1 /* 2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. 4 * 5 * This copyrighted material is made available to anyone wishing to use, 6 * modify, copy, or redistribute it subject to the terms and conditions 7 * of the GNU General Public License version 2. 8 */ 9 10 #include <linux/slab.h> 11 #include <linux/spinlock.h> 12 #include <linux/completion.h> 13 #include <linux/buffer_head.h> 14 #include <linux/pagemap.h> 15 #include <linux/uio.h> 16 #include <linux/blkdev.h> 17 #include <linux/mm.h> 18 #include <linux/mount.h> 19 #include <linux/fs.h> 20 #include <linux/gfs2_ondisk.h> 21 #include <linux/ext2_fs.h> 22 #include <linux/falloc.h> 23 #include <linux/swap.h> 24 #include <linux/crc32.h> 25 #include <linux/writeback.h> 26 #include <asm/uaccess.h> 27 #include <linux/dlm.h> 28 #include <linux/dlm_plock.h> 29 30 #include "gfs2.h" 31 #include "incore.h" 32 #include "bmap.h" 33 #include "dir.h" 34 #include "glock.h" 35 #include "glops.h" 36 #include "inode.h" 37 #include "log.h" 38 #include "meta_io.h" 39 #include "quota.h" 40 #include "rgrp.h" 41 #include "trans.h" 42 #include "util.h" 43 44 /** 45 * gfs2_llseek - seek to a location in a file 46 * @file: the file 47 * @offset: the offset 48 * @origin: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END) 49 * 50 * SEEK_END requires the glock for the file because it references the 51 * file's size. 52 * 53 * Returns: The new offset, or errno 54 */ 55 56 static loff_t gfs2_llseek(struct file *file, loff_t offset, int origin) 57 { 58 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 59 struct gfs2_holder i_gh; 60 loff_t error; 61 62 if (origin == 2) { 63 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, 64 &i_gh); 65 if (!error) { 66 error = generic_file_llseek_unlocked(file, offset, origin); 67 gfs2_glock_dq_uninit(&i_gh); 68 } 69 } else 70 error = generic_file_llseek_unlocked(file, offset, origin); 71 72 return error; 73 } 74 75 /** 76 * gfs2_readdir - Read directory entries from a directory 77 * @file: The directory to read from 78 * @dirent: Buffer for dirents 79 * @filldir: Function used to do the copying 80 * 81 * Returns: errno 82 */ 83 84 static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir) 85 { 86 struct inode *dir = file->f_mapping->host; 87 struct gfs2_inode *dip = GFS2_I(dir); 88 struct gfs2_holder d_gh; 89 u64 offset = file->f_pos; 90 int error; 91 92 gfs2_holder_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh); 93 error = gfs2_glock_nq(&d_gh); 94 if (error) { 95 gfs2_holder_uninit(&d_gh); 96 return error; 97 } 98 99 error = gfs2_dir_read(dir, &offset, dirent, filldir); 100 101 gfs2_glock_dq_uninit(&d_gh); 102 103 file->f_pos = offset; 104 105 return error; 106 } 107 108 /** 109 * fsflags_cvt 110 * @table: A table of 32 u32 flags 111 * @val: a 32 bit value to convert 112 * 113 * This function can be used to convert between fsflags values and 114 * GFS2's own flags values. 115 * 116 * Returns: the converted flags 117 */ 118 static u32 fsflags_cvt(const u32 *table, u32 val) 119 { 120 u32 res = 0; 121 while(val) { 122 if (val & 1) 123 res |= *table; 124 table++; 125 val >>= 1; 126 } 127 return res; 128 } 129 130 static const u32 fsflags_to_gfs2[32] = { 131 [3] = GFS2_DIF_SYNC, 132 [4] = GFS2_DIF_IMMUTABLE, 133 [5] = GFS2_DIF_APPENDONLY, 134 [7] = GFS2_DIF_NOATIME, 135 [12] = GFS2_DIF_EXHASH, 136 [14] = GFS2_DIF_INHERIT_JDATA, 137 }; 138 139 static const u32 gfs2_to_fsflags[32] = { 140 [gfs2fl_Sync] = FS_SYNC_FL, 141 [gfs2fl_Immutable] = FS_IMMUTABLE_FL, 142 [gfs2fl_AppendOnly] = FS_APPEND_FL, 143 [gfs2fl_NoAtime] = FS_NOATIME_FL, 144 [gfs2fl_ExHash] = FS_INDEX_FL, 145 [gfs2fl_InheritJdata] = FS_JOURNAL_DATA_FL, 146 }; 147 148 static int gfs2_get_flags(struct file *filp, u32 __user *ptr) 149 { 150 