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/falloc.h> 22 #include <linux/swap.h> 23 #include <linux/crc32.h> 24 #include <linux/writeback.h> 25 #include <asm/uaccess.h> 26 #include <linux/dlm.h> 27 #include <linux/dlm_plock.h> 28 29 #include "gfs2.h" 30 #include "incore.h" 31 #include "bmap.h" 32 #include "dir.h" 33 #include "glock.h" 34 #include "glops.h" 35 #include "inode.h" 36 #include "log.h" 37 #include "meta_io.h" 38 #include "quota.h" 39 #include "rgrp.h" 40 #include "trans.h" 41 #include "util.h" 42 43 /** 44 * gfs2_llseek - seek to a location in a file 45 * @file: the file 46 * @offset: the offset 47 * @origin: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END) 48 * 49 * SEEK_END requires the glock for the file because it references the 50 * file's size. 51 * 52 * Returns: The new offset, or errno 53 */ 54 55 static loff_t gfs2_llseek(struct file *file, loff_t offset, int origin) 56 { 57 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 58 struct gfs2_holder i_gh; 59 loff_t error; 60 61 switch (origin) { 62 case SEEK_END: /* These reference inode->i_size */ 63 case SEEK_DATA: 64 case SEEK_HOLE: 65 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, 66 &i_gh); 67 if (!error) { 68 error = generic_file_llseek(file, offset, origin); 69 gfs2_glock_dq_uninit(&i_gh); 70 } 71 break; 72 case SEEK_CUR: 73 case SEEK_SET: 74 error = generic_file_llseek(file, offset, origin); 75 break; 76 default: 77 error = -EINVAL; 78 } 79 80 return error; 81 } 82 83 /** 84 * gfs2_readdir - Read directory entries from a directory 85 * @file: The directory to read from 86 * @dirent: Buffer for dirents 87 * @filldir: Function used to do the copying 88 * 89 * Returns: errno 90 */ 91 92 static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir) 93 { 94 struct inode *dir = file->f_mapping->host; 95 struct gfs2_inode *dip = GFS2_I(dir); 96 struct gfs2_holder d_gh; 97 u64 offset = file->f_pos; 98 int error; 99 100 gfs2_holder_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh); 101 error = gfs2_glock_nq(&d_gh); 102 if (error) { 103 gfs2_holder_uninit(&d_gh); 104 return error; 105 } 106 107 error = gfs2_dir_read(dir, &offset, dirent, filldir, &file->f_ra); 108 109 gfs2_glock_dq_uninit(&d_gh); 110 111 file->f_pos = offset; 112 113 return error; 114 } 115 116 /** 117 * fsflags_cvt 118 * @table: A table of 32 u32 flags 119 * @val: a 32 bit value to convert 120 * 121 * This function can be used to convert between fsflags values and 122 * GFS2's own flags values. 123 * 124 * Returns: the converted flags 125 */ 126 static u32 fsflags_cvt(const u32 *table, u32 val) 127 { 128 u32 res = 0; 129 while(val) { 130 if (val & 1) 131 res |= *table; 132 table++; 133 val >>= 1; 134 } 135 return res; 136 } 137 138 static const u32 fsflags_to_gfs2[32] = { 139 [3] = GFS2_DIF_SYNC, 140 [4] = GFS2_DIF_IMMUTABLE, 141 [5] = GFS2_DIF_APPENDONLY, 142 [7] = GFS2_DIF_NOATIME, 143 [12] = GFS2_DIF_EXHASH, 144 [14] = GFS2_DIF_INHERIT_JDATA, 145 [17] = GFS2_DIF_TOPDIR, 146 }; 147 148 static const u32 gfs2_to_fsflags[32] = { 149 [gfs2fl_Sync] = FS_SYNC_FL, 150 [gfs2fl_Immutable] = FS_IMMUTABLE_FL, 151 [gfs2fl_AppendOnly] = FS_APPEND_FL, 152 [gfs2fl_NoAtime] = FS_NOATIME_FL, 153 [gfs2fl_ExHash] = FS_INDEX_FL, 154 [gfs2fl_TopLevel] = FS_TOPDIR_FL, 155 [gfs2fl_InheritJdata] = FS_JOURNAL_DATA_FL, 156 }; 157 