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 #include <linux/aio.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 * @whence: 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 whence) 57 { 58 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 59 struct gfs2_holder i_gh; 60 loff_t error; 61 62 switch (whence) { 63 case SEEK_END: /* These reference inode->i_size */ 64 case SEEK_DATA: 65 case SEEK_HOLE: 66 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, 67 &i_gh); 68 if (!error) { 69 error = generic_file_llseek(file, offset, whence); 70 gfs2_glock_dq_uninit(&i_gh); 71 } 72 break; 73 case SEEK_CUR: 74 case SEEK_SET: 75 error = generic_file_llseek(file, offset, whence); 76 break; 77 default: 78 error = -EINVAL; 79 } 80 81 return error; 82 } 83 84 /** 85 * gfs2_readdir - Read directory entries from a directory 86 * @file: The directory to read from 87 * @dirent: Buffer for dirents 88 * @filldir: Function used to do the copying 89 * 90 * Returns: errno 91 */ 92 93 static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir) 94 { 95 struct inode *dir = file->f_mapping->host; 96 struct gfs2_inode *dip = GFS2_I(dir); 97 struct gfs2_holder d_gh; 98 u64 offset = file->f_pos; 99 int error; 100 101 gfs2_holder_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh); 102 error = gfs2_glock_nq(&d_gh); 103 if (error) { 104 gfs2_holder_uninit(&d_gh); 105 return error; 106 } 107 108 error = gfs2_dir_read(dir, &offset, dirent, filldir, &file->f_ra); 109 110 gfs2_glock_dq_uninit(&d_gh); 111 112 file->f_pos = offset; 113 114 return error; 115 } 116 117 /** 118 * fsflags_cvt 119 * @table: A table of 32 u32 flags 120 * @val: a 32 bit value to convert 121 * 122 * This function can be used to convert between fsflags values and 123 * GFS2's own flags values. 124 * 125 * Returns: the converted flags 126 */ 127 static u32 fsflags_cvt(const u32 *table, u32 val) 128 { 129 u32 res = 0; 130 while(val) { 131 if (val & 1) 132 res |= *table; 133 table++; 134 val >>= 1; 135 } 136 return res; 137 } 138 139 static const u32 fsflags_to_gfs2[32] = { 140 [3] = GFS2_DIF_SYNC, 141 [4] = GFS2_DIF_IMMUTABLE, 142 [5] = GFS2_DIF_APPENDONLY, 143 [7] = GFS2_DIF_NOATIME, 144 [12] = GFS2_DIF_EXHASH, 145 [14] = GFS2_DIF_INHERIT_JDATA, 146 [17] = GFS2_DIF_TOPDIR, 147 }; 148 149 static const u32 gfs2_to_fsflags[32] = { 150 [gfs2fl_Sync] = FS_SYNC_FL, 151 [gfs2fl_Immutable] = FS_IMMUTABLE_FL, 152 [gfs2fl_AppendOnly] = FS_APPEND_FL, 153 [gfs2fl_NoAtime] = FS_NOATIME_FL, 154 [gfs2fl_ExHash] = FS_INDEX_FL, 155 [gfs2fl_TopLevel] = FS_TOPDIR_FL, 156 [gfs2fl_InheritJdata] = FS_JOURNAL_DATA_FL, 157 }; 158 159 static int gfs2_get_flags(struct file *filp, u32 __user *ptr) 160 { 161 struct inode *inode = file_inode(filp); 162 struct gfs2_inode *ip = GFS2_I(inode); 163 struct gfs2_holder gh; 164 int error; 165 u32 fsflags; 166 167 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); 168 error = gfs2_glock_nq(&gh); 169 if (error) 170 return error; 171 172 fsflags = fsflags_cvt(gfs2_to_fsflags, ip->i_diskflags); 173 if (!S_ISDIR(inode->i_mode) && ip->i_diskflags & GFS2_DIF_JDATA) 174 fsflags |= FS_JOURNAL_DATA_FL; 175 if (put_user(fsflags, ptr)) 176 error = -EFAULT; 177 178 gfs2_glock_dq(&gh); 179 gfs2_holder_uninit(&gh); 180 return error; 181 } 182 183 void gfs2_set_inode_flags(struct inode *inode) 184 { 185 struct gfs2_inode *ip = GFS2_I(inode); 186 unsigned int flags = inode->i_flags; 187 188 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC); 189 if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode)) 