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 | |
Christoph Hellwig | 907f455 | 2010-03-03 09:05:06 -0500 | [diff] [blame] | 110 | if (file->f_mode & FMODE_WRITE) |
Christoph Hellwig | 871a293 | 2010-03-03 09:05:07 -0500 | [diff] [blame] | 111 | dquot_initialize(inode); |
Christoph Hellwig | 907f455 | 2010-03-03 09:05:06 -0500 | [diff] [blame] | 112 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 113 | spin_lock(&oi->ip_lock); |
| 114 | |
| 115 | /* Check that the inode hasn't been wiped from disk by another |
| 116 | * node. If it hasn't then we're safe as long as we hold the |
| 117 | * spin lock until our increment of open count. */ |
| 118 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { |
| 119 | spin_unlock(&oi->ip_lock); |
| 120 | |
| 121 | status = -ENOENT; |
| 122 | goto leave; |
| 123 | } |
| 124 | |
| 125 | if (mode & O_DIRECT) |
| 126 | oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT; |
| 127 | |
| 128 | oi->ip_open_count++; |
| 129 | spin_unlock(&oi->ip_lock); |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 130 | |
| 131 | status = ocfs2_init_file_private(inode, file); |
| 132 | if (status) { |
| 133 | /* |
| 134 | * We want to set open count back if we're failing the |
| 135 | * open. |
| 136 | */ |
| 137 | spin_lock(&oi->ip_lock); |
| 138 | oi->ip_open_count--; |
| 139 | spin_unlock(&oi->ip_lock); |
| 140 | } |
| 141 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 142 | leave: |
| 143 | mlog_exit(status); |
| 144 | return status; |
| 145 | } |
| 146 | |
| 147 | static int ocfs2_file_release(struct inode *inode, struct file *file) |
| 148 | { |
| 149 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 150 | |
| 151 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 152 | file->f_path.dentry->d_name.len, |
| 153 | file->f_path.dentry->d_name.name); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 154 | |
| 155 | spin_lock(&oi->ip_lock); |
| 156 | if (!--oi->ip_open_count) |
| 157 | oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT; |
| 158 | spin_unlock(&oi->ip_lock); |
| 159 | |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 160 | ocfs2_free_file_private(inode, file); |
| 161 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 162 | mlog_exit(0); |
| 163 | |
| 164 | return 0; |
| 165 | } |
| 166 | |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 167 | static int ocfs2_dir_open(struct inode *inode, struct file *file) |
| 168 | { |
| 169 | return ocfs2_init_file_private(inode, file); |
| 170 | } |
| 171 | |
| 172 | static int ocfs2_dir_release(struct inode *inode, struct file *file) |
| 173 | { |
| 174 | ocfs2_free_file_private(inode, file); |
| 175 | return 0; |
| 176 | } |
| 177 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 178 | static int ocfs2_sync_file(struct file *file, |
| 179 | struct dentry *dentry, |
| 180 | int datasync) |
| 181 | { |
| 182 | int err = 0; |
| 183 | journal_t *journal; |
| 184 | struct inode *inode = dentry->d_inode; |
| 185 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 186 | |
| 187 | mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync, |
| 188 | dentry->d_name.len, dentry->d_name.name); |
| 189 | |
| 190 | err = ocfs2_sync_inode(dentry->d_inode); |
| 191 | if (err) |
| 192 | goto bail; |
| 193 | |
Hisashi Hifumi | e04cc15 | 2009-06-09 16:47:45 +0900 | [diff] [blame] | 194 | if (datasync && !(inode->i_state & I_DIRTY_DATASYNC)) |
| 195 | goto bail; |
| 196 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 197 | journal = osb->journal->j_journal; |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 198 | err = jbd2_journal_force_commit(journal); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 199 | |
| 200 | bail: |
| 201 | mlog_exit(err); |
| 202 | |
| 203 | return (err < 0) ? -EIO : 0; |
| 204 | } |
| 205 | |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 206 | int ocfs2_should_update_atime(struct inode *inode, |
| 207 | struct vfsmount *vfsmnt) |
| 208 | { |
| 209 | struct timespec now; |
| 210 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 211 | |
| 212 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) |
| 213 | return 0; |
| 214 | |
| 215 | if ((inode->i_flags & S_NOATIME) || |
| 216 | ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))) |
| 217 | return 0; |
| 218 | |
Mark Fasheh | 6c2aad0 | 2006-12-19 15:25:52 -0800 | [diff] [blame] | 219 | /* |
| 220 | * We can be called with no vfsmnt structure - NFSD will |
| 221 | * sometimes do this. |
| 222 | * |
| 223 | * Note that our action here is different than touch_atime() - |
| 224 | * if we can't tell whether this is a noatime mount, then we |
| 225 | * don't know whether to trust the value of s_atime_quantum. |
| 226 | */ |
| 227 | if (vfsmnt == NULL) |
| 228 | return 0; |
| 229 | |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 230 | if ((vfsmnt->mnt_flags & MNT_NOATIME) || |
| 231 | ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) |
| 232 | return 0; |
| 233 | |
Mark Fasheh | 7e913c5 | 2006-12-13 00:34:35 -0800 | [diff] [blame] | 234 | if (vfsmnt->mnt_flags & MNT_RELATIME) { |
| 235 | if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) || |
| 236 | (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0)) |
| 237 | return 1; |
| 238 | |
| 239 | return 0; |
| 240 | } |
| 241 | |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 242 | now = CURRENT_TIME; |
| 243 | if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum)) |
| 244 | return 0; |
| 245 | else |
| 246 | return 1; |
| 247 | } |
| 248 | |
| 249 | int ocfs2_update_inode_atime(struct inode *inode, |
| 250 | struct buffer_head *bh) |
| 251 | { |
| 252 | int ret; |
| 253 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 254 | handle_t *handle; |
Mark Fasheh | c11e9fa | 2007-07-20 11:24:53 -0700 | [diff] [blame] | 255 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data; |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 256 | |
| 257 | mlog_entry_void(); |
| 258 | |
| 259 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
Jan Kara | fa38e92 | 2008-10-20 19:23:51 +0200 | [diff] [blame] | 260 | if (IS_ERR(handle)) { |
| 261 | ret = PTR_ERR(handle); |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 262 | mlog_errno(ret); |
| 263 | goto out; |
| 264 | } |
| 265 | |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 266 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh, |
Joel Becker | 13723d0 | 2008-10-17 19:25:01 -0700 | [diff] [blame] | 267 | OCFS2_JOURNAL_ACCESS_WRITE); |
Mark Fasheh | c11e9fa | 2007-07-20 11:24:53 -0700 | [diff] [blame] | 268 | if (ret) { |
| 269 | mlog_errno(ret); |
| 270 | goto out_commit; |
| 271 | } |
| 272 | |
| 273 | /* |
| 274 | * Don't use ocfs2_mark_inode_dirty() here as we don't always |
| 275 | * have i_mutex to guard against concurrent changes to other |
| 276 | * inode fields. |
| 277 | */ |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 278 | inode->i_atime = CURRENT_TIME; |
Mark Fasheh | c11e9fa | 2007-07-20 11:24:53 -0700 | [diff] [blame] | 279 | di->i_atime = cpu_to_le64(inode->i_atime.tv_sec); |
| 280 | di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec); |
Joel Becker | ec20cec | 2010-03-19 14:13:52 -0700 | [diff] [blame] | 281 | ocfs2_journal_dirty(handle, bh); |
Tiger Yang | 7f1a37e | 2006-11-15 15:48:42 +0800 | [diff] [blame] | 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 | |
Tao Ma | 37f8a2b | 2009-08-26 09:47:28 +0800 | [diff] [blame] | 338 | static int ocfs2_cow_file_pos(struct inode *inode, |
| 339 | struct buffer_head *fe_bh, |
| 340 | u64 offset) |
| 341 | { |
| 342 | int status; |
| 343 | u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits; |
| 344 | unsigned int num_clusters = 0; |
| 345 | unsigned int ext_flags = 0; |
| 346 | |
| 347 | /* |
| 348 | * If the new offset is aligned to the range of the cluster, there is |
| 349 | * no space for ocfs2_zero_range_for_truncate to fill, so no need to |
| 350 | * CoW either. |
| 351 | */ |
| 352 | if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0) |
| 353 | return 0; |
| 354 | |
| 355 | status = ocfs2_get_clusters(inode, cpos, &phys, |
| 356 | &num_clusters, &ext_flags); |
| 357 | if (status) { |
| 358 | mlog_errno(status); |
| 359 | goto out; |
| 360 | } |
| 361 | |
| 362 | if (!(ext_flags & OCFS2_EXT_REFCOUNTED)) |
| 363 | goto out; |
| 364 | |
| 365 | return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1); |
| 366 | |
| 367 | out: |
| 368 | return status; |
| 369 | } |
| 370 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 371 | static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb, |
| 372 | struct inode *inode, |
| 373 | struct buffer_head *fe_bh, |
| 374 | u64 new_i_size) |
| 375 | { |
| 376 | int status; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 377 | handle_t *handle; |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 378 | struct ocfs2_dinode *di; |
Mark Fasheh | 35edec1 | 2007-07-06 14:41:18 -0700 | [diff] [blame] | 379 | u64 cluster_bytes; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 380 | |
| 381 | mlog_entry_void(); |
| 382 | |
Tao Ma | 37f8a2b | 2009-08-26 09:47:28 +0800 | [diff] [blame] | 383 | /* |
| 384 | * We need to CoW the cluster contains the offset if it is reflinked |
| 385 | * since we will call ocfs2_zero_range_for_truncate later which will |
| 386 | * write "0" from offset to the end of the cluster. |
| 387 | */ |
| 388 | status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size); |
| 389 | if (status) { |
| 390 | mlog_errno(status); |
| 391 | return status; |
| 392 | } |
| 393 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 394 | /* TODO: This needs to actually orphan the inode in this |
| 395 | * transaction. */ |
| 396 | |
Mark Fasheh | 65eff9c | 2006-10-09 17:26:22 -0700 | [diff] [blame] | 397 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 398 | if (IS_ERR(handle)) { |
| 399 | status = PTR_ERR(handle); |
| 400 | mlog_errno(status); |
| 401 | goto out; |
| 402 | } |
| 403 | |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 404 | status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh, |
Joel Becker | 13723d0 | 2008-10-17 19:25:01 -0700 | [diff] [blame] | 405 | OCFS2_JOURNAL_ACCESS_WRITE); |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 406 | if (status < 0) { |
| 407 | mlog_errno(status); |
| 408 | goto out_commit; |
| 409 | } |
| 410 | |
| 411 | /* |
| 412 | * Do this before setting i_size. |
| 413 | */ |
Mark Fasheh | 35edec1 | 2007-07-06 14:41:18 -0700 | [diff] [blame] | 414 | cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size); |
| 415 | status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size, |
| 416 | cluster_bytes); |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 417 | if (status) { |
| 418 | mlog_errno(status); |
| 419 | goto out_commit; |
| 420 | } |
| 421 | |
| 422 | i_size_write(inode, new_i_size); |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 423 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
| 424 | |
| 425 | di = (struct ocfs2_dinode *) fe_bh->b_data; |
| 426 | di->i_size = cpu_to_le64(new_i_size); |
| 427 | di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec); |
| 428 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); |
| 429 | |
Joel Becker | ec20cec | 2010-03-19 14:13:52 -0700 | [diff] [blame] | 430 | ocfs2_journal_dirty(handle, fe_bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 431 | |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 432 | out_commit: |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 433 | ocfs2_commit_trans(osb, handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 434 | out: |
Mark Fasheh | 60b1139 | 2007-02-16 11:46:50 -0800 | [diff] [blame] | 435 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 436 | mlog_exit(status); |
| 437 | return status; |
| 438 | } |
| 439 | |
| 440 | static int ocfs2_truncate_file(struct inode *inode, |
| 441 | struct buffer_head *di_bh, |
| 442 | u64 new_i_size) |
| 443 | { |
| 444 | int status = 0; |
| 445 | struct ocfs2_dinode *fe = NULL; |
| 446 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 447 | |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 448 | mlog_entry("(inode = %llu, new_i_size = %llu\n", |
| 449 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
| 450 | (unsigned long long)new_i_size); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 451 | |
Joel Becker | b657c95 | 2008-11-13 14:49:11 -0800 | [diff] [blame] | 452 | /* We trust di_bh because it comes from ocfs2_inode_lock(), which |
| 453 | * already validated it */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 454 | fe = (struct ocfs2_dinode *) di_bh->b_data; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 455 | |
| 456 | 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] | 457 | "Inode %llu, inode i_size = %lld != di " |