158 static int gfs2_get_flags(struct file *filp, u32 __user *ptr) 159 { 160 struct inode *inode = filp->f_path.dentry->d_inode; 161 struct gfs2_inode *ip = GFS2_I(inode); 162 struct gfs2_holder gh; 163 int error; 164 u32 fsflags; 165 166 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); 167 error = gfs2_glock_nq(&gh); 168 if (error) 169 return error; 170 171 fsflags = fsflags_cvt(gfs2_to_fsflags, ip->i_diskflags); 172 if (!S_ISDIR(inode->i_mode) && ip->i_diskflags & GFS2_DIF_JDATA) 173 fsflags |= FS_JOURNAL_DATA_FL; 174 if (put_user(fsflags, ptr)) 175 error = -EFAULT; 176 177 gfs2_glock_dq(&gh); 178 gfs2_holder_uninit(&gh); 179 return error; 180 } 181 182 void gfs2_set_inode_flags(struct inode *inode) 183 { 184 struct gfs2_inode *ip = GFS2_I(inode); 185 unsigned int flags = inode->i_flags; 186 187 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC); 188 if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode)) 189 inode->i_flags |= S_NOSEC; 190 if (ip->i_diskflags & GFS2_DIF_IMMUTABLE) 191 flags |= S_IMMUTABLE; 192 if (ip->i_diskflags & GFS2_DIF_APPENDONLY) 193 flags |= S_APPEND; 194 if (ip->i_diskflags & GFS2_DIF_NOATIME) 195 flags |= S_NOATIME; 196 if (ip->i_diskflags & GFS2_DIF_SYNC) 197 flags |= S_SYNC; 198 inode->i_flags = flags; 199 } 200 201 /* Flags that can be set by user space */ 202 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \ 203 GFS2_DIF_IMMUTABLE| \ 204 GFS2_DIF_APPENDONLY| \ 205 GFS2_DIF_NOATIME| \ 206 GFS2_DIF_SYNC| \ 207 GFS2_DIF_SYSTEM| \ 208 GFS2_DIF_TOPDIR| \ 209 GFS2_DIF_INHERIT_JDATA) 210 211 /** 212 * gfs2_set_flags - set flags on an inode 213 * @inode: The inode 214 * @flags: The flags to set 215 * @mask: Indicates which flags are valid 216 * 217 */ 218 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask) 219 { 220 struct inode *inode = filp->f_path.dentry->d_inode; 221 struct gfs2_inode *ip = GFS2_I(inode); 222 struct gfs2_sbd *sdp = GFS2_SB(inode); 223 struct buffer_head *bh; 224 struct gfs2_holder gh; 225 int error; 226 u32 new_flags, flags; 227 228 error = mnt_want_write_file(filp); 229 if (error) 230 return error; 231 232 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 233 if (error) 234 goto out_drop_write; 235 236 error = -EACCES; 237 if (!inode_owner_or_capable(inode)) 238 goto out; 239 240 error = 0; 241 flags = ip->i_diskflags; 242 new_flags = (flags & ~mask) | (reqflags & mask); 243 if ((new_flags ^ flags) == 0) 244 goto out; 245 246 error = -EINVAL; 247 if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET) 248 goto out; 249 250 error = -EPERM; 251 if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE)) 252 goto out; 253 if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY)) 254 goto out; 255 if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) && 256 !capable(CAP_LINUX_IMMUTABLE)) 257 goto out; 258 if (!IS_IMMUTABLE(inode)) { 259 error = gfs2_permission(inode, MAY_WRITE); 260 if (error) 261 goto out; 262 } 263 if ((flags ^ new_flags) & GFS2_DIF_JDATA) { 264 if (flags & GFS2_DIF_JDATA) 265 gfs2_log_flush(sdp, ip->i_gl); 266 error = filemap_fdatawrite(inode->i_mapping); 267 if (error) 268 goto out; 269 error = filemap_fdatawait(inode->i_mapping); 270 if (error) 271 goto out; 272 } 273 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 274 if (error) 275 goto out; 276 error = gfs2_meta_inode_buffer(ip, &bh); 277 if (error) 278 goto out_trans_end; 279 gfs2_trans_add_bh(ip->i_gl, bh, 1); 280 ip->i_diskflags = new_flags; 281 gfs2_dinode_out(ip, bh->b_data); 282 brelse(bh); 283 gfs2_set_inode_flags(inode); 284 gfs2_set_aops(inode); 285 out_trans_end: 286 gfs2_trans_end(sdp); 287 out: 288 gfs2_glock_dq_uninit(&gh); 289 out_drop_write: 290 mnt_drop_write_file(filp); 291 return error; 292 } 293 294 static int gfs2_set_flags(struct file *filp, u32 __user *ptr) 295 { 296 struct inode *inode = filp->f_path.dentry->d_inode; 297 u32 fsflags, gfsflags; 298 299 if (get_user(fsflags, ptr)) 300 return -EFAULT; 301 302 gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags); 303 if (!S_ISDIR(inode->i_mode)) { 304 gfsflags &= ~GFS2_DIF_TOPDIR; 305 if (gfsflags & GFS2_DIF_INHERIT_JDATA) 306 gfsflags ^= (GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA); 307 return do_gfs2_set_flags(filp, gfsflags, ~0); 308 } 309 return do_gfs2_set_flags(filp, gfsflags, ~GFS2_DIF_JDATA); 310 } 311 312 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) 313 { 314 switch(cmd) { 315 case FS_IOC_GETFLAGS: 316 return gfs2_get_flags(filp, (u32 __user *)arg); 317 case FS_IOC_SETFLAGS: 318 return gfs2_set_flags(filp, (u32 __user *)arg); 319 case FITRIM: 320 return gfs2_fitrim(filp, (void __user *)arg); 321 } 322 return -ENOTTY; 323 } 324 325 /** 326 * gfs2_allocate_page_backing - Use bmap to allocate blocks 327 * @page: The (locked) page to allocate backing for 328 * 329 * We try to allocate all the blocks required for the page in 330 * one go. This might fail for various reasons, so we keep 331 * trying until all the blocks to back this page are allocated. 332 * If some of the blocks are already allocated, thats ok too. 333 */ 334 335 static int gfs2_allocate_page_backing(struct page *page) 336 { 337 struct inode *inode = page->mapping->host; 338 struct buffer_head bh; 339 unsigned long size = PAGE_CACHE_SIZE; 340 u64 lblock = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 341 342 do { 343 bh.b_state = 0; 344 bh.b_size = size; 345 gfs2_block_map(inode, lblock, &bh, 1); 346 if (!buffer_mapped(&bh)) 347 return -EIO; 348 size -= bh.b_size; 349 lblock += (bh.b_size >> inode->i_blkbits); 350 } while(size > 0); 351 return 0; 352 } 353 354 /** 355 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable 356 * @vma: The virtual memory area 357 * @page: The page which is about to become writable 358 * 359 * When the page becomes writable, we need to ensure that we have 360 * blocks allocated on disk to back that page. 361 */ 362 363 static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) 364 { 365 struct page *page = vmf->page; 366 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 367 struct gfs2_inode *ip = GFS2_I(inode); 368 struct gfs2_sbd *sdp = GFS2_SB(inode); 369 unsigned long last_index; 370 u64 pos = page->index << PAGE_CACHE_SHIFT; 371 unsigned int data_blocks, ind_blocks, rblocks; 372 struct gfs2_holder gh; 373 loff_t size; 374 int ret; 375 376 /* Wait if fs is frozen. This is racy so we check again later on 377 * and retry if the fs has been frozen after the page lock has 378 * been acquired 379 */ 380 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); 381 382 ret = gfs2_rs_alloc(ip); 383 if (ret) 384 return ret; 385 386 atomic_set(&ip->i_res->rs_sizehint, 387 