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_size_hint - Give a hint to the size of a write request 327 * @file: The struct file 328 * @offset: The file offset of the write 329 * @size: The length of the write 330 * 331 * When we are about to do a write, this function records the total 332 * write size in order to provide a suitable hint to the lower layers 333 * about how many blocks will be required. 334 * 335 */ 336 337 static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size) 338 { 339 struct inode *inode = filep->f_dentry->d_inode; 340 struct gfs2_sbd *sdp = GFS2_SB(inode); 341 struct gfs2_inode *ip = GFS2_I(inode); 342 size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift; 343 int hint = min_t(size_t, INT_MAX, blks); 344 345 atomic_set(&ip->i_res->rs_sizehint, hint); 346 } 347 348 /** 349 * gfs2_allocate_page_backing - Use bmap to allocate blocks 350 * @page: The (locked) page to allocate backing for 351 * 352 * We try to allocate all the blocks required for the page in 353 * one go. This might fail for various reasons, so we keep 354 * trying until all the blocks to back this page are allocated. 355 * If some of the blocks are already allocated, thats ok too. 356 */ 357 358 static int gfs2_allocate_page_backing(struct page *page) 359 { 360 struct inode *inode = page->mapping->host; 361 struct buffer_head bh; 362 unsigned long size = PAGE_CACHE_SIZE; 363 u64 lblock = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 364 365 do { 366 bh.b_state = 0; 367 bh.b_size = size; 368 gfs2_block_map(inode, lblock, &bh, 1); 369 if (!buffer_mapped(&bh)) 370 return -EIO; 371 size -= bh.b_size; 372 lblock += (bh.b_size >> inode->i_blkbits); 373 } while(size > 0); 374 return 0; 375 } 376 377 /** 378 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable 379 * @vma: The virtual memory area 380 * @page: The page which is about to become writable 381 * 382 * When the page becomes writable, we need to ensure that we have 383 * blocks allocated on disk to back that page. 384 */ 385 386 static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) 387 { 388 struct page *page = vmf->page; 389 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 390 struct gfs2_inode *ip = GFS2_I(inode); 391 struct gfs2_sbd *sdp = GFS2_SB(inode); 392 unsigned long last_index; 393 u64 pos = page->index << PAGE_CACHE_SHIFT; 394 unsigned int data_blocks, ind_blocks, rblocks; 395 struct gfs2_holder gh; 396 loff_t size; 397 int ret; 398 399 sb_start_pagefault(inode->i_sb); 400 401 /* Update file times before taking page lock */ 402 file_update_time(vma->vm_file); 403 404 ret = gfs2_rs_alloc(ip); 405 if (ret) 406 return ret; 407 408 gfs2_size_hint(vma->vm_file, pos, PAGE_CACHE_SIZE); 409 410 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 411 ret = gfs2_glock_nq(&gh); 412 if (ret) 413 goto out; 414 415 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags); 416 set_bit(GIF_SW_PAGED, &ip->i_flags); 417 418 if (!gfs2_write_alloc_required(ip, pos, PAGE_CACHE_SIZE)) { 419 lock_page(page); 420 if (!PageUptodate(page) || page->mapping != inode->i_mapping) { 421 ret = -EAGAIN; 422 unlock_page(page); 423 } 424 goto out_unlock; 425 } 426 427 ret = gfs2_rindex_update(sdp); 428 if (ret) 429 goto out_unlock; 430 431 ret = gfs2_quota_lock_check(ip); 432 if (ret) 433 goto out_unlock; 434 gfs2_write_calc_reserv(ip, PAGE_CACHE_SIZE, &data_blocks, &ind_blocks); 435 ret = gfs2_inplace_reserve(ip, data_blocks + ind_blocks); 436 if (ret) 437 goto out_quota_unlock; 438 439 rblocks = RES_DINODE + ind_blocks; 440 if (gfs2_is_jdata(ip)) 441 rblocks += data_blocks ? data_blocks : 1; 442 if (ind_blocks || data_blocks) { 443 rblocks += RES_STATFS + RES_QUOTA; 444 rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks); 445 } 446 ret = gfs2_trans_begin(sdp, rblocks, 0); 447 if (ret) 448 goto out_trans_fail; 449 450 lock_page(page); 451 ret = -EINVAL; 452 size = i_size_read(inode); 453 last_index = (size - 1) >> PAGE_CACHE_SHIFT; 454 /* Check page index against inode size */ 455 if (size == 0 || (page->index > last_index)) 456 goto out_trans_end; 457 458 ret = -EAGAIN; 459 /* If truncated, we must retry the operation, we may have raced 460 * with the glock demotion code. 461 */ 462 if (!PageUptodate(page) || page->mapping != inode->i_mapping) 463 goto out_trans_end; 464 465 /* Unstuff, if required, and allocate backing blocks for page */ 466 ret = 0; 467 if (gfs2_is_stuffed(ip)) 468 ret = gfs2_unstuff_dinode(ip, page); 469 if (ret == 0) 470 ret = gfs2_allocate_page_backing(page); 471 472 out_trans_end: 473 if (ret) 474 unlock_page(page); 475 gfs2_trans_end(sdp); 476 out_trans_fail: 477 gfs2_inplace_release(ip); 478 out_quota_unlock: 479 gfs2_quota_unlock(ip); 480 out_unlock: 481 gfs2_glock_dq(&gh); 482 out: 483 gfs2_holder_uninit(&gh); 484 if (ret == 0) { 485 set_page_dirty(page); 486 wait_on_page_writeback(page); 487 } 488 sb_end_pagefault(inode->i_sb); 489 return block_page_mkwrite_return(ret); 490 } 491 492 static const struct vm_operations_struct gfs2_vm_ops = { 493 .fault = filemap_fault, 494 .page_mkwrite = gfs2_page_mkwrite, 495 .remap_pages = generic_file_remap_pages, 496 }; 497 498 /** 499 * gfs2_mmap - 500 * @file: The file to map 501 * @vma: The VMA which described the mapping 502 * 503 * There is no need to get a lock here unless we should be updating 504 * atime. We ignore any locking errors since the only consequence is 505 * a missed atime update (which will just be deferred until later). 506 * 507 * Returns: 0 508 */ 509 510 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma) 511 { 512 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 513 514 if (!(file->f_flags & O_NOATIME) && 515 !IS_NOATIME(&ip->i_inode)) { 516 struct gfs2_holder i_gh; 517 int error; 518 519 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh); 520 error = gfs2_glock_nq(&i_gh); 521 if (error == 0) { 522 file_accessed(file); 523 gfs2_glock_dq(&i_gh); 524 } 525 gfs2_holder_uninit(&i_gh); 526 if (error) 527 return error; 528 } 529 vma->vm_ops = &gfs2_vm_ops; 530 531 return 0; 532 } 533 534 /** 535 * gfs2_open - open a file 536 * @inode: the inode to open 537 * @file: the struct file for this opening 538 * 539 * Returns: errno 540 */ 541 542 static int gfs2_open(struct inode *inode, struct file *file) 543 { 544 struct gfs2_inode *ip = GFS2_I(inode); 545 struct gfs2_holder i_gh; 546 struct gfs2_file *fp; 547 int error; 548 549 fp = kzalloc(sizeof(struct gfs2_file), GFP_KERNEL); 550 if (!fp) 551 return -ENOMEM; 552 553 mutex_init(&fp->f_fl_mutex); 554 555 gfs2_assert_warn(GFS2_SB(inode), !file->private_data); 556 file->private_data = fp; 557 558 if (S_ISREG(ip->i_inode.i_mode)) { 559 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, 560 &i_gh); 561 if (error) 562 goto fail; 563 564 if (!