Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3a8a9a10 | 3 | * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
4 | * |
5 | * This copyrighted material is made available to anyone wishing to use, | |
6 | * modify, copy, or redistribute it subject to the terms and conditions | |
e9fc2aa0 | 7 | * of the GNU General Public License version 2. |
b3b94faa DT |
8 | */ |
9 | ||
b3b94faa DT |
10 | #include <linux/slab.h> |
11 | #include <linux/spinlock.h> | |
12 | #include <linux/completion.h> | |
13 | #include <linux/buffer_head.h> | |
14 | #include <linux/pagemap.h> | |
15 | #include <linux/uio.h> | |
16 | #include <linux/blkdev.h> | |
17 | #include <linux/mm.h> | |
f58ba889 | 18 | #include <linux/mount.h> |
18ec7d5c | 19 | #include <linux/fs.h> |
5c676f6d | 20 | #include <linux/gfs2_ondisk.h> |
2fe17c10 CH |
21 | #include <linux/falloc.h> |
22 | #include <linux/swap.h> | |
71b86f56 | 23 | #include <linux/crc32.h> |
33c3de32 | 24 | #include <linux/writeback.h> |
b3b94faa | 25 | #include <asm/uaccess.h> |
f057f6cd SW |
26 | #include <linux/dlm.h> |
27 | #include <linux/dlm_plock.h> | |
2ddfbdd6 | 28 | #include <linux/delay.h> |
b3b94faa DT |
29 | |
30 | #include "gfs2.h" | |
5c676f6d | 31 | #include "incore.h" |
b3b94faa DT |
32 | #include "bmap.h" |
33 | #include "dir.h" | |
34 | #include "glock.h" | |
35 | #include "glops.h" | |
36 | #include "inode.h" | |
b3b94faa DT |
37 | #include "log.h" |
38 | #include "meta_io.h" | |
b3b94faa DT |
39 | #include "quota.h" |
40 | #include "rgrp.h" | |
41 | #include "trans.h" | |
5c676f6d | 42 | #include "util.h" |
b3b94faa | 43 | |
b3b94faa DT |
44 | /** |
45 | * gfs2_llseek - seek to a location in a file | |
46 | * @file: the file | |
47 | * @offset: the offset | |
965c8e59 | 48 | * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END) |
b3b94faa DT |
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 | ||
965c8e59 | 56 | static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence) |
b3b94faa | 57 | { |
feaa7bba | 58 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); |
b3b94faa DT |
59 | struct gfs2_holder i_gh; |
60 | loff_t error; | |
61 | ||
965c8e59 | 62 | switch (whence) { |
9453615a SW |
63 | case SEEK_END: /* These reference inode->i_size */ |
64 | case SEEK_DATA: | |
65 | case SEEK_HOLE: | |
b3b94faa DT |
66 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, |
67 | &i_gh); | |
68 | if (!error) { | |
965c8e59 | 69 | error = generic_file_llseek(file, offset, whence); |
b3b94faa DT |
70 | gfs2_glock_dq_uninit(&i_gh); |
71 | } | |
9453615a SW |
72 | break; |
73 | case SEEK_CUR: | |
74 | case SEEK_SET: | |
965c8e59 | 75 | error = generic_file_llseek(file, offset, whence); |
9453615a SW |
76 | break; |
77 | default: | |
78 | error = -EINVAL; | |
79 | } | |
b3b94faa DT |
80 | |
81 | return error; | |
82 | } | |
83 | ||
b3b94faa | 84 | /** |
d81a8ef5 | 85 | * gfs2_readdir - Iterator for a directory |
b3b94faa | 86 | * @file: The directory to read from |
d81a8ef5 | 87 | * @ctx: What to feed directory entries to |
b3b94faa DT |
88 | * |
89 | * Returns: errno | |
90 | */ | |
91 | ||
d81a8ef5 | 92 | static int gfs2_readdir(struct file *file, struct dir_context *ctx) |
b3b94faa | 93 | { |
71b86f56 | 94 | struct inode *dir = file->f_mapping->host; |
feaa7bba | 95 | struct gfs2_inode *dip = GFS2_I(dir); |
b3b94faa | 96 | struct gfs2_holder d_gh; |
b3b94faa DT |
97 | int error; |
98 | ||
d81a8ef5 AV |
99 | error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh); |
100 | if (error) | |
b3b94faa | 101 | return error; |
b3b94faa | 102 | |
d81a8ef5 | 103 | error = gfs2_dir_read(dir, ctx, &file->f_ra); |
b3b94faa DT |
104 | |
105 | gfs2_glock_dq_uninit(&d_gh); | |
106 | ||
b3b94faa DT |
107 | return error; |
108 | } | |
109 | ||
128e5eba SW |
110 | /** |
111 | * fsflags_cvt | |
112 | * @table: A table of 32 u32 flags | |
113 | * @val: a 32 bit value to convert | |
114 | * | |
115 | * This function can be used to convert between fsflags values and | |
116 | * GFS2's own flags values. | |
117 | * | |
118 | * Returns: the converted flags | |
119 | */ | |
120 | static u32 fsflags_cvt(const u32 *table, u32 val) | |
121 | { | |
122 | u32 res = 0; | |
123 | while(val) { | |
124 | if (val & 1) | |
125 | res |= *table; | |
126 | table++; | |
127 | val >>= 1; | |
128 | } | |
129 | return res; | |
130 | } | |
b3b94faa | 131 | |
128e5eba SW |
132 | static const u32 fsflags_to_gfs2[32] = { |
133 | [3] = GFS2_DIF_SYNC, | |
134 | [4] = GFS2_DIF_IMMUTABLE, | |
135 | [5] = GFS2_DIF_APPENDONLY, | |
136 | [7] = GFS2_DIF_NOATIME, | |
137 | [12] = GFS2_DIF_EXHASH, | |
b9af7ca6 | 138 | [14] = GFS2_DIF_INHERIT_JDATA, |
23d0bb83 | 139 | [17] = GFS2_DIF_TOPDIR, |
71b86f56 SW |
140 | }; |
141 | ||
128e5eba SW |
142 | static const u32 gfs2_to_fsflags[32] = { |
143 | [gfs2fl_Sync] = FS_SYNC_FL, | |
144 | [gfs2fl_Immutable] = FS_IMMUTABLE_FL, | |
145 | [gfs2fl_AppendOnly] = FS_APPEND_FL, | |
146 | [gfs2fl_NoAtime] = FS_NOATIME_FL, | |
147 | [gfs2fl_ExHash] = FS_INDEX_FL, | |
23d0bb83 | 148 | [gfs2fl_TopLevel] = FS_TOPDIR_FL, |
128e5eba | 149 | [gfs2fl_InheritJdata] = FS_JOURNAL_DATA_FL, |
7ea9ea83 | 150 | }; |
71b86f56 | 151 | |
b09e593d | 152 | static int gfs2_get_flags(struct file *filp, u32 __user *ptr) |
71b86f56 | 153 | { |
496ad9aa | 154 | struct inode *inode = file_inode(filp); |
feaa7bba | 155 | struct gfs2_inode *ip = GFS2_I(inode); |
71b86f56 SW |
156 | struct gfs2_holder gh; |
157 | int error; | |
128e5eba | 158 | u32 fsflags; |
71b86f56 | 159 | |
719ee344 SW |
160 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); |
161 | error = gfs2_glock_nq(&gh); | |
71b86f56 | 162 | if (error) |
9c7fe835 | 163 | goto out_uninit; |
907b9bce | 164 | |
383f01fb SW |