190 inode->i_flags |= S_NOSEC; 191 if (ip->i_diskflags & GFS2_DIF_IMMUTABLE) 192 flags |= S_IMMUTABLE; 193 if (ip->i_diskflags & GFS2_DIF_APPENDONLY) 194 flags |= S_APPEND; 195 if (ip->i_diskflags & GFS2_DIF_NOATIME) 196 flags |= S_NOATIME; 197 if (ip->i_diskflags & GFS2_DIF_SYNC) 198 flags |= S_SYNC; 199 inode->i_flags = flags; 200 } 201 202 /* Flags that can be set by user space */ 203 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \ 204 GFS2_DIF_IMMUTABLE| \ 205 GFS2_DIF_APPENDONLY| \ 206 GFS2_DIF_NOATIME| \ 207 GFS2_DIF_SYNC| \ 208 GFS2_DIF_SYSTEM| \ 209 GFS2_DIF_TOPDIR| \ 210 GFS2_DIF_INHERIT_JDATA) 211 212 /** 213 * gfs2_set_flags - set flags on an inode 214 * @inode: The inode 215 * @flags: The flags to set 216 * @mask: Indicates which flags are valid 217 * 218 */ 219 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask) 220 { 221 struct inode *inode = file_inode(filp); 222 struct gfs2_inode *ip = GFS2_I(inode); 223 struct gfs2_sbd *sdp = GFS2_SB(inode); 224 struct buffer_head *bh; 225 struct gfs2_holder gh; 226 int error; 227 u32 new_flags, flags; 228 229 error = mnt_want_write_file(filp); 230 if (error) 231 return error; 232 233 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 234 if (error) 235 goto out_drop_write; 236 237 error = -EACCES; 238 if (!inode_owner_or_capable(inode)) 239 goto out; 240 241 error = 0; 242 flags = ip->i_diskflags; 243 new_flags = (flags & ~mask) | (reqflags & mask); 244 if ((new_flags ^ flags) == 0) 245 goto out; 246 247 error = -EINVAL; 248 if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET) 249 goto out; 250 251 error = -EPERM; 252 if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE)) 253 goto out; 254 if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY)) 255 goto out; 256 if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) && 257 !capable(CAP_LINUX_IMMUTABLE)) 258 goto out; 259 if (!IS_IMMUTABLE(inode)) { 260 error = gfs2_permission(inode, MAY_WRITE); 261 if (error) 262 goto out; 263 } 264 if ((flags ^ new_flags) & GFS2_DIF_JDATA) { 265 if (flags & GFS2_DIF_JDATA) 266 gfs2_log_flush(sdp, ip->i_gl); 267 error = filemap_fdatawrite(inode->i_mapping); 268 if (error) 269 goto out; 270 error = filemap_fdatawait(inode->i_mapping); 271 if (error) 272 goto out; 273 } 274 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 275 if (error) 276 goto out; 277 error = gfs2_meta_inode_buffer(ip, &bh); 278 if (error) 279 goto out_trans_end; 280 gfs2_trans_add_meta(ip->i_gl, bh); 281 ip->i_diskflags = new_flags; 282 gfs2_dinode_out(ip, bh->b_data); 283 brelse(bh); 284 gfs2_set_inode_flags(inode); 285 gfs2_set_aops(inode); 286 out_trans_end: 287 gfs2_trans_end(sdp); 288 out: 289 gfs2_glock_dq_uninit(&gh); 290 out_drop_write: 291 mnt_drop_write_file(filp); 292 return error; 293 } 294 295 static int gfs2_set_flags(struct file *filp, u32 __user *ptr) 296 { 297 struct inode *inode = file_inode(filp); 298 u32 fsflags, gfsflags; 299 300 if (get_user(fsflags, ptr)) 301 return -EFAULT; 302 303 gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags); 304 if (!S_ISDIR(inode->i_mode)) { 305 gfsflags &= ~GFS2_DIF_TOPDIR; 306 if (gfsflags & GFS2_DIF_INHERIT_JDATA) 307 gfsflags ^= (GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA); 308 return do_gfs2_set_flags(filp, gfsflags, ~0); 309 } 310 return do_gfs2_set_flags(filp, gfsflags, ~GFS2_DIF_JDATA); 311 } 312 313 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) 314 { 315 switch(cmd) { 316 case FS_IOC_GETFLAGS: 317 return gfs2_get_flags(filp, (u32 __user *)arg); 318 