xref: /openbmc/linux/fs/ext4/indirect.c (revision 188c299e2a26cc33747187f87c9e044dfd85a782)
1b2441318SGreg Kroah-Hartman // SPDX-License-Identifier: GPL-2.0
2dae1e52cSAmir Goldstein /*
3dae1e52cSAmir Goldstein  *  linux/fs/ext4/indirect.c
4dae1e52cSAmir Goldstein  *
5dae1e52cSAmir Goldstein  *  from
6dae1e52cSAmir Goldstein  *
7dae1e52cSAmir Goldstein  *  linux/fs/ext4/inode.c
8dae1e52cSAmir Goldstein  *
9dae1e52cSAmir Goldstein  * Copyright (C) 1992, 1993, 1994, 1995
10dae1e52cSAmir Goldstein  * Remy Card (card@masi.ibp.fr)
11dae1e52cSAmir Goldstein  * Laboratoire MASI - Institut Blaise Pascal
12dae1e52cSAmir Goldstein  * Universite Pierre et Marie Curie (Paris VI)
13dae1e52cSAmir Goldstein  *
14dae1e52cSAmir Goldstein  *  from
15dae1e52cSAmir Goldstein  *
16dae1e52cSAmir Goldstein  *  linux/fs/minix/inode.c
17dae1e52cSAmir Goldstein  *
18dae1e52cSAmir Goldstein  *  Copyright (C) 1991, 1992  Linus Torvalds
19dae1e52cSAmir Goldstein  *
20dae1e52cSAmir Goldstein  *  Goal-directed block allocation by Stephen Tweedie
21dae1e52cSAmir Goldstein  *	(sct@redhat.com), 1993, 1998
22dae1e52cSAmir Goldstein  */
23dae1e52cSAmir Goldstein 
24dae1e52cSAmir Goldstein #include "ext4_jbd2.h"
25dae1e52cSAmir Goldstein #include "truncate.h"
26c94c2acfSMatthew Wilcox #include <linux/dax.h>
27e2e40f2cSChristoph Hellwig #include <linux/uio.h>
28dae1e52cSAmir Goldstein 
29dae1e52cSAmir Goldstein #include <trace/events/ext4.h>
30dae1e52cSAmir Goldstein 
31dae1e52cSAmir Goldstein typedef struct {
32dae1e52cSAmir Goldstein 	__le32	*p;
33dae1e52cSAmir Goldstein 	__le32	key;
34dae1e52cSAmir Goldstein 	struct buffer_head *bh;
35dae1e52cSAmir Goldstein } Indirect;
36dae1e52cSAmir Goldstein 
37dae1e52cSAmir Goldstein static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
38dae1e52cSAmir Goldstein {
39dae1e52cSAmir Goldstein 	p->key = *(p->p = v);
40dae1e52cSAmir Goldstein 	p->bh = bh;
41dae1e52cSAmir Goldstein }
42dae1e52cSAmir Goldstein 
43dae1e52cSAmir Goldstein /**
44dae1e52cSAmir Goldstein  *	ext4_block_to_path - parse the block number into array of offsets
45dae1e52cSAmir Goldstein  *	@inode: inode in question (we are only interested in its superblock)
46dae1e52cSAmir Goldstein  *	@i_block: block number to be parsed
47dae1e52cSAmir Goldstein  *	@offsets: array to store the offsets in
48dae1e52cSAmir Goldstein  *	@boundary: set this non-zero if the referred-to block is likely to be
49dae1e52cSAmir Goldstein  *	       followed (on disk) by an indirect block.
50dae1e52cSAmir Goldstein  *
51dae1e52cSAmir Goldstein  *	To store the locations of file's data ext4 uses a data structure common
52dae1e52cSAmir Goldstein  *	for UNIX filesystems - tree of pointers anchored in the inode, with
53dae1e52cSAmir Goldstein  *	data blocks at leaves and indirect blocks in intermediate nodes.
54dae1e52cSAmir Goldstein  *	This function translates the block number into path in that tree -
55dae1e52cSAmir Goldstein  *	return value is the path length and @offsets[n] is the offset of
56dae1e52cSAmir Goldstein  *	pointer to (n+1)th node in the nth one. If @block is out of range
57dae1e52cSAmir Goldstein  *	(negative or too large) warning is printed and zero returned.
58dae1e52cSAmir Goldstein  *
59dae1e52cSAmir Goldstein  *	Note: function doesn't find node addresses, so no IO is needed. All
60dae1e52cSAmir Goldstein  *	we need to know is the capacity of indirect blocks (taken from the
61dae1e52cSAmir Goldstein  *	inode->i_sb).
62dae1e52cSAmir Goldstein  */
63dae1e52cSAmir Goldstein 
64dae1e52cSAmir Goldstein /*
65dae1e52cSAmir Goldstein  * Portability note: the last comparison (check that we fit into triple
66dae1e52cSAmir Goldstein  * indirect block) is spelled differently, because otherwise on an
67dae1e52cSAmir Goldstein  * architecture with 32-bit longs and 8Kb pages we might get into trouble
68dae1e52cSAmir Goldstein  * if our filesystem had 8Kb blocks. We might use long long, but that would
69dae1e52cSAmir Goldstein  * kill us on x86. Oh, well, at least the sign propagation does not matter -
70dae1e52cSAmir Goldstein  * i_block would have to be negative in the very beginning, so we would not
71dae1e52cSAmir Goldstein  * get there at all.
72dae1e52cSAmir Goldstein  */
73dae1e52cSAmir Goldstein 
74dae1e52cSAmir Goldstein static int ext4_block_to_path(struct inode *inode,
75dae1e52cSAmir Goldstein 			      ext4_lblk_t i_block,
76dae1e52cSAmir Goldstein 			      ext4_lblk_t offsets[4], int *boundary)
77dae1e52cSAmir Goldstein {
78dae1e52cSAmir Goldstein 	int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
79dae1e52cSAmir Goldstein 	int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
80dae1e52cSAmir Goldstein 	const long direct_blocks = EXT4_NDIR_BLOCKS,
81dae1e52cSAmir Goldstein 		indirect_blocks = ptrs,
82dae1e52cSAmir Goldstein 		double_blocks = (1 << (ptrs_bits * 2));
83dae1e52cSAmir Goldstein 	int n = 0;
84dae1e52cSAmir Goldstein 	int final = 0;
85dae1e52cSAmir Goldstein 
86dae1e52cSAmir Goldstein 	if (i_block < direct_blocks) {
87dae1e52cSAmir Goldstein 		offsets[n++] = i_block;
88dae1e52cSAmir Goldstein 		final = direct_blocks;
89dae1e52cSAmir Goldstein 	} else if ((i_block -= direct_blocks) < indirect_blocks) {
90dae1e52cSAmir Goldstein 		offsets[n++] = EXT4_IND_BLOCK;
91dae1e52cSAmir Goldstein 		offsets[n++] = i_block;
92dae1e52cSAmir Goldstein 		final = ptrs;
93dae1e52cSAmir Goldstein 	} else if ((i_block -= indirect_blocks) < double_blocks) {
94dae1e52cSAmir Goldstein 		offsets[n++] = EXT4_DIND_BLOCK;
95dae1e52cSAmir Goldstein 		offsets[n++] = i_block >> ptrs_bits;
96dae1e52cSAmir Goldstein 		offsets[n++] = i_block & (ptrs - 1);
97dae1e52cSAmir Goldstein 		final = ptrs;
98dae1e52cSAmir Goldstein 	} else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
99dae1e52cSAmir Goldstein 		offsets[n++] = EXT4_TIND_BLOCK;
100dae1e52cSAmir Goldstein 		offsets[n++] = i_block >> (ptrs_bits * 2);
101dae1e52cSAmir Goldstein 		offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
102dae1e52cSAmir Goldstein 		offsets[n++] = i_block & (ptrs - 1);
103dae1e52cSAmir Goldstein 		final = ptrs;
104dae1e52cSAmir Goldstein 	} else {
105dae1e52cSAmir Goldstein 		ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
106dae1e52cSAmir Goldstein 			     i_block + direct_blocks +
107dae1e52cSAmir Goldstein 			     indirect_blocks + double_blocks, inode->i_ino);
108dae1e52cSAmir Goldstein 	}
109dae1e52cSAmir Goldstein 	if (boundary)
110dae1e52cSAmir Goldstein 		*boundary = final - 1 - (i_block & (ptrs - 1));
111dae1e52cSAmir Goldstein 	return n;
112dae1e52cSAmir Goldstein }
113dae1e52cSAmir Goldstein 
114dae1e52cSAmir Goldstein /**
115dae1e52cSAmir Goldstein  *	ext4_get_branch - read the chain of indirect blocks leading to data
116dae1e52cSAmir Goldstein  *	@inode: inode in question
117dae1e52cSAmir Goldstein  *	@depth: depth of the chain (1 - direct pointer, etc.)
118dae1e52cSAmir Goldstein  *	@offsets: offsets of pointers in inode/indirect blocks
119dae1e52cSAmir Goldstein  *	@chain: place to store the result
120dae1e52cSAmir Goldstein  *	@err: here we store the error value
121dae1e52cSAmir Goldstein  *
122dae1e52cSAmir Goldstein  *	Function fills the array of triples <key, p, bh> and returns %NULL
123dae1e52cSAmir Goldstein  *	if everything went OK or the pointer to the last filled triple
124dae1e52cSAmir Goldstein  *	(incomplete one) otherwise. Upon the return chain[i].key contains
125dae1e52cSAmir Goldstein  *	the number of (i+1)-th block in the chain (as it is stored in memory,
126dae1e52cSAmir Goldstein  *	i.e. little-endian 32-bit), chain[i].p contains the address of that
127dae1e52cSAmir Goldstein  *	number (it points into struct inode for i==0 and into the bh->b_data
128dae1e52cSAmir Goldstein  *	for i>0) and chain[i].bh points to the buffer_head of i-th indirect
129dae1e52cSAmir Goldstein  *	block for i>0 and NULL for i==0. In other words, it holds the block
130dae1e52cSAmir Goldstein  *	numbers of the chain, addresses they were taken from (and where we can
131dae1e52cSAmir Goldstein  *	verify that chain did not change) and buffer_heads hosting these
132dae1e52cSAmir Goldstein  *	numbers.
133dae1e52cSAmir Goldstein  *
134dae1e52cSAmir Goldstein  *	Function stops when it stumbles upon zero pointer (absent block)
135dae1e52cSAmir Goldstein  *		(pointer to last triple returned, *@err == 0)
136dae1e52cSAmir Goldstein  *	or when it gets an IO error reading an indirect block
137dae1e52cSAmir Goldstein  *		(ditto, *@err == -EIO)
138dae1e52cSAmir Goldstein  *	or when it reads all @depth-1 indirect blocks successfully and finds
139dae1e52cSAmir Goldstein  *	the whole chain, all way to the data (returns %NULL, *err == 0).
140dae1e52cSAmir Goldstein  *
141dae1e52cSAmir Goldstein  *      Need to be called with
142dae1e52cSAmir Goldstein  *      down_read(&EXT4_I(inode)->i_data_sem)
143dae1e52cSAmir Goldstein  */
144dae1e52cSAmir Goldstein static Indirect *ext4_get_branch(struct inode *inode, int depth,
145dae1e52cSAmir Goldstein 				 ext4_lblk_t  *offsets,
146dae1e52cSAmir Goldstein 				 Indirect chain[4], int *err)
147dae1e52cSAmir Goldstein {
148dae1e52cSAmir Goldstein 	struct super_block *sb = inode->i_sb;
149dae1e52cSAmir Goldstein 	Indirect *p = chain;
150dae1e52cSAmir Goldstein 	struct buffer_head *bh;
151860d21e2STheodore Ts'o 	int ret = -EIO;
152dae1e52cSAmir Goldstein 
153dae1e52cSAmir Goldstein 	*err = 0;
154dae1e52cSAmir Goldstein 	/* i_data is not going away, no lock needed */
155dae1e52cSAmir Goldstein 	add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
156dae1e52cSAmir Goldstein 	if (!p->key)
157dae1e52cSAmir Goldstein 		goto no_block;
158dae1e52cSAmir Goldstein 	while (--depth) {
159dae1e52cSAmir Goldstein 		bh = sb_getblk(sb, le32_to_cpu(p->key));
160860d21e2STheodore Ts'o 		if (unlikely(!bh)) {
161860d21e2STheodore Ts'o 			ret = -ENOMEM;
162dae1e52cSAmir Goldstein 			goto failure;
163860d21e2STheodore Ts'o 		}
164dae1e52cSAmir Goldstein 
165dae1e52cSAmir Goldstein 		if (!bh_uptodate_or_lock(bh)) {
1662d069c08Szhangyi (F) 			if (ext4_read_bh(bh, 0, NULL) < 0) {
167dae1e52cSAmir Goldstein 				put_bh(bh);
168dae1e52cSAmir Goldstein 				goto failure;
169dae1e52cSAmir Goldstein 			}
170dae1e52cSAmir Goldstein 			/* validate block references */
171dae1e52cSAmir Goldstein 			if (ext4_check_indirect_blockref(inode, bh)) {
172dae1e52cSAmir Goldstein 				put_bh(bh);
173dae1e52cSAmir Goldstein 				goto failure;
174dae1e52cSAmir Goldstein 			}
175dae1e52cSAmir Goldstein 		}
176dae1e52cSAmir Goldstein 
177dae1e52cSAmir Goldstein 		add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
178dae1e52cSAmir Goldstein 		/* Reader: end */
179dae1e52cSAmir Goldstein 		if (!p->key)
180dae1e52cSAmir Goldstein 			goto no_block;
181dae1e52cSAmir Goldstein 	}
182dae1e52cSAmir Goldstein 	return NULL;
183dae1e52cSAmir Goldstein 
184dae1e52cSAmir Goldstein failure:
185860d21e2STheodore Ts'o 	*err = ret;
186dae1e52cSAmir Goldstein no_block:
187dae1e52cSAmir Goldstein 	return p;
188dae1e52cSAmir Goldstein }
189dae1e52cSAmir Goldstein 
190dae1e52cSAmir Goldstein /**
191dae1e52cSAmir Goldstein  *	ext4_find_near - find a place for allocation with sufficient locality
192dae1e52cSAmir Goldstein  *	@inode: owner
193dae1e52cSAmir Goldstein  *	@ind: descriptor of indirect block.
