blob: 2dcbd56988843ccdfb017c9b14ca8f5b05a31b62 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/ext2/inode.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/inode.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * Goal-directed block allocation by Stephen Tweedie
16 * (sct@dcs.ed.ac.uk), 1993, 1998
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 * 64-bit file support on 64-bit platforms by Jakub Jelinek
20 * (jj@sunsite.ms.mff.cuni.cz)
21 *
22 * Assorted race fixes, rewrite of ext2_get_block() by Al Viro, 2000
23 */
24
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/time.h>
26#include <linux/highuid.h>
27#include <linux/pagemap.h>
Matthew Wilcoxc94c2ac2015-09-08 14:58:40 -070028#include <linux/dax.h>
Matthew Wilcox3dc29162016-03-15 11:20:41 -060029#include <linux/blkdev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/quotaops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/writeback.h>
32#include <linux/buffer_head.h>
33#include <linux/mpage.h>
Josef Bacik68c9d702008-10-03 17:32:43 -040034#include <linux/fiemap.h>
Christoph Hellwig25f4e702016-09-19 11:30:29 +100035#include <linux/iomap.h>
Duane Griffin8d6d0c42008-12-19 20:47:13 +000036#include <linux/namei.h>
Christoph Hellwige2e40f22015-02-22 08:58:50 -080037#include <linux/uio.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include "ext2.h"
39#include "acl.h"
Jan Karac288d292013-03-13 12:57:08 +010040#include "xattr.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Christoph Hellwiga9185b42010-03-05 09:21:37 +010042static int __ext2_write_inode(struct inode *inode, int do_sync);
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044/*
45 * Test whether an inode is a fast symlink.
46 */
47static inline int ext2_inode_is_fast_symlink(struct inode *inode)
48{
49 int ea_blocks = EXT2_I(inode)->i_file_acl ?
50 (inode->i_sb->s_blocksize >> 9) : 0;
51
52 return (S_ISLNK(inode->i_mode) &&
53 inode->i_blocks - ea_blocks == 0);
54}
55
npiggin@suse.de737f2e92010-05-27 01:05:37 +100056static void ext2_truncate_blocks(struct inode *inode, loff_t offset);
57
58static void ext2_write_failed(struct address_space *mapping, loff_t to)
59{
60 struct inode *inode = mapping->host;
61
62 if (to > inode->i_size) {
Kirill A. Shutemov7caef262013-09-12 15:13:56 -070063 truncate_pagecache(inode, inode->i_size);
npiggin@suse.de737f2e92010-05-27 01:05:37 +100064 ext2_truncate_blocks(inode, inode->i_size);
65 }
66}
67
Linus Torvalds1da177e2005-04-16 15:20:36 -070068/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 * Called at the last iput() if i_nlink is zero.
70 */
Al Viro72edc4d2010-06-04 23:32:28 -040071void ext2_evict_inode(struct inode * inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072{
Al Viro72edc4d2010-06-04 23:32:28 -040073 struct ext2_block_alloc_info *rsv;
74 int want_delete = 0;
75
76 if (!inode->i_nlink && !is_bad_inode(inode)) {
77 want_delete = 1;
Christoph Hellwig871a2932010-03-03 09:05:07 -050078 dquot_initialize(inode);
Al Viro72edc4d2010-06-04 23:32:28 -040079 } else {
80 dquot_drop(inode);
81 }
82
Johannes Weiner91b0abe2014-04-03 14:47:49 -070083 truncate_inode_pages_final(&inode->i_data);
Mark Fashehfef26652005-09-09 13:01:31 -070084
Al Viro72edc4d2010-06-04 23:32:28 -040085 if (want_delete) {
Jan Kara1e8b2122012-06-12 16:20:46 +020086 sb_start_intwrite(inode->i_sb);
Al Viro72edc4d2010-06-04 23:32:28 -040087 /* set dtime */
88 EXT2_I(inode)->i_dtime = get_seconds();
89 mark_inode_dirty(inode);
90 __ext2_write_inode(inode, inode_needs_sync(inode));
91 /* truncate to 0 */
92 inode->i_size = 0;
93 if (inode->i_blocks)
94 ext2_truncate_blocks(inode, 0);
Jan Karac288d292013-03-13 12:57:08 +010095 ext2_xattr_delete_inode(inode);
Al Viro72edc4d2010-06-04 23:32:28 -040096 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
Al Viro72edc4d2010-06-04 23:32:28 -040098 invalidate_inode_buffers(inode);
Jan Karadbd57682012-05-03 14:48:02 +020099 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100
Al Viro72edc4d2010-06-04 23:32:28 -0400101 ext2_discard_reservation(inode);
102 rsv = EXT2_I(inode)->i_block_alloc_info;
103 EXT2_I(inode)->i_block_alloc_info = NULL;
104 if (unlikely(rsv))
105 kfree(rsv);
106
Jan Kara1e8b2122012-06-12 16:20:46 +0200107 if (want_delete) {
Al Viro72edc4d2010-06-04 23:32:28 -0400108 ext2_free_inode(inode);
Jan Kara1e8b2122012-06-12 16:20:46 +0200109 sb_end_intwrite(inode->i_sb);
110 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111}
112
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113typedef struct {
114 __le32 *p;
115 __le32 key;
116 struct buffer_head *bh;
117} Indirect;
118
119static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
120{
121 p->key = *(p->p = v);
122 p->bh = bh;
123}
124
125static inline int verify_chain(Indirect *from, Indirect *to)
126{
127 while (from <= to && from->key == *from->p)
128 from++;
129 return (from > to);
130}
131
132/**
133 * ext2_block_to_path - parse the block number into array of offsets
134 * @inode: inode in question (we are only interested in its superblock)
135 * @i_block: block number to be parsed
136 * @offsets: array to store the offsets in
137 * @boundary: set this non-zero if the referred-to block is likely to be
138 * followed (on disk) by an indirect block.
139 * To store the locations of file's data ext2 uses a data structure common
140 * for UNIX filesystems - tree of pointers anchored in the inode, with
141 * data blocks at leaves and indirect blocks in intermediate nodes.
142 * This function translates the block number into path in that tree -
143 * return value is the path length and @offsets[n] is the offset of
144 * pointer to (n+1)th node in the nth one. If @block is out of range
145 * (negative or too large) warning is printed and zero returned.
146 *
147 * Note: function doesn't find node addresses, so no IO is needed. All
148 * we need to know is the capacity of indirect blocks (taken from the
149 * inode->i_sb).
150 */
151
152/*
153 * Portability note: the last comparison (check that we fit into triple
154 * indirect block) is spelled differently, because otherwise on an
155 * architecture with 32-bit longs and 8Kb pages we might get into trouble
156 * if our filesystem had 8Kb blocks. We might use long long, but that would
157 * kill us on x86. Oh, well, at least the sign propagation does not matter -
158 * i_block would have to be negative in the very beginning, so we would not
159 * get there at all.
160 */
161
162static int ext2_block_to_path(struct inode *inode,
163 long i_block, int offsets[4], int *boundary)
164{
165 int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb);
166 int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb);
167 const long direct_blocks = EXT2_NDIR_BLOCKS,
168 indirect_blocks = ptrs,
169 double_blocks = (1 << (ptrs_bits * 2));
170 int n = 0;
171 int final = 0;
172
173 if (i_block < 0) {
Alexey Fisher2314b072009-11-19 19:12:51 +0100174 ext2_msg(inode->i_sb, KERN_WARNING,
175 "warning: %s: block < 0", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 } else if (i_block < direct_blocks) {
177 offsets[n++] = i_block;
178 final = direct_blocks;
179 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
180 offsets[n++] = EXT2_IND_BLOCK;
181 offsets[n++] = i_block;
182 final = ptrs;
183 } else if ((i_block -= indirect_blocks) < double_blocks) {
184 offsets[n++] = EXT2_DIND_BLOCK;
185 offsets[n++] = i_block >> ptrs_bits;
186 offsets[n++] = i_block & (ptrs - 1);
187 final = ptrs;
188 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
189 offsets[n++] = EXT2_TIND_BLOCK;
190 offsets[n++] = i_block >> (ptrs_bits * 2);
191 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
192 offsets[n++] = i_block & (ptrs - 1);
193 final = ptrs;
194 } else {
Alexey Fisher2314b072009-11-19 19:12:51 +0100195 ext2_msg(inode->i_sb, KERN_WARNING,
196 "warning: %s: block is too big", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 }
198 if (boundary)
Martin J. Bligha686cd82007-10-16 23:30:46 -0700199 *boundary = final - 1 - (i_block & (ptrs - 1));
200
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 return n;
202}
203
204/**
205 * ext2_get_branch - read the chain of indirect blocks leading to data
206 * @inode: inode in question
207 * @depth: depth of the chain (1 - direct pointer, etc.)