case FS_IOC_SETFLAGS: 319 return gfs2_set_flags(filp, (u32 __user *)arg); 320 case FITRIM: 321 return gfs2_fitrim(filp, (void __user *)arg); 322 } 323 return -ENOTTY; 324 } 325 326 /** 327 * gfs2_size_hint - Give a hint to the size of a write request 328 * @file: The struct file 329 * @offset: The file offset of the write 330 * @size: The length of the write 331 * 332 * When we are about to do a write, this function records the total 333 * write size in order to provide a suitable hint to the lower layers 334 * about how many blocks will be required. 335 * 336 */ 337 338 static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size) 339 { 340 struct inode *inode = file_inode(filep); 341 struct gfs2_sbd *sdp = GFS2_SB(inode); 342 struct gfs2_inode *ip = GFS2_I(inode); 343 size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift; 344 int hint = min_t(size_t, INT_MAX, blks); 345 346 atomic_set(&ip->i_res->rs_sizehint, hint); 347 } 348 349 /** 350 * gfs2_allocate_page_backing - Use bmap to allocate blocks 351 * @page: The (locked) page to allocate backing for 352 * 353 * We try to allocate all the blocks required for the page in 354 * one go. This might fail for various reasons, so we keep 355 * trying until all the blocks to back this page are allocated. 356 * If some of the blocks are already allocated, thats ok too. 357 */ 358 359 static int gfs2_allocate_page_backing(struct page *page) 360 { 361 struct inode *inode = page->mapping->host; 362 struct buffer_head bh; 363 unsigned long size = PAGE_CACHE_SIZE; 364 u64 lblock = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 365 366 do { 367 bh.b_state = 0; 368 bh.b_size = size; 369 gfs2_block_map(inode, lblock, &bh, 1); 370 if (!buffer_mapped(&bh)) 371 return -EIO; 372 size -= bh.b_size; 373 lblock += (bh.b_size >> inode->i_blkbits); 374 } while(size > 0); 375 return 0; 376 } 377 378 /** 379 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable 380 * @vma: The virtual memory area 381 * @page: The page which is about to become writable 382 * 383 * When the page becomes writable, we need to ensure that we have 384 * blocks allocated on disk to back that page. 385 */ 386 387 static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) 388 { 389 struct page *page = vmf->page; 390 struct inode *inode = file_inode(vma->vm_file); 391 struct gfs2_inode *ip = GFS2_I(inode); 392 struct gfs2_sbd *sdp = GFS2_SB(inode); 393 unsigned long last_index; 394 u64 pos = page->index << PAGE_CACHE_SHIFT; 395 unsigned int data_blocks, ind_blocks, rblocks; 396 struct gfs2_holder gh; 397 loff_t size; 398 int ret; 399 400 sb_start_pagefault(inode->i_sb); 401 402 /* Update file times before taking page lock */ 403 file_update_time(vma->vm_file); 404 405 ret = gfs2_rs_alloc(ip); 406 if (ret) 407 return ret; 408 409 gfs2_size_hint(vma->vm_file, pos, PAGE_CACHE_SIZE); 410 411 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 412 ret = gfs2_glock_nq(&gh); 413 if (ret) 414 goto out; 415 416 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags); 417 set_bit(GIF_SW_PAGED, &ip->i_flags); 418 419 if (!gfs2_write_alloc_required(ip, pos, PAGE_CACHE_SIZE)) { 420 lock_page(page); 421 if (!PageUptodate(page) || page->mapping != inode->i_mapping) { 422 ret = -EAGAIN; 423 unlock_page(page); 424 } 425 goto out_unlock; 426 } 427 428 ret = gfs2_rindex_update(sdp); 429 if (ret) 430 goto out_unlock; 431 432 ret = gfs2_quota_lock_check(ip); 433 if (ret) 434 goto out_unlock; 435 gfs2_write_calc_reserv(ip, PAGE_CACHE_SIZE, &data_blocks, &ind_blocks); 436 ret = gfs2_inplace_reserve(ip, data_blocks + ind_blocks, 0); 437 if (ret) 438 goto out_quota_unlock; 439 440 rblocks = RES_DINODE + ind_blocks; 441 if (gfs2_is_jdata(ip)) 442 rblocks += data_blocks ? data_blocks : 1; 443 if (ind_blocks || data_blocks) { 444 rblocks += RES_STATFS + RES_QUOTA; 445 rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks); 446 } 447 ret = gfs2_trans_begin(sdp, rblocks, 0); 448 if (ret) 449 goto out_trans_fail; 450 451 lock_page(page); 452 ret = -EINVAL; 453 size = i_size_read(inode); 454 last_index = (size - 1) >> PAGE_CACHE_SHIFT; 455 /* Check page index against inode size */ 456 if (size == 0 || (page->index > last_index)) 457 goto out_trans_end; 458 459 ret = -EAGAIN; 460 /* If truncated, we must retry the operation, we may have raced 461 * with the glock demotion code. 462 */ 463 if (!PageUptodate(page) || page->mapping != inode->i_mapping) 464 goto out_trans_end; 465 466 /* Unstuff, if required, and allocate backing blocks for page */ 467 ret = 0; 468 if (gfs2_is_stuffed(ip)) 469 ret = gfs2_unstuff_dinode(ip, page); 470 if (ret == 0) 471 ret = gfs2_allocate_page_backing(page); 472 473 out_trans_end: 474 if (ret) 475 unlock_page(page); 476 gfs2_trans_end(sdp); 477 out_trans_fail: 478 gfs2_inplace_release(ip); 479 out_quota_unlock: 480 gfs2_quota_unlock(ip); 481 out_unlock: 482 gfs2_glock_dq(&gh); 483 out: 484 gfs2_holder_uninit(&gh); 485 if (ret == 0) { 486 set_page_dirty(page); 487 wait_for_stable_page(page); 488 } 489 sb_end_pagefault(inode->i_sb); 490 return block_page_mkwrite_return(ret); 491 } 492 493 static const struct vm_operations_struct gfs2_vm_ops = { 494 .fault = filemap_fault, 495 .page_mkwrite = gfs2_page_mkwrite, 496 .remap_pages = generic_file_remap_pages, 497 }; 498 499 /** 500 * gfs2_mmap - 501 * @file: The file to map 502 * @vma: The VMA which described the mapping 503 * 504 * There is no need to get a lock here unless we should be updating 505 * atime. We ignore any locking errors since the only consequence is 506 * a missed atime update (which will just be deferred until later). 507 * 508 * Returns: 0 509 */ 510 511 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma) 512 { 513 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 514 515 if (!(file->f_flags & O_NOATIME) && 516 !IS_NOATIME(&ip->i_inode)) { 517 struct gfs2_holder i_gh; 518 int error; 519 520 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, 521 &i_gh); 522 if (error) 523 return error; 524 /* grab lock to update inode */ 525 gfs2_glock_dq_uninit(&i_gh); 526 file_accessed(file); 527 } 528 vma->vm_ops = &gfs2_vm_ops; 529 530 return 0; 531 } 532 533 /** 534 * gfs2_open - open a file 535 * @inode: the inode to open 536 * @file: the struct file for this opening 537 * 538 * Returns: errno 539 */ 540 541 static int gfs2_open(struct inode *inode, struct file *file) 542 { 543 struct gfs2_inode *ip = GFS2_I(inode); 544 struct gfs2_holder i_gh; 545 struct gfs2_file *fp; 546 int error; 547 548 fp = kzalloc(sizeof(struct gfs2_file), GFP_KERNEL); 549 if (!fp) 550 return -ENOMEM; 551 552 mutex_init(&fp->f_fl_mutex); 553 554 gfs2_assert_warn(GFS2_SB(inode), !file->private_data); 555 file->private_data = fp; 556 557 if (S_ISREG(ip->i_inode.i_mode)) { 558 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, 559 &i_gh); 560 if (error) 561 goto fail; 562 563 if (!