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/delay.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 - Iterator for a directory 86 * @file: The directory to read from 87 * @ctx: What to feed directory entries to 88 * 89 * Returns: errno 90 */ 91 92 static int gfs2_readdir(struct file *file, struct dir_context *ctx) 93 { 94 struct inode *dir = file->f_mapping->host; 95 struct gfs2_inode *dip = GFS2_I(dir); 96 struct gfs2_holder d_gh; 97 int error; 98 99 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh); 100 if (error) 101 return error; 102 103 error = gfs2_dir_read(dir, ctx, &file->f_ra); 104 105 gfs2_glock_dq_uninit(&d_gh); 106 107 return error; 108 } 109 110 /** 111 * fsflags_cvt 112 * @table: A table of 32 u32 flags 113 * @val: a 32 bit value to convert 114 * 115 * This function can be used to convert between fsflags values and 116 * GFS2's own flags values. 117 * 118 * Returns: the converted flags 119 */ 120 static u32 fsflags_cvt(const u32 *table, u32 val) 121 { 122 u32 res = 0; 123 while(val) { 124 if (val & 1) 125 res |= *table; 126 table++; 127 val >>= 1; 128 } 129 return res; 130 } 131 132 static const u32 fsflags_to_gfs2[32] = { 133 [3] = GFS2_DIF_SYNC, 134 [4] = GFS2_DIF_IMMUTABLE, 135 [5] = GFS2_DIF_APPENDONLY, 136 [7] = GFS2_DIF_NOATIME, 137 [12] = GFS2_DIF_EXHASH, 138 [14] = GFS2_DIF_INHERIT_JDATA, 139 [17] = GFS2_DIF_TOPDIR, 140 }; 141 142 static const u32 gfs2_to_fsflags[32] = { 143 [gfs2fl_Sync] = FS_SYNC_FL, 144 [gfs2fl_Immutable] = FS_IMMUTABLE_FL, 145 [gfs2fl_AppendOnly] = FS_APPEND_FL, 146 [gfs2fl_NoAtime] = FS_NOATIME_FL, 147 [gfs2fl_ExHash] = FS_INDEX_FL, 148 [gfs2fl_TopLevel] = FS_TOPDIR_FL, 149 [gfs2fl_InheritJdata] = FS_JOURNAL_DATA_FL, 150 }; 151 152 static int gfs2_get_flags(struct file *filp, u32 __user *ptr) 153 { 154 struct inode *inode = file_inode(filp); 155 struct gfs2_inode *ip = GFS2_I(inode); 156 struct gfs2_holder gh; 157 int error; 158 u32 fsflags; 159 160 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); 161 error = gfs2_glock_nq(&gh); 162 if (error) 163 return error; 164 165 fsflags = fsflags_cvt(gfs2_to_fsflags, ip->i_diskflags); 166 if (!S_ISDIR(inode->i_mode) && ip->i_diskflags & GFS2_DIF_JDATA) 167 fsflags |= FS_JOURNAL_DATA_FL; 168 if (put_user(fsflags, ptr)) 169 error = -EFAULT; 170 171 gfs2_glock_dq(&gh); 172 gfs2_holder_uninit(&gh); 173 return error; 174 } 175 176 void gfs2_set_inode_flags(struct inode *inode) 177 { 178 struct gfs2_inode *ip = GFS2_I(inode); 179 unsigned int flags = inode->i_flags; 180 181 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC); 182 if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode)) 183 flags |= S_NOSEC; 184 if (ip->i_diskflags & GFS2_DIF_IMMUTABLE) 185 flags |= S_IMMUTABLE; 186 if (ip->i_diskflags & GFS2_DIF_APPENDONLY) 187 flags |= S_APPEND; 188 if (ip->i_diskflags & GFS2_DIF_NOATIME) 189 flags |= S_NOATIME; 190 if (ip->i_diskflags & GFS2_DIF_SYNC) 191 flags |= S_SYNC; 192 inode->i_flags = flags; 193 } 194 195 /* Flags that can be set by user space */ 196 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \ 197 GFS2_DIF_IMMUTABLE| \ 198 GFS2_DIF_APPENDONLY| \ 199 GFS2_DIF_NOATIME| \ 200 GFS2_DIF_SYNC| \ 201 GFS2_DIF_SYSTEM| \ 202 GFS2_DIF_TOPDIR| \ 203 GFS2_DIF_INHERIT_JDATA) 204 205 /** 206 * do_gfs2_set_flags - set flags on an inode 207 * @filp: file pointer 208 * @reqflags: The flags to set 209 * @mask: Indicates which flags are valid 210 * 211 */ 212 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask) 213 { 214 struct inode *inode = file_inode(filp); 215 struct gfs2_inode *ip = GFS2_I(inode); 216 struct gfs2_sbd *sdp = GFS2_SB(inode); 217 struct buffer_head *bh; 218 struct gfs2_holder gh; 219 int error; 220 u32 new_flags, flags; 221 222 error = mnt_want_write_file(filp); 223 if (error) 224 return