struct inode *inode = filp->f_path.dentry->d_inode; 151 struct gfs2_inode *ip = GFS2_I(inode); 152 struct gfs2_holder gh; 153 int error; 154 u32 fsflags; 155 156 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); 157 error = gfs2_glock_nq(&gh); 158 if (error) 159 return error; 160 161 fsflags = fsflags_cvt(gfs2_to_fsflags, ip->i_diskflags); 162 if (!S_ISDIR(inode->i_mode) && ip->i_diskflags & GFS2_DIF_JDATA) 163 fsflags |= FS_JOURNAL_DATA_FL; 164 if (put_user(fsflags, ptr)) 165 error = -EFAULT; 166 167 gfs2_glock_dq(&gh); 168 gfs2_holder_uninit(&gh); 169 return error; 170 } 171 172 void gfs2_set_inode_flags(struct inode *inode) 173 { 174 struct gfs2_inode *ip = GFS2_I(inode); 175 unsigned int flags = inode->i_flags; 176 177 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); 178 if (ip->i_diskflags & GFS2_DIF_IMMUTABLE) 179 flags |= S_IMMUTABLE; 180 if (ip->i_diskflags & GFS2_DIF_APPENDONLY) 181 flags |= S_APPEND; 182 if (ip->i_diskflags & GFS2_DIF_NOATIME) 183 flags |= S_NOATIME; 184 if (ip->i_diskflags & GFS2_DIF_SYNC) 185 flags |= S_SYNC; 186 inode->i_flags = flags; 187 } 188 189 /* Flags that can be set by user space */ 190 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \ 191 GFS2_DIF_IMMUTABLE| \ 192 GFS2_DIF_APPENDONLY| \ 193 GFS2_DIF_NOATIME| \ 194 GFS2_DIF_SYNC| \ 195 GFS2_DIF_SYSTEM| \ 196 GFS2_DIF_INHERIT_JDATA) 197 198 /** 199 * gfs2_set_flags - set flags on an inode 200 * @inode: The inode 201 * @flags: The flags to set 202 * @mask: Indicates which flags are valid 203 * 204 */ 205 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask) 206 { 207 struct inode *inode = filp->f_path.dentry->d_inode; 208 struct gfs2_inode *ip = GFS2_I(inode); 209 struct gfs2_sbd *sdp = GFS2_SB(inode); 210 struct buffer_head *bh; 211 struct gfs2_holder gh; 212 int error; 213 u32 new_flags, flags; 214 215 error = mnt_want_write(filp->f_path.mnt); 216 if (error) 217 return error; 218 219 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 220 if (error) 221 goto out_drop_write; 222 223 error = -EACCES; 224 if (!inode_owner_or_capable(inode)) 225 goto out; 226 227 error = 0; 228 flags = ip->i_diskflags; 229 new_flags = (flags & ~mask) | (reqflags & mask); 230 if ((new_flags ^ flags) == 0) 231 goto out; 232 233 error = -EINVAL; 234 if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET) 235 goto out; 236 237 error = -EPERM; 238 if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE)) 239 goto out; 240 if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY)) 241 goto out; 242 if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) && 243 !capable(CAP_LINUX_IMMUTABLE)) 244 goto out; 245 if (!IS_IMMUTABLE(inode)) { 246 error = gfs2_permission(inode, MAY_WRITE, 0); 247 if (error) 248 goto out; 249 } 250 if ((flags ^ new_flags) & GFS2_DIF_JDATA) { 251 if (flags & GFS2_DIF_JDATA) 252 gfs2_log_flush(sdp, ip->i_gl); 253 error = filemap_fdatawrite(inode->i_mapping); 254 if (error) 255 goto out; 256 error = filemap_fdatawait(inode->i_mapping); 257 if (error) 258 goto out; 259 } 260 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 261 if (error) 262 goto out; 263 error = gfs2_meta_inode_buffer(ip, &bh); 264 if (error) 265 goto out_trans_end; 266 gfs2_trans_add_bh(ip->i_gl, bh, 1); 267 ip->i_diskflags = new_flags; 268 gfs2_dinode_out(ip, bh->b_data); 269 brelse(bh); 270 gfs2_set_inode_flags(inode); 271 gfs2_set_aops(inode); 272 out_trans_end: 273 gfs2_trans_end(sdp); 274 out: 275 gfs2_glock_dq_uninit(&gh); 276 out_drop_write: 277 mnt_drop_write(filp->f_path.mnt); 278 return error; 279 } 280 281 static int gfs2_set_flags(struct file *filp, u32 __user *ptr) 282 { 283 struct inode *inode = filp->f_path.dentry->d_inode; 284 u32 fsflags, gfsflags; 285 286 if (get_user(fsflags, ptr)) 287 