode->i_mutex); |
| 2096 | |
| 2097 | if (err) |
| 2098 | ret = err; |
| 2099 | } |
| 2100 | balance_dirty_pages_ratelimited_nr(mapping, nr_pages); |
| 2101 | } |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 2102 | |
| 2103 | mlog_exit(ret); |
| 2104 | return ret; |
| 2105 | } |
| 2106 | |
| 2107 | static ssize_t ocfs2_file_splice_read(struct file *in, |
| 2108 | loff_t *ppos, |
| 2109 | struct pipe_inode_info *pipe, |
| 2110 | size_t len, |
| 2111 | unsigned int flags) |
| 2112 | { |
Tao Ma | 1962f39 | 2009-06-19 15:36:52 +0800 | [diff] [blame] | 2113 | int ret = 0, lock_level = 0; |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 2114 | struct inode *inode = in->f_path.dentry->d_inode; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 2115 | |
| 2116 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe, |
| 2117 | (unsigned int)len, |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 2118 | in->f_path.dentry->d_name.len, |
| 2119 | in->f_path.dentry->d_name.name); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 2120 | |
| 2121 | /* |
| 2122 | * See the comment in ocfs2_file_aio_read() |
| 2123 | */ |
Tao Ma | 1962f39 | 2009-06-19 15:36:52 +0800 | [diff] [blame] | 2124 | ret = ocfs2_inode_lock_atime(inode, in->f_vfsmnt, &lock_level); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 2125 | if (ret < 0) { |
| 2126 | mlog_errno(ret); |
| 2127 | goto bail; |
| 2128 | } |
Tao Ma | 1962f39 | 2009-06-19 15:36:52 +0800 | [diff] [blame] | 2129 | ocfs2_inode_unlock(inode, lock_level); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 2130 | |
| 2131 | ret = generic_file_splice_read(in, ppos, pipe, len, flags); |
| 2132 | |
| 2133 | bail: |
| 2134 | mlog_exit(ret); |
| 2135 | return ret; |
| 2136 | } |
| 2137 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2138 | static ssize_t ocfs2_file_aio_read(struct kiocb *iocb, |
Badari Pulavarty | 027445c | 2006-09-30 23:28:46 -0700 | [diff] [blame] | 2139 | const struct iovec *iov, |
| 2140 | unsigned long nr_segs, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2141 | loff_t pos) |
| 2142 | { |
Tiger Yang | 25899de | 2006-11-15 15:49:02 +0800 | [diff] [blame] | 2143 | 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] | 2144 | struct file *filp = iocb->ki_filp; |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 2145 | struct inode *inode = filp->f_path.dentry->d_inode; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2146 | |
Badari Pulavarty | 027445c | 2006-09-30 23:28:46 -0700 | [diff] [blame] | 2147 | mlog_entry("(0x%p, %u, '%.*s')\n", filp, |
| 2148 | (unsigned int)nr_segs, |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 2149 | filp->f_path.dentry->d_name.len, |
| 2150 | filp->f_path.dentry->d_name.name); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2151 | |
| 2152 | if (!inode) { |
| 2153 | ret = -EINVAL; |
| 2154 | mlog_errno(ret); |
| 2155 | goto bail; |
| 2156 | } |
| 2157 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2158 | /* |
| 2159 | * buffered reads protect themselves in ->readpage(). O_DIRECT reads |
| 2160 | * need locks to protect pending reads from racing with truncate. |
| 2161 | */ |
| 2162 | if (filp->f_flags & O_DIRECT) { |
| 2163 | down_read(&inode->i_alloc_sem); |
| 2164 | have_alloc_sem = 1; |
| 2165 | |
| 2166 | ret = ocfs2_rw_lock(inode, 0); |
| 2167 | if (ret < 0) { |
| 2168 | mlog_errno(ret); |
| 2169 | goto bail; |
| 2170 | } |
| 2171 | rw_level = 0; |
| 2172 | /* communicate with ocfs2_dio_end_io */ |
Mark Fasheh | 7cdfc3a | 2007-04-16 17:28:51 -0700 | [diff] [blame] | 2173 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2174 | } |
| 2175 | |
Mark Fasheh | c4374f8 | 2006-05-05 19:04:35 -0700 | [diff] [blame] | 2176 | /* |