error; 225 226 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 227 if (error) 228 goto out_drop_write; 229 230 error = -EACCES; 231 if (!inode_owner_or_capable(inode)) 232 goto out; 233 234 error = 0; 235 flags = ip->i_diskflags; 236 new_flags = (flags & ~mask) | (reqflags & mask); 237 if ((new_flags ^ flags) == 0) 238 goto out; 239 240 error = -EINVAL; 241 if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET) 242 goto out; 243 244 error = -EPERM; 245 if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE)) 246 goto out; 247 if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY)) 248 goto out; 249 if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) && 250 !capable(CAP_LINUX_IMMUTABLE)) 251 goto out; 252 if (!IS_IMMUTABLE(inode)) { 253 error = gfs2_permission(inode, MAY_WRITE); 254 if (error) 255 goto out; 256 } 257 if ((flags ^ new_flags) & GFS2_DIF_JDATA) { 258 if (flags & GFS2_DIF_JDATA) 259 gfs2_log_flush(sdp, ip->i_gl, NORMAL_FLUSH); 260 error = filemap_fdatawrite(inode->i_mapping); 261 if (error) 262 goto out; 263 error = filemap_fdatawait(inode->i_mapping); 264 if (error) 265 goto out; 266 } 267 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 268 if (error) 269 goto out; 270 error = gfs2_meta_inode_buffer(ip, &bh); 271 if (error) 272 goto out_trans_end; 273 gfs2_trans_add_meta(ip->i_gl, bh); 274 ip->i_diskflags = new_flags; 275 gfs2_dinode_out(ip, bh->b_data); 276 brelse(bh); 277 gfs2_set_inode_flags(inode); 278 gfs2_set_aops(inode); 279 out_trans_end: 280 gfs2_trans_end(sdp); 281 out: 282 gfs2_glock_dq_uninit(&gh); 283 out_drop_write: 284 mnt_drop_write_file(filp); 285 return error; 286 } 287 288 static int gfs2_set_flags(struct file *filp, u32 __user *ptr) 289 { 290 struct inode *inode = file_inode(filp); 291 u32 fsflags, gfsflags; 292 293 if (get_user(fsflags, ptr)) 294 return -EFAULT; 295 296 gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags); 297 if (!S_ISDIR(inode->i_mode)) { 298 gfsflags &= ~GFS2_DIF_TOPDIR; 299 if (gfsflags & GFS2_DIF_INHERIT_JDATA) 300 gfsflags ^= (GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA); 301 return do_gfs2_set_flags(filp, gfsflags, ~GFS2_DIF_SYSTEM); 302 } 303 return do_gfs2_set_flags(filp, gfsflags, ~(GFS2_DIF_SYSTEM | GFS2_DIF_JDATA)); 304 } 305 306 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) 307 { 308 switch(cmd) { 309 case FS_IOC_GETFLAGS: 310 return gfs2_get_flags(filp, (u32 __user *)arg); 311 case FS_IOC_SETFLAGS: 312 return gfs2_set_flags(filp, (u32 __user *)arg); 313 case FITRIM: 314 return gfs2_fitrim(filp, (void __user *)arg); 315 } 316 return -ENOTTY; 317 } 318 319 /** 320 * gfs2_size_hint - Give a hint to the size of a write request 321 * @filep: The struct file 322 * @offset: The file offset of the write 323 * @size: The length of the write 324 * 325 * When we are about to do a write, this function records the total 326 * write size in order to provide a suitable hint to the lower layers 327 * about how many blocks will be required. 328 * 329 */ 330 331 static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size) 332 { 333 struct inode *inode = file_inode(filep); 334 struct gfs2_sbd *sdp = GFS2_SB(inode); 335 struct gfs2_inode *ip = GFS2_I(inode); 336 size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift; 337 int hint = min_t(size_t, INT_MAX, blks); 338 339 if (hint > atomic_read(&ip->i_res.rs_sizehint)) 340 atomic_set(&ip->i_res.rs_sizehint, hint); 341 } 342 343 /** 344 * gfs2_allocate_page_backing - Use bmap to allocate blocks 345 * @page: The (locked) page to allocate backing for 346 * 347 * We try to allocate all the blocks required for the page in 348 * one go. This might fail for various reasons, so we keep 349 * trying until all the blocks to back this page are allocated. 