return -EFAULT; 288 289 gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags); 290 if (!S_ISDIR(inode->i_mode)) { 291 if (gfsflags & GFS2_DIF_INHERIT_JDATA) 292 gfsflags ^= (GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA); 293 return do_gfs2_set_flags(filp, gfsflags, ~0); 294 } 295 return do_gfs2_set_flags(filp, gfsflags, ~GFS2_DIF_JDATA); 296 } 297 298 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) 299 { 300 switch(cmd) { 301 case FS_IOC_GETFLAGS: 302 return gfs2_get_flags(filp, (u32 __user *)arg); 303 case FS_IOC_SETFLAGS: 304 return gfs2_set_flags(filp, (u32 __user *)arg); 305 } 306 return -ENOTTY; 307 } 308 309 /** 310 * gfs2_allocate_page_backing - Use bmap to allocate blocks 311 * @page: The (locked) page to allocate backing for 312 * 313 * We try to allocate all the blocks required for the page in 314 * one go. This might fail for various reasons, so we keep 315 * trying until all the blocks to back this page are allocated. 316 * If some of the blocks are already allocated, thats ok too. 317 */ 318 319 static int gfs2_allocate_page_backing(struct page *page) 320 { 321 struct inode *inode = page->mapping->host; 322 struct buffer_head bh; 323 unsigned long size = PAGE_CACHE_SIZE; 324 u64 lblock = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 325 326 do { 327 bh.b_state = 0; 328 bh.b_size = size; 329 gfs2_block_map(inode, lblock, &bh, 1); 330 if (!buffer_mapped(&bh)) 331 return -EIO; 332 size -= bh.b_size; 333 lblock += (bh.b_size >> inode->i_blkbits); 334 } while(size > 0); 335 return 0; 336 } 337 338 /** 339 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable 340 * @vma: The virtual memory area 341 * @page: The page which is about to become writable 342 * 343 * When the page becomes writable, we need to ensure that we have 344 * blocks allocated on disk to back that page. 345 */ 346 347 static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) 348 { 349 struct page *page = vmf->page; 350 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 351 struct gfs2_inode *ip = GFS2_I(inode); 352 struct gfs2_sbd *sdp = GFS2_SB(inode); 353 unsigned long last_index; 354 u64 pos = page->index << PAGE_CACHE_SHIFT; 355 unsigned int data_blocks, ind_blocks, rblocks; 356 struct gfs2_holder gh; 357 struct gfs2_alloc *al; 358 int ret; 359 360 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 361 ret = gfs2_glock_nq(&gh); 362 if (ret) 363 goto out; 364 365 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags); 366 set_bit(GIF_SW_PAGED, &ip->i_flags); 367 368 if (!gfs2_write_alloc_required(ip, pos, PAGE_CACHE_SIZE)) 369 goto out_unlock; 370 ret = -ENOMEM; 371 al = gfs2_alloc_get(ip); 372 if (al == NULL) 373 goto out_unlock; 374 375 ret = gfs2_quota_lock_check(ip); 376 if (ret) 377 goto out_alloc_put; 378 gfs2_write_calc_reserv(ip, PAGE_CACHE_SIZE, &data_blocks, &ind_blocks); 379 al->al_requested = data_blocks + ind_blocks; 380 ret = gfs2_inplace_reserve(ip); 381 if (ret) 382 goto out_quota_unlock; 383 384 rblocks = RES_DINODE + ind_blocks; 385 if (gfs2_is_jdata(ip)) 386 rblocks += data_blocks ? data_blocks : 1; 387 if (ind_blocks || data_blocks) { 388 rblocks += RES_STATFS + RES_QUOTA; 389 rblocks += gfs2_rg_blocks(al); 390 } 391 ret = gfs2_trans_begin(sdp, rblocks, 0); 392 if (ret) 393 goto out_trans_fail; 394 395 lock_page(page); 396 ret = -EINVAL; 397 last_index = ip->i_inode.i_size >> PAGE_CACHE_SHIFT; 398 if (page->index > last_index) 399 goto out_unlock_page; 400 ret = 0; 401 if (!PageUptodate(page) || page->mapping != ip->i_inode.i_mapping) 402 goto out_unlock_page; 403 if (gfs2_is_stuffed(ip)) { 404 ret = gfs2_unstuff_dinode(ip, page); 405 if (ret) 406 goto out_unlock_page; 407 } 408 ret = gfs2_allocate_page_backing(page); 409 410 out_unlock_page: 