xref: /openbmc/linux/fs/ext4/indirect.c (revision de25d6e9)
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 
add_chain(Indirect * p,struct buffer_head * bh,__le32 * v)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 
ext4_block_to_path(struct inode * inode,ext4_lblk_t i_block,ext4_lblk_t offsets[4],int * boundary)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  */
ext4_get_branch(struct inode * inode,int depth,ext4_lblk_t * offsets,Indirect chain[4],int * err)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;
15126d75a16SLuís Henriques 	unsigned int key;
152860d21e2STheodore Ts'o 	int ret = -EIO;
153dae1e52cSAmir Goldstein 
154dae1e52cSAmir Goldstein 	*err = 0;
155dae1e52cSAmir Goldstein 	/* i_data is not going away, no lock needed */
156dae1e52cSAmir Goldstein 	add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
157dae1e52cSAmir Goldstein 	if (!p->key)
158dae1e52cSAmir Goldstein 		goto no_block;
159dae1e52cSAmir Goldstein 	while (--depth) {
16026d75a16SLuís Henriques 		key = le32_to_cpu(p->key);
16126d75a16SLuís Henriques 		if (key > ext4_blocks_count(EXT4_SB(sb)->s_es)) {
16226d75a16SLuís Henriques 			/* the block was out of range */
16326d75a16SLuís Henriques 			ret = -EFSCORRUPTED;
16426d75a16SLuís Henriques 			goto failure;
16526d75a16SLuís Henriques 		}
16626d75a16SLuís Henriques 		bh = sb_getblk(sb, key);
167860d21e2STheodore Ts'o 		if (unlikely(!bh)) {
168860d21e2STheodore Ts'o 			ret = -ENOMEM;
169dae1e52cSAmir Goldstein 			goto failure;
170860d21e2STheodore Ts'o 		}
171dae1e52cSAmir Goldstein 
172dae1e52cSAmir Goldstein 		if (!bh_uptodate_or_lock(bh)) {
1732d069c08Szhangyi (F) 			if (ext4_read_bh(bh, 0, NULL) < 0) {
174dae1e52cSAmir Goldstein 				put_bh(bh);
175dae1e52cSAmir Goldstein 				goto failure;
176dae1e52cSAmir Goldstein 			}
177dae1e52cSAmir Goldstein 			/* validate block references */
178dae1e52cSAmir Goldstein 			if (ext4_check_indirect_blockref(inode, bh)) {
179dae1e52cSAmir Goldstein 				put_bh(bh);
180dae1e52cSAmir Goldstein 				goto failure;
181dae1e52cSAmir Goldstein 			}
182dae1e52cSAmir Goldstein 		}
183dae1e52cSAmir Goldstein 
184dae1e52cSAmir Goldstein 		add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
185dae1e52cSAmir Goldstein 		/* Reader: end */
186dae1e52cSAmir Goldstein 		if (!p->key)
187dae1e52cSAmir Goldstein 			goto no_block;
188dae1e52cSAmir Goldstein 	}
189dae1e52cSAmir Goldstein 	return NULL;
190dae1e52cSAmir Goldstein 
191dae1e52cSAmir Goldstein failure:
192860d21e2STheodore Ts'o 	*err = ret;
193dae1e52cSAmir Goldstein no_block:
194dae1e52cSAmir Goldstein 	return p;
195dae1e52cSAmir Goldstein }
196dae1e52cSAmir Goldstein 
197dae1e52cSAmir Goldstein /**
198dae1e52cSAmir Goldstein  *	ext4_find_near - find a place for allocation with sufficient locality
199dae1e52cSAmir Goldstein  *	@inode: owner
200dae1e52cSAmir Goldstein  *	@ind: descriptor of indirect block.
201dae1e52cSAmir Goldstein  *
202dae1e52cSAmir Goldstein  *	This function returns the preferred place for block allocation.
203dae1e52cSAmir Goldstein  *	It is used when heuristic for sequential allocation fails.
204dae1e52cSAmir Goldstein  *	Rules are:
205dae1e52cSAmir Goldstein  *	  + if there is a block to the left of our position - allocate near it.
206dae1e52cSAmir Goldstein  *	  + if pointer will live in indirect block - allocate near that block.
207dae1e52cSAmir Goldstein  *	  + if pointer will live in inode - allocate in the same
208dae1e52cSAmir Goldstein  *	    cylinder group.
209dae1e52cSAmir Goldstein  *
210dae1e52cSAmir Goldstein  * In the latter case we colour the starting block by the callers PID to
211dae1e52cSAmir Goldstein  * prevent it from clashing with concurrent allocations for a different inode
212dae1e52cSAmir Goldstein  * in the same block group.   The PID is used here so that functionally related
213dae1e52cSAmir Goldstein  * files will be close-by on-disk.
214dae1e52cSAmir Goldstein  *
215dae1e52cSAmir Goldstein  *	Caller must make sure that @ind is valid and will stay that way.
216dae1e52cSAmir Goldstein  */
ext4_find_near(struct inode * inode,Indirect * ind)217dae1e52cSAmir Goldstein static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
218dae1e52cSAmir Goldstein {
219dae1e52cSAmir Goldstein 	struct ext4_inode_info *ei = EXT4_I(inode);
220dae1e52cSAmir Goldstein 	__le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
221dae1e52cSAmir Goldstein 	__le32 *p;
222dae1e52cSAmir Goldstein 
223dae1e52cSAmir Goldstein 	/* Try to find previous block */
224dae1e52cSAmir Goldstein 	for (p = ind->p - 1; p >= start; p--) {
225dae1e52cSAmir Goldstein 		if (*p)
226dae1e52cSAmir Goldstein 			return le32_to_cpu(*p);
227dae1e52cSAmir Goldstein 	}
228dae1e52cSAmir Goldstein 
229dae1e52cSAmir Goldstein 	/* No such thing, so let's try location of indirect block */
230dae1e52cSAmir Goldstein 	if (ind->bh)
231dae1e52cSAmir Goldstein 		return ind->bh->b_blocknr;
232dae1e52cSAmir Goldstein 
233dae1e52cSAmir Goldstein 	/*
234dae1e52cSAmir Goldstein 	 * It is going to be referred to from the inode itself? OK, just put it
235dae1e52cSAmir Goldstein 	 * into the same cylinder group then.
236dae1e52cSAmir Goldstein 	 */
237f86186b4SEric Sandeen 	return ext4_inode_to_goal_block(inode);
238dae1e52cSAmir Goldstein }
239dae1e52cSAmir Goldstein 
240dae1e52cSAmir Goldstein /**
241dae1e52cSAmir Goldstein  *	ext4_find_goal - find a preferred place for allocation.
242dae1e52cSAmir Goldstein  *	@inode: owner
243dae1e52cSAmir Goldstein  *	@block:  block we want
244dae1e52cSAmir Goldstein  *	@partial: pointer to the last triple within a chain
245dae1e52cSAmir Goldstein  *
246dae1e52cSAmir Goldstein  *	Normally this function find the preferred place for block allocation,
247dae1e52cSAmir Goldstein  *	returns it.
248dae1e52cSAmir Goldstein  *	Because this is only used for non-extent files, we limit the block nr
249dae1e52cSAmir Goldstein  *	to 32 bits.
250dae1e52cSAmir Goldstein  */
ext4_find_goal(struct inode * inode,ext4_lblk_t block,Indirect * partial)251dae1e52cSAmir Goldstein static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
252dae1e52cSAmir Goldstein 				   Indirect *partial)
253dae1e52cSAmir Goldstein {
254dae1e52cSAmir Goldstein 	ext4_fsblk_t goal;
255dae1e52cSAmir Goldstein 
256dae1e52cSAmir Goldstein 	/*
257dae1e52cSAmir Goldstein 	 * XXX need to get goal block from mballoc's data structures
258dae1e52cSAmir Goldstein 	 */
259dae1e52cSAmir Goldstein 
260dae1e52cSAmir Goldstein 	goal = ext4_find_near(inode, partial);
261dae1e52cSAmir Goldstein 	goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
262dae1e52cSAmir Goldstein 	return goal;
263dae1e52cSAmir Goldstein }
264dae1e52cSAmir Goldstein 
265dae1e52cSAmir Goldstein /**
266dae1e52cSAmir Goldstein  *	ext4_blks_to_allocate - Look up the block map and count the number
267dae1e52cSAmir Goldstein  *	of direct blocks need to be allocated for the given branch.
268dae1e52cSAmir Goldstein  *
269dae1e52cSAmir Goldstein  *	@branch: chain of indirect blocks
270dae1e52cSAmir Goldstein  *	@k: number of blocks need for indirect blocks
271dae1e52cSAmir Goldstein  *	@blks: number of data blocks to be mapped.
272dae1e52cSAmir Goldstein  *	@blocks_to_boundary:  the offset in the indirect block
273dae1e52cSAmir Goldstein  *
274dae1e52cSAmir Goldstein  *	return the total number of blocks to be allocate, including the
275dae1e52cSAmir Goldstein  *	direct and indirect blocks.
276dae1e52cSAmir Goldstein  */
ext4_blks_to_allocate(Indirect * branch,int k,unsigned int blks,int blocks_to_boundary)277dae1e52cSAmir Goldstein static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
278dae1e52cSAmir Goldstein 				 int blocks_to_boundary)
279dae1e52cSAmir Goldstein {
280dae1e52cSAmir Goldstein 	unsigned int count = 0;
281dae1e52cSAmir Goldstein 
282dae1e52cSAmir Goldstein 	/*
283dae1e52cSAmir Goldstein 	 * Simple case, [t,d]Indirect block(s) has not allocated yet
284dae1e52cSAmir Goldstein 	 * then it's clear blocks on that path have not allocated
285dae1e52cSAmir Goldstein 	 */
286dae1e52cSAmir Goldstein 	if (k > 0) {
287dae1e52cSAmir Goldstein 		/* right now we don't handle cross boundary allocation */
288dae1e52cSAmir Goldstein 		if (blks < blocks_to_boundary + 1)
289dae1e52cSAmir Goldstein 			count += blks;
290dae1e52cSAmir Goldstein 		else
291dae1e52cSAmir Goldstein 			count += blocks_to_boundary + 1;
292dae1e52cSAmir Goldstein 		return count;
293dae1e52cSAmir Goldstein 	}
294dae1e52cSAmir Goldstein 
295dae1e52cSAmir Goldstein 	count++;
296dae1e52cSAmir Goldstein 	while (count < blks && count <= blocks_to_boundary &&
297dae1e52cSAmir Goldstein 		le32_to_cpu(*(branch[0].p + count)) == 0) {
298dae1e52cSAmir Goldstein 		count++;
299dae1e52cSAmir Goldstein 	}
300dae1e52cSAmir Goldstein 	return count;
301dae1e52cSAmir Goldstein }
302dae1e52cSAmir Goldstein 
303dae1e52cSAmir Goldstein /**
304c60990b3STheodore Ts'o  * ext4_alloc_branch() - allocate and set up a chain of blocks
305dae1e52cSAmir Goldstein  * @handle: handle for this transaction
306c60990b3STheodore Ts'o  * @ar: structure describing the allocation request
307dae1e52cSAmir Goldstein  * @indirect_blks: number of allocated indirect blocks
308dae1e52cSAmir Goldstein  * @offsets: offsets (in the blocks) to store the pointers to next.
309dae1e52cSAmir Goldstein  * @branch: place to store the chain in.
310dae1e52cSAmir Goldstein  *
311dae1e52cSAmir Goldstein  *	This function allocates blocks, zeroes out all but the last one,
312dae1e52cSAmir Goldstein  *	links them into chain and (if we are synchronous) writes them to disk.
313dae1e52cSAmir Goldstein  *	In other words, it prepares a branch that can be spliced onto the
314dae1e52cSAmir Goldstein  *	inode. It stores the information about that chain in the branch[], in
315dae1e52cSAmir Goldstein  *	the same format as ext4_get_branch() would do. We are calling it after
316dae1e52cSAmir Goldstein  *	we had read the existing part of chain and partial points to the last
317dae1e52cSAmir Goldstein  *	triple of that (one with zero ->key). Upon the exit we have the same
318dae1e52cSAmir Goldstein  *	picture as after the successful ext4_get_block(), except that in one
319dae1e52cSAmir Goldstein  *	place chain is disconnected - *branch->p is still zero (we did not
320dae1e52cSAmir Goldstein  *	set the last link), but branch->key contains the number that should
321dae1e52cSAmir Goldstein  *	be placed into *branch->p to fill that gap.