PAGE_CACHE_SIZE >> sdp->sd_sb.sb_bsize_shift); 388 389 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 390 ret = gfs2_glock_nq(&gh); 391 if (ret) 392 goto out; 393 394 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags); 395 set_bit(GIF_SW_PAGED, &ip->i_flags); 396 397 if (!gfs2_write_alloc_required(ip, pos, PAGE_CACHE_SIZE)) { 398 lock_page(page); 399 if (!PageUptodate(page) || page->mapping != inode->i_mapping) { 400 ret = -EAGAIN; 401 unlock_page(page); 402 } 403 goto out_unlock; 404 } 405 406 ret = gfs2_rindex_update(sdp); 407 if (ret) 408 goto out_unlock; 409 410 ret = gfs2_quota_lock_check(ip); 411 if (ret) 412 goto out_unlock; 413 gfs2_write_calc_reserv(ip, PAGE_CACHE_SIZE, &data_blocks, &ind_blocks); 414 ret = gfs2_inplace_reserve(ip, data_blocks + ind_blocks); 415 if (ret) 416 goto out_quota_unlock; 417 418 rblocks = RES_DINODE + ind_blocks; 419 if (gfs2_is_jdata(ip)) 420 rblocks += data_blocks ? data_blocks : 1; 421 if (ind_blocks || data_blocks) { 422 rblocks += RES_STATFS + RES_QUOTA; 423 rblocks += gfs2_rg_blocks(ip); 424 } 425 ret = gfs2_trans_begin(sdp, rblocks, 0); 426 if (ret) 427 goto out_trans_fail; 428 429 lock_page(page); 430 ret = -EINVAL; 431 size = i_size_read(inode); 432 last_index = (size - 1) >> PAGE_CACHE_SHIFT; 433 /* Check page index against inode size */ 434 if (size == 0 || (page->index > last_index)) 435 goto out_trans_end; 436 437 ret = -EAGAIN; 438 /* If truncated, we must retry the operation, we may have raced 439 * with the glock demotion code. 440 */ 441 if (!PageUptodate(page) || page->mapping != inode->i_mapping) 442 goto out_trans_end; 443 444 /* Unstuff, if required, and allocate backing blocks for page */ 445 ret = 0; 446 if (gfs2_is_stuffed(ip)) 447 ret = gfs2_unstuff_dinode(ip, page); 448 if (ret == 0) 449 ret = gfs2_allocate_page_backing(page); 450 451 out_trans_end: 452 if (ret) 453 unlock_page(page); 454 gfs2_trans_end(sdp); 455 out_trans_fail: 456 gfs2_inplace_release(ip); 457 out_quota_unlock: 458 gfs2_quota_unlock(ip); 459 out_unlock: 460 gfs2_glock_dq(&gh); 461 out: 462 gfs2_holder_uninit(&gh); 463 if (ret == 0) { 464 set_page_dirty(page); 465 /* This check must be post dropping of transaction lock */ 466 if (inode->i_sb->s_frozen == SB_UNFROZEN) { 467 wait_on_page_writeback(page); 468 } else { 469 ret = -EAGAIN; 470 unlock_page(page); 471 } 472 } 473 return block_page_mkwrite_return(ret); 474 } 475 476 static const struct vm_operations_struct gfs2_vm_ops = { 477 .fault = filemap_fault, 478 .page_mkwrite = gfs2_page_mkwrite, 479 }; 480 481 /** 482 * gfs2_mmap - 483 * @file: The file to map 484 * @vma: The VMA which described the mapping 485 * 486 * There is no need to get a lock here unless we should be updating 487 * atime. We ignore any locking errors since the only consequence is 488 * a missed atime update (which will just be deferred until later). 489 * 490 * Returns: 0 491 */ 492 493 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma) 494 { 495 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 496 497 if (!