411 unlock_page(page); 412 gfs2_trans_end(sdp); 413 out_trans_fail: 414 gfs2_inplace_release(ip); 415 out_quota_unlock: 416 gfs2_quota_unlock(ip); 417 out_alloc_put: 418 gfs2_alloc_put(ip); 419 out_unlock: 420 gfs2_glock_dq(&gh); 421 out: 422 gfs2_holder_uninit(&gh); 423 if (ret == -ENOMEM) 424 ret = VM_FAULT_OOM; 425 else if (ret) 426 ret = VM_FAULT_SIGBUS; 427 return ret; 428 } 429 430 static const struct vm_operations_struct gfs2_vm_ops = { 431 .fault = filemap_fault, 432 .page_mkwrite = gfs2_page_mkwrite, 433 }; 434 435 /** 436 * gfs2_mmap - 437 * @file: The file to map 438 * @vma: The VMA which described the mapping 439 * 440 * There is no need to get a lock here unless we should be updating 441 * atime. We ignore any locking errors since the only consequence is 442 * a missed atime update (which will just be deferred until later). 443 * 444 * Returns: 0 445 */ 446 447 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma) 448 { 449 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 450 451 if (!(file->f_flags & O_NOATIME) && 452 !IS_NOATIME(&ip->i_inode)) { 453 struct gfs2_holder i_gh; 454 int error; 455 456 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh); 457 error = gfs2_glock_nq(&i_gh); 458 if (error == 0) { 459 file_accessed(file); 460 gfs2_glock_dq(&i_gh); 461 } 462 gfs2_holder_uninit(&i_gh); 463 if (error) 464 return error; 465 } 466 vma->vm_ops = &gfs2_vm_ops; 467 vma->vm_flags |= VM_CAN_NONLINEAR; 468 469 return 0; 470 } 471 472 /** 473 * gfs2_open - open a file 474 * @inode: the inode to open 475 * @file: the struct file for this opening 476 * 477 * Returns: errno 478 */ 479 480 static int gfs2_open(struct inode *inode, struct file *file) 481 { 482 struct gfs2_inode *ip = GFS2_I(inode); 483 struct gfs2_holder i_gh; 484 struct gfs2_file *fp; 485 int error; 486 487 fp = kzalloc(sizeof(struct gfs2_file), GFP_KERNEL); 488 if (!fp) 489 return -ENOMEM; 490 491 mutex_init(&fp->f_fl_mutex); 492 493 gfs2_assert_warn(GFS2_SB(inode), !file->private_data); 494 file->private_data = fp; 495 496 if (S_ISREG(ip->i_inode.i_mode)) { 497 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, 498 &i_gh); 499 if (error) 500 goto fail; 501 502 if (!(file->f_flags & O_LARGEFILE) && 503 i_size_read(inode) > MAX_NON_LFS) { 504 error = -EOVERFLOW; 505 goto fail_gunlock; 506 } 507 508 gfs2_glock_dq_uninit(&i_gh); 509 } 510 511 return 0; 512 513 fail_gunlock: 514 gfs2_glock_dq_uninit(&i_gh); 515 fail: 516 file->private_data = NULL; 517 kfree(fp); 518 return error; 519 } 520 521 /** 522 * gfs2_close - called to close a struct file 523 * @inode: the inode the struct file belongs to 524 * @file: the struct file being closed 525 * 526 * Returns: errno 527 */ 528 529 static int gfs2_close(struct inode *inode, struct file *file) 530 { 531 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info; 532 struct gfs2_file *fp; 533 534 fp = file->private_data; 535 file->private_data = NULL; 536 537 if (gfs2_assert_warn(sdp, fp)) 538 return -EIO; 539 540 kfree(fp); 541 542 return 0; 543 } 544 545 /** 546 * gfs2_fsync - sync the dirty data for a file (across the cluster) 547 * @file: the file that points to the dentry (we ignore this) 548 * @dentry: the dentry that points to the inode to sync 549 * 550 * The VFS will flush "normal" data for us. We only need to worry 551 * about metadata here. For journaled data, we just do a log flush 552 * as we can't avoid it. Otherwise we can just bale out if datasync 553 * is set. For stuffed inodes we must flush the log in order to 554 * ensure that all data is on disk. 555 * 556 * The call to write_inode_now() is there to write back metadata and 557 * the inode itself. It does also try and write the data, but thats 558 * (hopefully) a no-op due to the VFS having