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 gfs2_llseek(struct file * file,loff_t offset,int whence)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 gfs2_readdir(struct file * file,struct dir_context * ctx)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 */ fsflags_cvt(const u32 * table,u32 val)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 gfs2_get_flags(struct file * filp,u32 __user * ptr)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 goto out_uninit; 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 out_uninit: 173 gfs2_holder_uninit(&gh); 174 return error; 175 } 176 gfs2_set_inode_flags(struct inode * inode)177 void gfs2_set_inode_flags(struct inode *inode) 178 { 179 struct gfs2_inode *ip = GFS2_I(inode); 180 unsigned int flags = inode->i_flags; 181 182 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC); 183 if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode)) 184 flags |= S_NOSEC; 185 if (ip->i_diskflags & GFS2_DIF_IMMUTABLE) 186 flags |= S_IMMUTABLE; 187 if (ip->i_diskflags & GFS2_DIF_APPENDONLY) 188 flags |= S_APPEND; 189 if (ip->i_diskflags & GFS2_DIF_NOATIME) 190 flags |= S_NOATIME; 191 if (ip->i_diskflags & GFS2_DIF_SYNC) 192 flags |= S_SYNC; 193 inode->i_flags = flags; 194 } 195 196 /* Flags that can be set by user space */ 197 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \ 198 GFS2_DIF_IMMUTABLE| \ 199 GFS2_DIF_APPENDONLY| \ 200 GFS2_DIF_NOATIME| \ 201 GFS2_DIF_SYNC| \ 202 GFS2_DIF_SYSTEM| \ 203 GFS2_DIF_TOPDIR| \ 204 GFS2_DIF_INHERIT_JDATA) 205 206 /** 207 * do_gfs2_set_flags - set flags on an inode 208 * @filp: file pointer 209 * @reqflags: The flags to set 210 * @mask: Indicates which flags are valid 211 * 212 */ do_gfs2_set_flags(struct file * filp,u32 reqflags,u32 mask)213 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask) 214 { 215 struct inode *inode = file_inode(filp); 216 struct gfs2_inode *ip = GFS2_I(inode); 217 struct gfs2_sbd *sdp = GFS2_SB(inode); 218 struct buffer_head *bh; 219 struct gfs2_holder gh; 220 int error; 221 u32 new_flags, flags; 222 223 error = mnt_want_write_file(filp); 224 if (error) 225 return error; 226 227 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 228 if (error) 229 goto out_drop_write; 230 231 error = -EACCES; 232 if (!inode_owner_or_capable(inode)) 233 goto out; 234 235 error = 0; 236 flags = ip->i_diskflags; 237 new_flags = (flags & ~mask) | (reqflags & mask); 238 if ((new_flags ^ flags) == 0) 239 goto out; 240 241 error = -EINVAL; 242 if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET) 243 goto out; 244 245 error = -EPERM; 246 if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE)) 247 goto out; 248 if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY)) 249 goto out; 250 if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) && 251 !capable(CAP_LINUX_IMMUTABLE)) 252 goto out; 253 if (!IS_IMMUTABLE(inode)) { 254 error = gfs2_permission(inode, MAY_WRITE); 255 if (error) 256 goto out; 257 } 258 if ((flags ^ new_flags) & GFS2_DIF_JDATA) { 259 if (new_flags & GFS2_DIF_JDATA) 260 gfs2_log_flush(sdp, ip->i_gl, NORMAL_FLUSH); 261 error = filemap_fdatawrite(inode->i_mapping); 262 if (error) 263 goto out; 264 error = filemap_fdatawait(inode->i_mapping); 265 if (error) 266 goto out; 267 if (new_flags & GFS2_DIF_JDATA) 268 gfs2_ordered_del_inode(ip); 269 } 270 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 271 if (error) 272 goto out; 273 error = gfs2_meta_inode_buffer(ip, &bh); 274 if (error) 275 goto out_trans_end; 276 gfs2_trans_add_meta(ip->i_gl, bh); 277 ip->i_diskflags = new_flags; 278 gfs2_dinode_out(ip, bh->b_data); 279 brelse(bh); 280 gfs2_set_inode_flags(inode); 281 gfs2_set_aops(inode); 282 out_trans_end: 283 gfs2_trans_end(sdp); 284 out: 285 