(file->f_flags & O_NOATIME) && 498 !IS_NOATIME(&ip->i_inode)) { 499 struct gfs2_holder i_gh; 500 int error; 501 502 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh); 503 error = gfs2_glock_nq(&i_gh); 504 if (error == 0) { 505 file_accessed(file); 506 gfs2_glock_dq(&i_gh); 507 } 508 gfs2_holder_uninit(&i_gh); 509 if (error) 510 return error; 511 } 512 vma->vm_ops = &gfs2_vm_ops; 513 vma->vm_flags |= VM_CAN_NONLINEAR; 514 515 return 0; 516 } 517 518 /** 519 * gfs2_open - open a file 520 * @inode: the inode to open 521 * @file: the struct file for this opening 522 * 523 * Returns: errno 524 */ 525 526 static int gfs2_open(struct inode *inode, struct file *file) 527 { 528 struct gfs2_inode *ip = GFS2_I(inode); 529 struct gfs2_holder i_gh; 530 struct gfs2_file *fp; 531 int error; 532 533 fp = kzalloc(sizeof(struct gfs2_file), GFP_KERNEL); 534 if (!fp) 535 return -ENOMEM; 536 537 mutex_init(&fp->f_fl_mutex); 538 539 gfs2_assert_warn(GFS2_SB(inode), !file->private_data); 540 file->private_data = fp; 541 542 if (S_ISREG(ip->i_inode.i_mode)) { 543 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, 544 &i_gh); 545 if (error) 546 goto fail; 547 548 if (!(file->f_flags & O_LARGEFILE) && 549 i_size_read(inode) > MAX_NON_LFS) { 550 error = -EOVERFLOW; 551 goto fail_gunlock; 552 } 553 554 gfs2_glock_dq_uninit(&i_gh); 555 } 556 557 return 0; 558 559 fail_gunlock: 560 gfs2_glock_dq_uninit(&i_gh); 561 fail: 562 file->private_data = NULL; 563 kfree(fp); 564 return error; 565 } 566 567 /** 568 * gfs2_release - called to close a struct file 569 * @inode: the inode the struct file belongs to 570 * @file: the struct file being closed 571 * 572 * Returns: errno 573 */ 574 575 static int gfs2_release(struct inode *inode, struct file *file) 576 { 577 struct gfs2_inode *ip = GFS2_I(inode); 578 579 kfree(file->private_data); 580 file->private_data = NULL; 581 582 if ((file->f_mode & FMODE_WRITE) && 583 (atomic_read(&inode->i_writecount) == 1)) 584 gfs2_rs_delete(ip); 585 586 return 0; 587 } 588 589 /** 590 * gfs2_fsync - sync the dirty data for a file (across the cluster) 591 * @file: the file that points to the dentry 592 * @start: the start position in the file to sync 593 * @end: the end position in the file to sync 594 * @datasync: set if we can ignore timestamp changes 595 * 596 * We split the data flushing here so that we don't wait for the data 597 * until after we've also sent the metadata to disk. Note that for 598 * data=ordered, we will write & wait for the data at the log flush 599 * stage anyway, so this is unlikely to make much of a difference 600 * except in the data=writeback case. 601 * 602 * If the fdatawrite fails due to any reason except -EIO, we will 603 * continue the remainder of the fsync, although we'll still report 604 * the error at the end. This is to match filemap_write_and_wait_range() 605 * behaviour. 606 * 607 * Returns: errno 608 */ 609 610 static int gfs2_fsync(struct file *file, loff_t start, loff_t end, 611 int datasync) 612 { 613 struct address_space *mapping = file->f_mapping; 614 struct inode *inode = mapping->host; 615 int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC); 616 struct gfs2_inode *ip = GFS2_I(inode); 617 int ret = 0, ret1 = 0; 618 619 if (mapping->nrpages) { 620 ret1 = filemap_fdatawrite_range(mapping, start, end); 621 if (ret1 == -EIO) 622 return ret1; 623 } 624 625 if (datasync) 626 sync_state &= ~I_DIRTY_SYNC; 627 628 if (sync_state) { 629 ret = sync_inode_metadata(inode, 1); 630 if (ret) 631 return ret; 632 if (gfs2_is_jdata(ip)) 633 filemap_write_and_wait(mapping); 634 gfs2_ail_flush(ip->i_gl, 1); 635 } 636 637 if (mapping->nrpages) 638 ret = filemap_fdatawait_range(mapping, start, end); 639 640 return ret ? ret : ret1; 641 } 642 643 /** 644 * gfs2_file_aio_write - Perform a write to a file 645 * @iocb: The io context 646 * @iov: The data to write 647 * @nr_segs: Number of @iov segments 648 * @pos: The file position 649 * 650 * We have to do a lock/unlock here to refresh the inode size for 651 * O_APPEND writes, otherwise we can land up writing at the wrong 652 * offset. There is still a race, but provided the app is using its 653 * own file locking, this will make O_APPEND work as expected. 654 * 655 */ 656 657 static ssize_t gfs2_file_aio_write(struct kiocb *iocb, const struct iovec *iov, 658 unsigned long nr_segs, loff_t pos) 659 { 660 struct file *file = iocb->ki_filp; 661 size_t writesize = iov_length(iov, nr_segs); 662 struct dentry *dentry = file->f_dentry; 663 struct gfs2_inode *ip = GFS2_I(dentry->d_inode); 664 struct gfs2_sbd *sdp; 665 int ret; 666 667 sdp = GFS2_SB(file->f_mapping->host); 668 ret = gfs2_rs_alloc(ip); 669 if (ret) 670 return ret; 671 672 atomic_set(&ip->i_res->rs_sizehint, writesize >> sdp->sd_sb.sb_bsize_shift); 673 if (file->f_flags & O_APPEND) { 674 struct gfs2_holder gh; 675 676 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh); 677 if (ret) 678 return ret; 679 gfs2_glock_dq_uninit(&gh); 680 } 681 682 return generic_file_aio_write(iocb, iov, nr_segs, pos); 683 } 684 685 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len, 686 int mode) 687 { 688 struct gfs2_inode *ip = GFS2_I(inode); 689 struct buffer_head *dibh; 690 int error; 691 loff_t size = len; 692 unsigned int nr_blks; 693 sector_t lblock = offset >> inode->i_blkbits; 694 695 error = gfs2_meta_inode_buffer(ip, &dibh); 696 if (unlikely(error)) 697 return error; 698 699 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 700 701 if (gfs2_is_stuffed(ip)) { 702 error = gfs2_unstuff_dinode(ip, NULL); 703 if (unlikely(error)) 704 goto out; 705 } 706 707 while (len) { 708 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 }; 709 bh_map.b_size = len; 710 set_buffer_zeronew(&bh_map); 711 712 error = gfs2_block_map(inode, lblock, &bh_map, 1); 713 if (unlikely(error)) 714 goto out; 715 len -= bh_map.b_size; 716 nr_blks = bh_map.b_size >> inode->i_blkbits; 717 lblock += nr_blks; 718 if (!buffer_new(&bh_map)) 719 continue; 720 if (unlikely(!buffer_zeronew(&bh_map))) { 721 error = -EIO; 722 goto out; 723 } 724 } 725 if (offset + size > inode->i_size && !(mode & FALLOC_FL_KEEP_SIZE)) 726 i_size_write(inode, offset + size); 727 728 mark_inode_dirty(inode); 729 730 out: 731 brelse(dibh); 732 return error; 733 } 734 735 static void calc_max_reserv(struct gfs2_inode *ip, loff_t max, loff_t *len, 736 unsigned int *data_blocks, unsigned int *ind_blocks) 737 { 738 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 739 unsigned int max_blocks = ip->i_rgd->rd_free_clone; 740 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1); 741 742 for (tmp = max_data; tmp > sdp->sd_diptrs;) { 743 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs); 744 max_data -= tmp; 745 } 746 /* This calculation isn't the exact reverse of gfs2_write_calc_reserve, 747 so it might end up with fewer data blocks */ 748 if (max_data <= *data_blocks) 749 return; 750 *data_blocks = max_data; 751 *ind_blocks = max_blocks - max_data; 752 