~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/gfs2/file.c

Version: ~ [ linux-6.2-rc3 ] ~ [ linux-6.1.5 ] ~ [ linux-6.0.19 ] ~ [ linux-5.19.17 ] ~ [ linux-5.18.19 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.87 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.162 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.228 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.269 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.302 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.337 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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 #include <linux/fileattr.h>
 29 
 30 #include "gfs2.h"
 31 #include "incore.h"
 32 #include "bmap.h"
 33 #include "aops.h"
 34 #include "dir.h"
 35 #include "glock.h"
 36 #include "glops.h"
 37 #include "inode.h"
 38 #include "log.h"
 39 #include "meta_io.h"
 40 #include "quota.h"
 41 #include "rgrp.h"
 42 #include "trans.h"
 43 #include "util.h"
 44 
 45 /**
 46  * gfs2_llseek - seek to a location in a file
 47  * @file: the file
 48  * @offset: the offset
 49  * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
 50  *
 51  * SEEK_END requires the glock for the file because it references the
 52  * file's size.
 53  *
 54  * Returns: The new offset, or errno
 55  */
 56 
 57 static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence)
 58 {
 59         struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
 60         struct gfs2_holder i_gh;
 61         loff_t error;
 62 
 63         switch (whence) {
 64         case SEEK_END:
 65                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
 66                                            &i_gh);
 67                 if (!error) {
 68                         error = generic_file_llseek(file, offset, whence);
 69                         gfs2_glock_dq_uninit(&i_gh);
 70                 }
 71                 break;
 72 
 73         case SEEK_DATA:
 74                 error = gfs2_seek_data(file, offset);
 75                 break;
 76 
 77         case SEEK_HOLE:
 78                 error = gfs2_seek_hole(file, offset);
 79                 break;
 80 
 81         case SEEK_CUR:
 82         case SEEK_SET:
 83                 /*
 84                  * These don't reference inode->i_size and don't depend on the
 85                  * block mapping, so we don't need the glock.
 86                  */
 87                 error = generic_file_llseek(file, offset, whence);
 88                 break;
 89         default:
 90                 error = -EINVAL;
 91         }
 92 
 93         return error;
 94 }
 95 
 96 /**
 97  * gfs2_readdir - Iterator for a directory
 98  * @file: The directory to read from
 99  * @ctx: What to feed directory entries to
100  *
101  * Returns: errno
102  */
103 
104 static int gfs2_readdir(struct file *file, struct dir_context *ctx)
105 {
106         struct inode *dir = file->f_mapping->host;
107         struct gfs2_inode *dip = GFS2_I(dir);
108         struct gfs2_holder d_gh;
109         int error;
110 
111         error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
112         if (error)
113                 return error;
114 
115         error = gfs2_dir_read(dir, ctx, &file->f_ra);
116 
117         gfs2_glock_dq_uninit(&d_gh);
118 
119         return error;
120 }
121 
122 /*
123  * struct fsflag_gfs2flag
124  *
125  * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories,
126  * and to GFS2_DIF_JDATA for non-directories.
127  */
128 static struct {
129         u32 fsflag;
130         u32 gfsflag;
131 } fsflag_gfs2flag[] = {
132         {FS_SYNC_FL, GFS2_DIF_SYNC},
133         {FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE},
134         {FS_APPEND_FL, GFS2_DIF_APPENDONLY},
135         {FS_NOATIME_FL, GFS2_DIF_NOATIME},
136         {FS_INDEX_FL, GFS2_DIF_EXHASH},
137         {FS_TOPDIR_FL, GFS2_DIF_TOPDIR},
138         {FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA},
139 };
140 
141 static inline u32 gfs2_gfsflags_to_fsflags(struct inode *inode, u32 gfsflags)
142 {
143         int i;
144         u32 fsflags = 0;
145 
146         if (S_ISDIR(inode->i_mode))
147                 gfsflags &= ~GFS2_DIF_JDATA;
148         else
149                 gfsflags &= ~GFS2_DIF_INHERIT_JDATA;
150 
151         for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++)
152                 if (gfsflags & fsflag_gfs2flag[i].gfsflag)
153                         fsflags |= fsflag_gfs2flag[i].fsflag;
154         return fsflags;
155 }
156 
157 int gfs2_fileattr_get(struct dentry *dentry, struct fileattr *fa)
158 {
159         struct inode *inode = d_inode(dentry);
160         struct gfs2_inode *ip = GFS2_I(inode);
161         struct gfs2_holder gh;
162         int error;
163         u32 fsflags;
164 
165         if (d_is_special(dentry))
166                 return -ENOTTY;
167 
168         gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
169         error = gfs2_glock_nq(&gh);
170         if (error)
171                 goto out_uninit;
172 
173         fsflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags);
174 
175         fileattr_fill_flags(fa, fsflags);
176 
177         gfs2_glock_dq(&gh);
178 out_uninit:
179         gfs2_holder_uninit(&gh);
180         return error;
181 }
182 
183 void gfs2_set_inode_flags(struct inode *inode)
184 {
185         struct gfs2_inode *ip = GFS2_I(inode);
186         unsigned int flags = inode->i_flags;
187 
188         flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
189         if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
190                 flags |= S_NOSEC;
191         if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
192                 flags |= S_IMMUTABLE;
193         if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
194                 flags |= S_APPEND;
195         if (ip->i_diskflags & GFS2_DIF_NOATIME)
196                 flags |= S_NOATIME;
197         if (ip->i_diskflags & GFS2_DIF_SYNC)
198                 flags |= S_SYNC;
199         inode->i_flags = flags;
200 }
201 
202 /* Flags that can be set by user space */
203 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA|                    \
204                              GFS2_DIF_IMMUTABLE|                \
205                              GFS2_DIF_APPENDONLY|               \
206                              GFS2_DIF_NOATIME|                  \
207                              GFS2_DIF_SYNC|                     \
208                              GFS2_DIF_TOPDIR|                   \
209                              GFS2_DIF_INHERIT_JDATA)
210 
211 /**
212  * do_gfs2_set_flags - set flags on an inode
213  * @inode: The inode
214  * @reqflags: The flags to set
215  * @mask: Indicates which flags are valid