(file->f_flags & O_LARGEFILE) && 565 i_size_read(inode) > MAX_NON_LFS) { 566 error = -EOVERFLOW; 567 goto fail_gunlock; 568 } 569 570 gfs2_glock_dq_uninit(&i_gh); 571 } 572 573 return 0; 574 575 fail_gunlock: 576 gfs2_glock_dq_uninit(&i_gh); 577 fail: 578 file->private_data = NULL; 579 kfree(fp); 580 return error; 581 } 582 583 /** 584 * gfs2_release - called to close a struct file 585 * @inode: the inode the struct file belongs to 586 * @file: the struct file being closed 587 * 588 * Returns: errno 589 */ 590 591 static int gfs2_release(struct inode *inode, struct file *file) 592 { 593 struct gfs2_inode *ip = GFS2_I(inode); 594 595 kfree(file->private_data); 596 file->private_data = NULL; 597 598 if ((file->f_mode & FMODE_WRITE) && 599 (atomic_read(&inode->i_writecount) == 1)) 600 gfs2_rs_delete(ip); 601 602 return 0; 603 } 604 605 /** 606 * gfs2_fsync - sync the dirty data for a file (across the cluster) 607 * @file: the file that points to the dentry 608 * @start: the start position in the file to sync 609 * @end: the end position in the file to sync 610 * @datasync: set if we can ignore timestamp changes 611 * 612 * We split the data flushing here so that we don't wait for the data 613 * until after we've also sent the metadata to disk. Note that for 614 * data=ordered, we will write & wait for the data at the log flush 615 * stage anyway, so this is unlikely to make much of a difference 616 * except in the data=writeback case. 617 * 618 * If the fdatawrite fails due to any reason except -EIO, we will 619 * continue the remainder of the fsync, although we'll still report 620 * the error at the end. This is to match filemap_write_and_wait_range() 621 * behaviour. 622 * 623 * Returns: errno 624 */ 625 626 static int gfs2_fsync(struct file *file, loff_t start, loff_t end, 627 int datasync) 628 { 629 struct address_space *mapping = file->f_mapping; 630 struct inode *inode = mapping->host; 631 int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC); 632 struct gfs2_inode *ip = GFS2_I(inode); 633 int ret = 0, ret1 = 0; 634 635 if (mapping->nrpages) { 636 ret1 = filemap_fdatawrite_range(mapping, start, end); 637 if (ret1 == -EIO) 638 return ret1; 639 } 640 641 if (datasync) 642 sync_state &= ~I_DIRTY_SYNC; 643 644 if (sync_state) { 645 ret = sync_inode_metadata(inode, 1); 646 if (ret) 647 return ret; 648 if (gfs2_is_jdata(ip)) 649 filemap_write_and_wait(mapping); 650 gfs2_ail_flush(ip->i_gl, 1); 651 } 652 653 if (mapping->nrpages) 654 ret = filemap_fdatawait_range(mapping, start, end); 655 656 return ret ? ret : ret1; 657 } 658 659 /** 660 * gfs2_file_aio_write - Perform a write to a file 661 * @iocb: The io context 662 * @iov: The data to write 663 * @nr_segs: Number of @iov segments 664 * @pos: The file position 665 * 666 * We have to do a lock/unlock here to refresh the inode size for 667 * O_APPEND writes, otherwise we can land up writing at the wrong 668 * offset. There is still a race, but provided the app is using its 669 * own file locking, this will make O_APPEND work as expected. 670 * 671 */ 672 673 static ssize_t gfs2_file_aio_write(struct kiocb *iocb, const struct iovec *iov, 674 unsigned long nr_segs, loff_t pos) 675 { 676 struct file *file = iocb->ki_filp; 677 size_t writesize = iov_length(iov, nr_segs); 678 struct dentry *dentry = file->f_dentry; 679 struct gfs2_inode *ip = GFS2_I(dentry->d_inode); 680 struct gfs2_sbd *sdp; 681 int ret; 682 683 sdp = GFS2_SB(file->f_mapping->host); 684 ret = gfs2_rs_alloc(ip); 685 if (ret) 686 return ret; 687 688 gfs2_size_hint(file, pos, writesize); 689 690 if (file->f_flags & O_APPEND) { 691 struct gfs2_holder gh; 692 693 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh); 694 if (ret) 695 return ret; 696 gfs2_glock_dq_uninit(&gh); 697 } 698 699 return generic_file_aio_write(iocb, iov, nr_segs, pos); 700 } 701 702 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len, 703 int mode) 704 { 705 struct gfs2_inode *ip = GFS2_I(inode); 706 struct buffer_head *dibh; 707 int error; 708 loff_t size = len; 709 unsigned int nr_blks; 710 sector_t lblock = offset >> inode->i_blkbits; 711 712 error = gfs2_meta_inode_buffer(ip, &dibh); 713 if (unlikely(error)) 714 return error; 715 716 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 717 718 if (gfs2_is_stuffed(ip)) { 719 error = gfs2_unstuff_dinode(ip, NULL); 720 if (unlikely(error)) 721 goto out; 722 } 723 724 while (len) { 725 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 }; 726 bh_map.b_size = len; 727 set_buffer_zeronew(&bh_map); 728 729 error = gfs2_block_map(inode, lblock, &bh_map, 1); 730 if (unlikely(error)) 731 goto out; 732 len -= bh_map.b_size; 733 nr_blks = bh_map.b_size >> inode->i_blkbits; 734 lblock += nr_blks; 735 if (!buffer_new(&bh_map)) 736 continue; 737 if (unlikely(!buffer_zeronew(&bh_map))) { 738 error = -EIO; 739 goto out; 740 } 741 } 742 if (offset + size > inode->i_size && !(mode & FALLOC_FL_KEEP_SIZE)) 743 i_size_write(inode, offset + size); 744 745 mark_inode_dirty(inode); 746 747 out: 748 brelse(dibh); 749 return error; 750 } 751 752 static void calc_max_reserv(struct gfs2_inode *ip, loff_t max, loff_t *len, 753 unsigned int *data_blocks, unsigned int *ind_blocks) 754 { 755 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 756 unsigned int max_blocks = ip->i_rgd->rd_free_clone; 757 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1); 758 759 for (tmp = max_data; tmp > sdp->sd_diptrs;) { 760 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs); 761 max_data -= tmp; 762 } 763 /* This calculation isn't the exact reverse of gfs2_write_calc_reserve, 764 so it might end up with fewer data blocks */ 765 if (max_data <= *data_blocks) 766 return; 767 *data_blocks = max_data; 768 *ind_blocks = max_blocks - max_data; 769 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift; 770 if (*len > max) { 771 *len = max; 772 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks); 773 } 774 } 775 776 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, 777 loff_t len) 778 { 779 struct inode *inode = file->f_path.dentry->d_inode; 780 struct gfs2_sbd *sdp = GFS2_SB(inode); 781 struct gfs2_inode *ip = GFS2_I(inode); 782 unsigned int data_blocks = 0, ind_blocks = 0, rblocks; 783 loff_t bytes, max_bytes; 784 int error; 785 const loff_t pos = offset; 786 const loff_t count = len; 787 loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1); 788 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift; 789 loff_t max_chunk_size = UINT_MAX & bsize_mask; 790 next = (next + 1) << sdp->sd_sb.sb_bsize_shift; 791 792 /* We only support the FALLOC_FL_KEEP_SIZE mode */ 793 if (mode & ~FALLOC_FL_KEEP_SIZE) 794 return -EOPNOTSUPP; 795 796 offset &= bsize_mask; 797 798 len = next - offset; 799 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2; 800 if (!bytes) 801 bytes = UINT_MAX; 802 bytes &= bsize_mask; 803 if (bytes == 0) 804 bytes = sdp->sd_sb.sb_bsize; 805 806 error = gfs2_rs_alloc(ip); 807 if (error) 808 return error; 809 810 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh); 811 error = gfs2_glock_nq(&ip->i_gh); 812 if (unlikely(error)) 813 goto out_uninit; 814 815 gfs2_size_hint(file, offset, len); 816 817 while (len > 0) { 818 if (len < bytes) 819 bytes = len; 820 if (!gfs2_write_alloc_required(ip, offset, bytes)) { 821 len -= bytes; 822 offset += bytes; 823 continue; 824 } 825 error = gfs2_quota_lock_check(ip); 826 if (error) 827 goto out_unlock; 828 829 retry: 830 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks); 831 832 error = gfs2_inplace_reserve(ip, data_blocks + ind_blocks); 833 if (error) { 834 if (error == -ENOSPC && bytes > sdp->sd_sb.sb_bsize) { 835 bytes >>= 1; 836 bytes &= bsize_mask; 837 if (bytes == 0) 838 bytes = sdp->sd_sb.sb_bsize; 839 goto retry; 840 } 841 goto out_qunlock; 842 } 843 max_bytes = bytes; 844 calc_max_reserv(ip, (len > max_chunk_size)? max_chunk_size: len, 845 &max_bytes, &data_blocks, &ind_blocks); 846 847 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA + 848 RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks); 849 if (gfs2_is_jdata(ip)) 850 rblocks += data_blocks ? data_blocks : 1; 851 852 error = gfs2_trans_begin(sdp, rblocks, 853 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize); 854 if (error) 855 goto out_trans_fail; 856 857 error = fallocate_chunk(inode, offset, max_bytes, mode); 858 gfs2_trans_end(sdp); 859 860 if (error) 861 goto out_trans_fail; 862 863 len -= max_bytes; 864 offset += max_bytes; 865 gfs2_inplace_release(ip); 866 gfs2_quota_unlock(ip); 867 } 868 869 if (error == 0) 870 error = generic_write_sync(file, pos, count); 871 goto out_unlock; 872 873 out_trans_fail: 874 gfs2_inplace_release(ip); 875 out_qunlock: 876 gfs2_quota_unlock(ip); 877 out_unlock: 878 gfs2_glock_dq(&ip->i_gh); 879 out_uninit: 880 gfs2_holder_uninit(&ip->i_gh); 881 return error; 882 } 883 884 #ifdef CONFIG_GFS2_FS_LOCKING_DLM 885 886 /** 887 * gfs2_setlease - acquire/release a file lease 888 * @file: the file pointer 889 * @arg: lease type 890 * @fl: file lock 891 * 892 * We don't currently have a way to enforce a lease across the whole 893 * cluster; until we do, disable leases (by just returning -EINVAL), 894 * unless the administrator has requested purely local locking. 895 * 896 * Locking: called under lock_flocks 897 * 898 * Returns: errno 899 */ 900 901 static int gfs2_setlease(struct file *file, long arg, struct file_lock **fl) 902 { 903 return -EINVAL; 904 } 905 906 /** 907 * gfs2_lock - acquire/release a posix lock on a file 908 * @file: the file pointer 909 * @cmd: either modify or retrieve lock state, possibly wait 910 * @fl: type and range of lock 911 * 912 * Returns: errno 913 */ 914 915 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl) 916 { 917 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 918 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host); 919 struct lm_lockstruct *ls = &sdp->sd_lockstruct; 920 921 if (!(fl->fl_flags & FL_POSIX)) 922 return -ENOLCK; 923 if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK) 924 return -ENOLCK; 925 926 if (cmd == F_CANCELLK) { 927 /* Hack: */ 928 cmd = F_SETLK; 929 fl->fl_type = F_UNLCK; 930 } 931 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 932 return -EIO; 933 if (IS_GETLK(cmd)) 934 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl); 935 else if (fl->fl_type == F_UNLCK) 936 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl); 937 else 938 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl); 939 } 940 941 static int do_flock(struct file *file, int cmd, struct file_lock *fl) 942 { 943 struct gfs2_file *fp = file->private_data; 944 struct gfs2_holder *fl_gh = &fp->f_fl_gh; 945 struct gfs2_inode *ip = GFS2_I(file->f_path.dentry->d_inode); 946 struct gfs2_glock *gl; 947 unsigned int state; 948 int flags; 949 int error = 0; 950 951 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED; 952 