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 (!is_owner_or_cap(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 struct gfs2_holder i_gh; 453 int error; 454 455 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh); 456 error = gfs2_glock_nq(&i_gh); 457 file_accessed(file); 458 if (error == 0) 459 gfs2_glock_dq_uninit(&i_gh); 460 } 461 vma->vm_ops = &gfs2_vm_ops; 462 vma->vm_flags |= VM_CAN_NONLINEAR; 463 464 return 0; 465 } 466 467 /** 468 * gfs2_open - open a file 469 * @inode: the inode to open 470 * @file: the struct file for this opening 471 * 472 * Returns: errno 473 */ 474 475 static int gfs2_open(struct inode *inode, struct file *file) 476 { 477 struct gfs2_inode *ip = GFS2_I(inode); 478 struct gfs2_holder i_gh; 479 struct gfs2_file *fp; 480 int error; 481 482 fp = kzalloc(sizeof(struct gfs2_file), GFP_KERNEL); 483 if (!fp) 484 return -ENOMEM; 485 486 mutex_init(&fp->f_fl_mutex); 487 488 gfs2_assert_warn(GFS2_SB(inode), !file->private_data); 489 file->private_data = fp; 490 491 if (S_ISREG(ip->i_inode.i_mode)) { 492 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, 493 &i_gh); 494 if (error) 495 goto fail; 496 497 if (!(file->f_flags & O_LARGEFILE) && 498 i_size_read(inode) > MAX_NON_LFS) { 499 error = -EOVERFLOW; 500 goto fail_gunlock; 501 } 502 503 gfs2_glock_dq_uninit(&i_gh); 504 } 505 506 return 0; 507 508 fail_gunlock: 509 gfs2_glock_dq_uninit(&i_gh); 510 fail: 511 file->private_data = NULL; 512 kfree(fp); 513 return error; 514 } 515 516 /** 517 * gfs2_close - called to close a struct file 518 * @inode: the inode the struct file belongs to 519 * @file: the struct file being closed 520 * 521 * Returns: errno 522 */ 523 524 static int gfs2_close(struct inode *inode, struct file *file) 525 { 526 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info; 527 struct gfs2_file *fp; 528 529 fp = file->private_data; 530 file->private_data = NULL; 531 532 if (gfs2_assert_warn(sdp, fp)) 533 return -EIO; 534 535 kfree(fp); 536 537 return 0; 538 } 539 540 /** 541 * gfs2_fsync - sync the dirty data for a file (across the cluster) 542 * @file: the file that points to the dentry (we ignore this) 543 * @dentry: the dentry that points to the inode to sync 544 * 545 * The VFS will flush "normal" data for us. We only need to worry 546 * about metadata here. For journaled data, we just do a log flush 547 * as we can't avoid it. Otherwise we can just bale out if datasync 548 * is set. For stuffed inodes we must flush the log in order to 549 * ensure that all data is on disk. 550 * 551 * The call to write_inode_now() is there to write back metadata and 552 * the inode itself. It does also try and write the data, but thats 553 * (hopefully) a no-op due to the VFS having already called filemap_fdatawrite() 554 * for us. 555 * 556 * Returns: errno 557 */ 558 559 static int gfs2_fsync(struct file *file, int datasync) 560 { 561 struct inode *inode = file->f_mapping->host; 562 int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC); 563 int ret = 0; 564 565 if (gfs2_is_jdata(GFS2_I(inode))) { 566 gfs2_log_flush(GFS2_SB(inode), GFS2_I(inode)->i_gl); 567 return 0; 568 } 569 570 if (sync_state != 0) { 571 if (!datasync) 572 ret = write_inode_now(inode, 0); 573 574 if (gfs2_is_stuffed(GFS2_I(inode))) 575 gfs2_log_flush(GFS2_SB(inode), GFS2_I(inode)->i_gl); 