216  *
217  */
218 static int do_gfs2_set_flags(struct inode *inode, u32 reqflags, u32 mask)
219 {
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;
226 
227         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
228         if (error)
229                 return error;
230 
231         error = 0;
232         flags = ip->i_diskflags;
233         new_flags = (flags & ~mask) | (reqflags & mask);
234         if ((new_flags ^ flags) == 0)
235                 goto out;
236 
237         if (!IS_IMMUTABLE(inode)) {
238                 error = gfs2_permission(&init_user_ns, inode, MAY_WRITE);
239                 if (error)
240                         goto out;
241         }
242         if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
243                 if (new_flags & GFS2_DIF_JDATA)
244                         gfs2_log_flush(sdp, ip->i_gl,
245                                        GFS2_LOG_HEAD_FLUSH_NORMAL |
246                                        GFS2_LFC_SET_FLAGS);
247                 error = filemap_fdatawrite(inode->i_mapping);
248                 if (error)
249                         goto out;
250                 error = filemap_fdatawait(inode->i_mapping);
251                 if (error)
252                         goto out;
253                 if (new_flags & GFS2_DIF_JDATA)
254                         gfs2_ordered_del_inode(ip);
255         }
256         error = gfs2_trans_begin(sdp, RES_DINODE, 0);
257         if (error)
258                 goto out;
259         error = gfs2_meta_inode_buffer(ip, &bh);
260         if (error)
261                 goto out_trans_end;
262         inode->i_ctime = current_time(inode);
263         gfs2_trans_add_meta(ip->i_gl, bh);
264         ip->i_diskflags = new_flags;
265         gfs2_dinode_out(ip, bh->b_data);
266         brelse(bh);
267         gfs2_set_inode_flags(inode);
268         gfs2_set_aops(inode);
269 out_trans_end:
270         gfs2_trans_end(sdp);
271 out:
272         gfs2_glock_dq_uninit(&gh);
273         return error;
274 }
275 
276 int gfs2_fileattr_set(struct user_namespace *mnt_userns,
277                       struct dentry *dentry, struct fileattr *fa)
278 {
279         struct inode *inode = d_inode(dentry);
280         u32 fsflags = fa->flags, gfsflags = 0;
281         u32 mask;
282         int i;
283 
284         if (d_is_special(dentry))
285                 return -ENOTTY;
286 
287         if (fileattr_has_fsx(fa))
288                 return -EOPNOTSUPP;
289 
290         for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) {
291                 if (fsflags & fsflag_gfs2flag[i].fsflag) {
292                         fsflags &= ~fsflag_gfs2flag[i].fsflag;
293                         gfsflags |= fsflag_gfs2flag[i].gfsflag;
294                 }
295         }
296         if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET)
297                 return -EINVAL;
298 
299         mask = GFS2_FLAGS_USER_SET;
300         if (S_ISDIR(inode->i_mode)) {
301                 mask &= ~GFS2_DIF_JDATA;
302         } else {
303                 /* The GFS2_DIF_TOPDIR flag is only valid for directories. */
304                 if (gfsflags & GFS2_DIF_TOPDIR)
305                         return -EINVAL;
306                 mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA);
307         }
308 
309         return do_gfs2_set_flags(inode, gfsflags, mask);
310 }
311 
312 static int gfs2_getlabel(struct file *filp, char __user *label)
313 {
314         struct inode *inode = file_inode(filp);
315         struct gfs2_sbd *sdp = GFS2_SB(inode);
316 
317         if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN))
318                 return -EFAULT;
319 
320         return 0;
321 }
322 
323 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
324 {
325         switch(cmd) {
326         case FITRIM:
327                 return gfs2_fitrim(filp, (void __user *)arg);
328         case FS_IOC_GETFSLABEL:
329                 return gfs2_getlabel(filp, (char __user *)arg);
330         }
331 
332         return -ENOTTY;
333 }
334 
335 #ifdef CONFIG_COMPAT
336 static long gfs2_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
337 {
338         switch(cmd) {
339         /* Keep this list in sync with gfs2_ioctl */
340         case FITRIM:
341         case FS_IOC_GETFSLABEL:
342                 break;
343         default:
344                 return -ENOIOCTLCMD;
345         }
346 
347         return gfs2_ioctl(filp, cmd, (unsigned long)compat_ptr(arg));
348 }
349 #else
350 #define gfs2_compat_ioctl NULL
351 #endif
352 
353 /**
354  * gfs2_size_hint - Give a hint to the size of a write request
355  * @filep: The struct file
356  * @offset: The file offset of the write
357  * @size: The length of the write
358  *
359  * When we are about to do a write, this function records the total
360  * write size in order to provide a suitable hint to the lower layers
361  * about how many blocks will be required.
362  *
363  */
364 
365 static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
366 {
367         struct inode *inode = file_inode(filep);
368         struct gfs2_sbd *sdp = GFS2_SB(inode);
369         struct gfs2_inode *ip = GFS2_I(inode);
370         size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
371         int hint = min_t(size_t, INT_MAX, blks);
372 
373         if (hint > atomic_read(&ip->i_sizehint))
374                 atomic_set(&ip->i_sizehint, hint);
375 }
376 
377 /**
378  * gfs2_allocate_page_backing - Allocate blocks for a write fault
379  * @page: The (locked) page to allocate backing for
380  * @length: Size of the allocation
381  *
382  * We try to allocate all the blocks required for the page in one go.  This
383  * might fail for various reasons, so we keep trying until all the blocks to
384  * back this page are allocated.  If some of the blocks are already allocated,
385  * that is ok too.
386  */
387 static int gfs2_allocate_page_backing(struct page *page, unsigned int length)
388 {
389         u64 pos = page_offset(page);
390 
391         do {
392                 struct iomap iomap = { };
393 
394                 if (gfs2_iomap_alloc(page->mapping->host, pos, length, &iomap))
395                         return -EIO;
396 
397                 if (length < iomap.length)
398                         iomap.length = length;
399                 length -= iomap.length;
400                 pos += iomap.length;
401         } while (length > 0);
402 
403         return 0;
404 }
405 
406 /**
407  * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
408  * @vmf: The virtual memory fault containing the page to become writable
409  *
410  * When the page becomes writable, we need to ensure that we have
411  * blocks allocated on disk to back that page.