165 | fsflags = fsflags_cvt(gfs2_to_fsflags, ip->i_diskflags); |
166 | if (!S_ISDIR(inode->i_mode) && ip->i_diskflags & GFS2_DIF_JDATA) | |
c9f6a6bb | 167 | fsflags |= FS_JOURNAL_DATA_FL; |
128e5eba | 168 | if (put_user(fsflags, ptr)) |
71b86f56 SW |
169 | error = -EFAULT; |
170 | ||
3cc3f710 | 171 | gfs2_glock_dq(&gh); |
9c7fe835 | 172 | out_uninit: |
71b86f56 SW |
173 | gfs2_holder_uninit(&gh); |
174 | return error; | |
175 | } | |
176 | ||
6b124d8d SW |
177 | void gfs2_set_inode_flags(struct inode *inode) |
178 | { | |
179 | struct gfs2_inode *ip = GFS2_I(inode); | |
6b124d8d SW |
180 | unsigned int flags = inode->i_flags; |
181 | ||
9964afbb SW |
182 | flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC); |
183 | if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode)) | |
01e64ee4 | 184 | flags |= S_NOSEC; |
383f01fb | 185 | if (ip->i_diskflags & GFS2_DIF_IMMUTABLE) |
6b124d8d | 186 | flags |= S_IMMUTABLE; |
383f01fb | 187 | if (ip->i_diskflags & GFS2_DIF_APPENDONLY) |
6b124d8d | 188 | flags |= S_APPEND; |
383f01fb | 189 | if (ip->i_diskflags & GFS2_DIF_NOATIME) |
6b124d8d | 190 | flags |= S_NOATIME; |
383f01fb | 191 | if (ip->i_diskflags & GFS2_DIF_SYNC) |
6b124d8d SW |
192 | flags |= S_SYNC; |
193 | inode->i_flags = flags; | |
194 | } | |
195 | ||
71b86f56 SW |
196 | /* Flags that can be set by user space */ |
197 | #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \ | |
71b86f56 SW |
198 | GFS2_DIF_IMMUTABLE| \ |
199 | GFS2_DIF_APPENDONLY| \ | |
200 | GFS2_DIF_NOATIME| \ | |
201 | GFS2_DIF_SYNC| \ | |
202 | GFS2_DIF_SYSTEM| \ | |
23d0bb83 | 203 | GFS2_DIF_TOPDIR| \ |
71b86f56 SW |
204 | GFS2_DIF_INHERIT_JDATA) |
205 | ||
206 | /** | |
9dd868e1 FF |
207 | * do_gfs2_set_flags - set flags on an inode |
208 | * @filp: file pointer | |
209 | * @reqflags: The flags to set | |
71b86f56 SW |
210 | * @mask: Indicates which flags are valid |
211 | * | |
212 | */ | |
b09e593d | 213 | static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask) |
71b86f56 | 214 | { |
496ad9aa | 215 | struct inode *inode = file_inode(filp); |
feaa7bba SW |
216 | struct gfs2_inode *ip = GFS2_I(inode); |
217 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
71b86f56 SW |
218 | struct buffer_head *bh; |
219 | struct gfs2_holder gh; | |
220 | int error; | |
55eccc6d | 221 | u32 new_flags, flags; |
71b86f56 | 222 | |
a561be71 | 223 | error = mnt_want_write_file(filp); |
52f341cf | 224 | if (error) |
71b86f56 SW |
225 | return error; |
226 | ||
f58ba889 MS |
227 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); |
228 | if (error) | |
229 | goto out_drop_write; | |
230 | ||
7df0e039 | 231 | error = -EACCES; |
2e149670 | 232 | if (!inode_owner_or_capable(inode)) |
7df0e039 SW |
233 | goto out; |
234 | ||
235 | error = 0; | |
383f01fb | 236 | flags = ip->i_diskflags; |
55eccc6d | 237 | new_flags = (flags & ~mask) | (reqflags & mask); |
71b86f56 SW |
238 | if ((new_flags ^ flags) == 0) |
239 | goto out; | |
240 | ||
241 | error = -EINVAL; | |
242 | if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET) | |
243 | goto out; | |
244 | ||
71b86f56 SW |
245 | error = -EPERM; |
246 | if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE)) | |
247 | goto out; | |
248 | if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY)) | |
249 | goto out; | |
907b9bce | 250 | if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) && |
b9cb9813 | 251 | !capable(CAP_LINUX_IMMUTABLE)) |
71b86f56 | 252 | goto out; |
b9cb9813 | 253 | if (!IS_IMMUTABLE(inode)) { |
10556cb2 | 254 | error = gfs2_permission(inode, MAY_WRITE); |
b9cb9813 SW |
255 | if (error) |
256 | goto out; | |
257 | } | |
5561093e SW |
258 | if ((flags ^ new_flags) & GFS2_DIF_JDATA) { |
259 | if (flags & GFS2_DIF_JDATA) | |
24972557 | 260 | gfs2_log_flush(sdp, ip->i_gl, NORMAL_FLUSH); |
5561093e SW |
261 | error = filemap_fdatawrite(inode->i_mapping); |
262 | if (error) | |
263 | goto out; | |
264 | error = filemap_fdatawait(inode->i_mapping); | |
265 | if (error) | |
266 | goto out; | |
267 | } | |
55eccc6d | 268 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); |
71b86f56 SW |
269 | if (error) |
270 | goto out; | |
55eccc6d SW |
271 | error = gfs2_meta_inode_buffer(ip, &bh); |
272 | if (error) | |
273 | goto out_trans_end; | |
350a9b0a | 274 | gfs2_trans_add_meta(ip->i_gl, bh); |
383f01fb | 275 | ip->i_diskflags = new_flags; |
539e5d6b | 276 | gfs2_dinode_out(ip, bh->b_data); |
71b86f56 | 277 | brelse(bh); |
6b124d8d | 278 | gfs2_set_inode_flags(inode); |
5561093e | 279 | gfs2_set_aops(inode); |
55eccc6d SW |
280 | out_trans_end: |
281 | gfs2_trans_end(sdp); | |
71b86f56 SW |
282 | out: |
283 | gfs2_glock_dq_uninit(&gh); | |
f58ba889 | 284 | out_drop_write: |
2a79f17e | 285 | mnt_drop_write_file(filp); |
71b86f56 SW |
286 | return error; |
287 | } | |
288 | ||
b09e593d | 289 | static int gfs2_set_flags(struct file *filp, u32 __user *ptr) |
71b86f56 | 290 | { |
496ad9aa | 291 | struct inode *inode = file_inode(filp); |
128e5eba | 292 | u32 fsflags, gfsflags; |
7df0e039 | 293 | |
128e5eba | 294 | if (get_user(fsflags, ptr)) |
71b86f56 | 295 | return -EFAULT; |
7df0e039 | 296 | |
128e5eba | 297 | gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags); |
b9af7ca6 | 298 | if (!S_ISDIR(inode->i_mode)) { |
23d0bb83 | 299 | gfsflags &= ~GFS2_DIF_TOPDIR; |
b9af7ca6 SW |
300 | if (gfsflags & GFS2_DIF_INHERIT_JDATA) |
301 | gfsflags ^= (GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA); | |
acc546fd | 302 | return do_gfs2_set_flags(filp, gfsflags, ~GFS2_DIF_SYSTEM); |
b9af7ca6 | 303 | } |
acc546fd | 304 | return do_gfs2_set_flags(filp, gfsflags, ~(GFS2_DIF_SYSTEM | GFS2_DIF_JDATA)); |
71b86f56 SW |
305 | } |
306 | ||
