fs/gfs2/file.c

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
gfs2_llseek
gfs2_readdir
gfs2_gfsflags_to_fsflags
gfs2_get_flags
gfs2_set_inode_flags
do_gfs2_set_flags
gfs2_set_flags
gfs2_getlabel
gfs2_ioctl
gfs2_compat_ioctl
gfs2_size_hint
gfs2_allocate_page_backing
gfs2_page_mkwrite
gfs2_mmap
gfs2_open_common
gfs2_open
gfs2_release
gfs2_fsync
gfs2_file_direct_read
gfs2_file_direct_write
gfs2_file_read_iter
gfs2_file_write_iter
fallocate_chunk
calc_max_reserv
__gfs2_fallocate
gfs2_fallocate
gfs2_file_splice_write
gfs2_lock
do_flock
do_unflock
gfs2_flock
   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/compat.h>
  10 #include <linux/completion.h>
  11 #include <linux/buffer_head.h>
  12 #include <linux/pagemap.h>
  13 #include <linux/uio.h>
  14 #include <linux/blkdev.h>
  15 #include <linux/mm.h>
  16 #include <linux/mount.h>
  17 #include <linux/fs.h>
  18 #include <linux/gfs2_ondisk.h>
  19 #include <linux/falloc.h>
  20 #include <linux/swap.h>
  21 #include <linux/crc32.h>
  22 #include <linux/writeback.h>
  23 #include <linux/uaccess.h>
  24 #include <linux/dlm.h>
  25 #include <linux/dlm_plock.h>
  26 #include <linux/delay.h>
  27 #include <linux/backing-dev.h>
  28 
  29 #include "gfs2.h"
  30 #include "incore.h"
  31 #include "bmap.h"
  32 #include "aops.h"
  33 #include "dir.h"
  34 #include "glock.h"
  35 #include "glops.h"
  36 #include "inode.h"
  37 #include "log.h"
  38 #include "meta_io.h"
  39 #include "quota.h"
  40 #include "rgrp.h"
  41 #include "trans.h"
  42 #include "util.h"
  43 
  44 /**
  45  * gfs2_llseek - seek to a location in a file
  46  * @file: the file
  47  * @offset: the offset
  48  * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
  49  *
  50  * SEEK_END requires the glock for the file because it references the
  51  * file's size.
  52  *
  53  * Returns: The new offset, or errno
  54  */
  55 
  56 static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence)
  57 {
  58         struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
  59         struct gfs2_holder i_gh;
  60         loff_t error;
  61 
  62         switch (whence) {
  63         case SEEK_END:
  64                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
  65                                            &i_gh);
  66                 if (!error) {
  67                         error = generic_file_llseek(file, offset, whence);
  68                         gfs2_glock_dq_uninit(&i_gh);
  69                 }
  70                 break;
  71 
  72         case SEEK_DATA:
  73                 error = gfs2_seek_data(file, offset);
  74                 break;
  75 
  76         case SEEK_HOLE:
  77                 error = gfs2_seek_hole(file, offset);
  78                 break;
  79 
  80         case SEEK_CUR:
  81         case SEEK_SET:
  82                 /*
  83                  * These don't reference inode->i_size and don't depend on the
  84                  * block mapping, so we don't need the glock.
  85                  */
  86                 error = generic_file_llseek(file, offset, whence);
  87                 break;
  88         default:
  89                 error = -EINVAL;
  90         }
  91 
  92         return error;
  93 }
  94 
  95 /**
  96  * gfs2_readdir - Iterator for a directory
  97  * @file: The directory to read from
  98  * @ctx: What to feed directory entries to
  99  *
 100  * Returns: errno
 101  */
 102 
 103 static int gfs2_readdir(struct file *file, struct dir_context *ctx)
 104 {
 105         struct inode *dir = file->f_mapping->host;
 106         struct gfs2_inode *dip = GFS2_I(dir);
 107         struct gfs2_holder d_gh;
 108         int error;
 109 
 110         error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
 111         if (error)
 112                 return error;
 113 
 114         error = gfs2_dir_read(dir, ctx, &file->f_ra);
 115 
 116         gfs2_glock_dq_uninit(&d_gh);
 117 
 118         return error;
 119 }
 120 
 121 /**
 122  * fsflag_gfs2flag
 123  *
 124  * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories,
 125  * and to GFS2_DIF_JDATA for non-directories.