412  */
413 
414 static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf)
415 {
416         struct page *page = vmf->page;
417         struct inode *inode = file_inode(vmf->vma->vm_file);
418         struct gfs2_inode *ip = GFS2_I(inode);
419         struct gfs2_sbd *sdp = GFS2_SB(inode);
420         struct gfs2_alloc_parms ap = { .aflags = 0, };
421         u64 offset = page_offset(page);
422         unsigned int data_blocks, ind_blocks, rblocks;
423         vm_fault_t ret = VM_FAULT_LOCKED;
424         struct gfs2_holder gh;
425         unsigned int length;
426         loff_t size;
427         int err;
428 
429         sb_start_pagefault(inode->i_sb);
430 
431         gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
432         err = gfs2_glock_nq(&gh);
433         if (err) {
434                 ret = block_page_mkwrite_return(err);
435                 goto out_uninit;
436         }
437 
438         /* Check page index against inode size */
439         size = i_size_read(inode);
440         if (offset >= size) {
441                 ret = VM_FAULT_SIGBUS;
442                 goto out_unlock;
443         }
444 
445         /* Update file times before taking page lock */
446         file_update_time(vmf->vma->vm_file);
447 
448         /* page is wholly or partially inside EOF */
449         if (size - offset < PAGE_SIZE)
450                 length = size - offset;
451         else
452                 length = PAGE_SIZE;
453 
454         gfs2_size_hint(vmf->vma->vm_file, offset, length);
455 
456         set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
457         set_bit(GIF_SW_PAGED, &ip->i_flags);
458 
459         /*
460          * iomap_writepage / iomap_writepages currently don't support inline
461          * files, so always unstuff here.
462          */
463 
464         if (!gfs2_is_stuffed(ip) &&
465             !gfs2_write_alloc_required(ip, offset, length)) {
466                 lock_page(page);
467                 if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
468                         ret = VM_FAULT_NOPAGE;
469                         unlock_page(page);
470                 }
471                 goto out_unlock;
472         }
473 
474         err = gfs2_rindex_update(sdp);
475         if (err) {
476                 ret = block_page_mkwrite_return(err);
477                 goto out_unlock;
478         }
479 
480         gfs2_write_calc_reserv(ip, length, &data_blocks, &ind_blocks);
481         ap.target = data_blocks + ind_blocks;
482         err = gfs2_quota_lock_check(ip, &ap);
483         if (err) {
484                 ret = block_page_mkwrite_return(err);
485                 goto out_unlock;
486         }
487         err = gfs2_inplace_reserve(ip, &ap);
488         if (err) {
489                 ret = block_page_mkwrite_return(err);
490                 goto out_quota_unlock;
491         }
492 
493         rblocks = RES_DINODE + ind_blocks;
494         if (gfs2_is_jdata(ip))
495                 rblocks += data_blocks ? data_blocks : 1;
496         if (ind_blocks || data_blocks) {
497                 rblocks += RES_STATFS + RES_QUOTA;
498                 rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
499         }
500         err = gfs2_trans_begin(sdp, rblocks, 0);
501         if (err) {
502                 ret = block_page_mkwrite_return(err);
503                 goto out_trans_fail;
504         }
505 
506         /* Unstuff, if required, and allocate backing blocks for page */
507         if (gfs2_is_stuffed(ip)) {
508                 err = gfs2_unstuff_dinode(ip);
509                 if (err) {
510                         ret = block_page_mkwrite_return(err);
511                         goto out_trans_end;
512                 }
513         }
514 
515         lock_page(page);
516         /* If truncated, we must retry the operation, we may have raced
517          * with the glock demotion code.
518          */
519         if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
520                 ret = VM_FAULT_NOPAGE;
521                 goto out_page_locked;
522         }
523 
524         err = gfs2_allocate_page_backing(page, length);
525         if (err)
526                 ret = block_page_mkwrite_return(err);
527 
528 out_page_locked:
529         if (ret != VM_FAULT_LOCKED)
530                 unlock_page(page);
531 out_trans_end:
532         gfs2_trans_end(sdp);
533 out_trans_fail:
534         gfs2_inplace_release(ip);
535 out_quota_unlock:
536         gfs2_quota_unlock(ip);
537 out_unlock:
538         gfs2_glock_dq(&gh);
539 out_uninit:
540         gfs2_holder_uninit(&gh);
541         if (ret == VM_FAULT_LOCKED) {
542                 set_page_dirty(page);
543                 wait_for_stable_page(page);
544         }
545         sb_end_pagefault(inode->i_sb);
546         return ret;
547 }
548 
549 static vm_fault_t gfs2_fault(struct vm_fault *vmf)
550 {
551         struct inode *inode = file_inode(vmf->vma->vm_file);
552         struct gfs2_inode *ip = GFS2_I(inode);
553         struct gfs2_holder gh;
554         vm_fault_t ret;
555         int err;
556 
557         gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
558         err = gfs2_glock_nq(&gh);
559         if (err) {
560                 ret = block_page_mkwrite_return(err);
561                 goto out_uninit;
562         }
563         ret = filemap_fault(vmf);
564         gfs2_glock_dq(&gh);
565 out_uninit:
566         gfs2_holder_uninit(&gh);
567         return ret;
568 }
569 
570 static const struct vm_operations_struct gfs2_vm_ops = {
571         .fault = gfs2_fault,
572         .map_pages = filemap_map_pages,
573         .page_mkwrite = gfs2_page_mkwrite,
574 };
575 
576 /**
577  * gfs2_mmap
578  * @file: The file to map
579  * @vma: The VMA which described the mapping
580  *
581  * There is no need to get a lock here unless we should be updating
582  * atime. We ignore any locking errors since the only consequence is
583  * a missed atime update (which will just be deferred until later).
584  *
585  * Returns: 0
586  */
587 
588 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
589 {
590         struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
591 
592         if (!(file->f_flags & O_NOATIME) &&
593             !IS_NOATIME(&ip->i_inode)) {
594                 struct gfs2_holder i_gh;
595                 int error;
596 
597                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
598                                            &i_gh);
599                 if (error)
600                         return error;
601                 /* grab lock to update inode */
602                 gfs2_glock_dq_uninit(&i_gh);
603                 file_accessed(file);
604         }
605         vma->vm_ops = &gfs2_vm_ops;
606 
607         return 0;
608 }
609 
610 /**
611  * gfs2_open_common - This is common to open and atomic_open
612  * @inode: The inode being opened
613  * @file: The file being opened
614  *
615  * This maybe called under a glock or not depending upon how it has
616  * been called. We must always be called under a glock for regular
617  * files, however. For other file types, it does not matter whether
618  * we hold the glock or not.
619  *
620  * Returns: Error code or 0 for success
621  */
622 
623 int gfs2_open_common(struct inode *inode, struct file *file)
624 {
625         struct gfs2_file *fp;
626         int ret;
627 
628         if (S_ISREG(inode->i_mode)) {
629                 ret = generic_file_open(inode, file);
630                 if (ret)
631                         return ret;
632         }
633 
634         fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
635         if (!fp)
636                 return -ENOMEM;
637 
638         mutex_init(&fp->f_fl_mutex);
639 
640         gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
641         file->private_data = fp;
642         if (file->f_mode & FMODE_WRITE) {
643                 ret = gfs2_qa_get(GFS2_I(inode));
644                 if (ret)
645                         goto fail;
646         }
647         return 0;
648 
649 fail:
650         kfree(file->private_data);
651         file->private_data = NULL;
652         return ret;
653 }
654 
655 /**
656  * gfs2_open - open a file
657  * @inode: the inode to open
658  * @file: the struct file for this opening
659  *
660  * After atomic_open, this function is only used for opening files
661  * which are already cached. We must still get the glock for regular
662  * files to ensure that we have the file size uptodate for the large
663  * file check which is in the common code. That is only an issue for
664  * regular files though.