b09e593d | 307 | static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) |
71b86f56 SW |
308 | { |
309 | switch(cmd) { | |
128e5eba | 310 | case FS_IOC_GETFLAGS: |
b09e593d | 311 | return gfs2_get_flags(filp, (u32 __user *)arg); |
128e5eba | 312 | case FS_IOC_SETFLAGS: |
b09e593d | 313 | return gfs2_set_flags(filp, (u32 __user *)arg); |
66fc061b SW |
314 | case FITRIM: |
315 | return gfs2_fitrim(filp, (void __user *)arg); | |
71b86f56 SW |
316 | } |
317 | return -ENOTTY; | |
318 | } | |
319 | ||
da1dfb6a SW |
320 | /** |
321 | * gfs2_size_hint - Give a hint to the size of a write request | |
9dd868e1 | 322 | * @filep: The struct file |
da1dfb6a SW |
323 | * @offset: The file offset of the write |
324 | * @size: The length of the write | |
325 | * | |
326 | * When we are about to do a write, this function records the total | |
327 | * write size in order to provide a suitable hint to the lower layers | |
328 | * about how many blocks will be required. | |
329 | * | |
330 | */ | |
331 | ||
332 | static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size) | |
333 | { | |
496ad9aa | 334 | struct inode *inode = file_inode(filep); |
da1dfb6a SW |
335 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
336 | struct gfs2_inode *ip = GFS2_I(inode); | |
337 | size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift; | |
338 | int hint = min_t(size_t, INT_MAX, blks); | |
339 | ||
a097dc7e BP |
340 | if (hint > atomic_read(&ip->i_res.rs_sizehint)) |
341 | atomic_set(&ip->i_res.rs_sizehint, hint); | |
da1dfb6a SW |
342 | } |
343 | ||
3cc3f710 SW |
344 | /** |
345 | * gfs2_allocate_page_backing - Use bmap to allocate blocks | |
346 | * @page: The (locked) page to allocate backing for | |
347 | * | |
348 | * We try to allocate all the blocks required for the page in | |
349 | * one go. This might fail for various reasons, so we keep | |
350 | * trying until all the blocks to back this page are allocated. | |
351 | * If some of the blocks are already allocated, thats ok too. | |
352 | */ | |
353 | ||
354 | static int gfs2_allocate_page_backing(struct page *page) | |
355 | { | |
356 | struct inode *inode = page->mapping->host; | |
357 | struct buffer_head bh; | |
09cbfeaf KS |
358 | unsigned long size = PAGE_SIZE; |
359 | u64 lblock = page->index << (PAGE_SHIFT - inode->i_blkbits); | |
3cc3f710 SW |
360 | |
361 | do { | |
362 | bh.b_state = 0; | |
363 | bh.b_size = size; | |
e9e1ef2b | 364 | gfs2_block_map(inode, lblock, &bh, 1); |
3cc3f710 SW |
365 | if (!buffer_mapped(&bh)) |
366 | return -EIO; | |
367 | size -= bh.b_size; | |
368 | lblock += (bh.b_size >> inode->i_blkbits); | |
369 | } while(size > 0); | |
370 | return 0; | |
371 | } | |
372 | ||
373 | /** | |
374 | * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable | |
375 | * @vma: The virtual memory area | |
9dd868e1 | 376 | * @vmf: The virtual memory fault containing the page to become writable |
3cc3f710 SW |
377 | * |
378 | * When the page becomes writable, we need to ensure that we have | |
379 | * blocks allocated on disk to back that page. | |
380 | */ | |
381 | ||
c2ec175c | 382 | static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
3cc3f710 | 383 | { |
c2ec175c | 384 | struct page *page = vmf->page; |
496ad9aa | 385 | struct inode *inode = file_inode(vma->vm_file); |
3cc3f710 SW |
386 | struct gfs2_inode *ip = GFS2_I(inode); |
387 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
7b9cff46 | 388 | struct gfs2_alloc_parms ap = { .aflags = 0, }; |
3cc3f710 | 389 | unsigned long last_index; |
09cbfeaf | 390 | u64 pos = page->index << PAGE_SHIFT; |
3cc3f710 | 391 | unsigned int data_blocks, ind_blocks, rblocks; |
3cc3f710 | 392 | struct gfs2_holder gh; |
13d921e3 | 393 | loff_t size; |
3cc3f710 SW |
394 | int ret; |
395 | ||
39263d5e | 396 | sb_start_pagefault(inode->i_sb); |
13d921e3 | 397 | |
a63e9b2e JK |
398 | /* Update file times before taking page lock */ |
399 | file_update_time(vma->vm_file); | |
13d921e3 | 400 | |
b54e9a0b | 401 | ret = gfs2_rsqa_alloc(ip); |
0a305e49 | 402 | if (ret) |
a097dc7e | 403 | goto out; |
0a305e49 | 404 | |
09cbfeaf | 405 | gfs2_size_hint(vma->vm_file, pos, PAGE_SIZE); |
8e2e0047 | 406 | |
719ee344 SW |
407 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); |
408 | ret = gfs2_glock_nq(&gh); | |
3cc3f710 | 409 | if (ret) |
2b3dcf35 | 410 | goto out_uninit; |
3cc3f710 | 411 | |
9c538837 SW |
412 | set_bit(GLF_DIRTY, &ip->i_gl->gl_flags); |
413 | set_bit(GIF_SW_PAGED, &ip->i_flags); | |
414 | ||
09cbfeaf | 415 | if (!gfs2_write_alloc_required(ip, pos, PAGE_SIZE)) { |
13d921e3 SW |
416 | lock_page(page); |
417 | if (!PageUptodate(page) || page->mapping != inode->i_mapping) { | |
418 | ret = -EAGAIN; | |
419 | unlock_page(page); | |
420 | } | |
3cc3f710 | 421 | goto out_unlock; |
13d921e3 SW |
422 | } |
423 | ||
5407e242 BP |
424 | ret = gfs2_rindex_update(sdp); |
425 | if (ret) | |
6dbd8224 SW |
426 | goto out_unlock; |
427 | ||
09cbfeaf | 428 | gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks); |
7b9cff46 | 429 | ap.target = data_blocks + ind_blocks; |
b8fbf471 AD |
430 | ret = gfs2_quota_lock_check(ip, &ap); |
431 | if (ret) | |
432 | goto out_unlock; | |
7b9cff46 | 433 | ret = gfs2_inplace_reserve(ip, &ap); |
3cc3f710 SW |
434 | if (ret) |
435 | goto out_quota_unlock; | |
436 | ||
437 | rblocks = RES_DINODE + ind_blocks; | |
438 | if (gfs2_is_jdata(ip)) | |
439 | rblocks += data_blocks ? data_blocks : 1; | |
bf97b673 | 440 | if (ind_blocks || data_blocks) { |
3cc3f710 | 441 | rblocks += RES_STATFS + RES_QUOTA; |
71f890f7 | 442 | rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks); |
bf97b673 | 443 | } |
3cc3f710 SW |
444 | ret = gfs2_trans_begin(sdp, rblocks, 0); |
445 | if (ret) | |