| 458 | "i_size = %llu, i_flags = 0x%x\n", |
| 459 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 460 | i_size_read(inode), |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 461 | (unsigned long long)le64_to_cpu(fe->i_size), |
| 462 | le32_to_cpu(fe->i_flags)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 463 | |
| 464 | if (new_i_size > le64_to_cpu(fe->i_size)) { |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 465 | mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n", |
| 466 | (unsigned long long)le64_to_cpu(fe->i_size), |
| 467 | (unsigned long long)new_i_size); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 468 | status = -EINVAL; |
| 469 | mlog_errno(status); |
| 470 | goto bail; |
| 471 | } |
| 472 | |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 473 | mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n", |
| 474 | (unsigned long long)le64_to_cpu(fe->i_blkno), |
| 475 | (unsigned long long)le64_to_cpu(fe->i_size), |
| 476 | (unsigned long long)new_i_size); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 477 | |
| 478 | /* lets handle the simple truncate cases before doing any more |
| 479 | * cluster locking. */ |
| 480 | if (new_i_size == le64_to_cpu(fe->i_size)) |
| 481 | goto bail; |
| 482 | |
Mark Fasheh | 2e89b2e | 2007-05-09 13:40:18 -0700 | [diff] [blame] | 483 | down_write(&OCFS2_I(inode)->ip_alloc_sem); |
| 484 | |
Mark Fasheh | 4fe370a | 2009-12-07 13:15:40 -0800 | [diff] [blame] | 485 | ocfs2_resv_discard(&osb->osb_la_resmap, |
| 486 | &OCFS2_I(inode)->ip_la_data_resv); |
| 487 | |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 488 | /* |
| 489 | * The inode lock forced other nodes to sync and drop their |
| 490 | * pages, which (correctly) happens even if we have a truncate |
| 491 | * without allocation change - ocfs2 cluster sizes can be much |
| 492 | * greater than page size, so we have to truncate them |
| 493 | * anyway. |
| 494 | */ |
Mark Fasheh | 2e89b2e | 2007-05-09 13:40:18 -0700 | [diff] [blame] | 495 | unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1); |
| 496 | truncate_inode_pages(inode->i_mapping, new_i_size); |
| 497 | |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 498 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
| 499 | status = ocfs2_truncate_inline(inode, di_bh, new_i_size, |
Mark Fasheh | b1967d0 | 2007-11-20 11:56:39 -0800 | [diff] [blame] | 500 | i_size_read(inode), 1); |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 501 | if (status) |
| 502 | mlog_errno(status); |
| 503 | |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 504 | goto bail_unlock_sem; |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 505 | } |
| 506 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 507 | /* alright, we're going to need to do a full blown alloc size |
| 508 | * change. Orphan the inode so that recovery can complete the |
| 509 | * truncate if necessary. This does the task of marking |
| 510 | * i_size. */ |
| 511 | status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size); |
| 512 | if (status < 0) { |
| 513 | mlog_errno(status); |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 514 | goto bail_unlock_sem; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 515 | } |
| 516 | |
Tristan Ye | 78f9467 | 2010-05-11 17:54:42 +0800 | [diff] [blame] | 517 | status = ocfs2_commit_truncate(osb, inode, di_bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 518 | if (status < 0) { |
| 519 | mlog_errno(status); |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 520 | goto bail_unlock_sem; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 521 | } |
| 522 | |
| 523 | /* TODO: orphan dir cleanup here. */ |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 524 | bail_unlock_sem: |
Mark Fasheh | 2e89b2e | 2007-05-09 13:40:18 -0700 | [diff] [blame] | 525 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
| 526 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 527 | bail: |
Tao Ma | 8b2c0db | 2009-08-18 11:43:49 +0800 | [diff] [blame] | 528 | if (!status && OCFS2_I(inode)->ip_clusters == 0) |
| 529 | status = ocfs2_try_remove_refcount_tree(inode, di_bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 530 | |
| 531 | mlog_exit(status); |
| 532 | return status; |
| 533 | } |
| 534 | |
| 535 | /* |
Tao Ma | 0eb8d47 | 2008-08-18 17:38:45 +0800 | [diff] [blame] | 536 | * extend file allocation only here. |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 537 | * we'll update all the disk stuff, and oip->alloc_size |
| 538 | * |
| 539 | * expect stuff to be locked, a transaction started and enough data / |
| 540 | * metadata reservations in the contexts. |
| 541 | * |
| 542 | * Will return -EAGAIN, and a reason if a restart is needed. |
| 543 | * If passed in, *reason will always be set, even in error. |
| 544 | */ |
Tao Ma | 0eb8d47 | 2008-08-18 17:38:45 +0800 | [diff] [blame] | 545 | int ocfs2_add_inode_data(struct ocfs2_super *osb, |
| 546 | struct inode *inode, |
| 547 | u32 *logical_offset, |
| 548 | u32 clusters_to_add, |
| 549 | int mark_unwritten, |
| 550 | struct buffer_head *fe_bh, |
| 551 | handle_t *handle, |
| 552 | struct ocfs2_alloc_context *data_ac, |
| 553 | struct ocfs2_alloc_context *meta_ac, |
| 554 | enum ocfs2_alloc_restarted *reason_ret) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 555 | { |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 556 | int ret; |
| 557 | struct ocfs2_extent_tree et; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 558 | |
Joel Becker | 5e404e9 | 2009-02-13 03:54:22 -0800 | [diff] [blame] | 559 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh); |
Joel Becker | cbee7e1 | 2009-02-13 03:34:15 -0800 | [diff] [blame] | 560 | ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset, |
| 561 | clusters_to_add, mark_unwritten, |
| 562 | data_ac, meta_ac, reason_ret); |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 563 | |
| 564 | return ret; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 565 | } |
| 566 | |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 567 | static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start, |
| 568 | u32 clusters_to_add, int mark_unwritten) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 569 | { |
| 570 | int status = 0; |
| 571 | int restart_func = 0; |
Mark Fasheh | abf8b15 | 2007-01-17 13:07:24 -0800 | [diff] [blame] | 572 | int credits; |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 573 | u32 prev_clusters; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 574 | struct buffer_head *bh = NULL; |
| 575 | struct ocfs2_dinode *fe = NULL; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 576 | handle_t *handle = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 577 | struct ocfs2_alloc_context *data_ac = NULL; |
| 578 | struct ocfs2_alloc_context *meta_ac = NULL; |
| 579 | enum ocfs2_alloc_restarted why; |
| 580 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 581 | struct ocfs2_extent_tree et; |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 582 | int did_quota = 0; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 583 | |
| 584 | mlog_entry("(clusters_to_add = %u)\n", clusters_to_add); |
| 585 | |
Mark Fasheh | dcd0538 | 2007-01-16 11:32:23 -0800 | [diff] [blame] | 586 | /* |
| 587 | * This function only exists for file systems which don't |
| 588 | * support holes. |
| 589 | */ |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 590 | BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb)); |
Mark Fasheh | dcd0538 | 2007-01-16 11:32:23 -0800 | [diff] [blame] | 591 | |
Joel Becker | b657c95 | 2008-11-13 14:49:11 -0800 | [diff] [blame] | 592 | status = ocfs2_read_inode_block(inode, &bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 593 | if (status < 0) { |
| 594 | mlog_errno(status); |
| 595 | goto leave; |
| 596 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 597 | fe = (struct ocfs2_dinode *) bh->b_data; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 598 | |
| 599 | restart_all: |
| 600 | BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters); |
| 601 | |
Tao Ma | e7d4cb6 | 2008-08-18 17:38:44 +0800 | [diff] [blame] | 602 | mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, " |
| 603 | "clusters_to_add = %u\n", |
| 604 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
| 605 | (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters), |
| 606 | clusters_to_add); |
Joel Becker | 5e404e9 | 2009-02-13 03:54:22 -0800 | [diff] [blame] | 607 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh); |
Joel Becker | f99b9b7 | 2008-08-20 19:36:33 -0700 | [diff] [blame] | 608 | status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0, |
| 609 | &data_ac, &meta_ac); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 610 | if (status) { |
| 611 | mlog_errno(status); |
| 612 | goto leave; |
| 613 | } |
| 614 | |
Tao Ma | 811f933 | 2008-08-18 17:38:43 +0800 | [diff] [blame] | 615 | credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list, |
| 616 | clusters_to_add); |
Mark Fasheh | 65eff9c | 2006-10-09 17:26:22 -0700 | [diff] [blame] | 617 | handle = ocfs2_start_trans(osb, credits); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 618 | if (IS_ERR(handle)) { |
| 619 | status = PTR_ERR(handle); |
| 620 | handle = NULL; |
| 621 | mlog_errno(status); |
| 622 | goto leave; |
| 623 | } |
| 624 | |
| 625 | restarted_transaction: |
Christoph Hellwig | 5dd4056 | 2010-03-03 09:05:00 -0500 | [diff] [blame] | 626 | status = dquot_alloc_space_nodirty(inode, |
| 627 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_add)); |
| 628 | if (status) |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 629 | goto leave; |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 630 | did_quota = 1; |
| 631 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 632 | /* reserve a write to the file entry early on - that we if we |
| 633 | * run out of credits in the allocation path, we can still |
| 634 | * update i_size. */ |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 635 | status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh, |
Joel Becker | 13723d0 | 2008-10-17 19:25:01 -0700 | [diff] [blame] | 636 | OCFS2_JOURNAL_ACCESS_WRITE); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 637 | if (status < 0) { |
| 638 | mlog_errno(status); |
| 639 | goto leave; |
| 640 | } |
| 641 | |
| 642 | prev_clusters = OCFS2_I(inode)->ip_clusters; |
| 643 | |
Tao Ma | 0eb8d47 | 2008-08-18 17:38:45 +0800 | [diff] [blame] | 644 | status = ocfs2_add_inode_data(osb, |
| 645 | inode, |
| 646 | &logical_start, |
| 647 | clusters_to_add, |
| 648 | mark_unwritten, |
| 649 | bh, |
| 650 | handle, |
| 651 | data_ac, |
| 652 | meta_ac, |
| 653 | &why); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 654 | if ((status < 0) && (status != -EAGAIN)) { |
| 655 | if (status != -ENOSPC) |
| 656 | mlog_errno(status); |
| 657 | goto leave; |
| 658 | } |
| 659 | |
Joel Becker | ec20cec | 2010-03-19 14:13:52 -0700 | [diff] [blame] | 660 | ocfs2_journal_dirty(handle, bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 661 | |
| 662 | spin_lock(&OCFS2_I(inode)->ip_lock); |
| 663 | clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters); |
| 664 | spin_unlock(&OCFS2_I(inode)->ip_lock); |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 665 | /* Release unused quota reservation */ |
Christoph Hellwig | 5dd4056 | 2010-03-03 09:05:00 -0500 | [diff] [blame] | 666 | dquot_free_space(inode, |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 667 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_add)); |
| 668 | did_quota = 0; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 669 | |
| 670 | if (why != RESTART_NONE && clusters_to_add) { |
| 671 | if (why == RESTART_META) { |
| 672 | mlog(0, "restarting function.\n"); |
| 673 | restart_func = 1; |
| 674 | } else { |
| 675 | BUG_ON(why != RESTART_TRANS); |
| 676 | |
| 677 | mlog(0, "restarting transaction.\n"); |
| 678 | /* TODO: This can be more intelligent. */ |
| 679 | credits = ocfs2_calc_extend_credits(osb->sb, |
Tao Ma | 811f933 | 2008-08-18 17:38:43 +0800 | [diff] [blame] | 680 | &fe->id2.i_list, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 681 | clusters_to_add); |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 682 | status = ocfs2_extend_trans(handle, credits); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 683 | if (status < 0) { |
| 684 | /* handle still has to be committed at |
| 685 | * this point. */ |
| 686 | status = -ENOMEM; |
| 687 | mlog_errno(status); |
| 688 | goto leave; |
| 689 | } |
| 690 | goto restarted_transaction; |
| 691 | } |
| 692 | } |
| 693 | |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 694 | mlog(0, "fe: i_clusters = %u, i_size=%llu\n", |
Mark Fasheh | 1ca1a11 | 2007-04-27 16:01:25 -0700 | [diff] [blame] | 695 | le32_to_cpu(fe->i_clusters), |