already called filemap_fdatawrite() 559 * for us. 560 * 561 * Returns: errno 562 */ 563 564 static int gfs2_fsync(struct file *file, int datasync) 565 { 566 struct inode *inode = file->f_mapping->host; 567 int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC); 568 int ret = 0; 569 570 if (gfs2_is_jdata(GFS2_I(inode))) { 571 gfs2_log_flush(GFS2_SB(inode), GFS2_I(inode)->i_gl); 572 return 0; 573 } 574 575 if (sync_state != 0) { 576 if (!datasync) 577 ret = write_inode_now(inode, 0); 578 579 if (gfs2_is_stuffed(GFS2_I(inode))) 580 gfs2_log_flush(GFS2_SB(inode), GFS2_I(inode)->i_gl); 581 } 582 583 return ret; 584 } 585 586 /** 587 * gfs2_file_aio_write - Perform a write to a file 588 * @iocb: The io context 589 * @iov: The data to write 590 * @nr_segs: Number of @iov segments 591 * @pos: The file position 592 * 593 * We have to do a lock/unlock here to refresh the inode size for 594 * O_APPEND writes, otherwise we can land up writing at the wrong 595 * offset. There is still a race, but provided the app is using its 596 * own file locking, this will make O_APPEND work as expected. 597 * 598 */ 599 600 static ssize_t gfs2_file_aio_write(struct kiocb *iocb, const struct iovec *iov, 601 unsigned long nr_segs, loff_t pos) 602 { 603 struct file *file = iocb->ki_filp; 604 605 if (file->f_flags & O_APPEND) { 606 struct dentry *dentry = file->f_dentry; 607 struct gfs2_inode *ip = GFS2_I(dentry->d_inode); 608 struct gfs2_holder gh; 609 int ret; 610 611 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh); 612 if (ret) 613 return ret; 614 gfs2_glock_dq_uninit(&gh); 615 } 616 617 return generic_file_aio_write(iocb, iov, nr_segs, pos); 618 } 619 620 static int empty_write_end(struct page *page, unsigned from, 621 unsigned to, int mode) 622 { 623 struct inode *inode = page->mapping->host; 624 struct gfs2_inode *ip = GFS2_I(inode); 625 struct buffer_head *bh; 626 unsigned offset, blksize = 1 << inode->i_blkbits; 627 pgoff_t end_index = i_size_read(inode) >> PAGE_CACHE_SHIFT; 628 629 zero_user(page, from, to-from); 630 mark_page_accessed(page); 631 632 if (page->index < end_index || !(mode & FALLOC_FL_KEEP_SIZE)) { 633 if (!gfs2_is_writeback(ip)) 634 gfs2_page_add_databufs(ip, page, from, to); 635 636 block_commit_write(page, from, to); 637 return 0; 638 } 639 640 offset = 0; 641 bh = page_buffers(page); 642 while (offset < to) { 643 if (offset >= from) { 644 set_buffer_uptodate(bh); 645 mark_buffer_dirty(bh); 646 clear_buffer_new(bh); 647 write_dirty_buffer(bh, WRITE); 648 } 649 offset += blksize; 650 bh = bh->b_this_page; 651 } 652 653 offset = 0; 654 bh = page_buffers(page); 655 while (offset < to) { 656 if (offset >= from) { 657 wait_on_buffer(bh); 658 if (!buffer_uptodate(bh)) 659 return -EIO; 660 } 661 offset += blksize; 662 bh = bh->b_this_page; 663 } 664 return 0; 665 } 666 667 static int needs_empty_write(sector_t block, struct inode *inode) 668 { 669 int error; 670 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 }; 671 672 bh_map.b_size = 1 << inode->i_blkbits; 673 error = gfs2_block_map(inode, block, &bh_map, 0); 674 if (unlikely(error)) 675 return error; 676 return !buffer_mapped(&bh_map); 677 } 678 679 static int write_empty_blocks(struct page *page, unsigned from, unsigned to, 680 int mode) 681 { 682 struct inode *inode = page->mapping->host; 683 unsigned start, end, next, blksize; 684 sector_t block = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 685 int ret; 686 687 blksize = 1 << inode->i_blkbits; 688 next = end = 0; 689 while (next < from) { 690 next += blksize; 691 block++; 692 } 693 start = next; 694 do { 695 next += blksize; 696 ret = needs_empty_write(block, inode); 697 if (unlikely(ret < 0)) 698 return ret; 699 if (ret == 0) { 700 if (end) { 701 ret = __block_write_begin(page, start, end - start, 702 gfs2_block_map); 703 if (unlikely(ret)) 704 return ret; 705 ret = empty_write_end(page, start, end, mode); 706 if (unlikely(ret)) 707 return ret; 708 end = 0; 709 } 710 start = next; 711 } 712 else 713 end = next; 714 block++; 715 } while (next < to); 716 717 if (end) { 718 ret = __block_write_begin(page, start, end - start, gfs2_block_map); 719 if (unlikely(ret)) 720 return ret; 721 ret = empty_write_end(page, start, end, mode); 722 if (unlikely(ret)) 723 return ret; 724 } 725 726 return 0; 727 } 728 729 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len, 730 int mode) 731 { 732 struct gfs2_inode *ip = GFS2_I(inode); 733 struct buffer_head *dibh; 734 int error; 735 u64 start = offset >> PAGE_CACHE_SHIFT; 736 unsigned int start_offset = offset & ~PAGE_CACHE_MASK; 737 u64 end = (offset + len - 1) >> PAGE_CACHE_SHIFT; 738 pgoff_t curr; 739 struct page *page; 740 unsigned int end_offset = (offset + len) & ~PAGE_CACHE_MASK; 741 unsigned int from, to; 742 743 if (!end_offset) 744 end_offset = PAGE_CACHE_SIZE; 745 746 error = gfs2_meta_inode_buffer(ip, &dibh); 747 if (unlikely(error)) 748 goto out; 749 750 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 751 752 if (gfs2_is_stuffed(ip)) { 753 error = gfs2_unstuff_dinode(ip, NULL); 754 if (unlikely(error)) 755 goto out; 756 } 757 758 curr = start; 759 offset = start << PAGE_CACHE_SHIFT; 760 from = start_offset; 761 to = PAGE_CACHE_SIZE; 762 while (curr <= end) { 763 page = grab_cache_page_write_begin(inode->i_mapping, curr, 764 AOP_FLAG_NOFS); 765 if (unlikely(!page)) { 766 error = -ENOMEM; 767 goto out; 768 } 769 770 if (curr == end) 771 to = end_offset; 772 error = write_empty_blocks(page, from, to, mode); 773 if (!error && offset + to > inode->i_size && 774 !(mode & FALLOC_FL_KEEP_SIZE)) { 775 i_size_write(inode, offset + to); 776 } 777 unlock_page(page); 778 page_cache_release(page); 779 if (error) 780 goto out; 781 curr++; 782 offset += PAGE_CACHE_SIZE; 783 from = 0; 784 } 785 786 gfs2_dinode_out(ip, dibh->b_data); 787 mark_inode_dirty(inode); 788 789 brelse(dibh); 790 791 out: 792 return error; 793 } 794 795 static void calc_max_reserv(struct gfs2_inode *ip, loff_t max, loff_t *len, 796 unsigned int *data_blocks, unsigned int *ind_blocks) 797 { 798 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 799 unsigned int max_blocks = ip->i_alloc->al_rgd->rd_free_clone; 800 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1); 801 802 for (tmp = max_data; tmp > sdp->sd_diptrs;) { 803 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs); 804 max_data -= tmp; 805 } 806 /* This calculation isn't the exact reverse of gfs2_write_calc_reserve, 807 so it might end up with fewer data blocks */ 808 if (max_data <= *data_blocks) 809 return; 810 *data_blocks = max_data; 811 *ind_blocks = max_blocks - max_data; 812 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift; 813 if (*len > max) { 814 *len = max; 815 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks); 816 } 817 } 818 819 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, 820 loff_t len) 821 { 822 struct inode *inode = file->f_path.dentry->d_inode; 823 struct gfs2_sbd *sdp = GFS2_SB(inode); 824 struct gfs2_inode *ip = GFS2_I(inode); 825 unsigned int data_blocks = 0, ind_blocks = 0, rblocks; 826 loff_t bytes, max_bytes; 827 struct gfs2_alloc *al; 828 int error; 829 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift; 830 next = (next + 1) << sdp->sd_sb.sb_bsize_shift; 831 832 /* We only support the FALLOC_FL_KEEP_SIZE mode */ 833 if (mode & ~FALLOC_FL_KEEP_SIZE) 834 return -EOPNOTSUPP; 835 836 offset = (offset >> sdp->sd_sb.sb_bsize_shift) << 837 sdp->sd_sb.sb_bsize_shift; 838 839 len = next - offset; 840 