350 * If some of the blocks are already allocated, thats ok too. 351 */ 352 353 static int gfs2_allocate_page_backing(struct page *page) 354 { 355 struct inode *inode = page->mapping->host; 356 struct buffer_head bh; 357 unsigned long size = PAGE_CACHE_SIZE; 358 u64 lblock = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 359 360 do { 361 bh.b_state = 0; 362 bh.b_size = size; 363 gfs2_block_map(inode, lblock, &bh, 1); 364 if (!buffer_mapped(&bh)) 365 return -EIO; 366 size -= bh.b_size; 367 lblock += (bh.b_size >> inode->i_blkbits); 368 } while(size > 0); 369 return 0; 370 } 371 372 /** 373 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable 374 * @vma: The virtual memory area 375 * @vmf: The virtual memory fault containing the page to become writable 376 * 377 * When the page becomes writable, we need to ensure that we have 378 * blocks allocated on disk to back that page. 379 */ 380 381 static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) 382 { 383 struct page *page = vmf->page; 384 struct inode *inode = file_inode(vma->vm_file); 385 struct gfs2_inode *ip = GFS2_I(inode); 386 struct gfs2_sbd *sdp = GFS2_SB(inode); 387 struct gfs2_alloc_parms ap = { .aflags = 0, }; 388 unsigned long last_index; 389 u64 pos = page->index << PAGE_CACHE_SHIFT; 390 unsigned int data_blocks, ind_blocks, rblocks; 391 struct gfs2_holder gh; 392 loff_t size; 393 int ret; 394 395 sb_start_pagefault(inode->i_sb); 396 397 /* Update file times before taking page lock */ 398 file_update_time(vma->vm_file); 399 400 ret = gfs2_rsqa_alloc(ip); 401 if (ret) 402 goto out; 403 404 gfs2_size_hint(vma->vm_file, pos, PAGE_CACHE_SIZE); 405 406 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 407 ret = gfs2_glock_nq(&gh); 408 if (ret) 409 goto out_uninit; 410 411 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags); 412 set_bit(GIF_SW_PAGED, &ip->i_flags); 413 414 if (!gfs2_write_alloc_required(ip, pos, PAGE_CACHE_SIZE)) { 415 lock_page(page); 416 if (!PageUptodate(page) || page->mapping != inode->i_mapping) { 417 ret = -EAGAIN; 418 unlock_page(page); 419 } 420 goto out_unlock; 421 } 422 423 ret = gfs2_rindex_update(sdp); 424 if (ret) 425 goto out_unlock; 426 427 gfs2_write_calc_reserv(ip, PAGE_CACHE_SIZE, &data_blocks, &ind_blocks); 428 ap.target = data_blocks + ind_blocks; 429 ret = gfs2_quota_lock_check(ip, &ap); 430 if (ret) 431 goto out_unlock; 432 ret = gfs2_inplace_reserve(ip, &ap); 433 if (ret) 434 goto out_quota_unlock; 435 436 rblocks = RES_DINODE + ind_blocks; 437 if (gfs2_is_jdata(ip)) 438 rblocks += data_blocks ? data_blocks : 1; 439 if (ind_blocks || data_blocks) { 440 rblocks += RES_STATFS + RES_QUOTA; 441 rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks); 442 } 443 ret = gfs2_trans_begin(sdp, rblocks, 0); 444 if (ret) 445 goto out_trans_fail; 446 447 lock_page(page); 448 ret = -EINVAL; 449 size = i_size_read(inode); 450 last_index = (size - 1) >> PAGE_CACHE_SHIFT; 451 /* Check page index against inode size */ 452 if (size == 0 || (page->index > last_index)) 453 goto out_trans_end; 454 455 ret = -EAGAIN; 456 /* If truncated, we must retry the operation, we may have raced 457 * with the glock demotion code. 458 */ 459 if (!PageUptodate(page) || page->mapping != inode->i_mapping) 460 goto out_trans_end; 461 462 /* Unstuff, if required, and allocate backing blocks for page */ 463 ret = 0; 464 if (gfs2_is_stuffed(ip)) 465 ret = gfs2_unstuff_dinode(ip, page); 466 if (ret == 0) 467 ret = gfs2_allocate_page_backing(page); 468 469 out_trans_end: 470 if (ret) 471 unlock_page(page); 472 gfs2_trans_end(sdp); 473 out_trans_fail: 474 gfs2_inplace_release(ip); 475 out_quota_unlock: 476 gfs2_quota_unlock(ip); 477 out_unlock: 478 gfs2_glock_dq(&gh); 479 out_uninit: 480 gfs2_holder_uninit(&gh); 481 if (ret == 0) { 482 set_page_dirty(page); 483 wait_for_stable_page(page); 484 } 485 out: 486 sb_end_pagefault(inode->i_sb); 487 return block_page_mkwrite_return(ret); 488 } 489 490 static const struct vm_operations_struct gfs2_vm_ops = { 491 .fault = filemap_fault, 492 .map_pages = filemap_map_pages, 493 .page_mkwrite = gfs2_page_mkwrite, 494 }; 495 496 /** 497 * gfs2_mmap - 498 * @file: The file to map 499 * @vma: The VMA which described the mapping 500 * 501 * There is no need to get a lock here unless we should be updating 502 * atime. We ignore any locking errors since the only consequence is 503 * a missed atime update (which will just be deferred until later). 504 * 505 * Returns: 0 506 */ 507 508 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma) 509 { 510 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 511 512 if (!(file->f_flags & O_NOATIME) && 513 !IS_NOATIME(&ip->i_inode)) { 514 struct gfs2_holder i_gh; 515 int error; 516 517 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, 518 &i_gh); 519 if (error) 520 return error; 521 /* grab lock to update inode */ 522 gfs2_glock_dq_uninit(&i_gh); 523 file_accessed(file); 524 } 525 vma->vm_ops = &gfs2_vm_ops; 526 527 return 0; 528 } 529 530 /** 531 * gfs2_open_common - This is common to open and atomic_open 532 * @inode: The inode being opened 533 * @file: The file being opened 534 * 535 * This maybe called under a glock or not depending upon how it has 536 * been called. We must always be called under a glock for regular 537 * files, however. For other file types, it does not matter whether 538 * we hold the glock or not. 539 * 540 * Returns: Error code or 0 for success 541 */ 542 543 int gfs2_open_common(struct inode *inode, struct file *file) 544 { 545 struct gfs2_file *fp; 546 int ret; 547 548 if (S_ISREG(inode->i_mode)) { 549 ret = generic_file_open(inode, file); 550 if (ret) 551 return ret; 552 } 553 554 fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS); 555 if (!fp) 556 return -ENOMEM; 557 558 mutex_init(&fp->f_fl_mutex); 559 560 gfs2_assert_warn(GFS2_SB(inode), !file->private_data); 561 file->private_data = fp; 562 return 0; 563 } 564 565 /** 566 * gfs2_open - open a file 567 * @inode: the inode to open 568 * @file: the struct file for this opening 569 * 570 * After atomic_open, this function is only used for opening files 571 * which are already cached. We must still get the glock for regular 572 * files to ensure that we have the file size uptodate for the large 573 * file check which is in the common code. That is only an issue for 574 * regular files though. 575 * 576 * Returns: errno 577 */ 578 579 static int gfs2_open(struct inode *inode, struct file *file) 580 { 581 struct gfs2_inode *ip = GFS2_I(inode); 582 struct gfs2_holder i_gh; 583 int error; 584 bool need_unlock = false; 585 586 if (S_ISREG(ip->i_inode.i_mode)) { 587 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, 588 &i_gh); 589 if (error) 590 return error; 591 need_unlock = true; 592 } 593 594 error = gfs2_open_common(inode, file); 595 596 if (need_unlock) 597 gfs2_glock_dq_uninit(&i_gh); 598 599 return error; 600 } 601 602 /** 603 * gfs2_release - called to close a struct file 604 * @inode: the inode the struct file belongs to 605 * @file: the struct file being closed 606 * 607 * Returns: errno 608 */ 609 610 static int gfs2_release(struct inode *inode, struct file *file) 611 { 612 struct gfs2_inode *ip = GFS2_I(inode); 613 614 kfree(file->private_data); 615 file->private_data = NULL; 616 617 if (!(file->f_mode & FMODE_WRITE)) 618 return 0; 619 620 gfs2_rsqa_delete(ip, &inode->i_writecount); 621 return 0; 622 } 623 624 /** 625 * gfs2_fsync - sync the dirty data for a file (across the cluster) 626 * @file: the file that points to the dentry 627 * @start: the start position in the file to sync 628 * @end: the end position in the file to sync 629 * @datasync: set if we can ignore timestamp changes 630 * 631 * We split the data flushing here so that we don't wait for the data 632 * until after we've also sent the metadata to disk. Note that for 633 * data=ordered, we will write & wait for the data at the log flush 634 * stage anyway, so this is unlikely to make much of a difference 635 * except in the data=writeback case. 636 * 637 * If the fdatawrite fails due to any reason except -EIO, we will 638 * continue the remainder of the fsync, although we'll still report 639 * the error at the end. This is to match filemap_write_and_wait_range() 640 * behaviour. 641 * 642 * Returns: errno 643 */ 644 645 static int gfs2_fsync(struct file *file, loff_t start, loff_t end, 646 int datasync) 647 { 648 struct address_space *mapping = file->f_mapping; 649 struct inode *inode = mapping->host; 650 int sync_state = inode->i_state & I_DIRTY_ALL; 651 struct gfs2_inode *ip = GFS2_I(inode); 652 int ret = 0, ret1 = 0; 653 654 if (mapping->nrpages) { 655 ret1 = filemap_fdatawrite_range(mapping, start, end); 656 if (ret1 == -EIO) 657 return ret1; 658 } 659 660 if (!gfs2_is_jdata(ip)) 661 sync_state &= ~I_DIRTY_PAGES; 662 if (datasync) 663 sync_state &= ~(I_DIRTY_SYNC | I_DIRTY_TIME); 664 665 if (sync_state) { 666 ret = sync_inode_metadata(inode, 1); 667 if (ret) 668 return ret; 669 if (gfs2_is_jdata(ip)) 670 filemap_write_and_wait(mapping); 671 gfs2_ail_flush(ip->i_gl, 1); 672 } 673 674 if (mapping->nrpages) 675 ret = filemap_fdatawait_range(mapping, start, end); 676 677 return ret ? ret : ret1; 678 } 679 680 /** 681 * gfs2_file_write_iter - Perform a write to a file 682 * @iocb: The io context 683 * @iov: The data to write 684 * @nr_segs: Number of @iov segments 685 * @pos: The file position 686 * 687 * We have to do a lock/unlock here to refresh the inode size for 688 * O_APPEND writes, otherwise we can land up writing at the wrong 689 * offset. There is still a race, but provided the app is using its 690 * own file locking, this will make O_APPEND work as expected. 691 * 692 */ 693 694 static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from) 695 { 696 struct file *file = iocb->ki_filp; 697 struct gfs2_inode *ip = GFS2_I(file_inode(file)); 698 int ret; 699 700 ret = gfs2_rsqa_alloc(ip); 701 if (ret) 702 return ret; 703 704 gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from)); 705 706 if (iocb->ki_flags & IOCB_APPEND) { 707 struct gfs2_holder gh; 708 709 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh); 710 if (ret) 711 return ret; 712 gfs2_glock_dq_uninit(&gh); 713 } 714 715 return generic_file_write_iter(iocb, from); 716 } 717 718 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len, 719 int mode) 720 { 721 struct gfs2_inode *ip = GFS2_I(inode); 722 struct buffer_head *dibh; 723 int error; 724 unsigned int nr_blks; 725 sector_t lblock = offset >> inode->i_blkbits; 726 727 error = gfs2_meta_inode_buffer(ip, &dibh); 728 if (unlikely(error)) 729 return error; 730 731 gfs2_trans_add_meta(ip->i_gl, dibh); 732 733 if (gfs2_is_stuffed(ip)) { 734 error = gfs2_unstuff_dinode(ip, NULL); 735 if (unlikely(error)) 736 goto out; 737 } 738 739 while (len) { 740 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 }; 741 bh_map.b_size = len; 742 set_buffer_zeronew(&bh_map); 743 744 error = gfs2_block_map(inode, lblock, &bh_map, 1); 745 if (unlikely(error)) 746 goto out; 747 len -= bh_map.b_size; 748 nr_blks = bh_map.b_size >> inode->i_blkbits; 749 lblock += nr_blks; 750 if (!buffer_new(&bh_map)) 751 continue; 752 if (unlikely(!buffer_zeronew(&bh_map))) { 753 error = -EIO; 754 goto out; 755 } 756 } 757 out: 758 brelse(dibh); 759 return error; 760 } 761 /** 762 * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of 763 * blocks, determine how many bytes can be written. 764 * @ip: The inode in question. 765 * @len: Max cap of bytes. What we return in *len must be <= this. 766 * @data_blocks: Compute and return the number of data blocks needed 767 * @ind_blocks: Compute and return the number of indirect blocks needed 768 * @max_blocks: The total blocks available to work with. 769 * 770 * Returns: void, but @len, @data_blocks and @ind_blocks are filled in. 771 */ 772 static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len, 773 unsigned int *data_blocks, unsigned int *ind_blocks, 774 unsigned int max_blocks) 775 { 776 loff_t max = *len; 777 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 778 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1); 779 780 for (tmp = max_data; tmp > sdp->sd_diptrs;) { 781 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs); 782 max_data -= tmp; 783 } 784 785 *data_blocks = max_data; 786 *ind_blocks = max_blocks - max_data; 787 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift; 788 if (*len > max) { 789 *len = max; 790 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks); 791 } 792 } 793 794 static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len) 795 { 796 struct inode *inode = file_inode(file); 797 struct gfs2_sbd *sdp = GFS2_SB(inode); 798 struct gfs2_inode *ip = GFS2_I(inode); 799 struct gfs2_alloc_parms ap = { .aflags = 0, }; 800 unsigned int data_blocks = 0, ind_blocks = 0, rblocks; 801 loff_t bytes, max_bytes, max_blks = UINT_MAX; 802 int error; 803 const loff_t pos = offset; 804 const loff_t count = len; 805 loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1); 806 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift; 807 loff_t max_chunk_size = UINT_MAX & bsize_mask; 808 809 next = (next + 1) << sdp->sd_sb.sb_bsize_shift; 810 811 offset &= bsize_mask; 812 813 len = next - offset; 814 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2; 815 if (!bytes) 816 bytes = UINT_MAX; 817 bytes &= bsize_mask; 818 if (bytes == 0) 819 bytes = sdp->sd_sb.sb_bsize; 820 821 gfs2_size_hint(file, offset, len); 822 823 gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks); 824 ap.min_target = data_blocks + ind_blocks; 825 826 while (len > 0) { 827 if (len < bytes) 828 bytes = len; 829 if (!gfs2_write_alloc_required(ip, offset, bytes)) { 830 len -= bytes; 831 offset += bytes; 832 continue; 833 } 834 835 /* We need to determine how many bytes we can actually 836 * fallocate without exceeding quota or going over the 837 * end of the fs. We start off optimistically by assuming 838 * we can write max_bytes */ 839 max_bytes = (len > max_chunk_size) ? max_chunk_size : len; 840 841 /* Since max_bytes is most likely a theoretical max, we 842 * calculate a more realistic 'bytes' to serve as a good 843 * starting point for the number of bytes we may be able 844 * to write */ 845 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks); 846 ap.target = data_blocks + ind_blocks; 847 848 error = gfs2_quota_lock_check(ip, &ap); 849 if (error) 850 return error; 851 /* ap.allowed tells us how many blocks quota will allow 852 * us to write. Check if this reduces max_blks */ 853 if (ap.allowed && ap.allowed < max_blks) 854 max_blks = ap.allowed; 855 856 error = gfs2_inplace_reserve(ip, &ap); 857 if (error) 858 goto out_qunlock; 859 860 /* check if the selected rgrp limits our max_blks further */ 861 if (ap.allowed && ap.allowed < max_blks) 862 max_blks = ap.allowed; 863 864 /* Almost done. Calculate bytes that can be written using 865 * max_blks. We also recompute max_bytes, data_blocks and 866 * ind_blocks */ 867 calc_max_reserv(ip, &max_bytes, &data_blocks, 868 &ind_blocks, max_blks); 869 870 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA + 871 RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks); 872 if (gfs2_is_jdata(ip)) 873 rblocks += data_blocks ? data_blocks : 1; 874 875 error = gfs2_trans_begin(sdp, rblocks, 876 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize); 877 if (error) 878 goto out_trans_fail; 879 880 error = fallocate_chunk(inode, offset, max_bytes, mode); 881 gfs2_trans_end(sdp); 882 883 if (error) 884 goto out_trans_fail; 885 886 len -= max_bytes; 887 offset += max_bytes; 888 gfs2_inplace_release(ip); 889 gfs2_quota_unlock(ip); 890 } 891 892 if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size) { 893 i_size_write(inode, pos + count); 894 file_update_time(file); 895 mark_inode_dirty(inode); 896 } 897 898 return generic_write_sync(file, pos, count); 899 900 out_trans_fail: 901 gfs2_inplace_release(ip); 902 out_qunlock: 903 gfs2_quota_unlock(ip); 904 return error; 905 } 906 907 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len) 908 { 909 struct inode *inode = file_inode(file); 910 struct gfs2_inode *ip = GFS2_I(inode); 911 struct gfs2_holder gh; 912 int ret; 913 914 if ((mode & ~FALLOC_FL_KEEP_SIZE) || gfs2_is_jdata(ip)) 915 return -EOPNOTSUPP; 916 917 inode_lock(inode); 918 919 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 920 ret = gfs2_glock_nq(&gh); 921 if (ret) 922 goto out_uninit; 923 924 if (!(mode & FALLOC_FL_KEEP_SIZE) && 925 (offset + len) > inode->i_size) { 926 ret = inode_newsize_ok(inode, offset + len); 927 if (ret) 928 goto out_unlock; 929 } 930 931 ret = get_write_access(inode); 932 if (ret) 933 goto out_unlock; 934 935 ret = gfs2_rsqa_alloc(ip); 936 if (ret) 937 goto out_putw; 938 939 ret = __gfs2_fallocate(file, mode, offset, len); 940 if (ret) 941 gfs2_rs_deltree(&ip->i_res); 942 943 out_putw: 944 put_write_access(inode); 945 out_unlock: 946 gfs2_glock_dq(&gh); 947 out_uninit: 948 gfs2_holder_uninit(&gh); 949 inode_unlock(inode); 950 return ret; 951 } 952 953 static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe, 954 struct file *out, loff_t *ppos, 955 size_t len, unsigned int flags) 956 { 957 int error; 958 struct gfs2_inode *ip = GFS2_I(out->f_mapping->host); 959 960 error = gfs2_rsqa_alloc(ip); 961 if (error) 962 return (ssize_t)error; 963 964 gfs2_size_hint(out, *ppos, len); 965 966 return iter_file_splice_write(pipe, out, ppos, len, flags); 967 } 968 969 #ifdef CONFIG_GFS2_FS_LOCKING_DLM 970 971 /** 972 * gfs2_lock - acquire/release a posix lock on a file 973 * @file: the file pointer 974 * @cmd: either modify or retrieve lock state, possibly wait 975 * @fl: type and range of lock 976 * 977 * Returns: errno 978 */ 979 980 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl) 981 { 982 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 983 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host); 984 struct lm_lockstruct *ls = &sdp->sd_lockstruct; 985 986 if (!(fl->fl_flags & FL_POSIX)) 987 return -ENOLCK; 988 if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK) 989 return -ENOLCK; 990 991 if (cmd == F_CANCELLK) { 992 /* Hack: */ 993 cmd = F_SETLK; 994 fl->fl_type = F_UNLCK; 995 } 996 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) { 997 if (fl->fl_type == F_UNLCK) 998 locks_lock_file_wait(file, fl); 999 return -EIO; 1000 } 1001 if (IS_GETLK(cmd)) 1002 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl); 1003 else if (fl->fl_type == F_UNLCK) 1004 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl); 1005 else 1006 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl); 1007 } 1008 1009 static int do_flock(struct file *file, int cmd, struct file_lock *fl) 1010 { 1011 struct gfs2_file *fp = file->private_data; 1012 struct gfs2_holder *fl_gh = &fp->f_fl_gh; 1013 struct gfs2_inode *ip = GFS2_I(file_inode(file)); 1014 struct gfs2_glock *gl; 1015 unsigned int state; 1016 u16 flags; 1017 int error = 0; 1018 int sleeptime; 1019 1020 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED; 1021 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT; 1022 1023 mutex_lock(&fp->f_fl_mutex); 1024 1025 gl = fl_gh->gh_gl; 1026 if (gl) { 1027 if (fl_gh->gh_state == state) 1028 goto out; 1029 locks_lock_file_wait(file, 1030 &(struct file_lock) { 1031 .fl_type = F_UNLCK, 1032 .fl_flags = FL_FLOCK 1033 }); 1034 gfs2_glock_dq(fl_gh); 1035 gfs2_holder_reinit(state, flags, fl_gh); 1036 } else { 1037 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr, 1038 &gfs2_flock_glops, CREATE, &gl); 1039 if (error) 1040 goto out; 1041 gfs2_holder_init(gl, state, flags, fl_gh); 1042 gfs2_glock_put(gl); 1043 } 1044 for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) { 1045 error = gfs2_glock_nq(fl_gh); 1046 if (error != GLR_TRYFAILED) 1047 break; 1048 fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT; 1049 fl_gh->gh_error = 0; 1050 msleep(sleeptime); 1051 } 1052 if (error) { 1053 gfs2_holder_uninit(fl_gh); 1054 if (error == GLR_TRYFAILED) 1055 error = -EAGAIN; 1056 } else { 1057 error = locks_lock_file_wait(file, fl); 1058 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); 1059 } 1060 1061 out: 1062 mutex_unlock(&fp->f_fl_mutex); 1063 return error; 1064 } 1065 1066 static void do_unflock(struct file *file, struct file_lock *fl) 1067 { 1068 struct gfs2_file *fp = file->private_data; 1069 struct gfs2_holder *fl_gh = &fp->f_fl_gh; 1070 1071 mutex_lock(&fp->f_fl_mutex); 1072 locks_lock_file_wait(file, fl); 1073 if (fl_gh->gh_gl) { 1074 gfs2_glock_dq(fl_gh); 1075 gfs2_holder_uninit(fl_gh); 1076 } 1077 mutex_unlock(&fp->f_fl_mutex); 1078 } 1079 1080 /** 1081 * gfs2_flock - acquire/release a flock lock on a file 1082 * @file: the file pointer 1083 * @cmd: either modify or retrieve lock state, possibly wait 1084 * @fl: type and range of lock 1085 * 1086 * Returns: errno 1087 */ 1088 1089 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl) 1090 { 1091 if (!(fl->fl_flags & FL_FLOCK)) 1092 return -ENOLCK; 1093 if (fl->fl_type & LOCK_MAND) 1094 return -EOPNOTSUPP; 1095 1096 if (fl->fl_type == F_UNLCK) { 1097 do_unflock(file, fl); 1098 return 0; 1099 } else { 1100 return do_flock(file, cmd, fl); 1101 } 1102 } 1103 1104 const struct file_operations gfs2_file_fops = { 1105 .llseek = gfs2_llseek, 1106 .read_iter = generic_file_read_iter, 1107 .write_iter = gfs2_file_write_iter, 1108 .unlocked_ioctl = gfs2_ioctl, 1109 .mmap = gfs2_mmap, 1110 .open = gfs2_open, 1111 .release = gfs2_release, 1112 .fsync = gfs2_fsync, 1113 .lock = gfs2_lock, 1114 .flock = gfs2_flock, 1115 .splice_read = generic_file_splice_read, 1116 .splice_write = gfs2_file_splice_write, 1117 .setlease = simple_nosetlease, 1118 .fallocate = gfs2_fallocate, 1119 }; 1120 1121 const struct file_operations gfs2_dir_fops = { 1122 .iterate = gfs2_readdir, 1123 .unlocked_ioctl = gfs2_ioctl, 1124 .open = gfs2_open, 1125 .release = gfs2_release, 1126 .fsync = gfs2_fsync, 1127 .lock = gfs2_lock, 1128 .flock = gfs2_flock, 1129 .llseek = default_llseek, 1130 }; 1131 1132 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */ 1133 1134 const struct file_operations gfs2_file_fops_nolock = { 1135 .llseek = gfs2_llseek, 1136 .read_iter = generic_file_read_iter, 1137 .write_iter = gfs2_file_write_iter, 1138 .unlocked_ioctl = gfs2_ioctl, 1139 .mmap = gfs2_mmap, 1140 .open = gfs2_open, 1141 .release = gfs2_release, 1142 .fsync = gfs2_fsync, 1143 .splice_read = generic_file_splice_read, 1144 .splice_write = gfs2_file_splice_write, 1145 .setlease = generic_setlease, 1146 .fallocate = gfs2_fallocate, 1147 }; 1148 1149 const struct file_operations gfs2_dir_fops_nolock = { 1150 .iterate = gfs2_readdir, 1151 .unlocked_ioctl = gfs2_ioctl, 1152 .open = gfs2_open, 1153 .release = gfs2_release, 1154 .fsync = gfs2_fsync, 1155 .llseek = default_llseek, 1156 }; 1157 1158