| 696 | (unsigned long long)le64_to_cpu(fe->i_size)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 697 | mlog(0, "inode: ip_clusters=%u, i_size=%lld\n", |
Jan Kara | 634bf74 | 2007-12-19 15:25:42 +0100 | [diff] [blame] | 698 | OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 699 | |
| 700 | leave: |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 701 | if (status < 0 && did_quota) |
Christoph Hellwig | 5dd4056 | 2010-03-03 09:05:00 -0500 | [diff] [blame] | 702 | dquot_free_space(inode, |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 703 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_add)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 704 | if (handle) { |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 705 | ocfs2_commit_trans(osb, handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 706 | handle = NULL; |
| 707 | } |
| 708 | if (data_ac) { |
| 709 | ocfs2_free_alloc_context(data_ac); |
| 710 | data_ac = NULL; |
| 711 | } |
| 712 | if (meta_ac) { |
| 713 | ocfs2_free_alloc_context(meta_ac); |
| 714 | meta_ac = NULL; |
| 715 | } |
| 716 | if ((!status) && restart_func) { |
| 717 | restart_func = 0; |
| 718 | goto restart_all; |
| 719 | } |
Mark Fasheh | a81cb88 | 2008-10-07 14:25:16 -0700 | [diff] [blame] | 720 | brelse(bh); |
| 721 | bh = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 722 | |
| 723 | mlog_exit(status); |
| 724 | return status; |
| 725 | } |
| 726 | |
| 727 | /* Some parts of this taken from generic_cont_expand, which turned out |
| 728 | * 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] | 729 | * worry about recursive locking in ->write_begin() and ->write_end(). */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 730 | static int ocfs2_write_zero_page(struct inode *inode, |
| 731 | u64 size) |
| 732 | { |
| 733 | struct address_space *mapping = inode->i_mapping; |
| 734 | struct page *page; |
| 735 | unsigned long index; |
| 736 | unsigned int offset; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 737 | handle_t *handle = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 738 | int ret; |
| 739 | |
| 740 | offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */ |
Sunil Mushran | 2bd6321 | 2010-01-25 16:57:38 -0800 | [diff] [blame] | 741 | /* ugh. in prepare/commit_write, if from==to==start of block, we |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 742 | ** skip the prepare. make sure we never send an offset for the start |
| 743 | ** of a block |
| 744 | */ |
| 745 | if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) { |
| 746 | offset++; |
| 747 | } |
| 748 | index = size >> PAGE_CACHE_SHIFT; |
| 749 | |
| 750 | page = grab_cache_page(mapping, index); |
| 751 | if (!page) { |
| 752 | ret = -ENOMEM; |
| 753 | mlog_errno(ret); |
| 754 | goto out; |
| 755 | } |
| 756 | |
Mark Fasheh | 53013cb | 2006-05-05 19:04:03 -0700 | [diff] [blame] | 757 | ret = ocfs2_prepare_write_nolock(inode, page, offset, offset); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 758 | if (ret < 0) { |
| 759 | mlog_errno(ret); |
| 760 | goto out_unlock; |
| 761 | } |
| 762 | |
| 763 | if (ocfs2_should_order_data(inode)) { |
| 764 | handle = ocfs2_start_walk_page_trans(inode, page, offset, |
| 765 | offset); |
| 766 | if (IS_ERR(handle)) { |
| 767 | ret = PTR_ERR(handle); |
| 768 | handle = NULL; |
| 769 | goto out_unlock; |
| 770 | } |
| 771 | } |
| 772 | |
| 773 | /* must not update i_size! */ |
| 774 | ret = block_commit_write(page, offset, offset); |
| 775 | if (ret < 0) |
| 776 | mlog_errno(ret); |
| 777 | else |
| 778 | ret = 0; |
| 779 | |
| 780 | if (handle) |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 781 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 782 | out_unlock: |
| 783 | unlock_page(page); |
| 784 | page_cache_release(page); |
| 785 | out: |
| 786 | return ret; |
| 787 | } |
| 788 | |
| 789 | static int ocfs2_zero_extend(struct inode *inode, |
| 790 | u64 zero_to_size) |
| 791 | { |
| 792 | int ret = 0; |
| 793 | u64 start_off; |
| 794 | struct super_block *sb = inode->i_sb; |
| 795 | |
| 796 | start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode)); |
| 797 | while (start_off < zero_to_size) { |
| 798 | ret = ocfs2_write_zero_page(inode, start_off); |
| 799 | if (ret < 0) { |
| 800 | mlog_errno(ret); |
| 801 | goto out; |
| 802 | } |
| 803 | |
| 804 | start_off += sb->s_blocksize; |
Mark Fasheh | e2057c5 | 2006-10-03 17:53:05 -0700 | [diff] [blame] | 805 | |
| 806 | /* |
| 807 | * Very large extends have the potential to lock up |
| 808 | * the cpu for extended periods of time. |
| 809 | */ |
| 810 | cond_resched(); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 811 | } |
| 812 | |
| 813 | out: |
| 814 | return ret; |
| 815 | } |
| 816 | |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 817 | int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to) |
| 818 | { |
| 819 | int ret; |
| 820 | u32 clusters_to_add; |
| 821 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 822 | |
| 823 | clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size); |
| 824 | if (clusters_to_add < oi->ip_clusters) |
| 825 | clusters_to_add = 0; |
| 826 | else |
| 827 | clusters_to_add -= oi->ip_clusters; |
| 828 | |
| 829 | if (clusters_to_add) { |
| 830 | ret = __ocfs2_extend_allocation(inode, oi->ip_clusters, |
| 831 | clusters_to_add, 0); |
| 832 | if (ret) { |
| 833 | mlog_errno(ret); |
| 834 | goto out; |
| 835 | } |
| 836 | } |
| 837 | |
| 838 | /* |
| 839 | * Call this even if we don't add any clusters to the tree. We |
| 840 | * still need to zero the area between the old i_size and the |
| 841 | * new i_size. |
| 842 | */ |
| 843 | ret = ocfs2_zero_extend(inode, zero_to); |
| 844 | if (ret < 0) |
| 845 | mlog_errno(ret); |
| 846 | |
| 847 | out: |
| 848 | return ret; |
| 849 | } |
| 850 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 851 | static int ocfs2_extend_file(struct inode *inode, |
| 852 | struct buffer_head *di_bh, |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 853 | u64 new_i_size) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 854 | { |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 855 | int ret = 0; |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 856 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 857 | |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 858 | BUG_ON(!di_bh); |
Mark Fasheh | 53013cb | 2006-05-05 19:04:03 -0700 | [diff] [blame] | 859 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 860 | /* setattr sometimes calls us like this. */ |
| 861 | if (new_i_size == 0) |
| 862 | goto out; |
| 863 | |
| 864 | if (i_size_read(inode) == new_i_size) |
| 865 | goto out; |
| 866 | BUG_ON(new_i_size < i_size_read(inode)); |
| 867 | |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 868 | /* |
| 869 | * Fall through for converting inline data, even if the fs |
| 870 | * supports sparse files. |
| 871 | * |
| 872 | * The check for inline data here is legal - nobody can add |
| 873 | * the feature since we have i_mutex. We must check it again |
| 874 | * after acquiring ip_alloc_sem though, as paths like mmap |
| 875 | * might have raced us to converting the inode to extents. |
| 876 | */ |
| 877 | if (!(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) |
| 878 | && ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) |
Mark Fasheh | 3a0782d | 2007-01-17 12:53:31 -0800 | [diff] [blame] | 879 | goto out_update_size; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 880 | |
Mark Fasheh | 0effef7 | 2006-10-03 17:44:42 -0700 | [diff] [blame] | 881 | /* |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 882 | * The alloc sem blocks people in read/write from reading our |
| 883 | * allocation until we're done changing it. We depend on |
| 884 | * i_mutex to block other extend/truncate calls while we're |
| 885 | * here. |
Mark Fasheh | 0effef7 | 2006-10-03 17:44:42 -0700 | [diff] [blame] | 886 | */ |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 887 | down_write(&oi->ip_alloc_sem); |
| 888 | |
| 889 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
| 890 | /* |
| 891 | * We can optimize small extends by keeping the inodes |
| 892 | * inline data. |
| 893 | */ |
| 894 | if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) { |
| 895 | up_write(&oi->ip_alloc_sem); |
| 896 | goto out_update_size; |
| 897 | } |
| 898 | |
| 899 | ret = ocfs2_convert_inline_data_to_extents(inode, di_bh); |
| 900 | if (ret) { |
| 901 | up_write(&oi->ip_alloc_sem); |
| 902 | |
| 903 | mlog_errno(ret); |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 904 | goto out; |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 905 | } |
| 906 | } |
| 907 | |
| 908 | if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) |
| 909 | ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size); |
| 910 | |
| 911 | up_write(&oi->ip_alloc_sem); |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 912 | |
Mark Fasheh | 0effef7 | 2006-10-03 17:44:42 -0700 | [diff] [blame] | 913 | if (ret < 0) { |
| 914 | mlog_errno(ret); |
Mark Fasheh | c934a92 | 2007-10-18 15:23:46 -0700 | [diff] [blame] | 915 | goto out; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 916 | } |
| 917 | |
Mark Fasheh | 3a0782d | 2007-01-17 12:53:31 -0800 | [diff] [blame] | 918 | out_update_size: |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 919 | ret = ocfs2_simple_size_update(inode, di_bh, new_i_size); |
| 920 | if (ret < 0) |
| 921 | mlog_errno(ret); |
Mark Fasheh | 53013cb | 2006-05-05 19:04:03 -0700 | [diff] [blame] | 922 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 923 | out: |
| 924 | return ret; |
| 925 | } |
| 926 | |
| 927 | int ocfs2_setattr(struct dentry *dentry, struct iattr *attr) |
| 928 | { |
| 929 | int status = 0, size_change; |
| 930 | struct inode *inode = dentry->d_inode; |
| 931 | struct super_block *sb = inode->i_sb; |
| 932 | struct ocfs2_super *osb = OCFS2_SB(sb); |
| 933 | struct buffer_head *bh = NULL; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 934 | handle_t *handle = NULL; |
Jan Kara | 65bac57 | 2009-06-02 14:24:01 +0200 | [diff] [blame] | 935 | int qtype; |
| 936 | struct dquot *transfer_from[MAXQUOTAS] = { }; |
| 937 | struct dquot *transfer_to[MAXQUOTAS] = { }; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 938 | |
| 939 | mlog_entry("(0x%p, '%.*s')\n", dentry, |
| 940 | dentry->d_name.len, dentry->d_name.name); |
| 941 | |
Sunil Mushran | bc53580 | 2008-04-18 10:23:53 -0700 | [diff] [blame] | 942 | /* ensuring we don't even attempt to truncate a symlink */ |
| 943 | if (S_ISLNK(inode->i_mode)) |
| 944 | attr->ia_valid &= ~ATTR_SIZE; |
| 945 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 946 | if (attr->ia_valid & ATTR_MODE) |
| 947 | mlog(0, "mode change: %d\n", attr->ia_mode); |
| 948 | if (attr->ia_valid & ATTR_UID) |
| 949 | mlog(0, "uid change: %d\n", attr->ia_uid); |
| 950 | if (attr->ia_valid & ATTR_GID) |
| 951 | mlog(0, "gid change: %d\n", attr->ia_gid); |
| 952 | if (attr->ia_valid & ATTR_SIZE) |
| 953 | mlog(0, "size change...\n"); |
| 954 | if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME)) |
| 955 | mlog(0, "time change...\n"); |
| 956 | |
| 957 | #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \ |
| 958 | | ATTR_GID | ATTR_UID | ATTR_MODE) |
| 959 | if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) { |
| 960 | mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid); |
| 961 | return 0; |
| 962 | } |
| 963 | |
| 964 | status = inode_change_ok(inode, attr); |
| 965 | if (status) |
| 966 | return status; |
| 967 | |
| 968 | size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE; |
| 969 | if (size_change) { |
Christoph Hellwig | 871a293 | 2010-03-03 09:05:07 -0500 | [diff] [blame] | 970 | dquot_initialize(inode); |
Christoph Hellwig | 907f455 | 2010-03-03 09:05:06 -0500 | [diff] [blame] | 971 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 972 | status = ocfs2_rw_lock(inode, 1); |
| 973 | if (status < 0) { |
| 974 | mlog_errno(status); |
| 975 | goto bail; |
| 976 | } |
| 977 | } |
| 978 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 979 | status = ocfs2_inode_lock(inode, &bh, 1); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 980 | if (status < 0) { |
| 981 | if (status != -ENOENT) |
| 982 | mlog_errno(status); |
| 983 | goto bail_unlock_rw; |
| 984 | } |
| 985 | |
| 986 | if (size_change && attr->ia_size != i_size_read(inode)) { |
Wengang Wang | 5051f76 | 2010-02-26 18:18:25 +0800 | [diff] [blame] | 987 | status = inode_newsize_ok(inode, attr->ia_size); |
| 988 | if (status) |
Mark Fasheh | ce76fd3 | 2007-07-20 12:02:14 -0700 | [diff] [blame] | 989 | goto bail_unlock; |
Mark Fasheh | ce76fd3 | 2007-07-20 12:02:14 -0700 | [diff] [blame] | 990 | |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 991 | if (i_size_read(inode) > attr->ia_size) { |
