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 * @datasync: set if we can ignore timestamp changes 549 * 550 * The VFS will flush data for us. We only need to worry 551 * about metadata here. 552 * 553 * Returns: errno 554 */ 555 556 static int gfs2_fsync(struct file *file, int datasync) 557 { 558 struct inode *inode = file->f_mapping->host; 559 int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC); 560 struct gfs2_inode *ip = GFS2_I(inode); 561 int ret; 562 563 if (datasync) 564 sync_state &= ~I_DIRTY_SYNC; 565 566 if (sync_state) { 567 ret = sync_inode_metadata(inode, 1); 568 if (ret) 569 return ret; 570 gfs2_ail_flush(ip->i_gl); 571 } 572 573 return 0; 574 } 575 576 /** 577 * gfs2_file_aio_write - Perform a write to a file 578 * @iocb: The io context 579 * @iov: The data to write 580 * @nr_segs: Number of @iov segments 581 * @pos: The file position 582 * 583 * We have to do a lock/unlock here to refresh the inode size for 584 * O_APPEND writes, otherwise we can land up writing at the wrong 585 * offset. There is still a race, but provided the app is using its 586 * own file locking, this will make O_APPEND work as expected. 587 * 588 */ 589 590 static ssize_t gfs2_file_aio_write(struct kiocb *iocb, const struct iovec *iov, 591 unsigned long nr_segs, loff_t pos) 592 { 593 struct file *file = iocb->ki_filp; 594 595 if (file->f_flags & O_APPEND) { 596 struct dentry *dentry = file->f_dentry; 597 struct gfs2_inode *ip = GFS2_I(dentry->d_inode); 598 struct gfs2_holder gh; 599 int ret; 600 601 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh); 602 if (ret) 603 return ret; 604 gfs2_glock_dq_uninit(&gh); 605 } 606 607 return generic_file_aio_write(iocb, iov, nr_segs, pos); 608 } 609 610 static int empty_write_end(struct page *page, unsigned from, 611 unsigned to, int mode) 612 { 613 struct inode *inode = page->mapping->host; 614 struct gfs2_inode *ip = GFS2_I(inode); 615 struct buffer_head *bh; 616 unsigned offset, blksize = 1 << inode->i_blkbits; 617 pgoff_t end_index = i_size_read(inode) >> PAGE_CACHE_SHIFT; 618 619 zero_user(page, from, to-from); 620 mark_page_accessed(page); 621 622 if (page->index < end_index || !(mode & FALLOC_FL_KEEP_SIZE)) { 623 if (!gfs2_is_writeback(ip)) 624 gfs2_page_add_databufs(ip, page, from, to); 625 626 block_commit_write(page, from, to); 627 return 0; 628 } 629 630 offset = 0; 631 bh = page_buffers(page); 632 while (offset < to) { 633 if (offset >= from) { 634 set_buffer_uptodate(bh); 635 mark_buffer_dirty(bh); 636 clear_buffer_new(bh); 637 write_dirty_buffer(bh, WRITE); 638 } 639 offset += blksize; 640 bh = bh->b_this_page; 641 } 642 643 offset = 0; 644 bh = page_buffers(page); 645 while (offset < to) { 646 if (offset >= from) { 647 wait_on_buffer(bh); 648 if (!buffer_uptodate(bh)) 649 return -EIO; 650 } 651 offset += blksize; 652 bh = bh->b_this_page; 653 } 654 return 0; 655 } 656 657 static int needs_empty_write(sector_t block, struct inode *inode) 658 { 659 int error; 660 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 }; 661 662 bh_map.b_size = 1 << inode->i_blkbits; 663 error = gfs2_block_map(inode, block, &bh_map, 0); 664 if (unlikely(error)) 665 return error; 666 return !buffer_mapped(&bh_map); 667 } 668 669 static int write_empty_blocks(struct page *page, unsigned from, unsigned to, 670 int mode) 671 { 672 struct inode *inode = page->mapping->host; 673 unsigned start, end, next, blksize; 674 sector_t block = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 675 int ret; 676 677 blksize = 1 << inode->i_blkbits; 678 next = end = 0; 679 while (next < from) { 680 next += blksize; 681 block++; 682 } 683 start = next; 684 do { 685 next += blksize; 686 ret = needs_empty_write(block, inode); 687 if (unlikely(ret < 0)) 688 return ret; 689 if (ret == 0) { 690 if (end) { 691 ret = __block_write_begin(page, start, end - start, 692 gfs2_block_map); 693 if (unlikely(ret)) 694 return ret; 695 ret = empty_write_end(page, start, end, mode); 696 if (unlikely(ret)) 697 return ret; 698 end = 0; 699 } 700 start = next; 701 } 702 else 703 end = next; 704 block++; 705 } while (next < to); 706 707 if (end) { 708 ret = __block_write_begin(page, start, end - start, gfs2_block_map); 709 if (unlikely(ret)) 710 return ret; 711 ret = empty_write_end(page, start, end, mode); 712 if (unlikely(ret)) 713 return ret; 714 } 715 716 return 0; 717 } 718 719 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len, 720 int mode) 721 { 722 struct gfs2_inode *ip = GFS2_I(inode); 723 struct buffer_head *dibh; 724 int error; 725 u64 start = offset >> PAGE_CACHE_SHIFT; 726 unsigned int start_offset = offset & ~PAGE_CACHE_MASK; 727 u64 end = (offset + len - 1) >> PAGE_CACHE_SHIFT; 728 pgoff_t curr; 729 struct page *page; 730 unsigned int end_offset = (offset + len) & ~PAGE_CACHE_MASK; 731 unsigned int from, to; 732 733 if (!end_offset) 734 end_offset = PAGE_CACHE_SIZE; 735 736 error = gfs2_meta_inode_buffer(ip, &dibh); 737 if (unlikely(error)) 738 goto out; 739 740 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 741 742 if (gfs2_is_stuffed(ip)) { 743 error = gfs2_unstuff_dinode(ip, NULL); 744 if (unlikely(error)) 745 goto out; 746 } 747 748 curr = start; 749 offset = start << PAGE_CACHE_SHIFT; 750 from = start_offset; 751 to = PAGE_CACHE_SIZE; 752 while (curr <= end) { 753 page = grab_cache_page_write_begin(inode->i_mapping, curr, 754 AOP_FLAG_NOFS); 755 if (unlikely(!page)) { 756 error = -ENOMEM; 757 goto out; 758 } 759 760 if (curr == end) 761 to = end_offset; 762 error = write_empty_blocks(page, from, to, mode); 763 if (!error && offset + to > inode->i_size && 764 !(mode & FALLOC_FL_KEEP_SIZE)) { 765 i_size_write(inode, offset + to); 766 } 767 unlock_page(page); 768 page_cache_release(page); 769 if (error) 770 goto out; 771 curr++; 772 offset += PAGE_CACHE_SIZE; 773 from = 0; 774 } 775 776 gfs2_dinode_out(ip, dibh->b_data); 777 mark_inode_dirty(inode); 778 779 brelse(dibh); 780 781 out: 782 return error; 783 } 784 785 static void calc_max_reserv(struct gfs2_inode *ip, loff_t max, loff_t *len, 786 unsigned int *data_blocks, unsigned int *ind_blocks) 787 { 788 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 789 unsigned int max_blocks = ip->i_alloc->al_rgd->rd_free_clone; 790 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1); 791 792 for (tmp = max_data; tmp > sdp->sd_diptrs;) { 793 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs); 794 max_data -= tmp; 795 } 796 /* This calculation isn't the exact reverse of gfs2_write_calc_reserve, 797 so it might end up with fewer data blocks */ 798 if (max_data <= *data_blocks) 799 return; 800 *data_blocks = max_data; 801 *ind_blocks = max_blocks - max_data; 802 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift; 803 if (*len > max) { 804 *len = max; 805 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks); 806 } 807 } 808 809 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, 810 loff_t len) 811 { 812 struct inode *inode = file->f_path.dentry->d_inode; 813 struct gfs2_sbd *sdp = GFS2_SB(inode); 814 struct gfs2_inode *ip = GFS2_I(inode); 815 unsigned int data_blocks = 0, ind_blocks = 0, rblocks; 816 loff_t bytes, max_bytes; 817 struct gfs2_alloc *al; 818 int error; 819 loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1); 820 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift; 821 next = (next + 1) << sdp->sd_sb.sb_bsize_shift; 822 823 /* We only support the FALLOC_FL_KEEP_SIZE mode */ 824 if (mode & ~FALLOC_FL_KEEP_SIZE) 825 return -EOPNOTSUPP; 826 827 offset &= bsize_mask; 828 829 len = next - offset; 830 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2; 831 