gfs2_glock_dq_uninit(&gh); 286 out_drop_write: 287 mnt_drop_write_file(filp); 288 return error; 289 } 290 gfs2_set_flags(struct file * filp,u32 __user * ptr)291 static int gfs2_set_flags(struct file *filp, u32 __user *ptr) 292 { 293 struct inode *inode = file_inode(filp); 294 u32 fsflags, gfsflags; 295 296 if (get_user(fsflags, ptr)) 297 return -EFAULT; 298 299 gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags); 300 if (!S_ISDIR(inode->i_mode)) { 301 gfsflags &= ~GFS2_DIF_TOPDIR; 302 if (gfsflags & GFS2_DIF_INHERIT_JDATA) 303 gfsflags ^= (GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA); 304 return do_gfs2_set_flags(filp, gfsflags, ~GFS2_DIF_SYSTEM); 305 } 306 return do_gfs2_set_flags(filp, gfsflags, ~(GFS2_DIF_SYSTEM | GFS2_DIF_JDATA)); 307 } 308 gfs2_ioctl(struct file * filp,unsigned int cmd,unsigned long arg)309 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) 310 { 311 switch(cmd) { 312 case FS_IOC_GETFLAGS: 313 return gfs2_get_flags(filp, (u32 __user *)arg); 314 case FS_IOC_SETFLAGS: 315 return gfs2_set_flags(filp, (u32 __user *)arg); 316 case FITRIM: 317 return gfs2_fitrim(filp, (void __user *)arg); 318 } 319 return -ENOTTY; 320 } 321 322 /** 323 * gfs2_size_hint - Give a hint to the size of a write request 324 * @filep: The struct file 325 * @offset: The file offset of the write 326 * @size: The length of the write 327 * 328 * When we are about to do a write, this function records the total 329 * write size in order to provide a suitable hint to the lower layers 330 * about how many blocks will be required. 331 * 332 */ 333 gfs2_size_hint(struct file * filep,loff_t offset,size_t size)334 static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size) 335 { 336 struct inode *inode = file_inode(filep); 337 struct gfs2_sbd *sdp = GFS2_SB(inode); 338 struct gfs2_inode *ip = GFS2_I(inode); 339 size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift; 340 int hint = min_t(size_t, INT_MAX, blks); 341 342 if (hint > atomic_read(&ip->i_res.rs_sizehint)) 343 atomic_set(&ip->i_res.rs_sizehint, hint); 344 } 345 346 /** 347 * gfs2_allocate_page_backing - Use bmap to allocate blocks 348 * @page: The (locked) page to allocate backing for 349 * 350 * We try to allocate all the blocks required for the page in 351 * one go. This might fail for various reasons, so we keep 352 * trying until all the blocks to back this page are allocated. 353 * If some of the blocks are already allocated, thats ok too. 354 */ 355 gfs2_allocate_page_backing(struct page * page)356 static int gfs2_allocate_page_backing(struct page *page) 357 { 358 struct inode *inode = page->mapping->host; 359 struct buffer_head bh; 360 unsigned long size = PAGE_SIZE; 361 u64 lblock = page->index << (PAGE_SHIFT - inode->i_blkbits); 362 363 do { 364 bh.b_state = 0; 365 bh.b_size = size; 366 gfs2_block_map(inode, lblock, &bh, 1); 367 if (!buffer_mapped(&bh)) 368 return -EIO; 369 size -= bh.b_size; 370 lblock += (bh.b_size >> inode->i_blkbits); 371 } while(size > 0); 372 return 0; 373 } 374 375 /** 376 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable 377 * @vma: The virtual memory area 378 * @vmf: The virtual memory fault containing the page to become writable 379 * 380 * When the page becomes writable, we need to ensure that we have 381 * blocks allocated on disk to back that page. 382 */ 383 gfs2_page_mkwrite(struct vm_area_struct * vma,struct vm_fault * vmf)384 static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) 385 { 386 struct page *page = vmf->page; 387 struct inode *inode = file_inode(vma->vm_file); 388 struct gfs2_inode *ip = GFS2_I(inode); 389 struct gfs2_sbd *sdp = GFS2_SB(inode); 390 struct gfs2_alloc_parms ap = { .aflags = 0, }; 391 unsigned long last_index; 392 u64 pos = page->index << PAGE_SHIFT; 