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift; 753 if (*len > max) { 754 *len = max; 755 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks); 756 } 757 } 758 759 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, 760 loff_t len) 761 { 762 struct inode *inode = file->f_path.dentry->d_inode; 763 struct gfs2_sbd *sdp = GFS2_SB(inode); 764 struct gfs2_inode *ip = GFS2_I(inode); 765 unsigned int data_blocks = 0, ind_blocks = 0, rblocks; 766 loff_t bytes, max_bytes; 767 int error; 768 const loff_t pos = offset; 769 const loff_t count = len; 770 loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1); 771 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift; 772 loff_t max_chunk_size = UINT_MAX & bsize_mask; 773 next = (next + 1) << sdp->sd_sb.sb_bsize_shift; 774 775 /* We only support the FALLOC_FL_KEEP_SIZE mode */ 776 if (mode & ~FALLOC_FL_KEEP_SIZE) 777 return -EOPNOTSUPP; 778 779 offset &= bsize_mask; 780 781 len = next - offset; 782 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2; 783 if (!bytes) 784 bytes = UINT_MAX; 785 bytes &= bsize_mask; 786 if (bytes == 0) 787 bytes = sdp->sd_sb.sb_bsize; 788 789 error = gfs2_rs_alloc(ip); 790 if (error) 791 return error; 792 793 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh); 794 error = gfs2_glock_nq(&ip->i_gh); 795 if (unlikely(error)) 796 goto out_uninit; 797 798 atomic_set(&ip->i_res->rs_sizehint, len >> sdp->sd_sb.sb_bsize_shift); 799 800 while (len > 0) { 801 if (len < bytes) 802 bytes = len; 803 if (!gfs2_write_alloc_required(ip, offset, bytes)) { 804 len -= bytes; 805 offset += bytes; 806 continue; 807 } 808 error = gfs2_quota_lock_check(ip); 809 if (error) 810 goto out_unlock; 811 812 retry: 813 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks); 814 815 error = gfs2_inplace_reserve(ip, data_blocks + ind_blocks); 816 if (error) { 817 if (error == -ENOSPC && bytes > sdp->sd_sb.sb_bsize) { 818 bytes >>= 1; 819 bytes &= bsize_mask; 820 if (bytes == 0) 821 bytes = sdp->sd_sb.sb_bsize; 822 goto retry; 823 } 824 goto out_qunlock; 825 } 826 max_bytes = bytes; 827 calc_max_reserv(ip, (len > max_chunk_size)? max_chunk_size: len, 828 &max_bytes, &data_blocks, &ind_blocks); 829 830 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA + 831 RES_RG_HDR + gfs2_rg_blocks(ip); 832 if (gfs2_is_jdata(ip)) 833 rblocks += data_blocks ? data_blocks : 1; 834 835 error = gfs2_trans_begin(sdp, rblocks, 836 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize); 837 if (error) 838 goto out_trans_fail; 839 840 error = fallocate_chunk(inode, offset, max_bytes, mode); 841 gfs2_trans_end(sdp); 842 843 if (error) 844 goto out_trans_fail; 845 846 len -= max_bytes; 847 offset += max_bytes; 848 gfs2_inplace_release(ip); 849 gfs2_quota_unlock(ip); 850 } 851 852 if (error == 0) 853 error = generic_write_sync(file, pos, count); 854 goto out_unlock; 855 856 out_trans_fail: 857 gfs2_inplace_release(ip); 858 out_qunlock: 859 gfs2_quota_unlock(ip); 860 out_unlock: 861 gfs2_glock_dq(&ip->i_gh); 862 out_uninit: 863 gfs2_holder_uninit(&ip->i_gh); 864 return error; 865 } 866 867 #ifdef CONFIG_GFS2_FS_LOCKING_DLM 868 869 /** 870 * gfs2_setlease - acquire/release a file lease 871 * @file: the file pointer 872 * @arg: lease type 873 * @fl: file lock 874 * 875 * We don't currently have a way to enforce a lease across the whole 876 * cluster; until we do, disable leases (by just returning -EINVAL), 877 * unless the administrator has requested purely local locking. 