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 inline u32 gfs2_gfsflags_to_fsflags(struct inode *inode, u32 gfsflags) 140 { 141 int i; 142 u32 fsflags = 0; 143 144 if (S_ISDIR(inode->i_mode)) 145 gfsflags &= ~GFS2_DIF_JDATA; 146 else 147 gfsflags &= ~GFS2_DIF_INHERIT_JDATA; 148 149 for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) 150 if (gfsflags & fsflag_gfs2flag[i].gfsflag) 151 fsflags |= fsflag_gfs2flag[i].fsflag; 152 return fsflags; 153 } 154 155 static int gfs2_get_flags(struct file *filp, u32 __user *ptr) 156 { 157 struct inode *inode = file_inode(filp); 158 struct gfs2_inode *ip = GFS2_I(inode); 159 struct gfs2_holder gh; 160 int error; 161 u32 fsflags; 162 163 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); 164 error = gfs2_glock_nq(&gh); 165 if (error) 166 goto out_uninit; 167 168 fsflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags); 169 170 if (put_user(fsflags, ptr)) 171 error = -EFAULT; 172 173 gfs2_glock_dq(&gh); 174 out_uninit: 175 gfs2_holder_uninit(&gh); 176 return error; 177 } 178 179 void gfs2_set_inode_flags(struct inode *inode) 180 { 181 struct gfs2_inode *ip = GFS2_I(inode); 182 unsigned int flags = inode->i_flags; 183 184 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC); 185 if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode)) 186 flags |= S_NOSEC; 187 if (ip->i_diskflags & GFS2_DIF_IMMUTABLE) 188 flags |= S_IMMUTABLE; 189 if (ip->i_diskflags & GFS2_DIF_APPENDONLY) 190 flags |= S_APPEND; 191 if (ip->i_diskflags & GFS2_DIF_NOATIME) 192 flags |= S_NOATIME; 193 if (ip->i_diskflags & GFS2_DIF_SYNC) 194 flags |= S_SYNC; 195 inode->i_flags = flags; 196 } 197 198 /* Flags that can be set by user space */ 199 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \ 200 GFS2_DIF_IMMUTABLE| \ 201 GFS2_DIF_APPENDONLY| \ 202 GFS2_DIF_NOATIME| \ 203 GFS2_DIF_SYNC| \ 204 GFS2_DIF_TOPDIR| \ 205 GFS2_DIF_INHERIT_JDATA) 206 207 /** 208 * do_gfs2_set_flags - set flags on an inode 209 * @filp: file pointer 210 * @reqflags: The flags to set 211 * @mask: Indicates which flags are valid 212 * @fsflags: The FS_* inode flags passed in 213 * 214 */ 215 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask, 216 const u32 fsflags) 217 { 218 struct inode *inode = file_inode(filp); 219 struct gfs2_inode *ip = GFS2_I(inode); 220 struct gfs2_sbd *sdp = GFS2_SB(inode); 221 struct buffer_head *bh; 222 struct gfs2_holder gh; 223 int error; 224 u32 new_flags, flags, oldflags; 225 226 error = mnt_want_write_file(filp); 227 if (error) 228 return error; 229 230 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 231 if (error) 232 goto out_drop_write; 233 234 oldflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags); 235 error = vfs_ioc_setflags_prepare(inode, oldflags, fsflags); 236 if (error) 237 goto out; 238 239 error = -EACCES; 240 if (!inode_owner_or_capable(inode)) 241 goto out; 242 243 error = 0; 244 flags = ip->i_diskflags; 245 new_flags = (flags & ~mask) | (reqflags & mask); 246 if ((new_flags ^ flags) == 0) 247 goto out; 248 249 error = -EPERM; 250 if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE)) 251 goto out; 252 if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY)) 253 goto out; 254 if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) && 255 !capable(CAP_LINUX_IMMUTABLE)) 256 goto out; 257 if (!IS_IMMUTABLE(inode)) { 258 error = gfs2_permission(inode, MAY_WRITE); 259 if (error) 260 goto out; 261 } 262 if ((flags ^ new_flags) & GFS2_DIF_JDATA) { 263 if (new_flags & GFS2_DIF_JDATA) 264 gfs2_log_flush(sdp, ip->i_gl, 265 GFS2_LOG_HEAD_FLUSH_NORMAL | 266 GFS2_LFC_SET_FLAGS); 267 error = filemap_fdatawrite(inode->i_mapping); 268 if (error) 269 goto out; 270 error = filemap_fdatawait(inode->i_mapping); 271 if (error) 272 goto out; 273 if (new_flags & GFS2_DIF_JDATA) 274 gfs2_ordered_del_inode(ip); 275 } 276 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 277 if (error) 278 goto out; 279 error = gfs2_meta_inode_buffer(ip, &bh); 280 if (error) 281 goto out_trans_end; 282 inode->i_ctime = current_time(inode); 283 gfs2_trans_add_meta(ip->i_gl, bh); 284 ip->i_diskflags = new_flags; 285 gfs2_dinode_out(ip, bh->b_data); 286 brelse(bh); 287 gfs2_set_inode_flags(inode); 288 gfs2_set_aops(inode); 289 out_trans_end: 290 gfs2_trans_end(sdp); 291 