(file->f_flags & O_LARGEFILE) && 564 i_size_read(inode) > MAX_NON_LFS) { 565 error = -EOVERFLOW; 566 goto fail_gunlock; 567 } 568 569 gfs2_glock_dq_uninit(&i_gh); 570 } 571 572 return 0; 573 574 fail_gunlock: 575 gfs2_glock_dq_uninit(&i_gh); 576 fail: 577 file->private_data = NULL; 578 kfree(fp); 579 return error; 580 } 581 582 /** 583 * gfs2_release - called to close a struct file 584 * @inode: the inode the struct file belongs to 585 * @file: the struct file being closed 586 * 587 * Returns: errno 588 */ 589 590 static int gfs2_release(struct inode *inode, struct file *file) 591 { 592 struct gfs2_inode *ip = GFS2_I(inode); 593 594 kfree(file->private_data); 595 file->private_data = NULL; 596 597 if ((file->f_mode & FMODE_WRITE) && 598 (atomic_read(&inode->i_writecount) == 1)) 599 gfs2_rs_delete(ip); 600 601 return 0; 602 } 603 604 /** 605 * gfs2_fsync - sync the dirty data for a file (across the cluster) 606 * @file: the file that points to the dentry 607 * @start: the start position in the file to sync 608 * @end: the end position in the file to sync 609 * @datasync: set if we can ignore timestamp changes 610 * 611 * We split the data flushing here so that we don't wait for the data 612 * until after we've also sent the metadata to disk. Note that for 613 * data=ordered, we will write & wait for the data at the log flush 614 * stage anyway, so this is unlikely to make much of a difference 615 * except in the data=writeback case. 616 * 617 * If the fdatawrite fails due to any reason except -EIO, we will 618 * continue the remainder of the fsync, although we'll still report 619 * the error at the end. This is to match filemap_write_and_wait_range() 620 * behaviour. 621 * 622 * Returns: errno 623 */ 624 625 static int gfs2_fsync(struct file *file, loff_t start, loff_t end, 626 int datasync) 627 { 628 struct address_space *mapping = file->f_mapping; 629 struct inode *inode = mapping->host; 630 int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC); 631 struct gfs2_inode *ip = GFS2_I(inode); 632 int ret = 0, ret1 = 0; 633 634 if (mapping->nrpages) { 635 ret1 = filemap_fdatawrite_range(mapping, start, end); 636 if (ret1 == -EIO) 637 return ret1; 638 } 639 640 if (datasync) 641 sync_state &= ~I_DIRTY_SYNC; 642 643 if (sync_state) { 644 ret = sync_inode_metadata(inode, 1); 645 if (ret) 646 return ret; 647 if (gfs2_is_jdata(ip)) 648 filemap_write_and_wait(mapping); 649 gfs2_ail_flush(ip->i_gl, 1); 650 } 651 652 if (mapping->nrpages) 653 ret = filemap_fdatawait_range(mapping, start, end); 654 655 return ret ? ret : ret1; 656 } 657 658 /** 659 * gfs2_file_aio_write - Perform a write to a file 660 * @iocb: The io context 661 * @iov: The data to write 662 * @nr_segs: Number of @iov segments 663 * @pos: The file position 664 * 665 * We have to do a lock/unlock here to refresh the inode size for 666 * O_APPEND writes, otherwise we can land up writing at the wrong 667 * offset. There is still a race, but provided the app is using its 668 * own file locking, this will make O_APPEND work as expected. 669 * 670 */ 671 672 static ssize_t gfs2_file_aio_write(struct kiocb *iocb, const struct iovec *iov, 673 unsigned long nr_segs, loff_t pos) 674 { 675 struct file *file = iocb->ki_filp; 676 size_t writesize = iov_length(iov, nr_segs); 677 struct gfs2_inode *ip = GFS2_I(file_inode(file)); 678 int ret; 679 680 ret = gfs2_rs_alloc(ip); 681 if (ret) 682 return ret; 683 684 gfs2_size_hint(file, pos, writesize); 685 686 if (file->f_flags & O_APPEND) { 687 struct gfs2_holder gh; 688 689 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh); 690 if (ret) 691 return ret; 692 gfs2_glock_dq_uninit(&gh); 693 } 694 695 return generic_file_aio_write(iocb, iov, nr_segs, pos); 696 } 697 698 