194dae1e52cSAmir Goldstein  *
195dae1e52cSAmir Goldstein  *	This function returns the preferred place for block allocation.
196dae1e52cSAmir Goldstein  *	It is used when heuristic for sequential allocation fails.
197dae1e52cSAmir Goldstein  *	Rules are:
198dae1e52cSAmir Goldstein  *	  + if there is a block to the left of our position - allocate near it.
199dae1e52cSAmir Goldstein  *	  + if pointer will live in indirect block - allocate near that block.
200dae1e52cSAmir Goldstein  *	  + if pointer will live in inode - allocate in the same
201dae1e52cSAmir Goldstein  *	    cylinder group.
202dae1e52cSAmir Goldstein  *
203dae1e52cSAmir Goldstein  * In the latter case we colour the starting block by the callers PID to
204dae1e52cSAmir Goldstein  * prevent it from clashing with concurrent allocations for a different inode
205dae1e52cSAmir Goldstein  * in the same block group.   The PID is used here so that functionally related
206dae1e52cSAmir Goldstein  * files will be close-by on-disk.
207dae1e52cSAmir Goldstein  *
208dae1e52cSAmir Goldstein  *	Caller must make sure that @ind is valid and will stay that way.
209dae1e52cSAmir Goldstein  */
210dae1e52cSAmir Goldstein static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
211dae1e52cSAmir Goldstein {
212dae1e52cSAmir Goldstein 	struct ext4_inode_info *ei = EXT4_I(inode);
213dae1e52cSAmir Goldstein 	__le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
214dae1e52cSAmir Goldstein 	__le32 *p;
215dae1e52cSAmir Goldstein 
216dae1e52cSAmir Goldstein 	/* Try to find previous block */
217dae1e52cSAmir Goldstein 	for (p = ind->p - 1; p >= start; p--) {
218dae1e52cSAmir Goldstein 		if (*p)
219dae1e52cSAmir Goldstein 			return le32_to_cpu(*p);
220dae1e52cSAmir Goldstein 	}
221dae1e52cSAmir Goldstein 
222dae1e52cSAmir Goldstein 	/* No such thing, so let's try location of indirect block */
223dae1e52cSAmir Goldstein 	if (ind->bh)
224dae1e52cSAmir Goldstein 		return ind->bh->b_blocknr;
225dae1e52cSAmir Goldstein 
226dae1e52cSAmir Goldstein 	/*
227dae1e52cSAmir Goldstein 	 * It is going to be referred to from the inode itself? OK, just put it
228dae1e52cSAmir Goldstein 	 * into the same cylinder group then.
229dae1e52cSAmir Goldstein 	 */
230f86186b4SEric Sandeen 	return ext4_inode_to_goal_block(inode);
231dae1e52cSAmir Goldstein }
232dae1e52cSAmir Goldstein 
233dae1e52cSAmir Goldstein /**
234dae1e52cSAmir Goldstein  *	ext4_find_goal - find a preferred place for allocation.
235dae1e52cSAmir Goldstein  *	@inode: owner
236dae1e52cSAmir Goldstein  *	@block:  block we want
237dae1e52cSAmir Goldstein  *	@partial: pointer to the last triple within a chain
238dae1e52cSAmir Goldstein  *
239dae1e52cSAmir Goldstein  *	Normally this function find the preferred place for block allocation,
240dae1e52cSAmir Goldstein  *	returns it.
241dae1e52cSAmir Goldstein  *	Because this is only used for non-extent files, we limit the block nr
242dae1e52cSAmir Goldstein  *	to 32 bits.
243dae1e52cSAmir Goldstein  */
244dae1e52cSAmir Goldstein static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
245dae1e52cSAmir Goldstein 				   Indirect *partial)
246dae1e52cSAmir Goldstein {
247dae1e52cSAmir Goldstein 	ext4_fsblk_t goal;
248dae1e52cSAmir Goldstein 
249dae1e52cSAmir Goldstein 	/*
250dae1e52cSAmir Goldstein 	 * XXX need to get goal block from mballoc's data structures
251dae1e52cSAmir Goldstein 	 */
252dae1e52cSAmir Goldstein 
253dae1e52cSAmir Goldstein 	goal = ext4_find_near(inode, partial);
254dae1e52cSAmir Goldstein 	goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
255dae1e52cSAmir Goldstein 	return goal;
256dae1e52cSAmir Goldstein }
257dae1e52cSAmir Goldstein 
258dae1e52cSAmir Goldstein /**
259dae1e52cSAmir Goldstein  *	ext4_blks_to_allocate - Look up the block map and count the number
260dae1e52cSAmir Goldstein  *	of direct blocks need to be allocated for the given branch.
261dae1e52cSAmir Goldstein  *
262dae1e52cSAmir Goldstein  *	@branch: chain of indirect blocks
263dae1e52cSAmir Goldstein  *	@k: number of blocks need for indirect blocks
264dae1e52cSAmir Goldstein  *	@blks: number of data blocks to be mapped.
265dae1e52cSAmir Goldstein  *	@blocks_to_boundary:  the offset in the indirect block
266dae1e52cSAmir Goldstein  *
267dae1e52cSAmir Goldstein  *	return the total number of blocks to be allocate, including the
268dae1e52cSAmir Goldstein  *	direct and indirect blocks.
269dae1e52cSAmir Goldstein  */
270dae1e52cSAmir Goldstein static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
271dae1e52cSAmir Goldstein 				 int blocks_to_boundary)
272dae1e52cSAmir Goldstein {
273dae1e52cSAmir Goldstein 	unsigned int count = 0;
274dae1e52cSAmir Goldstein 
275dae1e52cSAmir Goldstein 	/*
276dae1e52cSAmir Goldstein 	 * Simple case, [t,d]Indirect block(s) has not allocated yet
277dae1e52cSAmir Goldstein 	 * then it's clear blocks on that path have not allocated
278dae1e52cSAmir Goldstein 	 */
279dae1e52cSAmir Goldstein 	if (k > 0) {
280dae1e52cSAmir Goldstein 		/* right now we don't handle cross boundary allocation */
281dae1e52cSAmir Goldstein 		if (blks < blocks_to_boundary + 1)
282dae1e52cSAmir Goldstein 			count += blks;
283dae1e52cSAmir Goldstein 		else
284dae1e52cSAmir Goldstein 			count += blocks_to_boundary + 1;
285dae1e52cSAmir Goldstein 		return count;
286dae1e52cSAmir Goldstein 	}
287dae1e52cSAmir Goldstein 
288dae1e52cSAmir Goldstein 	count++;
289dae1e52cSAmir Goldstein 	while (count < blks && count <= blocks_to_boundary &&
290dae1e52cSAmir Goldstein 		le32_to_cpu(*(branch[0].p + count)) == 0) {
291dae1e52cSAmir Goldstein 		count++;
292dae1e52cSAmir Goldstein 	}
293dae1e52cSAmir Goldstein 	return count;
294dae1e52cSAmir Goldstein }
295dae1e52cSAmir Goldstein 
296dae1e52cSAmir Goldstein /**
297c60990b3STheodore Ts'o  * ext4_alloc_branch() - allocate and set up a chain of blocks
298dae1e52cSAmir Goldstein  * @handle: handle for this transaction
299c60990b3STheodore Ts'o  * @ar: structure describing the allocation request
300dae1e52cSAmir Goldstein  * @indirect_blks: number of allocated indirect blocks
301dae1e52cSAmir Goldstein  * @offsets: offsets (in the blocks) to store the pointers to next.
302dae1e52cSAmir Goldstein  * @branch: place to store the chain in.
303dae1e52cSAmir Goldstein  *
304dae1e52cSAmir Goldstein  *	This function allocates blocks, zeroes out all but the last one,
305dae1e52cSAmir Goldstein  *	links them into chain and (if we are synchronous) writes them to disk.
306dae1e52cSAmir Goldstein  *	In other words, it prepares a branch that can be spliced onto the
307dae1e52cSAmir Goldstein  *	inode. It stores the information about that chain in the branch[], in
308dae1e52cSAmir Goldstein  *	the same format as ext4_get_branch() would do. We are calling it after
309dae1e52cSAmir Goldstein  *	we had read the existing part of chain and partial points to the last
310dae1e52cSAmir Goldstein  *	triple of that (one with zero ->key). Upon the exit we have the same
311dae1e52cSAmir Goldstein  *	picture as after the successful ext4_get_block(), except that in one
312dae1e52cSAmir Goldstein  *	place chain is disconnected - *branch->p is still zero (we did not
313dae1e52cSAmir Goldstein  *	set the last link), but branch->key contains the number that should
314dae1e52cSAmir Goldstein  *	be placed into *branch->p to fill that gap.
315dae1e52cSAmir Goldstein  *
316dae1e52cSAmir Goldstein  *	If allocation fails we free all blocks we've allocated (and forget
317dae1e52cSAmir Goldstein  *	their buffer_heads) and return the error value the from failed
318dae1e52cSAmir Goldstein  *	ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
319dae1e52cSAmir Goldstein  *	as described above and return 0.
320dae1e52cSAmir Goldstein  */
321a5211002STheodore Ts'o static int ext4_alloc_branch(handle_t *handle,
322a5211002STheodore Ts'o 			     struct ext4_allocation_request *ar,
323a5211002STheodore Ts'o 			     int indirect_blks, ext4_lblk_t *offsets,
324a5211002STheodore Ts'o 			     Indirect *branch)
325dae1e52cSAmir Goldstein {
326dae1e52cSAmir Goldstein 	struct buffer_head *		bh;
327781f143eSTheodore Ts'o 	ext4_fsblk_t			b, new_blocks[4];
328781f143eSTheodore Ts'o 	__le32				*p;
329781f143eSTheodore Ts'o 	int				i, j, err, len = 1;
330dae1e52cSAmir Goldstein 
331781f143eSTheodore Ts'o 	for (i = 0; i <= indirect_blks; i++) {
332781f143eSTheodore Ts'o 		if (i == indirect_blks) {
333a5211002STheodore Ts'o 			new_blocks[i] = ext4_mb_new_blocks(handle, ar, &err);
334f2890730SJan Kara 		} else {
335a5211002STheodore Ts'o 			ar->goal = new_blocks[i] = ext4_new_meta_blocks(handle,
336e3cf5d5dSTheodore Ts'o 					ar->inode, ar->goal,
337e3cf5d5dSTheodore Ts'o 					ar->flags & EXT4_MB_DELALLOC_RESERVED,
338e3cf5d5dSTheodore Ts'o 					NULL, &err);
339f2890730SJan Kara 			/* Simplify error cleanup... */
340f2890730SJan Kara 			branch[i+1].bh = NULL;
341f2890730SJan Kara 		}
342781f143eSTheodore Ts'o 		if (err) {
343781f143eSTheodore Ts'o 			i--;
344781f143eSTheodore Ts'o 			goto failed;
345781f143eSTheodore Ts'o 		}
346781f143eSTheodore Ts'o 		branch[i].key = cpu_to_le32(new_blocks[i]);
347781f143eSTheodore Ts'o 		if (i == 0)
348781f143eSTheodore Ts'o 			continue;
349781f143eSTheodore Ts'o 
350a5211002STheodore Ts'o 		bh = branch[i].bh = sb_getblk(ar->inode->i_sb, new_blocks[i-1]);
351dae1e52cSAmir Goldstein 		if (unlikely(!bh)) {
352860d21e2STheodore Ts'o 			err = -ENOMEM;
353dae1e52cSAmir Goldstein 			goto failed;
354dae1e52cSAmir Goldstein 		}
355dae1e52cSAmir Goldstein 		lock_buffer(bh);
356dae1e52cSAmir Goldstein 		BUFFER_TRACE(bh, "call get_create_access");
357*188c299eSJan Kara 		err = ext4_journal_get_create_access(handle, ar->inode->i_sb,
358*188c299eSJan Kara 						     bh, EXT4_JTR_NONE);
359dae1e52cSAmir Goldstein 		if (err) {
360dae1e52cSAmir Goldstein 			unlock_buffer(bh);
361dae1e52cSAmir Goldstein 			goto failed;
362dae1e52cSAmir Goldstein 		}
363dae1e52cSAmir Goldstein 
364781f143eSTheodore Ts'o 		memset(bh->b_data, 0, bh->b_size);
365781f143eSTheodore Ts'o 		p = branch[i].p = (__le32 *) bh->b_data + offsets[i];
366781f143eSTheodore Ts'o 		b = new_blocks[i];
367781f143eSTheodore Ts'o 
368781f143eSTheodore Ts'o 		if (i == indirect_blks)
369a5211002STheodore Ts'o 			len = ar->len;
370781f143eSTheodore Ts'o 		for (j = 0; j < len; j++)
371781f143eSTheodore Ts'o 			*p++ = cpu_to_le32(b++);
372781f143eSTheodore Ts'o 
373dae1e52cSAmir Goldstein 		BUFFER_TRACE(bh, "marking uptodate");
374dae1e52cSAmir Goldstein 		set_buffer_uptodate(bh);
375dae1e52cSAmir Goldstein 		unlock_buffer(bh);
376dae1e52cSAmir Goldstein 
377dae1e52cSAmir Goldstein 		BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
378a5211002STheodore Ts'o 		err = ext4_handle_dirty_metadata(handle, ar->inode, bh);
379dae1e52cSAmir Goldstein 		if (err)
380dae1e52cSAmir Goldstein 			goto failed;
381dae1e52cSAmir Goldstein 	}
382781f143eSTheodore Ts'o 	return 0;
383dae1e52cSAmir Goldstein failed:
384f2890730SJan Kara 	if (i == indirect_blks) {
385f2890730SJan Kara 		/* Free data blocks */
386f2890730SJan Kara 		ext4_free_blocks(handle, ar->inode, NULL, new_blocks[i],
387f2890730SJan Kara 				 ar->len, 0);
388f2890730SJan Kara 		i--;
389f2890730SJan Kara 	}
390781f143eSTheodore Ts'o 	for (; i >= 0; i--) {
391c5c7b8ddSJan Kara 		/*
392c5c7b8ddSJan Kara 		 * We want to ext4_forget() only freshly allocated indirect
393f2890730SJan Kara 		 * blocks. Buffer for new_blocks[i] is at branch[i+1].bh
394f2890730SJan Kara 		 * (buffer at branch[0].bh is indirect block / inode already
395f2890730SJan Kara 		 * existing before ext4_alloc_branch() was called). Also
396f2890730SJan Kara 		 * because blocks are freshly allocated, we don't need to
397f2890730SJan Kara 		 * revoke them which is why we don't set
398f2890730SJan Kara 		 * EXT4_FREE_BLOCKS_METADATA.