208 * @offsets: offsets of pointers in inode/indirect blocks
209 * @chain: place to store the result
210 * @err: here we store the error value
211 *
212 * Function fills the array of triples <key, p, bh> and returns %NULL
213 * if everything went OK or the pointer to the last filled triple
214 * (incomplete one) otherwise. Upon the return chain[i].key contains
215 * the number of (i+1)-th block in the chain (as it is stored in memory,
216 * i.e. little-endian 32-bit), chain[i].p contains the address of that
217 * number (it points into struct inode for i==0 and into the bh->b_data
218 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
219 * block for i>0 and NULL for i==0. In other words, it holds the block
220 * numbers of the chain, addresses they were taken from (and where we can
221 * verify that chain did not change) and buffer_heads hosting these
222 * numbers.
223 *
224 * Function stops when it stumbles upon zero pointer (absent block)
225 * (pointer to last triple returned, *@err == 0)
226 * or when it gets an IO error reading an indirect block
227 * (ditto, *@err == -EIO)
228 * or when it notices that chain had been changed while it was reading
229 * (ditto, *@err == -EAGAIN)
230 * or when it reads all @depth-1 indirect blocks successfully and finds
231 * the whole chain, all way to the data (returns %NULL, *err == 0).
232 */
233static Indirect *ext2_get_branch(struct inode *inode,
234 int depth,
235 int *offsets,
236 Indirect chain[4],
237 int *err)
238{
239 struct super_block *sb = inode->i_sb;
240 Indirect *p = chain;
241 struct buffer_head *bh;
242
243 *err = 0;
244 /* i_data is not going away, no lock needed */
245 add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets);
246 if (!p->key)
247 goto no_block;
248 while (--depth) {
249 bh = sb_bread(sb, le32_to_cpu(p->key));
250 if (!bh)
251 goto failure;
252 read_lock(&EXT2_I(inode)->i_meta_lock);
253 if (!verify_chain(chain, p))
254 goto changed;
255 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
256 read_unlock(&EXT2_I(inode)->i_meta_lock);
257 if (!p->key)
258 goto no_block;
259 }
260 return NULL;
261
262changed:
263 read_unlock(&EXT2_I(inode)->i_meta_lock);
264 brelse(bh);
265 *err = -EAGAIN;
266 goto no_block;
267failure:
268 *err = -EIO;
269no_block:
270 return p;
271}
272
273/**
274 * ext2_find_near - find a place for allocation with sufficient locality
275 * @inode: owner
276 * @ind: descriptor of indirect block.
277 *
Benoit Boissinot1cc8dcf2008-04-21 22:45:55 +0000278 * This function returns the preferred place for block allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 * It is used when heuristic for sequential allocation fails.
280 * Rules are:
281 * + if there is a block to the left of our position - allocate near it.
282 * + if pointer will live in indirect block - allocate near that block.
283 * + if pointer will live in inode - allocate in the same cylinder group.
284 *
285 * In the latter case we colour the starting block by the callers PID to
286 * prevent it from clashing with concurrent allocations for a different inode
287 * in the same block group. The PID is used here so that functionally related
288 * files will be close-by on-disk.
289 *
290 * Caller must make sure that @ind is valid and will stay that way.
291 */
292
Akinobu Mita4c8b3122008-04-28 02:16:02 -0700293static ext2_fsblk_t ext2_find_near(struct inode *inode, Indirect *ind)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294{
295 struct ext2_inode_info *ei = EXT2_I(inode);
296 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
297 __le32 *p;
Akinobu Mita4c8b3122008-04-28 02:16:02 -0700298 ext2_fsblk_t bg_start;
299 ext2_fsblk_t colour;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300
301 /* Try to find previous block */
302 for (p = ind->p - 1; p >= start; p--)
303 if (*p)
304 return le32_to_cpu(*p);
305
306 /* No such thing, so let's try location of indirect block */
307 if (ind->bh)
308 return ind->bh->b_blocknr;
309
310 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300311 * It is going to be referred from inode itself? OK, just put it into
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 * the same cylinder group then.
313 */
Akinobu Mita24097d12008-04-28 02:16:01 -0700314 bg_start = ext2_group_first_block_no(inode->i_sb, ei->i_block_group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 colour = (current->pid % 16) *
316 (EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16);
317 return bg_start + colour;
318}
319
320/**
Benoit Boissinot1cc8dcf2008-04-21 22:45:55 +0000321 * ext2_find_goal - find a preferred place for allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 * @inode: owner
323 * @block: block we want
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 * @partial: pointer to the last triple within a chain
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 *
Martin J. Bligha686cd82007-10-16 23:30:46 -0700326 * Returns preferred place for a block (the goal).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 */
328
Akinobu Mita4c8b3122008-04-28 02:16:02 -0700329static inline ext2_fsblk_t ext2_find_goal(struct inode *inode, long block,
330 Indirect *partial)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331{
Martin J. Bligha686cd82007-10-16 23:30:46 -0700332 struct ext2_block_alloc_info *block_i;
333
334 block_i = EXT2_I(inode)->i_block_alloc_info;
335
336 /*
337 * try the heuristic for sequential allocation,
338 * failing that at least try to get decent locality.
339 */
340 if (block_i && (block == block_i->last_alloc_logical_block + 1)
341 && (block_i->last_alloc_physical_block != 0)) {
342 return block_i->last_alloc_physical_block + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 }
Martin J. Bligha686cd82007-10-16 23:30:46 -0700344
345 return ext2_find_near(inode, partial);
346}
347
348/**
349 * ext2_blks_to_allocate: Look up the block map and count the number
350 * of direct blocks need to be allocated for the given branch.
351 *
352 * @branch: chain of indirect blocks
353 * @k: number of blocks need for indirect blocks
354 * @blks: number of data blocks to be mapped.
355 * @blocks_to_boundary: the offset in the indirect block
356 *
357 * return the total number of blocks to be allocate, including the
358 * direct and indirect blocks.
359 */
360static int
361ext2_blks_to_allocate(Indirect * branch, int k, unsigned long blks,
362 int blocks_to_boundary)
363{
364 unsigned long count = 0;
365
366 /*
367 * Simple case, [t,d]Indirect block(s) has not allocated yet
368 * then it's clear blocks on that path have not allocated
369 */
370 if (k > 0) {
371 /* right now don't hanel cross boundary allocation */
372 if (blks < blocks_to_boundary + 1)
373 count += blks;
374 else
375 count += blocks_to_boundary + 1;
376 return count;
377 }
378
379 count++;
380 while (count < blks && count <= blocks_to_boundary
381 && le32_to_cpu(*(branch[0].p + count)) == 0) {
382 count++;
383 }
384 return count;
385}
386
387/**
388 * ext2_alloc_blocks: multiple allocate blocks needed for a branch
389 * @indirect_blks: the number of blocks need to allocate for indirect
390 * blocks
391 *
392 * @new_blocks: on return it will store the new block numbers for
393 * the indirect blocks(if needed) and the first direct block,
394 * @blks: on return it will store the total number of allocated
395 * direct blocks
396 */
397static int ext2_alloc_blocks(struct inode *inode,
398 ext2_fsblk_t goal, int indirect_blks, int blks,
399 ext2_fsblk_t new_blocks[4], int *err)
400{
401 int target, i;
402 unsigned long count = 0;
403 int index = 0;
404 ext2_fsblk_t current_block = 0;
405 int ret = 0;
406
407 /*
408 * Here we try to allocate the requested multiple blocks at once,
409 * on a best-effort basis.
410 * To build a branch, we should allocate blocks for
411 * the indirect blocks(if not allocated yet), and at least
412 * the first direct block of this branch. That's the
413 * minimum number of blocks need to allocate(required)
414 */
415 target = blks + indirect_blks;
416
417 while (1) {
418 count = target;
419 /* allocating blocks for indirect blocks and direct blocks */
420 current_block = ext2_new_blocks(inode,goal,&count,err);
421 if (*err)
422 goto failed_out;
423
424 target -= count;
425 /* allocate blocks for indirect blocks */
426 while (index < indirect_blks && count) {
427 new_blocks[index++] = current_block++;
428 count--;
429 }
430
431 if (count > 0)
432 break;
433 }
434
435 /* save the new block number for the first direct block */
436 new_blocks[index] = current_block;
437
438 /* total number of blocks allocated for direct blocks */
439 ret = count;
440 *err = 0;
441 return ret;
442failed_out:
443 for (i = 0; i <index; i++)
444 ext2_free_blocks(inode, new_blocks[i], 1);
Al Viroaddacc72010-07-22 01:19:42 +0400445 if (index)
446 mark_inode_dirty(inode);
Martin J. Bligha686cd82007-10-16 23:30:46 -0700447 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448}
449
450/**
451 * ext2_alloc_branch - allocate and set up a chain of blocks.
452 * @inode: owner
453 * @num: depth of the chain (number of blocks to allocate)
454 * @offsets: offsets (in the blocks) to store the pointers to next.