393 unsigned int data_blocks, ind_blocks, rblocks; 394 struct gfs2_holder gh; 395 loff_t size; 396 int ret; 397 398 sb_start_pagefault(inode->i_sb); 399 400 ret = gfs2_rsqa_alloc(ip); 401 if (ret) 402 goto out; 403 404 gfs2_size_hint(vma->vm_file, pos, PAGE_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 /* Update file times before taking page lock */ 412 file_update_time(vma->vm_file); 413 414 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags); 415 set_bit(GIF_SW_PAGED, &ip->i_flags); 416 417 if (!gfs2_write_alloc_required(ip, pos, PAGE_SIZE)) { 418 lock_page(page); 419 if (!PageUptodate(page) || page->mapping != inode->i_mapping) { 420 ret = -EAGAIN; 421 unlock_page(page); 422 } 423 goto out_unlock; 424 } 425 426 ret = gfs2_rindex_update(sdp); 427 if (ret) 428 goto out_unlock; 429 430 gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks); 431 ap.target = data_blocks + ind_blocks; 432 ret = gfs2_quota_lock_check(ip, &ap); 433 if (ret) 434 goto out_unlock; 435 ret = gfs2_inplace_reserve(ip, &ap); 436 if (ret) 437 goto out_quota_unlock; 438 439 rblocks = RES_DINODE + ind_blocks; 440 if (gfs2_is_jdata(ip)) 441 rblocks += data_blocks ? data_blocks : 1; 442 if (ind_blocks || data_blocks) { 443 rblocks += RES_STATFS + RES_QUOTA; 444 rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks); 445 } 446 ret = gfs2_trans_begin(sdp, rblocks, 0); 447 if (ret) 448 goto out_trans_fail; 449 450 lock_page(page); 451 ret = -EINVAL; 452 size = i_size_read(inode); 453 last_index = (size - 1) >> PAGE_SHIFT; 454 /* Check page index against inode size */ 455 if (size == 0 || (page->index > last_index)) 456 goto out_trans_end; 457 458 ret = -EAGAIN; 459 /* If truncated, we must retry the operation, we may have raced 460 * with the glock demotion code. 461 */ 462 if (!PageUptodate(page) || page->mapping != inode->i_mapping) 463 goto out_trans_end; 464 465 /* Unstuff, if required, and allocate backing blocks for page */ 466 ret = 0; 467 if (gfs2_is_stuffed(ip)) 468 ret = gfs2_unstuff_dinode(ip, page); 469 if (ret == 0) 470 ret = gfs2_allocate_page_backing(page); 471 472 out_trans_end: 473 if (ret) 474 unlock_page(page); 475 gfs2_trans_end(sdp); 476 out_trans_fail: 477 gfs2_inplace_release(ip); 478 out_quota_unlock: 479 gfs2_quota_unlock(ip); 480 out_unlock: 481 gfs2_glock_dq(&gh); 482 out_uninit: 483 gfs2_holder_uninit(&gh); 484 if (ret == 0) { 485 set_page_dirty(page); 486 wait_for_stable_page(page); 487 } 488 out: 489 sb_end_pagefault(inode->i_sb); 490 return block_page_mkwrite_return(ret); 491 } 492 493 static const struct vm_operations_struct gfs2_vm_ops = { 494 .fault = filemap_fault, 495 .map_pages = filemap_map_pages, 496 .page_mkwrite = gfs2_page_mkwrite, 497 }; 498 499 /** 500 * gfs2_mmap - 501 * @file: The file to map 502 * @vma: The VMA which described the mapping 503 * 504 * There is no need to get a lock here unless we should be updating 505 * atime. We ignore any locking errors since the only consequence is 506 * a missed atime update (which will just be deferred until later). 507 * 508 * Returns: 0 509 */ 510 gfs2_mmap(struct file * file,struct vm_area_struct * vma)511 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma) 512 { 513 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 514 515 if (!