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len, 699 int mode) 700 { 701 struct gfs2_inode *ip = GFS2_I(inode); 702 struct buffer_head *dibh; 703 int error; 704 loff_t size = len; 705 unsigned int nr_blks; 706 sector_t lblock = offset >> inode->i_blkbits; 707 708 error = gfs2_meta_inode_buffer(ip, &dibh); 709 if (unlikely(error)) 710 return error; 711 712 gfs2_trans_add_meta(ip->i_gl, dibh); 713 714 if (gfs2_is_stuffed(ip)) { 715 error = gfs2_unstuff_dinode(ip, NULL); 716 if (unlikely(error)) 717 goto out; 718 } 719 720 while (len) { 721 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 }; 722 bh_map.b_size = len; 723 set_buffer_zeronew(&bh_map); 724 725 error = gfs2_block_map(inode, lblock, &bh_map, 1); 726 if (unlikely(error)) 727 goto out; 728 len -= bh_map.b_size; 729 nr_blks = bh_map.b_size >> inode->i_blkbits; 730 lblock += nr_blks; 731 if (!buffer_new(&bh_map)) 732 continue; 733 if (unlikely(!buffer_zeronew(&bh_map))) { 734 error = -EIO; 735 goto out; 736 } 737 } 738 if (offset + size > inode->i_size && !(mode & FALLOC_FL_KEEP_SIZE)) 739 i_size_write(inode, offset + size); 740 741 mark_inode_dirty(inode); 742 743 out: 744 brelse(dibh); 745 return error; 746 } 747 748 static void calc_max_reserv(struct gfs2_inode *ip, loff_t max, loff_t *len, 749 unsigned int *data_blocks, unsigned int *ind_blocks) 750 { 751 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 752 unsigned int max_blocks = ip->i_rgd->rd_free_clone; 753 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1); 754 755 for (tmp = max_data; tmp > sdp->sd_diptrs;) { 756 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs); 757 max_data -= tmp; 758 } 759 /* This calculation isn't the exact reverse of gfs2_write_calc_reserve, 760 so it might end up with fewer data blocks */ 761 if (max_data <= *data_blocks) 762 return; 763 *data_blocks = max_data; 764 *ind_blocks = max_blocks - max_data; 765 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift; 766 if (*len > max) { 767 *len = max; 768 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks); 769 } 770 } 771 772 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, 773 loff_t len) 774 { 775 struct inode *inode = file_inode(file); 776 struct gfs2_sbd *sdp = GFS2_SB(inode); 777 struct gfs2_inode *ip = GFS2_I(inode); 778 unsigned int data_blocks = 0, ind_blocks = 0, rblocks; 779 loff_t bytes, max_bytes; 780 int error; 781 const loff_t pos = offset; 782 const loff_t count = len; 783 loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1); 784 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift; 785 loff_t max_chunk_size = UINT_MAX & bsize_mask; 786 next = (next + 1) << sdp->sd_sb.sb_bsize_shift; 787 788 /* We only support the FALLOC_FL_KEEP_SIZE mode */ 789 if (mode & ~FALLOC_FL_KEEP_SIZE) 790 return -EOPNOTSUPP; 791 792 offset &= bsize_mask; 793 794 len = next - offset; 795 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2; 796 if (!bytes) 797 bytes = UINT_MAX; 798 bytes &= bsize_mask; 799 if (bytes == 0) 800 bytes = sdp->sd_sb.sb_bsize; 801 802 error = gfs2_rs_alloc(ip); 803 if (error) 804 return error; 805 806 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh); 807 error = gfs2_glock_nq(&ip->i_gh); 808 if (unlikely(error)) 809 goto out_uninit; 810 811 gfs2_size_hint(file, offset, len); 812 813 while (len > 0) { 814 if (len < bytes) 815 bytes = len; 816 if (!gfs2_write_alloc_required(ip, offset, bytes)) { 817 len -= bytes; 818 offset += bytes; 819 continue; 820 } 821 error = gfs2_quota_lock_check(ip); 822 if (error) 823 goto out_unlock; 824 825 