446 | goto out_trans_fail; | |
447 | ||
448 | lock_page(page); | |
449 | ret = -EINVAL; | |
13d921e3 | 450 | size = i_size_read(inode); |
09cbfeaf | 451 | last_index = (size - 1) >> PAGE_SHIFT; |
13d921e3 SW |
452 | /* Check page index against inode size */ |
453 | if (size == 0 || (page->index > last_index)) | |
454 | goto out_trans_end; | |
455 | ||
456 | ret = -EAGAIN; | |
457 | /* If truncated, we must retry the operation, we may have raced | |
458 | * with the glock demotion code. | |
459 | */ | |
460 | if (!PageUptodate(page) || page->mapping != inode->i_mapping) | |
461 | goto out_trans_end; | |
462 | ||
463 | /* Unstuff, if required, and allocate backing blocks for page */ | |
b7fe2e39 | 464 | ret = 0; |
13d921e3 | 465 | if (gfs2_is_stuffed(ip)) |
3cc3f710 | 466 | ret = gfs2_unstuff_dinode(ip, page); |
13d921e3 SW |
467 | if (ret == 0) |
468 | ret = gfs2_allocate_page_backing(page); | |
3cc3f710 | 469 | |
13d921e3 SW |
470 | out_trans_end: |
471 | if (ret) | |
472 | unlock_page(page); | |
3cc3f710 SW |
473 | gfs2_trans_end(sdp); |
474 | out_trans_fail: | |
475 | gfs2_inplace_release(ip); | |
476 | out_quota_unlock: | |
477 | gfs2_quota_unlock(ip); | |
3cc3f710 SW |
478 | out_unlock: |
479 | gfs2_glock_dq(&gh); | |
2b3dcf35 | 480 | out_uninit: |
3cc3f710 | 481 | gfs2_holder_uninit(&gh); |
13d921e3 SW |
482 | if (ret == 0) { |
483 | set_page_dirty(page); | |
1d1d1a76 | 484 | wait_for_stable_page(page); |
13d921e3 | 485 | } |
2b3dcf35 | 486 | out: |
39263d5e | 487 | sb_end_pagefault(inode->i_sb); |
13d921e3 | 488 | return block_page_mkwrite_return(ret); |
3cc3f710 SW |
489 | } |
490 | ||
f0f37e2f | 491 | static const struct vm_operations_struct gfs2_vm_ops = { |
3cc3f710 | 492 | .fault = filemap_fault, |
f1820361 | 493 | .map_pages = filemap_map_pages, |
3cc3f710 SW |
494 | .page_mkwrite = gfs2_page_mkwrite, |
495 | }; | |
496 | ||
b3b94faa DT |
497 | /** |
498 | * gfs2_mmap - | |
499 | * @file: The file to map | |
500 | * @vma: The VMA which described the mapping | |
501 | * | |
48bf2b17 SW |
502 | * There is no need to get a lock here unless we should be updating |
503 | * atime. We ignore any locking errors since the only consequence is | |
504 | * a missed atime update (which will just be deferred until later). | |
505 | * | |
506 | * Returns: 0 | |
b3b94faa DT |
507 | */ |
508 | ||
509 | static int gfs2_mmap(struct file *file, struct vm_area_struct *vma) | |
510 | { | |
feaa7bba | 511 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); |
b3b94faa | 512 | |
b9c93bb7 SW |
513 | if (!(file->f_flags & O_NOATIME) && |
514 | !IS_NOATIME(&ip->i_inode)) { | |
48bf2b17 SW |
515 | struct gfs2_holder i_gh; |
516 | int error; | |
b3b94faa | 517 | |
3d162688 BM |
518 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, |
519 | &i_gh); | |
b9c93bb7 SW |
520 | if (error) |
521 | return error; | |
3d162688 BM |
522 | /* grab lock to update inode */ |
523 | gfs2_glock_dq_uninit(&i_gh); | |
524 | file_accessed(file); | |
48bf2b17 | 525 | } |
3cc3f710 | 526 | vma->vm_ops = &gfs2_vm_ops; |
b3b94faa | 527 | |
48bf2b17 | 528 | return 0; |
b3b94faa DT |
529 | } |
530 | ||
531 | /** | |
6d4ade98 SW |
532 | * gfs2_open_common - This is common to open and atomic_open |
533 | * @inode: The inode being opened | |
534 | * @file: The file being opened | |
b3b94faa | 535 | * |
6d4ade98 SW |
536 | * This maybe called under a glock or not depending upon how it has |
537 | * been called. We must always be called under a glock for regular | |
538 | * files, however. For other file types, it does not matter whether | |
539 | * we hold the glock or not. | |
540 | * | |
541 | * Returns: Error code or 0 for success | |
b3b94faa DT |
542 | */ |
543 | ||
6d4ade98 | 544 | int gfs2_open_common(struct inode *inode, struct file *file) |
b3b94faa | 545 | { |
b3b94faa | 546 | struct gfs2_file *fp; |
6d4ade98 SW |
547 | int ret; |
548 | ||
549 | if (S_ISREG(inode->i_mode)) { | |
550 | ret = generic_file_open(inode, file); | |
551 | if (ret) | |
552 | return ret; | |
553 | } | |
b3b94faa | 554 | |
6d4ade98 | 555 | fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS); |
b3b94faa DT |
556 | if (!fp) |
557 | return -ENOMEM; | |
558 | ||
f55ab26a | 559 | mutex_init(&fp->f_fl_mutex); |
b3b94faa | 560 | |
feaa7bba | 561 | gfs2_assert_warn(GFS2_SB(inode), !file->private_data); |
5c676f6d | 562 | file->private_data = fp; |
6d4ade98 SW |
563 | return 0; |
564 | } | |
565 | ||
566 | /** | |
567 | * gfs2_open - open a file | |
568 | * @inode: the inode to open | |
569 | * @file: the struct file for this opening | |
570 | * | |
571 | * After atomic_open, this function is only used for opening files | |
572 | * which are already cached. We must still get the glock for regular | |
573 | * files to ensure that we have the file size uptodate for the large | |
574 | * file check which is in the common code. That is only an issue for | |
575 | * regular files though. | |
576 | * | |
577 | * Returns: errno | |
578 | */ | |
579 | ||
580 | static int gfs2_open(struct inode *inode, struct file *file) | |
581 | { | |
582 | struct gfs2_inode *ip = GFS2_I(inode); | |
583 | struct gfs2_holder i_gh; | |
584 | int error; | |
585 | bool need_unlock = false; | |
b3b94faa | 586 | |
b60623c2 | 587 | if (S_ISREG(ip->i_inode.i_mode)) { |
b3b94faa DT |
588 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, |
589 | &i_gh); | |
590 | if (error) | |
6d4ade98 SW |
591 | return error; |
592 | need_unlock = true; | |
593 | } | |
b3b94faa | 594 | |
6d4ade98 | 595 | error = gfs2_open_common(inode, file); |
b3b94faa | 596 | |
6d4ade98 | 597 | if (need_unlock) |
b3b94faa | 598 | gfs2_glock_dq_uninit(&i_gh); |
b3b94faa | 599 | |
b3b94faa DT |
600 | return error; |
601 | } | |