| 992 | if (ocfs2_should_order_data(inode)) { |
| 993 | status = ocfs2_begin_ordered_truncate(inode, |
| 994 | attr->ia_size); |
| 995 | if (status) |
| 996 | goto bail_unlock; |
| 997 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 998 | status = ocfs2_truncate_file(inode, bh, attr->ia_size); |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 999 | } else |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1000 | status = ocfs2_extend_file(inode, bh, attr->ia_size); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1001 | if (status < 0) { |
| 1002 | if (status != -ENOSPC) |
| 1003 | mlog_errno(status); |
| 1004 | status = -ENOSPC; |
| 1005 | goto bail_unlock; |
| 1006 | } |
| 1007 | } |
| 1008 | |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 1009 | if ((attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) || |
| 1010 | (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) { |
Jan Kara | 65bac57 | 2009-06-02 14:24:01 +0200 | [diff] [blame] | 1011 | /* |
| 1012 | * Gather pointers to quota structures so that allocation / |
| 1013 | * freeing of quota structures happens here and not inside |
Christoph Hellwig | b43fa82 | 2010-03-03 09:05:03 -0500 | [diff] [blame] | 1014 | * dquot_transfer() where we have problems with lock ordering |
Jan Kara | 65bac57 | 2009-06-02 14:24:01 +0200 | [diff] [blame] | 1015 | */ |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 1016 | if (attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid |
| 1017 | && OCFS2_HAS_RO_COMPAT_FEATURE(sb, |
| 1018 | OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) { |
Jan Kara | 65bac57 | 2009-06-02 14:24:01 +0200 | [diff] [blame] | 1019 | transfer_to[USRQUOTA] = dqget(sb, attr->ia_uid, |
| 1020 | USRQUOTA); |
| 1021 | transfer_from[USRQUOTA] = dqget(sb, inode->i_uid, |
| 1022 | USRQUOTA); |
| 1023 | if (!transfer_to[USRQUOTA] || !transfer_from[USRQUOTA]) { |
| 1024 | status = -ESRCH; |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 1025 | goto bail_unlock; |
Jan Kara | 65bac57 | 2009-06-02 14:24:01 +0200 | [diff] [blame] | 1026 | } |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 1027 | } |
| 1028 | if (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid |
| 1029 | && OCFS2_HAS_RO_COMPAT_FEATURE(sb, |
| 1030 | OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) { |
Jan Kara | 65bac57 | 2009-06-02 14:24:01 +0200 | [diff] [blame] | 1031 | transfer_to[GRPQUOTA] = dqget(sb, attr->ia_gid, |
| 1032 | GRPQUOTA); |
| 1033 | transfer_from[GRPQUOTA] = dqget(sb, inode->i_gid, |
| 1034 | GRPQUOTA); |
| 1035 | if (!transfer_to[GRPQUOTA] || !transfer_from[GRPQUOTA]) { |
| 1036 | status = -ESRCH; |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 1037 | goto bail_unlock; |
Jan Kara | 65bac57 | 2009-06-02 14:24:01 +0200 | [diff] [blame] | 1038 | } |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 1039 | } |
Jan Kara | 65bac57 | 2009-06-02 14:24:01 +0200 | [diff] [blame] | 1040 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS + |
| 1041 | 2 * ocfs2_quota_trans_credits(sb)); |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 1042 | if (IS_ERR(handle)) { |
| 1043 | status = PTR_ERR(handle); |
| 1044 | mlog_errno(status); |
| 1045 | goto bail_unlock; |
| 1046 | } |
Christoph Hellwig | b43fa82 | 2010-03-03 09:05:03 -0500 | [diff] [blame] | 1047 | status = dquot_transfer(inode, attr); |
Jan Kara | a90714c | 2008-10-09 19:38:40 +0200 | [diff] [blame] | 1048 | if (status < 0) |
| 1049 | goto bail_commit; |
| 1050 | } else { |
| 1051 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
| 1052 | if (IS_ERR(handle)) { |
| 1053 | status = PTR_ERR(handle); |
| 1054 | mlog_errno(status); |
| 1055 | goto bail_unlock; |
| 1056 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1057 | } |
| 1058 | |
Mark Fasheh | 7307de8 | 2007-05-09 15:16:19 -0700 | [diff] [blame] | 1059 | /* |
| 1060 | * This will intentionally not wind up calling vmtruncate(), |
| 1061 | * since all the work for a size change has been done above. |
| 1062 | * Otherwise, we could get into problems with truncate as |
| 1063 | * ip_alloc_sem is used there to protect against i_size |
| 1064 | * changes. |
| 1065 | */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1066 | status = inode_setattr(inode, attr); |
| 1067 | if (status < 0) { |
| 1068 | mlog_errno(status); |
| 1069 | goto bail_commit; |
| 1070 | } |
| 1071 | |
| 1072 | status = ocfs2_mark_inode_dirty(handle, inode, bh); |
| 1073 | if (status < 0) |
| 1074 | mlog_errno(status); |
| 1075 | |
| 1076 | bail_commit: |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 1077 | ocfs2_commit_trans(osb, handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1078 | bail_unlock: |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1079 | ocfs2_inode_unlock(inode, 1); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1080 | bail_unlock_rw: |
| 1081 | if (size_change) |
| 1082 | ocfs2_rw_unlock(inode, 1); |
| 1083 | bail: |
Mark Fasheh | a81cb88 | 2008-10-07 14:25:16 -0700 | [diff] [blame] | 1084 | brelse(bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1085 | |
Jan Kara | 65bac57 | 2009-06-02 14:24:01 +0200 | [diff] [blame] | 1086 | /* Release quota pointers in case we acquired them */ |
| 1087 | for (qtype = 0; qtype < MAXQUOTAS; qtype++) { |
| 1088 | dqput(transfer_to[qtype]); |
| 1089 | dqput(transfer_from[qtype]); |
| 1090 | } |
| 1091 | |
Tiger Yang | 060bc66 | 2008-11-14 11:17:29 +0800 | [diff] [blame] | 1092 | if (!status && attr->ia_valid & ATTR_MODE) { |
| 1093 | status = ocfs2_acl_chmod(inode); |
| 1094 | if (status < 0) |
| 1095 | mlog_errno(status); |
| 1096 | } |
| 1097 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1098 | mlog_exit(status); |
| 1099 | return status; |
| 1100 | } |
| 1101 | |
| 1102 | int ocfs2_getattr(struct vfsmount *mnt, |
| 1103 | struct dentry *dentry, |
| 1104 | struct kstat *stat) |
| 1105 | { |
| 1106 | struct inode *inode = dentry->d_inode; |
| 1107 | struct super_block *sb = dentry->d_inode->i_sb; |
| 1108 | struct ocfs2_super *osb = sb->s_fs_info; |
| 1109 | int err; |
| 1110 | |
| 1111 | mlog_entry_void(); |
| 1112 | |
| 1113 | err = ocfs2_inode_revalidate(dentry); |
| 1114 | if (err) { |
| 1115 | if (err != -ENOENT) |
| 1116 | mlog_errno(err); |
| 1117 | goto bail; |
| 1118 | } |
| 1119 | |
| 1120 | generic_fillattr(inode, stat); |
| 1121 | |
| 1122 | /* We set the blksize from the cluster size for performance */ |
| 1123 | stat->blksize = osb->s_clustersize; |
| 1124 | |
| 1125 | bail: |
| 1126 | mlog_exit(err); |
| 1127 | |
| 1128 | return err; |
| 1129 | } |
| 1130 | |
Al Viro | e6305c4 | 2008-07-15 21:03:57 -0400 | [diff] [blame] | 1131 | int ocfs2_permission(struct inode *inode, int mask) |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 1132 | { |
| 1133 | int ret; |
| 1134 | |
| 1135 | mlog_entry_void(); |
| 1136 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1137 | ret = ocfs2_inode_lock(inode, NULL, 0); |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 1138 | if (ret) { |
Mark Fasheh | a9f5f70 | 2007-04-26 11:43:43 -0700 | [diff] [blame] | 1139 | if (ret != -ENOENT) |
| 1140 | mlog_errno(ret); |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 1141 | goto out; |
| 1142 | } |
| 1143 | |
Tiger Yang | 23fc270 | 2008-11-14 11:17:18 +0800 | [diff] [blame] | 1144 | ret = generic_permission(inode, mask, ocfs2_check_acl); |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 1145 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1146 | ocfs2_inode_unlock(inode, 0); |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 1147 | out: |
| 1148 | mlog_exit(ret); |
| 1149 | return ret; |
| 1150 | } |
| 1151 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1152 | static int __ocfs2_write_remove_suid(struct inode *inode, |
| 1153 | struct buffer_head *bh) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1154 | { |
| 1155 | int ret; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 1156 | handle_t *handle; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1157 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1158 | struct ocfs2_dinode *di; |
| 1159 | |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 1160 | mlog_entry("(Inode %llu, mode 0%o)\n", |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1161 | (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1162 | |
Mark Fasheh | 65eff9c | 2006-10-09 17:26:22 -0700 | [diff] [blame] | 1163 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
Jan Kara | fa38e92 | 2008-10-20 19:23:51 +0200 | [diff] [blame] | 1164 | if (IS_ERR(handle)) { |
| 1165 | ret = PTR_ERR(handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1166 | mlog_errno(ret); |
| 1167 | goto out; |
| 1168 | } |
| 1169 | |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 1170 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh, |
Joel Becker | 13723d0 | 2008-10-17 19:25:01 -0700 | [diff] [blame] | 1171 | OCFS2_JOURNAL_ACCESS_WRITE); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1172 | if (ret < 0) { |
| 1173 | mlog_errno(ret); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1174 | goto out_trans; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1175 | } |
| 1176 | |
| 1177 | inode->i_mode &= ~S_ISUID; |
| 1178 | if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP)) |
| 1179 | inode->i_mode &= ~S_ISGID; |
| 1180 | |
| 1181 | di = (struct ocfs2_dinode *) bh->b_data; |
| 1182 | di->i_mode = cpu_to_le16(inode->i_mode); |
| 1183 | |
Joel Becker | ec20cec | 2010-03-19 14:13:52 -0700 | [diff] [blame] | 1184 | ocfs2_journal_dirty(handle, bh); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1185 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1186 | out_trans: |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 1187 | ocfs2_commit_trans(osb, handle); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1188 | out: |
| 1189 | mlog_exit(ret); |
| 1190 | return ret; |
| 1191 | } |
| 1192 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1193 | /* |
| 1194 | * Will look for holes and unwritten extents in the range starting at |
| 1195 | * pos for count bytes (inclusive). |
| 1196 | */ |
| 1197 | static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos, |
| 1198 | size_t count) |
| 1199 | { |
| 1200 | int ret = 0; |
Mark Fasheh | 49cb8d2 | 2007-03-09 16:21:46 -0800 | [diff] [blame] | 1201 | unsigned int extent_flags; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1202 | u32 cpos, clusters, extent_len, phys_cpos; |
| 1203 | struct super_block *sb = inode->i_sb; |
| 1204 | |
| 1205 | cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits; |
| 1206 | clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos; |
| 1207 | |
| 1208 | while (clusters) { |
Mark Fasheh | 49cb8d2 | 2007-03-09 16:21:46 -0800 | [diff] [blame] | 1209 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len, |
| 1210 | &extent_flags); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1211 | if (ret < 0) { |
| 1212 | mlog_errno(ret); |
| 1213 | goto out; |
| 1214 | } |
| 1215 | |
Mark Fasheh | 49cb8d2 | 2007-03-09 16:21:46 -0800 | [diff] [blame] | 1216 | if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) { |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1217 | ret = 1; |
| 1218 | break; |
| 1219 | } |
| 1220 | |
| 1221 | if (extent_len > clusters) |
| 1222 | extent_len = clusters; |
| 1223 | |
| 1224 | clusters -= extent_len; |
| 1225 | cpos += extent_len; |
| 1226 | } |
| 1227 | out: |
| 1228 | return ret; |
| 1229 | } |
| 1230 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1231 | static int ocfs2_write_remove_suid(struct inode *inode) |
| 1232 | { |
| 1233 | int ret; |
| 1234 | struct buffer_head *bh = NULL; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1235 | |
Joel Becker | b657c95 | 2008-11-13 14:49:11 -0800 | [diff] [blame] | 1236 | ret = ocfs2_read_inode_block(inode, &bh); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1237 | if (ret < 0) { |
| 1238 | mlog_errno(ret); |
| 1239 | goto out; |
| 1240 | } |
| 1241 | |
| 1242 | ret = __ocfs2_write_remove_suid(inode, bh); |
| 1243 | out: |
| 1244 | brelse(bh); |
| 1245 | return ret; |
| 1246 | } |
| 1247 | |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 1248 | /* |
| 1249 | * Allocate enough extents to cover the region starting at byte offset |
| 1250 | * start for len bytes. Existing extents are skipped, any extents |
| 1251 | * added are marked as "unwritten". |
| 1252 | */ |
| 1253 | static int ocfs2_allocate_unwritten_extents(struct inode *inode, |
| 1254 | u64 start, u64 len) |
| 1255 | { |
| 1256 | int ret; |
| 1257 | u32 cpos, phys_cpos, clusters, alloc_size; |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1258 | u64 end = start + len; |
| 1259 | struct buffer_head *di_bh = NULL; |
| 1260 | |
| 1261 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