665  *
666  * Returns: errno
667  */
668 
669 static int gfs2_open(struct inode *inode, struct file *file)
670 {
671         struct gfs2_inode *ip = GFS2_I(inode);
672         struct gfs2_holder i_gh;
673         int error;
674         bool need_unlock = false;
675 
676         if (S_ISREG(ip->i_inode.i_mode)) {
677                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
678                                            &i_gh);
679                 if (error)
680                         return error;
681                 need_unlock = true;
682         }
683 
684         error = gfs2_open_common(inode, file);
685 
686         if (need_unlock)
687                 gfs2_glock_dq_uninit(&i_gh);
688 
689         return error;
690 }
691 
692 /**
693  * gfs2_release - called to close a struct file
694  * @inode: the inode the struct file belongs to
695  * @file: the struct file being closed
696  *
697  * Returns: errno
698  */
699 
700 static int gfs2_release(struct inode *inode, struct file *file)
701 {
702         struct gfs2_inode *ip = GFS2_I(inode);
703 
704         kfree(file->private_data);
705         file->private_data = NULL;
706 
707         if (file->f_mode & FMODE_WRITE) {
708                 if (gfs2_rs_active(&ip->i_res))
709                         gfs2_rs_delete(ip);
710                 gfs2_qa_put(ip);
711         }
712         return 0;
713 }
714 
715 /**
716  * gfs2_fsync - sync the dirty data for a file (across the cluster)
717  * @file: the file that points to the dentry
718  * @start: the start position in the file to sync
719  * @end: the end position in the file to sync
720  * @datasync: set if we can ignore timestamp changes
721  *
722  * We split the data flushing here so that we don't wait for the data
723  * until after we've also sent the metadata to disk. Note that for
724  * data=ordered, we will write & wait for the data at the log flush
725  * stage anyway, so this is unlikely to make much of a difference
726  * except in the data=writeback case.
727  *
728  * If the fdatawrite fails due to any reason except -EIO, we will
729  * continue the remainder of the fsync, although we'll still report
730  * the error at the end. This is to match filemap_write_and_wait_range()
731  * behaviour.
732  *
733  * Returns: errno
734  */
735 
736 static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
737                       int datasync)
738 {
739         struct address_space *mapping = file->f_mapping;
740         struct inode *inode = mapping->host;
741         int sync_state = inode->i_state & I_DIRTY;
742         struct gfs2_inode *ip = GFS2_I(inode);
743         int ret = 0, ret1 = 0;
744 
745         if (mapping->nrpages) {
746                 ret1 = filemap_fdatawrite_range(mapping, start, end);
747                 if (ret1 == -EIO)
748                         return ret1;
749         }
750 
751         if (!gfs2_is_jdata(ip))
752                 sync_state &= ~I_DIRTY_PAGES;
753         if (datasync)
754                 sync_state &= ~I_DIRTY_SYNC;
755 
756         if (sync_state) {
757                 ret = sync_inode_metadata(inode, 1);
758                 if (ret)
759                         return ret;
760                 if (gfs2_is_jdata(ip))
761                         ret = file_write_and_wait(file);
762                 if (ret)
763                         return ret;
764                 gfs2_ail_flush(ip->i_gl, 1);
765         }
766 
767         if (mapping->nrpages)
768                 ret = file_fdatawait_range(file, start, end);
769 
770         return ret ? ret : ret1;
771 }
772 
773 static inline bool should_fault_in_pages(ssize_t ret, struct iov_iter *i,
774                                          size_t *prev_count,
775                                          size_t *window_size)
776 {
777         size_t count = iov_iter_count(i);
778         char __user *p;
779         int pages = 1;
780 
781         if (likely(!count))
782                 return false;
783         if (ret <= 0 && ret != -EFAULT)
784                 return false;
785         if (!iter_is_iovec(i))
786                 return false;
787 
788         if (*prev_count != count || !*window_size) {
789                 int pages, nr_dirtied;
790 
791                 pages = min_t(int, BIO_MAX_VECS, DIV_ROUND_UP(count, PAGE_SIZE));
792                 nr_dirtied = max(current->nr_dirtied_pause -
793                                  current->nr_dirtied, 1);
794                 pages = min(pages, nr_dirtied);
795         }
796 
797         *prev_count = count;
798         p = i->iov[0].iov_base + i->iov_offset;
799         *window_size = (size_t)PAGE_SIZE * pages - offset_in_page(p);
800         return true;
801 }
802 
803 static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to,
804                                      struct gfs2_holder *gh)
805 {
806         struct file *file = iocb->ki_filp;
807         struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
808         size_t prev_count = 0, window_size = 0;
809         size_t written = 0;
810         ssize_t ret;
811 
812         /*
813          * In this function, we disable page faults when we're holding the
814          * inode glock while doing I/O.  If a page fault occurs, we indicate
815          * that the inode glock may be dropped, fault in the pages manually,
816          * and retry.
817          *
818          * Unlike generic_file_read_iter, for reads, iomap_dio_rw can trigger
819          * physical as well as manual page faults, and we need to disable both
820          * kinds.
821          *
822          * For direct I/O, gfs2 takes the inode glock in deferred mode.  This
823          * locking mode is compatible with other deferred holders, so multiple
824          * processes and nodes can do direct I/O to a file at the same time.
825          * There's no guarantee that reads or writes will be atomic.  Any
826          * coordination among readers and writers needs to happen externally.
