1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 4 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. 5 */ 6 7 #include <linux/slab.h> 8 #include <linux/spinlock.h> 9 #include <linux/completion.h> 10 #include <linux/buffer_head.h> 11 #include <linux/pagemap.h> 12 #include <linux/uio.h> 13 #include <linux/blkdev.h> 14 #include <linux/mm.h> 15 #include <linux/mount.h> 16 #include <linux/fs.h> 17 #include <linux/gfs2_ondisk.h> 18 #include <linux/falloc.h> 19 #include <linux/swap.h> 20 #include <linux/crc32.h> 21 #include <linux/writeback.h> 22 #include <linux/uaccess.h> 23 #include <linux/dlm.h> 24 #include <linux/dlm_plock.h> 25 #include <linux/delay.h> 26 #include <linux/backing-dev.h> 27 28 #include "gfs2.h" 29 #include "incore.h" 30 #include "bmap.h" 31 #include "aops.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 * @whence: 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 whence) 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 (whence) { 62 case SEEK_END: 63 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, 64 &i_gh); 65 if (!error) { 66 error = generic_file_llseek(file, offset, whence); 67 gfs2_glock_dq_uninit(&i_gh); 68 } 69 break; 70 71 case SEEK_DATA: 72 error = gfs2_seek_data(file, offset); 73 break; 74 75 case SEEK_HOLE: 76 error = gfs2_seek_hole(file, offset); 77 break; 78 79 case SEEK_CUR: 80 case SEEK_SET: 81 /* 82 * These don't reference inode->i_size and don't depend on the 83 * block mapping, so we don't need the glock. 84 */ 85 error = generic_file_llseek(file, offset, whence); 86 break; 87 default: 88 error = -EINVAL; 89 } 90 91 return error; 92 } 93 94 /** 95 * gfs2_readdir - Iterator for a directory 96 * @file: The directory to read from 97 * @ctx: What to feed directory entries to 98 * 99 * Returns: errno 100 */ 101 102 static int gfs2_readdir(struct file *file, struct dir_context *ctx) 103 { 104 struct inode *dir = file->f_mapping->host; 105 struct gfs2_inode *dip = GFS2_I(dir); 106 struct gfs2_holder d_gh; 107 int error; 108 109 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh); 110 if (error) 111 return error; 112 113 error = gfs2_dir_read(dir, ctx, &file->f_ra); 114 115 gfs2_glock_dq_uninit(&d_gh); 116 117 return error; 118 } 119 120 /** 121 * fsflag_gfs2flag 122 * 123 * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories, 124 * and to GFS2_DIF_JDATA for non-directories. 125 */ 126 static struct { 127 u32 fsflag; 128 u32 gfsflag; 129 } fsflag_gfs2flag[] = { 130 {FS_SYNC_FL, GFS2_DIF_SYNC}, 131 {FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE}, 132 {FS_APPEND_FL, GFS2_DIF_APPENDONLY}, 133 {FS_NOATIME_FL, GFS2_DIF_NOATIME}, 134 {FS_INDEX_FL, GFS2_DIF_EXHASH}, 135 {FS_TOPDIR_FL, GFS2_DIF_TOPDIR}, 136 {FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA}, 137 }; 138 139 static int gfs2_get_flags(struct file *filp, u32 __user *ptr) 140 { 141 struct inode *inode = file_inode(filp); 142 struct gfs2_inode *ip = GFS2_I(inode); 143 struct gfs2_holder gh; 144 int i, error; 145 u32 gfsflags, fsflags = 0; 146 147 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); 148 error = gfs2_glock_nq(&gh); 149 if (error) 150 goto out_uninit; 151 152 gfsflags = ip->i_diskflags; 153 if (S_ISDIR(inode->i_mode)) 154 gfsflags &= ~GFS2_DIF_JDATA; 155 else 156 gfsflags &= ~GFS2_DIF_INHERIT_JDATA; 157 for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) 158 if (gfsflags & fsflag_gfs2flag[i].gfsflag) 159 fsflags |= fsflag_gfs2flag[i].fsflag; 160 161 if (put_user(fsflags, ptr)) 162 error = -EFAULT; 163 164 gfs2_glock_dq(&gh); 165 out_uninit: 166 gfs2_holder_uninit(&gh); 167 return error; 168 } 169 170 void gfs2_set_inode_flags(struct inode *inode) 171 { 172 struct gfs2_inode *ip = GFS2_I(inode); 173 unsigned int flags = inode->i_flags; 174 175 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC); 176 if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode)) 177 flags |= S_NOSEC; 178 if (ip->i_diskflags & GFS2_DIF_IMMUTABLE) 179 flags |= S_IMMUTABLE; 180 if (ip->i_diskflags & GFS2_DIF_APPENDONLY) 181 flags |= S_APPEND; 182 if (ip->i_diskflags & GFS2_DIF_NOATIME) 183 flags |= S_NOATIME; 184 if (ip->i_diskflags & GFS2_DIF_SYNC) 185 flags |= S_SYNC; 186 inode->i_flags = flags; 187 } 188 189 /* Flags that can be set by user space */ 190 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \ 191 GFS2_DIF_IMMUTABLE| \ 192 GFS2_DIF_APPENDONLY| \ 193 GFS2_DIF_NOATIME| \ 194 GFS2_DIF_SYNC| \ 195 GFS2_DIF_TOPDIR| \ 196 GFS2_DIF_INHERIT_JDATA) 197 198 /** 199 * do_gfs2_set_flags - set flags on an inode 200 * @filp: file pointer 201 * @reqflags: The flags