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE; 953 954 mutex_lock(&fp->f_fl_mutex); 955 956 gl = fl_gh->gh_gl; 957 if (gl) { 958 if (fl_gh->gh_state == state) 959 goto out; 960 flock_lock_file_wait(file, 961 &(struct file_lock){.fl_type = F_UNLCK}); 962 gfs2_glock_dq_wait(fl_gh); 963 gfs2_holder_reinit(state, flags, fl_gh); 964 } else { 965 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr, 966 &gfs2_flock_glops, CREATE, &gl); 967 if (error) 968 goto out; 969 gfs2_holder_init(gl, state, flags, fl_gh); 970 gfs2_glock_put(gl); 971 } 972 error = gfs2_glock_nq(fl_gh); 973 if (error) { 974 gfs2_holder_uninit(fl_gh); 975 if (error == GLR_TRYFAILED) 976 error = -EAGAIN; 977 } else { 978 error = flock_lock_file_wait(file, fl); 979 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); 980 } 981 982 out: 983 mutex_unlock(&fp->f_fl_mutex); 984 return error; 985 } 986 987 static void do_unflock(struct file *file, struct file_lock *fl) 988 { 989 struct gfs2_file *fp = file->private_data; 990 struct gfs2_holder *fl_gh = &fp->f_fl_gh; 991 992 mutex_lock(&fp->f_fl_mutex); 993 flock_lock_file_wait(file, fl); 994 if (fl_gh->gh_gl) { 995 gfs2_glock_dq_wait(fl_gh); 996 gfs2_holder_uninit(fl_gh); 997 } 998 mutex_unlock(&fp->f_fl_mutex); 999 } 1000 1001 /** 1002 * gfs2_flock - acquire/release a flock lock on a file 1003 * @file: the file pointer 1004 * @cmd: either modify or retrieve lock state, possibly wait 1005 * @fl: type and range of lock 1006 * 1007 * Returns: errno 1008 */ 1009 1010 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl) 1011 { 1012 if (!(fl->fl_flags & FL_FLOCK)) 1013 return -ENOLCK; 1014 if (fl->fl_type & LOCK_MAND) 1015 return -EOPNOTSUPP; 1016 1017 if (fl->fl_type == F_UNLCK) { 1018 do_unflock(file, fl); 1019 return 0; 1020 } else { 1021 return do_flock(file, cmd, fl); 1022 } 1023 } 1024 1025 const struct file_operations gfs2_file_fops = { 1026 .llseek = gfs2_llseek, 1027 .read = do_sync_read, 1028 .aio_read = generic_file_aio_read, 1029 .write = do_sync_write, 1030 .aio_write = gfs2_file_aio_write, 1031 .unlocked_ioctl = gfs2_ioctl, 1032 .mmap = gfs2_mmap, 1033 .open = gfs2_open, 1034 .release = gfs2_release, 1035 .fsync = gfs2_fsync, 1036 .lock = gfs2_lock, 1037 .flock = gfs2_flock, 1038 .splice_read = generic_file_splice_read, 1039 .splice_write = generic_file_splice_write, 1040 .setlease = gfs2_setlease, 1041 .fallocate = gfs2_fallocate, 1042 }; 1043 1044 const struct file_operations gfs2_dir_fops = { 1045 .readdir = gfs2_readdir, 1046 .unlocked_ioctl = gfs2_ioctl, 1047 .open = gfs2_open, 1048 .release = gfs2_release, 1049 .fsync = gfs2_fsync, 1050 .lock = gfs2_lock, 1051 .flock = gfs2_flock, 1052 .llseek = default_llseek, 1053 }; 1054 1055 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */ 1056 1057 const struct file_operations gfs2_file_fops_nolock = { 1058 .llseek = gfs2_llseek, 1059 .read = do_sync_read, 1060 .aio_read = generic_file_aio_read, 1061 .write = do_sync_write, 1062 .aio_write = gfs2_file_aio_write, 1063 .unlocked_ioctl = gfs2_ioctl, 1064 .mmap = gfs2_mmap, 1065 .open = gfs2_open, 1066 .release = gfs2_release, 1067 .fsync = gfs2_fsync, 1068 .splice_read = generic_file_splice_read, 1069 .splice_write = generic_file_splice_write, 1070 .setlease = generic_setlease, 1071 .fallocate = gfs2_fallocate, 1072 }; 1073 1074 const struct file_operations gfs2_dir_fops_nolock = { 1075 .readdir = gfs2_readdir, 1076 .unlocked_ioctl = gfs2_ioctl, 1077 .open = gfs2_open, 1078 .release = gfs2_release, 1079 .fsync = gfs2_fsync, 1080 .llseek = default_llseek, 1081 }; 1082 1083