 126  */
 127 static struct {
 128         u32 fsflag;
 129         u32 gfsflag;
 130 } fsflag_gfs2flag[] = {
 131         {FS_SYNC_FL, GFS2_DIF_SYNC},
 132         {FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE},
 133         {FS_APPEND_FL, GFS2_DIF_APPENDONLY},
 134         {FS_NOATIME_FL, GFS2_DIF_NOATIME},
 135         {FS_INDEX_FL, GFS2_DIF_EXHASH},
 136         {FS_TOPDIR_FL, GFS2_DIF_TOPDIR},
 137         {FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA},
 138 };
 139 
 140 static inline u32 gfs2_gfsflags_to_fsflags(struct inode *inode, u32 gfsflags)
 141 {
 142         int i;
 143         u32 fsflags = 0;
 144 
 145         if (S_ISDIR(inode->i_mode))
 146                 gfsflags &= ~GFS2_DIF_JDATA;
 147         else
 148                 gfsflags &= ~GFS2_DIF_INHERIT_JDATA;
 149 
 150         for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++)
 151                 if (gfsflags & fsflag_gfs2flag[i].gfsflag)
 152                         fsflags |= fsflag_gfs2flag[i].fsflag;
 153         return fsflags;
 154 }
 155 
 156 static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
 157 {
 158         struct inode *inode = file_inode(filp);
 159         struct gfs2_inode *ip = GFS2_I(inode);
 160         struct gfs2_holder gh;
 161         int error;
 162         u32 fsflags;
 163 
 164         gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 165         error = gfs2_glock_nq(&gh);
 166         if (error)
 167                 goto out_uninit;
 168 
 169         fsflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags);
 170 
 171         if (put_user(fsflags, ptr))
 172                 error = -EFAULT;
 173 
 174         gfs2_glock_dq(&gh);
 175 out_uninit:
 176         gfs2_holder_uninit(&gh);
 177         return error;
 178 }
 179 
 180 void gfs2_set_inode_flags(struct inode *inode)
 181 {
 182         struct gfs2_inode *ip = GFS2_I(inode);
 183         unsigned int flags = inode->i_flags;
 184 
 185         flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
 186         if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
 187                 flags |= S_NOSEC;
 188         if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
 189                 flags |= S_IMMUTABLE;
 190         if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
 191                 flags |= S_APPEND;
 192         if (ip->i_diskflags & GFS2_DIF_NOATIME)
 193                 flags |= S_NOATIME;
 194         if (ip->i_diskflags & GFS2_DIF_SYNC)
 195                 flags |= S_SYNC;
 196         inode->i_flags = flags;
 197 }
 198 
 199 /* Flags that can be set by user space */
 200 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA|                    \
 201                              GFS2_DIF_IMMUTABLE|                \
 202                              GFS2_DIF_APPENDONLY|               \
 203                              GFS2_DIF_NOATIME|                  \
 204                              GFS2_DIF_SYNC|                     \
 205                              GFS2_DIF_TOPDIR|                   \
 206                              GFS2_DIF_INHERIT_JDATA)
 207 
 208 /**
 209  * do_gfs2_set_flags - set flags on an inode
 210  * @filp: file pointer
 211  * @reqflags: The flags to set
 212  * @mask: Indicates which flags are valid
 213  * @fsflags: The FS_* inode flags passed in
 214  *
 215  */
 216 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask,
 217                              const u32 fsflags)
 218 {
 219         struct inode *inode = file_inode(filp);
 220         struct gfs2_inode *ip = GFS2_I(inode);
 221         struct gfs2_sbd *sdp = GFS2_SB(inode);
 222         struct buffer_head *bh;
 223         struct gfs2_holder gh;
 224         int error;
 225         u32 new_flags, flags, oldflags;
 226 
 227         error = mnt_want_write_file(filp);
 228         if (error)
 229                 return error;
 230 
 231         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 232         if (error)
 233                 goto out_drop_write;
 234 
 235         oldflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags);
 236         error = vfs_ioc_setflags_prepare(inode, oldflags, fsflags);
 237         if (error)
 238                 goto out;
 239 
 240         error = -EACCES;
 241         if (!inode_owner_or_capable(inode))
 242                 goto out;
 243 
 244         error = 0;
 245         flags = ip->i_diskflags;
 246         new_flags = (flags & ~mask) | (reqflags & mask);
 247         if ((new_flags ^ flags) == 0)
 248                 goto out;
 249 
 250         error = -EPERM;
 251         if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
 252                 goto out;
 253         if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
 254                 goto out;
 255         if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
 256             !capable(CAP_LINUX_IMMUTABLE))
 257                 goto out;
 258         if (!IS_IMMUTABLE(inode)) {
 259                 error = gfs2_permission(inode, MAY_WRITE);
 260                 if (error)
 261                         goto out;
 262         }
 263         if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
 264                 if (new_flags & GFS2_DIF_JDATA)
 265                         gfs2_log_flush(sdp, ip->i_gl,
 266                                        GFS2_LOG_HEAD_FLUSH_NORMAL |
 267                                        GFS2_LFC_SET_FLAGS);
 268                 error = filemap_fdatawrite(inode->i_mapping);
 269                 if (error)
 270                         goto out;
 271                 error = filemap_fdatawait(inode->i_mapping);
 272                 if (error)
 273                         goto out;
 274                 if (new_flags & GFS2_DIF_JDATA)
 275                         gfs2_ordered_del_inode(ip);
 276         }
 277         error = gfs2_trans_begin(sdp, RES_DINODE, 0);
 278         if (error)
 279                 goto out;
 280         error = gfs2_meta_inode_buffer(ip, &bh);
 281         if (error)
 282                 goto out_trans_end;
 283         inode->i_ctime = current_time(inode);
 284         gfs2_trans_add_meta(ip->i_gl, bh);
 285         ip->i_diskflags = new_flags;
 286         gfs2_dinode_out(ip, bh->b_data);
 287         brelse(bh);
 288         gfs2_set_inode_flags(inode);
 289         gfs2_set_aops(inode);
 290 out_trans_end:
 291         gfs2_trans_end(sdp);
 292 out:
 293         gfs2_glock_dq_uninit(&gh);
 294 out_drop_write:
 295         mnt_drop_write_file(filp);
 296         return error;
 297 }
 298 
 299 static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
 300 {
 301         struct inode *inode = file_inode(filp);
 302         u32 fsflags, gfsflags = 0;
 303         u32 mask;
 304         int i;
 305 
 306         if (get_user(fsflags, ptr))
 307                 return -EFAULT;
 308 
 309         for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) {
 310                 if (fsflags & fsflag_gfs2flag[i].fsflag) {
 311                         fsflags &= ~fsflag_gfs2flag[i].fsflag;
 312                         gfsflags |= fsflag_gfs2flag[i].gfsflag;
 313                 }
 314         }
 315         if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET)
 316                 return -EINVAL;
 317 
 318         mask = GFS2_FLAGS_USER_SET;
 319         if (S_ISDIR(inode->i_mode)) {