602 | ||
603 | /** | |
df3fd117 | 604 | * gfs2_release - called to close a struct file |
b3b94faa DT |
605 | * @inode: the inode the struct file belongs to |
606 | * @file: the struct file being closed | |
607 | * | |
608 | * Returns: errno | |
609 | */ | |
610 | ||
df3fd117 | 611 | static int gfs2_release(struct inode *inode, struct file *file) |
b3b94faa | 612 | { |
0a305e49 | 613 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa | 614 | |
8e2e0047 | 615 | kfree(file->private_data); |
5c676f6d | 616 | file->private_data = NULL; |
b3b94faa | 617 | |
2b3dcf35 BP |
618 | if (!(file->f_mode & FMODE_WRITE)) |
619 | return 0; | |
0a305e49 | 620 | |
b54e9a0b | 621 | gfs2_rsqa_delete(ip, &inode->i_writecount); |
b3b94faa DT |
622 | return 0; |
623 | } | |
624 | ||
625 | /** | |
626 | * gfs2_fsync - sync the dirty data for a file (across the cluster) | |
02c24a82 JB |
627 | * @file: the file that points to the dentry |
628 | * @start: the start position in the file to sync | |
629 | * @end: the end position in the file to sync | |
dba898b0 | 630 | * @datasync: set if we can ignore timestamp changes |
b3b94faa | 631 | * |
2f0264d5 SW |
632 | * We split the data flushing here so that we don't wait for the data |
633 | * until after we've also sent the metadata to disk. Note that for | |
634 | * data=ordered, we will write & wait for the data at the log flush | |
635 | * stage anyway, so this is unlikely to make much of a difference | |
636 | * except in the data=writeback case. | |
637 | * | |
638 | * If the fdatawrite fails due to any reason except -EIO, we will | |
639 | * continue the remainder of the fsync, although we'll still report | |
640 | * the error at the end. This is to match filemap_write_and_wait_range() | |
641 | * behaviour. | |
34126f9f | 642 | * |
b3b94faa DT |
643 | * Returns: errno |
644 | */ | |
645 | ||
02c24a82 JB |
646 | static int gfs2_fsync(struct file *file, loff_t start, loff_t end, |
647 | int datasync) | |
b3b94faa | 648 | { |
2f0264d5 SW |
649 | struct address_space *mapping = file->f_mapping; |
650 | struct inode *inode = mapping->host; | |
0ae45f63 | 651 | int sync_state = inode->i_state & I_DIRTY_ALL; |
dba898b0 | 652 | struct gfs2_inode *ip = GFS2_I(inode); |
87654896 | 653 | int ret = 0, ret1 = 0; |
b3b94faa | 654 | |
2f0264d5 SW |
655 | if (mapping->nrpages) { |
656 | ret1 = filemap_fdatawrite_range(mapping, start, end); | |
657 | if (ret1 == -EIO) | |
658 | return ret1; | |
659 | } | |
02c24a82 | 660 | |
0c901809 BM |
661 | if (!gfs2_is_jdata(ip)) |
662 | sync_state &= ~I_DIRTY_PAGES; | |
dba898b0 | 663 | if (datasync) |
0ae45f63 | 664 | sync_state &= ~(I_DIRTY_SYNC | I_DIRTY_TIME); |
b3b94faa | 665 | |
dba898b0 SW |
666 | if (sync_state) { |
667 | ret = sync_inode_metadata(inode, 1); | |
b5b24d7a | 668 | if (ret) |
dba898b0 | 669 | return ret; |
f1818529 SW |
670 | if (gfs2_is_jdata(ip)) |
671 | filemap_write_and_wait(mapping); | |
b5b24d7a | 672 | gfs2_ail_flush(ip->i_gl, 1); |
33c3de32 SW |
673 | } |
674 | ||
2f0264d5 SW |
675 | if (mapping->nrpages) |
676 | ret = filemap_fdatawait_range(mapping, start, end); | |
677 | ||
678 | return ret ? ret : ret1; | |
b3b94faa DT |
679 | } |
680 | ||
56aa616a | 681 | /** |
da56e45b | 682 | * gfs2_file_write_iter - Perform a write to a file |
56aa616a SW |
683 | * @iocb: The io context |
684 | * @iov: The data to write | |
685 | * @nr_segs: Number of @iov segments | |
686 | * @pos: The file position | |
687 | * | |
688 | * We have to do a lock/unlock here to refresh the inode size for | |
689 | * O_APPEND writes, otherwise we can land up writing at the wrong | |
690 | * offset. There is still a race, but provided the app is using its | |
691 | * own file locking, this will make O_APPEND work as expected. | |
692 | * | |
693 | */ | |
694 | ||
da56e45b | 695 | static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
56aa616a SW |
696 | { |
697 | struct file *file = iocb->ki_filp; | |
496ad9aa | 698 | struct gfs2_inode *ip = GFS2_I(file_inode(file)); |
0a305e49 BP |
699 | int ret; |
700 | ||
b54e9a0b | 701 | ret = gfs2_rsqa_alloc(ip); |
0a305e49 BP |
702 | if (ret) |
703 | return ret; | |
56aa616a | 704 | |
da56e45b | 705 | gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from)); |
da1dfb6a | 706 | |
2ba48ce5 | 707 | if (iocb->ki_flags & IOCB_APPEND) { |
56aa616a | 708 | struct gfs2_holder gh; |
56aa616a SW |
709 | |
710 | ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh); | |
711 | if (ret) | |
712 | return ret; | |
713 | gfs2_glock_dq_uninit(&gh); | |
714 | } | |
715 | ||
da56e45b | 716 | return generic_file_write_iter(iocb, from); |
56aa616a SW |
717 | } |
718 | ||
2fe17c10 CH |
719 | static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len, |
720 | int mode) | |
721 | { | |
722 | struct gfs2_inode *ip = GFS2_I(inode); | |
723 | struct buffer_head *dibh; | |
724 | int error; | |
64dd153c BM |
725 | unsigned int nr_blks; |
726 | sector_t lblock = offset >> inode->i_blkbits; | |
2fe17c10 CH |
727 | |
728 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
729 | if (unlikely(error)) | |
64dd153c | 730 | return error; |
2fe17c10 | 731 | |
350a9b0a | 732 | gfs2_trans_add_meta(ip->i_gl, dibh); |
2fe17c10 CH |
733 | |
734 | if (gfs2_is_stuffed(ip)) { | |
735 | error = gfs2_unstuff_dinode(ip, NULL); | |
736 | if (unlikely(error)) | |
737 | goto out; | |
738 | } | |
739 | ||
64dd153c BM |
740 | while (len) { |
741 | struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 }; | |
742 | bh_map.b_size = len; | |
743 | set_buffer_zeronew(&bh_map); | |
2fe17c10 | 744 | |
64dd153c BM |
745 | error = gfs2_block_map(inode, lblock, &bh_map, 1); |
746 | if (unlikely(error)) | |
747 | goto out; | |