399c5c7b8ddSJan Kara 		 */
400f2890730SJan Kara 		ext4_free_blocks(handle, ar->inode, branch[i+1].bh,
401f2890730SJan Kara 				 new_blocks[i], 1,
402f2890730SJan Kara 				 branch[i+1].bh ? EXT4_FREE_BLOCKS_FORGET : 0);
403dae1e52cSAmir Goldstein 	}
404dae1e52cSAmir Goldstein 	return err;
405dae1e52cSAmir Goldstein }
406dae1e52cSAmir Goldstein 
407dae1e52cSAmir Goldstein /**
408c60990b3STheodore Ts'o  * ext4_splice_branch() - splice the allocated branch onto inode.
409dae1e52cSAmir Goldstein  * @handle: handle for this transaction
410c60990b3STheodore Ts'o  * @ar: structure describing the allocation request
411dae1e52cSAmir Goldstein  * @where: location of missing link
412dae1e52cSAmir Goldstein  * @num:   number of indirect blocks we are adding
413dae1e52cSAmir Goldstein  *
414dae1e52cSAmir Goldstein  * This function fills the missing link and does all housekeeping needed in
415dae1e52cSAmir Goldstein  * inode (->i_blocks, etc.). In case of success we end up with the full
416dae1e52cSAmir Goldstein  * chain to new block and return 0.
417dae1e52cSAmir Goldstein  */
418a5211002STheodore Ts'o static int ext4_splice_branch(handle_t *handle,
419a5211002STheodore Ts'o 			      struct ext4_allocation_request *ar,
420a5211002STheodore Ts'o 			      Indirect *where, int num)
421dae1e52cSAmir Goldstein {
422dae1e52cSAmir Goldstein 	int i;
423dae1e52cSAmir Goldstein 	int err = 0;
424dae1e52cSAmir Goldstein 	ext4_fsblk_t current_block;
425dae1e52cSAmir Goldstein 
426dae1e52cSAmir Goldstein 	/*
427dae1e52cSAmir Goldstein 	 * If we're splicing into a [td]indirect block (as opposed to the
428dae1e52cSAmir Goldstein 	 * inode) then we need to get write access to the [td]indirect block
429dae1e52cSAmir Goldstein 	 * before the splice.
430dae1e52cSAmir Goldstein 	 */
431dae1e52cSAmir Goldstein 	if (where->bh) {
432dae1e52cSAmir Goldstein 		BUFFER_TRACE(where->bh, "get_write_access");
433*188c299eSJan Kara 		err = ext4_journal_get_write_access(handle, ar->inode->i_sb,
434*188c299eSJan Kara 						    where->bh, EXT4_JTR_NONE);
435dae1e52cSAmir Goldstein 		if (err)
436dae1e52cSAmir Goldstein 			goto err_out;
437dae1e52cSAmir Goldstein 	}
438dae1e52cSAmir Goldstein 	/* That's it */
439dae1e52cSAmir Goldstein 
440dae1e52cSAmir Goldstein 	*where->p = where->key;
441dae1e52cSAmir Goldstein 
442dae1e52cSAmir Goldstein 	/*
443dae1e52cSAmir Goldstein 	 * Update the host buffer_head or inode to point to more just allocated
444dae1e52cSAmir Goldstein 	 * direct blocks blocks
445dae1e52cSAmir Goldstein 	 */
446a5211002STheodore Ts'o 	if (num == 0 && ar->len > 1) {
447dae1e52cSAmir Goldstein 		current_block = le32_to_cpu(where->key) + 1;
448a5211002STheodore Ts'o 		for (i = 1; i < ar->len; i++)
449dae1e52cSAmir Goldstein 			*(where->p + i) = cpu_to_le32(current_block++);
450dae1e52cSAmir Goldstein 	}
451dae1e52cSAmir Goldstein 
452dae1e52cSAmir Goldstein 	/* We are done with atomic stuff, now do the rest of housekeeping */
453dae1e52cSAmir Goldstein 	/* had we spliced it onto indirect block? */
454dae1e52cSAmir Goldstein 	if (where->bh) {
455dae1e52cSAmir Goldstein 		/*
456dae1e52cSAmir Goldstein 		 * If we spliced it onto an indirect block, we haven't
457dae1e52cSAmir Goldstein 		 * altered the inode.  Note however that if it is being spliced
458dae1e52cSAmir Goldstein 		 * onto an indirect block at the very end of the file (the
459dae1e52cSAmir Goldstein 		 * file is growing) then we *will* alter the inode to reflect
460dae1e52cSAmir Goldstein 		 * the new i_size.  But that is not done here - it is done in
461dae1e52cSAmir Goldstein 		 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
462dae1e52cSAmir Goldstein 		 */
463dae1e52cSAmir Goldstein 		jbd_debug(5, "splicing indirect only\n");
464dae1e52cSAmir Goldstein 		BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
465a5211002STheodore Ts'o 		err = ext4_handle_dirty_metadata(handle, ar->inode, where->bh);
466dae1e52cSAmir Goldstein 		if (err)
467dae1e52cSAmir Goldstein 			goto err_out;
468dae1e52cSAmir Goldstein 	} else {
469dae1e52cSAmir Goldstein 		/*
470dae1e52cSAmir Goldstein 		 * OK, we spliced it into the inode itself on a direct block.
471dae1e52cSAmir Goldstein 		 */
4724209ae12SHarshad Shirwadkar 		err = ext4_mark_inode_dirty(handle, ar->inode);
4734209ae12SHarshad Shirwadkar 		if (unlikely(err))
4744209ae12SHarshad Shirwadkar 			goto err_out;
475dae1e52cSAmir Goldstein 		jbd_debug(5, "splicing direct\n");
476dae1e52cSAmir Goldstein 	}
477dae1e52cSAmir Goldstein 	return err;
478dae1e52cSAmir Goldstein 
479dae1e52cSAmir Goldstein err_out:
480dae1e52cSAmir Goldstein 	for (i = 1; i <= num; i++) {
481dae1e52cSAmir Goldstein 		/*
482dae1e52cSAmir Goldstein 		 * branch[i].bh is newly allocated, so there is no
483dae1e52cSAmir Goldstein 		 * need to revoke the block, which is why we don't
484dae1e52cSAmir Goldstein 		 * need to set EXT4_FREE_BLOCKS_METADATA.
485dae1e52cSAmir Goldstein 		 */
486a5211002STheodore Ts'o 		ext4_free_blocks(handle, ar->inode, where[i].bh, 0, 1,
487dae1e52cSAmir Goldstein 				 EXT4_FREE_BLOCKS_FORGET);
488dae1e52cSAmir Goldstein 	}
489a5211002STheodore Ts'o 	ext4_free_blocks(handle, ar->inode, NULL, le32_to_cpu(where[num].key),
490a5211002STheodore Ts'o 			 ar->len, 0);
491dae1e52cSAmir Goldstein 
492dae1e52cSAmir Goldstein 	return err;
493dae1e52cSAmir Goldstein }
494dae1e52cSAmir Goldstein 
495dae1e52cSAmir Goldstein /*
496dae1e52cSAmir Goldstein  * The ext4_ind_map_blocks() function handles non-extents inodes
497dae1e52cSAmir Goldstein  * (i.e., using the traditional indirect/double-indirect i_blocks
498dae1e52cSAmir Goldstein  * scheme) for ext4_map_blocks().
499dae1e52cSAmir Goldstein  *
500dae1e52cSAmir Goldstein  * Allocation strategy is simple: if we have to allocate something, we will
501dae1e52cSAmir Goldstein  * have to go the whole way to leaf. So let's do it before attaching anything
502dae1e52cSAmir Goldstein  * to tree, set linkage between the newborn blocks, write them if sync is
503dae1e52cSAmir Goldstein  * required, recheck the path, free and repeat if check fails, otherwise
504dae1e52cSAmir Goldstein  * set the last missing link (that will protect us from any truncate-generated
505dae1e52cSAmir Goldstein  * removals - all blocks on the path are immune now) and possibly force the
506dae1e52cSAmir Goldstein  * write on the parent block.
507dae1e52cSAmir Goldstein  * That has a nice additional property: no special recovery from the failed
508dae1e52cSAmir Goldstein  * allocations is needed - we simply release blocks and do not touch anything
509dae1e52cSAmir Goldstein  * reachable from inode.
510dae1e52cSAmir Goldstein  *
511dae1e52cSAmir Goldstein  * `handle' can be NULL if create == 0.
512dae1e52cSAmir Goldstein  *
513dae1e52cSAmir Goldstein  * return > 0, # of blocks mapped or allocated.
514dae1e52cSAmir Goldstein  * return = 0, if plain lookup failed.
515dae1e52cSAmir Goldstein  * return < 0, error case.
516dae1e52cSAmir Goldstein  *
517dae1e52cSAmir Goldstein  * The ext4_ind_get_blocks() function should be called with
518dae1e52cSAmir Goldstein  * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
519dae1e52cSAmir Goldstein  * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
520dae1e52cSAmir Goldstein  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
521dae1e52cSAmir Goldstein  * blocks.