322dae1e52cSAmir Goldstein  *
323dae1e52cSAmir Goldstein  *	If allocation fails we free all blocks we've allocated (and forget
324dae1e52cSAmir Goldstein  *	their buffer_heads) and return the error value the from failed
325dae1e52cSAmir Goldstein  *	ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
326dae1e52cSAmir Goldstein  *	as described above and return 0.
327dae1e52cSAmir Goldstein  */
ext4_alloc_branch(handle_t * handle,struct ext4_allocation_request * ar,int indirect_blks,ext4_lblk_t * offsets,Indirect * branch)328a5211002STheodore Ts'o static int ext4_alloc_branch(handle_t *handle,
329a5211002STheodore Ts'o 			     struct ext4_allocation_request *ar,
330a5211002STheodore Ts'o 			     int indirect_blks, ext4_lblk_t *offsets,
331a5211002STheodore Ts'o 			     Indirect *branch)
332dae1e52cSAmir Goldstein {
333dae1e52cSAmir Goldstein 	struct buffer_head *		bh;
334781f143eSTheodore Ts'o 	ext4_fsblk_t			b, new_blocks[4];
335781f143eSTheodore Ts'o 	__le32				*p;
336781f143eSTheodore Ts'o 	int				i, j, err, len = 1;
337dae1e52cSAmir Goldstein 
338781f143eSTheodore Ts'o 	for (i = 0; i <= indirect_blks; i++) {
339781f143eSTheodore Ts'o 		if (i == indirect_blks) {
340a5211002STheodore Ts'o 			new_blocks[i] = ext4_mb_new_blocks(handle, ar, &err);
341f2890730SJan Kara 		} else {
342a5211002STheodore Ts'o 			ar->goal = new_blocks[i] = ext4_new_meta_blocks(handle,
343e3cf5d5dSTheodore Ts'o 					ar->inode, ar->goal,
344e3cf5d5dSTheodore Ts'o 					ar->flags & EXT4_MB_DELALLOC_RESERVED,
345e3cf5d5dSTheodore Ts'o 					NULL, &err);
346f2890730SJan Kara 			/* Simplify error cleanup... */
347f2890730SJan Kara 			branch[i+1].bh = NULL;
348f2890730SJan Kara 		}
349781f143eSTheodore Ts'o 		if (err) {
350781f143eSTheodore Ts'o 			i--;
351781f143eSTheodore Ts'o 			goto failed;
352781f143eSTheodore Ts'o 		}
353781f143eSTheodore Ts'o 		branch[i].key = cpu_to_le32(new_blocks[i]);
354781f143eSTheodore Ts'o 		if (i == 0)
355781f143eSTheodore Ts'o 			continue;
356781f143eSTheodore Ts'o 
357a5211002STheodore Ts'o 		bh = branch[i].bh = sb_getblk(ar->inode->i_sb, new_blocks[i-1]);
358dae1e52cSAmir Goldstein 		if (unlikely(!bh)) {
359860d21e2STheodore Ts'o 			err = -ENOMEM;
360dae1e52cSAmir Goldstein 			goto failed;
361dae1e52cSAmir Goldstein 		}
362dae1e52cSAmir Goldstein 		lock_buffer(bh);
363dae1e52cSAmir Goldstein 		BUFFER_TRACE(bh, "call get_create_access");
364188c299eSJan Kara 		err = ext4_journal_get_create_access(handle, ar->inode->i_sb,
365188c299eSJan Kara 						     bh, EXT4_JTR_NONE);
366dae1e52cSAmir Goldstein 		if (err) {
367dae1e52cSAmir Goldstein 			unlock_buffer(bh);
368dae1e52cSAmir Goldstein 			goto failed;
369dae1e52cSAmir Goldstein 		}
370dae1e52cSAmir Goldstein 
371781f143eSTheodore Ts'o 		memset(bh->b_data, 0, bh->b_size);
372781f143eSTheodore Ts'o 		p = branch[i].p = (__le32 *) bh->b_data + offsets[i];
373781f143eSTheodore Ts'o 		b = new_blocks[i];
374781f143eSTheodore Ts'o 
375781f143eSTheodore Ts'o 		if (i == indirect_blks)
376a5211002STheodore Ts'o 			len = ar->len;
377781f143eSTheodore Ts'o 		for (j = 0; j < len; j++)
378781f143eSTheodore Ts'o 			*p++ = cpu_to_le32(b++);
379781f143eSTheodore Ts'o 
380dae1e52cSAmir Goldstein 		BUFFER_TRACE(bh, "marking uptodate");
381dae1e52cSAmir Goldstein 		set_buffer_uptodate(bh);
382dae1e52cSAmir Goldstein 		unlock_buffer(bh);
383dae1e52cSAmir Goldstein 
384dae1e52cSAmir Goldstein 		BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
385a5211002STheodore Ts'o 		err = ext4_handle_dirty_metadata(handle, ar->inode, bh);
386dae1e52cSAmir Goldstein 		if (err)
387dae1e52cSAmir Goldstein 			goto failed;
388dae1e52cSAmir Goldstein 	}
389781f143eSTheodore Ts'o 	return 0;
390dae1e52cSAmir Goldstein failed:
391f2890730SJan Kara 	if (i == indirect_blks) {
392f2890730SJan Kara 		/* Free data blocks */
393f2890730SJan Kara 		ext4_free_blocks(handle, ar->inode, NULL, new_blocks[i],
394f2890730SJan Kara 				 ar->len, 0);
395f2890730SJan Kara 		i--;
396f2890730SJan Kara 	}
397781f143eSTheodore Ts'o 	for (; i >= 0; i--) {
398c5c7b8ddSJan Kara 		/*
399c5c7b8ddSJan Kara 		 * We want to ext4_forget() only freshly allocated indirect
400f2890730SJan Kara 		 * blocks. Buffer for new_blocks[i] is at branch[i+1].bh
401f2890730SJan Kara 		 * (buffer at branch[0].bh is indirect block / inode already
402f2890730SJan Kara 		 * existing before ext4_alloc_branch() was called). Also
403f2890730SJan Kara 		 * because blocks are freshly allocated, we don't need to
404f2890730SJan Kara 		 * revoke them which is why we don't set
405f2890730SJan Kara 		 * EXT4_FREE_BLOCKS_METADATA.
406c5c7b8ddSJan Kara 		 */
407f2890730SJan Kara 		ext4_free_blocks(handle, ar->inode, branch[i+1].bh,
408f2890730SJan Kara 				 new_blocks[i], 1,
409f2890730SJan Kara 				 branch[i+1].bh ? EXT4_FREE_BLOCKS_FORGET : 0);
410dae1e52cSAmir Goldstein 	}
411dae1e52cSAmir Goldstein 	return err;
412dae1e52cSAmir Goldstein }
413dae1e52cSAmir Goldstein 
414dae1e52cSAmir Goldstein /**
415c60990b3STheodore Ts'o  * ext4_splice_branch() - splice the allocated branch onto inode.
416dae1e52cSAmir Goldstein  * @handle: handle for this transaction
417c60990b3STheodore Ts'o  * @ar: structure describing the allocation request
418dae1e52cSAmir Goldstein  * @where: location of missing link
419dae1e52cSAmir Goldstein  * @num:   number of indirect blocks we are adding
420dae1e52cSAmir Goldstein  *
421dae1e52cSAmir Goldstein  * This function fills the missing link and does all housekeeping needed in
422dae1e52cSAmir Goldstein  * inode (->i_blocks, etc.). In case of success we end up with the full
423dae1e52cSAmir Goldstein  * chain to new block and return 0.
424dae1e52cSAmir Goldstein  */
ext4_splice_branch(handle_t * handle,struct ext4_allocation_request * ar,Indirect * where,int num)425a5211002STheodore Ts'o static int ext4_splice_branch(handle_t *handle,
426a5211002STheodore Ts'o 			      struct ext4_allocation_request *ar,
427a5211002STheodore Ts'o 			      Indirect *where, int num)
428dae1e52cSAmir Goldstein {
429dae1e52cSAmir Goldstein 	int i;
430dae1e52cSAmir Goldstein 	int err = 0;
431dae1e52cSAmir Goldstein 	ext4_fsblk_t current_block;
432dae1e52cSAmir Goldstein 
433dae1e52cSAmir Goldstein 	/*
434dae1e52cSAmir Goldstein 	 * If we're splicing into a [td]indirect block (as opposed to the
435dae1e52cSAmir Goldstein 	 * inode) then we need to get write access to the [td]indirect block
436dae1e52cSAmir Goldstein 	 * before the splice.
437dae1e52cSAmir Goldstein 	 */
438dae1e52cSAmir Goldstein 	if (where->bh) {
439dae1e52cSAmir Goldstein 		BUFFER_TRACE(where->bh, "get_write_access");
440188c299eSJan Kara 		err = ext4_journal_get_write_access(handle, ar->inode->i_sb,
441188c299eSJan Kara 						    where->bh, EXT4_JTR_NONE);
442dae1e52cSAmir Goldstein 		if (err)
443dae1e52cSAmir Goldstein 			goto err_out;
444dae1e52cSAmir Goldstein 	}
445dae1e52cSAmir Goldstein 	/* That's it */
446dae1e52cSAmir Goldstein 
447dae1e52cSAmir Goldstein 	*where->p = where->key;
448dae1e52cSAmir Goldstein 
449dae1e52cSAmir Goldstein 	/*
450dae1e52cSAmir Goldstein 	 * Update the host buffer_head or inode to point to more just allocated
451dae1e52cSAmir Goldstein 	 * direct blocks blocks
452dae1e52cSAmir Goldstein 	 */
453a5211002STheodore Ts'o 	if (num == 0 && ar->len > 1) {
454dae1e52cSAmir Goldstein 		current_block = le32_to_cpu(where->key) + 1;
455a5211002STheodore Ts'o 		for (i = 1; i < ar->len; i++)
456dae1e52cSAmir Goldstein 			*(where->p + i) = cpu_to_le32(current_block++);
457dae1e52cSAmir Goldstein 	}
458dae1e52cSAmir Goldstein 
459dae1e52cSAmir Goldstein 	/* We are done with atomic stuff, now do the rest of housekeeping */
460dae1e52cSAmir Goldstein 	/* had we spliced it onto indirect block? */
461dae1e52cSAmir Goldstein 	if (where->bh) {
462dae1e52cSAmir Goldstein 		/*
463dae1e52cSAmir Goldstein 		 * If we spliced it onto an indirect block, we haven't
464dae1e52cSAmir Goldstein 		 * altered the inode.  Note however that if it is being spliced
465dae1e52cSAmir Goldstein 		 * onto an indirect block at the very end of the file (the
466dae1e52cSAmir Goldstein 		 * file is growing) then we *will* alter the inode to reflect
467dae1e52cSAmir Goldstein 		 * the new i_size.  But that is not done here - it is done in
468dae1e52cSAmir Goldstein 		 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
469dae1e52cSAmir Goldstein 		 */
4704978c659SJan Kara 		ext4_debug("splicing indirect only\n");
471dae1e52cSAmir Goldstein 		BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
472a5211002STheodore Ts'o 		err = ext4_handle_dirty_metadata(handle, ar->inode, where->bh);
473dae1e52cSAmir Goldstein 		if (err)
474dae1e52cSAmir Goldstein 			goto err_out;
475dae1e52cSAmir Goldstein 	} else {
476dae1e52cSAmir Goldstein 		/*
477dae1e52cSAmir Goldstein 		 * OK, we spliced it into the inode itself on a direct block.