to set 202 * @mask: Indicates which flags are valid 203 * 204 */ 205 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask) 206 { 207 struct inode *inode = file_inode(filp); 208 struct gfs2_inode *ip = GFS2_I(inode); 209 struct gfs2_sbd *sdp = GFS2_SB(inode); 210 struct buffer_head *bh; 211 struct gfs2_holder gh; 212 int error; 213 u32 new_flags, flags; 214 215 error = mnt_want_write_file(filp); 216 if (error) 217 return error; 218 219 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 220 if (error) 221 goto out_drop_write; 222 223 error = -EACCES; 224 if (!inode_owner_or_capable(inode)) 225 goto out; 226 227 error = 0; 228 flags = ip->i_diskflags; 229 new_flags = (flags & ~mask) | (reqflags & mask); 230 if ((new_flags ^ flags) == 0) 231 goto out; 232 233 error = -EPERM; 234 if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE)) 235 goto out; 236 if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY)) 237 goto out; 238 if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) && 239 !capable(CAP_LINUX_IMMUTABLE)) 240 goto out; 241 if (!IS_IMMUTABLE(inode)) { 242 error = gfs2_permission(inode, MAY_WRITE); 243 if (error) 244 goto out; 245 } 246 if ((flags ^ new_flags) & GFS2_DIF_JDATA) { 247 if (new_flags & GFS2_DIF_JDATA) 248 gfs2_log_flush(sdp, ip->i_gl, 249 GFS2_LOG_HEAD_FLUSH_NORMAL | 250 GFS2_LFC_SET_FLAGS); 251 error = filemap_fdatawrite(inode->i_mapping); 252 if (error) 253 goto out; 254 error = filemap_fdatawait(inode->i_mapping); 255 if (error) 256 goto out; 257 if (new_flags & GFS2_DIF_JDATA) 258 gfs2_ordered_del_inode(ip); 259 } 260 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 261 if (error) 262 goto out; 263 error = gfs2_meta_inode_buffer(ip, &bh); 264 if (error) 265 goto out_trans_end; 266 inode->i_ctime = current_time(inode); 267 gfs2_trans_add_meta(ip->i_gl, bh); 268 ip->i_diskflags = new_flags; 269 gfs2_dinode_out(ip, bh->b_data); 270 brelse(bh); 271 gfs2_set_inode_flags(inode); 272 gfs2_set_aops(inode); 273 out_trans_end: 274 gfs2_trans_end(sdp); 275 out: 276 gfs2_glock_dq_uninit(&gh); 277 out_drop_write: 278 mnt_drop_write_file(filp); 279 return error; 280 } 281 282 static int gfs2_set_flags(struct file *filp, u32 __user *ptr) 283 { 284 struct inode *inode = file_inode(filp); 285 u32 fsflags, gfsflags = 0; 286 u32 mask; 287 int i; 288 289 if (get_user(fsflags, ptr)) 290 return -EFAULT; 291 292 for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) { 293 if (fsflags & fsflag_gfs2flag[i].fsflag) { 294 fsflags &= ~fsflag_gfs2flag[i].fsflag; 295 gfsflags |= fsflag_gfs2flag[i].gfsflag; 296 } 297 } 298 if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET) 299 return -EINVAL; 300 301 mask = GFS2_FLAGS_USER_SET; 302 if (S_ISDIR(inode->i_mode)) { 303 mask &= ~GFS2_DIF_JDATA; 304 } else { 305 /* The GFS2_DIF_TOPDIR flag is only valid for directories. */ 306 if (gfsflags & GFS2_DIF_TOPDIR) 307 return -EINVAL; 308 mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA); 309 } 310 311 return do_gfs2_set_flags(filp, gfsflags, mask); 312 } 313 314 static int gfs2_getlabel(struct file *filp, char __user *label) 315 { 316 struct inode *inode = file_inode(filp); 317 struct gfs2_sbd *sdp = GFS2_SB(inode); 318 319 if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN)) 320 return -EFAULT; 321 322 return 0; 323 } 324 325 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) 326 { 327 switch(cmd) { 328 case FS_IOC_GETFLAGS: 329 return gfs2_get_flags(filp, (u32 __user *)arg); 330 case FS_IOC_SETFLAGS: 331 return gfs2_set_flags(filp, (u32 __user *)arg); 332 case FITRIM: 333 return gfs2_fitrim(filp, (void __user *)arg); 334 case FS_IOC_GETFSLABEL: 335 return gfs2_getlabel(filp, (char __user *)arg); 336 } 337 338 return -ENOTTY; 339 } 340 341 /** 342 * gfs2_size_hint - Give a hint to the size of a write request 343 * @filep: The struct file 344 * @offset: The file offset of the write 345 * @size: The length of the write 346 * 347 * When we are about to do a write, this function records the total 348 * write size in order to provide a suitable hint to the lower layers 349 * about how many blocks will be required. 350 * 351 */ 352 353 static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size) 354 { 355 struct inode *inode = file_inode(filep); 356 struct gfs2_sbd *sdp = GFS2_SB(inode); 357 struct gfs2_inode *ip = GFS2_I(inode); 358 size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift; 359 int hint = min_t(size_t, INT_MAX, blks); 360 361 if (hint > atomic_read(&ip->i_sizehint)) 362 atomic_set(&ip->i_sizehint, hint); 363 } 364 365 /** 366 * gfs2_allocate_page_backing - Use bmap to allocate blocks 367 * @page: The (locked) page to allocate backing for 368 * 369 * We try to allocate all the blocks required for the page in 370 * one go. This might fail for various reasons, so we keep 371 * trying until all the blocks to back this page are allocated. 