576 } 577 578 return ret; 579 } 580 581 /** 582 * gfs2_file_aio_write - Perform a write to a file 583 * @iocb: The io context 584 * @iov: The data to write 585 * @nr_segs: Number of @iov segments 586 * @pos: The file position 587 * 588 * We have to do a lock/unlock here to refresh the inode size for 589 * O_APPEND writes, otherwise we can land up writing at the wrong 590 * offset. There is still a race, but provided the app is using its 591 * own file locking, this will make O_APPEND work as expected. 592 * 593 */ 594 595 static ssize_t gfs2_file_aio_write(struct kiocb *iocb, const struct iovec *iov, 596 unsigned long nr_segs, loff_t pos) 597 { 598 struct file *file = iocb->ki_filp; 599 600 if (file->f_flags & O_APPEND) { 601 struct dentry *dentry = file->f_dentry; 602 struct gfs2_inode *ip = GFS2_I(dentry->d_inode); 603 struct gfs2_holder gh; 604 int ret; 605 606 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh); 607 if (ret) 608 return ret; 609 gfs2_glock_dq_uninit(&gh); 610 } 611 612 return generic_file_aio_write(iocb, iov, nr_segs, pos); 613 } 614 615 static void empty_write_end(struct page *page, unsigned from, 616 unsigned to) 617 { 618 struct gfs2_inode *ip = GFS2_I(page->mapping->host); 619 620 page_zero_new_buffers(page, from, to); 621 flush_dcache_page(page); 622 mark_page_accessed(page); 623 624 if (!gfs2_is_writeback(ip)) 625 gfs2_page_add_databufs(ip, page, from, to); 626 627 block_commit_write(page, from, to); 628 } 629 630 static int write_empty_blocks(struct page *page, unsigned from, unsigned to) 631 { 632 unsigned start, end, next; 633 struct buffer_head *bh, *head; 634 int error; 635 636 if (!page_has_buffers(page)) { 637 error = __block_write_begin(page, from, to - from, gfs2_block_map); 638 if (unlikely(error)) 639 return error; 640 641 empty_write_end(page, from, to); 642 return 0; 643 } 644 645 bh = head = page_buffers(page); 646 next = end = 0; 647 while (next < from) { 648 next += bh->b_size; 649 bh = bh->b_this_page; 650 } 651 start = next; 652 do { 653 next += bh->b_size; 654 if (buffer_mapped(bh)) { 655 if (end) { 656 error = __block_write_begin(page, start, end - start, 657 gfs2_block_map); 658 if (unlikely(error)) 659 return error; 660 empty_write_end(page, start, end); 661 end = 0; 662 } 663 start = next; 664 } 665 else 666 end = next; 667 bh = bh->b_this_page; 668 } while (next < to); 669 670 if (end) { 671 error = __block_write_begin(page, start, end - start, gfs2_block_map); 672 if (unlikely(error)) 673 return error; 674 empty_write_end(page, start, end); 675 } 676 677 return 0; 678 } 679 680 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len, 681 int mode) 682 { 683 struct gfs2_inode *ip = GFS2_I(inode); 684 struct buffer_head *dibh; 685 int error; 686 u64 start = offset >> PAGE_CACHE_SHIFT; 687 unsigned int start_offset = offset & ~PAGE_CACHE_MASK; 688 u64 end = (offset + len - 1) >> PAGE_CACHE_SHIFT; 689 pgoff_t curr; 690 struct page *page; 691 unsigned int end_offset = (offset + len) & ~PAGE_CACHE_MASK; 692 unsigned int from, to; 693 694 if (!end_offset) 695 end_offset = PAGE_CACHE_SIZE; 696 697 error = gfs2_meta_inode_buffer(ip, &dibh); 698 if (unlikely(error)) 699 goto out; 700 701 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 702 703 if (gfs2_is_stuffed(ip)) { 704 