if (!bytes) 832 bytes = UINT_MAX; 833 bytes &= bsize_mask; 834 if (bytes == 0) 835 bytes = sdp->sd_sb.sb_bsize; 836 837 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh); 838 error = gfs2_glock_nq(&ip->i_gh); 839 if (unlikely(error)) 840 goto out_uninit; 841 842 if (!gfs2_write_alloc_required(ip, offset, len)) 843 goto out_unlock; 844 845 while (len > 0) { 846 if (len < bytes) 847 bytes = len; 848 al = gfs2_alloc_get(ip); 849 if (!al) { 850 error = -ENOMEM; 851 goto out_unlock; 852 } 853 854 error = gfs2_quota_lock_check(ip); 855 if (error) 856 goto out_alloc_put; 857 858 retry: 859 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks); 860 861 al->al_requested = data_blocks + ind_blocks; 862 error = gfs2_inplace_reserve(ip); 863 if (error) { 864 if (error == -ENOSPC && bytes > sdp->sd_sb.sb_bsize) { 865 bytes >>= 1; 866 bytes &= bsize_mask; 867 if (bytes == 0) 868 bytes = sdp->sd_sb.sb_bsize; 869 goto retry; 870 } 871 goto out_qunlock; 872 } 873 max_bytes = bytes; 874 calc_max_reserv(ip, len, &max_bytes, &data_blocks, &ind_blocks); 875 al->al_requested = data_blocks + ind_blocks; 876 877 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA + 878 RES_RG_HDR + gfs2_rg_blocks(al); 879 if (gfs2_is_jdata(ip)) 880 rblocks += data_blocks ? data_blocks : 1; 881 882 error = gfs2_trans_begin(sdp, rblocks, 883 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize); 884 if (error) 885 goto out_trans_fail; 886 887 error = fallocate_chunk(inode, offset, max_bytes, mode); 888 gfs2_trans_end(sdp); 889 890 if (error) 891 goto out_trans_fail; 892 893 len -= max_bytes; 894 offset += max_bytes; 895 gfs2_inplace_release(ip); 896 gfs2_quota_unlock(ip); 897 gfs2_alloc_put(ip); 898 } 899 goto out_unlock; 900 901 out_trans_fail: 902 gfs2_inplace_release(ip); 903 out_qunlock: 904 gfs2_quota_unlock(ip); 905 out_alloc_put: 906 gfs2_alloc_put(ip); 907 out_unlock: 908 gfs2_glock_dq(&ip->i_gh); 909 out_uninit: 910 gfs2_holder_uninit(&ip->i_gh); 911 return error; 912 } 913 914 #ifdef CONFIG_GFS2_FS_LOCKING_DLM 915 916 /** 917 * gfs2_setlease - acquire/release a file lease 918 * @file: the file pointer 919 * @arg: lease type 920 * @fl: file lock 921 * 922 * We don't currently have a way to enforce a lease across the whole 923 * cluster; until we do, disable leases (by just returning -EINVAL), 924 * unless the administrator has requested purely local locking. 925 * 926 * Locking: called under lock_flocks 927 * 928 * Returns: errno 929 */ 930 931 static int gfs2_setlease(struct file *file, long arg, struct file_lock **fl) 932 { 933 return -EINVAL; 934 } 935 936 /** 937 * gfs2_lock - acquire/release a posix lock on a file 938 * @file: the file pointer 939 * @cmd: either modify or retrieve lock state, possibly wait 940 * @fl: type and range of lock 941 * 942 * Returns: errno 943 */ 944 945 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl) 946 { 947 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 948 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host); 949 struct lm_lockstruct *ls = &sdp->sd_lockstruct; 950 951 if (!(fl->fl_flags & FL_POSIX)) 952 return -ENOLCK; 953 if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK) 954 return -ENOLCK; 955 956 if (cmd == F_CANCELLK) { 957 /* Hack: */ 958 cmd = F_SETLK; 959 fl->fl_type = F_UNLCK; 960 } 961 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 962 return -EIO; 963 if (IS_GETLK(cmd)) 964 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl); 965 else if (fl->fl_type == F_UNLCK) 966 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl); 967 else 968 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl); 969 } 970 971 static int do_flock(struct file *file, int cmd, struct file_lock *fl) 972 { 973 struct gfs2_file *fp = file->private_data; 974 struct gfs2_holder *fl_gh = &fp->f_fl_gh; 