(file->f_flags & O_NOATIME) && 516 !IS_NOATIME(&ip->i_inode)) { 517 struct gfs2_holder i_gh; 518 int error; 519 520 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, 521 &i_gh); 522 if (error) 523 return error; 524 /* grab lock to update inode */ 525 gfs2_glock_dq_uninit(&i_gh); 526 file_accessed(file); 527 } 528 vma->vm_ops = &gfs2_vm_ops; 529 530 return 0; 531 } 532 533 /** 534 * gfs2_open_common - This is common to open and atomic_open 535 * @inode: The inode being opened 536 * @file: The file being opened 537 * 538 * This maybe called under a glock or not depending upon how it has 539 * been called. We must always be called under a glock for regular 540 * files, however. For other file types, it does not matter whether 541 * we hold the glock or not. 542 * 543 * Returns: Error code or 0 for success 544 */ 545 gfs2_open_common(struct inode * inode,struct file * file)546 int gfs2_open_common(struct inode *inode, struct file *file) 547 { 548 struct gfs2_file *fp; 549 int ret; 550 551 if (S_ISREG(inode->i_mode)) { 552 ret = generic_file_open(inode, file); 553 if (ret) 554 return ret; 555 } 556 557 fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS); 558 if (!fp) 559 return -ENOMEM; 560 561 mutex_init(&fp->f_fl_mutex); 562 563 gfs2_assert_warn(GFS2_SB(inode), !file->private_data); 564 file->private_data = fp; 565 return 0; 566 } 567 568 /** 569 * gfs2_open - open a file 570 * @inode: the inode to open 571 * @file: the struct file for this opening 572 * 573 * After atomic_open, this function is only used for opening files 574 * which are already cached. We must still get the glock for regular 575 * files to ensure that we have the file size uptodate for the large 576 * file check which is in the common code. That is only an issue for 577 * regular files though. 578 * 579 * Returns: errno 580 */ 581 gfs2_open(struct inode * inode,struct file * file)582 static int gfs2_open(struct inode *inode, struct file *file) 583 { 584 struct gfs2_inode *ip = GFS2_I(inode); 585 struct gfs2_holder i_gh; 586 int error; 587 bool need_unlock = false; 588 589 if (S_ISREG(ip->i_inode.i_mode)) { 590 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, 591 &i_gh); 592 if (error) 593 return error; 594 need_unlock = true; 595 } 596 597 error = gfs2_open_common(inode, file); 598 599 if (need_unlock) 600 gfs2_glock_dq_uninit(&i_gh); 601 602 return error; 603 } 604 605 /** 606 * gfs2_release - called to close a struct file 607 * @inode: the inode the struct file belongs to 608 * @file: the struct file being closed 609 * 610 * Returns: errno 611 */ 612 gfs2_release(struct inode * inode,struct file * file)613 static int gfs2_release(struct inode *inode, struct file *file) 614 { 615 struct gfs2_inode *ip = GFS2_I(inode); 616 617 kfree(file->private_data); 618 file->private_data = NULL; 619 620 if (!(file->f_mode & FMODE_WRITE)) 621 return 0; 622 623 gfs2_rsqa_delete(ip, &inode->i_writecount); 624 return 0; 625 } 626 627 /** 628 * gfs2_fsync - sync the dirty data for a file (across the cluster) 629 * @file: the file that points to the dentry 630 * @start: the start position in the file to sync 631 * @end: the end position in the file to sync 632 * @datasync: set if we can ignore timestamp changes 633 * 634 * We split the data flushing here so that we don't wait for the data 635 * until after we've also sent the metadata to disk. Note that for 636 * data=ordered, we will write & wait for the data at the log flush 637 * stage anyway, so this is unlikely to make much of a difference 638 * except in the data=writeback case. 639 * 640 * If the fdatawrite fails due to any reason except -EIO, we will 641 * continue the remainder of the fsync, although we'll still report 642 * the error at the end. This is to match filemap_write_and_wait_range() 643 * behaviour. 