455 * @branch: place to store the chain in.
456 *
457 * This function allocates @num blocks, zeroes out all but the last one,
458 * links them into chain and (if we are synchronous) writes them to disk.
459 * In other words, it prepares a branch that can be spliced onto the
460 * inode. It stores the information about that chain in the branch[], in
461 * the same format as ext2_get_branch() would do. We are calling it after
462 * we had read the existing part of chain and partial points to the last
463 * triple of that (one with zero ->key). Upon the exit we have the same
Namhyung Kim72b43572010-09-15 21:46:02 +0900464 * picture as after the successful ext2_get_block(), except that in one
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 * place chain is disconnected - *branch->p is still zero (we did not
466 * set the last link), but branch->key contains the number that should
467 * be placed into *branch->p to fill that gap.
468 *
469 * If allocation fails we free all blocks we've allocated (and forget
470 * their buffer_heads) and return the error value the from failed
471 * ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain
472 * as described above and return 0.
473 */
474
475static int ext2_alloc_branch(struct inode *inode,
Martin J. Bligha686cd82007-10-16 23:30:46 -0700476 int indirect_blks, int *blks, ext2_fsblk_t goal,
477 int *offsets, Indirect *branch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478{
479 int blocksize = inode->i_sb->s_blocksize;
Martin J. Bligha686cd82007-10-16 23:30:46 -0700480 int i, n = 0;
481 int err = 0;
482 struct buffer_head *bh;
483 int num;
484 ext2_fsblk_t new_blocks[4];
485 ext2_fsblk_t current_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
Martin J. Bligha686cd82007-10-16 23:30:46 -0700487 num = ext2_alloc_blocks(inode, goal, indirect_blks,
488 *blks, new_blocks, &err);
489 if (err)
490 return err;
491
492 branch[0].key = cpu_to_le32(new_blocks[0]);
493 /*
494 * metadata blocks and data blocks are allocated.
495 */
496 for (n = 1; n <= indirect_blks; n++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 /*
Martin J. Bligha686cd82007-10-16 23:30:46 -0700498 * Get buffer_head for parent block, zero it out
499 * and set the pointer to new one, then send
500 * parent to disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 */
Martin J. Bligha686cd82007-10-16 23:30:46 -0700502 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
Wang Shilong8d8759e2013-01-12 01:19:32 -0800503 if (unlikely(!bh)) {
504 err = -ENOMEM;
505 goto failed;
506 }
Martin J. Bligha686cd82007-10-16 23:30:46 -0700507 branch[n].bh = bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 lock_buffer(bh);
509 memset(bh->b_data, 0, blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 branch[n].p = (__le32 *) bh->b_data + offsets[n];
Martin J. Bligha686cd82007-10-16 23:30:46 -0700511 branch[n].key = cpu_to_le32(new_blocks[n]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 *branch[n].p = branch[n].key;
Martin J. Bligha686cd82007-10-16 23:30:46 -0700513 if ( n == indirect_blks) {
514 current_block = new_blocks[n];
515 /*
516 * End of chain, update the last new metablock of
517 * the chain to point to the new allocated
518 * data blocks numbers
519 */
520 for (i=1; i < num; i++)
521 *(branch[n].p + i) = cpu_to_le32(++current_block);
522 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 set_buffer_uptodate(bh);
524 unlock_buffer(bh);
525 mark_buffer_dirty_inode(bh, inode);
526 /* We used to sync bh here if IS_SYNC(inode).
Jan Karaa2a735a2009-08-18 17:54:11 +0200527 * But we now rely upon generic_write_sync()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 * and b_inode_buffers. But not for directories.
529 */
530 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
531 sync_dirty_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 }
Martin J. Bligha686cd82007-10-16 23:30:46 -0700533 *blks = num;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 return err;
Wang Shilong8d8759e2013-01-12 01:19:32 -0800535
536failed:
537 for (i = 1; i < n; i++)
538 bforget(branch[i].bh);
539 for (i = 0; i < indirect_blks; i++)
540 ext2_free_blocks(inode, new_blocks[i], 1);
541 ext2_free_blocks(inode, new_blocks[i], num);
542 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543}
544
545/**
Martin J. Bligha686cd82007-10-16 23:30:46 -0700546 * ext2_splice_branch - splice the allocated branch onto inode.
547 * @inode: owner
548 * @block: (logical) number of block we are adding
Martin J. Bligha686cd82007-10-16 23:30:46 -0700549 * @where: location of missing link
550 * @num: number of indirect blocks we are adding
551 * @blks: number of direct blocks we are adding
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 *
Martin J. Bligha686cd82007-10-16 23:30:46 -0700553 * This function fills the missing link and does all housekeeping needed in
554 * inode (->i_blocks, etc.). In case of success we end up with the full
555 * chain to new block and return 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 */
Martin J. Bligha686cd82007-10-16 23:30:46 -0700557static void ext2_splice_branch(struct inode *inode,
558 long block, Indirect *where, int num, int blks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 int i;
Martin J. Bligha686cd82007-10-16 23:30:46 -0700561 struct ext2_block_alloc_info *block_i;
562 ext2_fsblk_t current_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
Martin J. Bligha686cd82007-10-16 23:30:46 -0700564 block_i = EXT2_I(inode)->i_block_alloc_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565
Martin J. Bligha686cd82007-10-16 23:30:46 -0700566 /* XXX LOCKING probably should have i_meta_lock ?*/
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 /* That's it */
568
569 *where->p = where->key;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
Martin J. Bligha686cd82007-10-16 23:30:46 -0700571 /*
572 * Update the host buffer_head or inode to point to more just allocated
573 * direct blocks blocks
574 */
575 if (num == 0 && blks > 1) {
576 current_block = le32_to_cpu(where->key) + 1;
577 for (i = 1; i < blks; i++)
578 *(where->p + i ) = cpu_to_le32(current_block++);
579 }
580
581 /*
582 * update the most recently allocated logical & physical block
583 * in i_block_alloc_info, to assist find the proper goal block for next
584 * allocation
585 */
586 if (block_i) {
587 block_i->last_alloc_logical_block = block + blks - 1;
588 block_i->last_alloc_physical_block =
589 le32_to_cpu(where[num].key) + blks - 1;
590 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
592 /* We are done with atomic stuff, now do the rest of housekeeping */
593
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 /* had we spliced it onto indirect block? */
595 if (where->bh)
596 mark_buffer_dirty_inode(where->bh, inode);
597
Deepa Dinamani02027d42016-09-14 07:48:05 -0700598 inode->i_ctime = current_time(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 mark_inode_dirty(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600}
601
602/*
603 * Allocation strategy is simple: if we have to allocate something, we will
604 * have to go the whole way to leaf. So let's do it before attaching anything
605 * to tree, set linkage between the newborn blocks, write them if sync is
606 * required, recheck the path, free and repeat if check fails, otherwise
607 * set the last missing link (that will protect us from any truncate-generated
608 * removals - all blocks on the path are immune now) and possibly force the
609 * write on the parent block.
610 * That has a nice additional property: no special recovery from the failed
611 * allocations is needed - we simply release blocks and do not touch anything
612 * reachable from inode.
Martin J. Bligha686cd82007-10-16 23:30:46 -0700613 *
614 * `handle' can be NULL if create == 0.
615 *
Martin J. Bligha686cd82007-10-16 23:30:46 -0700616 * return > 0, # of blocks mapped or allocated.
617 * return = 0, if plain lookup failed.
618 * return < 0, error case.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 */
Martin J. Bligha686cd82007-10-16 23:30:46 -0700620static int ext2_get_blocks(struct inode *inode,
621 sector_t iblock, unsigned long maxblocks,
Christoph Hellwig6750ad72016-09-19 11:28:39 +1000622 u32 *bno, bool *new, bool *boundary,
Martin J. Bligha686cd82007-10-16 23:30:46 -0700623 int create)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624{
Arnd Bergmanne9528132016-10-18 00:05:34 +0200625 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 int offsets[4];
627 Indirect chain[4];
628 Indirect *partial;
Martin J. Bligha686cd82007-10-16 23:30:46 -0700629 ext2_fsblk_t goal;
630 int indirect_blks;
631 int blocks_to_boundary = 0;
632 int depth;
633 struct ext2_inode_info *ei = EXT2_I(inode);
634 int count = 0;
635 ext2_fsblk_t first_block = 0;
636
Jan Kara7ba3ec52013-11-05 01:15:38 +0100637 BUG_ON(maxblocks == 0);
638
Martin J. Bligha686cd82007-10-16 23:30:46 -0700639 depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640
641 if (depth == 0)
Arnd Bergmanne9528132016-10-18 00:05:34 +0200642 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643
Jan Kara316cb4ef32009-04-13 14:40:14 -0700644 partial = ext2_get_branch(inode, depth, offsets, chain, &err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 /* Simplest case - block found, no allocation needed */
646 if (!partial) {
Martin J. Bligha686cd82007-10-16 23:30:46 -0700647 first_block = le32_to_cpu(chain[depth - 1].key);
Martin J. Bligha686cd82007-10-16 23:30:46 -0700648 count++;
649 /*map more blocks*/
650 while (count < maxblocks && count <= blocks_to_boundary) {
651 ext2_fsblk_t blk;
652
Jan Kara316cb4ef32009-04-13 14:40:14 -0700653 if (!verify_chain(chain, chain + depth - 1)) {
Martin J. Bligha686cd82007-10-16 23:30:46 -0700654 /*
655 * Indirect block might be removed by
656 * truncate while we were reading it.