372 * If some of the blocks are already allocated, thats ok too. 373 */ 374 375 static int gfs2_allocate_page_backing(struct page *page) 376 { 377 struct inode *inode = page->mapping->host; 378 struct buffer_head bh; 379 unsigned long size = PAGE_SIZE; 380 u64 lblock = page->index << (PAGE_SHIFT - inode->i_blkbits); 381 382 do { 383 bh.b_state = 0; 384 bh.b_size = size; 385 gfs2_block_map(inode, lblock, &bh, 1); 386 if (!buffer_mapped(&bh)) 387 return -EIO; 388 size -= bh.b_size; 389 lblock += (bh.b_size >> inode->i_blkbits); 390 } while(size > 0); 391 return 0; 392 } 393 394 /** 395 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable 396 * @vma: The virtual memory area 397 * @vmf: The virtual memory fault containing the page to become writable 398 * 399 * When the page becomes writable, we need to ensure that we have 400 * blocks allocated on disk to back that page. 401 */ 402 403 static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf) 404 { 405 struct page *page = vmf->page; 406 struct inode *inode = file_inode(vmf->vma->vm_file); 407 struct gfs2_inode *ip = GFS2_I(inode); 408 struct gfs2_sbd *sdp = GFS2_SB(inode); 409 struct gfs2_alloc_parms ap = { .aflags = 0, }; 410 unsigned long last_index; 411 u64 pos = page->index << PAGE_SHIFT; 412 unsigned int data_blocks, ind_blocks, rblocks; 413 struct gfs2_holder gh; 414 loff_t size; 415 int ret; 416 417 sb_start_pagefault(inode->i_sb); 418 419 ret = gfs2_rsqa_alloc(ip); 420 if (ret) 421 goto out; 422 423 gfs2_size_hint(vmf->vma->vm_file, pos, PAGE_SIZE); 424 425 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 426 ret = gfs2_glock_nq(&gh); 427 if (ret) 428 goto out_uninit; 429 430 /* Update file times before taking page lock */ 431 file_update_time(vmf->vma->vm_file); 432 433 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags); 434 set_bit(GIF_SW_PAGED, &ip->i_flags); 435 436 if (!gfs2_write_alloc_required(ip, pos, PAGE_SIZE)) { 437 lock_page(page); 438 if (!PageUptodate(page) || page->mapping != inode->i_mapping) { 439 ret = -EAGAIN; 440 unlock_page(page); 441 } 442 goto out_unlock; 443 } 444 445 ret = gfs2_rindex_update(sdp); 446 if (ret) 447 goto out_unlock; 448 449 gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks); 450 ap.target = data_blocks + ind_blocks; 451 ret = gfs2_quota_lock_check(ip, &ap); 452 if (ret) 453 goto out_unlock; 454 ret = gfs2_inplace_reserve(ip, &ap); 455 if (ret) 456 goto out_quota_unlock; 457 458 rblocks = RES_DINODE + ind_blocks; 459 if (gfs2_is_jdata(ip)) 460 rblocks += data_blocks ? data_blocks : 1; 461 if (ind_blocks || data_blocks) { 462 rblocks += RES_STATFS + RES_QUOTA; 463 rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks); 464 } 465 ret = gfs2_trans_begin(sdp, rblocks, 0); 466 if (ret) 467 goto out_trans_fail; 468 469 lock_page(page); 470 ret = -EINVAL; 471 size = i_size_read(inode); 472 last_index = (size - 1) >> PAGE_SHIFT; 473 /* Check page index against inode size */ 474 if (size == 0 || (page->index > last_index)) 475 goto out_trans_end; 476 477 ret = -EAGAIN; 478 /* If truncated, we must retry the operation, we may have raced 479 * with the glock demotion code. 480 */ 481 if (!PageUptodate(page) || page->mapping != inode->i_mapping) 482 goto out_trans_end; 483 484 /* Unstuff, if required, and allocate backing blocks for page */ 485 ret = 0; 486 if (gfs2_is_stuffed(ip)) 487 ret = gfs2_unstuff_dinode(ip, page); 488 if (ret == 0) 489 ret = gfs2_allocate_page_backing(page); 490 491 out_trans_end: 492 if (ret) 493 unlock_page(page); 494 gfs2_trans_end(sdp); 495 out_trans_fail: 496 gfs2_inplace_release(ip); 497 out_quota_unlock: 498 gfs2_quota_unlock(ip); 499 out_unlock: 500 gfs2_glock_dq(&gh); 501 out_uninit: 502 gfs2_holder_uninit(&gh); 503 if (ret == 0) { 504 set_page_dirty(page); 505 wait_for_stable_page(page); 506 } 507 out: 508 sb_end_pagefault(inode->i_sb); 509 return block_page_mkwrite_return(ret); 510 } 511 512 static const struct vm_operations_struct gfs2_vm_ops = { 513 .fault = filemap_fault, 514 .map_pages = filemap_map_pages, 515 .page_mkwrite = gfs2_page_mkwrite, 516 }; 517 518 /** 519 * gfs2_mmap - 520 * @file: The file to map 521 * @vma: The VMA which described the mapping 522 * 523 * There is no need to get a lock here unless we should be updating 524 * atime. We ignore any locking errors since the only consequence is 525 * a missed atime update (which will just be deferred until later). 