error = gfs2_unstuff_dinode(ip, NULL); 705 if (unlikely(error)) 706 goto out; 707 } 708 709 curr = start; 710 offset = start << PAGE_CACHE_SHIFT; 711 from = start_offset; 712 to = PAGE_CACHE_SIZE; 713 while (curr <= end) { 714 page = grab_cache_page_write_begin(inode->i_mapping, curr, 715 AOP_FLAG_NOFS); 716 if (unlikely(!page)) { 717 error = -ENOMEM; 718 goto out; 719 } 720 721 if (curr == end) 722 to = end_offset; 723 error = write_empty_blocks(page, from, to); 724 if (!error && offset + to > inode->i_size && 725 !(mode & FALLOC_FL_KEEP_SIZE)) { 726 i_size_write(inode, offset + to); 727 } 728 unlock_page(page); 729 page_cache_release(page); 730 if (error) 731 goto out; 732 curr++; 733 offset += PAGE_CACHE_SIZE; 734 from = 0; 735 } 736 737 gfs2_dinode_out(ip, dibh->b_data); 738 mark_inode_dirty(inode); 739 740 brelse(dibh); 741 742 out: 743 return error; 744 } 745 746 static void calc_max_reserv(struct gfs2_inode *ip, loff_t max, loff_t *len, 747 unsigned int *data_blocks, unsigned int *ind_blocks) 748 { 749 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 750 unsigned int max_blocks = ip->i_alloc->al_rgd->rd_free_clone; 751 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1); 752 753 for (tmp = max_data; tmp > sdp->sd_diptrs;) { 754 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs); 755 max_data -= tmp; 756 } 757 /* This calculation isn't the exact reverse of gfs2_write_calc_reserve, 758 so it might end up with fewer data blocks */ 759 if (max_data <= *data_blocks) 760 return; 761 *data_blocks = max_data; 762 *ind_blocks = max_blocks - max_data; 763 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift; 764 if (*len > max) { 765 *len = max; 766 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks); 767 } 768 } 769 770 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, 771 loff_t len) 772 { 773 struct inode *inode = file->f_path.dentry->d_inode; 774 struct gfs2_sbd *sdp = GFS2_SB(inode); 775 struct gfs2_inode *ip = GFS2_I(inode); 776 unsigned int data_blocks = 0, ind_blocks = 0, rblocks; 777 loff_t bytes, max_bytes; 778 struct gfs2_alloc *al; 779 int error; 780 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift; 781 next = (next + 1) << sdp->sd_sb.sb_bsize_shift; 782 783 /* We only support the FALLOC_FL_KEEP_SIZE mode */ 784 if (mode & ~FALLOC_FL_KEEP_SIZE) 785 return -EOPNOTSUPP; 786 787 offset = (offset >> sdp->sd_sb.sb_bsize_shift) << 788 sdp->sd_sb.sb_bsize_shift; 789 790 len = next - offset; 791 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2; 792 if (!bytes) 793 bytes = UINT_MAX; 794 795 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh); 796 error = gfs2_glock_nq(&ip->i_gh); 797 if (unlikely(error)) 798 goto out_uninit; 799 800 if (!gfs2_write_alloc_required(ip, offset, len)) 801 goto out_unlock; 802 803 while (len > 0) { 804 if (len < bytes) 805 bytes = len; 806 al = gfs2_alloc_get(ip); 807 if (!al) { 808 error = -ENOMEM; 809 goto out_unlock; 810 } 811 812 error = gfs2_quota_lock_check(ip); 813 if (error) 814 goto out_alloc_put; 815 816 retry: 817 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks); 818 819 al->al_requested = data_blocks + ind_blocks; 820 error = gfs2_inplace_reserve(ip); 821 if (error) { 822 if (error == -ENOSPC && bytes > sdp->sd_sb.sb_bsize) { 823 bytes >>= 1; 824 goto