657 * Handling of that case: forget what we've
658 * got now, go to reread.
659 */
Jan Kara316cb4ef32009-04-13 14:40:14 -0700660 err = -EAGAIN;
Martin J. Bligha686cd82007-10-16 23:30:46 -0700661 count = 0;
Jan Kara316cb4ef32009-04-13 14:40:14 -0700662 break;
Martin J. Bligha686cd82007-10-16 23:30:46 -0700663 }
664 blk = le32_to_cpu(*(chain[depth-1].p + count));
665 if (blk == first_block + count)
666 count++;
667 else
668 break;
669 }
Jan Kara316cb4ef32009-04-13 14:40:14 -0700670 if (err != -EAGAIN)
671 goto got_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 }
673
674 /* Next simple case - plain lookup or failed read of indirect block */
Martin J. Bligha686cd82007-10-16 23:30:46 -0700675 if (!create || err == -EIO)
676 goto cleanup;
677
678 mutex_lock(&ei->truncate_mutex);
Jan Kara316cb4ef32009-04-13 14:40:14 -0700679 /*
680 * If the indirect block is missing while we are reading
Namhyung Kim72b43572010-09-15 21:46:02 +0900681 * the chain(ext2_get_branch() returns -EAGAIN err), or
Jan Kara316cb4ef32009-04-13 14:40:14 -0700682 * if the chain has been changed after we grab the semaphore,
683 * (either because another process truncated this branch, or
684 * another get_block allocated this branch) re-grab the chain to see if
685 * the request block has been allocated or not.
686 *
687 * Since we already block the truncate/other get_block
688 * at this point, we will have the current copy of the chain when we
689 * splice the branch into the tree.
690 */
691 if (err == -EAGAIN || !verify_chain(chain, partial)) {
692 while (partial > chain) {
693 brelse(partial->bh);
694 partial--;
695 }
696 partial = ext2_get_branch(inode, depth, offsets, chain, &err);
697 if (!partial) {
698 count++;
699 mutex_unlock(&ei->truncate_mutex);
700 if (err)
701 goto cleanup;
Jan Kara316cb4ef32009-04-13 14:40:14 -0700702 goto got_it;
703 }
704 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705
706 /*
Martin J. Bligha686cd82007-10-16 23:30:46 -0700707 * Okay, we need to do block allocation. Lazily initialize the block
708 * allocation info here if necessary
709 */
710 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
711 ext2_init_block_alloc_info(inode);
712
Akinobu Mitafb01bfd2008-02-06 01:40:16 -0800713 goal = ext2_find_goal(inode, iblock, partial);
Martin J. Bligha686cd82007-10-16 23:30:46 -0700714
715 /* the number of blocks need to allocate for [d,t]indirect blocks */
716 indirect_blks = (chain + depth) - partial - 1;
717 /*
718 * Next look up the indirect map to count the totoal number of
719 * direct blocks to allocate for this branch.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 */
Martin J. Bligha686cd82007-10-16 23:30:46 -0700721 count = ext2_blks_to_allocate(partial, indirect_blks,
722 maxblocks, blocks_to_boundary);
723 /*
724 * XXX ???? Block out ext2_truncate while we alter the tree
725 */
726 err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
727 offsets + (partial - chain), partial);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728
Martin J. Bligha686cd82007-10-16 23:30:46 -0700729 if (err) {
730 mutex_unlock(&ei->truncate_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 goto cleanup;
Martin J. Bligha686cd82007-10-16 23:30:46 -0700732 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -0800734 if (IS_DAX(inode)) {
Jan Kara225c5162016-09-27 16:35:45 +0200735 /*
736 * We must unmap blocks before zeroing so that writeback cannot
737 * overwrite zeros with stale data from block device page cache.
738 */
Jan Kara69a9bea2016-11-04 18:08:14 +0100739 clean_bdev_aliases(inode->i_sb->s_bdev,
740 le32_to_cpu(chain[depth-1].key),
741 count);
Carsten Otte6d791252005-06-23 22:05:26 -0700742 /*
Matthew Wilcox289c6ae2015-02-16 15:58:59 -0800743 * block must be initialised before we put it in the tree
744 * so that it's not found by another thread before it's
745 * initialised
Carsten Otte6d791252005-06-23 22:05:26 -0700746 */
Matthew Wilcox3dc29162016-03-15 11:20:41 -0600747 err = sb_issue_zeroout(inode->i_sb,
748 le32_to_cpu(chain[depth-1].key), count,
749 GFP_NOFS);
Martin J. Bligha686cd82007-10-16 23:30:46 -0700750 if (err) {
751 mutex_unlock(&ei->truncate_mutex);
Carsten Otte6d791252005-06-23 22:05:26 -0700752 goto cleanup;
Martin J. Bligha686cd82007-10-16 23:30:46 -0700753 }
Christoph Hellwig6750ad72016-09-19 11:28:39 +1000754 }
Jan Karae568df62016-08-10 16:42:53 +0200755 *new = true;
Carsten Otte6d791252005-06-23 22:05:26 -0700756
Martin J. Bligha686cd82007-10-16 23:30:46 -0700757 ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
758 mutex_unlock(&ei->truncate_mutex);
Martin J. Bligha686cd82007-10-16 23:30:46 -0700759got_it:
Martin J. Bligha686cd82007-10-16 23:30:46 -0700760 if (count > blocks_to_boundary)
Christoph Hellwig6750ad72016-09-19 11:28:39 +1000761 *boundary = true;
Martin J. Bligha686cd82007-10-16 23:30:46 -0700762 err = count;
763 /* Clean up and exit */
764 partial = chain + depth - 1; /* the whole chain */
765cleanup:
766 while (partial > chain) {
767 brelse(partial->bh);
768 partial--;
769 }
Arnd Bergmanne9528132016-10-18 00:05:34 +0200770 if (err > 0)
771 *bno = le32_to_cpu(chain[depth-1].key);
Martin J. Bligha686cd82007-10-16 23:30:46 -0700772 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773}
774
Christoph Hellwig6750ad72016-09-19 11:28:39 +1000775int ext2_get_block(struct inode *inode, sector_t iblock,
776 struct buffer_head *bh_result, int create)
Martin J. Bligha686cd82007-10-16 23:30:46 -0700777{
778 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
Christoph Hellwig6750ad72016-09-19 11:28:39 +1000779 bool new = false, boundary = false;
780 u32 bno;
781 int ret;
782
783 ret = ext2_get_blocks(inode, iblock, max_blocks, &bno, &new, &boundary,
784 create);
785 if (ret <= 0)
786 return ret;
787
788 map_bh(bh_result, inode->i_sb, bno);
789 bh_result->b_size = (ret << inode->i_blkbits);
790 if (new)
791 set_buffer_new(bh_result);
792 if (boundary)
793 set_buffer_boundary(bh_result);
794 return 0;
Martin J. Bligha686cd82007-10-16 23:30:46 -0700795
796}
797
Christoph Hellwig25f4e702016-09-19 11:30:29 +1000798#ifdef CONFIG_FS_DAX
799static int ext2_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
800 unsigned flags, struct iomap *iomap)
801{
Dan Williamsfa5d9322017-01-27 12:04:59 -0800802 struct block_device *bdev;
Christoph Hellwig25f4e702016-09-19 11:30:29 +1000803 unsigned int blkbits = inode->i_blkbits;
804 unsigned long first_block = offset >> blkbits;
805 unsigned long max_blocks = (length + (1 << blkbits) - 1) >> blkbits;
806 bool new = false, boundary = false;
807 u32 bno;
808 int ret;
809
810 ret = ext2_get_blocks(inode, first_block, max_blocks,
811 &bno, &new, &boundary, flags & IOMAP_WRITE);
812 if (ret < 0)
813 return ret;
814
815 iomap->flags = 0;
Dan Williamsfa5d9322017-01-27 12:04:59 -0800816 bdev = inode->i_sb->s_bdev;
817 iomap->bdev = bdev;
Christoph Hellwigd5bfccd2016-10-03 09:46:04 +1100818 iomap->offset = (u64)first_block << blkbits;
Dan Williamsfa5d9322017-01-27 12:04:59 -0800819 if (blk_queue_dax(bdev->bd_queue))
Dan Williamsf5705aa8c2017-05-13 16:31:05 -0700820 iomap->dax_dev = fs_dax_get_by_host(bdev->bd_disk->disk_name);
Dan Williamsfa5d9322017-01-27 12:04:59 -0800821 else
822 iomap->dax_dev = NULL;
Christoph Hellwig25f4e702016-09-19 11:30:29 +1000823
824 if (ret == 0) {
825 iomap->type = IOMAP_HOLE;
826 iomap->blkno = IOMAP_NULL_BLOCK;
827 iomap->length = 1 << blkbits;
828 } else {
829 iomap->type = IOMAP_MAPPED;
830 iomap->blkno = (sector_t)bno << (blkbits - 9);
831 iomap->length = (u64)ret << blkbits;
832 iomap->flags |= IOMAP_F_MERGED;
833 }
834
835 if (new)
836 iomap->flags |= IOMAP_F_NEW;
837 return 0;
838}
839
840static int
841ext2_iomap_end(struct inode *inode, loff_t offset, loff_t length,