| 2177 | * We're fine letting folks race truncates and extending |
| 2178 | * writes with read across the cluster, just like they can |
| 2179 | * locally. Hence no rw_lock during read. |
| 2180 | * |
| 2181 | * Take and drop the meta data lock to update inode fields |
| 2182 | * like i_size. This allows the checks down below |
| 2183 | * generic_file_aio_read() a chance of actually working. |
| 2184 | */ |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 2185 | ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level); |
Mark Fasheh | c4374f8 | 2006-05-05 19:04:35 -0700 | [diff] [blame] | 2186 | if (ret < 0) { |
| 2187 | mlog_errno(ret); |
| 2188 | goto bail; |
| 2189 | } |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 2190 | ocfs2_inode_unlock(inode, lock_level); |
Mark Fasheh | c4374f8 | 2006-05-05 19:04:35 -0700 | [diff] [blame] | 2191 | |
Badari Pulavarty | 027445c | 2006-09-30 23:28:46 -0700 | [diff] [blame] | 2192 | ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2193 | if (ret == -EINVAL) |
Sunil Mushran | 56753bd | 2008-06-09 11:24:41 -0700 | [diff] [blame] | 2194 | mlog(0, "generic_file_aio_read returned -EINVAL\n"); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2195 | |
| 2196 | /* buffered aio wouldn't have proper lock coverage today */ |
| 2197 | BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT)); |
| 2198 | |
| 2199 | /* see ocfs2_file_aio_write */ |
| 2200 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { |
| 2201 | rw_level = -1; |
| 2202 | have_alloc_sem = 0; |
| 2203 | } |
| 2204 | |
| 2205 | bail: |
| 2206 | if (have_alloc_sem) |
| 2207 | up_read(&inode->i_alloc_sem); |
| 2208 | if (rw_level != -1) |
| 2209 | ocfs2_rw_unlock(inode, rw_level); |
| 2210 | mlog_exit(ret); |
| 2211 | |
| 2212 | return ret; |
| 2213 | } |
| 2214 | |
Arjan van de Ven | 92e1d5b | 2007-02-12 00:55:39 -0800 | [diff] [blame] | 2215 | const struct inode_operations ocfs2_file_iops = { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2216 | .setattr = ocfs2_setattr, |
| 2217 | .getattr = ocfs2_getattr, |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 2218 | .permission = ocfs2_permission, |
Tiger Yang | cf1d6c7 | 2008-08-18 17:11:00 +0800 | [diff] [blame] | 2219 | .setxattr = generic_setxattr, |
| 2220 | .getxattr = generic_getxattr, |
| 2221 | .listxattr = ocfs2_listxattr, |
| 2222 | .removexattr = generic_removexattr, |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 2223 | .fallocate = ocfs2_fallocate, |
Mark Fasheh | 00dc417 | 2008-10-03 17:32:11 -0400 | [diff] [blame] | 2224 | .fiemap = ocfs2_fiemap, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2225 | }; |
| 2226 | |
Arjan van de Ven | 92e1d5b | 2007-02-12 00:55:39 -0800 | [diff] [blame] | 2227 | const struct inode_operations ocfs2_special_file_iops = { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2228 | .setattr = ocfs2_setattr, |
| 2229 | .getattr = ocfs2_getattr, |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 2230 | .permission = ocfs2_permission, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2231 | }; |
| 2232 | |
Mark Fasheh | 53da493 | 2008-07-21 14:29:16 -0700 | [diff] [blame] | 2233 | /* |
| 2234 | * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with |
| 2235 | * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks! |
| 2236 | */ |
Arjan van de Ven | 4b6f5d2 | 2006-03-28 01:56:42 -0800 | [diff] [blame] | 2237 | const struct file_operations ocfs2_fops = { |
Jan Kara | 32c3c0e | 2007-12-19 15:24:52 +0100 | [diff] [blame] | 2238 | .llseek = generic_file_llseek, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2239 | .read = do_sync_read, |