827          */
828 
829         if (!iov_iter_count(to))
830                 return 0; /* skip atime */
831 
832         gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
833 retry:
834         ret = gfs2_glock_nq(gh);
835         if (ret)
836                 goto out_uninit;
837 retry_under_glock:
838         pagefault_disable();
839         to->nofault = true;
840         ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL,
841                            IOMAP_DIO_PARTIAL, written);
842         to->nofault = false;
843         pagefault_enable();
844         if (ret > 0)
845                 written = ret;
846 
847         if (should_fault_in_pages(ret, to, &prev_count, &window_size)) {
848                 size_t leftover;
849 
850                 gfs2_holder_allow_demote(gh);
851                 leftover = fault_in_iov_iter_writeable(to, window_size);
852                 gfs2_holder_disallow_demote(gh);
853                 if (leftover != window_size) {
854                         if (gfs2_holder_queued(gh))
855                                 goto retry_under_glock;
856                         goto retry;
857                 }
858         }
859         if (gfs2_holder_queued(gh))
860                 gfs2_glock_dq(gh);
861 out_uninit:
862         gfs2_holder_uninit(gh);
863         if (ret < 0)
864                 return ret;
865         return written;
866 }
867 
868 static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from,
869                                       struct gfs2_holder *gh)
870 {
871         struct file *file = iocb->ki_filp;
872         struct inode *inode = file->f_mapping->host;
873         struct gfs2_inode *ip = GFS2_I(inode);
874         size_t prev_count = 0, window_size = 0;
875         size_t read = 0;
876         ssize_t ret;
877 
878         /*
879          * In this function, we disable page faults when we're holding the
880          * inode glock while doing I/O.  If a page fault occurs, we indicate
881          * that the inode glock may be dropped, fault in the pages manually,
882          * and retry.
883          *
884          * For writes, iomap_dio_rw only triggers manual page faults, so we
885          * don't need to disable physical ones.
886          */
887 
888         /*
889          * Deferred lock, even if its a write, since we do no allocation on
890          * this path. All we need to change is the atime, and this lock mode
891          * ensures that other nodes have flushed their buffered read caches
892          * (i.e. their page cache entries for this inode). We do not,
893          * unfortunately, have the option of only flushing a range like the
894          * VFS does.
895          */
896         gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
897 retry:
898         ret = gfs2_glock_nq(gh);
899         if (ret)
900                 goto out_uninit;
901 retry_under_glock:
902         /* Silently fall back to buffered I/O when writing beyond EOF */
903         if (iocb->ki_pos + iov_iter_count(from) > i_size_read(&ip->i_inode))
904                 goto out;
905 
906         from->nofault = true;
907         ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL,
908                            IOMAP_DIO_PARTIAL, read);
909         from->nofault = false;
910 
911         if (ret == -ENOTBLK)
912                 ret = 0;
913         if (ret > 0)
914                 read = ret;
915 
916         if (should_fault_in_pages(ret, from, &prev_count, &window_size)) {
917                 size_t leftover;
918 
919                 gfs2_holder_allow_demote(gh);
920                 leftover = fault_in_iov_iter_readable(from, window_size);
921                 gfs2_holder_disallow_demote(gh);
922                 if (leftover != window_size) {
923                         if (gfs2_holder_queued(gh))
924                                 goto retry_under_glock;
925                         goto retry;
926                 }
927         }
928 out:
929         if (gfs2_holder_queued(gh))
930                 gfs2_glock_dq(gh);
931 out_uninit:
932         gfs2_holder_uninit(gh);
933         if (ret < 0)
934                 return ret;
935         return read;
936 }
937 
938 static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
939 {
940         struct gfs2_inode *ip;
941         struct gfs2_holder gh;
942         size_t prev_count = 0, window_size = 0;
943         size_t written = 0;
944         ssize_t ret;
945 
946         /*
947          * In this function, we disable page faults when we're holding the
948          * inode glock while doing I/O.  If a page fault occurs, we indicate
949          * that the inode glock may be dropped, fault in the pages manually,
950          * and retry.
951          */
952 
953         if (iocb->ki_flags & IOCB_DIRECT) {
954                 ret = gfs2_file_direct_read(iocb, to, &gh);
955                 if (likely(ret != -ENOTBLK))
956                         return ret;
957                 iocb->ki_flags &= ~IOCB_DIRECT;
958         }
959         pagefault_disable();
960         iocb->ki_flags |= IOCB_NOIO;
961         ret = generic_file_read_iter(iocb, to);
962         iocb->ki_flags &= ~IOCB_NOIO;
963         pagefault_enable();
964         if (ret >= 0) {
965                 if (!iov_iter_count(to))
966                         return ret;
967                 written = ret;
968         } else if (ret != -EFAULT) {
969                 if (ret != -EAGAIN)
970                         return ret;
971                 if (iocb->ki_flags & IOCB_NOWAIT)
972                         return ret;
973         }
974         ip = GFS2_I(iocb->ki_filp->f_mapping->host);
975         gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
976 retry:
977         ret = gfs2_glock_nq(&gh);
978         if (ret)
979                 goto out_uninit;
980 retry_under_glock:
981         pagefault_disable();
982         ret = generic_file_read_iter(iocb, to);
983         pagefault_enable();
984         if (ret > 0)
985                 written += ret;
986 
987         if (should_fault_in_pages(ret, to, &prev_count, &window_size)) {
988                 size_t leftover;
989 
990                 gfs2_holder_allow_demote(&gh);
991                 leftover = fault_in_iov_iter_writeable(to, window_size);
992                 gfs2_holder_disallow_demote(&gh);
993                 if (leftover != window_size) {
994                         if (gfs2_holder_queued(&gh))
995                                 goto retry_under_glock;
996                         goto retry;
997                 }
998         }
999         if (gfs2_holder_queued(&gh))
1000                 gfs2_glock_dq(&gh);
1001 out_uninit:
1002         gfs2_holder_uninit(&gh);
1003         return written ? written : ret;
1004 }
1005 
1006 static ssize_t gfs2_file_buffered_write(struct kiocb *iocb,
1007                                         struct iov_iter *from,
1008                                         struct gfs2_holder *gh)
1009 {
1010         struct file *file = iocb->ki_filp;
1011         struct inode *inode = file_inode(file);
1012         struct gfs2_inode *ip = GFS2_I(inode);
1013         struct gfs2_sbd *sdp = GFS2_SB(inode);
1014         struct gfs2_holder *statfs_gh = NULL;
1015         size_t prev_count = 0, window_size = 0;
1016         size_t orig_count = iov_iter_count(from);
1017         size_t read = 0;
1018         ssize_t ret;
1019 
1020         /*
1021          * In this function, we disable page faults when we're holding the
1022          * inode glock while doing I/O.  If a page fault occurs, we indicate
1023          * that the inode glock may be dropped, fault in the pages manually,
1024          * and retry.
