| 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 | |
| 26 | #include <linux/capability.h> |
| 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> |
| 33 | #include <linux/sched.h> |
| 34 | #include <linux/splice.h> |
| 35 | #include <linux/mount.h> |
| 36 | #include <linux/writeback.h> |
| 37 | #include <linux/falloc.h> |
| 38 | |
| 39 | #define MLOG_MASK_PREFIX ML_INODE |
| 40 | #include <cluster/masklog.h> |
| 41 | |
| 42 | #include "ocfs2.h" |
| 43 | |
| 44 | #include "alloc.h" |
| 45 | #include "aops.h" |
| 46 | #include "dir.h" |
| 47 | #include "dlmglue.h" |
| 48 | #include "extent_map.h" |
| 49 | #include "file.h" |
| 50 | #include "sysfile.h" |
| 51 | #include "inode.h" |
| 52 | #include "ioctl.h" |
| 53 | #include "journal.h" |
| 54 | #include "locks.h" |
| 55 | #include "mmap.h" |
| 56 | #include "suballoc.h" |
| 57 | #include "super.h" |
| 58 | #include "xattr.h" |
| 59 | |
| 60 | #include "buffer_head_io.h" |
| 61 | |
| 62 | static int ocfs2_sync_inode(struct inode *inode) |
| 63 | { |
| 64 | filemap_fdatawrite(inode->i_mapping); |
| 65 | return sync_mapping_buffers(inode->i_mapping); |
| 66 | } |
| 67 | |
| 68 | static int ocfs2_init_file_private(struct inode *inode, struct file *file) |
| 69 | { |
| 70 | struct ocfs2_file_private *fp; |
| 71 | |
| 72 | fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL); |
| 73 | if (!fp) |
| 74 | return -ENOMEM; |
| 75 | |
| 76 | fp->fp_file = file; |
| 77 | mutex_init(&fp->fp_mutex); |
| 78 | ocfs2_file_lock_res_init(&fp->fp_flock, fp); |
| 79 | file->private_data = fp; |
| 80 | |
| 81 | return 0; |
| 82 | } |
| 83 | |
| 84 | static void ocfs2_free_file_private(struct inode *inode, struct file *file) |
| 85 | { |
| 86 | struct ocfs2_file_private *fp = file->private_data; |
| 87 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 88 | |
| 89 | if (fp) { |
| 90 | ocfs2_simple_drop_lockres(osb, &fp->fp_flock); |
| 91 | ocfs2_lock_res_free(&fp->fp_flock); |
| 92 | kfree(fp); |
| 93 | file->private_data = NULL; |
| 94 | } |
| 95 | } |
| 96 | |
| 97 | static int ocfs2_file_open(struct inode *inode, struct file *file) |
| 98 | { |
| 99 | int status; |
| 100 | int mode = file->f_flags; |
| 101 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 102 | |
| 103 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, |
| 104 | file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name); |
| 105 | |
| 106 | spin_lock(&oi->ip_lock); |
| 107 | |
| 108 | /* Check that the inode hasn't been wiped from disk by another |
| 109 | * node. If it hasn't then we're safe as long as we hold the |
| 110 | * spin lock until our increment of open count. */ |
| 111 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { |
| 112 | spin_unlock(&oi->ip_lock); |
| 113 | |
| 114 | status = -ENOENT; |
| 115 | goto leave; |
| 116 | } |
| 117 | |
| 118 | if (mode & O_DIRECT) |
| 119 | oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT; |
| 120 | |
| 121 | oi->ip_open_count++; |
| 122 | spin_unlock(&oi->ip_lock); |
| 123 | |
| 124 | status = ocfs2_init_file_private(inode, file); |
| 125 | if (status) { |
| 126 | /* |
| 127 | * We want to set open count back if we're failing the |
| 128 | * open. |
| 129 | */ |
| 130 | spin_lock(&oi->ip_lock); |
| 131 | oi->ip_open_count--; |
| 132 | spin_unlock(&oi->ip_lock); |
| 133 | } |
| 134 | |
| 135 | leave: |
| 136 | mlog_exit(status); |
| 137 | return status; |
| 138 | } |
| 139 | |
| 140 | static int ocfs2_file_release(struct inode *inode, struct file *file) |
| 141 | { |
| 142 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 143 | |
| 144 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, |
| 145 | file->f_path.dentry->d_name.len, |
| 146 | file->f_path.dentry->d_name.name); |
| 147 | |
| 148 | spin_lock(&oi->ip_lock); |
| 149 | if (!--oi->ip_open_count) |
| 150 | oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT; |
| 151 | spin_unlock(&oi->ip_lock); |
| 152 | |
| 153 | ocfs2_free_file_private(inode, file); |
| 154 | |
| 155 | mlog_exit(0); |
| 156 | |
| 157 | return 0; |
| 158 | } |
| 159 | |
| 160 | static int ocfs2_dir_open(struct inode *inode, struct file *file) |
| 161 | { |
| 162 | return ocfs2_init_file_private(inode, file); |
| 163 | } |
| 164 | |
| 165 | static int ocfs2_dir_release(struct inode *inode, struct file *file) |
| 166 | { |
| 167 | ocfs2_free_file_private(inode, file); |
| 168 | return 0; |
| 169 | } |
| 170 | |
| 171 | static int ocfs2_sync_file(struct file *file, |
| 172 | struct dentry *dentry, |
| 173 | int datasync) |
| 174 | { |
| 175 | int err = 0; |
| 176 | journal_t *journal; |
| 177 | struct inode *inode = dentry->d_inode; |
| 178 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 179 | |
| 180 | mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync, |
| 181 | dentry->d_name.len, dentry->d_name.name); |
| 182 | |
| 183 | err = ocfs2_sync_inode(dentry->d_inode); |
| 184 | if (err) |
| 185 | goto bail; |
| 186 | |
| 187 | journal = osb->journal->j_journal; |
| 188 | err = journal_force_commit(journal); |
| 189 | |
| 190 | bail: |
| 191 | mlog_exit(err); |
| 192 | |
| 193 | return (err < 0) ? -EIO : 0; |
| 194 | } |
| 195 | |
| 196 | int ocfs2_should_update_atime(struct inode *inode, |
| 197 | struct vfsmount *vfsmnt) |
| 198 | { |
| 199 | struct timespec now; |
| 200 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 201 | |
| 202 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) |
| 203 | return 0; |
| 204 | |
| 205 | if ((inode->i_flags & S_NOATIME) || |
| 206 | ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))) |
| 207 | return 0; |
| 208 | |
| 209 | /* |
| 210 | * We can be called with no vfsmnt structure - NFSD will |
| 211 | * sometimes do this. |
| 212 | * |
| 213 | * Note that our action here is different than touch_atime() - |
| 214 | * if we can't tell whether this is a noatime mount, then we |
| 215 | * don't know whether to trust the value of s_atime_quantum. |
| 216 | */ |
| 217 | if (vfsmnt == NULL) |
| 218 | return 0; |
| 219 | |
| 220 | if ((vfsmnt->mnt_flags & MNT_NOATIME) || |
| 221 | ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) |
| 222 | return 0; |
| 223 | |
| 224 | if (vfsmnt->mnt_flags & MNT_RELATIME) { |
| 225 | if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) || |
| 226 | (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0)) |
| 227 | return 1; |
| 228 | |
| 229 | return 0; |
| 230 | } |
| 231 | |
| 232 | now = CURRENT_TIME; |
| 233 | if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum)) |
| 234 | return 0; |
| 235 | else |
| 236 | return 1; |
| 237 | } |
| 238 | |
| 239 | int ocfs2_update_inode_atime(struct inode *inode, |
| 240 | struct buffer_head *bh) |
| 241 | { |
| 242 | int ret; |
| 243 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 244 | handle_t *handle; |
| 245 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data; |
| 246 | |
| 247 | mlog_entry_void(); |
| 248 | |
| 249 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
| 250 | if (handle == NULL) { |
| 251 | ret = -ENOMEM; |
| 252 | mlog_errno(ret); |
| 253 | goto out; |
| 254 | } |
| 255 | |
| 256 | ret = ocfs2_journal_access(handle, inode, bh, |
| 257 | OCFS2_JOURNAL_ACCESS_WRITE); |
| 258 | if (ret) { |
| 259 | mlog_errno(ret); |
| 260 | goto out_commit; |
| 261 | } |
| 262 | |
| 263 | /* |
| 264 | * Don't use ocfs2_mark_inode_dirty() here as we don't always |
| 265 | * have i_mutex to guard against concurrent changes to other |
| 266 | * inode fields. |
| 267 | */ |
| 268 | inode->i_atime = CURRENT_TIME; |
| 269 | di->i_atime = cpu_to_le64(inode->i_atime.tv_sec); |
| 270 | di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec); |
| 271 | |
| 272 | ret = ocfs2_journal_dirty(handle, bh); |
| 273 | if (ret < 0) |
| 274 | mlog_errno(ret); |
| 275 | |
| 276 | out_commit: |
| 277 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); |
| 278 | out: |
| 279 | mlog_exit(ret); |
| 280 | return ret; |
| 281 | } |
| 282 | |
| 283 | static int ocfs2_set_inode_size(handle_t *handle, |
| 284 | struct inode *inode, |
| 285 | struct buffer_head *fe_bh, |
| 286 | u64 new_i_size) |
| 287 | { |
| 288 | int status; |
| 289 | |
| 290 | mlog_entry_void(); |
| 291 | i_size_write(inode, new_i_size); |
| 292 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
| 293 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
| 294 | |
| 295 | status = ocfs2_mark_inode_dirty(handle, inode, fe_bh); |
| 296 | if (status < 0) { |
| 297 | mlog_errno(status); |
| 298 | goto bail; |
| 299 | } |
| 300 | |
| 301 | bail: |
| 302 | mlog_exit(status); |
| 303 | return status; |
| 304 | } |
| 305 | |
| 306 | static int ocfs2_simple_size_update(struct inode *inode, |
| 307 | struct buffer_head *di_bh, |
| 308 | u64 new_i_size) |