retry; 825 } 826 goto out_qunlock; 827 } 828 max_bytes = bytes; 829 calc_max_reserv(ip, len, &max_bytes, &data_blocks, &ind_blocks); 830 al->al_requested = data_blocks + ind_blocks; 831 832 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA + 833 RES_RG_HDR + gfs2_rg_blocks(al); 834 if (gfs2_is_jdata(ip)) 835 rblocks += data_blocks ? data_blocks : 1; 836 837 error = gfs2_trans_begin(sdp, rblocks, 838 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize); 839 if (error) 840 goto out_trans_fail; 841 842 error = fallocate_chunk(inode, offset, max_bytes, mode); 843 gfs2_trans_end(sdp); 844 845 if (error) 846 goto out_trans_fail; 847 848 len -= max_bytes; 849 offset += max_bytes; 850 gfs2_inplace_release(ip); 851 gfs2_quota_unlock(ip); 852 gfs2_alloc_put(ip); 853 } 854 goto out_unlock; 855 856 out_trans_fail: 857 gfs2_inplace_release(ip); 858 out_qunlock: 859 gfs2_quota_unlock(ip); 860 out_alloc_put: 861 gfs2_alloc_put(ip); 862 out_unlock: 863 gfs2_glock_dq(&ip->i_gh); 864 out_uninit: 865 gfs2_holder_uninit(&ip->i_gh); 866 return error; 867 } 868 869 #ifdef CONFIG_GFS2_FS_LOCKING_DLM 870 871 /** 872 * gfs2_setlease - acquire/release a file lease 873 * @file: the file pointer 874 * @arg: lease type 875 * @fl: file lock 876 * 877 * We don't currently have a way to enforce a lease across the whole 878 * cluster; until we do, disable leases (by just returning -EINVAL), 879 * unless the administrator has requested purely local locking. 880 * 881 * Locking: called under lock_flocks 882 * 883 * Returns: errno 884 */ 885 886 static int gfs2_setlease(struct file *file, long arg, struct file_lock **fl) 887 { 888 return -EINVAL; 889 } 890 891 /** 892 * gfs2_lock - acquire/release a posix lock on a file 893 * @file: the file pointer 894 * @cmd: either modify or retrieve lock state, possibly wait 895 * @fl: type and range of lock 896 * 897 * Returns: errno 898 */ 899 900 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl) 901 { 902 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 903 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host); 904 struct lm_lockstruct *ls = &sdp->sd_lockstruct; 905 906 if (!(fl->fl_flags & FL_POSIX)) 907 return -ENOLCK; 908 if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK) 909 return -ENOLCK; 910 911 if (cmd == F_CANCELLK) { 912 /* Hack: */ 913 cmd = F_SETLK; 914 fl->fl_type = F_UNLCK; 915 } 916 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 917 return -EIO; 918 if (IS_GETLK(cmd)) 919 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl); 920 else if (fl->fl_type == F_UNLCK) 921 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl); 922 else 923 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl); 924 } 925 926 static int do_flock(struct file *file, int cmd, struct file_lock *fl) 927 { 928 struct gfs2_file *fp = file->private_data; 929 struct gfs2_holder *fl_gh = &fp->f_fl_gh; 930 struct gfs2_inode *ip = GFS2_I(file->f_path.dentry->d_inode); 931 struct gfs2_glock *gl; 932 unsigned int state; 933 int flags; 934 int error = 0; 935 936 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED; 937 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE; 938 939 mutex_lock(&fp->f_fl_mutex); 940 941 gl = fl_gh->gh_gl; 942 if (gl) { 943 if (fl_gh->gh_state == state) 944 goto out; 945 flock_lock_file_wait(file, 