out: 292 gfs2_glock_dq_uninit(&gh); 293 out_drop_write: 294 mnt_drop_write_file(filp); 295 return error; 296 } 297 298 static int gfs2_set_flags(struct file *filp, u32 __user *ptr) 299 { 300 struct inode *inode = file_inode(filp); 301 u32 fsflags, gfsflags = 0; 302 u32 mask; 303 int i; 304 305 if (get_user(fsflags, ptr)) 306 return -EFAULT; 307 308 for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) { 309 if (fsflags & fsflag_gfs2flag[i].fsflag) { 310 fsflags &= ~fsflag_gfs2flag[i].fsflag; 311 gfsflags |= fsflag_gfs2flag[i].gfsflag; 312 } 313 } 314 if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET) 315 return -EINVAL; 316 317 mask = GFS2_FLAGS_USER_SET; 318 if (S_ISDIR(inode->i_mode)) { 319 mask &= ~GFS2_DIF_JDATA; 320 } else { 321 /* The GFS2_DIF_TOPDIR flag is only valid for directories. */ 322 if (gfsflags & GFS2_DIF_TOPDIR) 323 return -EINVAL; 324 mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA); 325 } 326 327 return do_gfs2_set_flags(filp, gfsflags, mask, fsflags); 328 } 329 330 static int gfs2_getlabel(struct file *filp, char __user *label) 331 { 332 struct inode *inode = file_inode(filp); 333 struct gfs2_sbd *sdp = GFS2_SB(inode); 334 335 if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN)) 336 return -EFAULT; 337 338 return 0; 339 } 340 341 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) 342 { 343 switch(cmd) { 344 case FS_IOC_GETFLAGS: 345 return gfs2_get_flags(filp, (u32 __user *)arg); 346 case FS_IOC_SETFLAGS: 347 return gfs2_set_flags(filp, (u32 __user *)arg); 348 case FITRIM: 349 return gfs2_fitrim(filp, (void __user *)arg); 350 case FS_IOC_GETFSLABEL: 351 return gfs2_getlabel(filp, (char __user *)arg); 352 } 353 354 return -ENOTTY; 355 } 356 357 /** 358 * gfs2_size_hint - Give a hint to the size of a write request 359 * @filep: The struct file 360 * @offset: The file offset of the write 361 * @size: The length of the write 362 * 363 * When we are about to do a write, this function records the total 364 * write size in order to provide a suitable hint to the lower layers 365 * about how many blocks will be required. 366 * 367 */ 368 369 static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size) 370 { 371 struct inode *inode = file_inode(filep); 372 struct gfs2_sbd *sdp = GFS2_SB(inode); 373 struct gfs2_inode *ip = GFS2_I(inode); 374 size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift; 375 int hint = min_t(size_t, INT_MAX, blks); 376 377 if (hint > atomic_read(&ip->i_sizehint)) 378 atomic_set(&ip->i_sizehint, hint); 379 } 380 381 /** 382 * gfs2_allocate_page_backing - Allocate blocks for a write fault 383 * @page: The (locked) page to allocate backing for 384 * 385 * We try to allocate all the blocks required for the page in one go. This 386 * might fail for various reasons, so we keep trying until all the blocks to 387 * back this page are allocated. If some of the blocks are already allocated, 388 * that is ok too. 389 */ 390 static int gfs2_allocate_page_backing(struct page *page) 391 { 392 u64 pos = page_offset(page); 393 u64 size = PAGE_SIZE; 394 395 do { 396 struct iomap iomap = { }; 397 398 if (gfs2_iomap_get_alloc(page->mapping->host, pos, 1, &iomap)) 399 return -EIO; 400 401 iomap.length = min(iomap.length, size); 402 size -= iomap.length; 403 pos += iomap.length; 404 } while (size > 0); 405 406 return 0; 407 } 408 409 /** 410 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable 411 * @vma: The virtual memory area 412 * @vmf: The virtual memory fault containing the page to become writable 413 * 414 * When the page becomes writable, we need to ensure that we have 415 * blocks allocated on disk to back that page. 416 */ 417 418 static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf) 419 { 420 struct page *page = vmf->page; 421 struct inode *inode = file_inode(vmf->vma->vm_file); 422 struct gfs2_inode *ip = GFS2_I(inode); 423 struct gfs2_sbd *sdp = GFS2_SB(inode); 424 struct gfs2_alloc_parms ap = { .aflags = 0, }; 425 unsigned long last_index; 426 u64 pos = page_offset(page); 427 unsigned int data_blocks, ind_blocks, rblocks; 428 struct gfs2_holder gh; 429 loff_t size; 430 int ret; 431 432 sb_start_pagefault(inode->i_sb); 433 434 ret = gfs2_rsqa_alloc(ip); 435 if (ret) 436 goto out; 437 438 gfs2_size_hint(vmf->vma->vm_file, pos, PAGE_SIZE); 