| 309 | { |
| 310 | int ret; |
| 311 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 312 | handle_t *handle = NULL; |
| 313 | |
| 314 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
| 315 | if (handle == NULL) { |
| 316 | ret = -ENOMEM; |
| 317 | mlog_errno(ret); |
| 318 | goto out; |
| 319 | } |
| 320 | |
| 321 | ret = ocfs2_set_inode_size(handle, inode, di_bh, |
| 322 | new_i_size); |
| 323 | if (ret < 0) |
| 324 | mlog_errno(ret); |
| 325 | |
| 326 | ocfs2_commit_trans(osb, handle); |
| 327 | out: |
| 328 | return ret; |
| 329 | } |
| 330 | |
| 331 | static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb, |
| 332 | struct inode *inode, |
| 333 | struct buffer_head *fe_bh, |
| 334 | u64 new_i_size) |
| 335 | { |
| 336 | int status; |
| 337 | handle_t *handle; |
| 338 | struct ocfs2_dinode *di; |
| 339 | u64 cluster_bytes; |
| 340 | |
| 341 | mlog_entry_void(); |
| 342 | |
| 343 | /* TODO: This needs to actually orphan the inode in this |
| 344 | * transaction. */ |
| 345 | |
| 346 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
| 347 | if (IS_ERR(handle)) { |
| 348 | status = PTR_ERR(handle); |
| 349 | mlog_errno(status); |
| 350 | goto out; |
| 351 | } |
| 352 | |
| 353 | status = ocfs2_journal_access(handle, inode, fe_bh, |
| 354 | OCFS2_JOURNAL_ACCESS_WRITE); |
| 355 | if (status < 0) { |
| 356 | mlog_errno(status); |
| 357 | goto out_commit; |
| 358 | } |
| 359 | |
| 360 | /* |
| 361 | * Do this before setting i_size. |
| 362 | */ |
| 363 | cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size); |
| 364 | status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size, |
| 365 | cluster_bytes); |
| 366 | if (status) { |
| 367 | mlog_errno(status); |
| 368 | goto out_commit; |
| 369 | } |
| 370 | |
| 371 | i_size_write(inode, new_i_size); |
| 372 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
| 373 | |
| 374 | di = (struct ocfs2_dinode *) fe_bh->b_data; |
| 375 | di->i_size = cpu_to_le64(new_i_size); |
| 376 | di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec); |
| 377 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); |
| 378 | |
| 379 | status = ocfs2_journal_dirty(handle, fe_bh); |
| 380 | if (status < 0) |
| 381 | mlog_errno(status); |
| 382 | |
| 383 | out_commit: |
| 384 | ocfs2_commit_trans(osb, handle); |
| 385 | out: |
| 386 | |
| 387 | mlog_exit(status); |
| 388 | return status; |
| 389 | } |
| 390 | |
| 391 | static int ocfs2_truncate_file(struct inode *inode, |
| 392 | struct buffer_head *di_bh, |
| 393 | u64 new_i_size) |
| 394 | { |
| 395 | int status = 0; |
| 396 | struct ocfs2_dinode *fe = NULL; |
| 397 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 398 | struct ocfs2_truncate_context *tc = NULL; |
| 399 | |
| 400 | mlog_entry("(inode = %llu, new_i_size = %llu\n", |
| 401 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
| 402 | (unsigned long long)new_i_size); |
| 403 | |
| 404 | fe = (struct ocfs2_dinode *) di_bh->b_data; |
| 405 | if (!OCFS2_IS_VALID_DINODE(fe)) { |
| 406 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); |
| 407 | status = -EIO; |
| 408 | goto bail; |
| 409 | } |
| 410 | |
| 411 | mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode), |
| 412 | "Inode %llu, inode i_size = %lld != di " |
| 413 | "i_size = %llu, i_flags = 0x%x\n", |
| 414 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
| 415 | i_size_read(inode), |
| 416 | (unsigned long long)le64_to_cpu(fe->i_size), |
| 417 | le32_to_cpu(fe->i_flags)); |
| 418 | |
| 419 | if (new_i_size > le64_to_cpu(fe->i_size)) { |
| 420 | mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n", |
| 421 | (unsigned long long)le64_to_cpu(fe->i_size), |
| 422 | (unsigned long long)new_i_size); |
| 423 | status = -EINVAL; |
| 424 | mlog_errno(status); |
| 425 | goto bail; |
| 426 | } |
| 427 | |
| 428 | mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n", |
| 429 | (unsigned long long)le64_to_cpu(fe->i_blkno), |
| 430 | (unsigned long long)le64_to_cpu(fe->i_size), |
| 431 | (unsigned long long)new_i_size); |
| 432 | |
| 433 | /* lets handle the simple truncate cases before doing any more |
| 434 | * cluster locking. */ |
| 435 | if (new_i_size == le64_to_cpu(fe->i_size)) |
| 436 | goto bail; |
| 437 | |
| 438 | down_write(&OCFS2_I(inode)->ip_alloc_sem); |
| 439 | |
| 440 | /* |
| 441 | * The inode lock forced other nodes to sync and drop their |
| 442 | * pages, which (correctly) happens even if we have a truncate |
| 443 | * without allocation change - ocfs2 cluster sizes can be much |
| 444 | * greater than page size, so we have to truncate them |
| 445 | * anyway. |
| 446 | */ |
| 447 | unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1); |
| 448 | truncate_inode_pages(inode->i_mapping, new_i_size); |
| 449 | |
| 450 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
| 451 | status = ocfs2_truncate_inline(inode, di_bh, new_i_size, |
| 452 | i_size_read(inode), 1); |
| 453 | if (status) |
| 454 | mlog_errno(status); |
| 455 | |
| 456 | goto bail_unlock_sem; |
| 457 | } |
| 458 | |
| 459 | /* alright, we're going to need to do a full blown alloc size |
| 460 | * change. Orphan the inode so that recovery can complete the |
| 461 | * truncate if necessary. This does the task of marking |
| 462 | * i_size. */ |
| 463 | status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size); |
| 464 | if (status < 0) { |
| 465 | mlog_errno(status); |
| 466 | goto bail_unlock_sem; |
| 467 | } |
| 468 | |
| 469 | status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc); |
| 470 | if (status < 0) { |
| 471 | mlog_errno(status); |
| 472 | goto bail_unlock_sem; |
| 473 | } |
| 474 | |
| 475 | status = ocfs2_commit_truncate(osb, inode, di_bh, tc); |
| 476 | if (status < 0) { |
| 477 | mlog_errno(status); |
| 478 | goto bail_unlock_sem; |
| 479 | } |
| 480 | |
| 481 | /* TODO: orphan dir cleanup here. */ |
| 482 | bail_unlock_sem: |
| 483 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
| 484 | |
| 485 | bail: |
| 486 | |
| 487 | mlog_exit(status); |
| 488 | return status; |
| 489 | } |
| 490 | |
| 491 | /* |
| 492 | * extend file allocation only here. |
| 493 | * we'll update all the disk stuff, and oip->alloc_size |
| 494 | * |
| 495 | * expect stuff to be locked, a transaction started and enough data / |
| 496 | * metadata reservations in the contexts. |
| 497 | * |
| 498 | * Will return -EAGAIN, and a reason if a restart is needed. |
| 499 | * If passed in, *reason will always be set, even in error. |
| 500 | */ |
| 501 | int ocfs2_add_inode_data(struct ocfs2_super *osb, |
| 502 | struct inode *inode, |
| 503 | u32 *logical_offset, |
| 504 | u32 clusters_to_add, |
| 505 | int mark_unwritten, |
| 506 | struct buffer_head *fe_bh, |
| 507 | handle_t *handle, |
| 508 | struct ocfs2_alloc_context *data_ac, |
| 509 | struct ocfs2_alloc_context *meta_ac, |
| 510 | enum ocfs2_alloc_restarted *reason_ret) |
| 511 | { |
| 512 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data; |
| 513 | struct ocfs2_extent_list *el = &fe->id2.i_list; |
| 514 | |
| 515 | return ocfs2_add_clusters_in_btree(osb, inode, logical_offset, |
| 516 | clusters_to_add, mark_unwritten, |
| 517 | fe_bh, el, handle, |
| 518 | data_ac, meta_ac, reason_ret, |
| 519 | OCFS2_DINODE_EXTENT, NULL); |
| 520 | } |
| 521 | |
| 522 | static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start, |
| 523 | u32 clusters_to_add, int mark_unwritten) |
| 524 | { |
| 525 | int status = 0; |
| 526 | int restart_func = 0; |
| 527 | int credits; |
| 528 | u32 prev_clusters; |
| 529 | struct buffer_head *bh = NULL; |
| 530 | struct ocfs2_dinode *fe = NULL; |
| 531 | handle_t *handle = NULL; |
| 532 | struct ocfs2_alloc_context *data_ac = NULL; |
| 533 | struct ocfs2_alloc_context *meta_ac = NULL; |
| 534 | enum ocfs2_alloc_restarted why; |
| 535 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 536 | |
| 537 | mlog_entry("(clusters_to_add = %u)\n", clusters_to_add); |
| 538 | |
| 539 | /* |
| 540 | * This function only exists for file systems which don't |
| 541 | * support holes. |
| 542 | */ |
| 543 | BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb)); |
| 544 | |
| 545 | status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh, |
| 546 | OCFS2_BH_CACHED, inode); |
| 547 | if (status < 0) { |
| 548 | mlog_errno(status); |
| 549 | goto leave; |
| 550 | } |
| 551 | |
| 552 | fe = (struct ocfs2_dinode *) bh->b_data; |
| 553 | if (!OCFS2_IS_VALID_DINODE(fe)) { |
| 554 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); |
| 555 | status = -EIO; |
| 556 | goto leave; |
| 557 | } |
| 558 | |
| 559 | restart_all: |
| 560 | BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters); |
| 561 | |
| 562 | mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, " |