retry: 826 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks); 827 828 error = gfs2_inplace_reserve(ip, data_blocks + ind_blocks, 0); 829 if (error) { 830 if (error == -ENOSPC && bytes > sdp->sd_sb.sb_bsize) { 831 bytes >>= 1; 832 bytes &= bsize_mask; 833 if (bytes == 0) 834 bytes = sdp->sd_sb.sb_bsize; 835 goto retry; 836 } 837 goto out_qunlock; 838 } 839 max_bytes = bytes; 840 calc_max_reserv(ip, (len > max_chunk_size)? max_chunk_size: len, 841 &max_bytes, &data_blocks, &ind_blocks); 842 843 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA + 844 RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks); 845 if (gfs2_is_jdata(ip)) 846 rblocks += data_blocks ? data_blocks : 1; 847 848 error = gfs2_trans_begin(sdp, rblocks, 849 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize); 850 if (error) 851 goto out_trans_fail; 852 853 error = fallocate_chunk(inode, offset, max_bytes, mode); 854 gfs2_trans_end(sdp); 855 856 if (error) 857 goto out_trans_fail; 858 859 len -= max_bytes; 860 offset += max_bytes; 861 gfs2_inplace_release(ip); 862 gfs2_quota_unlock(ip); 863 } 864 865 if (error == 0) 866 error = generic_write_sync(file, pos, count); 867 goto out_unlock; 868 869 out_trans_fail: 870 gfs2_inplace_release(ip); 871 out_qunlock: 872 gfs2_quota_unlock(ip); 873 out_unlock: 874 gfs2_glock_dq(&ip->i_gh); 875 out_uninit: 876 gfs2_holder_uninit(&ip->i_gh); 877 return error; 878 } 879 880 #ifdef CONFIG_GFS2_FS_LOCKING_DLM 881 882 /** 883 * gfs2_setlease - acquire/release a file lease 884 * @file: the file pointer 885 * @arg: lease type 886 * @fl: file lock 887 * 888 * We don't currently have a way to enforce a lease across the whole 889 * cluster; until we do, disable leases (by just returning -EINVAL), 890 * unless the administrator has requested purely local locking. 891 * 892 * Locking: called under lock_flocks 893 * 894 * Returns: errno 895 */ 896 897 static int gfs2_setlease(struct file *file, long arg, struct file_lock **fl) 898 { 899 return -EINVAL; 900 } 901 902 /** 903 * gfs2_lock - acquire/release a posix lock on a file 904 * @file: the file pointer 905 * @cmd: either modify or retrieve lock state, possibly wait 906 * @fl: type and range of lock 907 * 908 * Returns: errno 909 */ 910 911 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl) 912 { 913 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 914 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host); 915 struct lm_lockstruct *ls = &sdp->sd_lockstruct; 916 917 if (!(fl->fl_flags & FL_POSIX)) 918 return -ENOLCK; 919 if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK) 920 return -ENOLCK; 921 922 if (cmd == F_CANCELLK) { 923 /* Hack: */ 924 cmd = F_SETLK; 925 fl->fl_type = F_UNLCK; 926 } 927 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) { 928 if (fl->fl_type == F_UNLCK) 929 posix_lock_file_wait(file, fl); 930 return -EIO; 931 } 932 if (IS_GETLK(cmd)) 933 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl); 934 else if (fl->fl_type == F_UNLCK) 935 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl); 936 else 937 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl); 938 } 939 940 static int do_flock(struct file *file, int cmd, struct file_lock *fl) 941 { 942 struct gfs2_file *fp = file->private_data; 943 struct gfs2_holder *fl_gh = &fp->f_fl_gh; 944 struct gfs2_inode *ip = GFS2_I(file_inode(file)); 945 struct gfs2_glock *gl; 946 unsigned int state; 947 int flags; 948 int error = 0; 949 950 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED; 951 