 320                 mask &= ~GFS2_DIF_JDATA;
 321         } else {
 322                 /* The GFS2_DIF_TOPDIR flag is only valid for directories. */
 323                 if (gfsflags & GFS2_DIF_TOPDIR)
 324                         return -EINVAL;
 325                 mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA);
 326         }
 327 
 328         return do_gfs2_set_flags(filp, gfsflags, mask, fsflags);
 329 }
 330 
 331 static int gfs2_getlabel(struct file *filp, char __user *label)
 332 {
 333         struct inode *inode = file_inode(filp);
 334         struct gfs2_sbd *sdp = GFS2_SB(inode);
 335 
 336         if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN))
 337                 return -EFAULT;
 338 
 339         return 0;
 340 }
 341 
 342 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 343 {
 344         switch(cmd) {
 345         case FS_IOC_GETFLAGS:
 346                 return gfs2_get_flags(filp, (u32 __user *)arg);
 347         case FS_IOC_SETFLAGS:
 348                 return gfs2_set_flags(filp, (u32 __user *)arg);
 349         case FITRIM:
 350                 return gfs2_fitrim(filp, (void __user *)arg);
 351         case FS_IOC_GETFSLABEL:
 352                 return gfs2_getlabel(filp, (char __user *)arg);
 353         }
 354 
 355         return -ENOTTY;
 356 }
 357 
 358 #ifdef CONFIG_COMPAT
 359 static long gfs2_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 360 {
 361         switch(cmd) {
 362         /* These are just misnamed, they actually get/put from/to user an int */
 363         case FS_IOC32_GETFLAGS:
 364                 cmd = FS_IOC_GETFLAGS;
 365                 break;
 366         case FS_IOC32_SETFLAGS:
 367                 cmd = FS_IOC_SETFLAGS;
 368                 break;
 369         /* Keep this list in sync with gfs2_ioctl */
 370         case FITRIM:
 371         case FS_IOC_GETFSLABEL:
 372                 break;
 373         default:
 374                 return -ENOIOCTLCMD;
 375         }
 376 
 377         return gfs2_ioctl(filp, cmd, (unsigned long)compat_ptr(arg));
 378 }
 379 #else
 380 #define gfs2_compat_ioctl NULL
 381 #endif
 382 
 383 /**
 384  * gfs2_size_hint - Give a hint to the size of a write request
 385  * @filep: The struct file
 386  * @offset: The file offset of the write
 387  * @size: The length of the write
 388  *
 389  * When we are about to do a write, this function records the total
 390  * write size in order to provide a suitable hint to the lower layers
 391  * about how many blocks will be required.
 392  *
 393  */
 394 
 395 static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
 396 {
 397         struct inode *inode = file_inode(filep);
 398         struct gfs2_sbd *sdp = GFS2_SB(inode);
 399         struct gfs2_inode *ip = GFS2_I(inode);
 400         size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
 401         int hint = min_t(size_t, INT_MAX, blks);
 402 
 403         if (hint > atomic_read(&ip->i_sizehint))
 404                 atomic_set(&ip->i_sizehint, hint);
 405 }
 406 
 407 /**
 408  * gfs2_allocate_page_backing - Allocate blocks for a write fault
 409  * @page: The (locked) page to allocate backing for
 410  * @length: Size of the allocation
 411  *
 412  * We try to allocate all the blocks required for the page in one go.  This
 413  * might fail for various reasons, so we keep trying until all the blocks to
 414  * back this page are allocated.  If some of the blocks are already allocated,
 415  * that is ok too.
 416  */
 417 static int gfs2_allocate_page_backing(struct page *page, unsigned int length)
 418 {
 419         u64 pos = page_offset(page);
 420 
 421         do {
 422                 struct iomap iomap = { };
 423 
 424                 if (gfs2_iomap_get_alloc(page->mapping->host, pos, length, &iomap))
 425                         return -EIO;
 426 
 427                 if (length < iomap.length)
 428                         iomap.length = length;
 429                 length -= iomap.length;
 430                 pos += iomap.length;
 431         } while (length > 0);
 432 
 433         return 0;
 434 }
 435 
 436 /**
 437  * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
 438  * @vma: The virtual memory area
 439  * @vmf: The virtual memory fault containing the page to become writable
 440  *
 441  * When the page becomes writable, we need to ensure that we have
 442  * blocks allocated on disk to back that page.
 443  */
 444 
 445 static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf)
 446 {
 447         struct page *page = vmf->page;
 448         struct inode *inode = file_inode(vmf->vma->vm_file);
 449         struct gfs2_inode *ip = GFS2_I(inode);
 450         struct gfs2_sbd *sdp = GFS2_SB(inode);
 451         struct gfs2_alloc_parms ap = { .aflags = 0, };
 452         unsigned long last_index;
 453         u64 pos = page_offset(page);
 454         unsigned int data_blocks, ind_blocks, rblocks;
 455         struct gfs2_holder gh;
 456         loff_t size;
 457         int ret;
 458 
 459         sb_start_pagefault(inode->i_sb);
 460 
 461         ret = gfs2_rsqa_alloc(ip);
 462         if (ret)
 463                 goto out;
 464 
 465         gfs2_size_hint(vmf->vma->vm_file, pos, PAGE_SIZE);
 466 
 467         gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
 468         ret = gfs2_glock_nq(&gh);
 469         if (ret)
 470                 goto out_uninit;
 471 
 472         /* Update file times before taking page lock */
 473         file_update_time(vmf->vma->vm_file);
 474 
 475         set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
 476         set_bit(GIF_SW_PAGED, &ip->i_flags);
 477 
 478         if (!gfs2_write_alloc_required(ip, pos, PAGE_SIZE)) {
 479                 lock_page(page);
 480                 if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
 481                         ret = -EAGAIN;
 482                         unlock_page(page);
 483                 }
 484                 goto out_unlock;
 485         }
 486 
 487         ret = gfs2_rindex_update(sdp);
 488         if (ret)
 489                 goto out_unlock;
 490 
 491         gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
 492         ap.target = data_blocks + ind_blocks;
 493         ret = gfs2_quota_lock_check(ip, &ap);
 494         if (ret)
 495                 goto out_unlock;
 496         ret = gfs2_inplace_reserve(ip, &ap);
 497         if (ret)
 498                 goto out_quota_unlock;
 499 
 500         rblocks = RES_DINODE + ind_blocks;
 501         if (gfs2_is_jdata(ip))
 502                 rblocks += data_blocks ? data_blocks : 1;
 503         if (ind_blocks || data_blocks) {
 504                 rblocks += RES_STATFS + RES_QUOTA;
 505                 rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
 506         }
 507         ret = gfs2_trans_begin(sdp, rblocks, 0);
 508         if (ret)
 509                 goto out_trans_fail;
 510 
 511         lock_page(page);
 512         ret = -EINVAL;
 513         size = i_size_read(inode);
 514         last_index = (size - 1) >> PAGE_SHIFT;
 515         /* Check page index against inode size */
 516         if (size == 0 || (page->index > last_index))
 517                 goto out_trans_end;
 518 
 519         ret = -EAGAIN;
 520         /* If truncated, we must retry the operation, we may have raced
 521          * with the glock demotion code.