975 struct gfs2_inode *ip = GFS2_I(file->f_path.dentry->d_inode); 976 struct gfs2_glock *gl; 977 unsigned int state; 978 int flags; 979 int error = 0; 980 981 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED; 982 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE; 983 984 mutex_lock(&fp->f_fl_mutex); 985 986 gl = fl_gh->gh_gl; 987 if (gl) { 988 if (fl_gh->gh_state == state) 989 goto out; 990 flock_lock_file_wait(file, 991 &(struct file_lock){.fl_type = F_UNLCK}); 992 gfs2_glock_dq_wait(fl_gh); 993 gfs2_holder_reinit(state, flags, fl_gh); 994 } else { 995 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr, 996 &gfs2_flock_glops, CREATE, &gl); 997 if (error) 998 goto out; 999 gfs2_holder_init(gl, state, flags, fl_gh); 1000 gfs2_glock_put(gl); 1001 } 1002 error = gfs2_glock_nq(fl_gh); 1003 if (error) { 1004 gfs2_holder_uninit(fl_gh); 1005 if (error == GLR_TRYFAILED) 1006 error = -EAGAIN; 1007 } else { 1008 error = flock_lock_file_wait(file, fl); 1009 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); 1010 } 1011 1012 out: 1013 mutex_unlock(&fp->f_fl_mutex); 1014 return error; 1015 } 1016 1017 static void do_unflock(struct file *file, struct file_lock *fl) 1018 { 1019 struct gfs2_file *fp = file->private_data; 1020 struct gfs2_holder *fl_gh = &fp->f_fl_gh; 1021 1022 mutex_lock(&fp->f_fl_mutex); 1023 flock_lock_file_wait(file, fl); 1024 if (fl_gh->gh_gl) { 1025 gfs2_glock_dq_wait(fl_gh); 1026 gfs2_holder_uninit(fl_gh); 1027 } 1028 mutex_unlock(&fp->f_fl_mutex); 1029 } 1030 1031 /** 1032 * gfs2_flock - acquire/release a flock lock on a file 1033 * @file: the file pointer 1034 * @cmd: either modify or retrieve lock state, possibly wait 1035 * @fl: type and range of lock 1036 * 1037 * Returns: errno 1038 */ 1039 1040 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl) 1041 { 1042 if (!(fl->fl_flags & FL_FLOCK)) 1043 return -ENOLCK; 1044 if (fl->fl_type & LOCK_MAND) 1045 return -EOPNOTSUPP; 1046 1047 if (fl->fl_type == F_UNLCK) { 1048 do_unflock(file, fl); 1049 return 0; 1050 } else { 1051 return do_flock(file, cmd, fl); 1052 } 1053 } 1054 1055 const struct file_operations gfs2_file_fops = { 1056 .llseek = gfs2_llseek, 1057 .read = do_sync_read, 1058 .aio_read = generic_file_aio_read, 1059 .write = do_sync_write, 1060 .aio_write = gfs2_file_aio_write, 1061 .unlocked_ioctl = gfs2_ioctl, 1062 .mmap = gfs2_mmap, 1063 .open = gfs2_open, 1064 .release = gfs2_close, 1065 .fsync = gfs2_fsync, 1066 .lock = gfs2_lock, 1067 .flock = gfs2_flock, 1068 .splice_read = generic_file_splice_read, 1069 .splice_write = generic_file_splice_write, 1070 .setlease = gfs2_setlease, 1071 .fallocate = gfs2_fallocate, 1072 }; 1073 1074 const struct file_operations gfs2_dir_fops = { 1075 .readdir = gfs2_readdir, 1076 .unlocked_ioctl = gfs2_ioctl, 1077 .open = gfs2_open, 1078 .release = gfs2_close, 1079 .fsync = gfs2_fsync, 1080 .lock = gfs2_lock, 1081 .flock = gfs2_flock, 1082 .llseek = default_llseek, 1083 }; 1084 1085 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */ 1086 1087 const struct file_operations gfs2_file_fops_nolock = { 1088 .llseek = gfs2_llseek, 1089 .read = do_sync_read, 1090 .aio_read = generic_file_aio_read, 1091 .write = do_sync_write, 1092 .aio_write = gfs2_file_aio_write, 1093 .unlocked_ioctl = gfs2_ioctl, 1094 .mmap = gfs2_mmap, 1095 .open = gfs2_open, 1096 .release = gfs2_close, 1097 .fsync = gfs2_fsync, 1098 .splice_read = generic_file_splice_read, 1099 .splice_write = generic_file_splice_write, 1100 .setlease = generic_setlease, 1101 .fallocate = gfs2_fallocate, 1102 }; 1103 1104 const struct file_operations gfs2_dir_fops_nolock = { 1105 .readdir = gfs2_readdir, 1106 .unlocked_ioctl = gfs2_ioctl, 1107 .open = gfs2_open, 1108 .release = gfs2_close, 1109 .fsync = gfs2_fsync, 1110 .llseek = default_llseek, 1111 }; 1112 1113