Joel Becker | b657c95 | 2008-11-13 14:49:11 -0800 | [diff] [blame] | 1262 | ret = ocfs2_read_inode_block(inode, &di_bh); |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1263 | if (ret) { |
| 1264 | mlog_errno(ret); |
| 1265 | goto out; |
| 1266 | } |
| 1267 | |
| 1268 | /* |
| 1269 | * Nothing to do if the requested reservation range |
| 1270 | * fits within the inode. |
| 1271 | */ |
| 1272 | if (ocfs2_size_fits_inline_data(di_bh, end)) |
| 1273 | goto out; |
| 1274 | |
| 1275 | ret = ocfs2_convert_inline_data_to_extents(inode, di_bh); |
| 1276 | if (ret) { |
| 1277 | mlog_errno(ret); |
| 1278 | goto out; |
| 1279 | } |
| 1280 | } |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 1281 | |
| 1282 | /* |
| 1283 | * We consider both start and len to be inclusive. |
| 1284 | */ |
| 1285 | cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits; |
| 1286 | clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len); |
| 1287 | clusters -= cpos; |
| 1288 | |
| 1289 | while (clusters) { |
| 1290 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, |
| 1291 | &alloc_size, NULL); |
| 1292 | if (ret) { |
| 1293 | mlog_errno(ret); |
| 1294 | goto out; |
| 1295 | } |
| 1296 | |
| 1297 | /* |
| 1298 | * Hole or existing extent len can be arbitrary, so |
| 1299 | * cap it to our own allocation request. |
| 1300 | */ |
| 1301 | if (alloc_size > clusters) |
| 1302 | alloc_size = clusters; |
| 1303 | |
| 1304 | if (phys_cpos) { |
| 1305 | /* |
| 1306 | * We already have an allocation at this |
| 1307 | * region so we can safely skip it. |
| 1308 | */ |
| 1309 | goto next; |
| 1310 | } |
| 1311 | |
| 1312 | ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1); |
| 1313 | if (ret) { |
| 1314 | if (ret != -ENOSPC) |
| 1315 | mlog_errno(ret); |
| 1316 | goto out; |
| 1317 | } |
| 1318 | |
| 1319 | next: |
| 1320 | cpos += alloc_size; |
| 1321 | clusters -= alloc_size; |
| 1322 | } |
| 1323 | |
| 1324 | ret = 0; |
| 1325 | out: |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1326 | |
| 1327 | brelse(di_bh); |
Mark Fasheh | 2ae99a6 | 2007-03-09 16:43:28 -0800 | [diff] [blame] | 1328 | return ret; |
| 1329 | } |
| 1330 | |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1331 | /* |
| 1332 | * Truncate a byte range, avoiding pages within partial clusters. This |
| 1333 | * preserves those pages for the zeroing code to write to. |
| 1334 | */ |
| 1335 | static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start, |
| 1336 | u64 byte_len) |
| 1337 | { |
| 1338 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1339 | loff_t start, end; |
| 1340 | struct address_space *mapping = inode->i_mapping; |
| 1341 | |
| 1342 | start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start); |
| 1343 | end = byte_start + byte_len; |
| 1344 | end = end & ~(osb->s_clustersize - 1); |
| 1345 | |
| 1346 | if (start < end) { |
| 1347 | unmap_mapping_range(mapping, start, end - start, 0); |
| 1348 | truncate_inode_pages_range(mapping, start, end - 1); |
| 1349 | } |
| 1350 | } |
| 1351 | |
| 1352 | static int ocfs2_zero_partial_clusters(struct inode *inode, |
| 1353 | u64 start, u64 len) |
| 1354 | { |
| 1355 | int ret = 0; |
| 1356 | u64 tmpend, end = start + len; |
| 1357 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1358 | unsigned int csize = osb->s_clustersize; |
| 1359 | handle_t *handle; |
| 1360 | |
| 1361 | /* |
| 1362 | * The "start" and "end" values are NOT necessarily part of |
| 1363 | * the range whose allocation is being deleted. Rather, this |
| 1364 | * is what the user passed in with the request. We must zero |
| 1365 | * partial clusters here. There's no need to worry about |
| 1366 | * physical allocation - the zeroing code knows to skip holes. |
| 1367 | */ |
| 1368 | mlog(0, "byte start: %llu, end: %llu\n", |
| 1369 | (unsigned long long)start, (unsigned long long)end); |
| 1370 | |
| 1371 | /* |
| 1372 | * If both edges are on a cluster boundary then there's no |
| 1373 | * zeroing required as the region is part of the allocation to |
| 1374 | * be truncated. |
| 1375 | */ |
| 1376 | if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0) |
| 1377 | goto out; |
| 1378 | |
| 1379 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
Jan Kara | fa38e92 | 2008-10-20 19:23:51 +0200 | [diff] [blame] | 1380 | if (IS_ERR(handle)) { |
| 1381 | ret = PTR_ERR(handle); |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1382 | mlog_errno(ret); |
| 1383 | goto out; |
| 1384 | } |
| 1385 | |
| 1386 | /* |
| 1387 | * We want to get the byte offset of the end of the 1st cluster. |
| 1388 | */ |
| 1389 | tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1)); |
| 1390 | if (tmpend > end) |
| 1391 | tmpend = end; |
| 1392 | |
| 1393 | mlog(0, "1st range: start: %llu, tmpend: %llu\n", |
| 1394 | (unsigned long long)start, (unsigned long long)tmpend); |
| 1395 | |
| 1396 | ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend); |
| 1397 | if (ret) |
| 1398 | mlog_errno(ret); |
| 1399 | |
| 1400 | if (tmpend < end) { |
| 1401 | /* |
| 1402 | * This may make start and end equal, but the zeroing |
| 1403 | * code will skip any work in that case so there's no |
| 1404 | * need to catch it up here. |
| 1405 | */ |
| 1406 | start = end & ~(osb->s_clustersize - 1); |
| 1407 | |
| 1408 | mlog(0, "2nd range: start: %llu, end: %llu\n", |
| 1409 | (unsigned long long)start, (unsigned long long)end); |
| 1410 | |
| 1411 | ret = ocfs2_zero_range_for_truncate(inode, handle, start, end); |
| 1412 | if (ret) |
| 1413 | mlog_errno(ret); |
| 1414 | } |
| 1415 | |
| 1416 | ocfs2_commit_trans(osb, handle); |
| 1417 | out: |
| 1418 | return ret; |
| 1419 | } |
| 1420 | |
| 1421 | static int ocfs2_remove_inode_range(struct inode *inode, |
| 1422 | struct buffer_head *di_bh, u64 byte_start, |
| 1423 | u64 byte_len) |
| 1424 | { |
Tristan Ye | e8aec06 | 2010-05-11 17:54:43 +0800 | [diff] [blame^] | 1425 | int ret = 0, flags = 0; |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1426 | u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size; |
| 1427 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1428 | struct ocfs2_cached_dealloc_ctxt dealloc; |
Mark Fasheh | b1967d0 | 2007-11-20 11:56:39 -0800 | [diff] [blame] | 1429 | struct address_space *mapping = inode->i_mapping; |
Mark Fasheh | fecc011 | 2008-11-12 15:16:38 -0800 | [diff] [blame] | 1430 | struct ocfs2_extent_tree et; |
Tristan Ye | e8aec06 | 2010-05-11 17:54:43 +0800 | [diff] [blame^] | 1431 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; |
| 1432 | u64 refcount_loc = le64_to_cpu(di->i_refcount_loc); |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1433 | |
Joel Becker | 5e404e9 | 2009-02-13 03:54:22 -0800 | [diff] [blame] | 1434 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh); |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1435 | ocfs2_init_dealloc_ctxt(&dealloc); |
| 1436 | |
| 1437 | if (byte_len == 0) |
| 1438 | return 0; |
| 1439 | |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1440 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
| 1441 | ret = ocfs2_truncate_inline(inode, di_bh, byte_start, |
Mark Fasheh | b1967d0 | 2007-11-20 11:56:39 -0800 | [diff] [blame] | 1442 | byte_start + byte_len, 0); |
| 1443 | if (ret) { |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1444 | mlog_errno(ret); |
Mark Fasheh | b1967d0 | 2007-11-20 11:56:39 -0800 | [diff] [blame] | 1445 | goto out; |
| 1446 | } |
| 1447 | /* |
| 1448 | * There's no need to get fancy with the page cache |
| 1449 | * truncate of an inline-data inode. We're talking |
| 1450 | * about less than a page here, which will be cached |
| 1451 | * in the dinode buffer anyway. |
| 1452 | */ |
| 1453 | unmap_mapping_range(mapping, 0, 0, 0); |
| 1454 | truncate_inode_pages(mapping, 0); |
| 1455 | goto out; |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1456 | } |
| 1457 | |
Tristan Ye | e8aec06 | 2010-05-11 17:54:43 +0800 | [diff] [blame^] | 1458 | /* |
| 1459 | * For reflinks, we may need to CoW 2 clusters which might be |
| 1460 | * partially zero'd later, if hole's start and end offset were |
| 1461 | * within one cluster(means is not exactly aligned to clustersize). |
| 1462 | */ |
| 1463 | |
| 1464 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) { |
| 1465 | |
| 1466 | ret = ocfs2_cow_file_pos(inode, di_bh, byte_start); |
| 1467 | if (ret) { |
| 1468 | mlog_errno(ret); |
| 1469 | goto out; |
| 1470 | } |
| 1471 | |
| 1472 | ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len); |
| 1473 | if (ret) { |
| 1474 | mlog_errno(ret); |
| 1475 | goto out; |
| 1476 | } |
| 1477 | } |
| 1478 | |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1479 | trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start); |
| 1480 | trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits; |
| 1481 | if (trunc_len >= trunc_start) |
| 1482 | trunc_len -= trunc_start; |
| 1483 | else |
| 1484 | trunc_len = 0; |
| 1485 | |
| 1486 | mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n", |
| 1487 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
| 1488 | (unsigned long long)byte_start, |
| 1489 | (unsigned long long)byte_len, trunc_start, trunc_len); |
| 1490 | |
| 1491 | ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len); |
| 1492 | if (ret) { |
| 1493 | mlog_errno(ret); |
| 1494 | goto out; |
| 1495 | } |
| 1496 | |
| 1497 | cpos = trunc_start; |
| 1498 | while (trunc_len) { |
| 1499 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, |
Tristan Ye | e8aec06 | 2010-05-11 17:54:43 +0800 | [diff] [blame^] | 1500 | &alloc_size, &flags); |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1501 | if (ret) { |
| 1502 | mlog_errno(ret); |
| 1503 | goto out; |
| 1504 | } |
| 1505 | |
| 1506 | if (alloc_size > trunc_len) |
| 1507 | alloc_size = trunc_len; |
| 1508 | |
| 1509 | /* Only do work for non-holes */ |
| 1510 | if (phys_cpos != 0) { |
Mark Fasheh | fecc011 | 2008-11-12 15:16:38 -0800 | [diff] [blame] | 1511 | ret = ocfs2_remove_btree_range(inode, &et, cpos, |
| 1512 | phys_cpos, alloc_size, |
Tristan Ye | e8aec06 | 2010-05-11 17:54:43 +0800 | [diff] [blame^] | 1513 | flags, &dealloc, |
| 1514 | refcount_loc); |
Mark Fasheh | 063c456 | 2007-07-03 13:34:11 -0700 | [diff] [blame] | 1515 | if (ret) { |
| 1516 | mlog_errno(ret); |
| 1517 | goto out; |
| 1518 | } |
| 1519 | } |
| 1520 | |
| 1521 | cpos += alloc_size; |
| 1522 | trunc_len -= alloc_size; |
| 1523 | } |
| 1524 | |
| 1525 | ocfs2_truncate_cluster_pages(inode, byte_start, byte_len); |
| 1526 | |
| 1527 | out: |
| 1528 | ocfs2_schedule_truncate_log_flush(osb, 1); |
| 1529 | ocfs2_run_deallocs(osb, &dealloc); |
| 1530 | |
| 1531 | return ret; |
| 1532 | } |
| 1533 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1534 | /* |
| 1535 | * Parts of this function taken from xfs_change_file_space() |
| 1536 | */ |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1537 | static int __ocfs2_change_file_space(struct file *file, struct inode *inode, |
| 1538 | loff_t f_pos, unsigned int cmd, |
| 1539 | struct ocfs2_space_resv *sr, |
| 1540 | int change_size) |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1541 | { |
| 1542 | int ret; |
| 1543 | s64 llen; |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1544 | loff_t size; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1545 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1546 | struct buffer_head *di_bh = NULL; |
| 1547 | handle_t *handle; |
Mark Fasheh | a00cce3 | 2007-07-20 11:28:30 -0700 | [diff] [blame] | 1548 | unsigned long long max_off = inode->i_sb->s_maxbytes; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1549 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1550 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) |
| 1551 | return -EROFS; |
| 1552 | |
| 1553 | mutex_lock(&inode->i_mutex); |
| 1554 | |
| 1555 | /* |
| 1556 | * This prevents concurrent writes on other nodes |
| 1557 | */ |
| 1558 | ret = ocfs2_rw_lock(inode, 1); |
| 1559 | if (ret) { |
| 1560 | mlog_errno(ret); |
| 1561 | goto out; |
| 1562 | } |
| 1563 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1564 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1565 | if (ret) { |
| 1566 | mlog_errno(ret); |
| 1567 | goto out_rw_unlock; |
| 1568 | } |
| 1569 | |
| 1570 | if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) { |
| 1571 | ret = -EPERM; |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1572 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1573 | } |
| 1574 | |
| 1575 | switch (sr->l_whence) { |
| 1576 | case 0: /*SEEK_SET*/ |
| 1577 | break; |
| 1578 | case 1: /*SEEK_CUR*/ |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1579 | sr->l_start += f_pos; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1580 | break; |
| 1581 | case 2: /*SEEK_END*/ |
| 1582 | sr->l_start += i_size_read(inode); |
| 1583 | break; |
| 1584 | default: |