842 ssize_t written, unsigned flags, struct iomap *iomap)
843{
Dan Williamsf5705aa8c2017-05-13 16:31:05 -0700844 fs_put_dax(iomap->dax_dev);
Christoph Hellwig25f4e702016-09-19 11:30:29 +1000845 if (iomap->type == IOMAP_MAPPED &&
846 written < length &&
847 (flags & IOMAP_WRITE))
848 ext2_write_failed(inode->i_mapping, offset + length);
849 return 0;
850}
851
Christoph Hellwig8ff6daa2017-01-27 23:20:26 -0800852const struct iomap_ops ext2_iomap_ops = {
Christoph Hellwig25f4e702016-09-19 11:30:29 +1000853 .iomap_begin = ext2_iomap_begin,
854 .iomap_end = ext2_iomap_end,
855};
Jan Kara00697ee2016-11-20 20:47:07 -0500856#else
857/* Define empty ops for !CONFIG_FS_DAX case to avoid ugly ifdefs */
Christoph Hellwig8ff6daa2017-01-27 23:20:26 -0800858const struct iomap_ops ext2_iomap_ops;
Christoph Hellwig25f4e702016-09-19 11:30:29 +1000859#endif /* CONFIG_FS_DAX */
860
Josef Bacik68c9d702008-10-03 17:32:43 -0400861int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
862 u64 start, u64 len)
863{
864 return generic_block_fiemap(inode, fieinfo, start, len,
865 ext2_get_block);
866}
867
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868static int ext2_writepage(struct page *page, struct writeback_control *wbc)
869{
870 return block_write_full_page(page, ext2_get_block, wbc);
871}
872
873static int ext2_readpage(struct file *file, struct page *page)
874{
875 return mpage_readpage(page, ext2_get_block);
876}
877
878static int
879ext2_readpages(struct file *file, struct address_space *mapping,
880 struct list_head *pages, unsigned nr_pages)
881{
882 return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
883}
884
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885static int
Nick Pigginf34fb6e2007-10-16 01:25:04 -0700886ext2_write_begin(struct file *file, struct address_space *mapping,
887 loff_t pos, unsigned len, unsigned flags,
888 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889{
npiggin@suse.de737f2e92010-05-27 01:05:37 +1000890 int ret;
891
Christoph Hellwig155130a2010-06-04 11:29:58 +0200892 ret = block_write_begin(mapping, pos, len, flags, pagep,
893 ext2_get_block);
npiggin@suse.de737f2e92010-05-27 01:05:37 +1000894 if (ret < 0)
895 ext2_write_failed(mapping, pos + len);
896 return ret;
897}
898
899static int ext2_write_end(struct file *file, struct address_space *mapping,
900 loff_t pos, unsigned len, unsigned copied,
901 struct page *page, void *fsdata)
902{
903 int ret;
904
905 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
906 if (ret < len)
907 ext2_write_failed(mapping, pos + len);
908 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909}
910
Nick Piggin03158cd2007-10-16 01:25:25 -0700911static int
912ext2_nobh_write_begin(struct file *file, struct address_space *mapping,
913 loff_t pos, unsigned len, unsigned flags,
914 struct page **pagep, void **fsdata)
915{
npiggin@suse.de737f2e92010-05-27 01:05:37 +1000916 int ret;
917
Christoph Hellwigea0f04e2010-06-04 11:29:54 +0200918 ret = nobh_write_begin(mapping, pos, len, flags, pagep, fsdata,
919 ext2_get_block);
npiggin@suse.de737f2e92010-05-27 01:05:37 +1000920 if (ret < 0)
921 ext2_write_failed(mapping, pos + len);
922 return ret;
Nick Piggin03158cd2007-10-16 01:25:25 -0700923}
924
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925static int ext2_nobh_writepage(struct page *page,
926 struct writeback_control *wbc)
927{
928 return nobh_writepage(page, ext2_get_block, wbc);
929}
930
931static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
932{
933 return generic_block_bmap(mapping,block,ext2_get_block);
934}
935
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936static ssize_t
Christoph Hellwigc8b8e322016-04-07 08:51:58 -0700937ext2_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938{
939 struct file *file = iocb->ki_filp;
npiggin@suse.de737f2e92010-05-27 01:05:37 +1000940 struct address_space *mapping = file->f_mapping;
941 struct inode *inode = mapping->host;
Al Viroa6cbcd42014-03-04 22:38:00 -0500942 size_t count = iov_iter_count(iter);
Christoph Hellwigc8b8e322016-04-07 08:51:58 -0700943 loff_t offset = iocb->ki_pos;
npiggin@suse.de737f2e92010-05-27 01:05:37 +1000944 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945
Christoph Hellwig25f4e702016-09-19 11:30:29 +1000946 if (WARN_ON_ONCE(IS_DAX(inode)))
947 return -EIO;
948
949 ret = blockdev_direct_IO(iocb, inode, iter, ext2_get_block);
Omar Sandoval6f673762015-03-16 04:33:52 -0700950 if (ret < 0 && iov_iter_rw(iter) == WRITE)
Al Viroa6cbcd42014-03-04 22:38:00 -0500951 ext2_write_failed(mapping, offset + count);
npiggin@suse.de737f2e92010-05-27 01:05:37 +1000952 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953}
954
955static int
956ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
957{
Ross Zwisler7f6d5b52016-02-26 15:19:55 -0800958#ifdef CONFIG_FS_DAX
959 if (dax_mapping(mapping)) {
960 return dax_writeback_mapping_range(mapping,
961 mapping->host->i_sb->s_bdev,
962 wbc);
963 }
964#endif
965
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 return mpage_writepages(mapping, wbc, ext2_get_block);
967}
968
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700969const struct address_space_operations ext2_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 .readpage = ext2_readpage,
971 .readpages = ext2_readpages,
972 .writepage = ext2_writepage,
Nick Pigginf34fb6e2007-10-16 01:25:04 -0700973 .write_begin = ext2_write_begin,
npiggin@suse.de737f2e92010-05-27 01:05:37 +1000974 .write_end = ext2_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 .bmap = ext2_bmap,
976 .direct_IO = ext2_direct_IO,
977 .writepages = ext2_writepages,
Christoph Lametere965f962006-02-01 03:05:41 -0800978 .migratepage = buffer_migrate_page,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -0700979 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +0200980 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981};
982
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700983const struct address_space_operations ext2_nobh_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 .readpage = ext2_readpage,
985 .readpages = ext2_readpages,
986 .writepage = ext2_nobh_writepage,
Nick Piggin03158cd2007-10-16 01:25:25 -0700987 .write_begin = ext2_nobh_write_begin,
988 .write_end = nobh_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 .bmap = ext2_bmap,
990 .direct_IO = ext2_direct_IO,
991 .writepages = ext2_writepages,
Christoph Lametere965f962006-02-01 03:05:41 -0800992 .migratepage = buffer_migrate_page,
Andi Kleenaa261f52009-09-16 11:50:16 +0200993 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994};
995
996/*
997 * Probably it should be a library function... search for first non-zero word
998 * or memcmp with zero_page, whatever is better for particular architecture.
999 * Linus?
1000 */
1001static inline int all_zeroes(__le32 *p, __le32 *q)
1002{
1003 while (p < q)
1004 if (*p++)
1005 return 0;
1006 return 1;
1007}
1008
1009/**
1010 * ext2_find_shared - find the indirect blocks for partial truncation.
1011 * @inode: inode in question
1012 * @depth: depth of the affected branch
1013 * @offsets: offsets of pointers in that branch (see ext2_block_to_path)
1014 * @chain: place to store the pointers to partial indirect blocks
1015 * @top: place to the (detached) top of branch
1016 *
1017 * This is a helper function used by ext2_truncate().
1018 *
1019 * When we do truncate() we may have to clean the ends of several indirect
1020 * blocks but leave the blocks themselves alive. Block is partially
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001021 * truncated if some data below the new i_size is referred from it (and
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 * it is on the path to the first completely truncated data block, indeed).