 522          */
 523         if (!PageUptodate(page) || page->mapping != inode->i_mapping)
 524                 goto out_trans_end;
 525 
 526         /* Unstuff, if required, and allocate backing blocks for page */
 527         ret = 0;
 528         if (gfs2_is_stuffed(ip))
 529                 ret = gfs2_unstuff_dinode(ip, page);
 530         if (ret == 0)
 531                 ret = gfs2_allocate_page_backing(page, PAGE_SIZE);
 532 
 533 out_trans_end:
 534         if (ret)
 535                 unlock_page(page);
 536         gfs2_trans_end(sdp);
 537 out_trans_fail:
 538         gfs2_inplace_release(ip);
 539 out_quota_unlock:
 540         gfs2_quota_unlock(ip);
 541 out_unlock:
 542         gfs2_glock_dq(&gh);
 543 out_uninit:
 544         gfs2_holder_uninit(&gh);
 545         if (ret == 0) {
 546                 set_page_dirty(page);
 547                 wait_for_stable_page(page);
 548         }
 549 out:
 550         sb_end_pagefault(inode->i_sb);
 551         return block_page_mkwrite_return(ret);
 552 }
 553 
 554 static const struct vm_operations_struct gfs2_vm_ops = {
 555         .fault = filemap_fault,
 556         .map_pages = filemap_map_pages,
 557         .page_mkwrite = gfs2_page_mkwrite,
 558 };
 559 
 560 /**
 561  * gfs2_mmap -
 562  * @file: The file to map
 563  * @vma: The VMA which described the mapping
 564  *
 565  * There is no need to get a lock here unless we should be updating
 566  * atime. We ignore any locking errors since the only consequence is
 567  * a missed atime update (which will just be deferred until later).
 568  *
 569  * Returns: 0
 570  */
 571 
 572 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
 573 {
 574         struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
 575 
 576         if (!(file->f_flags & O_NOATIME) &&
 577             !IS_NOATIME(&ip->i_inode)) {
 578                 struct gfs2_holder i_gh;
 579                 int error;
 580 
 581                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
 582                                            &i_gh);
 583                 if (error)
 584                         return error;
 585                 /* grab lock to update inode */
 586                 gfs2_glock_dq_uninit(&i_gh);
 587                 file_accessed(file);
 588         }
 589         vma->vm_ops = &gfs2_vm_ops;
 590 
 591         return 0;
 592 }
 593 
 594 /**
 595  * gfs2_open_common - This is common to open and atomic_open
 596  * @inode: The inode being opened
 597  * @file: The file being opened
 598  *
 599  * This maybe called under a glock or not depending upon how it has
 600  * been called. We must always be called under a glock for regular
 601  * files, however. For other file types, it does not matter whether
 602  * we hold the glock or not.
 603  *
 604  * Returns: Error code or 0 for success
 605  */
 606 
 607 int gfs2_open_common(struct inode *inode, struct file *file)
 608 {
 609         struct gfs2_file *fp;
 610         int ret;
 611 
 612         if (S_ISREG(inode->i_mode)) {
 613                 ret = generic_file_open(inode, file);
 614                 if (ret)
 615                         return ret;
 616         }
 617 
 618         fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
 619         if (!fp)
 620                 return -ENOMEM;
 621 
 622         mutex_init(&fp->f_fl_mutex);
 623 
 624         gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
 625         file->private_data = fp;
 626         return 0;
 627 }
 628 
 629 /**
 630  * gfs2_open - open a file
 631  * @inode: the inode to open
 632  * @file: the struct file for this opening
 633  *
 634  * After atomic_open, this function is only used for opening files
 635  * which are already cached. We must still get the glock for regular
 636  * files to ensure that we have the file size uptodate for the large
 637  * file check which is in the common code. That is only an issue for
 638  * regular files though.
 639  *
 640  * Returns: errno
 641  */
 642 
 643 static int gfs2_open(struct inode *inode, struct file *file)
 644 {
 645         struct gfs2_inode *ip = GFS2_I(inode);
 646         struct gfs2_holder i_gh;
 647         int error;
 648         bool need_unlock = false;
 649 
 650         if (S_ISREG(ip->i_inode.i_mode)) {
 651                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
 652                                            &i_gh);
 653                 if (error)
 654                         return error;
 655                 need_unlock = true;
 656         }
 657 
 658         error = gfs2_open_common(inode, file);
 659 
 660         if (need_unlock)
 661                 gfs2_glock_dq_uninit(&i_gh);
 662 
 663         return error;
 664 }
 665 
 666 /**
 667  * gfs2_release - called to close a struct file
 668  * @inode: the inode the struct file belongs to
 669  * @file: the struct file being closed
 670  *
 671  * Returns: errno
 672  */
 673 
 674 static int gfs2_release(struct inode *inode, struct file *file)
 675 {
 676         struct gfs2_inode *ip = GFS2_I(inode);
 677 
 678         kfree(file->private_data);
 679         file->private_data = NULL;
 680 
 681         if (!(file->f_mode & FMODE_WRITE))
 682                 return 0;
 683 
 684         gfs2_rsqa_delete(ip, &inode->i_writecount);
 685         return 0;
 686 }
 687 
 688 /**
 689  * gfs2_fsync - sync the dirty data for a file (across the cluster)
 690  * @file: the file that points to the dentry
 691  * @start: the start position in the file to sync
 692  * @end: the end position in the file to sync
 693  * @datasync: set if we can ignore timestamp changes
 694  *
 695  * We split the data flushing here so that we don't wait for the data
 696  * until after we've also sent the metadata to disk. Note that for
 697  * data=ordered, we will write & wait for the data at the log flush
 698  * stage anyway, so this is unlikely to make much of a difference
 699  * except in the data=writeback case.