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2; 841 if (!bytes) 842 bytes = UINT_MAX; 843 844 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh); 845 error = gfs2_glock_nq(&ip->i_gh); 846 if (unlikely(error)) 847 goto out_uninit; 848 849 if (!gfs2_write_alloc_required(ip, offset, len)) 850 goto out_unlock; 851 852 while (len > 0) { 853 if (len < bytes) 854 bytes = len; 855 al = gfs2_alloc_get(ip); 856 if (!al) { 857 error = -ENOMEM; 858 goto out_unlock; 859 } 860 861 error = gfs2_quota_lock_check(ip); 862 if (error) 863 goto out_alloc_put; 864 865 retry: 866 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks); 867 868 al->al_requested = data_blocks + ind_blocks; 869 error = gfs2_inplace_reserve(ip); 870 if (error) { 871 if (error == -ENOSPC && bytes > sdp->sd_sb.sb_bsize) { 872 bytes >>= 1; 873 goto retry; 874 } 875 goto out_qunlock; 876 } 877 max_bytes = bytes; 878 calc_max_reserv(ip, len, &max_bytes, &data_blocks, &ind_blocks); 879 al->al_requested = data_blocks + ind_blocks; 880 881 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA + 882 RES_RG_HDR + gfs2_rg_blocks(al); 883 if (gfs2_is_jdata(ip)) 884 rblocks += data_blocks ? data_blocks : 1; 885 886 error = gfs2_trans_begin(sdp, rblocks, 887 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize); 888 if (error) 889 goto out_trans_fail; 890 891 error = fallocate_chunk(inode, offset, max_bytes, mode); 892 gfs2_trans_end(sdp); 893 894 if (error) 895 goto out_trans_fail; 896 897 len -= max_bytes; 898 offset += max_bytes; 899 gfs2_inplace_release(ip); 900 gfs2_quota_unlock(ip); 901 gfs2_alloc_put(ip); 902 } 903 goto out_unlock; 904 905 out_trans_fail: 906 gfs2_inplace_release(ip); 907 out_qunlock: 908 gfs2_quota_unlock(ip); 909 out_alloc_put: 910 gfs2_alloc_put(ip); 911 out_unlock: 912 gfs2_glock_dq(&ip->i_gh); 913 out_uninit: 914 gfs2_holder_uninit(&ip->i_gh); 915 return error; 916 } 917 918 #ifdef CONFIG_GFS2_FS_LOCKING_DLM 919 920 /** 921 * gfs2_setlease - acquire/release a file lease 922 * @file: the file pointer 923 * @arg: lease type 924 * @fl: file lock 925 * 926 * We don't currently have a way to enforce a lease across the whole 927 * cluster; until we do, disable leases (by just returning -EINVAL), 928 * unless the administrator has requested purely local locking. 929 * 930 * Locking: called under lock_flocks 931 * 932 * Returns: errno 933 */ 934 935 static int gfs2_setlease(struct file *file, long arg, struct file_lock **fl) 936 { 937 return -EINVAL; 938 } 939 940 /** 941 * gfs2_lock - acquire/release a posix lock on a file 942 * @file: the file pointer 943 * @cmd: either modify or retrieve lock state, possibly wait 944 * @fl: type and range of lock 945 * 946 * Returns: errno 947 */ 948 949 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl) 950 { 951 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 952 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host); 953 struct lm_lockstruct *ls = &sdp->sd_lockstruct; 954 955 if (!(fl->fl_flags & FL_POSIX)) 956 return -ENOLCK; 957 if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK) 958 return -ENOLCK; 959 960 if (cmd == F_CANCELLK) { 961 /* Hack: */ 962 cmd = F_SETLK; 963 fl->fl_type = F_UNLCK; 964 } 965 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 966 return -EIO; 967 if (IS_GETLK(cmd)) 968 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl); 969 else if (fl->fl_type == F_UNLCK) 970 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl); 971 else 972 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl); 973 } 974 975 static int do_flock(struct file *file, int cmd, struct file_lock *fl) 976 { 977 struct gfs2_file *fp = file->private_data; 978 struct gfs2_holder *fl_gh = &fp->f_fl_gh; 979 struct gfs2_inode *ip = GFS2_I(file->f_path.dentry->d_inode); 