439 440 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 441 ret = gfs2_glock_nq(&gh); 442 if (ret) 443 goto out_uninit; 444 445 /* Update file times before taking page lock */ 446 file_update_time(vmf->vma->vm_file); 447 448 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags); 449 set_bit(GIF_SW_PAGED, &ip->i_flags); 450 451 if (!gfs2_write_alloc_required(ip, pos, PAGE_SIZE)) { 452 lock_page(page); 453 if (!PageUptodate(page) || page->mapping != inode->i_mapping) { 454 ret = -EAGAIN; 455 unlock_page(page); 456 } 457 goto out_unlock; 458 } 459 460 ret = gfs2_rindex_update(sdp); 461 if (ret) 462 goto out_unlock; 463 464 gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks); 465 ap.target = data_blocks + ind_blocks; 466 ret = gfs2_quota_lock_check(ip, &ap); 467 if (ret) 468 goto out_unlock; 469 ret = gfs2_inplace_reserve(ip, &ap); 470 if (ret) 471 goto out_quota_unlock; 472 473 rblocks = RES_DINODE + ind_blocks; 474 if (gfs2_is_jdata(ip)) 475 rblocks += data_blocks ? data_blocks : 1; 476 if (ind_blocks || data_blocks) { 477 rblocks += RES_STATFS + RES_QUOTA; 478 rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks); 479 } 480 ret = gfs2_trans_begin(sdp, rblocks, 0); 481 if (ret) 482 goto out_trans_fail; 483 484 lock_page(page); 485 ret = -EINVAL; 486 size = i_size_read(inode); 487 last_index = (size - 1) >> PAGE_SHIFT; 488 /* Check page index against inode size */ 489 if (size == 0 || (page->index > last_index)) 490 goto out_trans_end; 491 492 ret = -EAGAIN; 493 /* If truncated, we must retry the operation, we may have raced 494 * with the glock demotion code. 495 */ 496 if (!PageUptodate(page) || page->mapping != inode->i_mapping) 497 goto out_trans_end; 498 499 /* Unstuff, if required, and allocate backing blocks for page */ 500 ret = 0; 501 if (gfs2_is_stuffed(ip)) 502 ret = gfs2_unstuff_dinode(ip, page); 503 if (ret == 0) 504 ret = gfs2_allocate_page_backing(page); 505 506 out_trans_end: 507 if (ret) 508 unlock_page(page); 509 gfs2_trans_end(sdp); 510 out_trans_fail: 511 gfs2_inplace_release(ip); 512 out_quota_unlock: 513 gfs2_quota_unlock(ip); 514 out_unlock: 515 gfs2_glock_dq(&gh); 516 out_uninit: 517 gfs2_holder_uninit(&gh); 518 if (ret == 0) { 519 set_page_dirty(page); 520 wait_for_stable_page(page); 521 } 522 out: 523 sb_end_pagefault(inode->i_sb); 524 return block_page_mkwrite_return(ret); 525 } 526 527 static const struct vm_operations_struct gfs2_vm_ops = { 528 .fault = filemap_fault, 529 .map_pages = filemap_map_pages, 530 .page_mkwrite = gfs2_page_mkwrite, 531 }; 532 533 /** 534 * gfs2_mmap - 535 * @file: The file to map 536 * @vma: The VMA which described the mapping 537 * 538 * There is no need to get a lock here unless we should be updating 539 * atime. We ignore any locking errors since the only consequence is 540 * a missed atime update (which will just be deferred until later). 541 * 542 * Returns: 0 543 */ 544 545 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma) 546 { 547 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 548 549 if (!(file->f_flags & O_NOATIME) && 550 !IS_NOATIME(&ip->i_inode)) { 551 struct gfs2_holder i_gh; 552 int error; 553 554 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, 555 &i_gh); 556 if (error) 557 return error; 558 /* grab lock to update inode */ 559 gfs2_glock_dq_uninit(&i_gh); 560 file_accessed(file); 561 } 562 vma->vm_ops = &gfs2_vm_ops; 563 564 return 0; 565 } 566 567 /** 568 * gfs2_open_common - This is common to open and atomic_open 569 * @inode: The inode being opened 570 * @file: The file being opened 571 * 572 * This maybe called under a glock or not depending upon how it has 573 * been called. We must always be called under a glock for regular 574 * files, however. For other file types, it does not matter whether 575 * we hold the glock or not. 576 * 577 * Returns: Error code or 0 for success 578 */ 579 580 int gfs2_open_common(struct inode *inode, struct file *file) 581 { 582 struct gfs2_file *fp; 583 int ret; 584 585 if (S_ISREG(inode->i_mode)) { 586 ret = generic_file_open(inode, file); 587 if (ret) 588 return ret; 589 } 590 591 fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS); 592 if (!fp) 593 return -ENOMEM; 594 595 mutex_init(&fp->f_fl_mutex); 596 597 gfs2_assert_warn(GFS2_SB(inode), !file->private_data); 598 file->private_data = fp; 599 return 0; 600 } 601 602 /** 603 * gfs2_open - open a file 604 * @inode: the inode to open 605 * @file: the struct file for this opening 606 * 607 * After atomic_open, this function is only used for opening files 608 * which are already cached. We must still get the glock for regular 609 * files to ensure that we have the file size uptodate for the large 610 * file check which is in the common code. That is only an issue for 611 * regular files though. 