522dae1e52cSAmir Goldstein  */
523dae1e52cSAmir Goldstein int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
524dae1e52cSAmir Goldstein 			struct ext4_map_blocks *map,
525dae1e52cSAmir Goldstein 			int flags)
526dae1e52cSAmir Goldstein {
527a5211002STheodore Ts'o 	struct ext4_allocation_request ar;
528dae1e52cSAmir Goldstein 	int err = -EIO;
529dae1e52cSAmir Goldstein 	ext4_lblk_t offsets[4];
530dae1e52cSAmir Goldstein 	Indirect chain[4];
531dae1e52cSAmir Goldstein 	Indirect *partial;
532dae1e52cSAmir Goldstein 	int indirect_blks;
533dae1e52cSAmir Goldstein 	int blocks_to_boundary = 0;
534dae1e52cSAmir Goldstein 	int depth;
535dae1e52cSAmir Goldstein 	int count = 0;
536dae1e52cSAmir Goldstein 	ext4_fsblk_t first_block = 0;
537dae1e52cSAmir Goldstein 
538dae1e52cSAmir Goldstein 	trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
539837c23fbSChunguang Xu 	ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
540837c23fbSChunguang Xu 	ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
541dae1e52cSAmir Goldstein 	depth = ext4_block_to_path(inode, map->m_lblk, offsets,
542dae1e52cSAmir Goldstein 				   &blocks_to_boundary);
543dae1e52cSAmir Goldstein 
544dae1e52cSAmir Goldstein 	if (depth == 0)
545dae1e52cSAmir Goldstein 		goto out;
546dae1e52cSAmir Goldstein 
547dae1e52cSAmir Goldstein 	partial = ext4_get_branch(inode, depth, offsets, chain, &err);
548dae1e52cSAmir Goldstein 
549dae1e52cSAmir Goldstein 	/* Simplest case - block found, no allocation needed */
550dae1e52cSAmir Goldstein 	if (!partial) {
551dae1e52cSAmir Goldstein 		first_block = le32_to_cpu(chain[depth - 1].key);
552dae1e52cSAmir Goldstein 		count++;
553dae1e52cSAmir Goldstein 		/*map more blocks*/
554dae1e52cSAmir Goldstein 		while (count < map->m_len && count <= blocks_to_boundary) {
555dae1e52cSAmir Goldstein 			ext4_fsblk_t blk;
556dae1e52cSAmir Goldstein 
557dae1e52cSAmir Goldstein 			blk = le32_to_cpu(*(chain[depth-1].p + count));
558dae1e52cSAmir Goldstein 
559dae1e52cSAmir Goldstein 			if (blk == first_block + count)
560dae1e52cSAmir Goldstein 				count++;
561dae1e52cSAmir Goldstein 			else
562dae1e52cSAmir Goldstein 				break;
563dae1e52cSAmir Goldstein 		}
564dae1e52cSAmir Goldstein 		goto got_it;
565dae1e52cSAmir Goldstein 	}
566dae1e52cSAmir Goldstein 
567facab4d9SJan Kara 	/* Next simple case - plain lookup failed */
568facab4d9SJan Kara 	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
569facab4d9SJan Kara 		unsigned epb = inode->i_sb->s_blocksize / sizeof(u32);
570facab4d9SJan Kara 		int i;
571facab4d9SJan Kara 
5722ee3ee06SJan Kara 		/*
5732ee3ee06SJan Kara 		 * Count number blocks in a subtree under 'partial'. At each
5742ee3ee06SJan Kara 		 * level we count number of complete empty subtrees beyond
5752ee3ee06SJan Kara 		 * current offset and then descend into the subtree only
5762ee3ee06SJan Kara 		 * partially beyond current offset.
5772ee3ee06SJan Kara 		 */
5782ee3ee06SJan Kara 		count = 0;
5792ee3ee06SJan Kara 		for (i = partial - chain + 1; i < depth; i++)
5802ee3ee06SJan Kara 			count = count * epb + (epb - offsets[i] - 1);
5812ee3ee06SJan Kara 		count++;
582facab4d9SJan Kara 		/* Fill in size of a hole we found */
583facab4d9SJan Kara 		map->m_pblk = 0;
584facab4d9SJan Kara 		map->m_len = min_t(unsigned int, map->m_len, count);
585facab4d9SJan Kara 		goto cleanup;
586facab4d9SJan Kara 	}
587facab4d9SJan Kara 
588facab4d9SJan Kara 	/* Failed read of indirect block */
589facab4d9SJan Kara 	if (err == -EIO)
590dae1e52cSAmir Goldstein 		goto cleanup;
591dae1e52cSAmir Goldstein 
592dae1e52cSAmir Goldstein 	/*
593dae1e52cSAmir Goldstein 	 * Okay, we need to do block allocation.
594dae1e52cSAmir Goldstein 	*/
595e2b911c5SDarrick J. Wong 	if (ext4_has_feature_bigalloc(inode->i_sb)) {
596bab08ab9STheodore Ts'o 		EXT4_ERROR_INODE(inode, "Can't allocate blocks for "
597bab08ab9STheodore Ts'o 				 "non-extent mapped inodes with bigalloc");
5982be7d717SZhang Qilong 		err = -EFSCORRUPTED;
5992be7d717SZhang Qilong 		goto out;
600bab08ab9STheodore Ts'o 	}
601bab08ab9STheodore Ts'o 
602a5211002STheodore Ts'o 	/* Set up for the direct block allocation */
603a5211002STheodore Ts'o 	memset(&ar, 0, sizeof(ar));
604a5211002STheodore Ts'o 	ar.inode = inode;
605a5211002STheodore Ts'o 	ar.logical = map->m_lblk;
606a5211002STheodore Ts'o 	if (S_ISREG(inode->i_mode))
607a5211002STheodore Ts'o 		ar.flags = EXT4_MB_HINT_DATA;
608e3cf5d5dSTheodore Ts'o 	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
609e3cf5d5dSTheodore Ts'o 		ar.flags |= EXT4_MB_DELALLOC_RESERVED;
610c5e298aeSTheodore Ts'o 	if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
611c5e298aeSTheodore Ts'o 		ar.flags |= EXT4_MB_USE_RESERVED;
612a5211002STheodore Ts'o 
613a5211002STheodore Ts'o 	ar.goal = ext4_find_goal(inode, map->m_lblk, partial);
614dae1e52cSAmir Goldstein 
615dae1e52cSAmir Goldstein 	/* the number of blocks need to allocate for [d,t]indirect blocks */
616dae1e52cSAmir Goldstein 	indirect_blks = (chain + depth) - partial - 1;
617dae1e52cSAmir Goldstein 
618dae1e52cSAmir Goldstein 	/*
619dae1e52cSAmir Goldstein 	 * Next look up the indirect map to count the totoal number of
620dae1e52cSAmir Goldstein 	 * direct blocks to allocate for this branch.
621dae1e52cSAmir Goldstein 	 */
622a5211002STheodore Ts'o 	ar.len = ext4_blks_to_allocate(partial, indirect_blks,
623dae1e52cSAmir Goldstein 				       map->m_len, blocks_to_boundary);
624a5211002STheodore Ts'o 
625dae1e52cSAmir Goldstein 	/*
626dae1e52cSAmir Goldstein 	 * Block out ext4_truncate while we alter the tree
627dae1e52cSAmir Goldstein 	 */
628a5211002STheodore Ts'o 	err = ext4_alloc_branch(handle, &ar, indirect_blks,
629dae1e52cSAmir Goldstein 				offsets + (partial - chain), partial);
630dae1e52cSAmir Goldstein 
631dae1e52cSAmir Goldstein 	/*
632dae1e52cSAmir Goldstein 	 * The ext4_splice_branch call will free and forget any buffers
633dae1e52cSAmir Goldstein 	 * on the new chain if there is a failure, but that risks using
634dae1e52cSAmir Goldstein 	 * up transaction credits, especially for bitmaps where the
635dae1e52cSAmir Goldstein 	 * credits cannot be returned.  Can we handle this somehow?  We
636dae1e52cSAmir Goldstein 	 * may need to return -EAGAIN upwards in the worst case.  --sct
637dae1e52cSAmir Goldstein 	 */
638dae1e52cSAmir Goldstein 	if (!err)
639a5211002STheodore Ts'o 		err = ext4_splice_branch(handle, &ar, partial, indirect_blks);
640dae1e52cSAmir Goldstein 	if (err)
641dae1e52cSAmir Goldstein 		goto cleanup;
642dae1e52cSAmir Goldstein 
643dae1e52cSAmir Goldstein 	map->m_flags |= EXT4_MAP_NEW;
644dae1e52cSAmir Goldstein 
645dae1e52cSAmir Goldstein 	ext4_update_inode_fsync_trans(handle, inode, 1);
646a5211002STheodore Ts'o 	count = ar.len;
647dae1e52cSAmir Goldstein got_it:
648dae1e52cSAmir Goldstein 	map->m_flags |= EXT4_MAP_MAPPED;
649dae1e52cSAmir Goldstein 	map->m_pblk = le32_to_cpu(chain[depth-1].key);
650dae1e52cSAmir Goldstein 	map->m_len = count;
651dae1e52cSAmir Goldstein 	if (count > blocks_to_boundary)
652dae1e52cSAmir Goldstein 		map->m_flags |= EXT4_MAP_BOUNDARY;
653dae1e52cSAmir Goldstein 	err = count;
654dae1e52cSAmir Goldstein 	/* Clean up and exit */
655dae1e52cSAmir Goldstein 	partial = chain + depth - 1;	/* the whole chain */
656dae1e52cSAmir Goldstein cleanup:
657dae1e52cSAmir Goldstein 	while (partial > chain) {
658dae1e52cSAmir Goldstein 		BUFFER_TRACE(partial->bh, "call brelse");
659dae1e52cSAmir Goldstein 		brelse(partial->bh);
660dae1e52cSAmir Goldstein 		partial--;
661dae1e52cSAmir Goldstein 	}
662dae1e52cSAmir Goldstein out:
66321ddd568STheodore Ts'o 	trace_ext4_ind_map_blocks_exit(inode, flags, map, err);
664dae1e52cSAmir Goldstein 	return err;
665dae1e52cSAmir Goldstein }
666dae1e52cSAmir Goldstein 
667dae1e52cSAmir Goldstein /*
668fa55a0edSJan Kara  * Calculate number of indirect blocks touched by mapping @nrblocks logically
669fa55a0edSJan Kara  * contiguous blocks
670fa55a0edSJan Kara  */
671fa55a0edSJan Kara int ext4_ind_trans_blocks(struct inode *inode, int nrblocks)
672dae1e52cSAmir Goldstein {
673dae1e52cSAmir Goldstein 	/*
674dae1e52cSAmir Goldstein 	 * With N contiguous data blocks, we need at most
675dae1e52cSAmir Goldstein 	 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
676dae1e52cSAmir Goldstein 	 * 2 dindirect blocks, and 1 tindirect block
677dae1e52cSAmir Goldstein 	 */
678fa55a0edSJan Kara 	return DIV_ROUND_UP(nrblocks, EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
679dae1e52cSAmir Goldstein }
680dae1e52cSAmir Goldstein 
681a4130367SJan Kara static int ext4_ind_trunc_restart_fn(handle_t *handle, struct inode *inode,
682a4130367SJan Kara 				     struct buffer_head *bh, int *dropped)
683a4130367SJan Kara {
684a4130367SJan Kara 	int err;
685a4130367SJan Kara 
686a4130367SJan Kara 	if (bh) {
687a4130367SJan Kara 		BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
688a4130367SJan Kara 		err = ext4_handle_dirty_metadata(handle, inode, bh);
689a4130367SJan Kara 		if (unlikely(err))
690a4130367SJan Kara 			return err;
691a4130367SJan Kara 	}
692a4130367SJan Kara 	err = ext4_mark_inode_dirty(handle, inode);
693a4130367SJan Kara 	if (unlikely(err))
694a4130367SJan Kara 		return err;
695a4130367SJan Kara 	/*
696a4130367SJan Kara 	 * Drop i_data_sem to avoid deadlock with ext4_map_blocks.  At this
697a4130367SJan Kara 	 * moment, get_block can be called only for blocks inside i_size since
698a4130367SJan Kara 	 * page cache has been already dropped and writes are blocked by
699a4130367SJan Kara 	 * i_mutex. So we can safely drop the i_data_sem here.
700a4130367SJan Kara 	 */
701a4130367SJan Kara 	BUG_ON(EXT4_JOURNAL(inode) == NULL);
70227bc446eSbrookxu 	ext4_discard_preallocations(inode, 0);
703a4130367SJan Kara 	up_write(&EXT4_I(inode)->i_data_sem);
704a4130367SJan Kara 	*dropped = 1;
705a4130367SJan Kara 	return 0;
706a4130367SJan Kara }
707a4130367SJan Kara 
708dae1e52cSAmir Goldstein /*
709dae1e52cSAmir Goldstein  * Truncate transactions can be complex and absolutely huge.  So we need to
7103088e5a5SBhaskar Chowdhury  * be able to restart the transaction at a convenient checkpoint to make
711dae1e52cSAmir Goldstein  * sure we don't overflow the journal.
712dae1e52cSAmir Goldstein  *
713819c4920STheodore Ts'o  * Try to extend this transaction for the purposes of truncation.  If
714a4130367SJan Kara  * extend fails, we restart transaction.
715dae1e52cSAmir Goldstein  */
716a4130367SJan Kara static int ext4_ind_truncate_ensure_credits(handle_t *handle,
717a4130367SJan Kara 					    struct inode *inode,
71883448bdfSJan Kara 					    struct buffer_head *bh,
71983448bdfSJan Kara 					    int revoke_creds)
720dae1e52cSAmir Goldstein {
721a4130367SJan Kara 	int ret;
722a4130367SJan Kara 	int dropped = 0;
723a4130367SJan Kara 
724a4130367SJan Kara 	ret = ext4_journal_ensure_credits_fn(handle, EXT4_RESERVE_TRANS_BLOCKS,
72583448bdfSJan Kara 			ext4_blocks_for_truncate(inode), revoke_creds,
726a4130367SJan Kara 			ext4_ind_trunc_restart_fn(handle, inode, bh, &dropped));
727a4130367SJan Kara 	if (dropped)
728a4130367SJan Kara 		down_write(&EXT4_I(inode)->i_data_sem);
729a4130367SJan Kara 	if (ret <= 0)
730a4130367SJan Kara 		return ret;
731a4130367SJan Kara 	if (bh) {
732a4130367SJan Kara 		BUFFER_TRACE(bh, "retaking write access");
733*188c299eSJan Kara 		ret = ext4_journal_get_write_access(handle, inode->i_sb, bh,
734*188c299eSJan Kara 						    EXT4_JTR_NONE);
735a4130367SJan Kara 		if (unlikely(ret))
736a4130367SJan Kara 			return ret;
737a4130367SJan Kara 	}
738dae1e52cSAmir Goldstein 	return 0;
739dae1e52cSAmir Goldstein }
740dae1e52cSAmir Goldstein 
741dae1e52cSAmir Goldstein /*
742dae1e52cSAmir Goldstein  * Probably it should be a library function... search for first non-zero word
743dae1e52cSAmir Goldstein  * or memcmp with zero_page, whatever is better for particular architecture.