| 563 | "clusters_to_add = %u\n", |
| 564 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
| 565 | (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters), |
| 566 | clusters_to_add); |
| 567 | status = ocfs2_lock_allocators(inode, bh, &fe->id2.i_list, |
| 568 | clusters_to_add, 0, &data_ac, |
| 569 | &meta_ac, OCFS2_DINODE_EXTENT, NULL); |
| 570 | if (status) { |
| 571 | mlog_errno(status); |
| 572 | goto leave; |
| 573 | } |
| 574 | |
| 575 | credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list, |
| 576 | clusters_to_add); |
| 577 | handle = ocfs2_start_trans(osb, credits); |
| 578 | if (IS_ERR(handle)) { |
| 579 | status = PTR_ERR(handle); |
| 580 | handle = NULL; |
| 581 | mlog_errno(status); |
| 582 | goto leave; |
| 583 | } |
| 584 | |
| 585 | restarted_transaction: |
| 586 | /* reserve a write to the file entry early on - that we if we |
| 587 | * run out of credits in the allocation path, we can still |
| 588 | * update i_size. */ |
| 589 | status = ocfs2_journal_access(handle, inode, bh, |
| 590 | OCFS2_JOURNAL_ACCESS_WRITE); |
| 591 | if (status < 0) { |
| 592 | mlog_errno(status); |
| 593 | goto leave; |
| 594 | } |
| 595 | |
| 596 | prev_clusters = OCFS2_I(inode)->ip_clusters; |
| 597 | |
| 598 | status = ocfs2_add_inode_data(osb, |
| 599 | inode, |
| 600 | &logical_start, |
| 601 | clusters_to_add, |
| 602 | mark_unwritten, |
| 603 | bh, |
| 604 | handle, |
| 605 | data_ac, |
| 606 | meta_ac, |
| 607 | &why); |
| 608 | if ((status < 0) && (status != -EAGAIN)) { |
| 609 | if (status != -ENOSPC) |
| 610 | mlog_errno(status); |
| 611 | goto leave; |
| 612 | } |
| 613 | |
| 614 | status = ocfs2_journal_dirty(handle, bh); |
| 615 | if (status < 0) { |
| 616 | mlog_errno(status); |
| 617 | goto leave; |
| 618 | } |
| 619 | |
| 620 | spin_lock(&OCFS2_I(inode)->ip_lock); |
| 621 | clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters); |
| 622 | spin_unlock(&OCFS2_I(inode)->ip_lock); |
| 623 | |
| 624 | if (why != RESTART_NONE && clusters_to_add) { |
| 625 | if (why == RESTART_META) { |
| 626 | mlog(0, "restarting function.\n"); |
| 627 | restart_func = 1; |
| 628 | } else { |
| 629 | BUG_ON(why != RESTART_TRANS); |
| 630 | |
| 631 | mlog(0, "restarting transaction.\n"); |
| 632 | /* TODO: This can be more intelligent. */ |
| 633 | credits = ocfs2_calc_extend_credits(osb->sb, |
| 634 | &fe->id2.i_list, |
| 635 | clusters_to_add); |
| 636 | status = ocfs2_extend_trans(handle, credits); |
| 637 | if (status < 0) { |
| 638 | /* handle still has to be committed at |
| 639 | * this point. */ |
| 640 | status = -ENOMEM; |
| 641 | mlog_errno(status); |
| 642 | goto leave; |
| 643 | } |
| 644 | goto restarted_transaction; |
| 645 | } |
| 646 | } |
| 647 | |
| 648 | mlog(0, "fe: i_clusters = %u, i_size=%llu\n", |
| 649 | le32_to_cpu(fe->i_clusters), |
| 650 | (unsigned long long)le64_to_cpu(fe->i_size)); |
| 651 | mlog(0, "inode: ip_clusters=%u, i_size=%lld\n", |
| 652 | OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode)); |
| 653 | |
| 654 | leave: |
| 655 | if (handle) { |
| 656 | ocfs2_commit_trans(osb, handle); |
| 657 | handle = NULL; |
| 658 | } |
| 659 | if (data_ac) { |
| 660 | ocfs2_free_alloc_context(data_ac); |
| 661 | data_ac = NULL; |
| 662 | } |
| 663 | if (meta_ac) { |
| 664 | ocfs2_free_alloc_context(meta_ac); |
| 665 | meta_ac = NULL; |
| 666 | } |
| 667 | if ((!status) && restart_func) { |
| 668 | restart_func = 0; |
| 669 | goto restart_all; |
| 670 | } |
| 671 | if (bh) { |
| 672 | brelse(bh); |
| 673 | bh = NULL; |
| 674 | } |
| 675 | |
| 676 | mlog_exit(status); |
| 677 | return status; |
| 678 | } |
| 679 | |
| 680 | /* Some parts of this taken from generic_cont_expand, which turned out |
| 681 | * to be too fragile to do exactly what we need without us having to |
| 682 | * worry about recursive locking in ->prepare_write() and |
| 683 | * ->commit_write(). */ |
| 684 | static int ocfs2_write_zero_page(struct inode *inode, |
| 685 | u64 size) |
| 686 | { |
| 687 | struct address_space *mapping = inode->i_mapping; |
| 688 | struct page *page; |
| 689 | unsigned long index; |
| 690 | unsigned int offset; |
| 691 | handle_t *handle = NULL; |
| 692 | int ret; |
| 693 | |
| 694 | offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */ |
| 695 | /* ugh. in prepare/commit_write, if from==to==start of block, we |
| 696 | ** skip the prepare. make sure we never send an offset for the start |
| 697 | ** of a block |
| 698 | */ |
| 699 | if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) { |
| 700 | offset++; |
| 701 | } |
| 702 | index = size >> PAGE_CACHE_SHIFT; |
| 703 | |
| 704 | page = grab_cache_page(mapping, index); |
| 705 | if (!page) { |
| 706 | ret = -ENOMEM; |
| 707 | mlog_errno(ret); |
| 708 | goto out; |
| 709 | } |
| 710 | |
| 711 | ret = ocfs2_prepare_write_nolock(inode, page, offset, offset); |
| 712 | if (ret < 0) { |
| 713 | mlog_errno(ret); |
| 714 | goto out_unlock; |
| 715 | } |
| 716 | |
| 717 | if (ocfs2_should_order_data(inode)) { |
| 718 | handle = ocfs2_start_walk_page_trans(inode, page, offset, |
| 719 | offset); |
| 720 | if (IS_ERR(handle)) { |
| 721 | ret = PTR_ERR(handle); |
| 722 | handle = NULL; |
| 723 | goto out_unlock; |
| 724 | } |
| 725 | } |
| 726 | |
| 727 | /* must not update i_size! */ |
| 728 | ret = block_commit_write(page, offset, offset); |
| 729 | if (ret < 0) |
| 730 | mlog_errno(ret); |
| 731 | else |
| 732 | ret = 0; |
| 733 | |
| 734 | if (handle) |
| 735 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); |
| 736 | out_unlock: |
| 737 | unlock_page(page); |
| 738 | page_cache_release(page); |
| 739 | out: |
| 740 | return ret; |
| 741 | } |
| 742 | |
| 743 | static int ocfs2_zero_extend(struct inode *inode, |
| 744 | u64 zero_to_size) |
| 745 | { |
| 746 | int ret = 0; |
| 747 | u64 start_off; |
| 748 | struct super_block *sb = inode->i_sb; |
| 749 | |
| 750 | start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode)); |
| 751 | while (start_off < zero_to_size) { |
| 752 | ret = ocfs2_write_zero_page(inode, start_off); |
| 753 | if (ret < 0) { |
| 754 | mlog_errno(ret); |
| 755 | goto out; |
| 756 | } |
| 757 | |
| 758 | start_off += sb->s_blocksize; |
| 759 | |
| 760 | /* |
| 761 | * Very large extends have the potential to lock up |
| 762 | * the cpu for extended periods of time. |
| 763 | */ |
| 764 | cond_resched(); |
| 765 | } |
| 766 | |
| 767 | out: |
| 768 | return ret; |
| 769 | } |
| 770 | |
| 771 | int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to) |
| 772 | { |
| 773 | int ret; |
| 774 | u32 clusters_to_add; |
| 775 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 776 | |
| 777 | clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size); |
| 778 | if (clusters_to_add < oi->ip_clusters) |
| 779 | clusters_to_add = 0; |
| 780 | else |
| 781 | clusters_to_add -= oi->ip_clusters; |
| 782 | |
| 783 | if (clusters_to_add) { |
| 784 | ret = __ocfs2_extend_allocation(inode, oi->ip_clusters, |
| 785 | clusters_to_add, 0); |
| 786 | if (ret) { |
| 787 | mlog_errno(ret); |
| 788 | goto out; |
| 789 | } |
| 790 | } |
| 791 | |
| 792 | /* |
| 793 | * Call this even if we don't add any clusters to the tree. We |
| 794 | * still need to zero the area between the old i_size and the |
| 795 | * new i_size. |
| 796 | */ |
| 797 | ret = ocfs2_zero_extend(inode, zero_to); |
| 798 | if (ret < 0) |
| 799 | mlog_errno(ret); |
| 800 | |
| 801 | out: |
| 802 | return ret; |
| 803 | } |
| 804 | |
| 805 | static int ocfs2_extend_file(struct inode *inode, |
| 806 | struct buffer_head *di_bh, |
| 807 | u64 new_i_size) |
| 808 | { |
| 809 | int ret = 0; |
| 810 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 811 | |
| 812 | BUG_ON(!di_bh); |
| 813 | |
| 814 | /* setattr sometimes calls us like this. */ |
| 815 | if (new_i_size == 0) |
| 816 | goto out; |
| 817 | |
| 818 | if (i_size_read(inode) == new_i_size) |
| 819 | goto out; |
| 820 | BUG_ON(new_i_size < i_size_read(inode)); |
| 821 | |
| 822 | /* |
| 823 | * Fall through for converting inline data, even if the fs |
| 824 | * supports sparse files. |
| 825 | * |
| 826 | * The check for inline data here is legal - nobody can add |
| 827 | * the feature since we have i_mutex. We must check it again |
| 828 | * after acquiring ip_alloc_sem though, as paths like mmap |
| 829 | * might have raced us to converting the inode to extents. |
| 830 | */ |
| 831 | if (!(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) |
| 832 | && ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) |
| 833 | goto out_update_size; |
| 834 | |
| 835 | /* |
| 836 | * The alloc sem blocks people in read/write from reading our |
| 837 | * allocation until we're done changing it. We depend on |