1023 * We have to free the top of that path along with everything to the right
1024 * of the path. Since no allocation past the truncation point is possible
1025 * until ext2_truncate() finishes, we may safely do the latter, but top
1026 * of branch may require special attention - pageout below the truncation
1027 * point might try to populate it.
1028 *
1029 * We atomically detach the top of branch from the tree, store the block
1030 * number of its root in *@top, pointers to buffer_heads of partially
1031 * truncated blocks - in @chain[].bh and pointers to their last elements
1032 * that should not be removed - in @chain[].p. Return value is the pointer
1033 * to last filled element of @chain.
1034 *
1035 * The work left to caller to do the actual freeing of subtrees:
1036 * a) free the subtree starting from *@top
1037 * b) free the subtrees whose roots are stored in
1038 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1039 * c) free the subtrees growing from the inode past the @chain[0].p
1040 * (no partially truncated stuff there).
1041 */
1042
1043static Indirect *ext2_find_shared(struct inode *inode,
1044 int depth,
1045 int offsets[4],
1046 Indirect chain[4],
1047 __le32 *top)
1048{
1049 Indirect *partial, *p;
1050 int k, err;
1051
1052 *top = 0;
1053 for (k = depth; k > 1 && !offsets[k-1]; k--)
1054 ;
1055 partial = ext2_get_branch(inode, k, offsets, chain, &err);
1056 if (!partial)
1057 partial = chain + k-1;
1058 /*
1059 * If the branch acquired continuation since we've looked at it -
1060 * fine, it should all survive and (new) top doesn't belong to us.
1061 */
1062 write_lock(&EXT2_I(inode)->i_meta_lock);
1063 if (!partial->key && *partial->p) {
1064 write_unlock(&EXT2_I(inode)->i_meta_lock);
1065 goto no_top;
1066 }
1067 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
1068 ;
1069 /*
1070 * OK, we've found the last block that must survive. The rest of our
1071 * branch should be detached before unlocking. However, if that rest
1072 * of branch is all ours and does not grow immediately from the inode
1073 * it's easier to cheat and just decrement partial->p.
1074 */
1075 if (p == chain + k - 1 && p > chain) {
1076 p->p--;
1077 } else {
1078 *top = *p->p;
1079 *p->p = 0;
1080 }
1081 write_unlock(&EXT2_I(inode)->i_meta_lock);
1082
1083 while(partial > p)
1084 {
1085 brelse(partial->bh);
1086 partial--;
1087 }
1088no_top:
1089 return partial;
1090}
1091
1092/**
1093 * ext2_free_data - free a list of data blocks
1094 * @inode: inode we are dealing with
1095 * @p: array of block numbers
1096 * @q: points immediately past the end of array
1097 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001098 * We are freeing all blocks referred from that array (numbers are
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 * stored as little-endian 32-bit) and updating @inode->i_blocks
1100 * appropriately.
1101 */
1102static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
1103{
1104 unsigned long block_to_free = 0, count = 0;
1105 unsigned long nr;
1106
1107 for ( ; p < q ; p++) {
1108 nr = le32_to_cpu(*p);
1109 if (nr) {
1110 *p = 0;
1111 /* accumulate blocks to free if they're contiguous */
1112 if (count == 0)
1113 goto free_this;
1114 else if (block_to_free == nr - count)
1115 count++;
1116 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 ext2_free_blocks (inode, block_to_free, count);
Al Viroaddacc72010-07-22 01:19:42 +04001118 mark_inode_dirty(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 free_this:
1120 block_to_free = nr;
1121 count = 1;
1122 }
1123 }
1124 }
1125 if (count > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 ext2_free_blocks (inode, block_to_free, count);
Al Viroaddacc72010-07-22 01:19:42 +04001127 mark_inode_dirty(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 }
1129}
1130
1131/**
1132 * ext2_free_branches - free an array of branches
1133 * @inode: inode we are dealing with
1134 * @p: array of block numbers
1135 * @q: pointer immediately past the end of array
1136 * @depth: depth of the branches to free
1137 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001138 * We are freeing all blocks referred from these branches (numbers are
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 * stored as little-endian 32-bit) and updating @inode->i_blocks
1140 * appropriately.
1141 */
1142static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
1143{
1144 struct buffer_head * bh;
1145 unsigned long nr;
1146
1147 if (depth--) {
1148 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1149 for ( ; p < q ; p++) {
1150 nr = le32_to_cpu(*p);
1151 if (!nr)
1152 continue;
1153 *p = 0;
1154 bh = sb_bread(inode->i_sb, nr);
1155 /*
1156 * A read failure? Report error and clear slot
1157 * (should be rare).
1158 */
1159 if (!bh) {
1160 ext2_error(inode->i_sb, "ext2_free_branches",
1161 "Read failure, inode=%ld, block=%ld",
1162 inode->i_ino, nr);
1163 continue;
1164 }
1165 ext2_free_branches(inode,
1166 (__le32*)bh->b_data,
1167 (__le32*)bh->b_data + addr_per_block,
1168 depth);
1169 bforget(bh);
1170 ext2_free_blocks(inode, nr, 1);
1171 mark_inode_dirty(inode);
1172 }
1173 } else
1174 ext2_free_data(inode, p, q);
1175}
1176
Ross Zwisler5726b272015-10-13 16:25:37 -06001177/* dax_sem must be held when calling this function */
npiggin@suse.de737f2e92010-05-27 01:05:37 +10001178static void __ext2_truncate_blocks(struct inode *inode, loff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179{
1180 __le32 *i_data = EXT2_I(inode)->i_data;
Martin J. Bligha686cd82007-10-16 23:30:46 -07001181 struct ext2_inode_info *ei = EXT2_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1183 int offsets[4];
1184 Indirect chain[4];
1185 Indirect *partial;
1186 __le32 nr = 0;
1187 int n;
1188 long iblock;
1189 unsigned blocksize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 blocksize = inode->i_sb->s_blocksize;
npiggin@suse.de737f2e92010-05-27 01:05:37 +10001191 iblock = (offset + blocksize-1) >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192
Ross Zwisler5726b272015-10-13 16:25:37 -06001193#ifdef CONFIG_FS_DAX
1194 WARN_ON(!rwsem_is_locked(&ei->dax_sem));
1195#endif
1196
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 n = ext2_block_to_path(inode, iblock, offsets, NULL);
1198 if (n == 0)
1199 return;
1200
Martin J. Bligha686cd82007-10-16 23:30:46 -07001201 /*
1202 * From here we block out all ext2_get_block() callers who want to
1203 * modify the block allocation tree.
1204 */
1205 mutex_lock(&ei->truncate_mutex);
1206
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 if (n == 1) {
1208 ext2_free_data(inode, i_data+offsets[0],
1209 i_data + EXT2_NDIR_BLOCKS);
1210 goto do_indirects;
1211 }
1212
1213 partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1214 /* Kill the top of shared branch (already detached) */
1215 if (nr) {
1216 if (partial == chain)
1217 mark_inode_dirty(inode);
1218 else
1219 mark_buffer_dirty_inode(partial->bh, inode);
1220 ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1221 }
1222 /* Clear the ends of indirect blocks on the shared branch */
1223 while (partial > chain) {
1224 ext2_free_branches(inode,
1225 partial->p + 1,
1226 (__le32*)partial->bh->b_data+addr_per_block,
1227 (chain+n-1) - partial);
1228 mark_buffer_dirty_inode(partial->bh, inode);
1229 brelse (partial->bh);
1230 partial--;
1231 }
1232do_indirects:
1233 /* Kill the remaining (whole) subtrees */
1234 switch (offsets[0]) {
1235 default:
1236 nr = i_data[EXT2_IND_BLOCK];
1237 if (nr) {
1238 i_data[EXT2_IND_BLOCK] = 0;
1239 mark_inode_dirty(inode);
1240 ext2_free_branches(inode, &nr, &nr+1, 1);
1241 }
1242 case EXT2_IND_BLOCK:
1243 nr = i_data[EXT2_DIND_BLOCK];
1244 if (nr) {
1245 i_data[EXT2_DIND_BLOCK] = 0;
1246 mark_inode_dirty(inode);
1247 ext2_free_branches(inode, &nr, &nr+1, 2);
1248 }
1249 case EXT2_DIND_BLOCK:
1250 nr = i_data[EXT2_TIND_BLOCK];
1251 if (nr) {
1252 i_data[EXT2_TIND_BLOCK] = 0;
1253 mark_inode_dirty(inode);
1254 ext2_free_branches(inode, &nr, &nr+1, 3);
1255 }
1256 case EXT2_TIND_BLOCK:
1257 ;
1258 }
Martin J. Bligha686cd82007-10-16 23:30:46 -07001259
1260 ext2_discard_reservation(inode);
1261
1262 mutex_unlock(&ei->truncate_mutex);
npiggin@suse.de737f2e92010-05-27 01:05:37 +10001263}
1264
1265static void ext2_truncate_blocks(struct inode *inode, loff_t offset)
1266{
1267 /*
1268 * XXX: it seems like a bug here that we don't allow
1269 * IS_APPEND inode to have blocks-past-i_size trimmed off.