478dae1e52cSAmir Goldstein 		 */
4794209ae12SHarshad Shirwadkar 		err = ext4_mark_inode_dirty(handle, ar->inode);
4804209ae12SHarshad Shirwadkar 		if (unlikely(err))
4814209ae12SHarshad Shirwadkar 			goto err_out;
4824978c659SJan Kara 		ext4_debug("splicing direct\n");
483dae1e52cSAmir Goldstein 	}
484dae1e52cSAmir Goldstein 	return err;
485dae1e52cSAmir Goldstein 
486dae1e52cSAmir Goldstein err_out:
487dae1e52cSAmir Goldstein 	for (i = 1; i <= num; i++) {
488dae1e52cSAmir Goldstein 		/*
489dae1e52cSAmir Goldstein 		 * branch[i].bh is newly allocated, so there is no
490dae1e52cSAmir Goldstein 		 * need to revoke the block, which is why we don't
491dae1e52cSAmir Goldstein 		 * need to set EXT4_FREE_BLOCKS_METADATA.
492dae1e52cSAmir Goldstein 		 */
493a5211002STheodore Ts'o 		ext4_free_blocks(handle, ar->inode, where[i].bh, 0, 1,
494dae1e52cSAmir Goldstein 				 EXT4_FREE_BLOCKS_FORGET);
495dae1e52cSAmir Goldstein 	}
496a5211002STheodore Ts'o 	ext4_free_blocks(handle, ar->inode, NULL, le32_to_cpu(where[num].key),
497a5211002STheodore Ts'o 			 ar->len, 0);
498dae1e52cSAmir Goldstein 
499dae1e52cSAmir Goldstein 	return err;
500dae1e52cSAmir Goldstein }
501dae1e52cSAmir Goldstein 
502dae1e52cSAmir Goldstein /*
503dae1e52cSAmir Goldstein  * The ext4_ind_map_blocks() function handles non-extents inodes
504dae1e52cSAmir Goldstein  * (i.e., using the traditional indirect/double-indirect i_blocks
505dae1e52cSAmir Goldstein  * scheme) for ext4_map_blocks().
506dae1e52cSAmir Goldstein  *
507dae1e52cSAmir Goldstein  * Allocation strategy is simple: if we have to allocate something, we will
508dae1e52cSAmir Goldstein  * have to go the whole way to leaf. So let's do it before attaching anything
509dae1e52cSAmir Goldstein  * to tree, set linkage between the newborn blocks, write them if sync is
510dae1e52cSAmir Goldstein  * required, recheck the path, free and repeat if check fails, otherwise
511dae1e52cSAmir Goldstein  * set the last missing link (that will protect us from any truncate-generated
512dae1e52cSAmir Goldstein  * removals - all blocks on the path are immune now) and possibly force the
513dae1e52cSAmir Goldstein  * write on the parent block.
514dae1e52cSAmir Goldstein  * That has a nice additional property: no special recovery from the failed
515dae1e52cSAmir Goldstein  * allocations is needed - we simply release blocks and do not touch anything
516dae1e52cSAmir Goldstein  * reachable from inode.
517dae1e52cSAmir Goldstein  *
518dae1e52cSAmir Goldstein  * `handle' can be NULL if create == 0.
519dae1e52cSAmir Goldstein  *
520dae1e52cSAmir Goldstein  * return > 0, # of blocks mapped or allocated.
521dae1e52cSAmir Goldstein  * return = 0, if plain lookup failed.
522dae1e52cSAmir Goldstein  * return < 0, error case.
523dae1e52cSAmir Goldstein  *
524dae1e52cSAmir Goldstein  * The ext4_ind_get_blocks() function should be called with
525dae1e52cSAmir Goldstein  * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
526dae1e52cSAmir Goldstein  * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
527dae1e52cSAmir Goldstein  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
528dae1e52cSAmir Goldstein  * blocks.
529dae1e52cSAmir Goldstein  */
ext4_ind_map_blocks(handle_t * handle,struct inode * inode,struct ext4_map_blocks * map,int flags)530dae1e52cSAmir Goldstein int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
531dae1e52cSAmir Goldstein 			struct ext4_map_blocks *map,
532dae1e52cSAmir Goldstein 			int flags)
533dae1e52cSAmir Goldstein {
534a5211002STheodore Ts'o 	struct ext4_allocation_request ar;
535dae1e52cSAmir Goldstein 	int err = -EIO;
536dae1e52cSAmir Goldstein 	ext4_lblk_t offsets[4];
537dae1e52cSAmir Goldstein 	Indirect chain[4];
538dae1e52cSAmir Goldstein 	Indirect *partial;
539dae1e52cSAmir Goldstein 	int indirect_blks;
540dae1e52cSAmir Goldstein 	int blocks_to_boundary = 0;
541dae1e52cSAmir Goldstein 	int depth;
542dae1e52cSAmir Goldstein 	int count = 0;
543dae1e52cSAmir Goldstein 	ext4_fsblk_t first_block = 0;
544dae1e52cSAmir Goldstein 
545dae1e52cSAmir Goldstein 	trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
546837c23fbSChunguang Xu 	ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
547837c23fbSChunguang Xu 	ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
548dae1e52cSAmir Goldstein 	depth = ext4_block_to_path(inode, map->m_lblk, offsets,
549dae1e52cSAmir Goldstein 				   &blocks_to_boundary);
550dae1e52cSAmir Goldstein 
551dae1e52cSAmir Goldstein 	if (depth == 0)
552dae1e52cSAmir Goldstein 		goto out;
553dae1e52cSAmir Goldstein 
554dae1e52cSAmir Goldstein 	partial = ext4_get_branch(inode, depth, offsets, chain, &err);
555dae1e52cSAmir Goldstein 
556dae1e52cSAmir Goldstein 	/* Simplest case - block found, no allocation needed */
557dae1e52cSAmir Goldstein 	if (!partial) {
558dae1e52cSAmir Goldstein 		first_block = le32_to_cpu(chain[depth - 1].key);
559dae1e52cSAmir Goldstein 		count++;
560dae1e52cSAmir Goldstein 		/*map more blocks*/
561dae1e52cSAmir Goldstein 		while (count < map->m_len && count <= blocks_to_boundary) {
562dae1e52cSAmir Goldstein 			ext4_fsblk_t blk;
563dae1e52cSAmir Goldstein 
564dae1e52cSAmir Goldstein 			blk = le32_to_cpu(*(chain[depth-1].p + count));
565dae1e52cSAmir Goldstein 
566dae1e52cSAmir Goldstein 			if (blk == first_block + count)
567dae1e52cSAmir Goldstein 				count++;
568dae1e52cSAmir Goldstein 			else
569dae1e52cSAmir Goldstein 				break;
570dae1e52cSAmir Goldstein 		}
571dae1e52cSAmir Goldstein 		goto got_it;
572dae1e52cSAmir Goldstein 	}
573dae1e52cSAmir Goldstein 
574facab4d9SJan Kara 	/* Next simple case - plain lookup failed */
575facab4d9SJan Kara 	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
576facab4d9SJan Kara 		unsigned epb = inode->i_sb->s_blocksize / sizeof(u32);
577facab4d9SJan Kara 		int i;
578facab4d9SJan Kara 
5792ee3ee06SJan Kara 		/*
5802ee3ee06SJan Kara 		 * Count number blocks in a subtree under 'partial'. At each
5812ee3ee06SJan Kara 		 * level we count number of complete empty subtrees beyond
5822ee3ee06SJan Kara 		 * current offset and then descend into the subtree only
5832ee3ee06SJan Kara 		 * partially beyond current offset.
5842ee3ee06SJan Kara 		 */
5852ee3ee06SJan Kara 		count = 0;
5862ee3ee06SJan Kara 		for (i = partial - chain + 1; i < depth; i++)
5872ee3ee06SJan Kara 			count = count * epb + (epb - offsets[i] - 1);
5882ee3ee06SJan Kara 		count++;
589facab4d9SJan Kara 		/* Fill in size of a hole we found */
590facab4d9SJan Kara 		map->m_pblk = 0;
591facab4d9SJan Kara 		map->m_len = min_t(unsigned int, map->m_len, count);
592facab4d9SJan Kara 		goto cleanup;
593facab4d9SJan Kara 	}
594facab4d9SJan Kara 
595facab4d9SJan Kara 	/* Failed read of indirect block */
596facab4d9SJan Kara 	if (err == -EIO)
597dae1e52cSAmir Goldstein 		goto cleanup;
598dae1e52cSAmir Goldstein 
599dae1e52cSAmir Goldstein 	/*
600dae1e52cSAmir Goldstein 	 * Okay, we need to do block allocation.
601dae1e52cSAmir Goldstein 	*/
602e2b911c5SDarrick J. Wong 	if (ext4_has_feature_bigalloc(inode->i_sb)) {
603bab08ab9STheodore Ts'o 		EXT4_ERROR_INODE(inode, "Can't allocate blocks for "
604bab08ab9STheodore Ts'o 				 "non-extent mapped inodes with bigalloc");
6052be7d717SZhang Qilong 		err = -EFSCORRUPTED;
6062be7d717SZhang Qilong 		goto out;
607bab08ab9STheodore Ts'o 	}
608bab08ab9STheodore Ts'o 
609a5211002STheodore Ts'o 	/* Set up for the direct block allocation */
610a5211002STheodore Ts'o 	memset(&ar, 0, sizeof(ar));
611a5211002STheodore Ts'o 	ar.inode = inode;
612a5211002STheodore Ts'o 	ar.logical = map->m_lblk;
613a5211002STheodore Ts'o 	if (S_ISREG(inode->i_mode))
614a5211002STheodore Ts'o 		ar.flags = EXT4_MB_HINT_DATA;
615e3cf5d5dSTheodore Ts'o 	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
616e3cf5d5dSTheodore Ts'o 		ar.flags |= EXT4_MB_DELALLOC_RESERVED;
617c5e298aeSTheodore Ts'o 	if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
618c5e298aeSTheodore Ts'o 		ar.flags |= EXT4_MB_USE_RESERVED;
619a5211002STheodore Ts'o 
620a5211002STheodore Ts'o 	ar.goal = ext4_find_goal(inode, map->m_lblk, partial);
621dae1e52cSAmir Goldstein 
622dae1e52cSAmir Goldstein 	/* the number of blocks need to allocate for [d,t]indirect blocks */
623dae1e52cSAmir Goldstein 	indirect_blks = (chain + depth) - partial - 1;
624dae1e52cSAmir Goldstein 
625dae1e52cSAmir Goldstein 	/*
626dae1e52cSAmir Goldstein 	 * Next look up the indirect map to count the totoal number of
627dae1e52cSAmir Goldstein 	 * direct blocks to allocate for this branch.
628dae1e52cSAmir Goldstein 	 */
629a5211002STheodore Ts'o 	ar.len = ext4_blks_to_allocate(partial, indirect_blks,
630dae1e52cSAmir Goldstein 				       map->m_len, blocks_to_boundary);
631a5211002STheodore Ts'o 
632dae1e52cSAmir Goldstein 	/*
633dae1e52cSAmir Goldstein 	 * Block out ext4_truncate while we alter the tree
634dae1e52cSAmir Goldstein 	 */
635a5211002STheodore Ts'o 	err = ext4_alloc_branch(handle, &ar, indirect_blks,
636dae1e52cSAmir Goldstein 				offsets + (partial - chain), partial);
637dae1e52cSAmir Goldstein 
638dae1e52cSAmir Goldstein 	/*
639dae1e52cSAmir Goldstein 	 * The ext4_splice_branch call will free and forget any buffers
640dae1e52cSAmir Goldstein 	 * on the new chain if there is a failure, but that risks using
641dae1e52cSAmir Goldstein 	 * up transaction credits, especially for bitmaps where the
642dae1e52cSAmir Goldstein 	 * credits cannot be returned.  Can we handle this somehow?  We
643dae1e52cSAmir Goldstein 	 * may need to return -EAGAIN upwards in the worst case.  --sct
644dae1e52cSAmir Goldstein 	 */
645dae1e52cSAmir Goldstein 	if (!err)
646a5211002STheodore Ts'o 		err = ext4_splice_branch(handle, &ar, partial, indirect_blks);
647dae1e52cSAmir Goldstein 	if (err)
648dae1e52cSAmir Goldstein 		goto cleanup;
649dae1e52cSAmir Goldstein 
650dae1e52cSAmir Goldstein 	map->m_flags |= EXT4_MAP_NEW;
651dae1e52cSAmir Goldstein 
652dae1e52cSAmir Goldstein 	ext4_update_inode_fsync_trans(handle, inode, 1);
653a5211002STheodore Ts'o 	count = ar.len;
654*de25d6e9SBaokun Li 
655*de25d6e9SBaokun Li 	/*
656*de25d6e9SBaokun Li 	 * Update reserved blocks/metadata blocks after successful block
657*de25d6e9SBaokun Li 	 * allocation which had been deferred till now.