 700  *
 701  * If the fdatawrite fails due to any reason except -EIO, we will
 702  * continue the remainder of the fsync, although we'll still report
 703  * the error at the end. This is to match filemap_write_and_wait_range()
 704  * behaviour.
 705  *
 706  * Returns: errno
 707  */
 708 
 709 static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
 710                       int datasync)
 711 {
 712         struct address_space *mapping = file->f_mapping;
 713         struct inode *inode = mapping->host;
 714         int sync_state = inode->i_state & I_DIRTY_ALL;
 715         struct gfs2_inode *ip = GFS2_I(inode);
 716         int ret = 0, ret1 = 0;
 717 
 718         if (mapping->nrpages) {
 719                 ret1 = filemap_fdatawrite_range(mapping, start, end);
 720                 if (ret1 == -EIO)
 721                         return ret1;
 722         }
 723 
 724         if (!gfs2_is_jdata(ip))
 725                 sync_state &= ~I_DIRTY_PAGES;
 726         if (datasync)
 727                 sync_state &= ~(I_DIRTY_SYNC | I_DIRTY_TIME);
 728 
 729         if (sync_state) {
 730                 ret = sync_inode_metadata(inode, 1);
 731                 if (ret)
 732                         return ret;
 733                 if (gfs2_is_jdata(ip))
 734                         ret = file_write_and_wait(file);
 735                 if (ret)
 736                         return ret;
 737                 gfs2_ail_flush(ip->i_gl, 1);
 738         }
 739 
 740         if (mapping->nrpages)
 741                 ret = file_fdatawait_range(file, start, end);
 742 
 743         return ret ? ret : ret1;
 744 }
 745 
 746 static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to)
 747 {
 748         struct file *file = iocb->ki_filp;
 749         struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
 750         size_t count = iov_iter_count(to);
 751         struct gfs2_holder gh;
 752         ssize_t ret;
 753 
 754         if (!count)
 755                 return 0; /* skip atime */
 756 
 757         gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
 758         ret = gfs2_glock_nq(&gh);
 759         if (ret)
 760                 goto out_uninit;
 761 
 762         ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL);
 763 
 764         gfs2_glock_dq(&gh);
 765 out_uninit:
 766         gfs2_holder_uninit(&gh);
 767         return ret;
 768 }
 769 
 770 static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
 771 {
 772         struct file *file = iocb->ki_filp;
 773         struct inode *inode = file->f_mapping->host;
 774         struct gfs2_inode *ip = GFS2_I(inode);
 775         size_t len = iov_iter_count(from);
 776         loff_t offset = iocb->ki_pos;
 777         struct gfs2_holder gh;
 778         ssize_t ret;
 779 
 780         /*
 781          * Deferred lock, even if its a write, since we do no allocation on
 782          * this path. All we need to change is the atime, and this lock mode
 783          * ensures that other nodes have flushed their buffered read caches
 784          * (i.e. their page cache entries for this inode). We do not,
 785          * unfortunately, have the option of only flushing a range like the
 786          * VFS does.
 787          */
 788         gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
 789         ret = gfs2_glock_nq(&gh);
 790         if (ret)
 791                 goto out_uninit;
 792 
 793         /* Silently fall back to buffered I/O when writing beyond EOF */
 794         if (offset + len > i_size_read(&ip->i_inode))
 795                 goto out;
 796 
 797         ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL);
 798 
 799 out:
 800         gfs2_glock_dq(&gh);
 801 out_uninit:
 802         gfs2_holder_uninit(&gh);
 803         return ret;
 804 }
 805 
 806 static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
 807 {
 808         ssize_t ret;
 809 
 810         if (iocb->ki_flags & IOCB_DIRECT) {
 811                 ret = gfs2_file_direct_read(iocb, to);
 812                 if (likely(ret != -ENOTBLK))
 813                         return ret;
 814                 iocb->ki_flags &= ~IOCB_DIRECT;
 815         }
 816         return generic_file_read_iter(iocb, to);
 817 }
 818 
 819 /**
 820  * gfs2_file_write_iter - Perform a write to a file
 821  * @iocb: The io context
 822  * @from: The data to write
 823  *
 824  * We have to do a lock/unlock here to refresh the inode size for
 825  * O_APPEND writes, otherwise we can land up writing at the wrong
 826  * offset. There is still a race, but provided the app is using its
 827  * own file locking, this will make O_APPEND work as expected.