748 | len -= bh_map.b_size; | |
749 | nr_blks = bh_map.b_size >> inode->i_blkbits; | |
750 | lblock += nr_blks; | |
751 | if (!buffer_new(&bh_map)) | |
752 | continue; | |
753 | if (unlikely(!buffer_zeronew(&bh_map))) { | |
754 | error = -EIO; | |
2fe17c10 | 755 | goto out; |
64dd153c | 756 | } |
2fe17c10 | 757 | } |
2fe17c10 | 758 | out: |
64dd153c | 759 | brelse(dibh); |
2fe17c10 CH |
760 | return error; |
761 | } | |
d9be0cda AD |
762 | /** |
763 | * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of | |
764 | * blocks, determine how many bytes can be written. | |
765 | * @ip: The inode in question. | |
766 | * @len: Max cap of bytes. What we return in *len must be <= this. | |
767 | * @data_blocks: Compute and return the number of data blocks needed | |
768 | * @ind_blocks: Compute and return the number of indirect blocks needed | |
769 | * @max_blocks: The total blocks available to work with. | |
770 | * | |
771 | * Returns: void, but @len, @data_blocks and @ind_blocks are filled in. | |
772 | */ | |
773 | static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len, | |
774 | unsigned int *data_blocks, unsigned int *ind_blocks, | |
775 | unsigned int max_blocks) | |
2fe17c10 | 776 | { |
d9be0cda | 777 | loff_t max = *len; |
2fe17c10 | 778 | const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
2fe17c10 CH |
779 | unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1); |
780 | ||
781 | for (tmp = max_data; tmp > sdp->sd_diptrs;) { | |
782 | tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs); | |
783 | max_data -= tmp; | |
784 | } | |
d9be0cda | 785 | |
2fe17c10 CH |
786 | *data_blocks = max_data; |
787 | *ind_blocks = max_blocks - max_data; | |
788 | *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift; | |
789 | if (*len > max) { | |
790 | *len = max; | |
791 | gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks); | |
792 | } | |
793 | } | |
794 | ||
9c9f1159 | 795 | static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
2fe17c10 | 796 | { |
496ad9aa | 797 | struct inode *inode = file_inode(file); |
2fe17c10 CH |
798 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
799 | struct gfs2_inode *ip = GFS2_I(inode); | |
7b9cff46 | 800 | struct gfs2_alloc_parms ap = { .aflags = 0, }; |
2fe17c10 | 801 | unsigned int data_blocks = 0, ind_blocks = 0, rblocks; |
d9be0cda | 802 | loff_t bytes, max_bytes, max_blks = UINT_MAX; |
2fe17c10 | 803 | int error; |
4442f2e0 SW |
804 | const loff_t pos = offset; |
805 | const loff_t count = len; | |
6905d9e4 | 806 | loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1); |
2fe17c10 | 807 | loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift; |
64dd153c | 808 | loff_t max_chunk_size = UINT_MAX & bsize_mask; |
a0846a53 | 809 | |
2fe17c10 CH |
810 | next = (next + 1) << sdp->sd_sb.sb_bsize_shift; |
811 | ||
6905d9e4 | 812 | offset &= bsize_mask; |
2fe17c10 CH |
813 | |
814 | len = next - offset; | |
815 | bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2; | |
816 | if (!bytes) | |
817 | bytes = UINT_MAX; | |
6905d9e4 BM |
818 | bytes &= bsize_mask; |
819 | if (bytes == 0) | |
820 | bytes = sdp->sd_sb.sb_bsize; | |
2fe17c10 | 821 | |
da1dfb6a | 822 | gfs2_size_hint(file, offset, len); |
8e2e0047 | 823 | |
d9be0cda AD |
824 | gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks); |
825 | ap.min_target = data_blocks + ind_blocks; | |
826 | ||
2fe17c10 CH |
827 | while (len > 0) { |
828 | if (len < bytes) | |
829 | bytes = len; | |
58a7d5fb BM |
830 | if (!gfs2_write_alloc_required(ip, offset, bytes)) { |
831 | len -= bytes; | |
832 | offset += bytes; | |
833 | continue; | |
834 | } | |
d9be0cda AD |
835 | |
836 | /* We need to determine how many bytes we can actually | |
837 | * fallocate without exceeding quota or going over the | |
838 | * end of the fs. We start off optimistically by assuming | |
839 | * we can write max_bytes */ | |
840 | max_bytes = (len > max_chunk_size) ? max_chunk_size : len; | |
841 | ||
842 | /* Since max_bytes is most likely a theoretical max, we | |
843 | * calculate a more realistic 'bytes' to serve as a good | |
844 | * starting point for the number of bytes we may be able | |
845 | * to write */ | |
2fe17c10 | 846 | gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks); |
7b9cff46 | 847 | ap.target = data_blocks + ind_blocks; |
b8fbf471 AD |
848 | |
849 | error = gfs2_quota_lock_check(ip, &ap); | |
2fe17c10 | 850 | if (error) |
9c9f1159 | 851 | return error; |
d9be0cda AD |
852 | /* ap.allowed tells us how many blocks quota will allow |
853 | * us to write. Check if this reduces max_blks */ | |
854 | if (ap.allowed && ap.allowed < max_blks) | |
855 | max_blks = ap.allowed; | |
2fe17c10 | 856 | |
7b9cff46 | 857 | error = gfs2_inplace_reserve(ip, &ap); |
d9be0cda | 858 | if (error) |
2fe17c10 | 859 | goto out_qunlock; |
d9be0cda AD |
860 | |
861 | /* check if the selected rgrp limits our max_blks further */ | |
862 | if (ap.allowed && ap.allowed < max_blks) | |
863 | max_blks = ap.allowed; | |
864 | ||
865 | /* Almost done. Calculate bytes that can be written using | |
866 | * max_blks. We also recompute max_bytes, data_blocks and | |
867 | * ind_blocks */ | |
868 | calc_max_reserv(ip, &max_bytes, &data_blocks, | |
869 | &ind_blocks, max_blks); | |
2fe17c10 CH |
870 | |
871 | rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA + | |
71f890f7 | 872 | RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks); |
2fe17c10 CH |
873 | if (gfs2_is_jdata(ip)) |
874 | rblocks += data_blocks ? data_blocks : 1; | |
875 | ||
876 | error = gfs2_trans_begin(sdp, rblocks, | |
09cbfeaf | 877 | PAGE_SIZE/sdp->sd_sb.sb_bsize); |
2fe17c10 CH |
878 | if (error) |
879 | goto out_trans_fail; | |
880 | ||