526 * 527 * Returns: 0 528 */ 529 530 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma) 531 { 532 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 533 534 if (!(file->f_flags & O_NOATIME) && 535 !IS_NOATIME(&ip->i_inode)) { 536 struct gfs2_holder i_gh; 537 int error; 538 539 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, 540 &i_gh); 541 if (error) 542 return error; 543 /* grab lock to update inode */ 544 gfs2_glock_dq_uninit(&i_gh); 545 file_accessed(file); 546 } 547 vma->vm_ops = &gfs2_vm_ops; 548 549 return 0; 550 } 551 552 /** 553 * gfs2_open_common - This is common to open and atomic_open 554 * @inode: The inode being opened 555 * @file: The file being opened 556 * 557 * This maybe called under a glock or not depending upon how it has 558 * been called. We must always be called under a glock for regular 559 * files, however. For other file types, it does not matter whether 560 * we hold the glock or not. 561 * 562 * Returns: Error code or 0 for success 563 */ 564 565 int gfs2_open_common(struct inode *inode, struct file *file) 566 { 567 struct gfs2_file *fp; 568 int ret; 569 570 if (S_ISREG(inode->i_mode)) { 571 ret = generic_file_open(inode, file); 572 if (ret) 573 return ret; 574 } 575 576 fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS); 577 if (!fp) 578 return -ENOMEM; 579 580 mutex_init(&fp->f_fl_mutex); 581 582 gfs2_assert_warn(GFS2_SB(inode), !file->private_data); 583 file->private_data = fp; 584 return 0; 585 } 586 587 /** 588 * gfs2_open - open a file 589 * @inode: the inode to open 590 * @file: the struct file for this opening 591 * 592 * After atomic_open, this function is only used for opening files 593 * which are already cached. We must still get the glock for regular 594 * files to ensure that we have the file size uptodate for the large 595 * file check which is in the common code. That is only an issue for 596 * regular files though. 597 * 598 * Returns: errno 599 */ 600 601 static int gfs2_open(struct inode *inode, struct file *file) 602 { 603 struct gfs2_inode *ip = GFS2_I(inode); 604 struct gfs2_holder i_gh; 605 int error; 606 bool need_unlock = false; 607 608 if (S_ISREG(ip->i_inode.i_mode)) { 609 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, 610 &i_gh); 611 if (error) 612 return error; 613 need_unlock = true; 614 } 615 616 error = gfs2_open_common(inode, file); 617 618 if (need_unlock) 619 gfs2_glock_dq_uninit(&i_gh); 620 621 return error; 622 } 623 624 /** 625 * gfs2_release - called to close a struct file 626 * @inode: the inode the struct file belongs to 627 * @file: the struct file being closed 628 * 629 * Returns: errno 630 */ 631 632 static int gfs2_release(struct inode *inode, struct file *file) 633 { 634 struct gfs2_inode *ip = GFS2_I(inode); 635 636 kfree(file->private_data); 637 file->private_data = NULL; 638 639 if (!(file->f_mode & FMODE_WRITE)) 640 return 0; 641 642 gfs2_rsqa_delete(ip, &inode->i_writecount); 643 return 0; 644 } 645 646 /** 647 * gfs2_fsync - sync the dirty data for a file (across the cluster) 648 * @file: the file that points to the dentry 649 * @start: the start position in the file to sync 650 * @end: the end position in the file to sync 651 * @datasync: set if we can ignore timestamp changes 652 * 653 * We split the data flushing here so that we don't wait for the data 654 * until after we've also sent the metadata to disk. Note that for 655 * data=ordered, we will write & wait for the data at the log flush 656 * stage anyway, so this is unlikely to make much of a difference 657 * except in the data=writeback case. 658 * 659 * If the fdatawrite fails due to any reason except -EIO, we will 660 * continue the remainder of the fsync, although we'll still report 661 * the error at the end. This is to match filemap_write_and_wait_range() 662 * behaviour. 