1025          */
1026 
1027         if (inode == sdp->sd_rindex) {
1028                 statfs_gh = kmalloc(sizeof(*statfs_gh), GFP_NOFS);
1029                 if (!statfs_gh)
1030                         return -ENOMEM;
1031         }
1032 
1033         gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, gh);
1034 retry:
1035         ret = gfs2_glock_nq(gh);
1036         if (ret)
1037                 goto out_uninit;
1038 retry_under_glock:
1039         if (inode == sdp->sd_rindex) {
1040                 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
1041 
1042                 ret = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
1043                                          GL_NOCACHE, statfs_gh);
1044                 if (ret)
1045                         goto out_unlock;
1046         }
1047 
1048         current->backing_dev_info = inode_to_bdi(inode);
1049         pagefault_disable();
1050         ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
1051         pagefault_enable();
1052         current->backing_dev_info = NULL;
1053         if (ret > 0) {
1054                 iocb->ki_pos += ret;
1055                 read += ret;
1056         }
1057 
1058         if (inode == sdp->sd_rindex)
1059                 gfs2_glock_dq_uninit(statfs_gh);
1060 
1061         from->count = orig_count - read;
1062         if (should_fault_in_pages(ret, from, &prev_count, &window_size)) {
1063                 size_t leftover;
1064 
1065                 gfs2_holder_allow_demote(gh);
1066                 leftover = fault_in_iov_iter_readable(from, window_size);
1067                 gfs2_holder_disallow_demote(gh);
1068                 if (leftover != window_size) {
1069                         from->count = min(from->count, window_size - leftover);
1070                         if (gfs2_holder_queued(gh))
1071                                 goto retry_under_glock;
1072                         goto retry;
1073                 }
1074         }
1075 out_unlock:
1076         if (gfs2_holder_queued(gh))
1077                 gfs2_glock_dq(gh);
1078 out_uninit:
1079         gfs2_holder_uninit(gh);
1080         if (statfs_gh)
1081                 kfree(statfs_gh);
1082         from->count = orig_count - read;
1083         return read ? read : ret;
1084 }
1085 
1086 /**
1087  * gfs2_file_write_iter - Perform a write to a file
1088  * @iocb: The io context
1089  * @from: The data to write
1090  *
1091  * We have to do a lock/unlock here to refresh the inode size for
1092  * O_APPEND writes, otherwise we can land up writing at the wrong
1093  * offset. There is still a race, but provided the app is using its
1094  * own file locking, this will make O_APPEND work as expected.
1095  *
1096  */
1097 
1098 static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1099 {
1100         struct file *file = iocb->ki_filp;
1101         struct inode *inode = file_inode(file);
1102         struct gfs2_inode *ip = GFS2_I(inode);
1103         struct gfs2_holder gh;
1104         ssize_t ret;
1105 
1106         gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
1107 
1108         if (iocb->ki_flags & IOCB_APPEND) {
1109                 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1110                 if (ret)
1111                         return ret;
1112                 gfs2_glock_dq_uninit(&gh);
1113         }
1114 
1115         inode_lock(inode);
1116         ret = generic_write_checks(iocb, from);
1117         if (ret <= 0)
1118                 goto out_unlock;
1119 
1120         ret = file_remove_privs(file);
1121         if (ret)
1122                 goto out_unlock;
1123 
1124         ret = file_update_time(file);
1125         if (ret)
1126                 goto out_unlock;
1127 
1128         if (iocb->ki_flags & IOCB_DIRECT) {
1129                 struct address_space *mapping = file->f_mapping;
1130                 ssize_t buffered, ret2;
1131 
1132                 ret = gfs2_file_direct_write(iocb, from, &gh);
1133                 if (ret < 0 || !iov_iter_count(from))
1134                         goto out_unlock;
1135 
1136                 iocb->ki_flags |= IOCB_DSYNC;
1137                 buffered = gfs2_file_buffered_write(iocb, from, &gh);
1138                 if (unlikely(buffered <= 0)) {
1139                         if (!ret)
1140                                 ret = buffered;
1141                         goto out_unlock;
1142                 }
1143 
1144                 /*
1145                  * We need to ensure that the page cache pages are written to
1146                  * disk and invalidated to preserve the expected O_DIRECT
1147                  * semantics.  If the writeback or invalidate fails, only report
1148                  * the direct I/O range as we don't know if the buffered pages
1149                  * made it to disk.
1150                  */
1151                 ret2 = generic_write_sync(iocb, buffered);
1152                 invalidate_mapping_pages(mapping,
1153                                 (iocb->ki_pos - buffered) >> PAGE_SHIFT,
1154                                 (iocb->ki_pos - 1) >> PAGE_SHIFT);
1155                 if (!ret || ret2 > 0)
1156                         ret += ret2;
1157         } else {
1158                 ret = gfs2_file_buffered_write(iocb, from, &gh);
1159                 if (likely(ret > 0))
1160                         ret = generic_write_sync(iocb, ret);
1161         }
1162 
1163 out_unlock:
1164         inode_unlock(inode);
1165         return ret;
1166 }
1167 
1168 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
1169                            int mode)
1170 {
1171         struct super_block *sb = inode->i_sb;
1172         struct gfs2_inode *ip = GFS2_I(inode);
1173         loff_t end = offset + len;
1174         struct buffer_head *dibh;
1175         int error;
1176 
1177         error = gfs2_meta_inode_buffer(ip, &dibh);
1178         if (unlikely(error))
1179                 return error;
1180 
1181         gfs2_trans_add_meta(ip->i_gl, dibh);
1182 
1183         if (gfs2_is_stuffed(ip)) {
1184                 error = gfs2_unstuff_dinode(ip);
1185                 if (unlikely(error))
1186                         goto out;
1187         }
1188 
1189         while (offset < end) {
1190                 struct iomap iomap = { };
1191 
1192                 error = gfs2_iomap_alloc(inode, offset, end - offset, &iomap);
1193                 if (error)
1194                         goto out;
1195                 offset = iomap.offset + iomap.length;
1196                 if (!(iomap.flags & IOMAP_F_NEW))
1197                         continue;
1198                 error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
1199                                          iomap.length >> inode->i_blkbits,
1200                                          GFP_NOFS);
1201                 if (error) {
1202                         fs_err(GFS2_SB(inode), "Failed to zero data buffers\n");
1203                         goto out;
1204                 }
1205         }
1206 out:
1207         brelse(dibh);
1208         return error;
1209 }
1210 
1211 /**
1212  * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
1213  *                     blocks, determine how many bytes can be written.
1214  * @ip:          The inode in question.
1215  * @len:         Max cap of bytes. What we return in *len must be <= this.
1216  * @data_blocks: Compute and return the number of data blocks needed
1217  * @ind_blocks:  Compute and return the number of indirect blocks needed
1218  * @max_blocks:  The total blocks available to work with.