644 * 645 * Returns: errno 646 */ 647 gfs2_fsync(struct file * file,loff_t start,loff_t end,int datasync)648 static int gfs2_fsync(struct file *file, loff_t start, loff_t end, 649 int datasync) 650 { 651 struct address_space *mapping = file->f_mapping; 652 struct inode *inode = mapping->host; 653 int sync_state = inode->i_state & I_DIRTY_ALL; 654 struct gfs2_inode *ip = GFS2_I(inode); 655 int ret = 0, ret1 = 0; 656 657 if (mapping->nrpages) { 658 ret1 = filemap_fdatawrite_range(mapping, start, end); 659 if (ret1 == -EIO) 660 return ret1; 661 } 662 663 if (!gfs2_is_jdata(ip)) 664 sync_state &= ~I_DIRTY_PAGES; 665 if (datasync) 666 sync_state &= ~(I_DIRTY_SYNC | I_DIRTY_TIME); 667 668 if (sync_state) { 669 ret = sync_inode_metadata(inode, 1); 670 if (ret) 671 return ret; 672 if (gfs2_is_jdata(ip)) 673 filemap_write_and_wait(mapping); 674 gfs2_ail_flush(ip->i_gl, 1); 675 } 676 677 if (mapping->nrpages) 678 ret = filemap_fdatawait_range(mapping, start, end); 679 680 return ret ? ret : ret1; 681 } 682 683 /** 684 * gfs2_file_write_iter - Perform a write to a file 685 * @iocb: The io context 686 * @iov: The data to write 687 * @nr_segs: Number of @iov segments 688 * @pos: The file position 689 * 690 * We have to do a lock/unlock here to refresh the inode size for 691 * O_APPEND writes, otherwise we can land up writing at the wrong 692 * offset. There is still a race, but provided the app is using its 693 * own file locking, this will make O_APPEND work as expected. 694 * 695 */ 696 gfs2_file_write_iter(struct kiocb * iocb,struct iov_iter * from)697 static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from) 698 { 699 struct file *file = iocb->ki_filp; 700 struct gfs2_inode *ip = GFS2_I(file_inode(file)); 701 int ret; 702 703 ret = gfs2_rsqa_alloc(ip); 704 if (ret) 705 return ret; 706 707 gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from)); 708 709 if (iocb->ki_flags & IOCB_APPEND) { 710 struct gfs2_holder gh; 711 712 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh); 713 if (ret) 714 return ret; 715 gfs2_glock_dq_uninit(&gh); 716 } 717 718 return generic_file_write_iter(iocb, from); 719 } 720 fallocate_chunk(struct inode * inode,loff_t offset,loff_t len,int mode)721 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len, 722 int mode) 723 { 724 struct gfs2_inode *ip = GFS2_I(inode); 725 struct buffer_head *dibh; 726 int error; 727 unsigned int nr_blks; 728 sector_t lblock = offset >> inode->i_blkbits; 729 730 error = gfs2_meta_inode_buffer(ip, &dibh); 731 if (unlikely(error)) 732 return error; 733 734 gfs2_trans_add_meta(ip->i_gl, dibh); 735 736 if (gfs2_is_stuffed(ip)) { 737 error = gfs2_unstuff_dinode(ip, NULL); 738 if (unlikely(error)) 739 goto out; 740 } 741 742 while (len) { 743 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 }; 744 bh_map.b_size = len; 745 set_buffer_zeronew(&bh_map); 746 747 error = gfs2_block_map(inode, lblock, &bh_map, 1); 748 if (unlikely(error)) 749 goto out; 750 len -= bh_map.b_size; 751 nr_blks = bh_map.b_size >> inode->i_blkbits; 752 lblock += nr_blks; 753 if (!buffer_new(&bh_map)) 754 continue; 755 if (unlikely(!buffer_zeronew(&bh_map))) { 756 error = -EIO; 757 goto out; 758 } 759 } 760 out: 761 brelse(dibh); 762 return error; 763 } 764 /** 765 * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of 766 * blocks, determine how many bytes can be written. 767 * @ip: The inode in question. 768 * @len: Max cap of bytes. What we return in *len must be <= this. 769 * @data_blocks: Compute and return the number of data blocks needed 770 * @ind_blocks: Compute and return the number of indirect blocks needed 771 * @max_blocks: The total blocks available to work with. 772 * 773 * Returns: void, but @len, @data_blocks and @ind_blocks are filled in. 774 */ calc_max_reserv(struct gfs2_inode * ip,loff_t * len,unsigned int * data_blocks,unsigned int * ind_blocks,unsigned int max_blocks)775 static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len, 776 unsigned int *data_blocks, unsigned int *ind_blocks, 777 unsigned int max_blocks) 778 { 779 loff_t max = *len; 780 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 781 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1); 782 783 for (tmp = max_data; tmp > sdp->sd_diptrs;) { 784 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs); 785 max_data -= tmp; 786 } 787 788 *data_blocks = max_data; 789 *ind_blocks = max_blocks - max_data; 790 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift; 791 if (*len > max) { 792 *len = max; 793 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks); 794 } 795 } 796 __gfs2_fallocate(struct file * file,int mode,loff_t offset,loff_t len)797 static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len) 798 { 799 struct inode *inode = file_inode(file); 800 struct gfs2_sbd *sdp = GFS2_SB(inode); 801 struct gfs2_inode *ip = GFS2_I(inode); 802 struct gfs2_alloc_parms ap = { .aflags = 0, }; 803 unsigned int data_blocks = 0, ind_blocks = 0, rblocks; 804 loff_t bytes, max_bytes, max_blks = UINT_MAX; 805 int error; 806 const loff_t pos = offset; 807 const loff_t count = len; 808 loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1); 809 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift; 810 loff_t max_chunk_size = UINT_MAX & bsize_mask; 811 812 next = (next + 1) << sdp->sd_sb.sb_bsize_shift; 813 814 offset &= bsize_mask; 815 816 len = next - offset; 817 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2; 818 if (!bytes) 819 bytes = UINT_MAX; 820 bytes &= bsize_mask; 821 if (bytes == 0) 822 bytes = sdp->sd_sb.sb_bsize; 823 824 gfs2_size_hint(file, offset, len); 825 826 gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks); 827 ap.min_target = data_blocks + ind_blocks; 828 829 while (len > 0) { 830 if (len < bytes) 831 bytes = len; 832 if (!gfs2_write_alloc_required(ip, offset, bytes)) { 833 len -= bytes; 834 offset += bytes; 835 continue; 836 } 837 838 /* We need to determine how many bytes we can actually 839 * fallocate without exceeding quota or going over the 840 * end of the fs. We start off optimistically by assuming 841 * we can write max_bytes */ 842 max_bytes = (len > max_chunk_size) ? max_chunk_size : len; 843 844 /* Since max_bytes is most likely a theoretical max, we 845 * calculate a more realistic 'bytes' to serve as a good 846 * starting point for the number of bytes we may be able 847 * to write */ 848 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks); 849 ap.target = data_blocks + ind_blocks; 850 851 error = gfs2_quota_lock_check(ip, &ap); 852 if (error) 853 return error; 854 /* ap.allowed tells us how many blocks quota will allow 855 * us to write. Check if this reduces max_blks */ 856 if (ap.allowed && ap.allowed < max_blks) 857 max_blks = ap.allowed; 858 859 error = gfs2_inplace_reserve(ip, &ap); 860 if (error) 861 goto out_qunlock; 862 863 /* check if the selected rgrp limits our max_blks further */ 864 if (ap.allowed && ap.allowed < max_blks) 865 max_blks = ap.allowed; 866 867 /* Almost done. Calculate bytes that can be written using 868 * max_blks. We also recompute max_bytes, data_blocks and 869 * ind_blocks */ 870 calc_max_reserv(ip, &max_bytes, &data_blocks, 871 &ind_blocks, max_blks); 872 873 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA + 874 RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks); 875 if (gfs2_is_jdata(ip)) 876 rblocks += data_blocks ? data_blocks : 1; 877 878 error = gfs2_trans_begin(sdp, rblocks, 879 PAGE_SIZE/sdp->sd_sb.sb_bsize); 880 if (error) 881 goto out_trans_fail; 882 883 error = fallocate_chunk(inode, offset, max_bytes, mode); 884 gfs2_trans_end(sdp); 885 886 if (error) 887 goto out_trans_fail; 888 889 len -= max_bytes; 890 offset += max_bytes; 891 gfs2_inplace_release(ip); 892 gfs2_quota_unlock(ip); 893 } 894 895 if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size) { 896 i_size_write(inode, pos + count); 897 file_update_time(file); 898 mark_inode_dirty(inode); 899 } 900 901 if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host)) 902 return vfs_fsync_range(file, pos, pos + count - 1, 903 (file->f_flags & __O_SYNC) ? 