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE; 952 953 mutex_lock(&fp->f_fl_mutex); 954 955 gl = fl_gh->gh_gl; 956 if (gl) { 957 if (fl_gh->gh_state == state) 958 goto out; 959 flock_lock_file_wait(file, 960 &(struct file_lock){.fl_type = F_UNLCK}); 961 gfs2_glock_dq_wait(fl_gh); 962 gfs2_holder_reinit(state, flags, fl_gh); 963 } else { 964 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr, 965 &gfs2_flock_glops, CREATE, &gl); 966 if (error) 967 goto out; 968 gfs2_holder_init(gl, state, flags, fl_gh); 969 gfs2_glock_put(gl); 970 } 971 error = gfs2_glock_nq(fl_gh); 972 if (error) { 973 gfs2_holder_uninit(fl_gh); 974 if (error == GLR_TRYFAILED) 975 error = -EAGAIN; 976 } else { 977 error = flock_lock_file_wait(file, fl); 978 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); 979 } 980 981 out: 982 mutex_unlock(&fp->f_fl_mutex); 983 return error; 984 } 985 986 static void do_unflock(struct file *file, struct file_lock *fl) 987 { 988 struct gfs2_file *fp = file->private_data; 989 struct gfs2_holder *fl_gh = &fp->f_fl_gh; 990 991 mutex_lock(&fp->f_fl_mutex); 992 flock_lock_file_wait(file, fl); 993 if (fl_gh->gh_gl) { 994 gfs2_glock_dq_wait(fl_gh); 995 gfs2_holder_uninit(fl_gh); 996 } 997 mutex_unlock(&fp->f_fl_mutex); 998 } 999 1000 /** 1001 * gfs2_flock - acquire/release a flock lock on a file 1002 * @file: the file pointer 1003 * @cmd: either modify or retrieve lock state, possibly wait 1004 * @fl: type and range of lock 1005 * 1006 * Returns: errno 1007 */ 1008 1009 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl) 1010 { 1011 if (!(fl->fl_flags & FL_FLOCK)) 1012 return -ENOLCK; 1013 if (fl->fl_type & LOCK_MAND) 1014 return -EOPNOTSUPP; 1015 1016 if (fl->fl_type == F_UNLCK) { 1017 do_unflock(file, fl); 1018 return 0; 1019 } else { 1020 return do_flock(file, cmd, fl); 1021 } 1022 } 1023 1024 const struct file_operations gfs2_file_fops = { 1025 .llseek = gfs2_llseek, 1026 .read = do_sync_read, 1027 .aio_read = generic_file_aio_read, 1028 .write = do_sync_write, 1029 .aio_write = gfs2_file_aio_write, 1030 .unlocked_ioctl = gfs2_ioctl, 1031 .mmap = gfs2_mmap, 1032 .open = gfs2_open, 1033 .release = gfs2_release, 1034 .fsync = gfs2_fsync, 1035 .lock = gfs2_lock, 1036 .flock = gfs2_flock, 1037 .splice_read = generic_file_splice_read, 1038 .splice_write = generic_file_splice_write, 1039 .setlease = gfs2_setlease, 1040 .fallocate = gfs2_fallocate, 1041 }; 1042 1043 const struct file_operations gfs2_dir_fops = { 1044 .readdir = gfs2_readdir, 1045 .unlocked_ioctl = gfs2_ioctl, 1046 .open = gfs2_open, 1047 .release = gfs2_release, 1048 .fsync = gfs2_fsync, 1049 .lock = gfs2_lock, 1050 .flock = gfs2_flock, 1051 .llseek = default_llseek, 1052 }; 1053 1054 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */ 1055 1056 const struct file_operations gfs2_file_fops_nolock = { 1057 .llseek = gfs2_llseek, 1058 .read = do_sync_read, 1059 .aio_read = generic_file_aio_read, 1060 .write = do_sync_write, 1061 .aio_write = gfs2_file_aio_write, 1062 .unlocked_ioctl = gfs2_ioctl, 1063 .mmap = gfs2_mmap, 1064 .open = gfs2_open, 1065 .release = gfs2_release, 1066 .fsync = gfs2_fsync, 1067 .splice_read = generic_file_splice_read, 1068 .splice_write = generic_file_splice_write, 1069 .setlease = generic_setlease, 1070 .fallocate = gfs2_fallocate, 1071 }; 1072 1073 const struct file_operations gfs2_dir_fops_nolock = { 1074 .readdir = gfs2_readdir, 1075 .unlocked_ioctl = gfs2_ioctl, 1076 .open = gfs2_open, 1077 .release = gfs2_release, 1078 .fsync = gfs2_fsync, 1079 .llseek = default_llseek, 1080 }; 1081 1082