658*de25d6e9SBaokun Li 	 */
659*de25d6e9SBaokun Li 	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
660*de25d6e9SBaokun Li 		ext4_da_update_reserve_space(inode, count, 1);
661*de25d6e9SBaokun Li 
662dae1e52cSAmir Goldstein got_it:
663dae1e52cSAmir Goldstein 	map->m_flags |= EXT4_MAP_MAPPED;
664dae1e52cSAmir Goldstein 	map->m_pblk = le32_to_cpu(chain[depth-1].key);
665dae1e52cSAmir Goldstein 	map->m_len = count;
666dae1e52cSAmir Goldstein 	if (count > blocks_to_boundary)
667dae1e52cSAmir Goldstein 		map->m_flags |= EXT4_MAP_BOUNDARY;
668dae1e52cSAmir Goldstein 	err = count;
669dae1e52cSAmir Goldstein 	/* Clean up and exit */
670dae1e52cSAmir Goldstein 	partial = chain + depth - 1;	/* the whole chain */
671dae1e52cSAmir Goldstein cleanup:
672dae1e52cSAmir Goldstein 	while (partial > chain) {
673dae1e52cSAmir Goldstein 		BUFFER_TRACE(partial->bh, "call brelse");
674dae1e52cSAmir Goldstein 		brelse(partial->bh);
675dae1e52cSAmir Goldstein 		partial--;
676dae1e52cSAmir Goldstein 	}
677dae1e52cSAmir Goldstein out:
67821ddd568STheodore Ts'o 	trace_ext4_ind_map_blocks_exit(inode, flags, map, err);
679dae1e52cSAmir Goldstein 	return err;
680dae1e52cSAmir Goldstein }
681dae1e52cSAmir Goldstein 
682dae1e52cSAmir Goldstein /*
683fa55a0edSJan Kara  * Calculate number of indirect blocks touched by mapping @nrblocks logically
684fa55a0edSJan Kara  * contiguous blocks
685fa55a0edSJan Kara  */
ext4_ind_trans_blocks(struct inode * inode,int nrblocks)686fa55a0edSJan Kara int ext4_ind_trans_blocks(struct inode *inode, int nrblocks)
687dae1e52cSAmir Goldstein {
688dae1e52cSAmir Goldstein 	/*
689dae1e52cSAmir Goldstein 	 * With N contiguous data blocks, we need at most
690dae1e52cSAmir Goldstein 	 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
691dae1e52cSAmir Goldstein 	 * 2 dindirect blocks, and 1 tindirect block
692dae1e52cSAmir Goldstein 	 */
693fa55a0edSJan Kara 	return DIV_ROUND_UP(nrblocks, EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
694dae1e52cSAmir Goldstein }
695dae1e52cSAmir Goldstein 
ext4_ind_trunc_restart_fn(handle_t * handle,struct inode * inode,struct buffer_head * bh,int * dropped)696a4130367SJan Kara static int ext4_ind_trunc_restart_fn(handle_t *handle, struct inode *inode,
697a4130367SJan Kara 				     struct buffer_head *bh, int *dropped)
698a4130367SJan Kara {
699a4130367SJan Kara 	int err;
700a4130367SJan Kara 
701a4130367SJan Kara 	if (bh) {
702a4130367SJan Kara 		BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
703a4130367SJan Kara 		err = ext4_handle_dirty_metadata(handle, inode, bh);
704a4130367SJan Kara 		if (unlikely(err))
705a4130367SJan Kara 			return err;
706a4130367SJan Kara 	}
707a4130367SJan Kara 	err = ext4_mark_inode_dirty(handle, inode);
708a4130367SJan Kara 	if (unlikely(err))
709a4130367SJan Kara 		return err;
710a4130367SJan Kara 	/*
711a4130367SJan Kara 	 * Drop i_data_sem to avoid deadlock with ext4_map_blocks.  At this
712a4130367SJan Kara 	 * moment, get_block can be called only for blocks inside i_size since
713a4130367SJan Kara 	 * page cache has been already dropped and writes are blocked by
714f340b3d9Shongnanli 	 * i_rwsem. So we can safely drop the i_data_sem here.
715a4130367SJan Kara 	 */
716a4130367SJan Kara 	BUG_ON(EXT4_JOURNAL(inode) == NULL);
71727bc446eSbrookxu 	ext4_discard_preallocations(inode, 0);
718a4130367SJan Kara 	up_write(&EXT4_I(inode)->i_data_sem);
719a4130367SJan Kara 	*dropped = 1;
720a4130367SJan Kara 	return 0;
721a4130367SJan Kara }
722a4130367SJan Kara 
723dae1e52cSAmir Goldstein /*
724dae1e52cSAmir Goldstein  * Truncate transactions can be complex and absolutely huge.  So we need to
7253088e5a5SBhaskar Chowdhury  * be able to restart the transaction at a convenient checkpoint to make
726dae1e52cSAmir Goldstein  * sure we don't overflow the journal.
727dae1e52cSAmir Goldstein  *
728819c4920STheodore Ts'o  * Try to extend this transaction for the purposes of truncation.  If
729a4130367SJan Kara  * extend fails, we restart transaction.
730dae1e52cSAmir Goldstein  */
ext4_ind_truncate_ensure_credits(handle_t * handle,struct inode * inode,struct buffer_head * bh,int revoke_creds)731a4130367SJan Kara static int ext4_ind_truncate_ensure_credits(handle_t *handle,
732a4130367SJan Kara 					    struct inode *inode,
73383448bdfSJan Kara 					    struct buffer_head *bh,
73483448bdfSJan Kara 					    int revoke_creds)
735dae1e52cSAmir Goldstein {
736a4130367SJan Kara 	int ret;
737a4130367SJan Kara 	int dropped = 0;
738a4130367SJan Kara 
739a4130367SJan Kara 	ret = ext4_journal_ensure_credits_fn(handle, EXT4_RESERVE_TRANS_BLOCKS,
74083448bdfSJan Kara 			ext4_blocks_for_truncate(inode), revoke_creds,
741a4130367SJan Kara 			ext4_ind_trunc_restart_fn(handle, inode, bh, &dropped));
742a4130367SJan Kara 	if (dropped)
743a4130367SJan Kara 		down_write(&EXT4_I(inode)->i_data_sem);
744a4130367SJan Kara 	if (ret <= 0)
745a4130367SJan Kara 		return ret;
746a4130367SJan Kara 	if (bh) {
747a4130367SJan Kara 		BUFFER_TRACE(bh, "retaking write access");
748188c299eSJan Kara 		ret = ext4_journal_get_write_access(handle, inode->i_sb, bh,
749188c299eSJan Kara 						    EXT4_JTR_NONE);
750a4130367SJan Kara 		if (unlikely(ret))
751a4130367SJan Kara 			return ret;
752a4130367SJan Kara 	}
753dae1e52cSAmir Goldstein 	return 0;
754dae1e52cSAmir Goldstein }
755dae1e52cSAmir Goldstein 
756dae1e52cSAmir Goldstein /*
757dae1e52cSAmir Goldstein  * Probably it should be a library function... search for first non-zero word
758dae1e52cSAmir Goldstein  * or memcmp with zero_page, whatever is better for particular architecture.
759dae1e52cSAmir Goldstein  * Linus?
760dae1e52cSAmir Goldstein  */
all_zeroes(__le32 * p,__le32 * q)761dae1e52cSAmir Goldstein static inline int all_zeroes(__le32 *p, __le32 *q)
762dae1e52cSAmir Goldstein {
763dae1e52cSAmir Goldstein 	while (p < q)
764dae1e52cSAmir Goldstein 		if (*p++)
765dae1e52cSAmir Goldstein 			return 0;
766dae1e52cSAmir Goldstein 	return 1;
767dae1e52cSAmir Goldstein }
768dae1e52cSAmir Goldstein 
769dae1e52cSAmir Goldstein /**
770dae1e52cSAmir Goldstein  *	ext4_find_shared - find the indirect blocks for partial truncation.
771dae1e52cSAmir Goldstein  *	@inode:	  inode in question
772dae1e52cSAmir Goldstein  *	@depth:	  depth of the affected branch
773dae1e52cSAmir Goldstein  *	@offsets: offsets of pointers in that branch (see ext4_block_to_path)
774dae1e52cSAmir Goldstein  *	@chain:	  place to store the pointers to partial indirect blocks
775dae1e52cSAmir Goldstein  *	@top:	  place to the (detached) top of branch
776dae1e52cSAmir Goldstein  *
777dae1e52cSAmir Goldstein  *	This is a helper function used by ext4_truncate().
778dae1e52cSAmir Goldstein  *
779dae1e52cSAmir Goldstein  *	When we do truncate() we may have to clean the ends of several
780dae1e52cSAmir Goldstein  *	indirect blocks but leave the blocks themselves alive. Block is
781dae1e52cSAmir Goldstein  *	partially truncated if some data below the new i_size is referred
782dae1e52cSAmir Goldstein  *	from it (and it is on the path to the first completely truncated
783dae1e52cSAmir Goldstein  *	data block, indeed).  We have to free the top of that path along
784dae1e52cSAmir Goldstein  *	with everything to the right of the path. Since no allocation
785dae1e52cSAmir Goldstein  *	past the truncation point is possible until ext4_truncate()
786dae1e52cSAmir Goldstein  *	finishes, we may safely do the latter, but top of branch may
787dae1e52cSAmir Goldstein  *	require special attention - pageout below the truncation point
788dae1e52cSAmir Goldstein  *	might try to populate it.
789dae1e52cSAmir Goldstein  *
790dae1e52cSAmir Goldstein  *	We atomically detach the top of branch from the tree, store the
791dae1e52cSAmir Goldstein  *	block number of its root in *@top, pointers to buffer_heads of
792dae1e52cSAmir Goldstein  *	partially truncated blocks - in @chain[].bh and pointers to
793dae1e52cSAmir Goldstein  *	their last elements that should not be removed - in
794dae1e52cSAmir Goldstein  *	@chain[].p. Return value is the pointer to last filled element
795dae1e52cSAmir Goldstein  *	of @chain.
796dae1e52cSAmir Goldstein  *
797dae1e52cSAmir Goldstein  *	The work left to caller to do the actual freeing of subtrees:
798dae1e52cSAmir Goldstein  *		a) free the subtree starting from *@top
799dae1e52cSAmir Goldstein  *		b) free the subtrees whose roots are stored in
800dae1e52cSAmir Goldstein  *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
801dae1e52cSAmir Goldstein  *		c) free the subtrees growing from the inode past the @chain[0].
802dae1e52cSAmir Goldstein  *			(no partially truncated stuff there).  */
803dae1e52cSAmir Goldstein 
ext4_find_shared(struct inode * inode,int depth,ext4_lblk_t offsets[4],Indirect chain[4],__le32 * top)804dae1e52cSAmir Goldstein static Indirect *ext4_find_shared(struct inode *inode, int depth,
805dae1e52cSAmir Goldstein 				  ext4_lblk_t offsets[4], Indirect chain[4],
806dae1e52cSAmir Goldstein 				  __le32 *top)
807dae1e52cSAmir Goldstein {
808dae1e52cSAmir Goldstein 	Indirect *partial, *p;
809dae1e52cSAmir Goldstein 	int k, err;
810dae1e52cSAmir Goldstein 
811dae1e52cSAmir Goldstein 	*top = 0;
812dae1e52cSAmir Goldstein 	/* Make k index the deepest non-null offset + 1 */
813dae1e52cSAmir Goldstein 	for (k = depth; k > 1 && !offsets[k-1]; k--)
814dae1e52cSAmir Goldstein 		;
815dae1e52cSAmir Goldstein 	partial = ext4_get_branch(inode, k, offsets, chain, &err);
816dae1e52cSAmir Goldstein 	/* Writer: pointers */
817dae1e52cSAmir Goldstein 	if (!partial)
818dae1e52cSAmir Goldstein 		partial = chain + k-1;
819dae1e52cSAmir Goldstein 	/*
820dae1e52cSAmir Goldstein 	 * If the branch acquired continuation since we've looked at it -
821dae1e52cSAmir Goldstein 	 * fine, it should all survive and (new) top doesn't belong to us.