| 1585 | ret = -EINVAL; |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1586 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1587 | } |
| 1588 | sr->l_whence = 0; |
| 1589 | |
| 1590 | llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len; |
| 1591 | |
| 1592 | if (sr->l_start < 0 |
| 1593 | || sr->l_start > max_off |
| 1594 | || (sr->l_start + llen) < 0 |
| 1595 | || (sr->l_start + llen) > max_off) { |
| 1596 | ret = -EINVAL; |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1597 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1598 | } |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1599 | size = sr->l_start + sr->l_len; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1600 | |
| 1601 | if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) { |
| 1602 | if (sr->l_len <= 0) { |
| 1603 | ret = -EINVAL; |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1604 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1605 | } |
| 1606 | } |
| 1607 | |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1608 | if (file && should_remove_suid(file->f_path.dentry)) { |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1609 | ret = __ocfs2_write_remove_suid(inode, di_bh); |
| 1610 | if (ret) { |
| 1611 | mlog_errno(ret); |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1612 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1613 | } |
| 1614 | } |
| 1615 | |
| 1616 | down_write(&OCFS2_I(inode)->ip_alloc_sem); |
| 1617 | switch (cmd) { |
| 1618 | case OCFS2_IOC_RESVSP: |
| 1619 | case OCFS2_IOC_RESVSP64: |
| 1620 | /* |
| 1621 | * This takes unsigned offsets, but the signed ones we |
| 1622 | * pass have been checked against overflow above. |
| 1623 | */ |
| 1624 | ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start, |
| 1625 | sr->l_len); |
| 1626 | break; |
| 1627 | case OCFS2_IOC_UNRESVSP: |
| 1628 | case OCFS2_IOC_UNRESVSP64: |
| 1629 | ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start, |
| 1630 | sr->l_len); |
| 1631 | break; |
| 1632 | default: |
| 1633 | ret = -EINVAL; |
| 1634 | } |
| 1635 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
| 1636 | if (ret) { |
| 1637 | mlog_errno(ret); |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1638 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1639 | } |
| 1640 | |
| 1641 | /* |
| 1642 | * We update c/mtime for these changes |
| 1643 | */ |
| 1644 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
| 1645 | if (IS_ERR(handle)) { |
| 1646 | ret = PTR_ERR(handle); |
| 1647 | mlog_errno(ret); |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1648 | goto out_inode_unlock; |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1649 | } |
| 1650 | |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1651 | if (change_size && i_size_read(inode) < size) |
| 1652 | i_size_write(inode, size); |
| 1653 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1654 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
| 1655 | ret = ocfs2_mark_inode_dirty(handle, inode, di_bh); |
| 1656 | if (ret < 0) |
| 1657 | mlog_errno(ret); |
| 1658 | |
| 1659 | ocfs2_commit_trans(osb, handle); |
| 1660 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1661 | out_inode_unlock: |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1662 | brelse(di_bh); |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1663 | ocfs2_inode_unlock(inode, 1); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1664 | out_rw_unlock: |
| 1665 | ocfs2_rw_unlock(inode, 1); |
| 1666 | |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1667 | out: |
Julia Lawall | c259ae5 | 2008-07-21 09:59:15 +0200 | [diff] [blame] | 1668 | mutex_unlock(&inode->i_mutex); |
Mark Fasheh | b258010 | 2007-03-09 16:53:21 -0800 | [diff] [blame] | 1669 | return ret; |
| 1670 | } |
| 1671 | |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1672 | int ocfs2_change_file_space(struct file *file, unsigned int cmd, |
| 1673 | struct ocfs2_space_resv *sr) |
| 1674 | { |
| 1675 | struct inode *inode = file->f_path.dentry->d_inode; |
Fernando Carrijo | c19a28e | 2009-01-07 18:09:08 -0800 | [diff] [blame] | 1676 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 1677 | |
| 1678 | if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) && |
| 1679 | !ocfs2_writes_unwritten_extents(osb)) |
| 1680 | return -ENOTTY; |
| 1681 | else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) && |
| 1682 | !ocfs2_sparse_alloc(osb)) |
| 1683 | return -ENOTTY; |
| 1684 | |
| 1685 | if (!S_ISREG(inode->i_mode)) |
| 1686 | return -EINVAL; |
| 1687 | |
| 1688 | if (!(file->f_mode & FMODE_WRITE)) |
| 1689 | return -EBADF; |
| 1690 | |
| 1691 | return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0); |
| 1692 | } |
| 1693 | |
| 1694 | static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset, |
| 1695 | loff_t len) |
| 1696 | { |
| 1697 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1698 | struct ocfs2_space_resv sr; |
| 1699 | int change_size = 1; |
| 1700 | |
| 1701 | if (!ocfs2_writes_unwritten_extents(osb)) |
| 1702 | return -EOPNOTSUPP; |
| 1703 | |
| 1704 | if (S_ISDIR(inode->i_mode)) |
| 1705 | return -ENODEV; |
| 1706 | |
| 1707 | if (mode & FALLOC_FL_KEEP_SIZE) |
| 1708 | change_size = 0; |
| 1709 | |
| 1710 | sr.l_whence = 0; |
| 1711 | sr.l_start = (s64)offset; |
| 1712 | sr.l_len = (s64)len; |
| 1713 | |
| 1714 | return __ocfs2_change_file_space(NULL, inode, offset, |
| 1715 | OCFS2_IOC_RESVSP64, &sr, change_size); |
| 1716 | } |
| 1717 | |
Tao Ma | 293b2f7 | 2009-08-25 08:02:48 +0800 | [diff] [blame] | 1718 | int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos, |
| 1719 | size_t count) |
| 1720 | { |
| 1721 | int ret = 0; |
| 1722 | unsigned int extent_flags; |
| 1723 | u32 cpos, clusters, extent_len, phys_cpos; |
| 1724 | struct super_block *sb = inode->i_sb; |
| 1725 | |
| 1726 | if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) || |
Tao Ma | 2f48d59 | 2009-10-15 11:10:49 +0800 | [diff] [blame] | 1727 | !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) || |
| 1728 | OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) |
Tao Ma | 293b2f7 | 2009-08-25 08:02:48 +0800 | [diff] [blame] | 1729 | return 0; |
| 1730 | |
| 1731 | cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits; |
| 1732 | clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos; |
| 1733 | |
| 1734 | while (clusters) { |
| 1735 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len, |
| 1736 | &extent_flags); |
| 1737 | if (ret < 0) { |
| 1738 | mlog_errno(ret); |
| 1739 | goto out; |
| 1740 | } |
| 1741 | |
| 1742 | if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) { |
| 1743 | ret = 1; |
| 1744 | break; |
| 1745 | } |
| 1746 | |
| 1747 | if (extent_len > clusters) |
| 1748 | extent_len = clusters; |
| 1749 | |
| 1750 | clusters -= extent_len; |
| 1751 | cpos += extent_len; |
| 1752 | } |
| 1753 | out: |
| 1754 | return ret; |
| 1755 | } |
| 1756 | |
| 1757 | static int ocfs2_prepare_inode_for_refcount(struct inode *inode, |
| 1758 | loff_t pos, size_t count, |
| 1759 | int *meta_level) |
| 1760 | { |
| 1761 | int ret; |
| 1762 | struct buffer_head *di_bh = NULL; |
| 1763 | u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits; |
| 1764 | u32 clusters = |
| 1765 | ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos; |
| 1766 | |
| 1767 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
| 1768 | if (ret) { |
| 1769 | mlog_errno(ret); |
| 1770 | goto out; |
| 1771 | } |
| 1772 | |
| 1773 | *meta_level = 1; |
| 1774 | |
Tao Ma | 37f8a2b | 2009-08-26 09:47:28 +0800 | [diff] [blame] | 1775 | ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX); |
Tao Ma | 293b2f7 | 2009-08-25 08:02:48 +0800 | [diff] [blame] | 1776 | if (ret) |
| 1777 | mlog_errno(ret); |
| 1778 | out: |
| 1779 | brelse(di_bh); |
| 1780 | return ret; |
| 1781 | } |
| 1782 | |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1783 | static int ocfs2_prepare_inode_for_write(struct dentry *dentry, |
| 1784 | loff_t *ppos, |
| 1785 | size_t count, |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1786 | int appending, |
Tao Ma | 86470e9 | 2009-12-03 21:55:05 +0800 | [diff] [blame] | 1787 | int *direct_io, |
| 1788 | int *has_refcount) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1789 | { |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1790 | int ret = 0, meta_level = 0; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1791 | struct inode *inode = dentry->d_inode; |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1792 | loff_t saved_pos, end; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1793 | |
Sunil Mushran | 2bd6321 | 2010-01-25 16:57:38 -0800 | [diff] [blame] | 1794 | /* |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1795 | * We start with a read level meta lock and only jump to an ex |
| 1796 | * if we need to make modifications here. |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1797 | */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1798 | for(;;) { |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1799 | ret = ocfs2_inode_lock(inode, NULL, meta_level); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1800 | if (ret < 0) { |
| 1801 | meta_level = -1; |
| 1802 | mlog_errno(ret); |
| 1803 | goto out; |
| 1804 | } |
| 1805 | |
| 1806 | /* Clear suid / sgid if necessary. We do this here |
| 1807 | * instead of later in the write path because |
| 1808 | * remove_suid() calls ->setattr without any hint that |
| 1809 | * we may have already done our cluster locking. Since |
| 1810 | * ocfs2_setattr() *must* take cluster locks to |
| 1811 | * proceeed, this will lead us to recursively lock the |
| 1812 | * inode. There's also the dinode i_size state which |
| 1813 | * can be lost via setattr during extending writes (we |
| 1814 | * set inode->i_size at the end of a write. */ |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1815 | if (should_remove_suid(dentry)) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1816 | if (meta_level == 0) { |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1817 | ocfs2_inode_unlock(inode, meta_level); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1818 | meta_level = 1; |
| 1819 | continue; |
| 1820 | } |
| 1821 | |
| 1822 | ret = ocfs2_write_remove_suid(inode); |
| 1823 | if (ret < 0) { |
| 1824 | mlog_errno(ret); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1825 | goto out_unlock; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1826 | } |
| 1827 | } |
| 1828 | |
| 1829 | /* work on a copy of ppos until we're sure that we won't have |
| 1830 | * to recalculate it due to relocking. */ |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1831 | if (appending) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1832 | saved_pos = i_size_read(inode); |
| 1833 | mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos); |
| 1834 | } else { |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1835 | saved_pos = *ppos; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1836 | } |
Mark Fasheh | 3a0782d | 2007-01-17 12:53:31 -0800 | [diff] [blame] | 1837 | |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1838 | end = saved_pos + count; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1839 | |
Tao Ma | 293b2f7 | 2009-08-25 08:02:48 +0800 | [diff] [blame] | 1840 | ret = ocfs2_check_range_for_refcount(inode, saved_pos, count); |
| 1841 | if (ret == 1) { |
| 1842 | ocfs2_inode_unlock(inode, meta_level); |
| 1843 | meta_level = -1; |
| 1844 | |
| 1845 | ret = ocfs2_prepare_inode_for_refcount(inode, |
| 1846 | saved_pos, |
| 1847 | count, |
| 1848 | &meta_level); |
Tao Ma | 86470e9 | 2009-12-03 21:55:05 +0800 | [diff] [blame] | 1849 | if (has_refcount) |
| 1850 | *has_refcount = 1; |
Wengang Wang | 96a1cc7 | 2010-02-09 14:57:45 +0800 | [diff] [blame] | 1851 | if (direct_io) |
| 1852 | *direct_io = 0; |
Tao Ma | 293b2f7 | 2009-08-25 08:02:48 +0800 | [diff] [blame] | 1853 | } |
| 1854 | |
| 1855 | if (ret < 0) { |
| 1856 | mlog_errno(ret); |
| 1857 | goto out_unlock; |
| 1858 | } |
| 1859 | |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1860 | /* |
| 1861 | * Skip the O_DIRECT checks if we don't need |
| 1862 | * them. |
| 1863 | */ |
| 1864 | if (!direct_io || !(*direct_io)) |
| 1865 | break; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1866 | |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1867 | /* |
Mark Fasheh | 1afc32b | 2007-09-07 14:46:51 -0700 | [diff] [blame] | 1868 | * There's no sane way to do direct writes to an inode |
| 1869 | * with inline data. |
| 1870 | */ |
| 1871 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
| 1872 | *direct_io = 0; |
| 1873 | break; |
| 1874 | } |
| 1875 | |
| 1876 | /* |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1877 | * Allowing concurrent direct writes means |
| 1878 | * i_size changes wouldn't be synchronized, so |