| 838 | * i_mutex to block other extend/truncate calls while we're |
| 839 | * here. |
| 840 | */ |
| 841 | down_write(&oi->ip_alloc_sem); |
| 842 | |
| 843 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
| 844 | /* |
| 845 | * We can optimize small extends by keeping the inodes |
| 846 | * inline data. |
| 847 | */ |
| 848 | if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) { |
| 849 | up_write(&oi->ip_alloc_sem); |
| 850 | goto out_update_size; |
| 851 | } |
| 852 | |
| 853 | ret = ocfs2_convert_inline_data_to_extents(inode, di_bh); |
| 854 | if (ret) { |
| 855 | up_write(&oi->ip_alloc_sem); |
| 856 | |
| 857 | mlog_errno(ret); |
| 858 | goto out; |
| 859 | } |
| 860 | } |
| 861 | |
| 862 | if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) |
| 863 | ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size); |
| 864 | |
| 865 | up_write(&oi->ip_alloc_sem); |
| 866 | |
| 867 | if (ret < 0) { |
| 868 | mlog_errno(ret); |
| 869 | goto out; |
| 870 | } |
| 871 | |
| 872 | out_update_size: |
| 873 | ret = ocfs2_simple_size_update(inode, di_bh, new_i_size); |
| 874 | if (ret < 0) |
| 875 | mlog_errno(ret); |
| 876 | |
| 877 | out: |
| 878 | return ret; |
| 879 | } |
| 880 | |
| 881 | int ocfs2_setattr(struct dentry *dentry, struct iattr *attr) |
| 882 | { |
| 883 | int status = 0, size_change; |
| 884 | struct inode *inode = dentry->d_inode; |
| 885 | struct super_block *sb = inode->i_sb; |
| 886 | struct ocfs2_super *osb = OCFS2_SB(sb); |
| 887 | struct buffer_head *bh = NULL; |
| 888 | handle_t *handle = NULL; |
| 889 | |
| 890 | mlog_entry("(0x%p, '%.*s')\n", dentry, |
| 891 | dentry->d_name.len, dentry->d_name.name); |
| 892 | |
| 893 | /* ensuring we don't even attempt to truncate a symlink */ |
| 894 | if (S_ISLNK(inode->i_mode)) |
| 895 | attr->ia_valid &= ~ATTR_SIZE; |
| 896 | |
| 897 | if (attr->ia_valid & ATTR_MODE) |
| 898 | mlog(0, "mode change: %d\n", attr->ia_mode); |
| 899 | if (attr->ia_valid & ATTR_UID) |
| 900 | mlog(0, "uid change: %d\n", attr->ia_uid); |
| 901 | if (attr->ia_valid & ATTR_GID) |
| 902 | mlog(0, "gid change: %d\n", attr->ia_gid); |
| 903 | if (attr->ia_valid & ATTR_SIZE) |
| 904 | mlog(0, "size change...\n"); |
| 905 | if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME)) |
| 906 | mlog(0, "time change...\n"); |
| 907 | |
| 908 | #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \ |
| 909 | | ATTR_GID | ATTR_UID | ATTR_MODE) |
| 910 | if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) { |
| 911 | mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid); |
| 912 | return 0; |
| 913 | } |
| 914 | |
| 915 | status = inode_change_ok(inode, attr); |
| 916 | if (status) |
| 917 | return status; |
| 918 | |
| 919 | size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE; |
| 920 | if (size_change) { |
| 921 | status = ocfs2_rw_lock(inode, 1); |
| 922 | if (status < 0) { |
| 923 | mlog_errno(status); |
| 924 | goto bail; |
| 925 | } |
| 926 | } |
| 927 | |
| 928 | status = ocfs2_inode_lock(inode, &bh, 1); |
| 929 | if (status < 0) { |
| 930 | if (status != -ENOENT) |
| 931 | mlog_errno(status); |
| 932 | goto bail_unlock_rw; |
| 933 | } |
| 934 | |
| 935 | if (size_change && attr->ia_size != i_size_read(inode)) { |
| 936 | if (attr->ia_size > sb->s_maxbytes) { |
| 937 | status = -EFBIG; |
| 938 | goto bail_unlock; |
| 939 | } |
| 940 | |
| 941 | if (i_size_read(inode) > attr->ia_size) |
| 942 | status = ocfs2_truncate_file(inode, bh, attr->ia_size); |
| 943 | else |
| 944 | status = ocfs2_extend_file(inode, bh, attr->ia_size); |
| 945 | if (status < 0) { |
| 946 | if (status != -ENOSPC) |
| 947 | mlog_errno(status); |
| 948 | status = -ENOSPC; |
| 949 | goto bail_unlock; |
| 950 | } |
| 951 | } |
| 952 | |
| 953 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
| 954 | if (IS_ERR(handle)) { |
| 955 | status = PTR_ERR(handle); |
| 956 | mlog_errno(status); |
| 957 | goto bail_unlock; |
| 958 | } |
| 959 | |
| 960 | /* |
| 961 | * This will intentionally not wind up calling vmtruncate(), |
| 962 | * since all the work for a size change has been done above. |
| 963 | * Otherwise, we could get into problems with truncate as |
| 964 | * ip_alloc_sem is used there to protect against i_size |
| 965 | * changes. |
| 966 | */ |
| 967 | status = inode_setattr(inode, attr); |
| 968 | if (status < 0) { |
| 969 | mlog_errno(status); |
| 970 | goto bail_commit; |
| 971 | } |
| 972 | |
| 973 | status = ocfs2_mark_inode_dirty(handle, inode, bh); |
| 974 | if (status < 0) |
| 975 | mlog_errno(status); |
| 976 | |
| 977 | bail_commit: |
| 978 | ocfs2_commit_trans(osb, handle); |
| 979 | bail_unlock: |
| 980 | ocfs2_inode_unlock(inode, 1); |
| 981 | bail_unlock_rw: |
| 982 | if (size_change) |
| 983 | ocfs2_rw_unlock(inode, 1); |
| 984 | bail: |
| 985 | if (bh) |
| 986 | brelse(bh); |
| 987 | |
| 988 | mlog_exit(status); |
| 989 | return status; |
| 990 | } |
| 991 | |
| 992 | int ocfs2_getattr(struct vfsmount *mnt, |
| 993 | struct dentry *dentry, |
| 994 | struct kstat *stat) |
| 995 | { |
| 996 | struct inode *inode = dentry->d_inode; |
| 997 | struct super_block *sb = dentry->d_inode->i_sb; |
| 998 | struct ocfs2_super *osb = sb->s_fs_info; |
| 999 | int err; |
| 1000 | |
| 1001 | mlog_entry_void(); |
| 1002 | |
| 1003 | err = ocfs2_inode_revalidate(dentry); |
| 1004 | if (err) { |
| 1005 | if (err != -ENOENT) |
| 1006 | mlog_errno(err); |
| 1007 | goto bail; |
| 1008 | } |
| 1009 | |
| 1010 | generic_fillattr(inode, stat); |
| 1011 | |
| 1012 | /* We set the blksize from the cluster size for performance */ |
| 1013 | stat->blksize = osb->s_clustersize; |
| 1014 | |
| 1015 | bail: |
| 1016 | mlog_exit(err); |
| 1017 | |
| 1018 | return err; |
| 1019 | } |
| 1020 | |
| 1021 | int ocfs2_permission(struct inode *inode, int mask) |
| 1022 | { |
| 1023 | int ret; |
| 1024 | |
| 1025 | mlog_entry_void(); |
| 1026 | |
| 1027 | ret = ocfs2_inode_lock(inode, NULL, 0); |
| 1028 | if (ret) { |
| 1029 | if (ret != -ENOENT) |
| 1030 | mlog_errno(ret); |
| 1031 | goto out; |
| 1032 | } |
| 1033 | |
| 1034 | ret = generic_permission(inode, mask, NULL); |
| 1035 | |
| 1036 | ocfs2_inode_unlock(inode, 0); |
| 1037 | out: |
| 1038 | mlog_exit(ret); |
| 1039 | return ret; |
| 1040 | } |
| 1041 | |
| 1042 | static int __ocfs2_write_remove_suid(struct inode *inode, |
| 1043 | struct buffer_head *bh) |
| 1044 | { |
| 1045 | int ret; |
| 1046 | handle_t *handle; |
| 1047 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1048 | struct ocfs2_dinode *di; |
| 1049 | |
| 1050 | mlog_entry("(Inode %llu, mode 0%o)\n", |
| 1051 | (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode); |
| 1052 | |
| 1053 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
| 1054 | if (handle == NULL) { |
| 1055 | ret = -ENOMEM; |
| 1056 | mlog_errno(ret); |
| 1057 | goto out; |
| 1058 | } |
| 1059 | |
| 1060 | ret = ocfs2_journal_access(handle, inode, bh, |
| 1061 | OCFS2_JOURNAL_ACCESS_WRITE); |
| 1062 | if (ret < 0) { |
| 1063 | mlog_errno(ret); |
| 1064 | goto out_trans; |
| 1065 | } |
| 1066 | |
| 1067 | inode->i_mode &= ~S_ISUID; |
| 1068 | if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP)) |
| 1069 | inode->i_mode &= ~S_ISGID; |
| 1070 | |
| 1071 | di = (struct ocfs2_dinode *) bh->b_data; |
| 1072 | di->i_mode = cpu_to_le16(inode->i_mode); |
| 1073 | |
| 1074 | ret = ocfs2_journal_dirty(handle, bh); |
| 1075 | if (ret < 0) |
| 1076 | mlog_errno(ret); |
| 1077 | |
| 1078 | out_trans: |
| 1079 | ocfs2_commit_trans(osb, handle); |
| 1080 | out: |
| 1081 | mlog_exit(ret); |
| 1082 | return ret; |
| 1083 | } |
| 1084 | |
| 1085 | /* |
| 1086 | * Will look for holes and unwritten extents in the range starting at |
| 1087 | * pos for count bytes (inclusive). |
| 1088 | */ |
| 1089 | static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos, |
| 1090 | size_t count) |
| 1091 | { |
| 1092 | int ret = 0; |
| 1093 | unsigned int extent_flags; |
| 1094 | u32 cpos, clusters, extent_len, phys_cpos; |
| 1095 | struct super_block *sb = inode->i_sb; |
| 1096 | |
| 1097 | cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits; |
| 1098 | clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos; |
| 1099 | |
| 1100 | while (clusters) { |
| 1101 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len, |
| 1102 | &extent_flags); |
| 1103 | if (ret < 0) { |
| 1104 | mlog_errno(ret); |
| 1105 | goto out; |
| 1106 | } |
| 1107 | |
| 1108 | if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) { |
| 1109 | ret = 1; |
| 1110 | break; |
| 1111 | } |
| 1112 | |
| 1113 | if (extent_len > clusters) |
| 1114 | extent_len = clusters; |
| 1115 | |
| 1116 | clusters -= extent_len; |
| 1117 | cpos += extent_len; |
| 1118 | } |
| 1119 | out: |
| 1120 | return ret; |
| 1121 | } |
| 1122 | |
| 1123 | static int ocfs2_write_remove_suid(struct inode *inode) |
| 1124 | { |
| 1125 | int ret; |
| 1126 | struct buffer_head *bh = NULL; |
| 1127 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
| 1128 | |
| 1129 | ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), |
| 1130 | oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode); |
| 1131 | if (ret < 0) { |
| 1132 | mlog_errno(ret); |
| 1133 | goto out; |
| 1134 | } |
| 1135 | |
| 1136 | ret = __ocfs2_write_remove_suid(inode, bh); |
| 1137 | out: |
| 1138 | brelse(bh); |
| 1139 | return ret; |
| 1140 | } |
| 1141 | |
| 1142 | /* |
| 1143 | * Allocate enough extents to cover the region starting at byte offset |
| 1144 | * start for len bytes. Existing extents are skipped, any extents |
| 1145 | * added are marked as "unwritten". |
| 1146 | */ |
| 1147 | static int ocfs2_allocate_unwritten_extents(struct inode *inode, |
| 1148 | u64 start, u64 len) |
| 1149 | { |
| 1150 | int ret; |
| 1151 | u32 cpos, phys_cpos, clusters, alloc_size; |
| 1152 | u64 end = start + len; |
| 1153 | struct buffer_head *di_bh = NULL; |
| 1154 | |
| 1155 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
| 1156 | ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), |
| 1157 | OCFS2_I(inode)->ip_blkno, &di_bh, |
| 1158 | OCFS2_BH_CACHED, inode); |
| 1159 | if (ret) { |
| 1160 | mlog_errno(ret); |
| 1161 | goto out; |
| 1162 | } |
| 1163 | |
| 1164 | /* |
| 1165 | * Nothing to do if the requested reservation range |
| 1166 | * fits within the inode. |
| 1167 | */ |
| 1168 | if (ocfs2_size_fits_inline_data(di_bh, end)) |
| 1169 | goto out; |
| 1170 | |
| 1171 | ret = ocfs2_convert_inline_data_to_extents(inode, di_bh); |
| 1172 | if (ret) { |
| 1173 | mlog_errno(ret); |
| 1174 | goto out; |
| 1175 | } |
| 1176 | } |
| 1177 | |
| 1178 | /* |
| 1179 | * We consider both start and len to be inclusive. |
| 1180 | */ |
| 1181 | cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits; |
| 1182 | clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len); |
| 1183 | clusters -= cpos; |
| 1184 | |
| 1185 | while (clusters) { |
| 1186 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, |
| 1187 | &alloc_size, NULL); |
| 1188 | if (ret) { |
| 1189 | mlog_errno(ret); |
| 1190 | goto out; |
| 1191 | } |
| 1192 | |
| 1193 | /* |
| 1194 | * Hole or existing extent len can be arbitrary, so |
| 1195 | * cap it to our own allocation request. |
| 1196 | */ |
| 1197 | if (alloc_size > clusters) |
| 1198 | alloc_size = clusters; |
| 1199 | |
| 1200 | if (phys_cpos) { |
| 1201 | /* |
| 1202 | * We already have an allocation at this |
| 1203 | * region so we can safely skip it. |
| 1204 | */ |
| 1205 | goto next; |
| 1206 | } |
| 1207 | |
| 1208 | ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1); |
| 1209 | if (ret) { |
| 1210 | if (ret != -ENOSPC) |
| 1211 | mlog_errno(ret); |
| 1212 | goto out; |
| 1213 | } |
| 1214 | |
| 1215 | next: |
| 1216 | cpos += alloc_size; |
| 1217 | clusters -= alloc_size; |
| 1218 | } |
| 1219 | |
| 1220 | ret = 0; |
| 1221 | out: |
| 1222 | |
| 1223 | brelse(di_bh); |
| 1224 | return ret; |
| 1225 | } |
| 1226 | |
| 1227 | static int __ocfs2_remove_inode_range(struct inode *inode, |
| 1228 | struct buffer_head *di_bh, |
| 1229 | u32 cpos, u32 phys_cpos, u32 len, |
| 1230 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
| 1231 | { |
| 1232 | int ret; |
| 1233 | u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos); |
| 1234 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1235 | struct inode *tl_inode = osb->osb_tl_inode; |
| 1236 | handle_t *handle; |
| 1237 | struct ocfs2_alloc_context *meta_ac = NULL; |
| 1238 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; |
| 1239 | |
| 1240 | ret = ocfs2_lock_allocators(inode, di_bh, &di->id2.i_list, |
| 1241 | 0, 1, NULL, &meta_ac, |
| 1242 | OCFS2_DINODE_EXTENT, NULL); |
| 1243 | if (ret) { |
| 1244 | mlog_errno(ret); |
| 1245 | return ret; |
| 1246 | } |
| 1247 | |
| 1248 | mutex_lock(&tl_inode->i_mutex); |
| 1249 | |
| 1250 | if (ocfs2_truncate_log_needs_flush(osb)) { |
| 1251 | ret = __ocfs2_flush_truncate_log(osb); |
| 1252 | if (ret < 0) { |
| 1253 | mlog_errno(ret); |
| 1254 | goto out; |
| 1255 | } |
| 1256 | } |
| 1257 | |
| 1258 | handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS); |
| 1259 | if (handle == NULL) { |
| 1260 | ret = -ENOMEM; |
| 1261 | mlog_errno(ret); |
| 1262 | goto out; |
| 1263 | } |
| 1264 | |
| 1265 | ret = ocfs2_journal_access(handle, inode, di_bh, |
| 1266 | OCFS2_JOURNAL_ACCESS_WRITE); |
| 1267 | if (ret) { |
| 1268 | mlog_errno(ret); |
| 1269 | goto out; |
| 1270 | } |
| 1271 | |
| 1272 | ret = ocfs2_remove_extent(inode, di_bh, cpos, len, handle, meta_ac, |
| 1273 | dealloc, OCFS2_DINODE_EXTENT, NULL); |
| 1274 | if (ret) { |
| 1275 | mlog_errno(ret); |
| 1276 | goto out_commit; |
| 1277 | } |
| 1278 | |
| 1279 | OCFS2_I(inode)->ip_clusters -= len; |
| 1280 | di->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters); |
| 1281 | |
| 1282 | ret = ocfs2_journal_dirty(handle, di_bh); |
| 1283 | if (ret) { |
| 1284 | mlog_errno(ret); |
| 1285 | goto out_commit; |
| 1286 | } |
| 1287 | |
| 1288 | ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len); |
| 1289 | if (ret) |
| 1290 | mlog_errno(ret); |
| 1291 | |
| 1292 | out_commit: |
| 1293 | ocfs2_commit_trans(osb, handle); |
| 1294 | out: |
| 1295 | mutex_unlock(&tl_inode->i_mutex); |
| 1296 | |
| 1297 | if (meta_ac) |
| 1298 | ocfs2_free_alloc_context(meta_ac); |
| 1299 | |
| 1300 | return ret; |
| 1301 | } |
| 1302 | |
| 1303 | /* |
| 1304 | * Truncate a byte range, avoiding pages within partial clusters. This |
| 1305 | * preserves those pages for the zeroing code to write to. |
| 1306 | */ |
| 1307 | static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start, |
| 1308 | u64 byte_len) |
| 1309 | { |
| 1310 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1311 | loff_t start, end; |
| 1312 | struct address_space *mapping = inode->i_mapping; |
| 1313 | |
| 1314 | start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start); |
| 1315 | end = byte_start + byte_len; |
| 1316 | end = end & ~(osb->s_clustersize - 1); |
| 1317 | |
| 1318 | if (start < end) { |
| 1319 | unmap_mapping_range(mapping, start, end - start, 0); |
| 1320 | truncate_inode_pages_range(mapping, start, end - 1); |
| 1321 | } |
| 1322 | } |
| 1323 | |
| 1324 | static int ocfs2_zero_partial_clusters(struct inode *inode, |
| 1325 | u64 start, u64 len) |
| 1326 | { |
| 1327 | int ret = 0; |
| 1328 | u64 tmpend, end = start + len; |
| 1329 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1330 | unsigned int csize = osb->s_clustersize; |
| 1331 | handle_t *handle; |
| 1332 | |
| 1333 | /* |
| 1334 | * The "start" and "end" values are NOT necessarily part of |
| 1335 | * the range whose allocation is being deleted. Rather, this |
| 1336 | * is what the user passed in with the request. We must zero |
| 1337 | * partial clusters here. There's no need to worry about |
| 1338 | * physical allocation - the zeroing code knows to skip holes. |
| 1339 | */ |
| 1340 | mlog(0, "byte start: %llu, end: %llu\n", |
| 1341 | (unsigned long long)start, (unsigned long long)end); |
| 1342 | |
| 1343 | /* |
| 1344 | * If both edges are on a cluster boundary then there's no |
| 1345 | * zeroing required as the region is part of the allocation to |
| 1346 | * be truncated. |
| 1347 | */ |
| 1348 | if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0) |
| 1349 | goto out; |
| 1350 | |
| 1351 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
| 1352 | if (handle == NULL) { |
| 1353 | ret = -ENOMEM; |
| 1354 | mlog_errno(ret); |
| 1355 | goto out; |
| 1356 | } |
| 1357 | |
| 1358 | /* |
| 1359 | * We want to get the byte offset of the end of the 1st cluster. |
| 1360 | */ |
| 1361 | tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1)); |
| 1362 | if (tmpend > end) |
| 1363 | tmpend = end; |
| 1364 | |
| 1365 | mlog(0, "1st range: start: %llu, tmpend: %llu\n", |
| 1366 | (unsigned long long)start, (unsigned long long)tmpend); |
| 1367 | |
| 1368 | ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend); |
| 1369 | if (ret) |
| 1370 | mlog_errno(ret); |
| 1371 | |
| 1372 | if (tmpend < end) { |
| 1373 | /* |
| 1374 | * This may make start and end equal, but the zeroing |
| 1375 | * code will skip any work in that case so there's no |
| 1376 | * need to catch it up here. |
| 1377 | */ |
| 1378 | start = end & ~(osb->s_clustersize - 1); |
| 1379 | |
| 1380 | mlog(0, "2nd range: start: %llu, end: %llu\n", |
| 1381 | (unsigned long long)start, (unsigned long long)end); |
| 1382 | |
| 1383 | ret = ocfs2_zero_range_for_truncate(inode, handle, start, end); |
| 1384 | if (ret) |
| 1385 | mlog_errno(ret); |
| 1386 | } |
| 1387 | |
| 1388 | ocfs2_commit_trans(osb, handle); |
| 1389 | out: |
| 1390 | return ret; |
| 1391 | } |
| 1392 | |
| 1393 | static int ocfs2_remove_inode_range(struct inode *inode, |
| 1394 | struct buffer_head *di_bh, u64 byte_start, |
| 1395 | u64 byte_len) |
| 1396 | { |
| 1397 | int ret = 0; |
| 1398 | u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size; |
| 1399 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1400 | struct ocfs2_cached_dealloc_ctxt dealloc; |
| 1401 | struct address_space *mapping = inode->i_mapping; |
| 1402 | |
| 1403 | ocfs2_init_dealloc_ctxt(&dealloc); |
| 1404 | |
| 1405 | if (byte_len == 0) |
| 1406 | return 0; |
| 1407 | |
| 1408 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
| 1409 | ret = ocfs2_truncate_inline(inode, di_bh, byte_start, |
| 1410 | byte_start + byte_len, 0); |
| 1411 | if (ret) { |
| 1412 | mlog_errno(ret); |
| 1413 | goto out; |
| 1414 | } |
| 1415 | /* |
| 1416 | * There's no need to get fancy with the page cache |
| 1417 | * truncate of an inline-data inode. We're talking |
| 1418 | * about less than a page here, which will be cached |
| 1419 | * in the dinode buffer anyway. |
| 1420 | */ |
| 1421 | unmap_mapping_range(mapping, 0, 0, 0); |
| 1422 | truncate_inode_pages(mapping, 0); |
| 1423 | goto out; |
| 1424 | } |
| 1425 | |
| 1426 | trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start); |
| 1427 | trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits; |
| 1428 | if (trunc_len >= trunc_start) |
| 1429 | trunc_len -= trunc_start; |
| 1430 | else |
| 1431 | trunc_len = 0; |
| 1432 | |
| 1433 | mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n", |
| 1434 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
| 1435 | (unsigned long long)byte_start, |
| 1436 | (unsigned long long)byte_len, trunc_start, trunc_len); |
| 1437 | |
| 1438 | ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len); |
| 1439 | if (ret) { |
| 1440 | mlog_errno(ret); |
| 1441 | goto out; |
| 1442 | } |
| 1443 | |
| 1444 | cpos = trunc_start; |
| 1445 | while (trunc_len) { |
| 1446 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, |
| 1447 | &alloc_size, NULL); |
| 1448 | if (ret) { |
| 1449 | mlog_errno(ret); |
| 1450 | goto out; |
| 1451 | } |
| 1452 | |
| 1453 | if (alloc_size > trunc_len) |
| 1454 | alloc_size = trunc_len; |
| 1455 | |
| 1456 | /* Only do work for non-holes */ |
| 1457 | if (phys_cpos != 0) { |
| 1458 | ret = __ocfs2_remove_inode_range(inode, di_bh, cpos, |
| 1459 | phys_cpos, alloc_size, |
| 1460 | &dealloc); |
| 1461 | if (ret) { |
| 1462 | mlog_errno(ret); |
| 1463 | goto out; |
| 1464 | } |
| 1465 | } |
| 1466 | |
| 1467 | cpos += alloc_size; |
| 1468 | trunc_len -= alloc_size; |
| 1469 | } |
| 1470 | |
| 1471 | ocfs2_truncate_cluster_pages(inode, byte_start, byte_len); |
| 1472 | |
| 1473 | out: |
| 1474 | ocfs2_schedule_truncate_log_flush(osb, 1); |
| 1475 | ocfs2_run_deallocs(osb, &dealloc); |
| 1476 | |
| 1477 | return ret; |
| 1478 | } |
| 1479 | |
| 1480 | /* |
| 1481 | * Parts of this function taken from xfs_change_file_space() |
| 1482 | */ |
| 1483 | static int __ocfs2_change_file_space(struct file *file, struct inode *inode, |
| 1484 | loff_t f_pos, unsigned int cmd, |
| 1485 | struct ocfs2_space_resv *sr, |
| 1486 | int change_size) |
| 1487 | { |
| 1488 | int ret; |
| 1489 | s64 llen; |
| 1490 | loff_t size; |
| 1491 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1492 | struct buffer_head *di_bh = NULL; |
| 1493 | handle_t *handle; |
| 1494 | unsigned long long max_off = inode->i_sb->s_maxbytes; |
| 1495 | |
| 1496 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) |
| 1497 | return -EROFS; |
| 1498 | |
| 1499 | mutex_lock(&inode->i_mutex); |
| 1500 | |
| 1501 | /* |
| 1502 | * This prevents concurrent writes on other nodes |
| 1503 | */ |
| 1504 | ret = ocfs2_rw_lock(inode, 1); |
| 1505 | if (ret) { |
| 1506 | mlog_errno(ret); |
| 1507 | goto out; |
| 1508 | } |
| 1509 | |
| 1510 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
| 1511 | if (ret) { |
| 1512 | mlog_errno(ret); |
| 1513 | goto out_rw_unlock; |
| 1514 | } |
| 1515 | |
| 1516 | if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) { |
| 1517 | ret = -EPERM; |
| 1518 | goto out_inode_unlock; |
| 1519 | } |
| 1520 | |
| 1521 | switch (sr->l_whence) { |
| 1522 | case 0: /*SEEK_SET*/ |
| 1523 | break; |
| 1524 | case 1: /*SEEK_CUR*/ |
| 1525 | sr->l_start += f_pos; |
| 1526 | break; |
| 1527 | case 2: /*SEEK_END*/ |
| 1528 | sr->l_start += i_size_read(inode); |
| 1529 | break; |
| 1530 | default: |
| 1531 | ret = -EINVAL; |
| 1532 | goto out_inode_unlock; |
| 1533 | } |
| 1534 | sr->l_whence = 0; |
| 1535 | |
| 1536 | llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len; |
| 1537 | |
| 1538 | if (sr->l_start < 0 |
| 1539 | || sr->l_start > max_off |
| 1540 | || (sr->l_start + llen) < 0 |
| 1541 | || (sr->l_start + llen) > max_off) { |
| 1542 | ret = -EINVAL; |
| 1543 | goto out_inode_unlock; |
| 1544 | } |
| 1545 | size = sr->l_start + sr->l_len; |
| 1546 | |
| 1547 | if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) { |
| 1548 | if (sr->l_len <= 0) { |
| 1549 | ret = -EINVAL; |
| 1550 | goto out_inode_unlock; |
| 1551 | } |
| 1552 | } |
| 1553 | |
| 1554 | if (file && should_remove_suid(file->f_path.dentry)) { |
| 1555 | ret = __ocfs2_write_remove_suid(inode, di_bh); |
| 1556 | if (ret) { |
| 1557 | mlog_errno(ret); |
| 1558 | goto out_inode_unlock; |
| 1559 | } |
| 1560 | } |
| 1561 | |
| 1562 | down_write(&OCFS2_I(inode)->ip_alloc_sem); |
| 1563 | switch (cmd) { |
| 1564 | case OCFS2_IOC_RESVSP: |
| 1565 | case OCFS2_IOC_RESVSP64: |
| 1566 | /* |
| 1567 | * This takes unsigned offsets, but the signed ones we |
| 1568 | * pass have been checked against overflow above. |
| 1569 | */ |
| 1570 | ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start, |
| 1571 | sr->l_len); |
| 1572 | break; |
| 1573 | case OCFS2_IOC_UNRESVSP: |
| 1574 | case OCFS2_IOC_UNRESVSP64: |
| 1575 | ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start, |
| 1576 | sr->l_len); |
| 1577 | break; |
| 1578 | default: |
| 1579 | ret = -EINVAL; |
| 1580 | } |
| 1581 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
| 1582 | if (ret) { |
| 1583 | mlog_errno(ret); |
| 1584 | goto out_inode_unlock; |
| 1585 | } |
| 1586 | |
| 1587 | /* |
| 1588 | * We update c/mtime for these changes |
| 1589 | */ |
| 1590 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
| 1591 | if (IS_ERR(handle)) { |
| 1592 | ret = PTR_ERR(handle); |
| 1593 | mlog_errno(ret); |
| 1594 | goto out_inode_unlock; |
| 1595 | } |
| 1596 | |
| 1597 | if (change_size && i_size_read(inode) < size) |
| 1598 | i_size_write(inode, size); |
| 1599 | |
| 1600 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
| 1601 | ret = ocfs2_mark_inode_dirty(handle, inode, di_bh); |
| 1602 | if (ret < 0) |
| 1603 | mlog_errno(ret); |
| 1604 | |
| 1605 | ocfs2_commit_trans(osb, handle); |
| 1606 | |
| 1607 | out_inode_unlock: |
| 1608 | brelse(di_bh); |
| 1609 | ocfs2_inode_unlock(inode, 1); |
| 1610 | out_rw_unlock: |
| 1611 | ocfs2_rw_unlock(inode, 1); |
| 1612 | |
| 1613 | out: |
| 1614 | mutex_unlock(&inode->i_mutex); |
| 1615 | return ret; |
| 1616 | } |
| 1617 | |
| 1618 | int ocfs2_change_file_space(struct file *file, unsigned int cmd, |
| 1619 | struct ocfs2_space_resv *sr) |
| 1620 | { |
| 1621 | struct inode *inode = file->f_path.dentry->d_inode; |
| 1622 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);; |
| 1623 | |
| 1624 | if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) && |
| 1625 | !ocfs2_writes_unwritten_extents(osb)) |
| 1626 | return -ENOTTY; |
| 1627 | else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) && |
| 1628 | !ocfs2_sparse_alloc(osb)) |
| 1629 | return -ENOTTY; |
| 1630 | |
| 1631 | if (!S_ISREG(inode->i_mode)) |
| 1632 | return -EINVAL; |
| 1633 | |
| 1634 | if (!(file->f_mode & FMODE_WRITE)) |
| 1635 | return -EBADF; |
| 1636 | |
| 1637 | return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0); |
| 1638 | } |
| 1639 | |
| 1640 | static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset, |
| 1641 | loff_t len) |
| 1642 | { |
| 1643 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1644 | struct ocfs2_space_resv sr; |
| 1645 | int change_size = 1; |
| 1646 | |
| 1647 | if (!ocfs2_writes_unwritten_extents(osb)) |
| 1648 | return -EOPNOTSUPP; |
| 1649 | |
| 1650 | if (S_ISDIR(inode->i_mode)) |
| 1651 | return -ENODEV; |
| 1652 | |
| 1653 | if (mode & FALLOC_FL_KEEP_SIZE) |
| 1654 | change_size = 0; |
| 1655 | |
| 1656 | sr.l_whence = 0; |
| 1657 | sr.l_start = (s64)offset; |
| 1658 | sr.l_len = (s64)len; |