 828  *
 829  */
 830 
 831 static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 832 {
 833         struct file *file = iocb->ki_filp;
 834         struct inode *inode = file_inode(file);
 835         struct gfs2_inode *ip = GFS2_I(inode);
 836         ssize_t ret;
 837 
 838         ret = gfs2_rsqa_alloc(ip);
 839         if (ret)
 840                 return ret;
 841 
 842         gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
 843 
 844         if (iocb->ki_flags & IOCB_APPEND) {
 845                 struct gfs2_holder gh;
 846 
 847                 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
 848                 if (ret)
 849                         return ret;
 850                 gfs2_glock_dq_uninit(&gh);
 851         }
 852 
 853         inode_lock(inode);
 854         ret = generic_write_checks(iocb, from);
 855         if (ret <= 0)
 856                 goto out_unlock;
 857 
 858         ret = file_remove_privs(file);
 859         if (ret)
 860                 goto out_unlock;
 861 
 862         ret = file_update_time(file);
 863         if (ret)
 864                 goto out_unlock;
 865 
 866         if (iocb->ki_flags & IOCB_DIRECT) {
 867                 struct address_space *mapping = file->f_mapping;
 868                 ssize_t buffered, ret2;
 869 
 870                 ret = gfs2_file_direct_write(iocb, from);
 871                 if (ret < 0 || !iov_iter_count(from))
 872                         goto out_unlock;
 873 
 874                 iocb->ki_flags |= IOCB_DSYNC;
 875                 current->backing_dev_info = inode_to_bdi(inode);
 876                 buffered = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
 877                 current->backing_dev_info = NULL;
 878                 if (unlikely(buffered <= 0))
 879                         goto out_unlock;
 880 
 881                 /*
 882                  * We need to ensure that the page cache pages are written to
 883                  * disk and invalidated to preserve the expected O_DIRECT
 884                  * semantics.  If the writeback or invalidate fails, only report
 885                  * the direct I/O range as we don't know if the buffered pages
 886                  * made it to disk.
 887                  */
 888                 iocb->ki_pos += buffered;
 889                 ret2 = generic_write_sync(iocb, buffered);
 890                 invalidate_mapping_pages(mapping,
 891                                 (iocb->ki_pos - buffered) >> PAGE_SHIFT,
 892                                 (iocb->ki_pos - 1) >> PAGE_SHIFT);
 893                 if (!ret || ret2 > 0)
 894                         ret += ret2;
 895         } else {
 896                 current->backing_dev_info = inode_to_bdi(inode);
 897                 ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
 898                 current->backing_dev_info = NULL;
 899                 if (likely(ret > 0)) {
 900                         iocb->ki_pos += ret;
 901                         ret = generic_write_sync(iocb, ret);
 902                 }
 903         }
 904 
 905 out_unlock:
 906         inode_unlock(inode);
 907         return ret;
 908 }
 909 
 910 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
 911                            int mode)
 912 {
 913         struct super_block *sb = inode->i_sb;
 914         struct gfs2_inode *ip = GFS2_I(inode);
 915         loff_t end = offset + len;
 916         struct buffer_head *dibh;
 917         int error;
 918 
 919         error = gfs2_meta_inode_buffer(ip, &dibh);
 920         if (unlikely(error))
 921                 return error;
 922 
 923         gfs2_trans_add_meta(ip->i_gl, dibh);
 924 
 925         if (gfs2_is_stuffed(ip)) {
 926                 error = gfs2_unstuff_dinode(ip, NULL);
 927                 if (unlikely(error))
 928                         goto out;
 929         }
 930 
 931         while (offset < end) {
 932                 struct iomap iomap = { };
 933 
 934                 error = gfs2_iomap_get_alloc(inode, offset, end - offset,
 935                                              &iomap);
 936                 if (error)
 937                         goto out;
 938                 offset = iomap.offset + iomap.length;
 939                 if (!(iomap.flags & IOMAP_F_NEW))
 940                         continue;
 941                 error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
 942                                          iomap.length >> inode->i_blkbits,
 943                                          GFP_NOFS);
 944                 if (error) {
 945                         fs_err(GFS2_SB(inode), "Failed to zero data buffers\n");
 946                         goto out;
 947                 }
 948         }
 949 out:
 950         brelse(dibh);
 951         return error;
 952 }
 953 /**
 954  * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
 955  *                     blocks, determine how many bytes can be written.
 956  * @ip:          The inode in question.
 957  * @len:         Max cap of bytes. What we return in *len must be <= this.
 958  * @data_blocks: Compute and return the number of data blocks needed
 959  * @ind_blocks:  Compute and return the number of indirect blocks needed
 960  * @max_blocks:  The total blocks available to work with.
 961  *
 962  * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
 963  */
 964 static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
 965                             unsigned int *data_blocks, unsigned int *ind_blocks,
 966                             unsigned int max_blocks)
 967 {
 968         loff_t max = *len;
 969         const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 970         unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
 971 
 972         for (tmp = max_data; tmp > sdp->sd_diptrs;) {
 973                 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
 974                 max_data -= tmp;
 975         }
 976 
 977         *data_blocks = max_data;
 978         *ind_blocks = max_blocks - max_data;
 979         *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
 980         if (*len > max) {
 981                 *len = max;
 982                 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
 983         }
 984 }
 985 
 986 static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 987 {
 988         struct inode *inode = file_inode(file);
 989         struct gfs2_sbd *sdp = GFS2_SB(inode);
 990         struct gfs2_inode *ip = GFS2_I(inode);
 991         struct gfs2_alloc_parms ap = { .aflags = 0, };
 992         unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
 993         loff_t bytes, max_bytes, max_blks;
 994         int error;
 995         const loff_t pos = offset;
 996         const loff_t count = len;
 997         loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
 998         loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
 999         loff_t max_chunk_size = UINT_MAX & bsize_mask;
1000 
1001         next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