822dae1e52cSAmir Goldstein 	 */
823dae1e52cSAmir Goldstein 	if (!partial->key && *partial->p)
824dae1e52cSAmir Goldstein 		/* Writer: end */
825dae1e52cSAmir Goldstein 		goto no_top;
826dae1e52cSAmir Goldstein 	for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
827dae1e52cSAmir Goldstein 		;
828dae1e52cSAmir Goldstein 	/*
829dae1e52cSAmir Goldstein 	 * OK, we've found the last block that must survive. The rest of our
830dae1e52cSAmir Goldstein 	 * branch should be detached before unlocking. However, if that rest
831dae1e52cSAmir Goldstein 	 * of branch is all ours and does not grow immediately from the inode
832dae1e52cSAmir Goldstein 	 * it's easier to cheat and just decrement partial->p.
833dae1e52cSAmir Goldstein 	 */
834dae1e52cSAmir Goldstein 	if (p == chain + k - 1 && p > chain) {
835dae1e52cSAmir Goldstein 		p->p--;
836dae1e52cSAmir Goldstein 	} else {
837dae1e52cSAmir Goldstein 		*top = *p->p;
838dae1e52cSAmir Goldstein 		/* Nope, don't do this in ext4.  Must leave the tree intact */
839dae1e52cSAmir Goldstein #if 0
840dae1e52cSAmir Goldstein 		*p->p = 0;
841dae1e52cSAmir Goldstein #endif
842dae1e52cSAmir Goldstein 	}
843dae1e52cSAmir Goldstein 	/* Writer: end */
844dae1e52cSAmir Goldstein 
845dae1e52cSAmir Goldstein 	while (partial > p) {
846dae1e52cSAmir Goldstein 		brelse(partial->bh);
847dae1e52cSAmir Goldstein 		partial--;
848dae1e52cSAmir Goldstein 	}
849dae1e52cSAmir Goldstein no_top:
850dae1e52cSAmir Goldstein 	return partial;
851dae1e52cSAmir Goldstein }
852dae1e52cSAmir Goldstein 
853dae1e52cSAmir Goldstein /*
854dae1e52cSAmir Goldstein  * Zero a number of block pointers in either an inode or an indirect block.
855dae1e52cSAmir Goldstein  * If we restart the transaction we must again get write access to the
856dae1e52cSAmir Goldstein  * indirect block for further modification.
857dae1e52cSAmir Goldstein  *
858dae1e52cSAmir Goldstein  * We release `count' blocks on disk, but (last - first) may be greater
859dae1e52cSAmir Goldstein  * than `count' because there can be holes in there.
860dae1e52cSAmir Goldstein  *
861dae1e52cSAmir Goldstein  * Return 0 on success, 1 on invalid block range
862dae1e52cSAmir Goldstein  * and < 0 on fatal error.
863dae1e52cSAmir Goldstein  */
ext4_clear_blocks(handle_t * handle,struct inode * inode,struct buffer_head * bh,ext4_fsblk_t block_to_free,unsigned long count,__le32 * first,__le32 * last)864dae1e52cSAmir Goldstein static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
865dae1e52cSAmir Goldstein 			     struct buffer_head *bh,
866dae1e52cSAmir Goldstein 			     ext4_fsblk_t block_to_free,
867dae1e52cSAmir Goldstein 			     unsigned long count, __le32 *first,
868dae1e52cSAmir Goldstein 			     __le32 *last)
869dae1e52cSAmir Goldstein {
870dae1e52cSAmir Goldstein 	__le32 *p;
871981250caSTheodore Ts'o 	int	flags = EXT4_FREE_BLOCKS_VALIDATED;
872dae1e52cSAmir Goldstein 	int	err;
873dae1e52cSAmir Goldstein 
874ddfa17e4STahsin Erdogan 	if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
875ddfa17e4STahsin Erdogan 	    ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
876981250caSTheodore Ts'o 		flags |= EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_METADATA;
877981250caSTheodore Ts'o 	else if (ext4_should_journal_data(inode))
878981250caSTheodore Ts'o 		flags |= EXT4_FREE_BLOCKS_FORGET;
879dae1e52cSAmir Goldstein 
880ce9f24ccSJan Kara 	if (!ext4_inode_block_valid(inode, block_to_free, count)) {
881dae1e52cSAmir Goldstein 		EXT4_ERROR_INODE(inode, "attempt to clear invalid "
882dae1e52cSAmir Goldstein 				 "blocks %llu len %lu",
883dae1e52cSAmir Goldstein 				 (unsigned long long) block_to_free, count);
884dae1e52cSAmir Goldstein 		return 1;
885dae1e52cSAmir Goldstein 	}
886dae1e52cSAmir Goldstein 
88783448bdfSJan Kara 	err = ext4_ind_truncate_ensure_credits(handle, inode, bh,
88883448bdfSJan Kara 				ext4_free_data_revoke_credits(inode, count));
889a4130367SJan Kara 	if (err < 0)
890dae1e52cSAmir Goldstein 		goto out_err;
891dae1e52cSAmir Goldstein 
892dae1e52cSAmir Goldstein 	for (p = first; p < last; p++)
893dae1e52cSAmir Goldstein 		*p = 0;
894dae1e52cSAmir Goldstein 
895dae1e52cSAmir Goldstein 	ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
896dae1e52cSAmir Goldstein 	return 0;
897dae1e52cSAmir Goldstein out_err:
898dae1e52cSAmir Goldstein 	ext4_std_error(inode->i_sb, err);
899dae1e52cSAmir Goldstein 	return err;
900dae1e52cSAmir Goldstein }
901dae1e52cSAmir Goldstein 
902dae1e52cSAmir Goldstein /**
903dae1e52cSAmir Goldstein  * ext4_free_data - free a list of data blocks
904dae1e52cSAmir Goldstein  * @handle:	handle for this transaction
905dae1e52cSAmir Goldstein  * @inode:	inode we are dealing with
906dae1e52cSAmir Goldstein  * @this_bh:	indirect buffer_head which contains *@first and *@last
907dae1e52cSAmir Goldstein  * @first:	array of block numbers
908dae1e52cSAmir Goldstein  * @last:	points immediately past the end of array
909dae1e52cSAmir Goldstein  *
910dae1e52cSAmir Goldstein  * We are freeing all blocks referred from that array (numbers are stored as
911dae1e52cSAmir Goldstein  * little-endian 32-bit) and updating @inode->i_blocks appropriately.
912dae1e52cSAmir Goldstein  *
913dae1e52cSAmir Goldstein  * We accumulate contiguous runs of blocks to free.  Conveniently, if these
914dae1e52cSAmir Goldstein  * blocks are contiguous then releasing them at one time will only affect one
915dae1e52cSAmir Goldstein  * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
916dae1e52cSAmir Goldstein  * actually use a lot of journal space.
917dae1e52cSAmir Goldstein  *
918dae1e52cSAmir Goldstein  * @this_bh will be %NULL if @first and @last point into the inode's direct
919dae1e52cSAmir Goldstein  * block pointers.
920dae1e52cSAmir Goldstein  */
ext4_free_data(handle_t * handle,struct inode * inode,struct buffer_head * this_bh,__le32 * first,__le32 * last)921dae1e52cSAmir Goldstein static void ext4_free_data(handle_t *handle, struct inode *inode,
922dae1e52cSAmir Goldstein 			   struct buffer_head *this_bh,
923dae1e52cSAmir Goldstein 			   __le32 *first, __le32 *last)
924dae1e52cSAmir Goldstein {
925dae1e52cSAmir Goldstein 	ext4_fsblk_t block_to_free = 0;    /* Starting block # of a run */
926dae1e52cSAmir Goldstein 	unsigned long count = 0;	    /* Number of blocks in the run */
927dae1e52cSAmir Goldstein 	__le32 *block_to_free_p = NULL;	    /* Pointer into inode/ind
928dae1e52cSAmir Goldstein 					       corresponding to
929dae1e52cSAmir Goldstein 					       block_to_free */
930dae1e52cSAmir Goldstein 	ext4_fsblk_t nr;		    /* Current block # */
931dae1e52cSAmir Goldstein 	__le32 *p;			    /* Pointer into inode/ind
932dae1e52cSAmir Goldstein 					       for current block */
933dae1e52cSAmir Goldstein 	int err = 0;
934dae1e52cSAmir Goldstein 
935dae1e52cSAmir Goldstein 	if (this_bh) {				/* For indirect block */
936dae1e52cSAmir Goldstein 		BUFFER_TRACE(this_bh, "get_write_access");
937188c299eSJan Kara 		err = ext4_journal_get_write_access(handle, inode->i_sb,
938188c299eSJan Kara 						    this_bh, EXT4_JTR_NONE);
939dae1e52cSAmir Goldstein 		/* Important: if we can't update the indirect pointers
940dae1e52cSAmir Goldstein 		 * to the blocks, we can't free them. */
941dae1e52cSAmir Goldstein 		if (err)
942dae1e52cSAmir Goldstein 			return;
943dae1e52cSAmir Goldstein 	}
944dae1e52cSAmir Goldstein 
945dae1e52cSAmir Goldstein 	for (p = first; p < last; p++) {
946dae1e52cSAmir Goldstein 		nr = le32_to_cpu(*p);
947dae1e52cSAmir Goldstein 		if (nr) {
948dae1e52cSAmir Goldstein 			/* accumulate blocks to free if they're contiguous */
949dae1e52cSAmir Goldstein 			if (count == 0) {
950dae1e52cSAmir Goldstein 				block_to_free = nr;
951dae1e52cSAmir Goldstein 				block_to_free_p = p;
952dae1e52cSAmir Goldstein 				count = 1;
953dae1e52cSAmir Goldstein 			} else if (nr == block_to_free + count) {
954dae1e52cSAmir Goldstein 				count++;
955dae1e52cSAmir Goldstein 			} else {
956dae1e52cSAmir Goldstein 				err = ext4_clear_blocks(handle, inode, this_bh,
957dae1e52cSAmir Goldstein 						        block_to_free, count,
958dae1e52cSAmir Goldstein 						        block_to_free_p, p);
959dae1e52cSAmir Goldstein 				if (err)
960dae1e52cSAmir Goldstein 					break;
961dae1e52cSAmir Goldstein 				block_to_free = nr;
962dae1e52cSAmir Goldstein 				block_to_free_p = p;
963dae1e52cSAmir Goldstein 				count = 1;
964dae1e52cSAmir Goldstein 			}
965dae1e52cSAmir Goldstein 		}
966dae1e52cSAmir Goldstein 	}
967dae1e52cSAmir Goldstein 
968dae1e52cSAmir Goldstein 	if (!err && count > 0)
969dae1e52cSAmir Goldstein 		err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
970dae1e52cSAmir Goldstein 					count, block_to_free_p, p);
971dae1e52cSAmir Goldstein 	if (err < 0)
972dae1e52cSAmir Goldstein 		/* fatal error */
973dae1e52cSAmir Goldstein 		return;
974dae1e52cSAmir Goldstein 
975dae1e52cSAmir Goldstein 	if (this_bh) {
976dae1e52cSAmir Goldstein 		BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
977dae1e52cSAmir Goldstein 
978dae1e52cSAmir Goldstein 		/*
979dae1e52cSAmir Goldstein 		 * The buffer head should have an attached journal head at this
980dae1e52cSAmir Goldstein 		 * point. However, if the data is corrupted and an indirect
981dae1e52cSAmir Goldstein 		 * block pointed to itself, it would have been detached when
982dae1e52cSAmir Goldstein 		 * the block was cleared. Check for this instead of OOPSing.