744dae1e52cSAmir Goldstein  * Linus?
745dae1e52cSAmir Goldstein  */
746dae1e52cSAmir Goldstein static inline int all_zeroes(__le32 *p, __le32 *q)
747dae1e52cSAmir Goldstein {
748dae1e52cSAmir Goldstein 	while (p < q)
749dae1e52cSAmir Goldstein 		if (*p++)
750dae1e52cSAmir Goldstein 			return 0;
751dae1e52cSAmir Goldstein 	return 1;
752dae1e52cSAmir Goldstein }
753dae1e52cSAmir Goldstein 
754dae1e52cSAmir Goldstein /**
755dae1e52cSAmir Goldstein  *	ext4_find_shared - find the indirect blocks for partial truncation.
756dae1e52cSAmir Goldstein  *	@inode:	  inode in question
757dae1e52cSAmir Goldstein  *	@depth:	  depth of the affected branch
758dae1e52cSAmir Goldstein  *	@offsets: offsets of pointers in that branch (see ext4_block_to_path)
759dae1e52cSAmir Goldstein  *	@chain:	  place to store the pointers to partial indirect blocks
760dae1e52cSAmir Goldstein  *	@top:	  place to the (detached) top of branch
761dae1e52cSAmir Goldstein  *
762dae1e52cSAmir Goldstein  *	This is a helper function used by ext4_truncate().
763dae1e52cSAmir Goldstein  *
764dae1e52cSAmir Goldstein  *	When we do truncate() we may have to clean the ends of several
765dae1e52cSAmir Goldstein  *	indirect blocks but leave the blocks themselves alive. Block is
766dae1e52cSAmir Goldstein  *	partially truncated if some data below the new i_size is referred
767dae1e52cSAmir Goldstein  *	from it (and it is on the path to the first completely truncated
768dae1e52cSAmir Goldstein  *	data block, indeed).  We have to free the top of that path along
769dae1e52cSAmir Goldstein  *	with everything to the right of the path. Since no allocation
770dae1e52cSAmir Goldstein  *	past the truncation point is possible until ext4_truncate()
771dae1e52cSAmir Goldstein  *	finishes, we may safely do the latter, but top of branch may
772dae1e52cSAmir Goldstein  *	require special attention - pageout below the truncation point
773dae1e52cSAmir Goldstein  *	might try to populate it.
774dae1e52cSAmir Goldstein  *
775dae1e52cSAmir Goldstein  *	We atomically detach the top of branch from the tree, store the
776dae1e52cSAmir Goldstein  *	block number of its root in *@top, pointers to buffer_heads of
777dae1e52cSAmir Goldstein  *	partially truncated blocks - in @chain[].bh and pointers to
778dae1e52cSAmir Goldstein  *	their last elements that should not be removed - in
779dae1e52cSAmir Goldstein  *	@chain[].p. Return value is the pointer to last filled element
780dae1e52cSAmir Goldstein  *	of @chain.
781dae1e52cSAmir Goldstein  *
782dae1e52cSAmir Goldstein  *	The work left to caller to do the actual freeing of subtrees:
783dae1e52cSAmir Goldstein  *		a) free the subtree starting from *@top
784dae1e52cSAmir Goldstein  *		b) free the subtrees whose roots are stored in
785dae1e52cSAmir Goldstein  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
786dae1e52cSAmir Goldstein  *		c) free the subtrees growing from the inode past the @chain[0].
787dae1e52cSAmir Goldstein  *			(no partially truncated stuff there).  */
788dae1e52cSAmir Goldstein 
789dae1e52cSAmir Goldstein static Indirect *ext4_find_shared(struct inode *inode, int depth,
790dae1e52cSAmir Goldstein 				  ext4_lblk_t offsets[4], Indirect chain[4],
791dae1e52cSAmir Goldstein 				  __le32 *top)
792dae1e52cSAmir Goldstein {
793dae1e52cSAmir Goldstein 	Indirect *partial, *p;
794dae1e52cSAmir Goldstein 	int k, err;
795dae1e52cSAmir Goldstein 
796dae1e52cSAmir Goldstein 	*top = 0;
797dae1e52cSAmir Goldstein 	/* Make k index the deepest non-null offset + 1 */
798dae1e52cSAmir Goldstein 	for (k = depth; k > 1 && !offsets[k-1]; k--)
799dae1e52cSAmir Goldstein 		;
800dae1e52cSAmir Goldstein 	partial = ext4_get_branch(inode, k, offsets, chain, &err);
801dae1e52cSAmir Goldstein 	/* Writer: pointers */
802dae1e52cSAmir Goldstein 	if (!partial)
803dae1e52cSAmir Goldstein 		partial = chain + k-1;
804dae1e52cSAmir Goldstein 	/*
805dae1e52cSAmir Goldstein 	 * If the branch acquired continuation since we've looked at it -
806dae1e52cSAmir Goldstein 	 * fine, it should all survive and (new) top doesn't belong to us.
807dae1e52cSAmir Goldstein 	 */
808dae1e52cSAmir Goldstein 	if (!partial->key && *partial->p)
809dae1e52cSAmir Goldstein 		/* Writer: end */
810dae1e52cSAmir Goldstein 		goto no_top;
811dae1e52cSAmir Goldstein 	for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
812dae1e52cSAmir Goldstein 		;
813dae1e52cSAmir Goldstein 	/*
814dae1e52cSAmir Goldstein 	 * OK, we've found the last block that must survive. The rest of our
815dae1e52cSAmir Goldstein 	 * branch should be detached before unlocking. However, if that rest
816dae1e52cSAmir Goldstein 	 * of branch is all ours and does not grow immediately from the inode
817dae1e52cSAmir Goldstein 	 * it's easier to cheat and just decrement partial->p.
818dae1e52cSAmir Goldstein 	 */
819dae1e52cSAmir Goldstein 	if (p == chain + k - 1 && p > chain) {
820dae1e52cSAmir Goldstein 		p->p--;
821dae1e52cSAmir Goldstein 	} else {
822dae1e52cSAmir Goldstein 		*top = *p->p;
823dae1e52cSAmir Goldstein 		/* Nope, don't do this in ext4.  Must leave the tree intact */
824dae1e52cSAmir Goldstein #if 0
825dae1e52cSAmir Goldstein 		*p->p = 0;
826dae1e52cSAmir Goldstein #endif
827dae1e52cSAmir Goldstein 	}
828dae1e52cSAmir Goldstein 	/* Writer: end */
829dae1e52cSAmir Goldstein 
830dae1e52cSAmir Goldstein 	while (partial > p) {
831dae1e52cSAmir Goldstein 		brelse(partial->bh);
832dae1e52cSAmir Goldstein 		partial--;
833dae1e52cSAmir Goldstein 	}
834dae1e52cSAmir Goldstein no_top:
835dae1e52cSAmir Goldstein 	return partial;
836dae1e52cSAmir Goldstein }
837dae1e52cSAmir Goldstein 
838dae1e52cSAmir Goldstein /*
839dae1e52cSAmir Goldstein  * Zero a number of block pointers in either an inode or an indirect block.
840dae1e52cSAmir Goldstein  * If we restart the transaction we must again get write access to the
841dae1e52cSAmir Goldstein  * indirect block for further modification.
842dae1e52cSAmir Goldstein  *
843dae1e52cSAmir Goldstein  * We release `count' blocks on disk, but (last - first) may be greater
844dae1e52cSAmir Goldstein  * than `count' because there can be holes in there.
845dae1e52cSAmir Goldstein  *
846dae1e52cSAmir Goldstein  * Return 0 on success, 1 on invalid block range
847dae1e52cSAmir Goldstein  * and < 0 on fatal error.
848dae1e52cSAmir Goldstein  */
849dae1e52cSAmir Goldstein static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
850dae1e52cSAmir Goldstein 			     struct buffer_head *bh,
851dae1e52cSAmir Goldstein 			     ext4_fsblk_t block_to_free,
852dae1e52cSAmir Goldstein 			     unsigned long count, __le32 *first,
853dae1e52cSAmir Goldstein 			     __le32 *last)
854dae1e52cSAmir Goldstein {
855dae1e52cSAmir Goldstein 	__le32 *p;
856981250caSTheodore Ts'o 	int	flags = EXT4_FREE_BLOCKS_VALIDATED;
857dae1e52cSAmir Goldstein 	int	err;
858dae1e52cSAmir Goldstein 
859ddfa17e4STahsin Erdogan 	if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
860ddfa17e4STahsin Erdogan 	    ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
861981250caSTheodore Ts'o 		flags |= EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_METADATA;
862981250caSTheodore Ts'o 	else if (ext4_should_journal_data(inode))
863981250caSTheodore Ts'o 		flags |= EXT4_FREE_BLOCKS_FORGET;
864dae1e52cSAmir Goldstein 
865ce9f24ccSJan Kara 	if (!ext4_inode_block_valid(inode, block_to_free, count)) {
866dae1e52cSAmir Goldstein 		EXT4_ERROR_INODE(inode, "attempt to clear invalid "
867dae1e52cSAmir Goldstein 				 "blocks %llu len %lu",
868dae1e52cSAmir Goldstein 				 (unsigned long long) block_to_free, count);
869dae1e52cSAmir Goldstein 		return 1;
870dae1e52cSAmir Goldstein 	}
871dae1e52cSAmir Goldstein 
87283448bdfSJan Kara 	err = ext4_ind_truncate_ensure_credits(handle, inode, bh,
87383448bdfSJan Kara 				ext4_free_data_revoke_credits(inode, count));
874a4130367SJan Kara 	if (err < 0)
875dae1e52cSAmir Goldstein 		goto out_err;
876dae1e52cSAmir Goldstein 
877dae1e52cSAmir Goldstein 	for (p = first; p < last; p++)
878dae1e52cSAmir Goldstein 		*p = 0;
879dae1e52cSAmir Goldstein 
880dae1e52cSAmir Goldstein 	ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
881dae1e52cSAmir Goldstein 	return 0;
882dae1e52cSAmir Goldstein out_err:
883dae1e52cSAmir Goldstein 	ext4_std_error(inode->i_sb, err);
884dae1e52cSAmir Goldstein 	return err;
885dae1e52cSAmir Goldstein }
886dae1e52cSAmir Goldstein 
887dae1e52cSAmir Goldstein /**
888dae1e52cSAmir Goldstein  * ext4_free_data - free a list of data blocks
889dae1e52cSAmir Goldstein  * @handle:	handle for this transaction
890dae1e52cSAmir Goldstein  * @inode:	inode we are dealing with
891dae1e52cSAmir Goldstein  * @this_bh:	indirect buffer_head which contains *@first and *@last
892dae1e52cSAmir Goldstein  * @first:	array of block numbers
893dae1e52cSAmir Goldstein  * @last:	points immediately past the end of array
894dae1e52cSAmir Goldstein  *
895dae1e52cSAmir Goldstein  * We are freeing all blocks referred from that array (numbers are stored as
896dae1e52cSAmir Goldstein  * little-endian 32-bit) and updating @inode->i_blocks appropriately.
897dae1e52cSAmir Goldstein  *
898dae1e52cSAmir Goldstein  * We accumulate contiguous runs of blocks to free.  Conveniently, if these
899dae1e52cSAmir Goldstein  * blocks are contiguous then releasing them at one time will only affect one
900dae1e52cSAmir Goldstein  * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
901dae1e52cSAmir Goldstein  * actually use a lot of journal space.
902dae1e52cSAmir Goldstein  *
903dae1e52cSAmir Goldstein  * @this_bh will be %NULL if @first and @last point into the inode's direct
904dae1e52cSAmir Goldstein  * block pointers.