0 : 1); 904 return 0; 905 906 out_trans_fail: 907 gfs2_inplace_release(ip); 908 out_qunlock: 909 gfs2_quota_unlock(ip); 910 return error; 911 } 912 gfs2_fallocate(struct file * file,int mode,loff_t offset,loff_t len)913 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len) 914 { 915 struct inode *inode = file_inode(file); 916 struct gfs2_inode *ip = GFS2_I(inode); 917 struct gfs2_holder gh; 918 int ret; 919 920 if ((mode & ~FALLOC_FL_KEEP_SIZE) || gfs2_is_jdata(ip)) 921 return -EOPNOTSUPP; 922 923 inode_lock(inode); 924 925 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 926 ret = gfs2_glock_nq(&gh); 927 if (ret) 928 goto out_uninit; 929 930 if (!(mode & FALLOC_FL_KEEP_SIZE) && 931 (offset + len) > inode->i_size) { 932 ret = inode_newsize_ok(inode, offset + len); 933 if (ret) 934 goto out_unlock; 935 } 936 937 ret = get_write_access(inode); 938 if (ret) 939 goto out_unlock; 940 941 ret = gfs2_rsqa_alloc(ip); 942 if (ret) 943 goto out_putw; 944 945 ret = __gfs2_fallocate(file, mode, offset, len); 946 if (ret) 947 gfs2_rs_deltree(&ip->i_res); 948 949 out_putw: 950 put_write_access(inode); 951 out_unlock: 952 gfs2_glock_dq(&gh); 953 out_uninit: 954 gfs2_holder_uninit(&gh); 955 inode_unlock(inode); 956 return ret; 957 } 958 gfs2_file_splice_write(struct pipe_inode_info * pipe,struct file * out,loff_t * ppos,size_t len,unsigned int flags)959 static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe, 960 struct file *out, loff_t *ppos, 961 size_t len, unsigned int flags) 962 { 963 int error; 964 struct gfs2_inode *ip = GFS2_I(out->f_mapping->host); 965 966 error = gfs2_rsqa_alloc(ip); 967 if (error) 968 return (ssize_t)error; 969 970 gfs2_size_hint(out, *ppos, len); 971 972 return iter_file_splice_write(pipe, out, ppos, len, flags); 973 } 974 975 #ifdef CONFIG_GFS2_FS_LOCKING_DLM 976 977 /** 978 * gfs2_lock - acquire/release a posix lock on a file 979 * @file: the file pointer 980 * @cmd: either modify or retrieve lock state, possibly wait 981 * @fl: type and range of lock 982 * 983 * Returns: errno 984 */ 985 gfs2_lock(struct file * file,int cmd,struct file_lock * fl)986 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl) 987 { 988 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); 989 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host); 990 struct lm_lockstruct *ls = &sdp->sd_lockstruct; 991 992 if (!(fl->fl_flags & FL_POSIX)) 993 return -ENOLCK; 994 if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK) 995 return -ENOLCK; 996 997 if (cmd == F_CANCELLK) { 998 /* Hack: */ 999 cmd = F_SETLK; 1000 fl->fl_type = F_UNLCK; 1001 } 1002 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) { 1003 if (fl->fl_type == F_UNLCK) 1004 locks_lock_file_wait(file, fl); 1005 return -EIO; 1006 } 1007 if (IS_GETLK(cmd)) 1008 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl); 1009 else if (fl->fl_type == F_UNLCK) 1010 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl); 1011 else 1012 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl); 1013 } 1014 do_flock(struct file * file,int cmd,struct file_lock * fl)1015 static int do_flock(struct file *file, int cmd, struct file_lock *fl) 1016 { 1017 struct gfs2_file *fp = file->private_data; 1018 struct gfs2_holder *fl_gh = &fp->f_fl_gh; 1019 struct gfs2_inode *ip = GFS2_I(file_inode(file)); 1020 struct gfs2_glock *gl; 1021 unsigned int state; 1022 u16 flags; 1023 int error = 0; 1024 int sleeptime; 1025 1026 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED; 1027 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT; 1028 1029 mutex_lock(&fp->f_fl_mutex); 1030 1031 gl = fl_gh->gh_gl; 1032 if (gl) { 1033 if (fl_gh->gh_state == state) 1034 goto out; 1035 locks_lock_file_wait(file, 1036 &(struct file_lock) { 1037 .fl_type = F_UNLCK, 1038 .fl_flags = FL_FLOCK 1039 }); 1040 gfs2_glock_dq(fl_gh); 1041 gfs2_holder_reinit(state, flags, fl_gh); 1042 } else { 1043 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr, 1044 &gfs2_flock_glops, CREATE, &gl); 1045 if (error) 1046 goto out; 1047 gfs2_holder_init(gl, state, flags, fl_gh); 1048 gfs2_glock_put(gl); 1049 } 1050 for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) { 1051 error = gfs2_glock_nq(fl_gh); 1052 if (error != GLR_TRYFAILED) 1053 break; 1054 fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT; 1055 fl_gh->gh_error = 0; 1056 msleep(sleeptime); 1057 } 1058 if (error) { 1059 gfs2_holder_uninit(fl_gh); 1060 if (error == GLR_TRYFAILED) 1061 error = -EAGAIN; 1062 } else { 1063 error = locks_lock_file_wait(file, fl); 1064 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); 1065 } 1066 1067 out: 1068 mutex_unlock(&fp->f_fl_mutex); 1069 return error; 1070 } 1071 do_unflock(struct file * file,struct file_lock * fl)1072 static void do_unflock(struct file *file, struct file_lock *fl) 1073 { 1074 struct gfs2_file *fp = file->private_data; 1075 struct gfs2_holder *fl_gh = &fp->f_fl_gh; 1076 1077 mutex_lock(&fp->f_fl_mutex); 1078 locks_lock_file_wait(file, fl); 1079 if (gfs2_holder_initialized(fl_gh)) { 1080 gfs2_glock_dq(fl_gh); 1081 gfs2_holder_uninit(fl_gh); 1082 } 1083 mutex_unlock(&fp->f_fl_mutex); 1084 } 1085 1086 /** 1087 * gfs2_flock - acquire/release a flock lock on a file 1088 * @file: the file pointer 1089 * @cmd: either modify or retrieve lock state, possibly wait 1090 * @fl: type and range of lock 1091 * 1092 * Returns: errno 1093 */ 1094 gfs2_flock(struct file * file,int cmd,struct file_lock * fl)1095 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl) 1096 { 1097 if (!(fl->fl_flags & FL_FLOCK)) 1098 return -ENOLCK; 1099 if (fl->fl_type & LOCK_MAND) 1100 return -EOPNOTSUPP; 1101 1102 if (fl->fl_type == F_UNLCK) { 1103 do_unflock(file, fl); 1104 return 0; 1105 } else { 1106 return do_flock(file, cmd, fl); 1107 } 1108 } 1109 1110 const struct file_operations gfs2_file_fops = { 1111 .llseek = gfs2_llseek, 1112 .read_iter = generic_file_read_iter, 1113 .write_iter = gfs2_file_write_iter, 1114 .unlocked_ioctl = gfs2_ioctl, 1115 .mmap = gfs2_mmap, 1116 .open = gfs2_open, 1117 .release = gfs2_release, 1118 .fsync = gfs2_fsync, 1119 .lock = gfs2_lock, 1120 .flock = gfs2_flock, 1121 .splice_read = generic_file_splice_read, 1122 .splice_write = gfs2_file_splice_write, 1123 .setlease = simple_nosetlease, 1124 .fallocate = gfs2_fallocate, 1125 }; 1126 1127 const struct file_operations gfs2_dir_fops = { 1128 .iterate_shared = gfs2_readdir, 1129 .unlocked_ioctl = gfs2_ioctl, 1130 .open = gfs2_open, 1131 .release = gfs2_release, 1132 .fsync = gfs2_fsync, 1133 .lock = gfs2_lock, 1134 .flock = gfs2_flock, 1135 .llseek = default_llseek, 1136 }; 1137 1138 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */ 1139 1140 const struct file_operations gfs2_file_fops_nolock = { 1141 .llseek = gfs2_llseek, 1142 .read_iter = generic_file_read_iter, 1143 .write_iter = gfs2_file_write_iter, 1144 .unlocked_ioctl = gfs2_ioctl, 1145 .mmap = gfs2_mmap, 1146 .open = gfs2_open, 1147 .release = gfs2_release, 1148 .fsync = gfs2_fsync, 1149 .splice_read = generic_file_splice_read, 1150 .splice_write = gfs2_file_splice_write, 1151 .setlease = generic_setlease, 1152 .fallocate = gfs2_fallocate, 1153 }; 1154 1155 const struct file_operations gfs2_dir_fops_nolock = { 1156 .iterate_shared = gfs2_readdir, 1157 .unlocked_ioctl = gfs2_ioctl, 1158 .open = gfs2_open, 1159 .release = gfs2_release, 1160 .fsync = gfs2_fsync, 1161 .llseek = default_llseek, 1162 }; 1163 1164