983dae1e52cSAmir Goldstein 		 */
984dae1e52cSAmir Goldstein 		if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
985dae1e52cSAmir Goldstein 			ext4_handle_dirty_metadata(handle, inode, this_bh);
986dae1e52cSAmir Goldstein 		else
987dae1e52cSAmir Goldstein 			EXT4_ERROR_INODE(inode,
988dae1e52cSAmir Goldstein 					 "circular indirect block detected at "
989dae1e52cSAmir Goldstein 					 "block %llu",
990dae1e52cSAmir Goldstein 				(unsigned long long) this_bh->b_blocknr);
991dae1e52cSAmir Goldstein 	}
992dae1e52cSAmir Goldstein }
993dae1e52cSAmir Goldstein 
994dae1e52cSAmir Goldstein /**
995dae1e52cSAmir Goldstein  *	ext4_free_branches - free an array of branches
996dae1e52cSAmir Goldstein  *	@handle: JBD handle for this transaction
997dae1e52cSAmir Goldstein  *	@inode:	inode we are dealing with
998dae1e52cSAmir Goldstein  *	@parent_bh: the buffer_head which contains *@first and *@last
999dae1e52cSAmir Goldstein  *	@first:	array of block numbers
1000dae1e52cSAmir Goldstein  *	@last:	pointer immediately past the end of array
1001dae1e52cSAmir Goldstein  *	@depth:	depth of the branches to free
1002dae1e52cSAmir Goldstein  *
1003dae1e52cSAmir Goldstein  *	We are freeing all blocks referred from these branches (numbers are
1004dae1e52cSAmir Goldstein  *	stored as little-endian 32-bit) and updating @inode->i_blocks
1005dae1e52cSAmir Goldstein  *	appropriately.
1006dae1e52cSAmir Goldstein  */
ext4_free_branches(handle_t * handle,struct inode * inode,struct buffer_head * parent_bh,__le32 * first,__le32 * last,int depth)1007dae1e52cSAmir Goldstein static void ext4_free_branches(handle_t *handle, struct inode *inode,
1008dae1e52cSAmir Goldstein 			       struct buffer_head *parent_bh,
1009dae1e52cSAmir Goldstein 			       __le32 *first, __le32 *last, int depth)
1010dae1e52cSAmir Goldstein {
1011dae1e52cSAmir Goldstein 	ext4_fsblk_t nr;
1012dae1e52cSAmir Goldstein 	__le32 *p;
1013dae1e52cSAmir Goldstein 
1014dae1e52cSAmir Goldstein 	if (ext4_handle_is_aborted(handle))
1015dae1e52cSAmir Goldstein 		return;
1016dae1e52cSAmir Goldstein 
1017dae1e52cSAmir Goldstein 	if (depth--) {
1018dae1e52cSAmir Goldstein 		struct buffer_head *bh;
1019dae1e52cSAmir Goldstein 		int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1020dae1e52cSAmir Goldstein 		p = last;
1021dae1e52cSAmir Goldstein 		while (--p >= first) {
1022dae1e52cSAmir Goldstein 			nr = le32_to_cpu(*p);
1023dae1e52cSAmir Goldstein 			if (!nr)
1024dae1e52cSAmir Goldstein 				continue;		/* A hole */
1025dae1e52cSAmir Goldstein 
1026ce9f24ccSJan Kara 			if (!ext4_inode_block_valid(inode, nr, 1)) {
1027dae1e52cSAmir Goldstein 				EXT4_ERROR_INODE(inode,
1028dae1e52cSAmir Goldstein 						 "invalid indirect mapped "
1029dae1e52cSAmir Goldstein 						 "block %lu (level %d)",
1030dae1e52cSAmir Goldstein 						 (unsigned long) nr, depth);
1031dae1e52cSAmir Goldstein 				break;
1032dae1e52cSAmir Goldstein 			}
1033dae1e52cSAmir Goldstein 
1034dae1e52cSAmir Goldstein 			/* Go read the buffer for the next level down */
10350a846f49Szhangyi (F) 			bh = ext4_sb_bread(inode->i_sb, nr, 0);
1036dae1e52cSAmir Goldstein 
1037dae1e52cSAmir Goldstein 			/*
1038dae1e52cSAmir Goldstein 			 * A read failure? Report error and clear slot
1039dae1e52cSAmir Goldstein 			 * (should be rare).
1040dae1e52cSAmir Goldstein 			 */
10410a846f49Szhangyi (F) 			if (IS_ERR(bh)) {
10420a846f49Szhangyi (F) 				ext4_error_inode_block(inode, nr, -PTR_ERR(bh),
1043dae1e52cSAmir Goldstein 						       "Read failure");
1044dae1e52cSAmir Goldstein 				continue;
1045dae1e52cSAmir Goldstein 			}
1046dae1e52cSAmir Goldstein 
1047dae1e52cSAmir Goldstein 			/* This zaps the entire block.  Bottom up. */
1048dae1e52cSAmir Goldstein 			BUFFER_TRACE(bh, "free child branches");
1049dae1e52cSAmir Goldstein 			ext4_free_branches(handle, inode, bh,
1050dae1e52cSAmir Goldstein 					(__le32 *) bh->b_data,
1051dae1e52cSAmir Goldstein 					(__le32 *) bh->b_data + addr_per_block,
1052dae1e52cSAmir Goldstein 					depth);
1053dae1e52cSAmir Goldstein 			brelse(bh);
1054dae1e52cSAmir Goldstein 
1055dae1e52cSAmir Goldstein 			/*
1056b483bb77SRandy Dunlap 			 * Everything below this pointer has been
1057dae1e52cSAmir Goldstein 			 * released.  Now let this top-of-subtree go.
1058dae1e52cSAmir Goldstein 			 *
1059dae1e52cSAmir Goldstein 			 * We want the freeing of this indirect block to be
1060dae1e52cSAmir Goldstein 			 * atomic in the journal with the updating of the
1061dae1e52cSAmir Goldstein 			 * bitmap block which owns it.  So make some room in
1062dae1e52cSAmir Goldstein 			 * the journal.
1063dae1e52cSAmir Goldstein 			 *
1064dae1e52cSAmir Goldstein 			 * We zero the parent pointer *after* freeing its
1065dae1e52cSAmir Goldstein 			 * pointee in the bitmaps, so if extend_transaction()
1066dae1e52cSAmir Goldstein 			 * for some reason fails to put the bitmap changes and
1067dae1e52cSAmir Goldstein 			 * the release into the same transaction, recovery
1068dae1e52cSAmir Goldstein 			 * will merely complain about releasing a free block,
1069dae1e52cSAmir Goldstein 			 * rather than leaking blocks.
1070dae1e52cSAmir Goldstein 			 */
1071dae1e52cSAmir Goldstein 			if (ext4_handle_is_aborted(handle))
1072dae1e52cSAmir Goldstein 				return;
1073a4130367SJan Kara 			if (ext4_ind_truncate_ensure_credits(handle, inode,
107483448bdfSJan Kara 					NULL,
107583448bdfSJan Kara 					ext4_free_metadata_revoke_credits(
107683448bdfSJan Kara 							inode->i_sb, 1)) < 0)
1077a4130367SJan Kara 				return;
1078dae1e52cSAmir Goldstein 
1079dae1e52cSAmir Goldstein 			/*
1080dae1e52cSAmir Goldstein 			 * The forget flag here is critical because if
1081dae1e52cSAmir Goldstein 			 * we are journaling (and not doing data
1082dae1e52cSAmir Goldstein 			 * journaling), we have to make sure a revoke
1083dae1e52cSAmir Goldstein 			 * record is written to prevent the journal
1084dae1e52cSAmir Goldstein 			 * replay from overwriting the (former)
1085dae1e52cSAmir Goldstein 			 * indirect block if it gets reallocated as a
1086dae1e52cSAmir Goldstein 			 * data block.  This must happen in the same
1087dae1e52cSAmir Goldstein 			 * transaction where the data blocks are
1088dae1e52cSAmir Goldstein 			 * actually freed.
1089dae1e52cSAmir Goldstein 			 */
1090dae1e52cSAmir Goldstein 			ext4_free_blocks(handle, inode, NULL, nr, 1,
1091dae1e52cSAmir Goldstein 					 EXT4_FREE_BLOCKS_METADATA|
1092dae1e52cSAmir Goldstein 					 EXT4_FREE_BLOCKS_FORGET);
1093dae1e52cSAmir Goldstein 
1094dae1e52cSAmir Goldstein 			if (parent_bh) {
1095dae1e52cSAmir Goldstein 				/*
1096dae1e52cSAmir Goldstein 				 * The block which we have just freed is
1097dae1e52cSAmir Goldstein 				 * pointed to by an indirect block: journal it
1098dae1e52cSAmir Goldstein 				 */
1099dae1e52cSAmir Goldstein 				BUFFER_TRACE(parent_bh, "get_write_access");
1100dae1e52cSAmir Goldstein 				if (!ext4_journal_get_write_access(handle,
1101188c299eSJan Kara 						inode->i_sb, parent_bh,
1102188c299eSJan Kara 						EXT4_JTR_NONE)) {
1103dae1e52cSAmir Goldstein 					*p = 0;
1104dae1e52cSAmir Goldstein 					BUFFER_TRACE(parent_bh,
1105dae1e52cSAmir Goldstein 					"call ext4_handle_dirty_metadata");
1106dae1e52cSAmir Goldstein 					ext4_handle_dirty_metadata(handle,
1107dae1e52cSAmir Goldstein 								   inode,
1108dae1e52cSAmir Goldstein 								   parent_bh);
1109dae1e52cSAmir Goldstein 				}
1110dae1e52cSAmir Goldstein 			}
1111dae1e52cSAmir Goldstein 		}
1112dae1e52cSAmir Goldstein 	} else {
1113dae1e52cSAmir Goldstein 		/* We have reached the bottom of the tree. */
1114dae1e52cSAmir Goldstein 		BUFFER_TRACE(parent_bh, "free data blocks");
1115dae1e52cSAmir Goldstein 		ext4_free_data(handle, inode, parent_bh, first, last);
1116dae1e52cSAmir Goldstein 	}
1117dae1e52cSAmir Goldstein }
1118dae1e52cSAmir Goldstein 
ext4_ind_truncate(handle_t * handle,struct inode * inode)1119819c4920STheodore Ts'o void ext4_ind_truncate(handle_t *handle, struct inode *inode)
1120dae1e52cSAmir Goldstein {
1121dae1e52cSAmir Goldstein 	struct ext4_inode_info *ei = EXT4_I(inode);
1122dae1e52cSAmir Goldstein 	__le32 *i_data = ei->i_data;
1123dae1e52cSAmir Goldstein 	int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1124dae1e52cSAmir Goldstein 	ext4_lblk_t offsets[4];
1125dae1e52cSAmir Goldstein 	Indirect chain[4];
1126dae1e52cSAmir Goldstein 	Indirect *partial;
1127dae1e52cSAmir Goldstein 	__le32 nr = 0;
1128dae1e52cSAmir Goldstein 	int n = 0;
1129dae1e52cSAmir Goldstein 	ext4_lblk_t last_block, max_block;
1130dae1e52cSAmir Goldstein 	unsigned blocksize = inode->i_sb->s_blocksize;
1131dae1e52cSAmir Goldstein 
1132dae1e52cSAmir Goldstein 	last_block = (inode->i_size + blocksize-1)
1133dae1e52cSAmir Goldstein 					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1134dae1e52cSAmir Goldstein 	max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1135dae1e52cSAmir Goldstein 					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1136dae1e52cSAmir Goldstein 
1137dae1e52cSAmir Goldstein 	if (last_block != max_block) {
1138dae1e52cSAmir Goldstein 		n = ext4_block_to_path(inode, last_block, offsets, NULL);
1139dae1e52cSAmir Goldstein 		if (n == 0)
1140819c4920STheodore Ts'o 			return;
1141dae1e52cSAmir Goldstein 	}
1142dae1e52cSAmir Goldstein 
114351865fdaSZheng Liu 	ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block);
1144dae1e52cSAmir Goldstein 
1145dae1e52cSAmir Goldstein 	/*
1146dae1e52cSAmir Goldstein 	 * The orphan list entry will now protect us from any crash which
1147dae1e52cSAmir Goldstein 	 * occurs before the truncate completes, so it is now safe to propagate
1148dae1e52cSAmir Goldstein 	 * the new, shorter inode size (held for now in i_size) into the
1149dae1e52cSAmir Goldstein 	 * on-disk inode. We do this via i_disksize, which is the value which
1150dae1e52cSAmir Goldstein 	 * ext4 *really* writes onto the disk inode.