905dae1e52cSAmir Goldstein  */
906dae1e52cSAmir Goldstein static void ext4_free_data(handle_t *handle, struct inode *inode,
907dae1e52cSAmir Goldstein 			   struct buffer_head *this_bh,
908dae1e52cSAmir Goldstein 			   __le32 *first, __le32 *last)
909dae1e52cSAmir Goldstein {
910dae1e52cSAmir Goldstein 	ext4_fsblk_t block_to_free = 0;    /* Starting block # of a run */
911dae1e52cSAmir Goldstein 	unsigned long count = 0;	    /* Number of blocks in the run */
912dae1e52cSAmir Goldstein 	__le32 *block_to_free_p = NULL;	    /* Pointer into inode/ind
913dae1e52cSAmir Goldstein 					       corresponding to
914dae1e52cSAmir Goldstein 					       block_to_free */
915dae1e52cSAmir Goldstein 	ext4_fsblk_t nr;		    /* Current block # */
916dae1e52cSAmir Goldstein 	__le32 *p;			    /* Pointer into inode/ind
917dae1e52cSAmir Goldstein 					       for current block */
918dae1e52cSAmir Goldstein 	int err = 0;
919dae1e52cSAmir Goldstein 
920dae1e52cSAmir Goldstein 	if (this_bh) {				/* For indirect block */
921dae1e52cSAmir Goldstein 		BUFFER_TRACE(this_bh, "get_write_access");
922*188c299eSJan Kara 		err = ext4_journal_get_write_access(handle, inode->i_sb,
923*188c299eSJan Kara 						    this_bh, EXT4_JTR_NONE);
924dae1e52cSAmir Goldstein 		/* Important: if we can't update the indirect pointers
925dae1e52cSAmir Goldstein 		 * to the blocks, we can't free them. */
926dae1e52cSAmir Goldstein 		if (err)
927dae1e52cSAmir Goldstein 			return;
928dae1e52cSAmir Goldstein 	}
929dae1e52cSAmir Goldstein 
930dae1e52cSAmir Goldstein 	for (p = first; p < last; p++) {
931dae1e52cSAmir Goldstein 		nr = le32_to_cpu(*p);
932dae1e52cSAmir Goldstein 		if (nr) {
933dae1e52cSAmir Goldstein 			/* accumulate blocks to free if they're contiguous */
934dae1e52cSAmir Goldstein 			if (count == 0) {
935dae1e52cSAmir Goldstein 				block_to_free = nr;
936dae1e52cSAmir Goldstein 				block_to_free_p = p;
937dae1e52cSAmir Goldstein 				count = 1;
938dae1e52cSAmir Goldstein 			} else if (nr == block_to_free + count) {
939dae1e52cSAmir Goldstein 				count++;
940dae1e52cSAmir Goldstein 			} else {
941dae1e52cSAmir Goldstein 				err = ext4_clear_blocks(handle, inode, this_bh,
942dae1e52cSAmir Goldstein 						        block_to_free, count,
943dae1e52cSAmir Goldstein 						        block_to_free_p, p);
944dae1e52cSAmir Goldstein 				if (err)
945dae1e52cSAmir Goldstein 					break;
946dae1e52cSAmir Goldstein 				block_to_free = nr;
947dae1e52cSAmir Goldstein 				block_to_free_p = p;
948dae1e52cSAmir Goldstein 				count = 1;
949dae1e52cSAmir Goldstein 			}
950dae1e52cSAmir Goldstein 		}
951dae1e52cSAmir Goldstein 	}
952dae1e52cSAmir Goldstein 
953dae1e52cSAmir Goldstein 	if (!err && count > 0)
954dae1e52cSAmir Goldstein 		err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
955dae1e52cSAmir Goldstein 					count, block_to_free_p, p);
956dae1e52cSAmir Goldstein 	if (err < 0)
957dae1e52cSAmir Goldstein 		/* fatal error */
958dae1e52cSAmir Goldstein 		return;
959dae1e52cSAmir Goldstein 
960dae1e52cSAmir Goldstein 	if (this_bh) {
961dae1e52cSAmir Goldstein 		BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
962dae1e52cSAmir Goldstein 
963dae1e52cSAmir Goldstein 		/*
964dae1e52cSAmir Goldstein 		 * The buffer head should have an attached journal head at this
965dae1e52cSAmir Goldstein 		 * point. However, if the data is corrupted and an indirect
966dae1e52cSAmir Goldstein 		 * block pointed to itself, it would have been detached when
967dae1e52cSAmir Goldstein 		 * the block was cleared. Check for this instead of OOPSing.
968dae1e52cSAmir Goldstein 		 */
969dae1e52cSAmir Goldstein 		if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
970dae1e52cSAmir Goldstein 			ext4_handle_dirty_metadata(handle, inode, this_bh);
971dae1e52cSAmir Goldstein 		else
972dae1e52cSAmir Goldstein 			EXT4_ERROR_INODE(inode,
973dae1e52cSAmir Goldstein 					 "circular indirect block detected at "
974dae1e52cSAmir Goldstein 					 "block %llu",
975dae1e52cSAmir Goldstein 				(unsigned long long) this_bh->b_blocknr);
976dae1e52cSAmir Goldstein 	}
977dae1e52cSAmir Goldstein }
978dae1e52cSAmir Goldstein 
979dae1e52cSAmir Goldstein /**
980dae1e52cSAmir Goldstein  *	ext4_free_branches - free an array of branches
981dae1e52cSAmir Goldstein  *	@handle: JBD handle for this transaction
982dae1e52cSAmir Goldstein  *	@inode:	inode we are dealing with
983dae1e52cSAmir Goldstein  *	@parent_bh: the buffer_head which contains *@first and *@last
984dae1e52cSAmir Goldstein  *	@first:	array of block numbers
985dae1e52cSAmir Goldstein  *	@last:	pointer immediately past the end of array
986dae1e52cSAmir Goldstein  *	@depth:	depth of the branches to free
987dae1e52cSAmir Goldstein  *
988dae1e52cSAmir Goldstein  *	We are freeing all blocks referred from these branches (numbers are
989dae1e52cSAmir Goldstein  *	stored as little-endian 32-bit) and updating @inode->i_blocks
990dae1e52cSAmir Goldstein  *	appropriately.
991dae1e52cSAmir Goldstein  */
992dae1e52cSAmir Goldstein static void ext4_free_branches(handle_t *handle, struct inode *inode,
993dae1e52cSAmir Goldstein 			       struct buffer_head *parent_bh,
994dae1e52cSAmir Goldstein 			       __le32 *first, __le32 *last, int depth)
995dae1e52cSAmir Goldstein {
996dae1e52cSAmir Goldstein 	ext4_fsblk_t nr;
997dae1e52cSAmir Goldstein 	__le32 *p;
998dae1e52cSAmir Goldstein 
999dae1e52cSAmir Goldstein 	if (ext4_handle_is_aborted(handle))
1000dae1e52cSAmir Goldstein 		return;
1001dae1e52cSAmir Goldstein 
1002dae1e52cSAmir Goldstein 	if (depth--) {
1003dae1e52cSAmir Goldstein 		struct buffer_head *bh;
1004dae1e52cSAmir Goldstein 		int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1005dae1e52cSAmir Goldstein 		p = last;
1006dae1e52cSAmir Goldstein 		while (--p >= first) {
1007dae1e52cSAmir Goldstein 			nr = le32_to_cpu(*p);
1008dae1e52cSAmir Goldstein 			if (!nr)
1009dae1e52cSAmir Goldstein 				continue;		/* A hole */
1010dae1e52cSAmir Goldstein 
1011ce9f24ccSJan Kara 			if (!ext4_inode_block_valid(inode, nr, 1)) {
1012dae1e52cSAmir Goldstein 				EXT4_ERROR_INODE(inode,
1013dae1e52cSAmir Goldstein 						 "invalid indirect mapped "
1014dae1e52cSAmir Goldstein 						 "block %lu (level %d)",
1015dae1e52cSAmir Goldstein 						 (unsigned long) nr, depth);
1016dae1e52cSAmir Goldstein 				break;
1017dae1e52cSAmir Goldstein 			}
1018dae1e52cSAmir Goldstein 
1019dae1e52cSAmir Goldstein 			/* Go read the buffer for the next level down */
10200a846f49Szhangyi (F) 			bh = ext4_sb_bread(inode->i_sb, nr, 0);
1021dae1e52cSAmir Goldstein 
1022dae1e52cSAmir Goldstein 			/*
1023dae1e52cSAmir Goldstein 			 * A read failure? Report error and clear slot
1024dae1e52cSAmir Goldstein 			 * (should be rare).
1025dae1e52cSAmir Goldstein 			 */
10260a846f49Szhangyi (F) 			if (IS_ERR(bh)) {
10270a846f49Szhangyi (F) 				ext4_error_inode_block(inode, nr, -PTR_ERR(bh),
1028dae1e52cSAmir Goldstein 						       "Read failure");
1029dae1e52cSAmir Goldstein 				continue;
1030dae1e52cSAmir Goldstein 			}
1031dae1e52cSAmir Goldstein 
1032dae1e52cSAmir Goldstein 			/* This zaps the entire block.  Bottom up. */
1033dae1e52cSAmir Goldstein 			BUFFER_TRACE(bh, "free child branches");
1034dae1e52cSAmir Goldstein 			ext4_free_branches(handle, inode, bh,
1035dae1e52cSAmir Goldstein 					(__le32 *) bh->b_data,
1036dae1e52cSAmir Goldstein 					(__le32 *) bh->b_data + addr_per_block,
1037dae1e52cSAmir Goldstein 					depth);
1038dae1e52cSAmir Goldstein 			brelse(bh);
1039dae1e52cSAmir Goldstein 
1040dae1e52cSAmir Goldstein 			/*
1041b483bb77SRandy Dunlap 			 * Everything below this pointer has been
1042dae1e52cSAmir Goldstein 			 * released.  Now let this top-of-subtree go.
1043dae1e52cSAmir Goldstein 			 *
1044dae1e52cSAmir Goldstein 			 * We want the freeing of this indirect block to be
1045dae1e52cSAmir Goldstein 			 * atomic in the journal with the updating of the
1046dae1e52cSAmir Goldstein 			 * bitmap block which owns it.  So make some room in
1047dae1e52cSAmir Goldstein 			 * the journal.
1048dae1e52cSAmir Goldstein 			 *
1049dae1e52cSAmir Goldstein 			 * We zero the parent pointer *after* freeing its
1050dae1e52cSAmir Goldstein 			 * pointee in the bitmaps, so if extend_transaction()
1051dae1e52cSAmir Goldstein 			 * for some reason fails to put the bitmap changes and
1052dae1e52cSAmir Goldstein 			 * the release into the same transaction, recovery
1053dae1e52cSAmir Goldstein 			 * will merely complain about releasing a free block,
1054dae1e52cSAmir Goldstein 			 * rather than leaking blocks.
1055dae1e52cSAmir Goldstein 			 */
1056dae1e52cSAmir Goldstein 			if (ext4_handle_is_aborted(handle))
1057dae1e52cSAmir Goldstein 				return;
1058a4130367SJan Kara 			if (ext4_ind_truncate_ensure_credits(handle, inode,
105983448bdfSJan Kara 					NULL,
106083448bdfSJan Kara 					ext4_free_metadata_revoke_credits(
106183448bdfSJan Kara 							inode->i_sb, 1)) < 0)
1062a4130367SJan Kara 				return;
1063dae1e52cSAmir Goldstein 
1064dae1e52cSAmir Goldstein 			/*
1065dae1e52cSAmir Goldstein 			 * The forget flag here is critical because if
1066dae1e52cSAmir Goldstein 			 * we are journaling (and not doing data
1067dae1e52cSAmir Goldstein 			 * journaling), we have to make sure a revoke
1068dae1e52cSAmir Goldstein 			 * record is written to prevent the journal
1069dae1e52cSAmir Goldstein 			 * replay from overwriting the (former)
1070dae1e52cSAmir Goldstein 			 * indirect block if it gets reallocated as a
1071dae1e52cSAmir Goldstein 			 * data block.  This must happen in the same
1072dae1e52cSAmir Goldstein 			 * transaction where the data blocks are
1073dae1e52cSAmir Goldstein 			 * actually freed.