946 &(struct file_lock){.fl_type = F_UNLCK}); 947 gfs2_glock_dq_wait(fl_gh); 948 gfs2_holder_reinit(state, flags, fl_gh); 949 } else { 950 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr, 951 &gfs2_flock_glops, CREATE, &gl); 952 if (error) 953 goto out; 954 gfs2_holder_init(gl, state, flags, fl_gh); 955 gfs2_glock_put(gl); 956 } 957 error = gfs2_glock_nq(fl_gh); 958 if (error) { 959 gfs2_holder_uninit(fl_gh); 960 if (error == GLR_TRYFAILED) 961 error = -EAGAIN; 962 } else { 963 error = flock_lock_file_wait(file, fl); 964 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); 965 } 966 967 out: 968 mutex_unlock(&fp->f_fl_mutex); 969 return error; 970 } 971 972 static void do_unflock(struct file *file, struct file_lock *fl) 973 { 974 struct gfs2_file *fp = file->private_data; 975 struct gfs2_holder *fl_gh = &fp->f_fl_gh; 976 977 mutex_lock(&fp->f_fl_mutex); 978 flock_lock_file_wait(file, fl); 979 if (fl_gh->gh_gl) 980 gfs2_glock_dq_uninit(fl_gh); 981 mutex_unlock(&fp->f_fl_mutex); 982 } 983 984 /** 985 * gfs2_flock - acquire/release a flock lock on a file 986 * @file: the file pointer 987 * @cmd: either modify or retrieve lock state, possibly wait 988 * @fl: type and range of lock 989 * 990 * Returns: errno 991 */ 992 993 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl) 994 { 995 if (!(fl->fl_flags & FL_FLOCK)) 996 return -ENOLCK; 997 if (fl->fl_type & LOCK_MAND) 998 return -EOPNOTSUPP; 999 1000 if (fl->fl_type == F_UNLCK) { 1001 do_unflock(file, fl); 1002 return 0; 1003 } else { 1004 return do_flock(file, cmd, fl); 1005 } 1006 } 1007 1008 const struct file_operations gfs2_file_fops = { 1009 .llseek = gfs2_llseek, 1010 .read = do_sync_read, 1011 .aio_read = generic_file_aio_read, 1012 .write = do_sync_write, 1013 .aio_write = gfs2_file_aio_write, 1014 .unlocked_ioctl = gfs2_ioctl, 1015 .mmap = gfs2_mmap, 1016 .open = gfs2_open, 1017 .release = gfs2_close, 1018 .fsync = gfs2_fsync, 1019 .lock = gfs2_lock, 1020 .flock = gfs2_flock, 1021 .splice_read = generic_file_splice_read, 1022 .splice_write = generic_file_splice_write, 1023 .setlease = gfs2_setlease, 1024 .fallocate = gfs2_fallocate, 1025 }; 1026 1027 const struct file_operations gfs2_dir_fops = { 1028 .readdir = gfs2_readdir, 1029 .unlocked_ioctl = gfs2_ioctl, 1030 .open = gfs2_open, 1031 .release = gfs2_close, 1032 .fsync = gfs2_fsync, 1033 .lock = gfs2_lock, 1034 .flock = gfs2_flock, 1035 .llseek = default_llseek, 1036 }; 1037 1038 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */ 1039 1040 const struct file_operations gfs2_file_fops_nolock = { 1041 .llseek = gfs2_llseek, 1042 .read = do_sync_read, 1043 .aio_read = generic_file_aio_read, 1044 .write = do_sync_write, 1045 .aio_write = gfs2_file_aio_write, 1046 .unlocked_ioctl = gfs2_ioctl, 1047 .mmap = gfs2_mmap, 1048 .open = gfs2_open, 1049 .release = gfs2_close, 1050 .fsync = gfs2_fsync, 1051 .splice_read = generic_file_splice_read, 1052 .splice_write = generic_file_splice_write, 1053 .setlease = generic_setlease, 1054 .fallocate = gfs2_fallocate, 1055 }; 1056 1057 const struct file_operations gfs2_dir_fops_nolock = { 1058 .readdir = gfs2_readdir, 1059 .unlocked_ioctl = gfs2_ioctl, 1060 .open = gfs2_open, 1061 .release = gfs2_close, 1062 .fsync = gfs2_fsync, 1063 .llseek = default_llseek, 1064 }; 1065 1066