612 * 613 * Returns: errno 614 */ 615 616 static int gfs2_open(struct inode *inode, struct file *file) 617 { 618 struct gfs2_inode *ip = GFS2_I(inode); 619 struct gfs2_holder i_gh; 620 int error; 621 bool need_unlock = false; 622 623 if (S_ISREG(ip->i_inode.i_mode)) { 624 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, 625 &i_gh); 626 if (error) 627 return error; 628 need_unlock = true; 629 } 630 631 error = gfs2_open_common(inode, file); 632 633 if (need_unlock) 634 gfs2_glock_dq_uninit(&i_gh); 635 636 return error; 637 } 638 639 /** 640 * gfs2_release - called to close a struct file 641 * @inode: the inode the struct file belongs to 642 * @file: the struct file being closed 643 * 644 * Returns: errno 645 */ 646 647 static int gfs2_release(struct inode *inode, struct file *file) 648 { 649 struct gfs2_inode *ip = GFS2_I(inode); 650 651 kfree(file->private_data); 652 file->private_data = NULL; 653 654 if (!(file->f_mode & FMODE_WRITE)) 655 return 0; 656 657 gfs2_rsqa_delete(ip, &inode->i_writecount); 658 return 0; 659 } 660 661 /** 662 * gfs2_fsync - sync the dirty data for a file (across the cluster) 663 * @file: the file that points to the dentry 664 * @start: the start position in the file to sync 665 * @end: the end position in the file to sync 666 * @datasync: set if we can ignore timestamp changes 667 * 668 * We split the data flushing here so that we don't wait for the data 669 * until after we've also sent the metadata to disk. Note that for 670 * data=ordered, we will write & wait for the data at the log flush 671 * stage anyway, so this is unlikely to make much of a difference 672 * except in the data=writeback case. 673 * 674 * If the fdatawrite fails due to any reason except -EIO, we will 675 * continue the remainder of the fsync, although we'll still report 676 * the error at the end. This is to match filemap_write_and_wait_range() 677 * behaviour. 678 * 679 * Returns: errno 680 */ 681 682 static int gfs2_fsync(struct file *file, loff_t start, loff_t end, 683 int datasync) 684 { 685 struct address_space *mapping = file->f_mapping; 686 struct inode *inode = mapping->host; 687 int sync_state = inode->i_state & I_DIRTY_ALL; 688 struct gfs2_inode *ip = GFS2_I(inode); 689 int ret = 0, ret1 = 0; 690 691 if (mapping->nrpages) { 692 ret1 = filemap_fdatawrite_range(mapping, start, end); 693 if (ret1 == -EIO) 694 return ret1; 695 } 696 697 if (!gfs2_is_jdata(ip)) 698 sync_state &= ~I_DIRTY_PAGES; 699 if (datasync) 700 sync_state &= ~(I_DIRTY_SYNC | I_DIRTY_TIME); 701 702 if (sync_state) { 703 ret = sync_inode_metadata(inode, 1); 704 if (ret) 705 return ret; 706 if (gfs2_is_jdata(ip)) 707 ret = file_write_and_wait(file); 708 if (ret) 709 return ret; 710 gfs2_ail_flush(ip->i_gl, 1); 711 } 712 713 if (mapping->nrpages) 714 ret = file_fdatawait_range(file, start, end); 715 716 return ret ? ret : ret1; 717 } 718 719 static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to) 720 { 721 struct file *file = iocb->ki_filp; 722 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 723 size_t count = iov_iter_count(to); 724 struct gfs2_holder gh; 725 ssize_t ret; 726 727 if (!count) 728 return 0; /* skip atime */ 729 730 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh); 731 ret = gfs2_glock_nq(&gh); 732 if (ret) 733 goto out_uninit; 734 735 ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL); 736 737 gfs2_glock_dq(&gh); 738 out_uninit: 739 gfs2_holder_uninit(&gh); 740 return ret; 741 } 742 743 static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from) 744 { 745 struct file *file = iocb->ki_filp; 746 struct inode *inode = file->f_mapping->host; 747 struct gfs2_inode *ip = GFS2_I(inode); 748 size_t len = iov_iter_count(from); 749 loff_t offset = iocb->ki_pos; 750 struct gfs2_holder gh; 