1151dae1e52cSAmir Goldstein 	 */
1152dae1e52cSAmir Goldstein 	ei->i_disksize = inode->i_size;
1153dae1e52cSAmir Goldstein 
1154dae1e52cSAmir Goldstein 	if (last_block == max_block) {
1155dae1e52cSAmir Goldstein 		/*
1156dae1e52cSAmir Goldstein 		 * It is unnecessary to free any data blocks if last_block is
1157dae1e52cSAmir Goldstein 		 * equal to the indirect block limit.
1158dae1e52cSAmir Goldstein 		 */
1159819c4920STheodore Ts'o 		return;
1160dae1e52cSAmir Goldstein 	} else if (n == 1) {		/* direct blocks */
1161dae1e52cSAmir Goldstein 		ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1162dae1e52cSAmir Goldstein 			       i_data + EXT4_NDIR_BLOCKS);
1163dae1e52cSAmir Goldstein 		goto do_indirects;
1164dae1e52cSAmir Goldstein 	}
1165dae1e52cSAmir Goldstein 
1166dae1e52cSAmir Goldstein 	partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1167dae1e52cSAmir Goldstein 	/* Kill the top of shared branch (not detached) */
1168dae1e52cSAmir Goldstein 	if (nr) {
1169dae1e52cSAmir Goldstein 		if (partial == chain) {
1170dae1e52cSAmir Goldstein 			/* Shared branch grows from the inode */
1171dae1e52cSAmir Goldstein 			ext4_free_branches(handle, inode, NULL,
1172dae1e52cSAmir Goldstein 					   &nr, &nr+1, (chain+n-1) - partial);
1173dae1e52cSAmir Goldstein 			*partial->p = 0;
1174dae1e52cSAmir Goldstein 			/*
1175dae1e52cSAmir Goldstein 			 * We mark the inode dirty prior to restart,
1176dae1e52cSAmir Goldstein 			 * and prior to stop.  No need for it here.
1177dae1e52cSAmir Goldstein 			 */
1178dae1e52cSAmir Goldstein 		} else {
1179dae1e52cSAmir Goldstein 			/* Shared branch grows from an indirect block */
1180dae1e52cSAmir Goldstein 			BUFFER_TRACE(partial->bh, "get_write_access");
1181dae1e52cSAmir Goldstein 			ext4_free_branches(handle, inode, partial->bh,
1182dae1e52cSAmir Goldstein 					partial->p,
1183dae1e52cSAmir Goldstein 					partial->p+1, (chain+n-1) - partial);
1184dae1e52cSAmir Goldstein 		}
1185dae1e52cSAmir Goldstein 	}
1186dae1e52cSAmir Goldstein 	/* Clear the ends of indirect blocks on the shared branch */
1187dae1e52cSAmir Goldstein 	while (partial > chain) {
1188dae1e52cSAmir Goldstein 		ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1189dae1e52cSAmir Goldstein 				   (__le32*)partial->bh->b_data+addr_per_block,
1190dae1e52cSAmir Goldstein 				   (chain+n-1) - partial);
1191dae1e52cSAmir Goldstein 		BUFFER_TRACE(partial->bh, "call brelse");
1192dae1e52cSAmir Goldstein 		brelse(partial->bh);
1193dae1e52cSAmir Goldstein 		partial--;
1194dae1e52cSAmir Goldstein 	}
1195dae1e52cSAmir Goldstein do_indirects:
1196dae1e52cSAmir Goldstein 	/* Kill the remaining (whole) subtrees */
1197dae1e52cSAmir Goldstein 	switch (offsets[0]) {
1198dae1e52cSAmir Goldstein 	default:
1199dae1e52cSAmir Goldstein 		nr = i_data[EXT4_IND_BLOCK];
1200dae1e52cSAmir Goldstein 		if (nr) {
1201dae1e52cSAmir Goldstein 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1202dae1e52cSAmir Goldstein 			i_data[EXT4_IND_BLOCK] = 0;
1203dae1e52cSAmir Goldstein 		}
120470d7ced2SShijie Luo 		fallthrough;
1205dae1e52cSAmir Goldstein 	case EXT4_IND_BLOCK:
1206dae1e52cSAmir Goldstein 		nr = i_data[EXT4_DIND_BLOCK];
1207dae1e52cSAmir Goldstein 		if (nr) {
1208dae1e52cSAmir Goldstein 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1209dae1e52cSAmir Goldstein 			i_data[EXT4_DIND_BLOCK] = 0;
1210dae1e52cSAmir Goldstein 		}
121170d7ced2SShijie Luo 		fallthrough;
1212dae1e52cSAmir Goldstein 	case EXT4_DIND_BLOCK:
1213dae1e52cSAmir Goldstein 		nr = i_data[EXT4_TIND_BLOCK];
1214dae1e52cSAmir Goldstein 		if (nr) {
1215dae1e52cSAmir Goldstein 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1216dae1e52cSAmir Goldstein 			i_data[EXT4_TIND_BLOCK] = 0;
1217dae1e52cSAmir Goldstein 		}
121870d7ced2SShijie Luo 		fallthrough;
1219dae1e52cSAmir Goldstein 	case EXT4_TIND_BLOCK:
1220dae1e52cSAmir Goldstein 		;
1221dae1e52cSAmir Goldstein 	}
1222dae1e52cSAmir Goldstein }
1223dae1e52cSAmir Goldstein 
12244f579ae7SLukas Czerner /**
12254f579ae7SLukas Czerner  *	ext4_ind_remove_space - remove space from the range
12264f579ae7SLukas Czerner  *	@handle: JBD handle for this transaction
12274f579ae7SLukas Czerner  *	@inode:	inode we are dealing with
12284f579ae7SLukas Czerner  *	@start:	First block to remove
12294f579ae7SLukas Czerner  *	@end:	One block after the last block to remove (exclusive)
12304f579ae7SLukas Czerner  *
12314f579ae7SLukas Czerner  *	Free the blocks in the defined range (end is exclusive endpoint of
12324f579ae7SLukas Czerner  *	range). This is used by ext4_punch_hole().
12334f579ae7SLukas Czerner  */
ext4_ind_remove_space(handle_t * handle,struct inode * inode,ext4_lblk_t start,ext4_lblk_t end)12344f579ae7SLukas Czerner int ext4_ind_remove_space(handle_t *handle, struct inode *inode,
12354f579ae7SLukas Czerner 			  ext4_lblk_t start, ext4_lblk_t end)
12368bad6fc8SZheng Liu {
12374f579ae7SLukas Czerner 	struct ext4_inode_info *ei = EXT4_I(inode);
12384f579ae7SLukas Czerner 	__le32 *i_data = ei->i_data;
12398bad6fc8SZheng Liu 	int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
12404f579ae7SLukas Czerner 	ext4_lblk_t offsets[4], offsets2[4];
12414f579ae7SLukas Czerner 	Indirect chain[4], chain2[4];
12424f579ae7SLukas Czerner 	Indirect *partial, *partial2;
12435e86bddaSzhangyi (F) 	Indirect *p = NULL, *p2 = NULL;
12444f579ae7SLukas Czerner 	ext4_lblk_t max_block;
12454f579ae7SLukas Czerner 	__le32 nr = 0, nr2 = 0;
12464f579ae7SLukas Czerner 	int n = 0, n2 = 0;
12474f579ae7SLukas Czerner 	unsigned blocksize = inode->i_sb->s_blocksize;
12488bad6fc8SZheng Liu 
12494f579ae7SLukas Czerner 	max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
12504f579ae7SLukas Czerner 					>> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
12514f579ae7SLukas Czerner 	if (end >= max_block)
12524f579ae7SLukas Czerner 		end = max_block;
12534f579ae7SLukas Czerner 	if ((start >= end) || (start > max_block))
12544f579ae7SLukas Czerner 		return 0;
1255a93cd4cfSJan Kara 
12564f579ae7SLukas Czerner 	n = ext4_block_to_path(inode, start, offsets, NULL);
12574f579ae7SLukas Czerner 	n2 = ext4_block_to_path(inode, end, offsets2, NULL);
12584f579ae7SLukas Czerner 
12594f579ae7SLukas Czerner 	BUG_ON(n > n2);
12604f579ae7SLukas Czerner 
12614f579ae7SLukas Czerner 	if ((n == 1) && (n == n2)) {
12624f579ae7SLukas Czerner 		/* We're punching only within direct block range */
12634f579ae7SLukas Czerner 		ext4_free_data(handle, inode, NULL, i_data + offsets[0],
12644f579ae7SLukas Czerner 			       i_data + offsets2[0]);
12654f579ae7SLukas Czerner 		return 0;
12664f579ae7SLukas Czerner 	} else if (n2 > n) {
12674f579ae7SLukas Czerner 		/*
12684f579ae7SLukas Czerner 		 * Start and end are on a different levels so we're going to
12694f579ae7SLukas Czerner 		 * free partial block at start, and partial block at end of
12704f579ae7SLukas Czerner 		 * the range. If there are some levels in between then
12714f579ae7SLukas Czerner 		 * do_indirects label will take care of that.
12724f579ae7SLukas Czerner 		 */
12734f579ae7SLukas Czerner 
12744f579ae7SLukas Czerner 		if (n == 1) {
12754f579ae7SLukas Czerner 			/*
12764f579ae7SLukas Czerner 			 * Start is at the direct block level, free
12774f579ae7SLukas Czerner 			 * everything to the end of the level.
12784f579ae7SLukas Czerner 			 */
12794f579ae7SLukas Czerner 			ext4_free_data(handle, inode, NULL, i_data + offsets[0],
12804f579ae7SLukas Czerner 				       i_data + EXT4_NDIR_BLOCKS);
12814f579ae7SLukas Czerner 			goto end_range;
12828bad6fc8SZheng Liu 		}
12834f579ae7SLukas Czerner 
12844f579ae7SLukas Czerner 
12855e86bddaSzhangyi (F) 		partial = p = ext4_find_shared(inode, n, offsets, chain, &nr);
12864f579ae7SLukas Czerner 		if (nr) {
12874f579ae7SLukas Czerner 			if (partial == chain) {
12884f579ae7SLukas Czerner 				/* Shared branch grows from the inode */
12894f579ae7SLukas Czerner 				ext4_free_branches(handle, inode, NULL,
12904f579ae7SLukas Czerner 					   &nr, &nr+1, (chain+n-1) - partial);
12914f579ae7SLukas Czerner 				*partial->p = 0;
1292a93cd4cfSJan Kara 			} else {
12934f579ae7SLukas Czerner 				/* Shared branch grows from an indirect block */
12944f579ae7SLukas Czerner 				BUFFER_TRACE(partial->bh, "get_write_access");
12954f579ae7SLukas Czerner 				ext4_free_branches(handle, inode, partial->bh,
12964f579ae7SLukas Czerner 					partial->p,
12974f579ae7SLukas Czerner 					partial->p+1, (chain+n-1) - partial);
1298a93cd4cfSJan Kara 			}
12998bad6fc8SZheng Liu 		}
13008bad6fc8SZheng Liu 
13014f579ae7SLukas Czerner 		/*
13024f579ae7SLukas Czerner 		 * Clear the ends of indirect blocks on the shared branch
13034f579ae7SLukas Czerner 		 * at the start of the range
13044f579ae7SLukas Czerner 		 */
13054f579ae7SLukas Czerner 		while (partial > chain) {
13064f579ae7SLukas Czerner 			ext4_free_branches(handle, inode, partial->bh,
13074f579ae7SLukas Czerner 				partial->p + 1,
13084f579ae7SLukas Czerner 				(__le32 *)partial->bh->b_data+addr_per_block,
13094f579ae7SLukas Czerner 				(chain+n-1) - partial);
13104f579ae7SLukas Czerner 			partial--;
13118bad6fc8SZheng Liu 		}
13128bad6fc8SZheng Liu 
13134f579ae7SLukas Czerner end_range:
13145e86bddaSzhangyi (F) 		partial2 = p2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
13154f579ae7SLukas Czerner 		if (nr2) {
13164f579ae7SLukas Czerner 			if (partial2 == chain2) {
13174f579ae7SLukas Czerner 				/*
13184f579ae7SLukas Czerner 				 * Remember, end is exclusive so here we're at
13194f579ae7SLukas Czerner 				 * the start of the next level we're not going
13204f579ae7SLukas Czerner 				 * to free. Everything was covered by the start
13214f579ae7SLukas Czerner 				 * of the range.