881 | error = fallocate_chunk(inode, offset, max_bytes, mode); | |
882 | gfs2_trans_end(sdp); | |
883 | ||
884 | if (error) | |
885 | goto out_trans_fail; | |
886 | ||
887 | len -= max_bytes; | |
888 | offset += max_bytes; | |
889 | gfs2_inplace_release(ip); | |
890 | gfs2_quota_unlock(ip); | |
2fe17c10 | 891 | } |
4442f2e0 | 892 | |
1885867b AP |
893 | if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size) { |
894 | i_size_write(inode, pos + count); | |
98f1a696 | 895 | file_update_time(file); |
4b813f09 | 896 | mark_inode_dirty(inode); |
1885867b AP |
897 | } |
898 | ||
dde0c2e7 CH |
899 | if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host)) |
900 | return vfs_fsync_range(file, pos, pos + count - 1, | |
901 | (file->f_flags & __O_SYNC) ? 0 : 1); | |
902 | return 0; | |
2fe17c10 CH |
903 | |
904 | out_trans_fail: | |
905 | gfs2_inplace_release(ip); | |
906 | out_qunlock: | |
907 | gfs2_quota_unlock(ip); | |
9c9f1159 AP |
908 | return error; |
909 | } | |
910 | ||
911 | static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len) | |
912 | { | |
913 | struct inode *inode = file_inode(file); | |
914 | struct gfs2_inode *ip = GFS2_I(inode); | |
915 | struct gfs2_holder gh; | |
916 | int ret; | |
917 | ||
86066914 | 918 | if ((mode & ~FALLOC_FL_KEEP_SIZE) || gfs2_is_jdata(ip)) |
9c9f1159 AP |
919 | return -EOPNOTSUPP; |
920 | ||
5955102c | 921 | inode_lock(inode); |
9c9f1159 AP |
922 | |
923 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); | |
924 | ret = gfs2_glock_nq(&gh); | |
925 | if (ret) | |
926 | goto out_uninit; | |
927 | ||
928 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
929 | (offset + len) > inode->i_size) { | |
930 | ret = inode_newsize_ok(inode, offset + len); | |
931 | if (ret) | |
932 | goto out_unlock; | |
933 | } | |
934 | ||
935 | ret = get_write_access(inode); | |
936 | if (ret) | |
937 | goto out_unlock; | |
938 | ||
b54e9a0b | 939 | ret = gfs2_rsqa_alloc(ip); |
9c9f1159 AP |
940 | if (ret) |
941 | goto out_putw; | |
942 | ||
943 | ret = __gfs2_fallocate(file, mode, offset, len); | |
944 | if (ret) | |
a097dc7e BP |
945 | gfs2_rs_deltree(&ip->i_res); |
946 | ||
9c9f1159 AP |
947 | out_putw: |
948 | put_write_access(inode); | |
2fe17c10 | 949 | out_unlock: |
a0846a53 | 950 | gfs2_glock_dq(&gh); |
2fe17c10 | 951 | out_uninit: |
a0846a53 | 952 | gfs2_holder_uninit(&gh); |
5955102c | 953 | inode_unlock(inode); |
9c9f1159 | 954 | return ret; |
2fe17c10 CH |
955 | } |
956 | ||
61152675 AD |
957 | static ssize_t gfs2_file_splice_read(struct file *in, loff_t *ppos, |
958 | struct pipe_inode_info *pipe, size_t len, | |
959 | unsigned int flags) | |
960 | { | |
961 | struct inode *inode = in->f_mapping->host; | |
962 | struct gfs2_inode *ip = GFS2_I(inode); | |
963 | struct gfs2_holder gh; | |
964 | int ret; | |
965 | ||
80f4781d | 966 | inode_lock(inode); |
61152675 AD |
967 | |
968 | ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh); | |
969 | if (ret) { | |
80f4781d | 970 | inode_unlock(inode); |
61152675 AD |
971 | return ret; |
972 | } | |
973 | ||
974 | gfs2_glock_dq_uninit(&gh); | |
80f4781d | 975 | inode_unlock(inode); |
61152675 AD |
976 | |
977 | return generic_file_splice_read(in, ppos, pipe, len, flags); | |
978 | } | |
979 | ||
980 | ||
f1ea6f4e BP |
981 | static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe, |
982 | struct file *out, loff_t *ppos, | |
983 | size_t len, unsigned int flags) | |
984 | { | |
985 | int error; | |
986 | struct gfs2_inode *ip = GFS2_I(out->f_mapping->host); | |
987 | ||
b54e9a0b | 988 | error = gfs2_rsqa_alloc(ip); |
f1ea6f4e BP |
989 | if (error) |
990 | return (ssize_t)error; | |
991 | ||
992 | gfs2_size_hint(out, *ppos, len); | |
993 | ||
994 | return iter_file_splice_write(pipe, out, ppos, len, flags); | |
995 | } | |
996 | ||
f057f6cd SW |
997 | #ifdef CONFIG_GFS2_FS_LOCKING_DLM |
998 | ||
b3b94faa DT |
999 | /** |
1000 | * gfs2_lock - acquire/release a posix lock on a file | |
1001 | * @file: the file pointer | |
1002 | * @cmd: either modify or retrieve lock state, possibly wait | |
1003 | * @fl: type and range of lock | |
1004 | * | |
1005 | * Returns: errno | |
1006 | */ | |
1007 | ||
1008 | static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl) | |
1009 | { | |
feaa7bba SW |
1010 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); |
1011 | struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host); | |
f057f6cd | 1012 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
b3b94faa | 1013 | |
b3b94faa DT |
1014 | if (!(fl->fl_flags & FL_POSIX)) |
1015 | return -ENOLCK; | |
720e7749 | 1016 | if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK) |
b3b94faa DT |
1017 | return -ENOLCK; |
1018 | ||
586759f0 ME |
1019 | if (cmd == F_CANCELLK) { |
1020 | /* Hack: */ | |
1021 | cmd = F_SETLK; | |
1022 | fl->fl_type = F_UNLCK; | |
1023 | } | |
c2952d20 SW |
1024 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) { |
1025 | if (fl->fl_type == F_UNLCK) | |
4f656367 | 1026 | locks_lock_file_wait(file, fl); |
f057f6cd | 1027 | return -EIO; |
c2952d20 | 1028 | } |
b3b94faa | 1029 | if (IS_GETLK(cmd)) |
f057f6cd | 1030 | return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl); |
b3b94faa | 1031 | else if (fl->fl_type == F_UNLCK) |
f057f6cd | 1032 | return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl); |
b3b94faa | 1033 | else |
f057f6cd | 1034 | return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl); |
b3b94faa DT |
1035 | } |
1036 | ||
b3b94faa DT |
1037 | static int do_flock(struct file *file, int cmd, struct file_lock *fl) |
1038 | { | |
5c676f6d | 1039 | struct gfs2_file *fp = file->private_data; |
b3b94faa | 1040 | struct gfs2_holder *fl_gh = &fp->f_fl_gh; |