751 ssize_t ret; 752 753 /* 754 * Deferred lock, even if its a write, since we do no allocation on 755 * this path. All we need to change is the atime, and this lock mode 756 * ensures that other nodes have flushed their buffered read caches 757 * (i.e. their page cache entries for this inode). We do not, 758 * unfortunately, have the option of only flushing a range like the 759 * VFS does. 760 */ 761 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh); 762 ret = gfs2_glock_nq(&gh); 763 if (ret) 764 goto out_uninit; 765 766 /* Silently fall back to buffered I/O when writing beyond EOF */ 767 if (offset + len > i_size_read(&ip->i_inode)) 768 goto out; 769 770 ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL); 771 772 out: 773 gfs2_glock_dq(&gh); 774 out_uninit: 775 gfs2_holder_uninit(&gh); 776 return ret; 777 } 778 779 static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to) 780 { 781 ssize_t ret; 782 783 if (iocb->ki_flags & IOCB_DIRECT) { 784 ret = gfs2_file_direct_read(iocb, to); 785 if (likely(ret != -ENOTBLK)) 786 return ret; 787 iocb->ki_flags &= ~IOCB_DIRECT; 788 } 789 return generic_file_read_iter(iocb, to); 790 } 791 792 /** 793 * gfs2_file_write_iter - Perform a write to a file 794 * @iocb: The io context 795 * @from: The data to write 796 * 797 * We have to do a lock/unlock here to refresh the inode size for 798 * O_APPEND writes, otherwise we can land up writing at the wrong 799 * offset. There is still a race, but provided the app is using its 800 * own file locking, this will make O_APPEND work as expected. 801 * 802 */ 803 804 static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from) 805 { 806 struct file *file = iocb->ki_filp; 807 struct inode *inode = file_inode(file); 808 struct gfs2_inode *ip = GFS2_I(inode); 809 ssize_t written = 0, ret; 810 811 ret = gfs2_rsqa_alloc(ip); 812 if (ret) 813 return ret; 814 815 gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from)); 816 817 if (iocb->ki_flags & IOCB_APPEND) { 818 struct gfs2_holder gh; 819 820 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh); 821 if (ret) 822 return ret; 823 gfs2_glock_dq_uninit(&gh); 824 } 825 826 inode_lock(inode); 827 ret = generic_write_checks(iocb, from); 828 if (ret <= 0) 829 goto out; 830 831 /* We can write back this queue in page reclaim */ 832 current->backing_dev_info = inode_to_bdi(inode); 833 834 ret = file_remove_privs(file); 835 if (ret) 836 goto out2; 837 838 ret = file_update_time(file); 839 if (ret) 840 goto out2; 841 842 if (iocb->ki_flags & IOCB_DIRECT) { 843 struct address_space *mapping = file->f_mapping; 844 loff_t pos, endbyte; 845 ssize_t buffered; 846 847 written = gfs2_file_direct_write(iocb, from); 848 if (written < 0 || !iov_iter_count(from)) 849 goto out2; 850 851 ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops); 852 if (unlikely(ret < 0)) 853 goto out2; 854 buffered = ret; 855 856 /* 857 * We need to ensure that the page cache pages are written to 858 * disk and invalidated to preserve the expected O_DIRECT 859 * semantics. 860 */ 861 pos = iocb->ki_pos; 862 endbyte = pos + buffered - 1; 863 ret = filemap_write_and_wait_range(mapping, pos, endbyte); 864 if (!ret) { 865 iocb->ki_pos += buffered; 866 written += buffered; 867 invalidate_mapping_pages(mapping, 868 pos >> PAGE_SHIFT, 869 endbyte >> PAGE_SHIFT); 870 } else { 871 /* 872 * We don't know how much we wrote, so just return 873 * the number of bytes which were direct-written 874 */ 875 } 876 } else { 877 ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops); 878 if (likely(ret > 0)) 879 iocb->ki_pos += ret; 880 } 881 882 out2: 883 current->backing_dev_info = NULL; 884 out: 885 inode_unlock(inode); 886 if (likely(ret > 0)) { 887 /* Handle various SYNC-type writes */ 888 ret = generic_write_sync(iocb, ret); 889 } 890 return written ? written : ret; 891 } 892 893 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len, 894 int mode) 895 { 896 struct super_block *sb = inode->i_sb; 897 struct gfs2_inode *ip = GFS2_I(inode); 898 loff_t end = offset + len; 899 struct buffer_head *dibh; 900 int error; 901 902 error = gfs2_meta_inode_buffer(ip, &dibh); 903 if (unlikely(error)) 904 return error; 905 906 gfs2_trans_add_meta(ip->i_gl, dibh); 907 908 if (gfs2_is_stuffed(ip)) { 909 error = gfs2_unstuff_dinode(ip, NULL); 910 if (unlikely(error)) 911 goto out; 912 } 913 914 while (offset < end) { 915 struct iomap iomap = { }; 916 917 error = gfs2_iomap_get_alloc(inode, offset, end - offset, 918 &iomap); 919 if (error) 920 goto out; 921 offset = iomap.offset + iomap.length; 922 if (!