13224f579ae7SLukas Czerner 				 */
13236f30b7e3SOmar Sandoval 				goto do_indirects;
13248bad6fc8SZheng Liu 			}
13254f579ae7SLukas Czerner 		} else {
13264f579ae7SLukas Czerner 			/*
13274f579ae7SLukas Czerner 			 * ext4_find_shared returns Indirect structure which
13284f579ae7SLukas Czerner 			 * points to the last element which should not be
13294f579ae7SLukas Czerner 			 * removed by truncate. But this is end of the range
13304f579ae7SLukas Czerner 			 * in punch_hole so we need to point to the next element
13314f579ae7SLukas Czerner 			 */
13324f579ae7SLukas Czerner 			partial2->p++;
13334f579ae7SLukas Czerner 		}
13344f579ae7SLukas Czerner 
13354f579ae7SLukas Czerner 		/*
13364f579ae7SLukas Czerner 		 * Clear the ends of indirect blocks on the shared branch
13374f579ae7SLukas Czerner 		 * at the end of the range
13384f579ae7SLukas Czerner 		 */
13394f579ae7SLukas Czerner 		while (partial2 > chain2) {
13404f579ae7SLukas Czerner 			ext4_free_branches(handle, inode, partial2->bh,
13414f579ae7SLukas Czerner 					   (__le32 *)partial2->bh->b_data,
13424f579ae7SLukas Czerner 					   partial2->p,
13434f579ae7SLukas Czerner 					   (chain2+n2-1) - partial2);
13444f579ae7SLukas Czerner 			partial2--;
13454f579ae7SLukas Czerner 		}
13464f579ae7SLukas Czerner 		goto do_indirects;
13474f579ae7SLukas Czerner 	}
13484f579ae7SLukas Czerner 
13494f579ae7SLukas Czerner 	/* Punch happened within the same level (n == n2) */
13505e86bddaSzhangyi (F) 	partial = p = ext4_find_shared(inode, n, offsets, chain, &nr);
13515e86bddaSzhangyi (F) 	partial2 = p2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
13526f30b7e3SOmar Sandoval 
13536f30b7e3SOmar Sandoval 	/* Free top, but only if partial2 isn't its subtree. */
13546f30b7e3SOmar Sandoval 	if (nr) {
13556f30b7e3SOmar Sandoval 		int level = min(partial - chain, partial2 - chain2);
13566f30b7e3SOmar Sandoval 		int i;
13576f30b7e3SOmar Sandoval 		int subtree = 1;
13586f30b7e3SOmar Sandoval 
13596f30b7e3SOmar Sandoval 		for (i = 0; i <= level; i++) {
13606f30b7e3SOmar Sandoval 			if (offsets[i] != offsets2[i]) {
13616f30b7e3SOmar Sandoval 				subtree = 0;
13626f30b7e3SOmar Sandoval 				break;
13636f30b7e3SOmar Sandoval 			}
13646f30b7e3SOmar Sandoval 		}
13656f30b7e3SOmar Sandoval 
13666f30b7e3SOmar Sandoval 		if (!subtree) {
13676f30b7e3SOmar Sandoval 			if (partial == chain) {
13686f30b7e3SOmar Sandoval 				/* Shared branch grows from the inode */
13696f30b7e3SOmar Sandoval 				ext4_free_branches(handle, inode, NULL,
13706f30b7e3SOmar Sandoval 						   &nr, &nr+1,
13716f30b7e3SOmar Sandoval 						   (chain+n-1) - partial);
13726f30b7e3SOmar Sandoval 				*partial->p = 0;
13736f30b7e3SOmar Sandoval 			} else {
13746f30b7e3SOmar Sandoval 				/* Shared branch grows from an indirect block */
13756f30b7e3SOmar Sandoval 				BUFFER_TRACE(partial->bh, "get_write_access");
13766f30b7e3SOmar Sandoval 				ext4_free_branches(handle, inode, partial->bh,
13776f30b7e3SOmar Sandoval 						   partial->p,
13786f30b7e3SOmar Sandoval 						   partial->p+1,
13796f30b7e3SOmar Sandoval 						   (chain+n-1) - partial);
13806f30b7e3SOmar Sandoval 			}
13816f30b7e3SOmar Sandoval 		}
13826f30b7e3SOmar Sandoval 	}
13836f30b7e3SOmar Sandoval 
13846f30b7e3SOmar Sandoval 	if (!nr2) {
13854f579ae7SLukas Czerner 		/*
13864f579ae7SLukas Czerner 		 * ext4_find_shared returns Indirect structure which
13874f579ae7SLukas Czerner 		 * points to the last element which should not be
13884f579ae7SLukas Czerner 		 * removed by truncate. But this is end of the range
13894f579ae7SLukas Czerner 		 * in punch_hole so we need to point to the next element
13904f579ae7SLukas Czerner 		 */
13914f579ae7SLukas Czerner 		partial2->p++;
13926f30b7e3SOmar Sandoval 	}
13936f30b7e3SOmar Sandoval 
13946f30b7e3SOmar Sandoval 	while (partial > chain || partial2 > chain2) {
13956f30b7e3SOmar Sandoval 		int depth = (chain+n-1) - partial;
13966f30b7e3SOmar Sandoval 		int depth2 = (chain2+n2-1) - partial2;
13976f30b7e3SOmar Sandoval 
13986f30b7e3SOmar Sandoval 		if (partial > chain && partial2 > chain2 &&
13996f30b7e3SOmar Sandoval 		    partial->bh->b_blocknr == partial2->bh->b_blocknr) {
14006f30b7e3SOmar Sandoval 			/*
14016f30b7e3SOmar Sandoval 			 * We've converged on the same block. Clear the range,
14026f30b7e3SOmar Sandoval 			 * then we're done.
14036f30b7e3SOmar Sandoval 			 */
14044f579ae7SLukas Czerner 			ext4_free_branches(handle, inode, partial->bh,
14054f579ae7SLukas Czerner 					   partial->p + 1,
14064f579ae7SLukas Czerner 					   partial2->p,
14074f579ae7SLukas Czerner 					   (chain+n-1) - partial);
14085e86bddaSzhangyi (F) 			goto cleanup;
14094f579ae7SLukas Czerner 		}
14106f30b7e3SOmar Sandoval 
14114f579ae7SLukas Czerner 		/*
14126f30b7e3SOmar Sandoval 		 * The start and end partial branches may not be at the same
14136f30b7e3SOmar Sandoval 		 * level even though the punch happened within one level. So, we
14146f30b7e3SOmar Sandoval 		 * give them a chance to arrive at the same level, then walk
14156f30b7e3SOmar Sandoval 		 * them in step with each other until we converge on the same
14166f30b7e3SOmar Sandoval 		 * block.
14174f579ae7SLukas Czerner 		 */
14186f30b7e3SOmar Sandoval 		if (partial > chain && depth <= depth2) {
14194f579ae7SLukas Czerner 			ext4_free_branches(handle, inode, partial->bh,
14204f579ae7SLukas Czerner 					   partial->p + 1,
14214f579ae7SLukas Czerner 					   (__le32 *)partial->bh->b_data+addr_per_block,
14224f579ae7SLukas Czerner 					   (chain+n-1) - partial);
14234f579ae7SLukas Czerner 			partial--;
14244f579ae7SLukas Czerner 		}
14256f30b7e3SOmar Sandoval 		if (partial2 > chain2 && depth2 <= depth) {
14264f579ae7SLukas Czerner 			ext4_free_branches(handle, inode, partial2->bh,
14274f579ae7SLukas Czerner 					   (__le32 *)partial2->bh->b_data,
14284f579ae7SLukas Czerner 					   partial2->p,
14296f30b7e3SOmar Sandoval 					   (chain2+n2-1) - partial2);
14304f579ae7SLukas Czerner 			partial2--;
14318bad6fc8SZheng Liu 		}
14328bad6fc8SZheng Liu 	}
14335e86bddaSzhangyi (F) 
14345e86bddaSzhangyi (F) cleanup:
14355e86bddaSzhangyi (F) 	while (p && p > chain) {
14365e86bddaSzhangyi (F) 		BUFFER_TRACE(p->bh, "call brelse");
14375e86bddaSzhangyi (F) 		brelse(p->bh);
14385e86bddaSzhangyi (F) 		p--;
14395e86bddaSzhangyi (F) 	}
14405e86bddaSzhangyi (F) 	while (p2 && p2 > chain2) {
14415e86bddaSzhangyi (F) 		BUFFER_TRACE(p2->bh, "call brelse");
14425e86bddaSzhangyi (F) 		brelse(p2->bh);
14435e86bddaSzhangyi (F) 		p2--;
14445e86bddaSzhangyi (F) 	}
14456f30b7e3SOmar Sandoval 	return 0;
14468bad6fc8SZheng Liu 
14474f579ae7SLukas Czerner do_indirects:
14484f579ae7SLukas Czerner 	/* Kill the remaining (whole) subtrees */
14494f579ae7SLukas Czerner 	switch (offsets[0]) {
14504f579ae7SLukas Czerner 	default:
14514f579ae7SLukas Czerner 		if (++n >= n2)
14525e86bddaSzhangyi (F) 			break;
14534f579ae7SLukas Czerner 		nr = i_data[EXT4_IND_BLOCK];
14544f579ae7SLukas Czerner 		if (nr) {
14554f579ae7SLukas Czerner 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
14564f579ae7SLukas Czerner 			i_data[EXT4_IND_BLOCK] = 0;
14578bad6fc8SZheng Liu 		}
145870d7ced2SShijie Luo 		fallthrough;
14594f579ae7SLukas Czerner 	case EXT4_IND_BLOCK:
14604f579ae7SLukas Czerner 		if (++n >= n2)
14615e86bddaSzhangyi (F) 			break;
14624f579ae7SLukas Czerner 		nr = i_data[EXT4_DIND_BLOCK];
14634f579ae7SLukas Czerner 		if (nr) {
14644f579ae7SLukas Czerner 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
14654f579ae7SLukas Czerner 			i_data[EXT4_DIND_BLOCK] = 0;
14664f579ae7SLukas Czerner 		}
146770d7ced2SShijie Luo 		fallthrough;
14684f579ae7SLukas Czerner 	case EXT4_DIND_BLOCK:
14694f579ae7SLukas Czerner 		if (++n >= n2)
14705e86bddaSzhangyi (F) 			break;
14714f579ae7SLukas Czerner 		nr = i_data[EXT4_TIND_BLOCK];
14724f579ae7SLukas Czerner 		if (nr) {
14734f579ae7SLukas Czerner 			ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
14744f579ae7SLukas Czerner 			i_data[EXT4_TIND_BLOCK] = 0;
14754f579ae7SLukas Czerner 		}
147670d7ced2SShijie Luo 		fallthrough;
14774f579ae7SLukas Czerner 	case EXT4_TIND_BLOCK:
14784f579ae7SLukas Czerner 		;
14794f579ae7SLukas Czerner 	}
14805e86bddaSzhangyi (F) 	goto cleanup;
14814f579ae7SLukas Czerner }
1482