663 * 664 * Returns: errno 665 */ 666 667 static int gfs2_fsync(struct file *file, loff_t start, loff_t end, 668 int datasync) 669 { 670 struct address_space *mapping = file->f_mapping; 671 struct inode *inode = mapping->host; 672 int sync_state = inode->i_state & I_DIRTY_ALL; 673 struct gfs2_inode *ip = GFS2_I(inode); 674 int ret = 0, ret1 = 0; 675 676 if (mapping->nrpages) { 677 ret1 = filemap_fdatawrite_range(mapping, start, end); 678 if (ret1 == -EIO) 679 return ret1; 680 } 681 682 if (!gfs2_is_jdata(ip)) 683 sync_state &= ~I_DIRTY_PAGES; 684 if (datasync) 685 sync_state &= ~(I_DIRTY_SYNC | I_DIRTY_TIME); 686 687 if (sync_state) { 688 ret = sync_inode_metadata(inode, 1); 689 if (ret) 690 return ret; 691 if (gfs2_is_jdata(ip)) 692 ret = file_write_and_wait(file); 693 if (ret) 694 return ret; 695 gfs2_ail_flush(ip->i_gl, 1); 696 } 697 698 if (mapping->nrpages) 699 ret = file_fdatawait_range(file, start, end); 700 701 return ret ? ret : ret1; 702 } 703 704 static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to) 705 { 706 struct file *file = iocb->ki_filp; 707 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 708 size_t count = iov_iter_count(to); 709 struct gfs2_holder gh; 710 ssize_t ret; 711 712 if (!count) 713 return 0; /* skip atime */ 714 715 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh); 716 ret = gfs2_glock_nq(&gh); 717 if (ret) 718 goto out_uninit; 719 720 ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL); 721 722 gfs2_glock_dq(&gh); 723 out_uninit: 724 gfs2_holder_uninit(&gh); 725 return ret; 726 } 727 728 static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from) 729 { 730 struct file *file = iocb->ki_filp; 731 struct inode *inode = file->f_mapping->host; 732 struct gfs2_inode *ip = GFS2_I(inode); 733 size_t len = iov_iter_count(from); 734 loff_t offset = iocb->ki_pos; 735 struct gfs2_holder gh; 736 ssize_t ret; 737 738 /* 739 * Deferred lock, even if its a write, since we do no allocation on 740 * this path. All we need to change is the atime, and this lock mode 741 * ensures that other nodes have flushed their buffered read caches 742 * (i.e. their page cache entries for this inode). We do not, 743 * unfortunately, have the option of only flushing a range like the 744 * VFS does. 745 */ 746 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh); 747 ret = gfs2_glock_nq(&gh); 748 if (ret) 749 goto out_uninit; 750 751 /* Silently fall back to buffered I/O when writing beyond EOF */ 752 if (offset + len > i_size_read(&ip->i_inode)) 753 goto out; 754 755 ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL); 756 757 out: 758 gfs2_glock_dq(&gh); 759 out_uninit: 760 gfs2_holder_uninit(&gh); 761 return ret; 762 } 763 764 static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to) 765 { 766 ssize_t ret; 767 768 if (iocb->ki_flags & IOCB_DIRECT) { 769 ret = gfs2_file_direct_read(iocb, to); 770 if (likely(ret != -ENOTBLK)) 771 return ret; 772 iocb->ki_flags &= ~IOCB_DIRECT; 773 } 774 return generic_file_read_iter(iocb, to); 775 } 776 777 /** 778 * gfs2_file_write_iter - Perform a write to a file 779 * @iocb: The io context 780 * @from: The data to write 781 * 782 * We have to do a lock/unlock here to refresh the inode size for 783 * O_APPEND writes, otherwise we can land up writing at the wrong 784 * offset. There is still a race, but provided the app is using its 785 * own file locking, this will make O_APPEND work as expected. 786 * 787 */ 788 789 static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from) 790 { 791 struct file *file = iocb->ki_filp; 792 struct inode *inode = file_inode(file); 793 struct gfs2_inode *ip = GFS2_I(inode); 794 ssize_t written = 0, ret; 795 796 ret = gfs2_rsqa_alloc(ip); 797 if (ret) 798 return ret; 799 800 gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from)); 801 802 if (iocb->ki_flags & IOCB_APPEND) { 803 struct gfs2_holder gh; 804 805 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh); 806 if (ret) 807 return ret; 808 gfs2_glock_dq_uninit(&gh); 809 } 810 811 inode_lock(inode); 812 ret = generic_write_checks(iocb, from); 813 if (ret <= 0) 814 goto out; 815 816 /* We can write back this queue in page reclaim */ 817 current->backing_dev_info = inode_to_bdi(inode); 818 819 ret = file_remove_privs(file); 820 if (ret) 821 goto out2; 822 823 ret = file_update_time(file); 824 if (ret) 825 goto out2; 826 827 if (iocb->ki_flags & IOCB_DIRECT) { 828 struct address_space *mapping = file->f_mapping; 829 loff_t pos, endbyte; 830 ssize_t buffered; 831 832 written = gfs2_file_direct_write(iocb, from); 833 if (written < 0 || !iov_iter_count(from)) 834 goto out2; 835 836 ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops); 837 if (unlikely(ret < 0)) 838 goto out2; 839 buffered = ret; 840 841 /* 842 * We need to ensure that the page cache pages are written to 843 * disk and invalidated to preserve the expected O_DIRECT 844 * semantics. 