(iomap.flags & IOMAP_F_NEW)) 923 continue; 924 error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits, 925 iomap.length >> inode->i_blkbits, 926 GFP_NOFS); 927 if (error) { 928 fs_err(GFS2_SB(inode), "Failed to zero data buffers\n"); 929 goto out; 930 } 931 } 932 out: 933 brelse(dibh); 934 return error; 935 } 936 /** 937 * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of 938 * blocks, determine how many bytes can be written. 939 * @ip: The inode in question. 940 * @len: Max cap of bytes. What we return in *len must be <= this. 941 * @data_blocks: Compute and return the number of data blocks needed 942 * @ind_blocks: Compute and return the number of indirect blocks needed 943 * @max_blocks: The total blocks available to work with. 944 * 945 * Returns: void, but @len, @data_blocks and @ind_blocks are filled in. 946 */ 947 static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len, 948 unsigned int *data_blocks, unsigned int *ind_blocks, 949 unsigned int max_blocks) 950 { 951 loff_t max = *len; 952 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 953 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1); 954 955 for (tmp = max_data; tmp > sdp->sd_diptrs;) { 956 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs); 957 max_data -= tmp; 958 } 959 960 *data_blocks = max_data; 961 *ind_blocks = max_blocks - max_data; 962 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift; 963 if (*len > max) { 964 *len = max; 965 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks); 966 } 967 } 968 969 static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len) 970 { 971 struct inode *inode = file_inode(file); 972 struct gfs2_sbd *sdp = GFS2_SB(inode); 973 struct gfs2_inode *ip = GFS2_I(inode); 974 struct gfs2_alloc_parms ap = { .aflags = 0, }; 975 unsigned int data_blocks = 0, ind_blocks = 0, rblocks; 976 loff_t bytes, max_bytes, max_blks; 977 int error; 978 const loff_t pos = offset; 979 const loff_t count = len; 980 loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1); 981 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift; 982 loff_t max_chunk_size = UINT_MAX & bsize_mask; 983 984 next = (next + 1) << sdp->sd_sb.sb_bsize_shift; 985 986 offset &= bsize_mask; 987 988 len = next - offset; 989 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2; 990 if (!bytes) 991 bytes = UINT_MAX; 992 bytes &= bsize_mask; 993 if (bytes == 0) 994 bytes = sdp->sd_sb.sb_bsize; 995 996 gfs2_size_hint(file, offset, len); 997 998 gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks); 999 ap.min_target = data_blocks + ind_blocks; 1000 1001 while (len > 0) { 1002 if (len < bytes) 1003 bytes = len; 1004 if (!gfs2_write_alloc_required(ip, offset, bytes)) { 1005 len -= bytes; 1006 offset += bytes; 1007 continue; 1008 } 1009 1010 /* We need to determine how many bytes we can actually 1011 * fallocate without exceeding quota or going over the 1012 * end of the fs. We start off optimistically by assuming 1013 * we can write max_bytes */ 1014 max_bytes = (len > max_chunk_size) ? max_chunk_size : len; 1015 1016 /* Since max_bytes is most likely a theoretical max, we 1017 * calculate a more realistic 'bytes' to serve as a good 1018 * starting point for the number of bytes we may be able 1019 * to write */ 1020 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks); 1021 ap.target = data_blocks + ind_blocks; 1022 1023 error = gfs2_quota_lock_check(ip, &ap); 1024 if (error) 1025 return error; 1026 /* ap.allowed tells us how many blocks quota will allow 1027 * us to write. Check if this reduces max_blks */ 1028 max_blks = UINT_MAX; 1029 if (ap.allowed) 1030 max_blks = ap.allowed; 1031 1032 error = gfs2_inplace_reserve(ip, &ap); 1033 if (error) 1034 goto out_qunlock; 1035 1036 /* check if the selected rgrp limits our max_blks further */ 1037 if (ap.allowed && ap.allowed < max_blks) 1038 max_blks = ap.allowed; 1039 1040 /* Almost done. Calculate bytes that can be written using 1041 * max_blks. We also recompute max_bytes, data_blocks