| 1659 | |
| 1660 | return __ocfs2_change_file_space(NULL, inode, offset, |
| 1661 | OCFS2_IOC_RESVSP64, &sr, change_size); |
| 1662 | } |
| 1663 | |
| 1664 | static int ocfs2_prepare_inode_for_write(struct dentry *dentry, |
| 1665 | loff_t *ppos, |
| 1666 | size_t count, |
| 1667 | int appending, |
| 1668 | int *direct_io) |
| 1669 | { |
| 1670 | int ret = 0, meta_level = 0; |
| 1671 | struct inode *inode = dentry->d_inode; |
| 1672 | loff_t saved_pos, end; |
| 1673 | |
| 1674 | /* |
| 1675 | * We start with a read level meta lock and only jump to an ex |
| 1676 | * if we need to make modifications here. |
| 1677 | */ |
| 1678 | for(;;) { |
| 1679 | ret = ocfs2_inode_lock(inode, NULL, meta_level); |
| 1680 | if (ret < 0) { |
| 1681 | meta_level = -1; |
| 1682 | mlog_errno(ret); |
| 1683 | goto out; |
| 1684 | } |
| 1685 | |
| 1686 | /* Clear suid / sgid if necessary. We do this here |
| 1687 | * instead of later in the write path because |
| 1688 | * remove_suid() calls ->setattr without any hint that |
| 1689 | * we may have already done our cluster locking. Since |
| 1690 | * ocfs2_setattr() *must* take cluster locks to |
| 1691 | * proceeed, this will lead us to recursively lock the |
| 1692 | * inode. There's also the dinode i_size state which |
| 1693 | * can be lost via setattr during extending writes (we |
| 1694 | * set inode->i_size at the end of a write. */ |
| 1695 | if (should_remove_suid(dentry)) { |
| 1696 | if (meta_level == 0) { |
| 1697 | ocfs2_inode_unlock(inode, meta_level); |
| 1698 | meta_level = 1; |
| 1699 | continue; |
| 1700 | } |
| 1701 | |
| 1702 | ret = ocfs2_write_remove_suid(inode); |
| 1703 | if (ret < 0) { |
| 1704 | mlog_errno(ret); |
| 1705 | goto out_unlock; |
| 1706 | } |
| 1707 | } |
| 1708 | |
| 1709 | /* work on a copy of ppos until we're sure that we won't have |
| 1710 | * to recalculate it due to relocking. */ |
| 1711 | if (appending) { |
| 1712 | saved_pos = i_size_read(inode); |
| 1713 | mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos); |
| 1714 | } else { |
| 1715 | saved_pos = *ppos; |
| 1716 | } |
| 1717 | |
| 1718 | end = saved_pos + count; |
| 1719 | |
| 1720 | /* |
| 1721 | * Skip the O_DIRECT checks if we don't need |
| 1722 | * them. |
| 1723 | */ |
| 1724 | if (!direct_io || !(*direct_io)) |
| 1725 | break; |
| 1726 | |
| 1727 | /* |
| 1728 | * There's no sane way to do direct writes to an inode |
| 1729 | * with inline data. |
| 1730 | */ |
| 1731 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
| 1732 | *direct_io = 0; |
| 1733 | break; |
| 1734 | } |
| 1735 | |
| 1736 | /* |
| 1737 | * Allowing concurrent direct writes means |
| 1738 | * i_size changes wouldn't be synchronized, so |
| 1739 | * one node could wind up truncating another |
| 1740 | * nodes writes. |
| 1741 | */ |
| 1742 | if (end > i_size_read(inode)) { |
| 1743 | *direct_io = 0; |
| 1744 | break; |
| 1745 | } |
| 1746 | |
| 1747 | /* |
| 1748 | * We don't fill holes during direct io, so |
| 1749 | * check for them here. If any are found, the |
| 1750 | * caller will have to retake some cluster |
| 1751 | * locks and initiate the io as buffered. |
| 1752 | */ |
| 1753 | ret = ocfs2_check_range_for_holes(inode, saved_pos, count); |
| 1754 | if (ret == 1) { |
| 1755 | *direct_io = 0; |
| 1756 | ret = 0; |
| 1757 | } else if (ret < 0) |
| 1758 | mlog_errno(ret); |
| 1759 | break; |
| 1760 | } |
| 1761 | |
| 1762 | if (appending) |
| 1763 | *ppos = saved_pos; |
| 1764 | |
| 1765 | out_unlock: |
| 1766 | ocfs2_inode_unlock(inode, meta_level); |
| 1767 | |
| 1768 | out: |
| 1769 | return ret; |
| 1770 | } |
| 1771 | |
| 1772 | static ssize_t ocfs2_file_aio_write(struct kiocb *iocb, |
| 1773 | const struct iovec *iov, |
| 1774 | unsigned long nr_segs, |
| 1775 | loff_t pos) |
| 1776 | { |
| 1777 | int ret, direct_io, appending, rw_level, have_alloc_sem = 0; |
| 1778 | int can_do_direct; |
| 1779 | ssize_t written = 0; |
| 1780 | size_t ocount; /* original count */ |
| 1781 | size_t count; /* after file limit checks */ |
| 1782 | loff_t old_size, *ppos = &iocb->ki_pos; |
| 1783 | u32 old_clusters; |
| 1784 | struct file *file = iocb->ki_filp; |
| 1785 | struct inode *inode = file->f_path.dentry->d_inode; |
| 1786 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 1787 | |
| 1788 | mlog_entry("(0x%p, %u, '%.*s')\n", file, |
| 1789 | (unsigned int)nr_segs, |
| 1790 | file->f_path.dentry->d_name.len, |
| 1791 | file->f_path.dentry->d_name.name); |
| 1792 | |
| 1793 | if (iocb->ki_left == 0) |
| 1794 | return 0; |
| 1795 | |
| 1796 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); |
| 1797 | |
| 1798 | appending = file->f_flags & O_APPEND ? 1 : 0; |
| 1799 | direct_io = file->f_flags & O_DIRECT ? 1 : 0; |
| 1800 | |
| 1801 | mutex_lock(&inode->i_mutex); |
| 1802 | |
| 1803 | relock: |
| 1804 | /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */ |
| 1805 | if (direct_io) { |
| 1806 | down_read(&inode->i_alloc_sem); |
| 1807 | have_alloc_sem = 1; |
| 1808 | } |
| 1809 | |
| 1810 | /* concurrent O_DIRECT writes are allowed */ |
| 1811 | rw_level = !direct_io; |
| 1812 | ret = ocfs2_rw_lock(inode, rw_level); |
| 1813 | if (ret < 0) { |
| 1814 | mlog_errno(ret); |
| 1815 | goto out_sems; |
| 1816 | } |
| 1817 | |
| 1818 | can_do_direct = direct_io; |
| 1819 | ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos, |
| 1820 | iocb->ki_left, appending, |
| 1821 | &can_do_direct); |
| 1822 | if (ret < 0) { |
| 1823 | mlog_errno(ret); |
| 1824 | goto out; |
| 1825 | } |
| 1826 | |
| 1827 | /* |
| 1828 | * We can't complete the direct I/O as requested, fall back to |
| 1829 | * buffered I/O. |
| 1830 | */ |
| 1831 | if (direct_io && !can_do_direct) { |
| 1832 | ocfs2_rw_unlock(inode, rw_level); |
| 1833 | up_read(&inode->i_alloc_sem); |
| 1834 | |
| 1835 | have_alloc_sem = 0; |
| 1836 | rw_level = -1; |
| 1837 | |
| 1838 | direct_io = 0; |
| 1839 | goto relock; |
| 1840 | } |
| 1841 | |
| 1842 | /* |
| 1843 | * To later detect whether a journal commit for sync writes is |
| 1844 | * necessary, we sample i_size, and cluster count here. |
| 1845 | */ |
| 1846 | old_size = i_size_read(inode); |
| 1847 | old_clusters = OCFS2_I(inode)->ip_clusters; |
| 1848 | |
| 1849 | /* communicate with ocfs2_dio_end_io */ |
| 1850 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
| 1851 | |
| 1852 | if (direct_io) { |
| 1853 | ret = generic_segment_checks(iov, &nr_segs, &ocount, |
| 1854 | VERIFY_READ); |
| 1855 | if (ret) |
| 1856 | goto out_dio; |
| 1857 | |
| 1858 | ret = generic_write_checks(file, ppos, &count, |
| 1859 | S_ISBLK(inode->i_mode)); |
| 1860 | if (ret) |
| 1861 | goto out_dio; |
| 1862 | |
| 1863 | written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos, |
| 1864 | ppos, count, ocount); |
| 1865 | if (written < 0) { |
| 1866 | ret = written; |
| 1867 | goto out_dio; |
| 1868 | } |
| 1869 | } else { |
| 1870 | written = generic_file_aio_write_nolock(iocb, iov, nr_segs, |
| 1871 | *ppos); |
| 1872 | } |
| 1873 | |
| 1874 | out_dio: |
| 1875 | /* buffered aio wouldn't have proper lock coverage today */ |
| 1876 | BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT)); |
| 1877 | |
| 1878 | if ((file->f_flags & O_SYNC && !direct_io) || IS_SYNC(inode)) { |
| 1879 | /* |
| 1880 | * The generic write paths have handled getting data |
| 1881 | * to disk, but since we don't make use of the dirty |
| 1882 | * inode list, a manual journal commit is necessary |
| 1883 | * here. |
| 1884 | */ |
| 1885 | if (old_size != i_size_read(inode) || |
| 1886 | old_clusters != OCFS2_I(inode)->ip_clusters) { |
| 1887 | ret = journal_force_commit(osb->journal->j_journal); |
| 1888 | if (ret < 0) |
| 1889 | written = ret; |
| 1890 | } |
| 1891 | } |
| 1892 | |
| 1893 | /* |
| 1894 | * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io |
| 1895 | * function pointer which is called when o_direct io completes so that |
| 1896 | * it can unlock our rw lock. (it's the clustered equivalent of |
| 1897 | * i_alloc_sem; protects truncate from racing with pending ios). |
| 1898 | * Unfortunately there are error cases which call end_io and others |
| 1899 | * that don't. so we don't have to unlock the rw_lock if either an |
| 1900 | * async dio is going to do it in the future or an end_io after an |
| 1901 | * error has already done it. |
| 1902 | */ |
| 1903 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { |
| 1904 | rw_level = -1; |
| 1905 | have_alloc_sem = 0; |
| 1906 | } |
| 1907 | |
| 1908 | out: |
| 1909 | if (rw_level != -1) |
| 1910 | ocfs2_rw_unlock(inode, rw_level); |
| 1911 | |
| 1912 | out_sems: |
| 1913 | if (have_alloc_sem) |
| 1914 | up_read(&inode->i_alloc_sem); |
| 1915 | |
| 1916 | mutex_unlock(&inode->i_mutex); |
| 1917 | |