496ad9aa | 1041 | struct gfs2_inode *ip = GFS2_I(file_inode(file)); |
b3b94faa DT |
1042 | struct gfs2_glock *gl; |
1043 | unsigned int state; | |
b58bf407 | 1044 | u16 flags; |
b3b94faa | 1045 | int error = 0; |
2ddfbdd6 | 1046 | int sleeptime; |
b3b94faa DT |
1047 | |
1048 | state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED; | |
2ddfbdd6 | 1049 | flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT; |
b3b94faa | 1050 | |
f55ab26a | 1051 | mutex_lock(&fp->f_fl_mutex); |
b3b94faa DT |
1052 | |
1053 | gl = fl_gh->gh_gl; | |
1054 | if (gl) { | |
1055 | if (fl_gh->gh_state == state) | |
1056 | goto out; | |
4f656367 | 1057 | locks_lock_file_wait(file, |
a93a9983 JB |
1058 | &(struct file_lock) { |
1059 | .fl_type = F_UNLCK, | |
1060 | .fl_flags = FL_FLOCK | |
1061 | }); | |
5bef3e7c | 1062 | gfs2_glock_dq(fl_gh); |
b4c20166 | 1063 | gfs2_holder_reinit(state, flags, fl_gh); |
b3b94faa | 1064 | } else { |
6802e340 SW |
1065 | error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr, |
1066 | &gfs2_flock_glops, CREATE, &gl); | |
b3b94faa DT |
1067 | if (error) |
1068 | goto out; | |
b4c20166 AD |
1069 | gfs2_holder_init(gl, state, flags, fl_gh); |
1070 | gfs2_glock_put(gl); | |
b3b94faa | 1071 | } |
2ddfbdd6 BP |
1072 | for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) { |
1073 | error = gfs2_glock_nq(fl_gh); | |
1074 | if (error != GLR_TRYFAILED) | |
1075 | break; | |
1076 | fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT; | |
1077 | fl_gh->gh_error = 0; | |
1078 | msleep(sleeptime); | |
1079 | } | |
b3b94faa DT |
1080 | if (error) { |
1081 | gfs2_holder_uninit(fl_gh); | |
1082 | if (error == GLR_TRYFAILED) | |
1083 | error = -EAGAIN; | |
1084 | } else { | |
4f656367 | 1085 | error = locks_lock_file_wait(file, fl); |
feaa7bba | 1086 | gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); |
b3b94faa DT |
1087 | } |
1088 | ||
420b9e5e | 1089 | out: |
f55ab26a | 1090 | mutex_unlock(&fp->f_fl_mutex); |
b3b94faa DT |
1091 | return error; |
1092 | } | |
1093 | ||
1094 | static void do_unflock(struct file *file, struct file_lock *fl) | |
1095 | { | |
5c676f6d | 1096 | struct gfs2_file *fp = file->private_data; |
b3b94faa DT |
1097 | struct gfs2_holder *fl_gh = &fp->f_fl_gh; |
1098 | ||
f55ab26a | 1099 | mutex_lock(&fp->f_fl_mutex); |
4f656367 | 1100 | locks_lock_file_wait(file, fl); |
0a33443b | 1101 | if (fl_gh->gh_gl) { |
2ddfbdd6 | 1102 | gfs2_glock_dq(fl_gh); |
0a33443b SW |
1103 | gfs2_holder_uninit(fl_gh); |
1104 | } | |
f55ab26a | 1105 | mutex_unlock(&fp->f_fl_mutex); |
b3b94faa DT |
1106 | } |
1107 | ||
1108 | /** | |
1109 | * gfs2_flock - acquire/release a flock lock on a file | |
1110 | * @file: the file pointer | |
1111 | * @cmd: either modify or retrieve lock state, possibly wait | |
1112 | * @fl: type and range of lock | |
1113 | * | |
1114 | * Returns: errno | |
1115 | */ | |
1116 | ||
1117 | static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl) | |
1118 | { | |
b3b94faa DT |
1119 | if (!(fl->fl_flags & FL_FLOCK)) |
1120 | return -ENOLCK; | |
a12af1eb AD |
1121 | if (fl->fl_type & LOCK_MAND) |
1122 | return -EOPNOTSUPP; | |
b3b94faa | 1123 | |
b3b94faa DT |
1124 | if (fl->fl_type == F_UNLCK) { |
1125 | do_unflock(file, fl); | |
1126 | return 0; | |
d00223f1 | 1127 | } else { |
b3b94faa | 1128 | return do_flock(file, cmd, fl); |
d00223f1 | 1129 | } |
b3b94faa DT |
1130 | } |
1131 | ||
10d21988 | 1132 | const struct file_operations gfs2_file_fops = { |
26c1a574 | 1133 | .llseek = gfs2_llseek, |
aad4f8bb | 1134 | .read_iter = generic_file_read_iter, |
da56e45b | 1135 | .write_iter = gfs2_file_write_iter, |
26c1a574 SW |
1136 | .unlocked_ioctl = gfs2_ioctl, |
1137 | .mmap = gfs2_mmap, | |
1138 | .open = gfs2_open, | |
df3fd117 | 1139 | .release = gfs2_release, |
26c1a574 SW |
1140 | .fsync = gfs2_fsync, |
1141 | .lock = gfs2_lock, | |
26c1a574 | 1142 | .flock = gfs2_flock, |
61152675 | 1143 | .splice_read = gfs2_file_splice_read, |
f42a69fa | 1144 | .splice_write = gfs2_file_splice_write, |
1c994a09 | 1145 | .setlease = simple_nosetlease, |
2fe17c10 | 1146 | .fallocate = gfs2_fallocate, |
b3b94faa DT |
1147 | }; |
1148 | ||
10d21988 | 1149 | const struct file_operations gfs2_dir_fops = { |
1d1bb236 | 1150 | .iterate_shared = gfs2_readdir, |
26c1a574 SW |
1151 | .unlocked_ioctl = gfs2_ioctl, |
1152 | .open = gfs2_open, | |
df3fd117 | 1153 | .release = gfs2_release, |
26c1a574 SW |
1154 | .fsync = gfs2_fsync, |
1155 | .lock = gfs2_lock, | |
1156 | .flock = gfs2_flock, | |
6038f373 | 1157 | .llseek = default_llseek, |
b3b94faa DT |
1158 | }; |
1159 | ||
f057f6cd SW |
1160 | #endif /* CONFIG_GFS2_FS_LOCKING_DLM */ |
1161 | ||
10d21988 | 1162 | const struct file_operations gfs2_file_fops_nolock = { |
c97bfe43 | 1163 | .llseek = gfs2_llseek, |
aad4f8bb | 1164 | .read_iter = generic_file_read_iter, |
da56e45b | 1165 | .write_iter = gfs2_file_write_iter, |
c97bfe43 WC |
1166 | .unlocked_ioctl = gfs2_ioctl, |
1167 | .mmap = gfs2_mmap, | |
1168 | .open = gfs2_open, | |
df3fd117 | 1169 | .release = gfs2_release, |
c97bfe43 | 1170 | .fsync = gfs2_fsync, |
61152675 | 1171 | .splice_read = gfs2_file_splice_read, |
f42a69fa | 1172 | .splice_write = gfs2_file_splice_write, |
f057f6cd | 1173 | .setlease = generic_setlease, |
2fe17c10 | 1174 | .fallocate = gfs2_fallocate, |
c97bfe43 WC |
1175 | }; |
1176 | ||
10d21988 | 1177 | const struct file_operations gfs2_dir_fops_nolock = { |
1d1bb236 | 1178 | .iterate_shared = gfs2_readdir, |
c97bfe43 WC |
1179 | .unlocked_ioctl = gfs2_ioctl, |
1180 | .open = gfs2_open, | |
df3fd117 | 1181 | .release = gfs2_release, |
c97bfe43 | 1182 | .fsync = gfs2_fsync, |
6038f373 | 1183 | .llseek = default_llseek, |
c97bfe43 WC |
1184 | }; |
1185 |