and 1042 * ind_blocks */ 1043 calc_max_reserv(ip, &max_bytes, &data_blocks, 1044 &ind_blocks, max_blks); 1045 1046 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA + 1047 RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks); 1048 if (gfs2_is_jdata(ip)) 1049 rblocks += data_blocks ? data_blocks : 1; 1050 1051 error = gfs2_trans_begin(sdp, rblocks, 1052 PAGE_SIZE >> inode->i_blkbits); 1053 if (error) 1054 goto out_trans_fail; 1055 1056 error = fallocate_chunk(inode, offset, max_bytes, mode); 1057 gfs2_trans_end(sdp); 1058 1059 if (error) 1060 goto out_trans_fail; 1061 1062 len -= max_bytes; 1063 offset += max_bytes; 1064 gfs2_inplace_release(ip); 1065 gfs2_quota_unlock(ip); 1066 } 1067 1068 if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size) 1069 i_size_write(inode, pos + count); 1070 file_update_time(file); 1071 mark_inode_dirty(inode); 1072 1073 if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host)) 1074 return vfs_fsync_range(file, pos, pos + count - 1, 1075 (file->f_flags & __O_SYNC) ? 0 : 1); 1076 return 0; 1077 1078 out_trans_fail: 1079 gfs2_inplace_release(ip); 1080 out_qunlock: 1081 gfs2_quota_unlock(ip); 1082 return error; 1083 } 1084 1085 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len) 1086 { 1087 struct inode *inode = file_inode(file); 1088 struct gfs2_sbd *sdp = GFS2_SB(inode); 1089 struct gfs2_inode *ip = GFS2_I(inode); 1090 struct gfs2_holder gh; 1091 int ret; 1092 1093 if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)) 1094 return -EOPNOTSUPP; 1095 /* fallocate is needed by gfs2_grow to reserve space in the rindex */ 1096 if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex) 1097 return -EOPNOTSUPP; 1098 1099 inode_lock(inode); 1100 1101 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 1102 ret = gfs2_glock_nq(&gh); 1103 if (ret) 1104 goto out_uninit; 1105 1106 if (!(mode & FALLOC_FL_KEEP_SIZE) && 1107 (offset + len) > inode->i_size) { 1108 ret = inode_newsize_ok(inode, offset + len); 1109 if (ret) 1110 goto out_unlock; 1111 } 1112 1113 ret = get_write_access(inode); 1114 if (ret) 1115 goto out_unlock; 1116 1117 if (mode & FALLOC_FL_PUNCH_HOLE) { 1118 ret = __gfs2_punch_hole(file, offset, len); 1119 } else { 1120 ret = gfs2_rsqa_alloc(ip); 1121 if (ret) 1122 goto out_putw; 1123 1124 ret = __gfs2_fallocate(file, mode, offset, len); 1125 1126 if (ret) 1127 gfs2_rs_deltree(&ip->i_res); 1128 } 1129 1130 out_putw: 1131 put_write_access(inode); 1132 out_unlock: 1133 gfs2_glock_dq(&gh); 1134 out_uninit: 1135 gfs2_holder_uninit(&gh); 1136 inode_unlock(inode); 1137 return ret; 1138 } 1139 1140 static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe, 1141 struct file *out, loff_t *ppos, 1142 size_t len, unsigned int flags) 1143 { 1144 int error; 1145 struct gfs2_inode *ip = GFS2_I(out->f_mapping->host); 1146 1147 error = gfs2_rsqa_alloc(ip); 1148 if (error) 1149 return (ssize_t)error; 1150 1151 gfs2_size_hint(out, *ppos, len); 1152 1153 return iter_file_splice_write(pipe, out, ppos, len, flags); 1154 } 1155 1156 #ifdef CONFIG_GFS2_FS_LOCKING_DLM 1157 1158 /** 1159 * gfs2_lock - acquire/release a posix lock on a file 1160 * @file: the file pointer 1161 * @cmd: either modify or retrieve lock state, possibly wait 1162 * @fl: type and range of lock 1163 * 1164 * Returns: errno 1165 */ 1166 1167 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl) 1168 { 1169 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 1170 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host); 1171 struct lm_lockstruct *ls = &sdp->sd_lockstruct; 1172 1173 if (!(fl->fl_flags & FL_POSIX)) 1174 return -ENOLCK; 1175 if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK) 1176 return -ENOLCK; 1177 1178 if (cmd == F_CANCELLK) { 1179 /* Hack: */ 1180 cmd = F_SETLK; 1181 fl->fl_type = F_UNLCK; 1182 } 1183 if (unlikely(test_bit(SDF_WITHDRAWN, &sdp->sd_flags))) { 1184 if (fl->fl_type == F_UNLCK) 1185 locks_lock_file_wait(file, fl); 1186 return -EIO; 1187 } 1188 if (IS_GETLK(cmd)) 1189 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl); 1190 else if (fl->fl_type == F_UNLCK) 1191 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl); 1192 else 1193 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl); 1194 } 1195 1196 static int do_flock(struct file *file, int