845 */ 846 pos = iocb->ki_pos; 847 endbyte = pos + buffered - 1; 848 ret = filemap_write_and_wait_range(mapping, pos, endbyte); 849 if (!ret) { 850 iocb->ki_pos += buffered; 851 written += buffered; 852 invalidate_mapping_pages(mapping, 853 pos >> PAGE_SHIFT, 854 endbyte >> PAGE_SHIFT); 855 } else { 856 /* 857 * We don't know how much we wrote, so just return 858 * the number of bytes which were direct-written 859 */ 860 } 861 } else { 862 ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops); 863 if (likely(ret > 0)) 864 iocb->ki_pos += ret; 865 } 866 867 out2: 868 current->backing_dev_info = NULL; 869 out: 870 inode_unlock(inode); 871 if (likely(ret > 0)) { 872 /* Handle various SYNC-type writes */ 873 ret = generic_write_sync(iocb, ret); 874 } 875 return written ? written : ret; 876 } 877 878 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len, 879 int mode) 880 { 881 struct super_block *sb = inode->i_sb; 882 struct gfs2_inode *ip = GFS2_I(inode); 883 loff_t end = offset + len; 884 struct buffer_head *dibh; 885 int error; 886 887 error = gfs2_meta_inode_buffer(ip, &dibh); 888 if (unlikely(error)) 889 return error; 890 891 gfs2_trans_add_meta(ip->i_gl, dibh); 892 893 if (gfs2_is_stuffed(ip)) { 894 error = gfs2_unstuff_dinode(ip, NULL); 895 if (unlikely(error)) 896 goto out; 897 } 898 899 while (offset < end) { 900 struct iomap iomap = { }; 901 902 error = gfs2_iomap_get_alloc(inode, offset, end - offset, 903 &iomap); 904 if (error) 905 goto out; 906 offset = iomap.offset + iomap.length; 907 if (!(iomap.flags & IOMAP_F_NEW)) 908 continue; 909 error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits, 910 iomap.length >> inode->i_blkbits, 911 GFP_NOFS); 912 if (error) { 913 fs_err(GFS2_SB(inode), "Failed to zero data buffers\n"); 914 goto out; 915 } 916 } 917 out: 918 brelse(dibh); 919 return error; 920 } 921 /** 922 * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of 923 * blocks, determine how many bytes can be written. 924 * @ip: The inode in question. 925 * @len: Max cap of bytes. What we return in *len must be <= this. 926 * @data_blocks: Compute and return the number of data blocks needed 927 * @ind_blocks: Compute and return the number of indirect blocks needed 928 * @max_blocks: The total blocks available to work with. 929 * 930 * Returns: void, but @len, @data_blocks and @ind_blocks are filled in. 931 */ 932 static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len, 933 unsigned int *data_blocks, unsigned int *ind_blocks, 934 unsigned int max_blocks) 935 { 936 loff_t max = *len; 937 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 938 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1); 939 940 for (tmp = max_data; tmp > sdp->sd_diptrs;) { 941 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs); 942 max_data -= tmp; 943 } 944 945 *data_blocks = max_data; 946 *ind_blocks = max_blocks - max_data; 947 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift; 948 if (*len > max) { 949 *len = max; 950 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks); 951 } 952 } 953 954 static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len) 955 { 956 struct inode *inode = file_inode(file); 957 struct gfs2_sbd *sdp = GFS2_SB(inode); 958 struct gfs2_inode *ip = GFS2_I(inode); 959 struct gfs2_alloc_parms ap = { .aflags = 0, }; 960 unsigned int data_blocks = 0, ind_blocks = 0, rblocks; 961 loff_t bytes, max_bytes, max_blks; 962 int error; 963 const loff_t pos = offset; 964 const loff_t count = len; 965 loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1); 966 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift; 967 loff_t max_chunk_size = UINT_MAX & bsize_mask; 968 969 next = (next + 1) << sdp->sd_sb.sb_bsize_shift; 970 971 offset &= bsize_mask; 972 973 len = next - offset; 974 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2; 975 if (!bytes) 976 bytes = UINT_MAX; 977 bytes &= bsize_mask; 978 if (bytes == 0) 979 bytes = sdp->sd_sb.sb_bsize; 980 981 gfs2_size_hint(file, offset, len); 982 983 gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks); 984 ap.min_target = data_blocks + ind_blocks; 985 986 while (len > 0) { 987 if (len < bytes) 988 bytes = len; 989 if (!gfs2_write_alloc_required(ip, offset, bytes)) { 990 len -= bytes; 991 offset += bytes; 992 continue; 993 } 994 995 /* We need to determine how many bytes we can actually 996 * fallocate without exceeding quota or going over the 997 * end of the fs. We start off optimistically by assuming 998 * we can write max_bytes */ 999 max_bytes = (len > max_chunk_size) ? max_chunk_size : len; 1000 1001 /* Since max_bytes is most likely a theoretical max, we 1002 * calculate a more realistic 'bytes' to serve as a good 1003 * starting point for the number of bytes we may be able 1004 * to write */ 1005 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks); 1006 ap.target = data_blocks + ind_blocks; 1007 1008 error = gfs2_quota_lock_check(ip, &ap); 