1074dae1e52cSAmir Goldstein 			 */
1075dae1e52cSAmir Goldstein 			ext4_free_blocks(handle, inode, NULL, nr, 1,
1076dae1e52cSAmir Goldstein 					 EXT4_FREE_BLOCKS_METADATA|
1077dae1e52cSAmir Goldstein 					 EXT4_FREE_BLOCKS_FORGET);
1078dae1e52cSAmir Goldstein 
1079dae1e52cSAmir Goldstein 			if (parent_bh) {
1080dae1e52cSAmir Goldstein 				/*
1081dae1e52cSAmir Goldstein 				 * The block which we have just freed is
1082dae1e52cSAmir Goldstein 				 * pointed to by an indirect block: journal it
1083dae1e52cSAmir Goldstein 				 */
1084dae1e52cSAmir Goldstein 				BUFFER_TRACE(parent_bh, "get_write_access");
1085dae1e52cSAmir Goldstein 				if (!ext4_journal_get_write_access(handle,
1086*188c299eSJan Kara 						inode->i_sb, parent_bh,
1087*188c299eSJan Kara 						EXT4_JTR_NONE)) {
1088dae1e52cSAmir Goldstein 					*p = 0;
1089dae1e52cSAmir Goldstein 					BUFFER_TRACE(parent_bh,
1090dae1e52cSAmir Goldstein 					"call ext4_handle_dirty_metadata");
1091dae1e52cSAmir Goldstein 					ext4_handle_dirty_metadata(handle,
1092dae1e52cSAmir Goldstein 								   inode,
1093dae1e52cSAmir Goldstein 								   parent_bh);
1094dae1e52cSAmir Goldstein 				}
1095dae1e52cSAmir Goldstein 			}
1096dae1e52cSAmir Goldstein 		}
1097dae1e52cSAmir Goldstein 	} else {
1098dae1e52cSAmir Goldstein 		/* We have reached the bottom of the tree. */
1099dae1e52cSAmir Goldstein 		BUFFER_TRACE(parent_bh, "free data blocks");
1100dae1e52cSAmir Goldstein 		ext4_free_data(handle, inode, parent_bh, first, last);
1101dae1e52cSAmir Goldstein 	}
1102dae1e52cSAmir Goldstein }
1103dae1e52cSAmir Goldstein 
1104819c4920STheodore Ts'o void ext4_ind_truncate(handle_t *handle, struct inode *inode)
1105dae1e52cSAmir Goldstein {
1106dae1e52cSAmir Goldstein 	struct ext4_inode_info *ei = EXT4_I(inode);
1107dae1e52cSAmir Goldstein 	__le32 *i_data = ei->i_data;
1108dae1e52cSAmir Goldstein 	int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1109dae1e52cSAmir Goldstein 	ext4_lblk_t offsets[4];
1110dae1e52cSAmir Goldstein 	Indirect chain[4];
1111dae1e52cSAmir Goldstein 	Indirect *partial;
1112dae1e52cSAmir Goldstein 	__le32 nr = 0;
1113dae1e52cSAmir Goldstein 	int n = 0;
1114dae1e52cSAmir Goldstein 	ext4_lblk_t last_block, max_block;
1115dae1e52cSAmir Goldstein 	unsigned blocksize = inode->i_sb->s_blocksize;
1116dae1e52cSAmir Goldstein 
1117dae1e52cSAmir Goldstein 	last_block = (inode->i_size + blocksize-1)
1118dae1e52cSAmir Goldstein 					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1119dae1e52cSAmir Goldstein 	max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1120dae1e52cSAmir Goldstein 					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1121dae1e52cSAmir Goldstein 
1122dae1e52cSAmir Goldstein 	if (last_block != max_block) {
1123dae1e52cSAmir Goldstein 		n = ext4_block_to_path(inode, last_block, offsets, NULL);
1124dae1e52cSAmir Goldstein 		if (n == 0)
1125819c4920STheodore Ts'o 			return;
1126dae1e52cSAmir Goldstein 	}
1127dae1e52cSAmir Goldstein 
112851865fdaSZheng Liu 	ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block);
1129dae1e52cSAmir Goldstein 
1130dae1e52cSAmir Goldstein 	/*
1131dae1e52cSAmir Goldstein 	 * The orphan list entry will now protect us from any crash which
1132dae1e52cSAmir Goldstein 	 * occurs before the truncate completes, so it is now safe to propagate
1133dae1e52cSAmir Goldstein 	 * the new, shorter inode size (held for now in i_size) into the
1134dae1e52cSAmir Goldstein 	 * on-disk inode. We do this via i_disksize, which is the value which
1135dae1e52cSAmir Goldstein 	 * ext4 *really* writes onto the disk inode.
1136dae1e52cSAmir Goldstein 	 */
1137dae1e52cSAmir Goldstein 	ei->i_disksize = inode->i_size;
1138dae1e52cSAmir Goldstein 
1139dae1e52cSAmir Goldstein 	if (last_block == max_block) {
1140dae1e52cSAmir Goldstein 		/*
1141dae1e52cSAmir Goldstein 		 * It is unnecessary to free any data blocks if last_block is
1142dae1e52cSAmir Goldstein 		 * equal to the indirect block limit.
1143dae1e52cSAmir Goldstein 		 */
1144819c4920STheodore Ts'o 		return;
1145dae1e52cSAmir Goldstein 	} else if (n == 1) {		/* direct blocks */
1146dae1e52cSAmir Goldstein 		ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1147dae1e52cSAmir Goldstein 			       i_data + EXT4_NDIR_BLOCKS);
1148dae1e52cSAmir Goldstein 		goto do_indirects;
1149dae1e52cSAmir Goldstein 	}
1150dae1e52cSAmir Goldstein 
1151dae1e52cSAmir Goldstein 	partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1152dae1e52cSAmir Goldstein 	/* Kill the top of shared branch (not detached) */
1153dae1e52cSAmir Goldstein 	if (nr) {
1154dae1e52cSAmir Goldstein 		if (partial == chain) {
1155dae1e52cSAmir Goldstein 			/* Shared branch grows from the inode */
1156dae1e52cSAmir Goldstein 			ext4_free_branches(handle, inode, NULL,
1157dae1e52cSAmir Goldstein 					   &nr, &nr+1, (chain+n-1) - partial);
1158dae1e52cSAmir Goldstein 			*partial->p = 0;
1159dae1e52cSAmir Goldstein 			/*
1160dae1e52cSAmir Goldstein 			 * We mark the inode dirty prior to restart,
1161dae1e52cSAmir Goldstein 			 * and prior to stop.  No need for it here.
1162dae1e52cSAmir Goldstein 			 */
1163dae1e52cSAmir Goldstein 		} else {
1164dae1e52cSAmir Goldstein 			/* Shared branch grows from an indirect block */
1165dae1e52cSAmir Goldstein 			BUFFER_TRACE(partial->bh, "get_write_access");
1166dae1e52cSAmir Goldstein 			ext4_free_branches(handle, inode, partial->bh,
1167dae1e52cSAmir Goldstein 					partial->p,
1168dae1e52cSAmir Goldstein 					partial->p+1, (chain+n-1) - partial);
1169dae1e52cSAmir Goldstein 		}
1170dae1e52cSAmir Goldstein 	}
1171dae1e52cSAmir Goldstein 	/* Clear the ends of indirect blocks on the shared branch */
1172dae1e52cSAmir Goldstein 	while (partial > chain) {
1173dae1e52cSAmir Goldstein 		ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1174dae1e52cSAmir Goldstein 				   (__le32*)partial->bh->b_data+addr_per_block,
1175dae1e52cSAmir Goldstein 				   (chain+n-1) - partial);
1176dae1e52cSAmir Goldstein 		BUFFER_TRACE(partial->bh, "call brelse");
1177dae1e52cSAmir Goldstein 		brelse(partial->bh);
1178dae1e52cSAmir Goldstein 		partial--;
1179dae1e52cSAmir Goldstein 	}
1180dae1e52cSAmir Goldstein do_indirects:
1181dae1e52cSAmir Goldstein 	/* Kill the remaining (whole) subtrees */
1182dae1e52cSAmir Goldstein 	switch (offsets[0]) {
1183dae1e52cSAmir Goldstein 	default:
1184dae1e52cSAmir Goldstein 		nr = i_data[EXT4_IND_BLOCK];
1185dae1e52cSAmir Goldstein 		if (nr) {
1186dae1e52cSAmir Goldstein 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1187dae1e52cSAmir Goldstein 			i_data[EXT4_IND_BLOCK] = 0;
1188dae1e52cSAmir Goldstein 		}
118970d7ced2SShijie Luo 		fallthrough;
1190dae1e52cSAmir Goldstein 	case EXT4_IND_BLOCK:
1191dae1e52cSAmir Goldstein 		nr = i_data[EXT4_DIND_BLOCK];
1192dae1e52cSAmir Goldstein 		if (nr) {
1193dae1e52cSAmir Goldstein 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1194dae1e52cSAmir Goldstein 			i_data[EXT4_DIND_BLOCK] = 0;
1195dae1e52cSAmir Goldstein 		}
119670d7ced2SShijie Luo 		fallthrough;
1197dae1e52cSAmir Goldstein 	case EXT4_DIND_BLOCK:
1198dae1e52cSAmir Goldstein 		nr = i_data[EXT4_TIND_BLOCK];
1199dae1e52cSAmir Goldstein 		if (nr) {
1200dae1e52cSAmir Goldstein 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1201dae1e52cSAmir Goldstein 			i_data[EXT4_TIND_BLOCK] = 0;
1202dae1e52cSAmir Goldstein 		}
120370d7ced2SShijie Luo 		fallthrough;
1204dae1e52cSAmir Goldstein 	case EXT4_TIND_BLOCK:
1205dae1e52cSAmir Goldstein 		;
1206dae1e52cSAmir Goldstein 	}
1207dae1e52cSAmir Goldstein }
1208dae1e52cSAmir Goldstein 
12094f579ae7SLukas Czerner /**
12104f579ae7SLukas Czerner  *	ext4_ind_remove_space - remove space from the range
12114f579ae7SLukas Czerner  *	@handle: JBD handle for this transaction
12124f579ae7SLukas Czerner  *	@inode:	inode we are dealing with
12134f579ae7SLukas Czerner  *	@start:	First block to remove
12144f579ae7SLukas Czerner  *	@end:	One block after the last block to remove (exclusive)
12154f579ae7SLukas Czerner  *
12164f579ae7SLukas Czerner  *	Free the blocks in the defined range (end is exclusive endpoint of
12174f579ae7SLukas Czerner  *	range). This is used by ext4_punch_hole().
12184f579ae7SLukas Czerner  */
12194f579ae7SLukas Czerner int ext4_ind_remove_space(handle_t *handle, struct inode *inode,
12204f579ae7SLukas Czerner 			  ext4_lblk_t start, ext4_lblk_t end)
12218bad6fc8SZheng Liu {
12224f579ae7SLukas Czerner 	struct ext4_inode_info *ei = EXT4_I(inode);
12234f579ae7SLukas Czerner 	__le32 *i_data = ei->i_data;
12248bad6fc8SZheng Liu 	int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
12254f579ae7SLukas Czerner 	ext4_lblk_t offsets[4], offsets2[4];
12264f579ae7SLukas Czerner 	Indirect chain[4], chain2[4];
12274f579ae7SLukas Czerner 	Indirect *partial, *partial2;
12285e86bddaSzhangyi (F) 	Indirect *p = NULL, *p2 = NULL;
12294f579ae7SLukas Czerner 	ext4_lblk_t max_block;
12304f579ae7SLukas Czerner 	__le32 nr = 0, nr2 = 0;
12314f579ae7SLukas Czerner 	int n = 0, n2 = 0;
12324f579ae7SLukas Czerner 	unsigned blocksize = inode->i_sb->s_blocksize;
12338bad6fc8SZheng Liu 
12344f579ae7SLukas Czerner 	max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
12354f579ae7SLukas Czerner 					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
12364f579ae7SLukas Czerner 	if (end >= max_block)
12374f579ae7SLukas Czerner 		end = max_block;
12384f579ae7SLukas Czerner 	if ((start >= end) || (start > max_block))
12394f579ae7SLukas Czerner 		return 0;
1240a93cd4cfSJan Kara 
12414f579ae7SLukas Czerner 	n = ext4_block_to_path(inode, start, offsets, NULL);
12424f579ae7SLukas Czerner 	n2 = ext4_block_to_path(inode, end, offsets2, NULL);
12434f579ae7SLukas Czerner 
12444f579ae7SLukas Czerner 	BUG_ON(n > n2);
12454f579ae7SLukas Czerner 
12464f579ae7SLukas Czerner 	if ((n == 1) && (n == n2)) {
12474f579ae7SLukas Czerner 		/* We're punching only within direct block range */
12484f579ae7SLukas Czerner 		ext4_free_data(handle, inode, NULL, i_data + offsets[0],
12494f579ae7SLukas Czerner 			       i_data + offsets2[0]);
12504f579ae7SLukas Czerner 		return 0;
12514f579ae7SLukas Czerner 	} else if (n2 > n) {
12524f579ae7SLukas Czerner 		/*
12534f579ae7SLukas Czerner 		 * Start and end are on a different levels so we're going to
12544f579ae7SLukas Czerner 		 * free partial block at start, and partial block at end of
12554f579ae7SLukas Czerner 		 * the range. If there are some levels in between then
12564f579ae7SLukas Czerner 		 * do_indirects label will take care of that.
12574f579ae7SLukas Czerner 		 */
12584f579ae7SLukas Czerner 
12594f579ae7SLukas Czerner 		if (n == 1) {
12604f579ae7SLukas Czerner 			/*
12614f579ae7SLukas Czerner 			 * Start is at the direct block level, free
12624f579ae7SLukas Czerner 			 * everything to the end of the level.