| 1879 | * one node could wind up truncating another |
| 1880 | * nodes writes. |
| 1881 | */ |
| 1882 | if (end > i_size_read(inode)) { |
| 1883 | *direct_io = 0; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1884 | break; |
| 1885 | } |
| 1886 | |
Mark Fasheh | 3a0782d | 2007-01-17 12:53:31 -0800 | [diff] [blame] | 1887 | /* |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1888 | * We don't fill holes during direct io, so |
| 1889 | * check for them here. If any are found, the |
| 1890 | * caller will have to retake some cluster |
| 1891 | * locks and initiate the io as buffered. |
Mark Fasheh | 3a0782d | 2007-01-17 12:53:31 -0800 | [diff] [blame] | 1892 | */ |
Mark Fasheh | 65ed39d | 2007-08-28 17:13:23 -0700 | [diff] [blame] | 1893 | ret = ocfs2_check_range_for_holes(inode, saved_pos, count); |
| 1894 | if (ret == 1) { |
| 1895 | *direct_io = 0; |
| 1896 | ret = 0; |
| 1897 | } else if (ret < 0) |
| 1898 | mlog_errno(ret); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1899 | break; |
| 1900 | } |
| 1901 | |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1902 | if (appending) |
| 1903 | *ppos = saved_pos; |
| 1904 | |
| 1905 | out_unlock: |
Tao Ma | 293b2f7 | 2009-08-25 08:02:48 +0800 | [diff] [blame] | 1906 | if (meta_level >= 0) |
| 1907 | ocfs2_inode_unlock(inode, meta_level); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1908 | |
| 1909 | out: |
| 1910 | return ret; |
| 1911 | } |
| 1912 | |
| 1913 | static ssize_t ocfs2_file_aio_write(struct kiocb *iocb, |
| 1914 | const struct iovec *iov, |
| 1915 | unsigned long nr_segs, |
| 1916 | loff_t pos) |
| 1917 | { |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1918 | int ret, direct_io, appending, rw_level, have_alloc_sem = 0; |
Tao Ma | 86470e9 | 2009-12-03 21:55:05 +0800 | [diff] [blame] | 1919 | int can_do_direct, has_refcount = 0; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1920 | ssize_t written = 0; |
| 1921 | size_t ocount; /* original count */ |
| 1922 | size_t count; /* after file limit checks */ |
Mark Fasheh | 9ea2d32 | 2007-10-18 14:14:45 -0700 | [diff] [blame] | 1923 | loff_t old_size, *ppos = &iocb->ki_pos; |
| 1924 | u32 old_clusters; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1925 | struct file *file = iocb->ki_filp; |
| 1926 | struct inode *inode = file->f_path.dentry->d_inode; |
Mark Fasheh | 9ea2d32 | 2007-10-18 14:14:45 -0700 | [diff] [blame] | 1927 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1928 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1929 | mlog_entry("(0x%p, %u, '%.*s')\n", file, |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1930 | (unsigned int)nr_segs, |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1931 | file->f_path.dentry->d_name.len, |
| 1932 | file->f_path.dentry->d_name.name); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1933 | |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1934 | if (iocb->ki_left == 0) |
| 1935 | return 0; |
| 1936 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1937 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); |
| 1938 | |
| 1939 | appending = file->f_flags & O_APPEND ? 1 : 0; |
| 1940 | direct_io = file->f_flags & O_DIRECT ? 1 : 0; |
| 1941 | |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1942 | mutex_lock(&inode->i_mutex); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1943 | |
| 1944 | relock: |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1945 | /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */ |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1946 | if (direct_io) { |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1947 | down_read(&inode->i_alloc_sem); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1948 | have_alloc_sem = 1; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1949 | } |
| 1950 | |
| 1951 | /* concurrent O_DIRECT writes are allowed */ |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1952 | rw_level = !direct_io; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1953 | ret = ocfs2_rw_lock(inode, rw_level); |
| 1954 | if (ret < 0) { |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1955 | mlog_errno(ret); |
| 1956 | goto out_sems; |
| 1957 | } |
| 1958 | |
| 1959 | can_do_direct = direct_io; |
| 1960 | ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos, |
| 1961 | iocb->ki_left, appending, |
Tao Ma | 86470e9 | 2009-12-03 21:55:05 +0800 | [diff] [blame] | 1962 | &can_do_direct, &has_refcount); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1963 | if (ret < 0) { |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1964 | mlog_errno(ret); |
| 1965 | goto out; |
| 1966 | } |
| 1967 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1968 | /* |
| 1969 | * We can't complete the direct I/O as requested, fall back to |
| 1970 | * buffered I/O. |
| 1971 | */ |
| 1972 | if (direct_io && !can_do_direct) { |
| 1973 | ocfs2_rw_unlock(inode, rw_level); |
| 1974 | up_read(&inode->i_alloc_sem); |
| 1975 | |
| 1976 | have_alloc_sem = 0; |
| 1977 | rw_level = -1; |
| 1978 | |
| 1979 | direct_io = 0; |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1980 | goto relock; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 1981 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1982 | |
Mark Fasheh | 9ea2d32 | 2007-10-18 14:14:45 -0700 | [diff] [blame] | 1983 | /* |
| 1984 | * To later detect whether a journal commit for sync writes is |
| 1985 | * necessary, we sample i_size, and cluster count here. |
| 1986 | */ |
| 1987 | old_size = i_size_read(inode); |
| 1988 | old_clusters = OCFS2_I(inode)->ip_clusters; |
| 1989 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1990 | /* communicate with ocfs2_dio_end_io */ |
Mark Fasheh | 7cdfc3a | 2007-04-16 17:28:51 -0700 | [diff] [blame] | 1991 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1992 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 1993 | if (direct_io) { |
Nick Piggin | b6af1bc | 2007-10-16 01:25:24 -0700 | [diff] [blame] | 1994 | ret = generic_segment_checks(iov, &nr_segs, &ocount, |
| 1995 | VERIFY_READ); |
| 1996 | if (ret) |
| 1997 | goto out_dio; |
| 1998 | |
Goldwyn Rodrigues | cefcb80 | 2009-07-11 10:57:27 -0500 | [diff] [blame] | 1999 | count = ocount; |
Nick Piggin | b6af1bc | 2007-10-16 01:25:24 -0700 | [diff] [blame] | 2000 | ret = generic_write_checks(file, ppos, &count, |
| 2001 | S_ISBLK(inode->i_mode)); |
| 2002 | if (ret) |
| 2003 | goto out_dio; |
| 2004 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 2005 | written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos, |
| 2006 | ppos, count, ocount); |
| 2007 | if (written < 0) { |
Dmitri Monakhov | c435400 | 2008-10-27 13:01:49 -0700 | [diff] [blame] | 2008 | /* |
| 2009 | * direct write may have instantiated a few |
| 2010 | * blocks outside i_size. Trim these off again. |
| 2011 | * Don't need i_size_read because we hold i_mutex. |
| 2012 | */ |
| 2013 | if (*ppos + count > inode->i_size) |
| 2014 | vmtruncate(inode, inode->i_size); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 2015 | ret = written; |
| 2016 | goto out_dio; |
| 2017 | } |
| 2018 | } else { |
Jan Kara | 918941a | 2009-08-17 18:50:08 +0200 | [diff] [blame] | 2019 | written = __generic_file_aio_write(iocb, iov, nr_segs, ppos); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 2020 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2021 | |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 2022 | out_dio: |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2023 | /* buffered aio wouldn't have proper lock coverage today */ |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 2024 | BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2025 | |
Tao Ma | 60c4867 | 2010-02-03 09:56:04 +0800 | [diff] [blame] | 2026 | if (((file->f_flags & O_DSYNC) && !direct_io) || IS_SYNC(inode) || |
| 2027 | ((file->f_flags & O_DIRECT) && has_refcount)) { |
Jan Kara | 918941a | 2009-08-17 18:50:08 +0200 | [diff] [blame] | 2028 | ret = filemap_fdatawrite_range(file->f_mapping, pos, |
| 2029 | pos + count - 1); |
| 2030 | if (ret < 0) |
| 2031 | written = ret; |
| 2032 | |
| 2033 | if (!ret && (old_size != i_size_read(inode) || |
Tao Ma | 86470e9 | 2009-12-03 21:55:05 +0800 | [diff] [blame] | 2034 | old_clusters != OCFS2_I(inode)->ip_clusters || |
| 2035 | has_refcount)) { |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 2036 | ret = jbd2_journal_force_commit(osb->journal->j_journal); |
Mark Fasheh | 9ea2d32 | 2007-10-18 14:14:45 -0700 | [diff] [blame] | 2037 | if (ret < 0) |
| 2038 | written = ret; |
| 2039 | } |
Jan Kara | 918941a | 2009-08-17 18:50:08 +0200 | [diff] [blame] | 2040 | |
| 2041 | if (!ret) |
| 2042 | ret = filemap_fdatawait_range(file->f_mapping, pos, |
| 2043 | pos + count - 1); |
Mark Fasheh | 9ea2d32 | 2007-10-18 14:14:45 -0700 | [diff] [blame] | 2044 | } |
| 2045 | |
Sunil Mushran | 2bd6321 | 2010-01-25 16:57:38 -0800 | [diff] [blame] | 2046 | /* |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2047 | * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io |
| 2048 | * function pointer which is called when o_direct io completes so that |
| 2049 | * it can unlock our rw lock. (it's the clustered equivalent of |
| 2050 | * i_alloc_sem; protects truncate from racing with pending ios). |
| 2051 | * Unfortunately there are error cases which call end_io and others |
| 2052 | * that don't. so we don't have to unlock the rw_lock if either an |
| 2053 | * async dio is going to do it in the future or an end_io after an |
| 2054 | * error has already done it. |
| 2055 | */ |
Coly Li | 66b116c | 2010-02-25 14:57:13 +0800 | [diff] [blame] | 2056 | if ((ret == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2057 | rw_level = -1; |
| 2058 | have_alloc_sem = 0; |
| 2059 | } |
| 2060 | |
| 2061 | out: |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 2062 | if (rw_level != -1) |
| 2063 | ocfs2_rw_unlock(inode, rw_level); |
| 2064 | |
| 2065 | out_sems: |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2066 | if (have_alloc_sem) |
| 2067 | up_read(&inode->i_alloc_sem); |
Mark Fasheh | 9517bac | 2007-02-09 20:24:12 -0800 | [diff] [blame] | 2068 | |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 2069 | mutex_unlock(&inode->i_mutex); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2070 | |
Wengang Wang | 812e7a6 | 2009-07-10 13:26:04 +0800 | [diff] [blame] | 2071 | if (written) |
| 2072 | ret = written; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2073 | mlog_exit(ret); |
Wengang Wang | 812e7a6 | 2009-07-10 13:26:04 +0800 | [diff] [blame] | 2074 | return ret; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2075 | } |
| 2076 | |
Miklos Szeredi | 328eaab | 2009-04-14 19:48:39 +0200 | [diff] [blame] | 2077 | static int ocfs2_splice_to_file(struct pipe_inode_info *pipe, |
| 2078 | struct file *out, |
| 2079 | struct splice_desc *sd) |
| 2080 | { |
| 2081 | int ret; |
| 2082 | |
| 2083 | ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, &sd->pos, |
Tao Ma | 86470e9 | 2009-12-03 21:55:05 +0800 | [diff] [blame] | 2084 | sd->total_len, 0, NULL, NULL); |
Miklos Szeredi | 328eaab | 2009-04-14 19:48:39 +0200 | [diff] [blame] | 2085 | if (ret < 0) { |
| 2086 | mlog_errno(ret); |
| 2087 | return ret; |
| 2088 | } |
| 2089 | |
| 2090 | return splice_from_pipe_feed(pipe, sd, pipe_to_file); |
| 2091 | } |
| 2092 | |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 2093 | static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe, |
| 2094 | struct file *out, |
| 2095 | loff_t *ppos, |
| 2096 | size_t len, |
| 2097 | unsigned int flags) |
| 2098 | { |
| 2099 | int ret; |
Miklos Szeredi | 328eaab | 2009-04-14 19:48:39 +0200 | [diff] [blame] | 2100 | struct address_space *mapping = out->f_mapping; |
| 2101 | struct inode *inode = mapping->host; |
| 2102 | struct splice_desc sd = { |
| 2103 | .total_len = len, |
| 2104 | .flags = flags, |
| 2105 | .pos = *ppos, |
| 2106 | .u.file = out, |
| 2107 | }; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 2108 | |
| 2109 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe, |
| 2110 | (unsigned int)len, |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 2111 | out->f_path.dentry->d_name.len, |
| 2112 | out->f_path.dentry->d_name.name); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 2113 | |
Miklos Szeredi | 7bfac9e | 2009-04-06 17:41:00 +0200 | [diff] [blame] | 2114 | if (pipe->inode) |
Miklos Szeredi | 328eaab | 2009-04-14 19:48:39 +0200 | [diff] [blame] | 2115 | mutex_lock_nested(&pipe->inode->i_mutex, I_MUTEX_PARENT); |
| 2116 | |
| 2117 | splice_from_pipe_begin(&sd); |
| 2118 | do { |
| 2119 | ret = splice_from_pipe_next(pipe, &sd); |
| 2120 | if (ret <= 0) |
| 2121 | break; |
| 2122 | |
| 2123 | mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD); |
| 2124 | ret = ocfs2_rw_lock(inode, 1); |
| 2125 | if (ret < 0) |
| 2126 | mlog_errno(ret); |