| 1918 | mlog_exit(ret); |
| 1919 | return written ? written : ret; |
| 1920 | } |
| 1921 | |
| 1922 | static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe, |
| 1923 | struct file *out, |
| 1924 | loff_t *ppos, |
| 1925 | size_t len, |
| 1926 | unsigned int flags) |
| 1927 | { |
| 1928 | int ret; |
| 1929 | struct inode *inode = out->f_path.dentry->d_inode; |
| 1930 | |
| 1931 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe, |
| 1932 | (unsigned int)len, |
| 1933 | out->f_path.dentry->d_name.len, |
| 1934 | out->f_path.dentry->d_name.name); |
| 1935 | |
| 1936 | inode_double_lock(inode, pipe->inode); |
| 1937 | |
| 1938 | ret = ocfs2_rw_lock(inode, 1); |
| 1939 | if (ret < 0) { |
| 1940 | mlog_errno(ret); |
| 1941 | goto out; |
| 1942 | } |
| 1943 | |
| 1944 | ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0, |
| 1945 | NULL); |
| 1946 | if (ret < 0) { |
| 1947 | mlog_errno(ret); |
| 1948 | goto out_unlock; |
| 1949 | } |
| 1950 | |
| 1951 | ret = generic_file_splice_write_nolock(pipe, out, ppos, len, flags); |
| 1952 | |
| 1953 | out_unlock: |
| 1954 | ocfs2_rw_unlock(inode, 1); |
| 1955 | out: |
| 1956 | inode_double_unlock(inode, pipe->inode); |
| 1957 | |
| 1958 | mlog_exit(ret); |
| 1959 | return ret; |
| 1960 | } |
| 1961 | |
| 1962 | static ssize_t ocfs2_file_splice_read(struct file *in, |
| 1963 | loff_t *ppos, |
| 1964 | struct pipe_inode_info *pipe, |
| 1965 | size_t len, |
| 1966 | unsigned int flags) |
| 1967 | { |
| 1968 | int ret = 0; |
| 1969 | struct inode *inode = in->f_path.dentry->d_inode; |
| 1970 | |
| 1971 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe, |
| 1972 | (unsigned int)len, |
| 1973 | in->f_path.dentry->d_name.len, |
| 1974 | in->f_path.dentry->d_name.name); |
| 1975 | |
| 1976 | /* |
| 1977 | * See the comment in ocfs2_file_aio_read() |
| 1978 | */ |
| 1979 | ret = ocfs2_inode_lock(inode, NULL, 0); |
| 1980 | if (ret < 0) { |
| 1981 | mlog_errno(ret); |
| 1982 | goto bail; |
| 1983 | } |
| 1984 | ocfs2_inode_unlock(inode, 0); |
| 1985 | |
| 1986 | ret = generic_file_splice_read(in, ppos, pipe, len, flags); |
| 1987 | |
| 1988 | bail: |
| 1989 | mlog_exit(ret); |
| 1990 | return ret; |
| 1991 | } |
| 1992 | |
| 1993 | static ssize_t ocfs2_file_aio_read(struct kiocb *iocb, |
| 1994 | const struct iovec *iov, |
| 1995 | unsigned long nr_segs, |
| 1996 | loff_t pos) |
| 1997 | { |
| 1998 | int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0; |
| 1999 | struct file *filp = iocb->ki_filp; |
| 2000 | struct inode *inode = filp->f_path.dentry->d_inode; |
| 2001 | |
| 2002 | mlog_entry("(0x%p, %u, '%.*s')\n", filp, |
| 2003 | (unsigned int)nr_segs, |
| 2004 | filp->f_path.dentry->d_name.len, |
| 2005 | filp->f_path.dentry->d_name.name); |
| 2006 | |
| 2007 | if (!inode) { |
| 2008 | ret = -EINVAL; |
| 2009 | mlog_errno(ret); |
| 2010 | goto bail; |
| 2011 | } |
| 2012 | |
| 2013 | /* |
| 2014 | * buffered reads protect themselves in ->readpage(). O_DIRECT reads |
| 2015 | * need locks to protect pending reads from racing with truncate. |
| 2016 | */ |
| 2017 | if (filp->f_flags & O_DIRECT) { |
| 2018 | down_read(&inode->i_alloc_sem); |
| 2019 | have_alloc_sem = 1; |
| 2020 | |
| 2021 | ret = ocfs2_rw_lock(inode, 0); |
| 2022 | if (ret < 0) { |
| 2023 | mlog_errno(ret); |
| 2024 | goto bail; |
| 2025 | } |
| 2026 | rw_level = 0; |
| 2027 | /* communicate with ocfs2_dio_end_io */ |
| 2028 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
| 2029 | } |
| 2030 | |
| 2031 | /* |
| 2032 | * We're fine letting folks race truncates and extending |
| 2033 | * writes with read across the cluster, just like they can |
| 2034 | * locally. Hence no rw_lock during read. |
| 2035 | * |
| 2036 | * Take and drop the meta data lock to update inode fields |
| 2037 | * like i_size. This allows the checks down below |
| 2038 | * generic_file_aio_read() a chance of actually working. |
| 2039 | */ |
| 2040 | ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level); |
| 2041 | if (ret < 0) { |
| 2042 | mlog_errno(ret); |
| 2043 | goto bail; |
| 2044 | } |
| 2045 | ocfs2_inode_unlock(inode, lock_level); |
| 2046 | |
| 2047 | ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos); |
| 2048 | if (ret == -EINVAL) |
| 2049 | mlog(0, "generic_file_aio_read returned -EINVAL\n"); |
| 2050 | |
| 2051 | /* buffered aio wouldn't have proper lock coverage today */ |
| 2052 | BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT)); |
| 2053 | |
| 2054 | /* see ocfs2_file_aio_write */ |
| 2055 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { |
| 2056 | rw_level = -1; |
| 2057 | have_alloc_sem = 0; |
| 2058 | } |
| 2059 | |
| 2060 | bail: |
| 2061 | if (have_alloc_sem) |
| 2062 | up_read(&inode->i_alloc_sem); |
| 2063 | if (rw_level != -1) |
| 2064 | ocfs2_rw_unlock(inode, rw_level); |
| 2065 | mlog_exit(ret); |
| 2066 | |
| 2067 | return ret; |
| 2068 | } |
| 2069 | |
| 2070 | const struct inode_operations ocfs2_file_iops = { |
| 2071 | .setattr = ocfs2_setattr, |
| 2072 | .getattr = ocfs2_getattr, |
| 2073 | .permission = ocfs2_permission, |
| 2074 | .setxattr = generic_setxattr, |
| 2075 | .getxattr = generic_getxattr, |
| 2076 | .listxattr = ocfs2_listxattr, |
| 2077 | .removexattr = generic_removexattr, |
| 2078 | .fallocate = ocfs2_fallocate, |
| 2079 | .fiemap = ocfs2_fiemap, |
| 2080 | }; |
| 2081 | |
| 2082 | const struct inode_operations ocfs2_special_file_iops = { |
| 2083 | .setattr = ocfs2_setattr, |
| 2084 | .getattr = ocfs2_getattr, |
| 2085 | .permission = ocfs2_permission, |
| 2086 | }; |
| 2087 | |
| 2088 | /* |
| 2089 | * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with |
| 2090 | * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks! |
| 2091 | */ |
| 2092 | const struct file_operations ocfs2_fops = { |
| 2093 | .llseek = generic_file_llseek, |
| 2094 | .read = do_sync_read, |
| 2095 | .write = do_sync_write, |
| 2096 | .mmap = ocfs2_mmap, |
| 2097 | .fsync = ocfs2_sync_file, |
| 2098 | .release = ocfs2_file_release, |
| 2099 | .open = ocfs2_file_open, |
| 2100 | .aio_read = ocfs2_file_aio_read, |
| 2101 | .aio_write = ocfs2_file_aio_write, |
| 2102 | .unlocked_ioctl = ocfs2_ioctl, |
| 2103 | #ifdef CONFIG_COMPAT |
| 2104 | .compat_ioctl = ocfs2_compat_ioctl, |
| 2105 | #endif |
| 2106 | .lock = ocfs2_lock, |
| 2107 | .flock = ocfs2_flock, |
| 2108 | .splice_read = ocfs2_file_splice_read, |
| 2109 | .splice_write = ocfs2_file_splice_write, |
| 2110 | }; |
| 2111 | |
| 2112 | const struct file_operations ocfs2_dops = { |
| 2113 | .llseek = generic_file_llseek, |
| 2114 | .read = generic_read_dir, |
| 2115 | .readdir = ocfs2_readdir, |
| 2116 | .fsync = ocfs2_sync_file, |
| 2117 | .release = ocfs2_dir_release, |
| 2118 | .open = ocfs2_dir_open, |
| 2119 | .unlocked_ioctl = ocfs2_ioctl, |
| 2120 | #ifdef CONFIG_COMPAT |
| 2121 | .compat_ioctl = ocfs2_compat_ioctl, |
| 2122 | #endif |
| 2123 | .lock = ocfs2_lock, |
| 2124 | .flock = ocfs2_flock, |
| 2125 | }; |
| 2126 | |
| 2127 | /* |
| 2128 | * POSIX-lockless variants of our file_operations. |
| 2129 | * |
| 2130 | * These will be used if the underlying cluster stack does not support |
| 2131 | * posix file locking, if the user passes the "localflocks" mount |
| 2132 | * option, or if we have a local-only fs. |
| 2133 | * |
| 2134 | * ocfs2_flock is in here because all stacks handle UNIX file locks, |
| 2135 | * so we still want it in the case of no stack support for |
| 2136 | * plocks. Internally, it will do the right thing when asked to ignore |
| 2137 | * the cluster. |
| 2138 | */ |
| 2139 | const struct file_operations ocfs2_fops_no_plocks = { |
| 2140 | .llseek = generic_file_llseek, |
| 2141 | .read = do_sync_read, |
| 2142 | .write = do_sync_write, |
| 2143 | .mmap = ocfs2_mmap, |
| 2144 | .fsync = ocfs2_sync_file, |
| 2145 | .release = ocfs2_file_release, |
| 2146 | .open = ocfs2_file_open, |
| 2147 | .aio_read = ocfs2_file_aio_read, |
| 2148 | .aio_write = ocfs2_file_aio_write, |
| 2149 | .unlocked_ioctl = ocfs2_ioctl, |
| 2150 | #ifdef CONFIG_COMPAT |
| 2151 | .compat_ioctl = ocfs2_compat_ioctl, |
| 2152 | #endif |
| 2153 | .flock = ocfs2_flock, |
| 2154 | .splice_read = ocfs2_file_splice_read, |
| 2155 | .splice_write = ocfs2_file_splice_write, |
| 2156 | }; |
| 2157 | |
| 2158 | const struct file_operations ocfs2_dops_no_plocks = { |
| 2159 | .llseek = generic_file_llseek, |
| 2160 | .read = generic_read_dir, |
| 2161 | .readdir = ocfs2_readdir, |
| 2162 | .fsync = ocfs2_sync_file, |
| 2163 | .release = ocfs2_dir_release, |
| 2164 | .open = ocfs2_dir_open, |
| 2165 | .unlocked_ioctl = ocfs2_ioctl, |
| 2166 | #ifdef CONFIG_COMPAT |
| 2167 | .compat_ioctl = ocfs2_compat_ioctl, |
| 2168 | #endif |
| 2169 | .flock = ocfs2_flock, |
| 2170 | }; |