cmd, struct file_lock *fl) 1197 { 1198 struct gfs2_file *fp = file->private_data; 1199 struct gfs2_holder *fl_gh = &fp->f_fl_gh; 1200 struct gfs2_inode *ip = GFS2_I(file_inode(file)); 1201 struct gfs2_glock *gl; 1202 unsigned int state; 1203 u16 flags; 1204 int error = 0; 1205 int sleeptime; 1206 1207 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED; 1208 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT; 1209 1210 mutex_lock(&fp->f_fl_mutex); 1211 1212 if (gfs2_holder_initialized(fl_gh)) { 1213 struct file_lock request; 1214 if (fl_gh->gh_state == state) 1215 goto out; 1216 locks_init_lock(&request); 1217 request.fl_type = F_UNLCK; 1218 request.fl_flags = FL_FLOCK; 1219 locks_lock_file_wait(file, &request); 1220 gfs2_glock_dq(fl_gh); 1221 gfs2_holder_reinit(state, flags, fl_gh); 1222 } else { 1223 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr, 1224 &gfs2_flock_glops, CREATE, &gl); 1225 if (error) 1226 goto out; 1227 gfs2_holder_init(gl, state, flags, fl_gh); 1228 gfs2_glock_put(gl); 1229 } 1230 for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) { 1231 error = gfs2_glock_nq(fl_gh); 1232 if (error != GLR_TRYFAILED) 1233 break; 1234 fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT; 1235 fl_gh->gh_error = 0; 1236 msleep(sleeptime); 1237 } 1238 if (error) { 1239 gfs2_holder_uninit(fl_gh); 1240 if (error == GLR_TRYFAILED) 1241 error = -EAGAIN; 1242 } else { 1243 error = locks_lock_file_wait(file, fl); 1244 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); 1245 } 1246 1247 out: 1248 mutex_unlock(&fp->f_fl_mutex); 1249 return error; 1250 } 1251 1252 static void do_unflock(struct file *file, struct file_lock *fl) 1253 { 1254 struct gfs2_file *fp = file->private_data; 1255 struct gfs2_holder *fl_gh = &fp->f_fl_gh; 1256 1257 mutex_lock(&fp->f_fl_mutex); 1258 locks_lock_file_wait(file, fl); 1259 if (gfs2_holder_initialized(fl_gh)) { 1260 gfs2_glock_dq(fl_gh); 1261 gfs2_holder_uninit(fl_gh); 1262 } 1263 mutex_unlock(&fp->f_fl_mutex); 1264 } 1265 1266 /** 1267 * gfs2_flock - acquire/release a flock lock on a file 1268 * @file: the file pointer 1269 * @cmd: either modify or retrieve lock state, possibly wait 1270 * @fl: type and range of lock 1271 * 1272 * Returns: errno 1273 */ 1274 1275 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl) 1276 { 1277 if (!(fl->fl_flags & FL_FLOCK)) 1278 return -ENOLCK; 1279 if (fl->fl_type & LOCK_MAND) 1280 return -EOPNOTSUPP; 1281 1282 if (fl->fl_type == F_UNLCK) { 1283 do_unflock(file, fl); 1284 return 0; 1285 } else { 1286 return do_flock(file, cmd, fl); 1287 } 1288 } 1289 1290 const struct file_operations gfs2_file_fops = { 1291 .llseek = gfs2_llseek, 1292 .read_iter = gfs2_file_read_iter, 1293 .write_iter = gfs2_file_write_iter, 1294 .iopoll = iomap_dio_iopoll, 1295 .unlocked_ioctl = gfs2_ioctl, 1296 .mmap = gfs2_mmap, 1297 .open = gfs2_open, 1298 .release = gfs2_release, 1299 .fsync = gfs2_fsync, 1300 .lock = gfs2_lock, 1301 .flock = gfs2_flock, 1302 .splice_read = generic_file_splice_read, 1303 .splice_write = gfs2_file_splice_write, 1304 .setlease = simple_nosetlease, 1305 .fallocate = gfs2_fallocate, 1306 }; 1307 1308 const struct file_operations gfs2_dir_fops = { 1309 .iterate_shared = gfs2_readdir, 1310 .unlocked_ioctl = gfs2_ioctl, 1311 .open = gfs2_open, 1312 .release = gfs2_release, 1313 .fsync = gfs2_fsync, 1314 .lock = gfs2_lock, 1315 .flock = gfs2_flock, 1316 .llseek = default_llseek, 1317 }; 1318 1319 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */ 1320 1321 const struct file_operations gfs2_file_fops_nolock = { 1322 .llseek = gfs2_llseek, 1323 .read_iter = gfs2_file_read_iter, 1324 .write_iter = gfs2_file_write_iter, 1325 .iopoll = iomap_dio_iopoll, 1326 .unlocked_ioctl = gfs2_ioctl, 1327 .mmap = gfs2_mmap, 1328 .open = gfs2_open, 1329 .release = gfs2_release, 1330 .fsync = gfs2_fsync, 1331 .splice_read = generic_file_splice_read, 1332 .splice_write = gfs2_file_splice_write, 1333 .setlease = generic_setlease, 1334 .fallocate = gfs2_fallocate, 1335 }; 1336 1337 const struct file_operations gfs2_dir_fops_nolock = { 1338 .iterate_shared = gfs2_readdir, 1339 .unlocked_ioctl = gfs2_ioctl, 1340 .open = gfs2_open, 1341 .release = gfs2_release, 1342 .fsync = gfs2_fsync, 1343 .llseek = default_llseek, 1344 }; 1345 1346