| 2127 | else { |
| 2128 | ret = ocfs2_splice_to_file(pipe, out, &sd); |
| 2129 | ocfs2_rw_unlock(inode, 1); |
| 2130 | } |
| 2131 | mutex_unlock(&inode->i_mutex); |
| 2132 | } while (ret > 0); |
| 2133 | splice_from_pipe_end(pipe, &sd); |
| 2134 | |
Miklos Szeredi | 7bfac9e | 2009-04-06 17:41:00 +0200 | [diff] [blame] | 2135 | if (pipe->inode) |
| 2136 | mutex_unlock(&pipe->inode->i_mutex); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 2137 | |
Miklos Szeredi | 328eaab | 2009-04-14 19:48:39 +0200 | [diff] [blame] | 2138 | if (sd.num_spliced) |
| 2139 | ret = sd.num_spliced; |
| 2140 | |
| 2141 | if (ret > 0) { |
| 2142 | unsigned long nr_pages; |
Jan Kara | d23c937 | 2009-08-18 18:24:31 +0200 | [diff] [blame] | 2143 | int err; |
Miklos Szeredi | 328eaab | 2009-04-14 19:48:39 +0200 | [diff] [blame] | 2144 | |
Miklos Szeredi | 328eaab | 2009-04-14 19:48:39 +0200 | [diff] [blame] | 2145 | nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
| 2146 | |
Jan Kara | d23c937 | 2009-08-18 18:24:31 +0200 | [diff] [blame] | 2147 | err = generic_write_sync(out, *ppos, ret); |
| 2148 | if (err) |
| 2149 | ret = err; |
| 2150 | else |
| 2151 | *ppos += ret; |
Miklos Szeredi | 328eaab | 2009-04-14 19:48:39 +0200 | [diff] [blame] | 2152 | |
Miklos Szeredi | 328eaab | 2009-04-14 19:48:39 +0200 | [diff] [blame] | 2153 | balance_dirty_pages_ratelimited_nr(mapping, nr_pages); |
| 2154 | } |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 2155 | |
| 2156 | mlog_exit(ret); |
| 2157 | return ret; |
| 2158 | } |
| 2159 | |
| 2160 | static ssize_t ocfs2_file_splice_read(struct file *in, |
| 2161 | loff_t *ppos, |
| 2162 | struct pipe_inode_info *pipe, |
| 2163 | size_t len, |
| 2164 | unsigned int flags) |
| 2165 | { |
Tao Ma | 1962f39 | 2009-06-19 15:36:52 +0800 | [diff] [blame] | 2166 | int ret = 0, lock_level = 0; |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 2167 | struct inode *inode = in->f_path.dentry->d_inode; |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 2168 | |
| 2169 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe, |
| 2170 | (unsigned int)len, |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 2171 | in->f_path.dentry->d_name.len, |
| 2172 | in->f_path.dentry->d_name.name); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 2173 | |
| 2174 | /* |
| 2175 | * See the comment in ocfs2_file_aio_read() |
| 2176 | */ |
Tao Ma | 1962f39 | 2009-06-19 15:36:52 +0800 | [diff] [blame] | 2177 | ret = ocfs2_inode_lock_atime(inode, in->f_vfsmnt, &lock_level); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 2178 | if (ret < 0) { |
| 2179 | mlog_errno(ret); |
| 2180 | goto bail; |
| 2181 | } |
Tao Ma | 1962f39 | 2009-06-19 15:36:52 +0800 | [diff] [blame] | 2182 | ocfs2_inode_unlock(inode, lock_level); |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 2183 | |
| 2184 | ret = generic_file_splice_read(in, ppos, pipe, len, flags); |
| 2185 | |
| 2186 | bail: |
| 2187 | mlog_exit(ret); |
| 2188 | return ret; |
| 2189 | } |
| 2190 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2191 | static ssize_t ocfs2_file_aio_read(struct kiocb *iocb, |
Badari Pulavarty | 027445c | 2006-09-30 23:28:46 -0700 | [diff] [blame] | 2192 | const struct iovec *iov, |
| 2193 | unsigned long nr_segs, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2194 | loff_t pos) |
| 2195 | { |
Tiger Yang | 25899de | 2006-11-15 15:49:02 +0800 | [diff] [blame] | 2196 | 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] | 2197 | struct file *filp = iocb->ki_filp; |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 2198 | struct inode *inode = filp->f_path.dentry->d_inode; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2199 | |
Badari Pulavarty | 027445c | 2006-09-30 23:28:46 -0700 | [diff] [blame] | 2200 | mlog_entry("(0x%p, %u, '%.*s')\n", filp, |
| 2201 | (unsigned int)nr_segs, |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 2202 | filp->f_path.dentry->d_name.len, |
| 2203 | filp->f_path.dentry->d_name.name); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2204 | |
| 2205 | if (!inode) { |
| 2206 | ret = -EINVAL; |
| 2207 | mlog_errno(ret); |
| 2208 | goto bail; |
| 2209 | } |
| 2210 | |
Sunil Mushran | 2bd6321 | 2010-01-25 16:57:38 -0800 | [diff] [blame] | 2211 | /* |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2212 | * buffered reads protect themselves in ->readpage(). O_DIRECT reads |
| 2213 | * need locks to protect pending reads from racing with truncate. |
| 2214 | */ |
| 2215 | if (filp->f_flags & O_DIRECT) { |
| 2216 | down_read(&inode->i_alloc_sem); |
| 2217 | have_alloc_sem = 1; |
| 2218 | |
| 2219 | ret = ocfs2_rw_lock(inode, 0); |
| 2220 | if (ret < 0) { |
| 2221 | mlog_errno(ret); |
| 2222 | goto bail; |
| 2223 | } |
| 2224 | rw_level = 0; |
| 2225 | /* communicate with ocfs2_dio_end_io */ |
Mark Fasheh | 7cdfc3a | 2007-04-16 17:28:51 -0700 | [diff] [blame] | 2226 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2227 | } |
| 2228 | |
Mark Fasheh | c4374f8 | 2006-05-05 19:04:35 -0700 | [diff] [blame] | 2229 | /* |
| 2230 | * We're fine letting folks race truncates and extending |
| 2231 | * writes with read across the cluster, just like they can |
| 2232 | * locally. Hence no rw_lock during read. |
Sunil Mushran | 2bd6321 | 2010-01-25 16:57:38 -0800 | [diff] [blame] | 2233 | * |
Mark Fasheh | c4374f8 | 2006-05-05 19:04:35 -0700 | [diff] [blame] | 2234 | * Take and drop the meta data lock to update inode fields |
| 2235 | * like i_size. This allows the checks down below |
Sunil Mushran | 2bd6321 | 2010-01-25 16:57:38 -0800 | [diff] [blame] | 2236 | * generic_file_aio_read() a chance of actually working. |
Mark Fasheh | c4374f8 | 2006-05-05 19:04:35 -0700 | [diff] [blame] | 2237 | */ |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 2238 | ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level); |
Mark Fasheh | c4374f8 | 2006-05-05 19:04:35 -0700 | [diff] [blame] | 2239 | if (ret < 0) { |
| 2240 | mlog_errno(ret); |
| 2241 | goto bail; |
| 2242 | } |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 2243 | ocfs2_inode_unlock(inode, lock_level); |
Mark Fasheh | c4374f8 | 2006-05-05 19:04:35 -0700 | [diff] [blame] | 2244 | |
Badari Pulavarty | 027445c | 2006-09-30 23:28:46 -0700 | [diff] [blame] | 2245 | ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2246 | if (ret == -EINVAL) |
Sunil Mushran | 56753bd | 2008-06-09 11:24:41 -0700 | [diff] [blame] | 2247 | mlog(0, "generic_file_aio_read returned -EINVAL\n"); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2248 | |
| 2249 | /* buffered aio wouldn't have proper lock coverage today */ |
| 2250 | BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT)); |
| 2251 | |
| 2252 | /* see ocfs2_file_aio_write */ |
| 2253 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { |
| 2254 | rw_level = -1; |
| 2255 | have_alloc_sem = 0; |
| 2256 | } |
| 2257 | |
| 2258 | bail: |
| 2259 | if (have_alloc_sem) |
| 2260 | up_read(&inode->i_alloc_sem); |
Sunil Mushran | 2bd6321 | 2010-01-25 16:57:38 -0800 | [diff] [blame] | 2261 | if (rw_level != -1) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2262 | ocfs2_rw_unlock(inode, rw_level); |
| 2263 | mlog_exit(ret); |
| 2264 | |
| 2265 | return ret; |
| 2266 | } |
| 2267 | |
Arjan van de Ven | 92e1d5b | 2007-02-12 00:55:39 -0800 | [diff] [blame] | 2268 | const struct inode_operations ocfs2_file_iops = { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2269 | .setattr = ocfs2_setattr, |
| 2270 | .getattr = ocfs2_getattr, |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 2271 | .permission = ocfs2_permission, |
Tiger Yang | cf1d6c7 | 2008-08-18 17:11:00 +0800 | [diff] [blame] | 2272 | .setxattr = generic_setxattr, |
| 2273 | .getxattr = generic_getxattr, |
| 2274 | .listxattr = ocfs2_listxattr, |
| 2275 | .removexattr = generic_removexattr, |
Mark Fasheh | 385820a | 2007-07-19 00:14:38 -0700 | [diff] [blame] | 2276 | .fallocate = ocfs2_fallocate, |
Mark Fasheh | 00dc417 | 2008-10-03 17:32:11 -0400 | [diff] [blame] | 2277 | .fiemap = ocfs2_fiemap, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2278 | }; |
| 2279 | |
Arjan van de Ven | 92e1d5b | 2007-02-12 00:55:39 -0800 | [diff] [blame] | 2280 | const struct inode_operations ocfs2_special_file_iops = { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2281 | .setattr = ocfs2_setattr, |
| 2282 | .getattr = ocfs2_getattr, |
Tiger Yang | d38eb8d | 2006-11-27 09:59:21 +0800 | [diff] [blame] | 2283 | .permission = ocfs2_permission, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2284 | }; |
| 2285 | |
Mark Fasheh | 53da493 | 2008-07-21 14:29:16 -0700 | [diff] [blame] | 2286 | /* |
| 2287 | * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with |
| 2288 | * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks! |
| 2289 | */ |
Arjan van de Ven | 4b6f5d2 | 2006-03-28 01:56:42 -0800 | [diff] [blame] | 2290 | const struct file_operations ocfs2_fops = { |
Jan Kara | 32c3c0e | 2007-12-19 15:24:52 +0100 | [diff] [blame] | 2291 | .llseek = generic_file_llseek, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2292 | .read = do_sync_read, |
| 2293 | .write = do_sync_write, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2294 | .mmap = ocfs2_mmap, |
| 2295 | .fsync = ocfs2_sync_file, |
| 2296 | .release = ocfs2_file_release, |
| 2297 | .open = ocfs2_file_open, |
| 2298 | .aio_read = ocfs2_file_aio_read, |
| 2299 | .aio_write = ocfs2_file_aio_write, |
Andi Kleen | c9ec148 | 2008-01-27 03:17:17 +0100 | [diff] [blame] | 2300 | .unlocked_ioctl = ocfs2_ioctl, |
Mark Fasheh | 586d232 | 2007-03-09 15:56:28 -0800 | [diff] [blame] | 2301 | #ifdef CONFIG_COMPAT |
| 2302 | .compat_ioctl = ocfs2_compat_ioctl, |
| 2303 | #endif |
Mark Fasheh | 53da493 | 2008-07-21 14:29:16 -0700 | [diff] [blame] | 2304 | .lock = ocfs2_lock, |
| 2305 | .flock = ocfs2_flock, |
| 2306 | .splice_read = ocfs2_file_splice_read, |
| 2307 | .splice_write = ocfs2_file_splice_write, |
| 2308 | }; |
| 2309 | |
| 2310 | const struct file_operations ocfs2_dops = { |
| 2311 | .llseek = generic_file_llseek, |
| 2312 | .read = generic_read_dir, |
| 2313 | .readdir = ocfs2_readdir, |
| 2314 | .fsync = ocfs2_sync_file, |
| 2315 | .release = ocfs2_dir_release, |
| 2316 | .open = ocfs2_dir_open, |
| 2317 | .unlocked_ioctl = ocfs2_ioctl, |
| 2318 | #ifdef CONFIG_COMPAT |
| 2319 | .compat_ioctl = ocfs2_compat_ioctl, |
| 2320 | #endif |
| 2321 | .lock = ocfs2_lock, |
| 2322 | .flock = ocfs2_flock, |
| 2323 | }; |
| 2324 | |
| 2325 | /* |
| 2326 | * POSIX-lockless variants of our file_operations. |
| 2327 | * |
| 2328 | * These will be used if the underlying cluster stack does not support |
| 2329 | * posix file locking, if the user passes the "localflocks" mount |
| 2330 | * option, or if we have a local-only fs. |
| 2331 | * |
| 2332 | * ocfs2_flock is in here because all stacks handle UNIX file locks, |
| 2333 | * so we still want it in the case of no stack support for |
| 2334 | * plocks. Internally, it will do the right thing when asked to ignore |
| 2335 | * the cluster. |
| 2336 | */ |
| 2337 | const struct file_operations ocfs2_fops_no_plocks = { |
| 2338 | .llseek = generic_file_llseek, |
| 2339 | .read = do_sync_read, |
| 2340 | .write = do_sync_write, |
| 2341 | .mmap = ocfs2_mmap, |
| 2342 | .fsync = ocfs2_sync_file, |
| 2343 | .release = ocfs2_file_release, |
| 2344 | .open = ocfs2_file_open, |
| 2345 | .aio_read = ocfs2_file_aio_read, |
| 2346 | .aio_write = ocfs2_file_aio_write, |
| 2347 | .unlocked_ioctl = ocfs2_ioctl, |
| 2348 | #ifdef CONFIG_COMPAT |
| 2349 | .compat_ioctl = ocfs2_compat_ioctl, |
| 2350 | #endif |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 2351 | .flock = ocfs2_flock, |
Tiger Yang | 8659ac2 | 2006-10-17 18:29:52 -0700 | [diff] [blame] | 2352 | .splice_read = ocfs2_file_splice_read, |
| 2353 | .splice_write = ocfs2_file_splice_write, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2354 | }; |
| 2355 | |
Mark Fasheh | 53da493 | 2008-07-21 14:29:16 -0700 | [diff] [blame] | 2356 | const struct file_operations ocfs2_dops_no_plocks = { |
Jan Kara | 32c3c0e | 2007-12-19 15:24:52 +0100 | [diff] [blame] | 2357 | .llseek = generic_file_llseek, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2358 | .read = generic_read_dir, |
| 2359 | .readdir = ocfs2_readdir, |
| 2360 | .fsync = ocfs2_sync_file, |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 2361 | .release = ocfs2_dir_release, |
| 2362 | .open = ocfs2_dir_open, |
Andi Kleen | c9ec148 | 2008-01-27 03:17:17 +0100 | [diff] [blame] | 2363 | .unlocked_ioctl = ocfs2_ioctl, |
Mark Fasheh | 586d232 | 2007-03-09 15:56:28 -0800 | [diff] [blame] | 2364 | #ifdef CONFIG_COMPAT |
| 2365 | .compat_ioctl = ocfs2_compat_ioctl, |
| 2366 | #endif |
Mark Fasheh | 53fc622 | 2007-12-20 16:49:04 -0800 | [diff] [blame] | 2367 | .flock = ocfs2_flock, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2368 | }; |