980 struct gfs2_glock *gl; 981 unsigned int state; 982 int flags; 983 int error = 0; 984 985 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED; 986 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE; 987 988 mutex_lock(&fp->f_fl_mutex); 989 990 gl = fl_gh->gh_gl; 991 if (gl) { 992 if (fl_gh->gh_state == state) 993 goto out; 994 flock_lock_file_wait(file, 995 &(struct file_lock){.fl_type = F_UNLCK}); 996 gfs2_glock_dq_wait(fl_gh); 997 gfs2_holder_reinit(state, flags, fl_gh); 998 } else { 999 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr, 1000 &gfs2_flock_glops, CREATE, &gl); 1001 if (error) 1002 goto out; 1003 gfs2_holder_init(gl, state, flags, fl_gh); 1004 gfs2_glock_put(gl); 1005 } 1006 error = gfs2_glock_nq(fl_gh); 1007 if (error) { 1008 gfs2_holder_uninit(fl_gh); 1009 if (error == GLR_TRYFAILED) 1010 error = -EAGAIN; 1011 } else { 1012 error = flock_lock_file_wait(file, fl); 1013 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); 1014 } 1015 1016 out: 1017 mutex_unlock(&fp->f_fl_mutex); 1018 return error; 1019 } 1020 1021 static void do_unflock(struct file *file, struct file_lock *fl) 1022 { 1023 struct gfs2_file *fp = file->private_data; 1024 struct gfs2_holder *fl_gh = &fp->f_fl_gh; 1025 1026 mutex_lock(&fp->f_fl_mutex); 1027 flock_lock_file_wait(file, fl); 1028 if (fl_gh->gh_gl) { 1029 gfs2_glock_dq_wait(fl_gh); 1030 gfs2_holder_uninit(fl_gh); 1031 } 1032 mutex_unlock(&fp->f_fl_mutex); 1033 } 1034 1035 /** 1036 * gfs2_flock - acquire/release a flock lock on a file 1037 * @file: the file pointer 1038 * @cmd: either modify or retrieve lock state, possibly wait 1039 * @fl: type and range of lock 1040 * 1041 * Returns: errno 1042 */ 1043 1044 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl) 1045 { 1046 if (!(fl->fl_flags & FL_FLOCK)) 1047 return -ENOLCK; 1048 if (fl->fl_type & LOCK_MAND) 1049 return -EOPNOTSUPP; 1050 1051 if (fl->fl_type == F_UNLCK) { 1052 do_unflock(file, fl); 1053 return 0; 1054 } else { 1055 return do_flock(file, cmd, fl); 1056 } 1057 } 1058 1059 const struct file_operations gfs2_file_fops = { 1060 .llseek = gfs2_llseek, 1061 .read = do_sync_read, 1062 .aio_read = generic_file_aio_read, 1063 .write = do_sync_write, 1064 .aio_write = gfs2_file_aio_write, 1065 .unlocked_ioctl = gfs2_ioctl, 1066 .mmap = gfs2_mmap, 1067 .open = gfs2_open, 1068 .release = gfs2_close, 1069 .fsync = gfs2_fsync, 1070 .lock = gfs2_lock, 1071 .flock = gfs2_flock, 1072 .splice_read = generic_file_splice_read, 1073 .splice_write = generic_file_splice_write, 1074 .setlease = gfs2_setlease, 1075 .fallocate = gfs2_fallocate, 1076 }; 1077 1078 const struct file_operations gfs2_dir_fops = { 1079 .readdir = gfs2_readdir, 1080 .unlocked_ioctl = gfs2_ioctl, 1081 .open = gfs2_open, 1082 .release = gfs2_close, 1083 .fsync = gfs2_fsync, 1084 .lock = gfs2_lock, 1085 .flock = gfs2_flock, 1086 .llseek = default_llseek, 1087 }; 1088 1089 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */ 1090 1091 const struct file_operations gfs2_file_fops_nolock = { 1092 .llseek = gfs2_llseek, 1093 .read = do_sync_read, 1094 .aio_read = generic_file_aio_read, 1095 .write = do_sync_write, 1096 .aio_write = gfs2_file_aio_write, 1097 .unlocked_ioctl = gfs2_ioctl, 1098 .mmap = gfs2_mmap, 1099 .open = gfs2_open, 1100 .release = gfs2_close, 1101 .fsync = gfs2_fsync, 1102 .splice_read = generic_file_splice_read, 1103 .splice_write = generic_file_splice_write, 1104 .setlease = generic_setlease, 1105 .fallocate = gfs2_fallocate, 1106 }; 1107 1108 const struct file_operations gfs2_dir_fops_nolock = { 1109 .readdir = gfs2_readdir, 1110 .unlocked_ioctl = gfs2_ioctl, 1111 .open = gfs2_open, 1112 .release = gfs2_close, 1113 .fsync = gfs2_fsync, 1114 .llseek = default_llseek, 1115 }; 1116 1117