| 2240 | .write = do_sync_write, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2241 | .mmap = ocfs2_mmap, |
| 2242 | .fsync = ocfs2_sync_file, |
| 2243 | .release = ocfs2_file_release, |
| 2244 | .open = ocfs2_file_open, |
| 2245 | .aio_read = ocfs2_file_aio_read, |
| 2246 | .aio_write = ocfs2_file_aio_write, |
Andi Kleen | c9ec148 | 2008-01-27 03:17:17 +0100 | [diff] [blame] | 2247 | .unlocked_ioctl = ocfs2_ioctl, |
Mark Fasheh | 586d232 | 2007-03-09 15:56:28 -0800 | [diff] [blame] | 2248 | #ifdef CONFIG_COMPAT |
| 2249 | .compat_ioctl = ocfs2_compat_ioctl, |
| 2250 | #endif |
Mark Fasheh | 53da493 | 2008-07-21 14:29:16 -0700 | [diff] [blame] | 2251 | .lock = ocfs2_lock, |
| 2252 | .flock = ocfs2_flock, |
| 2253 | .splice_read = ocfs2_file_splice_read, |
| 2254 | .splice_write = ocfs2_file_splice_write, |
| 2255 | }; |
| 2256 | |
| 2257 | const struct file_operations ocfs2_dops = { |
| 2258 | .llseek = generic_file_llseek, |
| 2259 | .read = generic_read_dir, |
| 2260 | .readdir = ocfs2_readdir, |
| 2261 | .fsync = ocfs2_sync_file, |
| 2262 | .release = ocfs2_dir_release, |
| 2263 | .open = ocfs2_dir_open, |
| 2264 | .unlocked_ioctl = ocfs2_ioctl, |
| 2265 | #ifdef CONFIG_COMPAT |
| 2266 | .compat_ioctl = ocfs2_compat_ioctl, |
| 2267 | #endif |
| 2268 | .lock = ocfs2_lock, |
| 2269 | .flock = ocfs2_flock, |
| 2270 | }; |
| 2271 | |
| 2272 | /* |
| 2273 | * POSIX-lockless variants of our file_operations. |
| 2274 | * |
| 2275 | * These will be used if the underlying cluster stack does not support |
| 2276 | * posix file locking, if the user passes the "localflocks" mount |
| 2277 | * option, or if we have a local-only fs. |
| 2278 | * |
| 2279 | * ocfs2_flock is in here because all stacks handle UNIX file locks, |
| 2280 | * so we still want it in the case of no stack support for |
| 2281 | * plocks. Internally, it will do the right thing when asked to ignore |
| 2282 | * the cluster. |
| 2283 | */ |
| 2284 | const struct file_operations ocfs2_fops_no_plocks = { |
| 2285 | .llseek = generic_file_llseek, |
| 2286 | .read = do_sync_read, |
| 2287 | .write = do_sync_write, |
| 2288 | .mmap = ocfs2_mmap, |
| 2289 | .fsync = ocfs2_sync_file, |
| 2290 | .release = ocfs2_file_release, |
| 2291 | .open = ocfs2_file_open, |
| 2292 | .aio_read = ocfs2_file_aio_read, |
| 2293 | .aio_write = ocfs2_file_aio_write, |
| 2294 | .unlocked_ioctl = ocfs2_ioctl, |
| 2295 | #ifdef CONFIG_COMPAT |
| 2296 | .compat_ioctl = ocfs2_compat_ioctl, |
| 2297 | #endif |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 2298 | .flock = ocfs2_flock, |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 2299 | .splice_read = ocfs2_file_splice_read, |
| 2300 | .splice_write = ocfs2_file_splice_write, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2301 | }; |
| 2302 | |
Mark Fasheh | 53da493 | 2008-07-21 14:29:16 -0700 | [diff] [blame] | 2303 | const struct file_operations ocfs2_dops_no_plocks = { |
Jan Kara | 32c3c0e | 2007-12-19 15:24:52 +0100 | [diff] [blame] | 2304 | .llseek = generic_file_llseek, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2305 | .read = generic_read_dir, |
| 2306 | .readdir = ocfs2_readdir, |
| 2307 | .fsync = ocfs2_sync_file, |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 2308 | .release = ocfs2_dir_release, |
| 2309 | .open = ocfs2_dir_open, |
Andi Kleen | c9ec148 | 2008-01-27 03:17:17 +0100 | [diff] [blame] | 2310 | .unlocked_ioctl = ocfs2_ioctl, |
Mark Fasheh | 586d232 | 2007-03-09 15:56:28 -0800 | [diff] [blame] | 2311 | #ifdef CONFIG_COMPAT |
| 2312 | .compat_ioctl = ocfs2_compat_ioctl, |
| 2313 | #endif |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 2314 | .flock = ocfs2_flock, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2315 | }; |