1219  *
1220  * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
1221  */
1222 static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
1223                             unsigned int *data_blocks, unsigned int *ind_blocks,
1224                             unsigned int max_blocks)
1225 {
1226         loff_t max = *len;
1227         const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1228         unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
1229 
1230         for (tmp = max_data; tmp > sdp->sd_diptrs;) {
1231                 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
1232                 max_data -= tmp;
1233         }
1234 
1235         *data_blocks = max_data;
1236         *ind_blocks = max_blocks - max_data;
1237         *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
1238         if (*len > max) {
1239                 *len = max;
1240                 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
1241         }
1242 }
1243 
1244 static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1245 {
1246         struct inode *inode = file_inode(file);
1247         struct gfs2_sbd *sdp = GFS2_SB(inode);
1248         struct gfs2_inode *ip = GFS2_I(inode);
1249         struct gfs2_alloc_parms ap = { .aflags = 0, };
1250         unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
1251         loff_t bytes, max_bytes, max_blks;
1252         int error;
1253         const loff_t pos = offset;
1254         const loff_t count = len;
1255         loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
1256         loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
1257         loff_t max_chunk_size = UINT_MAX & bsize_mask;
1258 
1259         next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
1260 
1261         offset &= bsize_mask;
1262 
1263         len = next - offset;
1264         bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
1265         if (!bytes)
1266                 bytes = UINT_MAX;
1267         bytes &= bsize_mask;
1268         if (bytes == 0)
1269                 bytes = sdp->sd_sb.sb_bsize;
1270 
1271         gfs2_size_hint(file, offset, len);
1272 
1273         gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
1274         ap.min_target = data_blocks + ind_blocks;
1275 
1276         while (len > 0) {
1277                 if (len < bytes)
1278                         bytes = len;
1279                 if (!gfs2_write_alloc_required(ip, offset, bytes)) {
1280                         len -= bytes;
1281                         offset += bytes;
1282                         continue;
1283                 }
1284 
1285                 /* We need to determine how many bytes we can actually
1286                  * fallocate without exceeding quota or going over the
1287                  * end of the fs. We start off optimistically by assuming
1288                  * we can write max_bytes */
1289                 max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
1290 
1291                 /* Since max_bytes is most likely a theoretical max, we
1292                  * calculate a more realistic 'bytes' to serve as a good
1293                  * starting point for the number of bytes we may be able
1294                  * to write */
1295                 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
1296                 ap.target = data_blocks + ind_blocks;
1297 
1298                 error = gfs2_quota_lock_check(ip, &ap);
1299                 if (error)
1300                         return error;
1301                 /* ap.allowed tells us how many blocks quota will allow
1302                  * us to write. Check if this reduces max_blks */
1303                 max_blks = UINT_MAX;
1304                 if (ap.allowed)
1305                         max_blks = ap.allowed;
1306 
1307                 error = gfs2_inplace_reserve(ip, &ap);
1308                 if (error)
1309                         goto out_qunlock;
1310 
1311                 /* check if the selected rgrp limits our max_blks further */
1312                 if (ip->i_res.rs_reserved < max_blks)
1313                         max_blks = ip->i_res.rs_reserved;
1314 
1315                 /* Almost done. Calculate bytes that can be written using
1316                  * max_blks. We also recompute max_bytes, data_blocks and
1317                  * ind_blocks */
1318                 calc_max_reserv(ip, &max_bytes, &data_blocks,
1319                                 &ind_blocks, max_blks);
1320 
1321                 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
1322                           RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
1323                 if (gfs2_is_jdata(ip))
1324                         rblocks += data_blocks ? data_blocks : 1;
1325 
1326                 error = gfs2_trans_begin(sdp, rblocks,
1327                                          PAGE_SIZE >> inode->i_blkbits);
1328                 if (error)
1329                         goto out_trans_fail;
1330 
1331                 error = fallocate_chunk(inode, offset, max_bytes, mode);
1332                 gfs2_trans_end(sdp);
1333 
1334                 if (error)
1335                         goto out_trans_fail;
1336 
1337                 len -= max_bytes;
1338                 offset += max_bytes;
1339                 gfs2_inplace_release(ip);
1340                 gfs2_quota_unlock(ip);
1341         }
1342 
1343         if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size)
1344                 i_size_write(inode, pos + count);
1345         file_update_time(file);
1346         mark_inode_dirty(inode);
1347 
1348         if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host))
1349                 return vfs_fsync_range(file, pos, pos + count - 1,
1350                                (file->f_flags & __O_SYNC) ? 0 : 1);
1351         return 0;
1352 
1353 out_trans_fail:
1354         gfs2_inplace_release(ip);
1355 out_qunlock:
1356         gfs2_quota_unlock(ip);
1357         return error;
1358 }
1359 
1360 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1361 {
1362         struct inode *inode = file_inode(file);
1363         struct gfs2_sbd *sdp = GFS2_SB(inode);
1364         struct gfs2_inode *ip = GFS2_I(inode);
1365         struct gfs2_holder gh;
1366         int ret;
1367 
1368         if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))
1369                 return -EOPNOTSUPP;
1370         /* fallocate is needed by gfs2_grow to reserve space in the rindex */
1371         if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex)
1372                 return -EOPNOTSUPP;
1373 
1374         inode_lock(inode);
1375 
1376         gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1377         ret = gfs2_glock_nq(&gh);
1378         if (ret)
1379                 goto out_uninit;
1380 
1381         if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1382             (offset + len) > inode->i_size) {
1383                 ret = inode_newsize_ok(inode, offset + len);
1384                 if (ret)
1385                         goto out_unlock;
1386         }
1387 
1388         ret = get_write_access(inode);
1389         if (ret)
1390                 goto out_unlock;
1391 
1392         if (mode & FALLOC_FL_PUNCH_HOLE) {
1393                 ret = __gfs2_punch_hole(file, offset, len);
1394         } else {
1395                 ret = __gfs2_fallocate(file, mode, offset, len);
1396                 if (ret)
1397                         gfs2_rs_deltree(&ip->i_res);
1398         }
1399 
1400         put_write_access(inode);
1401 out_unlock:
1402         gfs2_glock_dq(&gh);
1403 out_uninit:
1404         gfs2_holder_uninit(&gh);
1405         inode_unlock(inode);
1406         return ret;
1407 }