1270 * review and fix this.
1271 *
1272 * Also would be nice to be able to handle IO errors and such,
1273 * but that's probably too much to ask.
1274 */
1275 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1276 S_ISLNK(inode->i_mode)))
1277 return;
1278 if (ext2_inode_is_fast_symlink(inode))
1279 return;
1280 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1281 return;
Ross Zwisler5726b272015-10-13 16:25:37 -06001282
1283 dax_sem_down_write(EXT2_I(inode));
npiggin@suse.de737f2e92010-05-27 01:05:37 +10001284 __ext2_truncate_blocks(inode, offset);
Ross Zwisler5726b272015-10-13 16:25:37 -06001285 dax_sem_up_write(EXT2_I(inode));
npiggin@suse.de737f2e92010-05-27 01:05:37 +10001286}
1287
Christoph Hellwig2c27c652010-06-04 11:30:04 +02001288static int ext2_setsize(struct inode *inode, loff_t newsize)
npiggin@suse.de737f2e92010-05-27 01:05:37 +10001289{
npiggin@suse.de737f2e92010-05-27 01:05:37 +10001290 int error;
1291
npiggin@suse.de737f2e92010-05-27 01:05:37 +10001292 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1293 S_ISLNK(inode->i_mode)))
1294 return -EINVAL;
1295 if (ext2_inode_is_fast_symlink(inode))
1296 return -EINVAL;
1297 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1298 return -EPERM;
1299
Christoph Hellwig562c72a2011-06-24 14:29:45 -04001300 inode_dio_wait(inode);
1301
Jan Kara00697ee2016-11-20 20:47:07 -05001302 if (IS_DAX(inode)) {
1303 error = iomap_zero_range(inode, newsize,
1304 PAGE_ALIGN(newsize) - newsize, NULL,
1305 &ext2_iomap_ops);
1306 } else if (test_opt(inode->i_sb, NOBH))
npiggin@suse.de737f2e92010-05-27 01:05:37 +10001307 error = nobh_truncate_page(inode->i_mapping,
1308 newsize, ext2_get_block);
1309 else
1310 error = block_truncate_page(inode->i_mapping,
1311 newsize, ext2_get_block);
1312 if (error)
1313 return error;
1314
Ross Zwisler5726b272015-10-13 16:25:37 -06001315 dax_sem_down_write(EXT2_I(inode));
Christoph Hellwig2c27c652010-06-04 11:30:04 +02001316 truncate_setsize(inode, newsize);
npiggin@suse.de737f2e92010-05-27 01:05:37 +10001317 __ext2_truncate_blocks(inode, newsize);
Ross Zwisler5726b272015-10-13 16:25:37 -06001318 dax_sem_up_write(EXT2_I(inode));
npiggin@suse.de737f2e92010-05-27 01:05:37 +10001319
Deepa Dinamani02027d42016-09-14 07:48:05 -07001320 inode->i_mtime = inode->i_ctime = current_time(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 if (inode_needs_sync(inode)) {
1322 sync_mapping_buffers(inode->i_mapping);
Christoph Hellwigc37650162010-10-06 10:48:20 +02001323 sync_inode_metadata(inode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 } else {
1325 mark_inode_dirty(inode);
1326 }
npiggin@suse.de737f2e92010-05-27 01:05:37 +10001327
1328 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329}
1330
1331static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1332 struct buffer_head **p)
1333{
1334 struct buffer_head * bh;
1335 unsigned long block_group;
1336 unsigned long block;
1337 unsigned long offset;
1338 struct ext2_group_desc * gdp;
1339
1340 *p = NULL;
1341 if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1342 ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1343 goto Einval;
1344
1345 block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
Eric Sandeenef2fb672007-10-16 23:26:30 -07001346 gdp = ext2_get_group_desc(sb, block_group, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 if (!gdp)
1348 goto Egdp;
1349 /*
1350 * Figure out the offset within the block group inode table
1351 */
1352 offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1353 block = le32_to_cpu(gdp->bg_inode_table) +
1354 (offset >> EXT2_BLOCK_SIZE_BITS(sb));
1355 if (!(bh = sb_bread(sb, block)))
1356 goto Eio;
1357
1358 *p = bh;
1359 offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1360 return (struct ext2_inode *) (bh->b_data + offset);
1361
1362Einval:
1363 ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1364 (unsigned long) ino);
1365 return ERR_PTR(-EINVAL);
1366Eio:
1367 ext2_error(sb, "ext2_get_inode",
1368 "unable to read inode block - inode=%lu, block=%lu",
1369 (unsigned long) ino, block);
1370Egdp:
1371 return ERR_PTR(-EIO);
1372}
1373
1374void ext2_set_inode_flags(struct inode *inode)
1375{
1376 unsigned int flags = EXT2_I(inode)->i_flags;
1377
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08001378 inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
1379 S_DIRSYNC | S_DAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 if (flags & EXT2_SYNC_FL)
1381 inode->i_flags |= S_SYNC;
1382 if (flags & EXT2_APPEND_FL)
1383 inode->i_flags |= S_APPEND;
1384 if (flags & EXT2_IMMUTABLE_FL)
1385 inode->i_flags |= S_IMMUTABLE;
1386 if (flags & EXT2_NOATIME_FL)
1387 inode->i_flags |= S_NOATIME;
1388 if (flags & EXT2_DIRSYNC_FL)
1389 inode->i_flags |= S_DIRSYNC;
Ross Zwisler0a6cf912016-02-26 15:19:46 -08001390 if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode))
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08001391 inode->i_flags |= S_DAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392}
1393
David Howells52fcf702008-02-07 00:15:35 -08001394struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395{
David Howells52fcf702008-02-07 00:15:35 -08001396 struct ext2_inode_info *ei;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 struct buffer_head * bh;
David Howells52fcf702008-02-07 00:15:35 -08001398 struct ext2_inode *raw_inode;
1399 struct inode *inode;
1400 long ret = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 int n;
Eric W. Biedermanb8a9f9e2012-02-07 15:39:12 -08001402 uid_t i_uid;
1403 gid_t i_gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404
David Howells52fcf702008-02-07 00:15:35 -08001405 inode = iget_locked(sb, ino);
1406 if (!inode)
1407 return ERR_PTR(-ENOMEM);
1408 if (!(inode->i_state & I_NEW))
1409 return inode;
1410
1411 ei = EXT2_I(inode);
Martin J. Bligha686cd82007-10-16 23:30:46 -07001412 ei->i_block_alloc_info = NULL;
1413
David Howells52fcf702008-02-07 00:15:35 -08001414 raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1415 if (IS_ERR(raw_inode)) {
1416 ret = PTR_ERR(raw_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 goto bad_inode;
David Howells52fcf702008-02-07 00:15:35 -08001418 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419
1420 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
Eric W. Biedermanb8a9f9e2012-02-07 15:39:12 -08001421 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1422 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 if (!(test_opt (inode->i_sb, NO_UID32))) {
Eric W. Biedermanb8a9f9e2012-02-07 15:39:12 -08001424 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1425 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 }
Eric W. Biedermanb8a9f9e2012-02-07 15:39:12 -08001427 i_uid_write(inode, i_uid);
1428 i_gid_write(inode, i_gid);
Miklos Szeredibfe86842011-10-28 14:13:29 +02001429 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 inode->i_size = le32_to_cpu(raw_inode->i_size);
Markus Rechberger4d7bf112007-05-08 00:23:39 -07001431 inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
1432 inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
1433 inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1435 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1436 /* We now have enough fields to check if the inode was active or not.
1437 * This is needed because nfsd might try to access dead inodes
1438 * the test is that same one that e2fsck uses
1439 * NeilBrown 1999oct15
1440 */
1441 if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1442 /* this inode is deleted */
1443 brelse (bh);
David Howells52fcf702008-02-07 00:15:35 -08001444 ret = -ESTALE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 goto bad_inode;
1446 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1448 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1449 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1450 ei->i_frag_no = raw_inode->i_frag;
1451 ei->i_frag_size = raw_inode->i_fsize;
1452 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1453 ei->i_dir_acl = 0;
Carlos Maiolinoff0031d2016-07-05 22:02:41 -04001454
1455 if (ei->i_file_acl &&
1456 !ext2_data_block_valid(EXT2_SB(sb), ei->i_file_acl, 1)) {
1457 ext2_error(sb, "ext2_iget", "bad extended attribute block %u",
1458 ei->i_file_acl);
1459 brelse(bh);
1460 ret = -EFSCORRUPTED;
1461 goto bad_inode;
1462 }
1463
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 if (S_ISREG(inode->i_mode))
1465 inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1466 else
1467 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
Darrick J. Wongb46dc032016-12-06 15:59:01 -08001468 if (i_size_read(inode) < 0) {
1469 ret = -EFSCORRUPTED;
1470 goto bad_inode;
1471 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 ei->i_dtime = 0;
1473 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1474 ei->i_state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1476 ei->i_dir_start_lookup = 0;
1477
1478 /*
1479 * NOTE! The in-memory inode i_data array is in little-endian order
1480 * even on big-endian machines: we do NOT byteswap the block numbers!
1481 */
1482 for (n = 0; n < EXT2_N_BLOCKS; n++)
1483 ei->i_data[n] = raw_inode->i_block[n];
1484
1485 if (S_ISREG(inode->i_mode)) {
1486 inode->i_op = &ext2_file_inode_operations;
Boaz Harroshbe64f882015-04-15 16:15:17 -07001487 if (test_opt(inode->i_sb, NOBH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 inode->i_mapping->a_ops = &ext2_nobh_aops;
Carsten Otte6d791252005-06-23 22:05:26 -07001489 inode->i_fop = &ext2_file_operations;
1490 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 inode->i_mapping->a_ops = &ext2_aops;
Carsten Otte6d791252005-06-23 22:05:26 -07001492 inode->i_fop = &ext2_file_operations;
1493 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 } else if (S_ISDIR(inode->i_mode)) {
1495 inode->i_op = &ext2_dir_inode_operations;
1496 inode->i_fop = &ext2_dir_operations;
1497 if (test_opt(inode->i_sb, NOBH))
1498 inode->i_mapping->a_ops = &ext2_nobh_aops;
1499 else
1500 inode->i_mapping->a_ops = &ext2_aops;
1501 } else if (S_ISLNK(inode->i_mode)) {
Duane Griffin8d6d0c42008-12-19 20:47:13 +00001502 if (ext2_inode_is_fast_symlink(inode)) {
Al Virocbe0fa32015-05-02 10:02:46 -04001503 inode->i_link = (char *)ei->i_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 inode->i_op = &ext2_fast_symlink_inode_operations;
Duane Griffin8d6d0c42008-12-19 20:47:13 +00001505 nd_terminate_link(ei->i_data, inode->i_size,
1506 sizeof(ei->i_data) - 1);
1507 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 inode->i_op = &ext2_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -05001509 inode_nohighmem(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 if (test_opt(inode->i_sb, NOBH))
1511 inode->i_mapping->a_ops = &ext2_nobh_aops;
1512 else
1513 inode->i_mapping->a_ops = &ext2_aops;
1514 }
1515 } else {
1516 inode->i_op = &ext2_special_inode_operations;
1517 if (raw_inode->i_block[0])
1518 init_special_inode(inode, inode->i_mode,
1519 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1520 else
1521 init_special_inode(inode, inode->i_mode,
1522 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1523 }
1524 brelse (bh);
1525 ext2_set_inode_flags(inode);
David Howells52fcf702008-02-07 00:15:35 -08001526 unlock_new_inode(inode);
1527 return inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528
1529bad_inode:
David Howells52fcf702008-02-07 00:15:35 -08001530 iget_failed(inode);
1531 return ERR_PTR(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532}
1533
Christoph Hellwiga9185b42010-03-05 09:21:37 +01001534static int __ext2_write_inode(struct inode *inode, int do_sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535{
1536 struct ext2_inode_info *ei = EXT2_I(inode);
1537 struct super_block *sb = inode->i_sb;
1538 ino_t ino = inode->i_ino;
Eric W. Biedermanb8a9f9e2012-02-07 15:39:12 -08001539 uid_t uid = i_uid_read(inode);
1540 gid_t gid = i_gid_read(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 struct buffer_head * bh;
1542 struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1543 int n;
1544 int err = 0;
1545
1546 if (IS_ERR(raw_inode))
1547 return -EIO;
1548
1549 /* For fields not not tracking in the in-memory inode,
1550 * initialise them to zero for new inodes. */
1551 if (ei->i_state & EXT2_STATE_NEW)
1552 memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1553
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1555 if (!(test_opt(sb, NO_UID32))) {
1556 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1557 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1558/*
1559 * Fix up interoperability with old kernels. Otherwise, old inodes get
1560 * re-used with the upper 16 bits of the uid/gid intact
1561 */
1562 if (!ei->i_dtime) {
1563 raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1564 raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1565 } else {
1566 raw_inode->i_uid_high = 0;
1567 raw_inode->i_gid_high = 0;
1568 }
1569 } else {
1570 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1571 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1572 raw_inode->i_uid_high = 0;
1573 raw_inode->i_gid_high = 0;
1574 }
1575 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1576 raw_inode->i_size = cpu_to_le32(inode->i_size);
1577 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1578 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1579 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1580
1581 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1582 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1583 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1584 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1585 raw_inode->i_frag = ei->i_frag_no;
1586 raw_inode->i_fsize = ei->i_frag_size;
1587 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1588 if (!S_ISREG(inode->i_mode))
1589 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1590 else {
1591 raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1592 if (inode->i_size > 0x7fffffffULL) {
1593 if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1594 EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1595 EXT2_SB(sb)->s_es->s_rev_level ==
1596 cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1597 /* If this is the first large file
1598 * created, add a flag to the superblock.
1599 */
Jan Blunckc15271f2010-04-14 14:38:38 +02001600 spin_lock(&EXT2_SB(sb)->s_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 ext2_update_dynamic_rev(sb);
1602 EXT2_SET_RO_COMPAT_FEATURE(sb,
1603 EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
Jan Blunckc15271f2010-04-14 14:38:38 +02001604 spin_unlock(&EXT2_SB(sb)->s_lock);
Jan Kara65547662017-04-05 13:32:52 +02001605 ext2_sync_super(sb, EXT2_SB(sb)->s_es, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 }
1607 }
1608 }
1609
1610 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1611 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1612 if (old_valid_dev(inode->i_rdev)) {
1613 raw_inode->i_block[0] =
1614 cpu_to_le32(old_encode_dev(inode->i_rdev));
1615 raw_inode->i_block[1] = 0;
1616 } else {
1617 raw_inode->i_block[0] = 0;
1618 raw_inode->i_block[1] =
1619 cpu_to_le32(new_encode_dev(inode->i_rdev));
1620 raw_inode->i_block[2] = 0;
1621 }
1622 } else for (n = 0; n < EXT2_N_BLOCKS; n++)
1623 raw_inode->i_block[n] = ei->i_data[n];
1624 mark_buffer_dirty(bh);
1625 if (do_sync) {
1626 sync_dirty_buffer(bh);
1627 if (buffer_req(bh) && !buffer_uptodate(bh)) {
1628 printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1629 sb->s_id, (unsigned long) ino);
1630 err = -EIO;
1631 }
1632 }
1633 ei->i_state &= ~EXT2_STATE_NEW;
1634 brelse (bh);
1635 return err;
1636}
1637
Christoph Hellwiga9185b42010-03-05 09:21:37 +01001638int ext2_write_inode(struct inode *inode, struct writeback_control *wbc)
1639{
1640 return __ext2_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1641}
1642
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1644{
David Howells2b0143b2015-03-17 22:25:59 +00001645 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 int error;
1647
Jan Kara31051c82016-05-26 16:55:18 +02001648 error = setattr_prepare(dentry, iattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 if (error)
1650 return error;
Christoph Hellwig907f4552010-03-03 09:05:06 -05001651
Jan Karac2edb302015-06-29 16:08:45 +02001652 if (is_quota_modification(inode, iattr)) {
1653 error = dquot_initialize(inode);
1654 if (error)
1655 return error;
1656 }
Eric W. Biedermanb8a9f9e2012-02-07 15:39:12 -08001657 if ((iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)) ||
1658 (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid))) {
Christoph Hellwigb43fa822010-03-03 09:05:03 -05001659 error = dquot_transfer(inode, iattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 if (error)
1661 return error;
1662 }
Nick Pigginaf5a30d2010-06-03 22:01:46 +10001663 if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) {
npiggin@suse.de737f2e92010-05-27 01:05:37 +10001664 error = ext2_setsize(inode, iattr->ia_size);
1665 if (error)
1666 return error;
1667 }
Christoph Hellwig6a1a90a2010-06-04 11:30:00 +02001668 setattr_copy(inode, iattr);
npiggin@suse.de737f2e92010-05-27 01:05:37 +10001669 if (iattr->ia_valid & ATTR_MODE)
Christoph Hellwig64e178a2013-12-20 05:16:44 -08001670 error = posix_acl_chmod(inode, inode->i_mode);
npiggin@suse.de737f2e92010-05-27 01:05:37 +10001671 mark_inode_dirty(inode);
1672
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 return error;
1674}