878 * 879 * Locking: called under lock_flocks 880 * 881 * Returns: errno 882 */ 883 884 static int gfs2_setlease(struct file *file, long arg, struct file_lock **fl) 885 { 886 return -EINVAL; 887 } 888 889 /** 890 * gfs2_lock - acquire/release a posix lock on a file 891 * @file: the file pointer 892 * @cmd: either modify or retrieve lock state, possibly wait 893 * @fl: type and range of lock 894 * 895 * Returns: errno 896 */ 897 898 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl) 899 { 900 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 901 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host); 902 struct lm_lockstruct *ls = &sdp->sd_lockstruct; 903 904 if (!(fl->fl_flags & FL_POSIX)) 905 return -ENOLCK; 906 if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK) 907 return -ENOLCK; 908 909 if (cmd == F_CANCELLK) { 910 /* Hack: */ 911 cmd = F_SETLK; 912 fl->fl_type = F_UNLCK; 913 } 914 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 915 return -EIO; 916 if (IS_GETLK(cmd)) 917 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl); 918 else if (fl->fl_type == F_UNLCK) 919 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl); 920 else 921 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl); 922 } 923 924 static int do_flock(struct file *file, int cmd, struct file_lock *fl) 925 { 926 struct gfs2_file *fp = file->private_data; 927 struct gfs2_holder *fl_gh = &fp->f_fl_gh; 928 struct gfs2_inode *ip = GFS2_I(file->f_path.dentry->d_inode); 929 struct gfs2_glock *gl; 930 unsigned int state; 931 int flags; 932 int error = 0; 933 934 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED; 935 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE; 936 937 mutex_lock(&fp->f_fl_mutex); 938 939 gl = fl_gh->gh_gl; 940 if (gl) { 941 if (fl_gh->gh_state == state) 942 goto out; 943 flock_lock_file_wait(file, 944 &(struct file_lock){.fl_type = F_UNLCK}); 945 gfs2_glock_dq_wait(fl_gh); 946 gfs2_holder_reinit(state, flags, fl_gh); 947 } else { 948 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr, 949 &gfs2_flock_glops, CREATE, &gl); 950 if (error) 951 goto out; 952 gfs2_holder_init(gl, state, flags, fl_gh); 953 gfs2_glock_put(gl); 954 } 955 error = gfs2_glock_nq(fl_gh); 956 if (error) { 957 gfs2_holder_uninit(fl_gh); 958 if (error == GLR_TRYFAILED) 959 error = -EAGAIN; 960 } else { 961 error = flock_lock_file_wait(file, fl); 962 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); 963 } 964 965 out: 966 mutex_unlock(&fp->f_fl_mutex); 967 return error; 968 } 969 970 static void do_unflock(struct file *file, struct file_lock *fl) 971 { 972 struct gfs2_file *fp = file->private_data; 973 struct gfs2_holder *fl_gh = &fp->f_fl_gh; 974 975 mutex_lock(&fp->f_fl_mutex); 976 flock_lock_file_wait(file, fl); 977 if (fl_gh->gh_gl) { 978 gfs2_glock_dq_wait(fl_gh); 979 gfs2_holder_uninit(fl_gh); 980 } 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_release, 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_release, 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_release, 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_release, 1062 .fsync = gfs2_fsync, 1063 .llseek = default_llseek, 1064 }; 1065 1066