1009 if (error) 1010 return error; 1011 /* ap.allowed tells us how many blocks quota will allow 1012 * us to write. Check if this reduces max_blks */ 1013 max_blks = UINT_MAX; 1014 if (ap.allowed) 1015 max_blks = ap.allowed; 1016 1017 error = gfs2_inplace_reserve(ip, &ap); 1018 if (error) 1019 goto out_qunlock; 1020 1021 /* check if the selected rgrp limits our max_blks further */ 1022 if (ap.allowed && ap.allowed < max_blks) 1023 max_blks = ap.allowed; 1024 1025 /* Almost done. Calculate bytes that can be written using 1026 * max_blks. We also recompute max_bytes, data_blocks and 1027 * ind_blocks */ 1028 calc_max_reserv(ip, &max_bytes, &data_blocks, 1029 &ind_blocks, max_blks); 1030 1031 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA + 1032 RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks); 1033 if (gfs2_is_jdata(ip)) 1034 rblocks += data_blocks ? data_blocks : 1; 1035 1036 error = gfs2_trans_begin(sdp, rblocks, 1037 PAGE_SIZE/sdp->sd_sb.sb_bsize); 1038 if (error) 1039 goto out_trans_fail; 1040 1041 error = fallocate_chunk(inode, offset, max_bytes, mode); 1042 gfs2_trans_end(sdp); 1043 1044 if (error) 1045 goto out_trans_fail; 1046 1047 len -= max_bytes; 1048 offset += max_bytes; 1049 gfs2_inplace_release(ip); 1050 gfs2_quota_unlock(ip); 1051 } 1052 1053 if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size) { 1054 i_size_write(inode, pos + count); 1055 file_update_time(file); 1056 mark_inode_dirty(inode); 1057 } 1058 1059 if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host)) 1060 return vfs_fsync_range(file, pos, pos + count - 1, 1061 (file->f_flags & __O_SYNC) ? 0 : 1); 1062 return 0; 1063 1064 out_trans_fail: 1065 gfs2_inplace_release(ip); 1066 out_qunlock: 1067 gfs2_quota_unlock(ip); 1068 return error; 1069 } 1070 1071 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len) 1072 { 1073 struct inode *inode = file_inode(file); 1074 struct gfs2_sbd *sdp = GFS2_SB(inode); 1075 struct gfs2_inode *ip = GFS2_I(inode); 1076 struct gfs2_holder gh; 1077 int ret; 1078 1079 if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)) 1080 return -EOPNOTSUPP; 1081 /* fallocate is needed by gfs2_grow to reserve space in the rindex */ 1082 if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex) 1083 return -EOPNOTSUPP; 1084 1085 inode_lock(inode); 1086 1087 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 1088 ret = gfs2_glock_nq(&gh); 1089 if (ret) 1090 goto out_uninit; 1091 1092 if (!(mode & FALLOC_FL_KEEP_SIZE) && 1093 (offset + len) > inode->i_size) { 1094 ret = inode_newsize_ok(inode, offset + len); 1095 if (ret) 1096 goto out_unlock; 1097 } 1098 1099 ret = get_write_access(inode); 1100 if (ret) 1101 goto out_unlock; 1102 1103 if (mode & FALLOC_FL_PUNCH_HOLE) { 1104 ret = __gfs2_punch_hole(file, offset, len); 1105 } else { 1106 ret = gfs2_rsqa_alloc(ip); 1107 if (ret) 1108 goto out_putw; 1109 1110 ret = __gfs2_fallocate(file, mode, offset, len); 1111 1112 if (ret) 1113 gfs2_rs_deltree(&ip->i_res); 1114 } 1115 1116 out_putw: 1117 put_write_access(inode); 1118 out_unlock: 1119 gfs2_glock_dq(&gh); 1120 out_uninit: 1121 gfs2_holder_uninit(&gh); 1122 inode_unlock(inode); 1123 return ret; 1124 } 1125 1126 static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe, 1127 struct file *out, loff_t *ppos, 1128 size_t len, unsigned int flags) 1129 { 1130 int error; 1131 struct gfs2_inode *ip = GFS2_I(out->f_mapping->host); 1132 1133 error = gfs2_rsqa_alloc(ip); 1134 if (error) 1135 return (ssize_t)error; 1136 1137 gfs2_size_hint(out, *ppos, len); 1138 1139 return iter_file_splice_write(pipe, out, ppos, len, flags); 1140 } 1141 1142 #ifdef CONFIG_GFS2_FS_LOCKING_DLM 1143 1144 /** 1145 * gfs2_lock - acquire/release a posix lock on a file 1146 * @file: the file pointer 1147 * @cmd: either modify or retrieve lock state, possibly wait 1148 * @fl: type and range of lock 1149 * 1150 * Returns: errno 1151 */ 1152 1153 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl) 1154 { 1155 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 1156 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host); 1157 struct lm_lockstruct *ls = &sdp->sd_lockstruct; 1158 1159 if (!