1002 
1003         offset &= bsize_mask;
1004 
1005         len = next - offset;
1006         bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
1007         if (!bytes)
1008                 bytes = UINT_MAX;
1009         bytes &= bsize_mask;
1010         if (bytes == 0)
1011                 bytes = sdp->sd_sb.sb_bsize;
1012 
1013         gfs2_size_hint(file, offset, len);
1014 
1015         gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
1016         ap.min_target = data_blocks + ind_blocks;
1017 
1018         while (len > 0) {
1019                 if (len < bytes)
1020                         bytes = len;
1021                 if (!gfs2_write_alloc_required(ip, offset, bytes)) {
1022                         len -= bytes;
1023                         offset += bytes;
1024                         continue;
1025                 }
1026 
1027                 /* We need to determine how many bytes we can actually
1028                  * fallocate without exceeding quota or going over the
1029                  * end of the fs. We start off optimistically by assuming
1030                  * we can write max_bytes */
1031                 max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
1032 
1033                 /* Since max_bytes is most likely a theoretical max, we
1034                  * calculate a more realistic 'bytes' to serve as a good
1035                  * starting point for the number of bytes we may be able
1036                  * to write */
1037                 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
1038                 ap.target = data_blocks + ind_blocks;
1039 
1040                 error = gfs2_quota_lock_check(ip, &ap);
1041                 if (error)
1042                         return error;
1043                 /* ap.allowed tells us how many blocks quota will allow
1044                  * us to write. Check if this reduces max_blks */
1045                 max_blks = UINT_MAX;
1046                 if (ap.allowed)
1047                         max_blks = ap.allowed;
1048 
1049                 error = gfs2_inplace_reserve(ip, &ap);
1050                 if (error)
1051                         goto out_qunlock;
1052 
1053                 /* check if the selected rgrp limits our max_blks further */
1054                 if (ap.allowed && ap.allowed < max_blks)
1055                         max_blks = ap.allowed;
1056 
1057                 /* Almost done. Calculate bytes that can be written using
1058                  * max_blks. We also recompute max_bytes, data_blocks and
1059                  * ind_blocks */
1060                 calc_max_reserv(ip, &max_bytes, &data_blocks,
1061                                 &ind_blocks, max_blks);
1062 
1063                 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
1064                           RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
1065                 if (gfs2_is_jdata(ip))
1066                         rblocks += data_blocks ? data_blocks : 1;
1067 
1068                 error = gfs2_trans_begin(sdp, rblocks,
1069                                          PAGE_SIZE >> inode->i_blkbits);
1070                 if (error)
1071                         goto out_trans_fail;
1072 
1073                 error = fallocate_chunk(inode, offset, max_bytes, mode);
1074                 gfs2_trans_end(sdp);
1075 
1076                 if (error)
1077                         goto out_trans_fail;
1078 
1079                 len -= max_bytes;
1080                 offset += max_bytes;
1081                 gfs2_inplace_release(ip);
1082                 gfs2_quota_unlock(ip);
1083         }
1084 
1085         if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size)
1086                 i_size_write(inode, pos + count);
1087         file_update_time(file);
1088         mark_inode_dirty(inode);
1089 
1090         if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host))
1091                 return vfs_fsync_range(file, pos, pos + count - 1,
1092                                (file->f_flags & __O_SYNC) ? 0 : 1);
1093         return 0;
1094 
1095 out_trans_fail:
1096         gfs2_inplace_release(ip);
1097 out_qunlock:
1098         gfs2_quota_unlock(ip);
1099         return error;
1100 }
1101 
1102 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1103 {
1104         struct inode *inode = file_inode(file);
1105         struct gfs2_sbd *sdp = GFS2_SB(inode);
1106         struct gfs2_inode *ip = GFS2_I(inode);
1107         struct gfs2_holder gh;
1108         int ret;
1109 
1110         if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))
1111                 return -EOPNOTSUPP;
1112         /* fallocate is needed by gfs2_grow to reserve space in the rindex */
1113         if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex)
1114                 return -EOPNOTSUPP;
1115 
1116         inode_lock(inode);
1117 
1118         gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1119         ret = gfs2_glock_nq(&gh);
1120         if (ret)
1121                 goto out_uninit;
1122 
1123         if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1124             (offset + len) > inode->i_size) {
1125                 ret = inode_newsize_ok(inode, offset + len);
1126                 if (ret)
1127                         goto out_unlock;
1128         }
1129 
1130         ret = get_write_access(inode);
1131         if (ret)
1132                 goto out_unlock;
1133 
1134         if (mode & FALLOC_FL_PUNCH_HOLE) {
1135                 ret = __gfs2_punch_hole(file, offset, len);
1136         } else {
1137                 ret = gfs2_rsqa_alloc(ip);
1138                 if (ret)
1139                         goto out_putw;
1140 
1141                 ret = __gfs2_fallocate(file, mode, offset, len);
1142 
1143                 if (ret)
1144                         gfs2_rs_deltree(&ip->i_res);
1145         }
1146 
1147 out_putw:
1148         put_write_access(inode);
1149 out_unlock:
1150         gfs2_glock_dq(&gh);
1151 out_uninit:
1152         gfs2_holder_uninit(&gh);
1153         inode_unlock(inode);
1154         return ret;
1155 }
1156 
1157 static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
1158                                       struct file *out, loff_t *ppos,
1159                                       size_t len, unsigned int flags)
1160 {
1161         int error;
1162         struct gfs2_inode *ip = GFS2_I(out->f_mapping->host);
1163 
1164         error = gfs2_rsqa_alloc(ip);
1165         if (error)
1166                 return (ssize_t)error;
1167 
1168         gfs2_size_hint(out, *ppos, len);
1169 
1170         return iter_file_splice_write(pipe, out, ppos, len, flags);
1171 }
1172 
1173 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1174 
1175 /**
1176  * gfs2_lock - acquire/release a posix lock on a file
1177  * @file: the file pointer
1178  * @cmd: either modify or retrieve lock state, possibly wait
1179  * @fl: type and range of lock
1180  *
1181  * Returns: errno
1182  */
1183 