12634f579ae7SLukas Czerner 			 */
12644f579ae7SLukas Czerner 			ext4_free_data(handle, inode, NULL, i_data + offsets[0],
12654f579ae7SLukas Czerner 				       i_data + EXT4_NDIR_BLOCKS);
12664f579ae7SLukas Czerner 			goto end_range;
12678bad6fc8SZheng Liu 		}
12684f579ae7SLukas Czerner 
12694f579ae7SLukas Czerner 
12705e86bddaSzhangyi (F) 		partial = p = ext4_find_shared(inode, n, offsets, chain, &nr);
12714f579ae7SLukas Czerner 		if (nr) {
12724f579ae7SLukas Czerner 			if (partial == chain) {
12734f579ae7SLukas Czerner 				/* Shared branch grows from the inode */
12744f579ae7SLukas Czerner 				ext4_free_branches(handle, inode, NULL,
12754f579ae7SLukas Czerner 					   &nr, &nr+1, (chain+n-1) - partial);
12764f579ae7SLukas Czerner 				*partial->p = 0;
1277a93cd4cfSJan Kara 			} else {
12784f579ae7SLukas Czerner 				/* Shared branch grows from an indirect block */
12794f579ae7SLukas Czerner 				BUFFER_TRACE(partial->bh, "get_write_access");
12804f579ae7SLukas Czerner 				ext4_free_branches(handle, inode, partial->bh,
12814f579ae7SLukas Czerner 					partial->p,
12824f579ae7SLukas Czerner 					partial->p+1, (chain+n-1) - partial);
1283a93cd4cfSJan Kara 			}
12848bad6fc8SZheng Liu 		}
12858bad6fc8SZheng Liu 
12864f579ae7SLukas Czerner 		/*
12874f579ae7SLukas Czerner 		 * Clear the ends of indirect blocks on the shared branch
12884f579ae7SLukas Czerner 		 * at the start of the range
12894f579ae7SLukas Czerner 		 */
12904f579ae7SLukas Czerner 		while (partial > chain) {
12914f579ae7SLukas Czerner 			ext4_free_branches(handle, inode, partial->bh,
12924f579ae7SLukas Czerner 				partial->p + 1,
12934f579ae7SLukas Czerner 				(__le32 *)partial->bh->b_data+addr_per_block,
12944f579ae7SLukas Czerner 				(chain+n-1) - partial);
12954f579ae7SLukas Czerner 			partial--;
12968bad6fc8SZheng Liu 		}
12978bad6fc8SZheng Liu 
12984f579ae7SLukas Czerner end_range:
12995e86bddaSzhangyi (F) 		partial2 = p2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
13004f579ae7SLukas Czerner 		if (nr2) {
13014f579ae7SLukas Czerner 			if (partial2 == chain2) {
13024f579ae7SLukas Czerner 				/*
13034f579ae7SLukas Czerner 				 * Remember, end is exclusive so here we're at
13044f579ae7SLukas Czerner 				 * the start of the next level we're not going
13054f579ae7SLukas Czerner 				 * to free. Everything was covered by the start
13064f579ae7SLukas Czerner 				 * of the range.
13074f579ae7SLukas Czerner 				 */
13086f30b7e3SOmar Sandoval 				goto do_indirects;
13098bad6fc8SZheng Liu 			}
13104f579ae7SLukas Czerner 		} else {
13114f579ae7SLukas Czerner 			/*
13124f579ae7SLukas Czerner 			 * ext4_find_shared returns Indirect structure which
13134f579ae7SLukas Czerner 			 * points to the last element which should not be
13144f579ae7SLukas Czerner 			 * removed by truncate. But this is end of the range
13154f579ae7SLukas Czerner 			 * in punch_hole so we need to point to the next element
13164f579ae7SLukas Czerner 			 */
13174f579ae7SLukas Czerner 			partial2->p++;
13184f579ae7SLukas Czerner 		}
13194f579ae7SLukas Czerner 
13204f579ae7SLukas Czerner 		/*
13214f579ae7SLukas Czerner 		 * Clear the ends of indirect blocks on the shared branch
13224f579ae7SLukas Czerner 		 * at the end of the range
13234f579ae7SLukas Czerner 		 */
13244f579ae7SLukas Czerner 		while (partial2 > chain2) {
13254f579ae7SLukas Czerner 			ext4_free_branches(handle, inode, partial2->bh,
13264f579ae7SLukas Czerner 					   (__le32 *)partial2->bh->b_data,
13274f579ae7SLukas Czerner 					   partial2->p,
13284f579ae7SLukas Czerner 					   (chain2+n2-1) - partial2);
13294f579ae7SLukas Czerner 			partial2--;
13304f579ae7SLukas Czerner 		}
13314f579ae7SLukas Czerner 		goto do_indirects;
13324f579ae7SLukas Czerner 	}
13334f579ae7SLukas Czerner 
13344f579ae7SLukas Czerner 	/* Punch happened within the same level (n == n2) */
13355e86bddaSzhangyi (F) 	partial = p = ext4_find_shared(inode, n, offsets, chain, &nr);
13365e86bddaSzhangyi (F) 	partial2 = p2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
13376f30b7e3SOmar Sandoval 
13386f30b7e3SOmar Sandoval 	/* Free top, but only if partial2 isn't its subtree. */
13396f30b7e3SOmar Sandoval 	if (nr) {
13406f30b7e3SOmar Sandoval 		int level = min(partial - chain, partial2 - chain2);
13416f30b7e3SOmar Sandoval 		int i;
13426f30b7e3SOmar Sandoval 		int subtree = 1;
13436f30b7e3SOmar Sandoval 
13446f30b7e3SOmar Sandoval 		for (i = 0; i <= level; i++) {
13456f30b7e3SOmar Sandoval 			if (offsets[i] != offsets2[i]) {
13466f30b7e3SOmar Sandoval 				subtree = 0;
13476f30b7e3SOmar Sandoval 				break;
13486f30b7e3SOmar Sandoval 			}
13496f30b7e3SOmar Sandoval 		}
13506f30b7e3SOmar Sandoval 
13516f30b7e3SOmar Sandoval 		if (!subtree) {
13526f30b7e3SOmar Sandoval 			if (partial == chain) {
13536f30b7e3SOmar Sandoval 				/* Shared branch grows from the inode */
13546f30b7e3SOmar Sandoval 				ext4_free_branches(handle, inode, NULL,
13556f30b7e3SOmar Sandoval 						   &nr, &nr+1,
13566f30b7e3SOmar Sandoval 						   (chain+n-1) - partial);
13576f30b7e3SOmar Sandoval 				*partial->p = 0;
13586f30b7e3SOmar Sandoval 			} else {
13596f30b7e3SOmar Sandoval 				/* Shared branch grows from an indirect block */
13606f30b7e3SOmar Sandoval 				BUFFER_TRACE(partial->bh, "get_write_access");
13616f30b7e3SOmar Sandoval 				ext4_free_branches(handle, inode, partial->bh,
13626f30b7e3SOmar Sandoval 						   partial->p,
13636f30b7e3SOmar Sandoval 						   partial->p+1,
13646f30b7e3SOmar Sandoval 						   (chain+n-1) - partial);
13656f30b7e3SOmar Sandoval 			}
13666f30b7e3SOmar Sandoval 		}
13676f30b7e3SOmar Sandoval 	}
13686f30b7e3SOmar Sandoval 
13696f30b7e3SOmar Sandoval 	if (!nr2) {
13704f579ae7SLukas Czerner 		/*
13714f579ae7SLukas Czerner 		 * ext4_find_shared returns Indirect structure which
13724f579ae7SLukas Czerner 		 * points to the last element which should not be
13734f579ae7SLukas Czerner 		 * removed by truncate. But this is end of the range
13744f579ae7SLukas Czerner 		 * in punch_hole so we need to point to the next element
13754f579ae7SLukas Czerner 		 */
13764f579ae7SLukas Czerner 		partial2->p++;
13776f30b7e3SOmar Sandoval 	}
13786f30b7e3SOmar Sandoval 
13796f30b7e3SOmar Sandoval 	while (partial > chain || partial2 > chain2) {
13806f30b7e3SOmar Sandoval 		int depth = (chain+n-1) - partial;
13816f30b7e3SOmar Sandoval 		int depth2 = (chain2+n2-1) - partial2;
13826f30b7e3SOmar Sandoval 
13836f30b7e3SOmar Sandoval 		if (partial > chain && partial2 > chain2 &&
13846f30b7e3SOmar Sandoval 		    partial->bh->b_blocknr == partial2->bh->b_blocknr) {
13856f30b7e3SOmar Sandoval 			/*
13866f30b7e3SOmar Sandoval 			 * We've converged on the same block. Clear the range,
13876f30b7e3SOmar Sandoval 			 * then we're done.
13886f30b7e3SOmar Sandoval 			 */
13894f579ae7SLukas Czerner 			ext4_free_branches(handle, inode, partial->bh,
13904f579ae7SLukas Czerner 					   partial->p + 1,
13914f579ae7SLukas Czerner 					   partial2->p,
13924f579ae7SLukas Czerner 					   (chain+n-1) - partial);
13935e86bddaSzhangyi (F) 			goto cleanup;
13944f579ae7SLukas Czerner 		}
13956f30b7e3SOmar Sandoval 
13964f579ae7SLukas Czerner 		/*
13976f30b7e3SOmar Sandoval 		 * The start and end partial branches may not be at the same
13986f30b7e3SOmar Sandoval 		 * level even though the punch happened within one level. So, we
13996f30b7e3SOmar Sandoval 		 * give them a chance to arrive at the same level, then walk
14006f30b7e3SOmar Sandoval 		 * them in step with each other until we converge on the same
14016f30b7e3SOmar Sandoval 		 * block.
14024f579ae7SLukas Czerner 		 */
14036f30b7e3SOmar Sandoval 		if (partial > chain && depth <= depth2) {
14044f579ae7SLukas Czerner 			ext4_free_branches(handle, inode, partial->bh,
14054f579ae7SLukas Czerner 					   partial->p + 1,
14064f579ae7SLukas Czerner 					   (__le32 *)partial->bh->b_data+addr_per_block,
14074f579ae7SLukas Czerner 					   (chain+n-1) - partial);
14084f579ae7SLukas Czerner 			partial--;
14094f579ae7SLukas Czerner 		}
14106f30b7e3SOmar Sandoval 		if (partial2 > chain2 && depth2 <= depth) {
14114f579ae7SLukas Czerner 			ext4_free_branches(handle, inode, partial2->bh,
14124f579ae7SLukas Czerner 					   (__le32 *)partial2->bh->b_data,
14134f579ae7SLukas Czerner 					   partial2->p,
14146f30b7e3SOmar Sandoval 					   (chain2+n2-1) - partial2);
14154f579ae7SLukas Czerner 			partial2--;
14168bad6fc8SZheng Liu 		}
14178bad6fc8SZheng Liu 	}
14185e86bddaSzhangyi (F) 
14195e86bddaSzhangyi (F) cleanup:
14205e86bddaSzhangyi (F) 	while (p && p > chain) {
14215e86bddaSzhangyi (F) 		BUFFER_TRACE(p->bh, "call brelse");
14225e86bddaSzhangyi (F) 		brelse(p->bh);
14235e86bddaSzhangyi (F) 		p--;
14245e86bddaSzhangyi (F) 	}
14255e86bddaSzhangyi (F) 	while (p2 && p2 > chain2) {
14265e86bddaSzhangyi (F) 		BUFFER_TRACE(p2->bh, "call brelse");
14275e86bddaSzhangyi (F) 		brelse(p2->bh);
14285e86bddaSzhangyi (F) 		p2--;
14295e86bddaSzhangyi (F) 	}
14306f30b7e3SOmar Sandoval 	return 0;
14318bad6fc8SZheng Liu 
14324f579ae7SLukas Czerner do_indirects:
14334f579ae7SLukas Czerner 	/* Kill the remaining (whole) subtrees */
14344f579ae7SLukas Czerner 	switch (offsets[0]) {
14354f579ae7SLukas Czerner 	default:
14364f579ae7SLukas Czerner 		if (++n >= n2)
14375e86bddaSzhangyi (F) 			break;
14384f579ae7SLukas Czerner 		nr = i_data[EXT4_IND_BLOCK];
14394f579ae7SLukas Czerner 		if (nr) {
14404f579ae7SLukas Czerner 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
14414f579ae7SLukas Czerner 			i_data[EXT4_IND_BLOCK] = 0;
14428bad6fc8SZheng Liu 		}
144370d7ced2SShijie Luo 		fallthrough;
14444f579ae7SLukas Czerner 	case EXT4_IND_BLOCK:
14454f579ae7SLukas Czerner 		if (++n >= n2)
14465e86bddaSzhangyi (F) 			break;
14474f579ae7SLukas Czerner 		nr = i_data[EXT4_DIND_BLOCK];
14484f579ae7SLukas Czerner 		if (nr) {
14494f579ae7SLukas Czerner 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
14504f579ae7SLukas Czerner 			i_data[EXT4_DIND_BLOCK] = 0;
14514f579ae7SLukas Czerner 		}
145270d7ced2SShijie Luo 		fallthrough;
14534f579ae7SLukas Czerner 	case EXT4_DIND_BLOCK:
14544f579ae7SLukas Czerner 		if (++n >= n2)
14555e86bddaSzhangyi (F) 			break;
14564f579ae7SLukas Czerner 		nr = i_data[EXT4_TIND_BLOCK];
14574f579ae7SLukas Czerner 		if (nr) {
14584f579ae7SLukas Czerner 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
14594f579ae7SLukas Czerner 			i_data[EXT4_TIND_BLOCK] = 0;
14604f579ae7SLukas Czerner 		}
146170d7ced2SShijie Luo 		fallthrough;
14624f579ae7SLukas Czerner 	case EXT4_TIND_BLOCK:
14634f579ae7SLukas Czerner 		;
14644f579ae7SLukas Czerner 	}
14655e86bddaSzhangyi (F) 	goto cleanup;
14664f579ae7SLukas Czerner }
1467