(fl->fl_flags & FL_POSIX)) 1160 return -ENOLCK; 1161 if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK) 1162 return -ENOLCK; 1163 1164 if (cmd == F_CANCELLK) { 1165 /* Hack: */ 1166 cmd = F_SETLK; 1167 fl->fl_type = F_UNLCK; 1168 } 1169 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) { 1170 if (fl->fl_type == F_UNLCK) 1171 locks_lock_file_wait(file, fl); 1172 return -EIO; 1173 } 1174 if (IS_GETLK(cmd)) 1175 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl); 1176 else if (fl->fl_type == F_UNLCK) 1177 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl); 1178 else 1179 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl); 1180 } 1181 1182 static int do_flock(struct file *file, int cmd, struct file_lock *fl) 1183 { 1184 struct gfs2_file *fp = file->private_data; 1185 struct gfs2_holder *fl_gh = &fp->f_fl_gh; 1186 struct gfs2_inode *ip = GFS2_I(file_inode(file)); 1187 struct gfs2_glock *gl; 1188 unsigned int state; 1189 u16 flags; 1190 int error = 0; 1191 int sleeptime; 1192 1193 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED; 1194 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT; 1195 1196 mutex_lock(&fp->f_fl_mutex); 1197 1198 if (gfs2_holder_initialized(fl_gh)) { 1199 struct file_lock request; 1200 if (fl_gh->gh_state == state) 1201 goto out; 1202 locks_init_lock(&request); 1203 request.fl_type = F_UNLCK; 1204 request.fl_flags = FL_FLOCK; 1205 locks_lock_file_wait(file, &request); 1206 gfs2_glock_dq(fl_gh); 1207 gfs2_holder_reinit(state, flags, fl_gh); 1208 } else { 1209 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr, 1210 &gfs2_flock_glops, CREATE, &gl); 1211 if (error) 1212 goto out; 1213 gfs2_holder_init(gl, state, flags, fl_gh); 1214 gfs2_glock_put(gl); 1215 } 1216 for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) { 1217 error = gfs2_glock_nq(fl_gh); 1218 if (error != GLR_TRYFAILED) 1219 break; 1220 fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT; 1221 fl_gh->gh_error = 0; 1222 msleep(sleeptime); 1223 } 1224 if (error) { 1225 gfs2_holder_uninit(fl_gh); 1226 if (error == GLR_TRYFAILED) 1227 error = -EAGAIN; 1228 } else { 1229 error = locks_lock_file_wait(file, fl); 1230 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); 1231 } 1232 1233 out: 1234 mutex_unlock(&fp->f_fl_mutex); 1235 return error; 1236 } 1237 1238 static void do_unflock(struct file *file, struct file_lock *fl) 1239 { 1240 struct gfs2_file *fp = file->private_data; 1241 struct gfs2_holder *fl_gh = &fp->f_fl_gh; 1242 1243 mutex_lock(&fp->f_fl_mutex); 1244 locks_lock_file_wait(file, fl); 1245 if (gfs2_holder_initialized(fl_gh)) { 1246 gfs2_glock_dq(fl_gh); 1247 gfs2_holder_uninit(fl_gh); 1248 } 1249 mutex_unlock(&fp->f_fl_mutex); 1250 } 1251 1252 /** 1253 * gfs2_flock - acquire/release a flock lock on a file 1254 * @file: the file pointer 1255 * @cmd: either modify or retrieve lock state, possibly wait 1256 * @fl: type and range of lock 1257 * 1258 * Returns: errno 1259 */ 1260 1261 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl) 1262 { 1263 if (!(fl->fl_flags & FL_FLOCK)) 1264 return -ENOLCK; 1265 if (fl->fl_type & LOCK_MAND) 1266 return -EOPNOTSUPP; 1267 1268 if (fl->fl_type == F_UNLCK) { 1269 do_unflock(file, fl); 1270 return 0; 1271 } else { 1272 return do_flock(file, cmd, fl); 1273 } 1274 } 1275 1276 const struct file_operations gfs2_file_fops = { 1277 .llseek = gfs2_llseek, 1278 .read_iter = gfs2_file_read_iter, 1279 .write_iter = gfs2_file_write_iter, 1280 .iopoll = iomap_dio_iopoll, 1281 .unlocked_ioctl = gfs2_ioctl, 1282 .mmap = gfs2_mmap, 1283 .open = gfs2_open, 1284 .release = gfs2_release, 1285 .fsync = gfs2_fsync, 1286 .lock = gfs2_lock, 1287 .flock = gfs2_flock, 1288 .splice_read = generic_file_splice_read, 1289 .splice_write = gfs2_file_splice_write, 1290 .setlease = simple_nosetlease, 1291 .fallocate = gfs2_fallocate, 1292 }; 1293 1294 const struct file_operations gfs2_dir_fops = { 1295 .iterate_shared = gfs2_readdir, 1296 .unlocked_ioctl = gfs2_ioctl, 1297 .open = gfs2_open, 1298 .release = gfs2_release, 1299 .fsync = gfs2_fsync, 1300 .lock = gfs2_lock, 1301 .flock = gfs2_flock, 1302 .llseek = default_llseek, 1303 }; 1304 1305 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */ 1306 1307 const struct file_operations gfs2_file_fops_nolock = { 1308 .llseek = gfs2_llseek, 1309 .read_iter = gfs2_file_read_iter, 1310 .write_iter = gfs2_file_write_iter, 1311 .iopoll = iomap_dio_iopoll, 1312 .unlocked_ioctl = gfs2_ioctl, 1313 .mmap = gfs2_mmap, 1314 .open = gfs2_open, 1315 .release = gfs2_release, 1316 .fsync = gfs2_fsync, 1317 .splice_read = generic_file_splice_read, 1318 .splice_write = gfs2_file_splice_write, 1319 .setlease = generic_setlease, 1320 .fallocate = gfs2_fallocate, 1321 }; 1322 1323 const struct file_operations gfs2_dir_fops_nolock = { 1324 .iterate_shared = gfs2_readdir, 1325 .unlocked_ioctl = gfs2_ioctl, 1326 .open = gfs2_open, 1327 .release = gfs2_release, 1328 .fsync = gfs2_fsync, 1329 .llseek = default_llseek, 1330 }; 1331 1332