1184 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
1185 {
1186         struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
1187         struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
1188         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1189 
1190         if (!(fl->fl_flags & FL_POSIX))
1191                 return -ENOLCK;
1192         if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
1193                 return -ENOLCK;
1194 
1195         if (cmd == F_CANCELLK) {
1196                 /* Hack: */
1197                 cmd = F_SETLK;
1198                 fl->fl_type = F_UNLCK;
1199         }
1200         if (unlikely(test_bit(SDF_WITHDRAWN, &sdp->sd_flags))) {
1201                 if (fl->fl_type == F_UNLCK)
1202                         locks_lock_file_wait(file, fl);
1203                 return -EIO;
1204         }
1205         if (IS_GETLK(cmd))
1206                 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1207         else if (fl->fl_type == F_UNLCK)
1208                 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1209         else
1210                 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1211 }
1212 
1213 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1214 {
1215         struct gfs2_file *fp = file->private_data;
1216         struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1217         struct gfs2_inode *ip = GFS2_I(file_inode(file));
1218         struct gfs2_glock *gl;
1219         unsigned int state;
1220         u16 flags;
1221         int error = 0;
1222         int sleeptime;
1223 
1224         state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1225         flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT;
1226 
1227         mutex_lock(&fp->f_fl_mutex);
1228 
1229         if (gfs2_holder_initialized(fl_gh)) {
1230                 struct file_lock request;
1231                 if (fl_gh->gh_state == state)
1232                         goto out;
1233                 locks_init_lock(&request);
1234                 request.fl_type = F_UNLCK;
1235                 request.fl_flags = FL_FLOCK;
1236                 locks_lock_file_wait(file, &request);
1237                 gfs2_glock_dq(fl_gh);
1238                 gfs2_holder_reinit(state, flags, fl_gh);
1239         } else {
1240                 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1241                                        &gfs2_flock_glops, CREATE, &gl);
1242                 if (error)
1243                         goto out;
1244                 gfs2_holder_init(gl, state, flags, fl_gh);
1245                 gfs2_glock_put(gl);
1246         }
1247         for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1248                 error = gfs2_glock_nq(fl_gh);
1249                 if (error != GLR_TRYFAILED)
1250                         break;
1251                 fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT;
1252                 fl_gh->gh_error = 0;
1253                 msleep(sleeptime);
1254         }
1255         if (error) {
1256                 gfs2_holder_uninit(fl_gh);
1257                 if (error == GLR_TRYFAILED)
1258                         error = -EAGAIN;
1259         } else {
1260                 error = locks_lock_file_wait(file, fl);
1261                 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1262         }
1263 
1264 out:
1265         mutex_unlock(&fp->f_fl_mutex);
1266         return error;
1267 }
1268 
1269 static void do_unflock(struct file *file, struct file_lock *fl)
1270 {
1271         struct gfs2_file *fp = file->private_data;
1272         struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1273 
1274         mutex_lock(&fp->f_fl_mutex);
1275         locks_lock_file_wait(file, fl);
1276         if (gfs2_holder_initialized(fl_gh)) {
1277                 gfs2_glock_dq(fl_gh);
1278                 gfs2_holder_uninit(fl_gh);
1279         }
1280         mutex_unlock(&fp->f_fl_mutex);
1281 }
1282 
1283 /**
1284  * gfs2_flock - acquire/release a flock lock on a file
1285  * @file: the file pointer
1286  * @cmd: either modify or retrieve lock state, possibly wait
1287  * @fl: type and range of lock
1288  *
1289  * Returns: errno
1290  */
1291 
1292 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1293 {
1294         if (!(fl->fl_flags & FL_FLOCK))
1295                 return -ENOLCK;
1296         if (fl->fl_type & LOCK_MAND)
1297                 return -EOPNOTSUPP;
1298 
1299         if (fl->fl_type == F_UNLCK) {
1300                 do_unflock(file, fl);
1301                 return 0;
1302         } else {
1303                 return do_flock(file, cmd, fl);
1304         }
1305 }
1306 
1307 const struct file_operations gfs2_file_fops = {
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         .compat_ioctl   = gfs2_compat_ioctl,
1314         .mmap           = gfs2_mmap,
1315         .open           = gfs2_open,
1316         .release        = gfs2_release,
1317         .fsync          = gfs2_fsync,
1318         .lock           = gfs2_lock,
1319         .flock          = gfs2_flock,
1320         .splice_read    = generic_file_splice_read,
1321         .splice_write   = gfs2_file_splice_write,
1322         .setlease       = simple_nosetlease,
1323         .fallocate      = gfs2_fallocate,
1324 };
1325 
1326 const struct file_operations gfs2_dir_fops = {
1327         .iterate_shared = gfs2_readdir,
1328         .unlocked_ioctl = gfs2_ioctl,
1329         .compat_ioctl   = gfs2_compat_ioctl,
1330         .open           = gfs2_open,
1331         .release        = gfs2_release,
1332         .fsync          = gfs2_fsync,
1333         .lock           = gfs2_lock,
1334         .flock          = gfs2_flock,
1335         .llseek         = default_llseek,
1336 };
1337 
1338 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1339 
1340 const struct file_operations gfs2_file_fops_nolock = {
1341         .llseek         = gfs2_llseek,
1342         .read_iter      = gfs2_file_read_iter,
1343         .write_iter     = gfs2_file_write_iter,
1344         .iopoll         = iomap_dio_iopoll,
1345         .unlocked_ioctl = gfs2_ioctl,
1346         .compat_ioctl   = gfs2_compat_ioctl,
1347         .mmap           = gfs2_mmap,
1348         .open           = gfs2_open,
1349         .release        = gfs2_release,
1350         .fsync          = gfs2_fsync,
1351         .splice_read    = generic_file_splice_read,
1352         .splice_write   = gfs2_file_splice_write,
1353         .setlease       = generic_setlease,
1354         .fallocate      = gfs2_fallocate,
1355 };
1356 
1357 const struct file_operations gfs2_dir_fops_nolock = {
1358         .iterate_shared = gfs2_readdir,
1359         .unlocked_ioctl = gfs2_ioctl,
1360         .compat_ioctl   = gfs2_compat_ioctl,
1361         .open           = gfs2_open,
1362         .release        = gfs2_release,
1363         .fsync          = gfs2_fsync,
1364         .llseek         = default_llseek,
1365 };
1366
/* [<][>][^][v][top][bottom][index][help] */
root/fs/gfs2/file.c

DEFINITIONS