1408 
1409 static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
1410                                       struct file *out, loff_t *ppos,
1411                                       size_t len, unsigned int flags)
1412 {
1413         ssize_t ret;
1414 
1415         gfs2_size_hint(out, *ppos, len);
1416 
1417         ret = iter_file_splice_write(pipe, out, ppos, len, flags);
1418         return ret;
1419 }
1420 
1421 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1422 
1423 /**
1424  * gfs2_lock - acquire/release a posix lock on a file
1425  * @file: the file pointer
1426  * @cmd: either modify or retrieve lock state, possibly wait
1427  * @fl: type and range of lock
1428  *
1429  * Returns: errno
1430  */
1431 
1432 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
1433 {
1434         struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
1435         struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
1436         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1437 
1438         if (!(fl->fl_flags & FL_POSIX))
1439                 return -ENOLCK;
1440         if (cmd == F_CANCELLK) {
1441                 /* Hack: */
1442                 cmd = F_SETLK;
1443                 fl->fl_type = F_UNLCK;
1444         }
1445         if (unlikely(gfs2_withdrawn(sdp))) {
1446                 if (fl->fl_type == F_UNLCK)
1447                         locks_lock_file_wait(file, fl);
1448                 return -EIO;
1449         }
1450         if (IS_GETLK(cmd))
1451                 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1452         else if (fl->fl_type == F_UNLCK)
1453                 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1454         else
1455                 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1456 }
1457 
1458 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1459 {
1460         struct gfs2_file *fp = file->private_data;
1461         struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1462         struct gfs2_inode *ip = GFS2_I(file_inode(file));
1463         struct gfs2_glock *gl;
1464         unsigned int state;
1465         u16 flags;
1466         int error = 0;
1467         int sleeptime;
1468 
1469         state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1470         flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT;
1471 
1472         mutex_lock(&fp->f_fl_mutex);
1473 
1474         if (gfs2_holder_initialized(fl_gh)) {
1475                 struct file_lock request;
1476                 if (fl_gh->gh_state == state)
1477                         goto out;
1478                 locks_init_lock(&request);
1479                 request.fl_type = F_UNLCK;
1480                 request.fl_flags = FL_FLOCK;
1481                 locks_lock_file_wait(file, &request);
1482                 gfs2_glock_dq(fl_gh);
1483                 gfs2_holder_reinit(state, flags, fl_gh);
1484         } else {
1485                 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1486                                        &gfs2_flock_glops, CREATE, &gl);
1487                 if (error)
1488                         goto out;
1489                 gfs2_holder_init(gl, state, flags, fl_gh);
1490                 gfs2_glock_put(gl);
1491         }
1492         for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1493                 error = gfs2_glock_nq(fl_gh);
1494                 if (error != GLR_TRYFAILED)
1495                         break;
1496                 fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT;
1497                 fl_gh->gh_error = 0;
1498                 msleep(sleeptime);
1499         }
1500         if (error) {
1501                 gfs2_holder_uninit(fl_gh);
1502                 if (error == GLR_TRYFAILED)
1503                         error = -EAGAIN;
1504         } else {
1505                 error = locks_lock_file_wait(file, fl);
1506                 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1507         }
1508 
1509 out:
1510         mutex_unlock(&fp->f_fl_mutex);
1511         return error;
1512 }
1513 
1514 static void do_unflock(struct file *file, struct file_lock *fl)
1515 {
1516         struct gfs2_file *fp = file->private_data;
1517         struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1518 
1519         mutex_lock(&fp->f_fl_mutex);
1520         locks_lock_file_wait(file, fl);
1521         if (gfs2_holder_initialized(fl_gh)) {
1522                 gfs2_glock_dq(fl_gh);
1523                 gfs2_holder_uninit(fl_gh);
1524         }
1525         mutex_unlock(&fp->f_fl_mutex);
1526 }
1527 
1528 /**
1529  * gfs2_flock - acquire/release a flock lock on a file
1530  * @file: the file pointer
1531  * @cmd: either modify or retrieve lock state, possibly wait
1532  * @fl: type and range of lock
1533  *
1534  * Returns: errno
1535  */
1536 
1537 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1538 {
1539         if (!(fl->fl_flags & FL_FLOCK))
1540                 return -ENOLCK;
1541 
1542         if (fl->fl_type == F_UNLCK) {
1543                 do_unflock(file, fl);
1544                 return 0;
1545         } else {
1546                 return do_flock(file, cmd, fl);
1547         }
1548 }
1549 
1550 const struct file_operations gfs2_file_fops = {
1551         .llseek         = gfs2_llseek,
1552         .read_iter      = gfs2_file_read_iter,
1553         .write_iter     = gfs2_file_write_iter,
1554         .iopoll         = iocb_bio_iopoll,
1555         .unlocked_ioctl = gfs2_ioctl,
1556         .compat_ioctl   = gfs2_compat_ioctl,
1557         .mmap           = gfs2_mmap,
1558         .open           = gfs2_open,
1559         .release        = gfs2_release,
1560         .fsync          = gfs2_fsync,
1561         .lock           = gfs2_lock,
1562         .flock          = gfs2_flock,
1563         .splice_read    = generic_file_splice_read,
1564         .splice_write   = gfs2_file_splice_write,
1565         .setlease       = simple_nosetlease,
1566         .fallocate      = gfs2_fallocate,
1567 };
1568 
1569 const struct file_operations gfs2_dir_fops = {
1570         .iterate_shared = gfs2_readdir,
1571         .unlocked_ioctl = gfs2_ioctl,
1572         .compat_ioctl   = gfs2_compat_ioctl,
1573         .open           = gfs2_open,
1574         .release        = gfs2_release,
1575         .fsync          = gfs2_fsync,
1576         .lock           = gfs2_lock,
1577         .flock          = gfs2_flock,
1578         .llseek         = default_llseek,
1579 };
1580 
1581 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1582 
1583 const struct file_operations gfs2_file_fops_nolock = {
1584         .llseek         = gfs2_llseek,
1585         .read_iter      = gfs2_file_read_iter,
1586         .write_iter     = gfs2_file_write_iter,
1587         .iopoll         = iocb_bio_iopoll,
1588         .unlocked_ioctl = gfs2_ioctl,
1589         .compat_ioctl   = gfs2_compat_ioctl,
1590         .mmap           = gfs2_mmap,
1591         .open           = gfs2_open,
1592         .release        = gfs2_release,
1593         .fsync          = gfs2_fsync,
1594         .splice_read    = generic_file_splice_read,
1595         .splice_write   = gfs2_file_splice_write,
1596         .setlease       = generic_setlease,
1597         .fallocate      = gfs2_fallocate,
1598 };
1599 
1600 const struct file_operations gfs2_dir_fops_nolock = {
1601         .iterate_shared = gfs2_readdir,
1602         .unlocked_ioctl = gfs2_ioctl,
1603         .compat_ioctl   = gfs2_compat_ioctl,
1604         .open           = gfs2_open,
1605         .release        = gfs2_release,
1606         .fsync          = gfs2_fsync,
1607         .llseek         = default_llseek,
1608 };
1609 
1610 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp