1dae1e52cSAmir Goldstein /* 2dae1e52cSAmir Goldstein * linux/fs/ext4/indirect.c 3dae1e52cSAmir Goldstein * 4dae1e52cSAmir Goldstein * from 5dae1e52cSAmir Goldstein * 6dae1e52cSAmir Goldstein * linux/fs/ext4/inode.c 7dae1e52cSAmir Goldstein * 8dae1e52cSAmir Goldstein * Copyright (C) 1992, 1993, 1994, 1995 9dae1e52cSAmir Goldstein * Remy Card (card@masi.ibp.fr) 10dae1e52cSAmir Goldstein * Laboratoire MASI - Institut Blaise Pascal 11dae1e52cSAmir Goldstein * Universite Pierre et Marie Curie (Paris VI) 12dae1e52cSAmir Goldstein * 13dae1e52cSAmir Goldstein * from 14dae1e52cSAmir Goldstein * 15dae1e52cSAmir Goldstein * linux/fs/minix/inode.c 16dae1e52cSAmir Goldstein * 17dae1e52cSAmir Goldstein * Copyright (C) 1991, 1992 Linus Torvalds 18dae1e52cSAmir Goldstein * 19dae1e52cSAmir Goldstein * Goal-directed block allocation by Stephen Tweedie 20dae1e52cSAmir Goldstein * (sct@redhat.com), 1993, 1998 21dae1e52cSAmir Goldstein */ 22dae1e52cSAmir Goldstein 23dae1e52cSAmir Goldstein #include "ext4_jbd2.h" 24dae1e52cSAmir Goldstein #include "truncate.h" 25c94c2acfSMatthew Wilcox #include <linux/dax.h> 26e2e40f2cSChristoph Hellwig #include <linux/uio.h> 27dae1e52cSAmir Goldstein 28dae1e52cSAmir Goldstein #include <trace/events/ext4.h> 29dae1e52cSAmir Goldstein 30dae1e52cSAmir Goldstein typedef struct { 31dae1e52cSAmir Goldstein __le32 *p; 32dae1e52cSAmir Goldstein __le32 key; 33dae1e52cSAmir Goldstein struct buffer_head *bh; 34dae1e52cSAmir Goldstein } Indirect; 35dae1e52cSAmir Goldstein 36dae1e52cSAmir Goldstein static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v) 37dae1e52cSAmir Goldstein { 38dae1e52cSAmir Goldstein p->key = *(p->p = v); 39dae1e52cSAmir Goldstein p->bh = bh; 40dae1e52cSAmir Goldstein } 41dae1e52cSAmir Goldstein 42dae1e52cSAmir Goldstein /** 43dae1e52cSAmir Goldstein * ext4_block_to_path - parse the block number into array of offsets 44dae1e52cSAmir Goldstein * @inode: inode in question (we are only interested in its superblock) 45dae1e52cSAmir Goldstein * @i_block: block number to be parsed 46dae1e52cSAmir Goldstein * @offsets: array to store the offsets in 47dae1e52cSAmir Goldstein * @boundary: set this non-zero if the referred-to block is likely to be 48dae1e52cSAmir Goldstein * followed (on disk) by an indirect block. 49dae1e52cSAmir Goldstein * 50dae1e52cSAmir Goldstein * To store the locations of file's data ext4 uses a data structure common 51dae1e52cSAmir Goldstein * for UNIX filesystems - tree of pointers anchored in the inode, with 52dae1e52cSAmir Goldstein * data blocks at leaves and indirect blocks in intermediate nodes. 53dae1e52cSAmir Goldstein * This function translates the block number into path in that tree - 54dae1e52cSAmir Goldstein * return value is the path length and @offsets[n] is the offset of 55dae1e52cSAmir Goldstein * pointer to (n+1)th node in the nth one. If @block is out of range 56dae1e52cSAmir Goldstein * (negative or too large) warning is printed and zero returned. 57dae1e52cSAmir Goldstein * 58dae1e52cSAmir Goldstein * Note: function doesn't find node addresses, so no IO is needed. All 59dae1e52cSAmir Goldstein * we need to know is the capacity of indirect blocks (taken from the 60dae1e52cSAmir Goldstein * inode->i_sb). 61dae1e52cSAmir Goldstein */ 62dae1e52cSAmir Goldstein 63dae1e52cSAmir Goldstein /* 64dae1e52cSAmir Goldstein * Portability note: the last comparison (check that we fit into triple 65dae1e52cSAmir Goldstein * indirect block) is spelled differently, because otherwise on an 66dae1e52cSAmir Goldstein * architecture with 32-bit longs and 8Kb pages we might get into trouble 67dae1e52cSAmir Goldstein * if our filesystem had 8Kb blocks. We might use long long, but that would 68dae1e52cSAmir Goldstein * kill us on x86. Oh, well, at least the sign propagation does not matter - 69dae1e52cSAmir Goldstein * i_block would have to be negative in the very beginning, so we would not 70dae1e52cSAmir Goldstein * get there at all. 71dae1e52cSAmir Goldstein */ 72dae1e52cSAmir Goldstein 73dae1e52cSAmir Goldstein static int ext4_block_to_path(struct inode *inode, 74dae1e52cSAmir Goldstein ext4_lblk_t i_block, 75dae1e52cSAmir Goldstein ext4_lblk_t offsets[4], int *boundary) 76dae1e52cSAmir Goldstein { 77dae1e52cSAmir Goldstein int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb); 78dae1e52cSAmir Goldstein int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb); 79dae1e52cSAmir Goldstein const long direct_blocks = EXT4_NDIR_BLOCKS, 80dae1e52cSAmir Goldstein indirect_blocks = ptrs, 81dae1e52cSAmir Goldstein double_blocks = (1 << (ptrs_bits * 2)); 82dae1e52cSAmir Goldstein int n = 0; 83dae1e52cSAmir Goldstein int final = 0; 84dae1e52cSAmir Goldstein 85dae1e52cSAmir Goldstein if (i_block < direct_blocks) { 86dae1e52cSAmir Goldstein offsets[n++] = i_block; 87dae1e52cSAmir Goldstein final = direct_blocks; 88dae1e52cSAmir Goldstein } else if ((i_block -= direct_blocks) < indirect_blocks) { 89dae1e52cSAmir Goldstein offsets[n++] = EXT4_IND_BLOCK; 90dae1e52cSAmir Goldstein offsets[n++] = i_block; 91dae1e52cSAmir Goldstein final = ptrs; 92dae1e52cSAmir Goldstein } else if ((i_block -= indirect_blocks) < double_blocks) { 93dae1e52cSAmir Goldstein offsets[n++] = EXT4_DIND_BLOCK; 94dae1e52cSAmir Goldstein offsets[n++] = i_block >> ptrs_bits; 95dae1e52cSAmir Goldstein offsets[n++] = i_block & (ptrs - 1); 96dae1e52cSAmir Goldstein final = ptrs; 97dae1e52cSAmir Goldstein } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) { 98dae1e52cSAmir Goldstein offsets[n++] = EXT4_TIND_BLOCK; 99dae1e52cSAmir Goldstein offsets[n++] = i_block >> (ptrs_bits * 2); 100dae1e52cSAmir Goldstein offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1); 101dae1e52cSAmir Goldstein offsets[n++] = i_block & (ptrs - 1); 102dae1e52cSAmir Goldstein final = ptrs; 103dae1e52cSAmir Goldstein } else { 104dae1e52cSAmir Goldstein ext4_warning(inode->i_sb, "block %lu > max in inode %lu", 105dae1e52cSAmir Goldstein i_block + direct_blocks + 106dae1e52cSAmir Goldstein indirect_blocks + double_blocks, inode->i_ino); 107dae1e52cSAmir Goldstein } 108dae1e52cSAmir Goldstein if (boundary) 109dae1e52cSAmir Goldstein *boundary = final - 1 - (i_block & (ptrs - 1)); 110dae1e52cSAmir Goldstein return n; 111dae1e52cSAmir Goldstein } 112dae1e52cSAmir Goldstein 113dae1e52cSAmir Goldstein /** 114dae1e52cSAmir Goldstein * ext4_get_branch - read the chain of indirect blocks leading to data 115dae1e52cSAmir Goldstein * @inode: inode in question 116dae1e52cSAmir Goldstein * @depth: depth of the chain (1 - direct pointer, etc.) 117dae1e52cSAmir Goldstein * @offsets: offsets of pointers in inode/indirect blocks 118dae1e52cSAmir Goldstein * @chain: place to store the result 119dae1e52cSAmir Goldstein * @err: here we store the error value 120dae1e52cSAmir Goldstein * 121dae1e52cSAmir Goldstein * Function fills the array of triples <key, p, bh> and returns %NULL 122dae1e52cSAmir Goldstein * if everything went OK or the pointer to the last filled triple 123dae1e52cSAmir Goldstein * (incomplete one) otherwise. Upon the return chain[i].key contains 124dae1e52cSAmir Goldstein * the number of (i+1)-th block in the chain (as it is stored in memory, 125dae1e52cSAmir Goldstein * i.e. little-endian 32-bit), chain[i].p contains the address of that 126dae1e52cSAmir Goldstein * number (it points into struct inode for i==0 and into the bh->b_data 127dae1e52cSAmir Goldstein * for i>0) and chain[i].bh points to the buffer_head of i-th indirect 128dae1e52cSAmir Goldstein * block for i>0 and NULL for i==0. In other words, it holds the block 129dae1e52cSAmir Goldstein * numbers of the chain, addresses they were taken from (and where we can 130dae1e52cSAmir Goldstein * verify that chain did not change) and buffer_heads hosting these 131dae1e52cSAmir Goldstein * numbers. 132dae1e52cSAmir Goldstein * 133dae1e52cSAmir Goldstein * Function stops when it stumbles upon zero pointer (absent block) 134dae1e52cSAmir Goldstein * (pointer to last triple returned, *@err == 0) 135dae1e52cSAmir Goldstein * or when it gets an IO error reading an indirect block 136dae1e52cSAmir Goldstein * (ditto, *@err == -EIO) 137dae1e52cSAmir Goldstein * or when it reads all @depth-1 indirect blocks successfully and finds 138dae1e52cSAmir Goldstein * the whole chain, all way to the data (returns %NULL, *err == 0). 139dae1e52cSAmir Goldstein * 140dae1e52cSAmir Goldstein * Need to be called with 141dae1e52cSAmir Goldstein * down_read(&EXT4_I(inode)->i_data_sem) 142dae1e52cSAmir Goldstein */ 143dae1e52cSAmir Goldstein static Indirect *ext4_get_branch(struct inode *inode, int depth, 144dae1e52cSAmir Goldstein ext4_lblk_t *offsets, 145dae1e52cSAmir Goldstein Indirect chain[4], int *err) 146dae1e52cSAmir Goldstein { 147dae1e52cSAmir Goldstein struct super_block *sb = inode->i_sb; 148dae1e52cSAmir Goldstein Indirect *p = chain; 149dae1e52cSAmir Goldstein struct buffer_head *bh; 150860d21e2STheodore Ts'o int ret = -EIO; 151dae1e52cSAmir Goldstein 152dae1e52cSAmir Goldstein *err = 0; 153dae1e52cSAmir Goldstein /* i_data is not going away, no lock needed */ 154dae1e52cSAmir Goldstein add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets); 155dae1e52cSAmir Goldstein if (!p->key) 156dae1e52cSAmir Goldstein goto no_block; 157dae1e52cSAmir Goldstein while (--depth) { 158dae1e52cSAmir Goldstein bh = sb_getblk(sb, le32_to_cpu(p->key)); 159860d21e2STheodore Ts'o if (unlikely(!bh)) { 160860d21e2STheodore Ts'o ret = -ENOMEM; 161dae1e52cSAmir Goldstein goto failure; 162860d21e2STheodore Ts'o } 163dae1e52cSAmir Goldstein 164dae1e52cSAmir Goldstein if (!bh_uptodate_or_lock(bh)) { 165dae1e52cSAmir Goldstein if (bh_submit_read(bh) < 0) { 166dae1e52cSAmir Goldstein put_bh(bh); 167dae1e52cSAmir Goldstein goto failure; 168dae1e52cSAmir Goldstein } 169dae1e52cSAmir Goldstein /* validate block references */ 170dae1e52cSAmir Goldstein if (ext4_check_indirect_blockref(inode, bh)) { 171dae1e52cSAmir Goldstein put_bh(bh); 172dae1e52cSAmir Goldstein goto failure; 173dae1e52cSAmir Goldstein } 174dae1e52cSAmir Goldstein } 175dae1e52cSAmir Goldstein 176dae1e52cSAmir Goldstein add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets); 177dae1e52cSAmir Goldstein /* Reader: end */ 178dae1e52cSAmir Goldstein if (!p->key) 179dae1e52cSAmir Goldstein goto no_block; 180dae1e52cSAmir Goldstein } 181dae1e52cSAmir Goldstein return NULL; 182dae1e52cSAmir Goldstein 183dae1e52cSAmir Goldstein failure: 184860d21e2STheodore Ts'o *err = ret; 185dae1e52cSAmir Goldstein no_block: 186dae1e52cSAmir Goldstein return p; 187dae1e52cSAmir Goldstein } 188dae1e52cSAmir Goldstein 189dae1e52cSAmir Goldstein /** 190dae1e52cSAmir Goldstein * ext4_find_near - find a place for allocation with sufficient locality 191dae1e52cSAmir Goldstein * @inode: owner 192dae1e52cSAmir Goldstein * @ind: descriptor of indirect block. 193dae1e52cSAmir Goldstein * 194dae1e52cSAmir Goldstein * This function returns the preferred place for block allocation. 195dae1e52cSAmir Goldstein * It is used when heuristic for sequential allocation fails. 196dae1e52cSAmir Goldstein * Rules are: 197dae1e52cSAmir Goldstein * + if there is a block to the left of our position - allocate near it. 198dae1e52cSAmir Goldstein * + if pointer will live in indirect block - allocate near that block. 199dae1e52cSAmir Goldstein * + if pointer will live in inode - allocate in the same 200dae1e52cSAmir Goldstein * cylinder group. 201dae1e52cSAmir Goldstein * 202dae1e52cSAmir Goldstein * In the latter case we colour the starting block by the callers PID to 203dae1e52cSAmir Goldstein * prevent it from clashing with concurrent allocations for a different inode 204dae1e52cSAmir Goldstein * in the same block group. The PID is used here so that functionally related 205dae1e52cSAmir Goldstein * files will be close-by on-disk. 206dae1e52cSAmir Goldstein * 207dae1e52cSAmir Goldstein * Caller must make sure that @ind is valid and will stay that way. 208dae1e52cSAmir Goldstein */ 209dae1e52cSAmir Goldstein static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind) 210dae1e52cSAmir Goldstein { 211dae1e52cSAmir Goldstein struct ext4_inode_info *ei = EXT4_I(inode); 212dae1e52cSAmir Goldstein __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data; 213dae1e52cSAmir Goldstein __le32 *p; 214dae1e52cSAmir Goldstein 215dae1e52cSAmir Goldstein /* Try to find previous block */ 216dae1e52cSAmir Goldstein for (p = ind->p - 1; p >= start; p--) { 217dae1e52cSAmir Goldstein if (*p) 218dae1e52cSAmir Goldstein return le32_to_cpu(*p); 219dae1e52cSAmir Goldstein } 220dae1e52cSAmir Goldstein 221dae1e52cSAmir Goldstein /* No such thing, so let's try location of indirect block */ 222dae1e52cSAmir Goldstein if (ind->bh) 223dae1e52cSAmir Goldstein return ind->bh->b_blocknr; 224dae1e52cSAmir Goldstein 225dae1e52cSAmir Goldstein /* 226dae1e52cSAmir Goldstein * It is going to be referred to from the inode itself? OK, just put it 227dae1e52cSAmir Goldstein * into the same cylinder group then. 228dae1e52cSAmir Goldstein */ 229f86186b4SEric Sandeen return ext4_inode_to_goal_block(inode); 230dae1e52cSAmir Goldstein } 231dae1e52cSAmir Goldstein 232dae1e52cSAmir Goldstein /** 233dae1e52cSAmir Goldstein * ext4_find_goal - find a preferred place for allocation. 234dae1e52cSAmir Goldstein * @inode: owner 235dae1e52cSAmir Goldstein * @block: block we want 236dae1e52cSAmir Goldstein * @partial: pointer to the last triple within a chain 237dae1e52cSAmir Goldstein * 238dae1e52cSAmir Goldstein * Normally this function find the preferred place for block allocation, 239dae1e52cSAmir Goldstein * returns it. 240dae1e52cSAmir Goldstein * Because this is only used for non-extent files, we limit the block nr 241dae1e52cSAmir Goldstein * to 32 bits. 242dae1e52cSAmir Goldstein */ 243dae1e52cSAmir Goldstein static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block, 244dae1e52cSAmir Goldstein Indirect *partial) 245dae1e52cSAmir Goldstein { 246dae1e52cSAmir Goldstein ext4_fsblk_t goal; 247dae1e52cSAmir Goldstein 248dae1e52cSAmir Goldstein /* 249dae1e52cSAmir Goldstein * XXX need to get goal block from mballoc's data structures 250dae1e52cSAmir Goldstein */ 251dae1e52cSAmir Goldstein 252dae1e52cSAmir Goldstein goal = ext4_find_near(inode, partial); 253dae1e52cSAmir Goldstein goal = goal & EXT4_MAX_BLOCK_FILE_PHYS; 254dae1e52cSAmir Goldstein return goal; 255dae1e52cSAmir Goldstein } 256dae1e52cSAmir Goldstein 257dae1e52cSAmir Goldstein /** 258dae1e52cSAmir Goldstein * ext4_blks_to_allocate - Look up the block map and count the number 259dae1e52cSAmir Goldstein * of direct blocks need to be allocated for the given branch. 260dae1e52cSAmir Goldstein * 261dae1e52cSAmir Goldstein * @branch: chain of indirect blocks 262dae1e52cSAmir Goldstein * @k: number of blocks need for indirect blocks 263dae1e52cSAmir Goldstein * @blks: number of data blocks to be mapped. 264dae1e52cSAmir Goldstein * @blocks_to_boundary: the offset in the indirect block 265dae1e52cSAmir Goldstein * 266dae1e52cSAmir Goldstein * return the total number of blocks to be allocate, including the 267dae1e52cSAmir Goldstein * direct and indirect blocks. 268dae1e52cSAmir Goldstein */ 269dae1e52cSAmir Goldstein static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks, 270dae1e52cSAmir Goldstein int blocks_to_boundary) 271dae1e52cSAmir Goldstein { 272dae1e52cSAmir Goldstein unsigned int count = 0; 273dae1e52cSAmir Goldstein 274dae1e52cSAmir Goldstein /* 275dae1e52cSAmir Goldstein * Simple case, [t,d]Indirect block(s) has not allocated yet 276dae1e52cSAmir Goldstein * then it's clear blocks on that path have not allocated 277dae1e52cSAmir Goldstein */ 278dae1e52cSAmir Goldstein if (k > 0) { 279dae1e52cSAmir Goldstein /* right now we don't handle cross boundary allocation */ 280dae1e52cSAmir Goldstein if (blks < blocks_to_boundary + 1) 281dae1e52cSAmir Goldstein count += blks; 282dae1e52cSAmir Goldstein else 283dae1e52cSAmir Goldstein count += blocks_to_boundary + 1; 284dae1e52cSAmir Goldstein return count; 285dae1e52cSAmir Goldstein } 286dae1e52cSAmir Goldstein 287dae1e52cSAmir Goldstein count++; 288dae1e52cSAmir Goldstein while (count < blks && count <= blocks_to_boundary && 289dae1e52cSAmir Goldstein le32_to_cpu(*(branch[0].p + count)) == 0) { 290dae1e52cSAmir Goldstein count++; 291dae1e52cSAmir Goldstein } 292dae1e52cSAmir Goldstein return count; 293dae1e52cSAmir Goldstein } 294dae1e52cSAmir Goldstein 295dae1e52cSAmir Goldstein /** 296dae1e52cSAmir Goldstein * ext4_alloc_branch - allocate and set up a chain of blocks. 297dae1e52cSAmir Goldstein * @handle: handle for this transaction 298dae1e52cSAmir Goldstein * @inode: owner 299dae1e52cSAmir Goldstein * @indirect_blks: number of allocated indirect blocks 300dae1e52cSAmir Goldstein * @blks: number of allocated direct blocks 301dae1e52cSAmir Goldstein * @goal: preferred place for allocation 302dae1e52cSAmir Goldstein * @offsets: offsets (in the blocks) to store the pointers to next. 303dae1e52cSAmir Goldstein * @branch: place to store the chain in. 304dae1e52cSAmir Goldstein * 305dae1e52cSAmir Goldstein * This function allocates blocks, zeroes out all but the last one, 306dae1e52cSAmir Goldstein * links them into chain and (if we are synchronous) writes them to disk. 307dae1e52cSAmir Goldstein * In other words, it prepares a branch that can be spliced onto the 308dae1e52cSAmir Goldstein * inode. It stores the information about that chain in the branch[], in 309dae1e52cSAmir Goldstein * the same format as ext4_get_branch() would do. We are calling it after 310dae1e52cSAmir Goldstein * we had read the existing part of chain and partial points to the last 311dae1e52cSAmir Goldstein * triple of that (one with zero ->key). Upon the exit we have the same 312dae1e52cSAmir Goldstein * picture as after the successful ext4_get_block(), except that in one 313dae1e52cSAmir Goldstein * place chain is disconnected - *branch->p is still zero (we did not 314dae1e52cSAmir Goldstein * set the last link), but branch->key contains the number that should 315dae1e52cSAmir Goldstein * be placed into *branch->p to fill that gap. 316dae1e52cSAmir Goldstein * 317dae1e52cSAmir Goldstein * If allocation fails we free all blocks we've allocated (and forget 318dae1e52cSAmir Goldstein * their buffer_heads) and return the error value the from failed 319dae1e52cSAmir Goldstein * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain 320dae1e52cSAmir Goldstein * as described above and return 0. 321dae1e52cSAmir Goldstein */ 322a5211002STheodore Ts'o static int ext4_alloc_branch(handle_t *handle, 323a5211002STheodore Ts'o struct ext4_allocation_request *ar, 324a5211002STheodore Ts'o int indirect_blks, ext4_lblk_t *offsets, 325a5211002STheodore Ts'o Indirect *branch) 326dae1e52cSAmir Goldstein { 327dae1e52cSAmir Goldstein struct buffer_head * bh; 328781f143eSTheodore Ts'o ext4_fsblk_t b, new_blocks[4]; 329781f143eSTheodore Ts'o __le32 *p; 330781f143eSTheodore Ts'o int i, j, err, len = 1; 331dae1e52cSAmir Goldstein 332781f143eSTheodore Ts'o for (i = 0; i <= indirect_blks; i++) { 333781f143eSTheodore Ts'o if (i == indirect_blks) { 334a5211002STheodore Ts'o new_blocks[i] = ext4_mb_new_blocks(handle, ar, &err); 335781f143eSTheodore Ts'o } else 336a5211002STheodore Ts'o ar->goal = new_blocks[i] = ext4_new_meta_blocks(handle, 337e3cf5d5dSTheodore Ts'o ar->inode, ar->goal, 338e3cf5d5dSTheodore Ts'o ar->flags & EXT4_MB_DELALLOC_RESERVED, 339e3cf5d5dSTheodore Ts'o NULL, &err); 340781f143eSTheodore Ts'o if (err) { 341781f143eSTheodore Ts'o i--; 342781f143eSTheodore Ts'o goto failed; 343781f143eSTheodore Ts'o } 344781f143eSTheodore Ts'o branch[i].key = cpu_to_le32(new_blocks[i]); 345781f143eSTheodore Ts'o if (i == 0) 346781f143eSTheodore Ts'o continue; 347781f143eSTheodore Ts'o 348a5211002STheodore Ts'o bh = branch[i].bh = sb_getblk(ar->inode->i_sb, new_blocks[i-1]); 349dae1e52cSAmir Goldstein if (unlikely(!bh)) { 350860d21e2STheodore Ts'o err = -ENOMEM; 351dae1e52cSAmir Goldstein goto failed; 352dae1e52cSAmir Goldstein } 353dae1e52cSAmir Goldstein lock_buffer(bh); 354dae1e52cSAmir Goldstein BUFFER_TRACE(bh, "call get_create_access"); 355dae1e52cSAmir Goldstein err = ext4_journal_get_create_access(handle, bh); 356dae1e52cSAmir Goldstein if (err) { 357dae1e52cSAmir Goldstein unlock_buffer(bh); 358dae1e52cSAmir Goldstein goto failed; 359dae1e52cSAmir Goldstein } 360dae1e52cSAmir Goldstein 361781f143eSTheodore Ts'o memset(bh->b_data, 0, bh->b_size); 362781f143eSTheodore Ts'o p = branch[i].p = (__le32 *) bh->b_data + offsets[i]; 363781f143eSTheodore Ts'o b = new_blocks[i]; 364781f143eSTheodore Ts'o 365781f143eSTheodore Ts'o if (i == indirect_blks) 366a5211002STheodore Ts'o len = ar->len; 367781f143eSTheodore Ts'o for (j = 0; j < len; j++) 368781f143eSTheodore Ts'o *p++ = cpu_to_le32(b++); 369781f143eSTheodore Ts'o 370dae1e52cSAmir Goldstein BUFFER_TRACE(bh, "marking uptodate"); 371dae1e52cSAmir Goldstein set_buffer_uptodate(bh); 372dae1e52cSAmir Goldstein unlock_buffer(bh); 373dae1e52cSAmir Goldstein 374dae1e52cSAmir Goldstein BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); 375a5211002STheodore Ts'o err = ext4_handle_dirty_metadata(handle, ar->inode, bh); 376dae1e52cSAmir Goldstein if (err) 377dae1e52cSAmir Goldstein goto failed; 378dae1e52cSAmir Goldstein } 379781f143eSTheodore Ts'o return 0; 380dae1e52cSAmir Goldstein failed: 381781f143eSTheodore Ts'o for (; i >= 0; i--) { 382c5c7b8ddSJan Kara /* 383c5c7b8ddSJan Kara * We want to ext4_forget() only freshly allocated indirect 384c5c7b8ddSJan Kara * blocks. Buffer for new_blocks[i-1] is at branch[i].bh and 385c5c7b8ddSJan Kara * buffer at branch[0].bh is indirect block / inode already 386c5c7b8ddSJan Kara * existing before ext4_alloc_branch() was called. 387c5c7b8ddSJan Kara */ 388c5c7b8ddSJan Kara if (i > 0 && i != indirect_blks && branch[i].bh) 389a5211002STheodore Ts'o ext4_forget(handle, 1, ar->inode, branch[i].bh, 390781f143eSTheodore Ts'o branch[i].bh->b_blocknr); 391a5211002STheodore Ts'o ext4_free_blocks(handle, ar->inode, NULL, new_blocks[i], 392a5211002STheodore Ts'o (i == indirect_blks) ? ar->len : 1, 0); 393dae1e52cSAmir Goldstein } 394dae1e52cSAmir Goldstein return err; 395dae1e52cSAmir Goldstein } 396dae1e52cSAmir Goldstein 397dae1e52cSAmir Goldstein /** 398dae1e52cSAmir Goldstein * ext4_splice_branch - splice the allocated branch onto inode. 399dae1e52cSAmir Goldstein * @handle: handle for this transaction 400dae1e52cSAmir Goldstein * @inode: owner 401dae1e52cSAmir Goldstein * @block: (logical) number of block we are adding 402dae1e52cSAmir Goldstein * @chain: chain of indirect blocks (with a missing link - see 403dae1e52cSAmir Goldstein * ext4_alloc_branch) 404dae1e52cSAmir Goldstein * @where: location of missing link 405dae1e52cSAmir Goldstein * @num: number of indirect blocks we are adding 406dae1e52cSAmir Goldstein * @blks: number of direct blocks we are adding 407dae1e52cSAmir Goldstein * 408dae1e52cSAmir Goldstein * This function fills the missing link and does all housekeeping needed in 409dae1e52cSAmir Goldstein * inode (->i_blocks, etc.). In case of success we end up with the full 410dae1e52cSAmir Goldstein * chain to new block and return 0. 411dae1e52cSAmir Goldstein */ 412a5211002STheodore Ts'o static int ext4_splice_branch(handle_t *handle, 413a5211002STheodore Ts'o struct ext4_allocation_request *ar, 414a5211002STheodore Ts'o Indirect *where, int num) 415dae1e52cSAmir Goldstein { 416dae1e52cSAmir Goldstein int i; 417dae1e52cSAmir Goldstein int err = 0; 418dae1e52cSAmir Goldstein ext4_fsblk_t current_block; 419dae1e52cSAmir Goldstein 420dae1e52cSAmir Goldstein /* 421dae1e52cSAmir Goldstein * If we're splicing into a [td]indirect block (as opposed to the 422dae1e52cSAmir Goldstein * inode) then we need to get write access to the [td]indirect block 423dae1e52cSAmir Goldstein * before the splice. 424dae1e52cSAmir Goldstein */ 425dae1e52cSAmir Goldstein if (where->bh) { 426dae1e52cSAmir Goldstein BUFFER_TRACE(where->bh, "get_write_access"); 427dae1e52cSAmir Goldstein err = ext4_journal_get_write_access(handle, where->bh); 428dae1e52cSAmir Goldstein if (err) 429dae1e52cSAmir Goldstein goto err_out; 430dae1e52cSAmir Goldstein } 431dae1e52cSAmir Goldstein /* That's it */ 432dae1e52cSAmir Goldstein 433dae1e52cSAmir Goldstein *where->p = where->key; 434dae1e52cSAmir Goldstein 435dae1e52cSAmir Goldstein /* 436dae1e52cSAmir Goldstein * Update the host buffer_head or inode to point to more just allocated 437dae1e52cSAmir Goldstein * direct blocks blocks 438dae1e52cSAmir Goldstein */ 439a5211002STheodore Ts'o if (num == 0 && ar->len > 1) { 440dae1e52cSAmir Goldstein current_block = le32_to_cpu(where->key) + 1; 441a5211002STheodore Ts'o for (i = 1; i < ar->len; i++) 442dae1e52cSAmir Goldstein *(where->p + i) = cpu_to_le32(current_block++); 443dae1e52cSAmir Goldstein } 444dae1e52cSAmir Goldstein 445dae1e52cSAmir Goldstein /* We are done with atomic stuff, now do the rest of housekeeping */ 446dae1e52cSAmir Goldstein /* had we spliced it onto indirect block? */ 447dae1e52cSAmir Goldstein if (where->bh) { 448dae1e52cSAmir Goldstein /* 449dae1e52cSAmir Goldstein * If we spliced it onto an indirect block, we haven't 450dae1e52cSAmir Goldstein * altered the inode. Note however that if it is being spliced 451dae1e52cSAmir Goldstein * onto an indirect block at the very end of the file (the 452dae1e52cSAmir Goldstein * file is growing) then we *will* alter the inode to reflect 453dae1e52cSAmir Goldstein * the new i_size. But that is not done here - it is done in 454dae1e52cSAmir Goldstein * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode. 455dae1e52cSAmir Goldstein */ 456dae1e52cSAmir Goldstein jbd_debug(5, "splicing indirect only\n"); 457dae1e52cSAmir Goldstein BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata"); 458a5211002STheodore Ts'o err = ext4_handle_dirty_metadata(handle, ar->inode, where->bh); 459dae1e52cSAmir Goldstein if (err) 460dae1e52cSAmir Goldstein goto err_out; 461dae1e52cSAmir Goldstein } else { 462dae1e52cSAmir Goldstein /* 463dae1e52cSAmir Goldstein * OK, we spliced it into the inode itself on a direct block. 464dae1e52cSAmir Goldstein */ 465a5211002STheodore Ts'o ext4_mark_inode_dirty(handle, ar->inode); 466dae1e52cSAmir Goldstein jbd_debug(5, "splicing direct\n"); 467dae1e52cSAmir Goldstein } 468dae1e52cSAmir Goldstein return err; 469dae1e52cSAmir Goldstein 470dae1e52cSAmir Goldstein err_out: 471dae1e52cSAmir Goldstein for (i = 1; i <= num; i++) { 472dae1e52cSAmir Goldstein /* 473dae1e52cSAmir Goldstein * branch[i].bh is newly allocated, so there is no 474dae1e52cSAmir Goldstein * need to revoke the block, which is why we don't 475dae1e52cSAmir Goldstein * need to set EXT4_FREE_BLOCKS_METADATA. 476dae1e52cSAmir Goldstein */ 477a5211002STheodore Ts'o ext4_free_blocks(handle, ar->inode, where[i].bh, 0, 1, 478dae1e52cSAmir Goldstein EXT4_FREE_BLOCKS_FORGET); 479dae1e52cSAmir Goldstein } 480a5211002STheodore Ts'o ext4_free_blocks(handle, ar->inode, NULL, le32_to_cpu(where[num].key), 481a5211002STheodore Ts'o ar->len, 0); 482dae1e52cSAmir Goldstein 483dae1e52cSAmir Goldstein return err; 484dae1e52cSAmir Goldstein } 485dae1e52cSAmir Goldstein 486dae1e52cSAmir Goldstein /* 487dae1e52cSAmir Goldstein * The ext4_ind_map_blocks() function handles non-extents inodes 488dae1e52cSAmir Goldstein * (i.e., using the traditional indirect/double-indirect i_blocks 489dae1e52cSAmir Goldstein * scheme) for ext4_map_blocks(). 490dae1e52cSAmir Goldstein * 491dae1e52cSAmir Goldstein * Allocation strategy is simple: if we have to allocate something, we will 492dae1e52cSAmir Goldstein * have to go the whole way to leaf. So let's do it before attaching anything 493dae1e52cSAmir Goldstein * to tree, set linkage between the newborn blocks, write them if sync is 494dae1e52cSAmir Goldstein * required, recheck the path, free and repeat if check fails, otherwise 495dae1e52cSAmir Goldstein * set the last missing link (that will protect us from any truncate-generated 496dae1e52cSAmir Goldstein * removals - all blocks on the path are immune now) and possibly force the 497dae1e52cSAmir Goldstein * write on the parent block. 498dae1e52cSAmir Goldstein * That has a nice additional property: no special recovery from the failed 499dae1e52cSAmir Goldstein * allocations is needed - we simply release blocks and do not touch anything 500dae1e52cSAmir Goldstein * reachable from inode. 501dae1e52cSAmir Goldstein * 502dae1e52cSAmir Goldstein * `handle' can be NULL if create == 0. 503dae1e52cSAmir Goldstein * 504dae1e52cSAmir Goldstein * return > 0, # of blocks mapped or allocated. 505dae1e52cSAmir Goldstein * return = 0, if plain lookup failed. 506dae1e52cSAmir Goldstein * return < 0, error case. 507dae1e52cSAmir Goldstein * 508dae1e52cSAmir Goldstein * The ext4_ind_get_blocks() function should be called with 509dae1e52cSAmir Goldstein * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem 510dae1e52cSAmir Goldstein * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or 511dae1e52cSAmir Goldstein * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system 512dae1e52cSAmir Goldstein * blocks. 513dae1e52cSAmir Goldstein */ 514dae1e52cSAmir Goldstein int ext4_ind_map_blocks(handle_t *handle, struct inode *inode, 515dae1e52cSAmir Goldstein struct ext4_map_blocks *map, 516dae1e52cSAmir Goldstein int flags) 517dae1e52cSAmir Goldstein { 518a5211002STheodore Ts'o struct ext4_allocation_request ar; 519dae1e52cSAmir Goldstein int err = -EIO; 520dae1e52cSAmir Goldstein ext4_lblk_t offsets[4]; 521dae1e52cSAmir Goldstein Indirect chain[4]; 522dae1e52cSAmir Goldstein Indirect *partial; 523dae1e52cSAmir Goldstein int indirect_blks; 524dae1e52cSAmir Goldstein int blocks_to_boundary = 0; 525dae1e52cSAmir Goldstein int depth; 526dae1e52cSAmir Goldstein int count = 0; 527dae1e52cSAmir Goldstein ext4_fsblk_t first_block = 0; 528dae1e52cSAmir Goldstein 529dae1e52cSAmir Goldstein trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); 530dae1e52cSAmir Goldstein J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))); 531dae1e52cSAmir Goldstein J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0); 532dae1e52cSAmir Goldstein depth = ext4_block_to_path(inode, map->m_lblk, offsets, 533dae1e52cSAmir Goldstein &blocks_to_boundary); 534dae1e52cSAmir Goldstein 535dae1e52cSAmir Goldstein if (depth == 0) 536dae1e52cSAmir Goldstein goto out; 537dae1e52cSAmir Goldstein 538dae1e52cSAmir Goldstein partial = ext4_get_branch(inode, depth, offsets, chain, &err); 539dae1e52cSAmir Goldstein 540dae1e52cSAmir Goldstein /* Simplest case - block found, no allocation needed */ 541dae1e52cSAmir Goldstein if (!partial) { 542dae1e52cSAmir Goldstein first_block = le32_to_cpu(chain[depth - 1].key); 543dae1e52cSAmir Goldstein count++; 544dae1e52cSAmir Goldstein /*map more blocks*/ 545dae1e52cSAmir Goldstein while (count < map->m_len && count <= blocks_to_boundary) { 546dae1e52cSAmir Goldstein ext4_fsblk_t blk; 547dae1e52cSAmir Goldstein 548dae1e52cSAmir Goldstein blk = le32_to_cpu(*(chain[depth-1].p + count)); 549dae1e52cSAmir Goldstein 550dae1e52cSAmir Goldstein if (blk == first_block + count) 551dae1e52cSAmir Goldstein count++; 552dae1e52cSAmir Goldstein else 553dae1e52cSAmir Goldstein break; 554dae1e52cSAmir Goldstein } 555dae1e52cSAmir Goldstein goto got_it; 556dae1e52cSAmir Goldstein } 557dae1e52cSAmir Goldstein 558facab4d9SJan Kara /* Next simple case - plain lookup failed */ 559facab4d9SJan Kara if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { 560facab4d9SJan Kara unsigned epb = inode->i_sb->s_blocksize / sizeof(u32); 561facab4d9SJan Kara int i; 562facab4d9SJan Kara 563facab4d9SJan Kara /* Count number blocks in a subtree under 'partial' */ 564facab4d9SJan Kara count = 1; 565facab4d9SJan Kara for (i = 0; partial + i != chain + depth - 1; i++) 566facab4d9SJan Kara count *= epb; 567facab4d9SJan Kara /* Fill in size of a hole we found */ 568facab4d9SJan Kara map->m_pblk = 0; 569facab4d9SJan Kara map->m_len = min_t(unsigned int, map->m_len, count); 570facab4d9SJan Kara goto cleanup; 571facab4d9SJan Kara } 572facab4d9SJan Kara 573facab4d9SJan Kara /* Failed read of indirect block */ 574facab4d9SJan Kara if (err == -EIO) 575dae1e52cSAmir Goldstein goto cleanup; 576dae1e52cSAmir Goldstein 577dae1e52cSAmir Goldstein /* 578dae1e52cSAmir Goldstein * Okay, we need to do block allocation. 579dae1e52cSAmir Goldstein */ 580e2b911c5SDarrick J. Wong if (ext4_has_feature_bigalloc(inode->i_sb)) { 581bab08ab9STheodore Ts'o EXT4_ERROR_INODE(inode, "Can't allocate blocks for " 582bab08ab9STheodore Ts'o "non-extent mapped inodes with bigalloc"); 5836a797d27SDarrick J. Wong return -EFSCORRUPTED; 584bab08ab9STheodore Ts'o } 585bab08ab9STheodore Ts'o 586a5211002STheodore Ts'o /* Set up for the direct block allocation */ 587a5211002STheodore Ts'o memset(&ar, 0, sizeof(ar)); 588a5211002STheodore Ts'o ar.inode = inode; 589a5211002STheodore Ts'o ar.logical = map->m_lblk; 590a5211002STheodore Ts'o if (S_ISREG(inode->i_mode)) 591a5211002STheodore Ts'o ar.flags = EXT4_MB_HINT_DATA; 592e3cf5d5dSTheodore Ts'o if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) 593e3cf5d5dSTheodore Ts'o ar.flags |= EXT4_MB_DELALLOC_RESERVED; 594c5e298aeSTheodore Ts'o if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL) 595c5e298aeSTheodore Ts'o ar.flags |= EXT4_MB_USE_RESERVED; 596a5211002STheodore Ts'o 597a5211002STheodore Ts'o ar.goal = ext4_find_goal(inode, map->m_lblk, partial); 598dae1e52cSAmir Goldstein 599dae1e52cSAmir Goldstein /* the number of blocks need to allocate for [d,t]indirect blocks */ 600dae1e52cSAmir Goldstein indirect_blks = (chain + depth) - partial - 1; 601dae1e52cSAmir Goldstein 602dae1e52cSAmir Goldstein /* 603dae1e52cSAmir Goldstein * Next look up the indirect map to count the totoal number of 604dae1e52cSAmir Goldstein * direct blocks to allocate for this branch. 605dae1e52cSAmir Goldstein */ 606a5211002STheodore Ts'o ar.len = ext4_blks_to_allocate(partial, indirect_blks, 607dae1e52cSAmir Goldstein map->m_len, blocks_to_boundary); 608a5211002STheodore Ts'o 609dae1e52cSAmir Goldstein /* 610dae1e52cSAmir Goldstein * Block out ext4_truncate while we alter the tree 611dae1e52cSAmir Goldstein */ 612a5211002STheodore Ts'o err = ext4_alloc_branch(handle, &ar, indirect_blks, 613dae1e52cSAmir Goldstein offsets + (partial - chain), partial); 614dae1e52cSAmir Goldstein 615dae1e52cSAmir Goldstein /* 616dae1e52cSAmir Goldstein * The ext4_splice_branch call will free and forget any buffers 617dae1e52cSAmir Goldstein * on the new chain if there is a failure, but that risks using 618dae1e52cSAmir Goldstein * up transaction credits, especially for bitmaps where the 619dae1e52cSAmir Goldstein * credits cannot be returned. Can we handle this somehow? We 620dae1e52cSAmir Goldstein * may need to return -EAGAIN upwards in the worst case. --sct 621dae1e52cSAmir Goldstein */ 622dae1e52cSAmir Goldstein if (!err) 623a5211002STheodore Ts'o err = ext4_splice_branch(handle, &ar, partial, indirect_blks); 624dae1e52cSAmir Goldstein if (err) 625dae1e52cSAmir Goldstein goto cleanup; 626dae1e52cSAmir Goldstein 627dae1e52cSAmir Goldstein map->m_flags |= EXT4_MAP_NEW; 628dae1e52cSAmir Goldstein 629dae1e52cSAmir Goldstein ext4_update_inode_fsync_trans(handle, inode, 1); 630a5211002STheodore Ts'o count = ar.len; 631dae1e52cSAmir Goldstein got_it: 632dae1e52cSAmir Goldstein map->m_flags |= EXT4_MAP_MAPPED; 633dae1e52cSAmir Goldstein map->m_pblk = le32_to_cpu(chain[depth-1].key); 634dae1e52cSAmir Goldstein map->m_len = count; 635dae1e52cSAmir Goldstein if (count > blocks_to_boundary) 636dae1e52cSAmir Goldstein map->m_flags |= EXT4_MAP_BOUNDARY; 637dae1e52cSAmir Goldstein err = count; 638dae1e52cSAmir Goldstein /* Clean up and exit */ 639dae1e52cSAmir Goldstein partial = chain + depth - 1; /* the whole chain */ 640dae1e52cSAmir Goldstein cleanup: 641dae1e52cSAmir Goldstein while (partial > chain) { 642dae1e52cSAmir Goldstein BUFFER_TRACE(partial->bh, "call brelse"); 643dae1e52cSAmir Goldstein brelse(partial->bh); 644dae1e52cSAmir Goldstein partial--; 645dae1e52cSAmir Goldstein } 646dae1e52cSAmir Goldstein out: 64721ddd568STheodore Ts'o trace_ext4_ind_map_blocks_exit(inode, flags, map, err); 648dae1e52cSAmir Goldstein return err; 649dae1e52cSAmir Goldstein } 650dae1e52cSAmir Goldstein 651dae1e52cSAmir Goldstein /* 652dae1e52cSAmir Goldstein * Calculate the number of metadata blocks need to reserve 653dae1e52cSAmir Goldstein * to allocate a new block at @lblocks for non extent file based file 654dae1e52cSAmir Goldstein */ 655dae1e52cSAmir Goldstein int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock) 656dae1e52cSAmir Goldstein { 657dae1e52cSAmir Goldstein struct ext4_inode_info *ei = EXT4_I(inode); 658dae1e52cSAmir Goldstein sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1); 659dae1e52cSAmir Goldstein int blk_bits; 660dae1e52cSAmir Goldstein 661dae1e52cSAmir Goldstein if (lblock < EXT4_NDIR_BLOCKS) 662dae1e52cSAmir Goldstein return 0; 663dae1e52cSAmir Goldstein 664dae1e52cSAmir Goldstein lblock -= EXT4_NDIR_BLOCKS; 665dae1e52cSAmir Goldstein 666dae1e52cSAmir Goldstein if (ei->i_da_metadata_calc_len && 667dae1e52cSAmir Goldstein (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) { 668dae1e52cSAmir Goldstein ei->i_da_metadata_calc_len++; 669dae1e52cSAmir Goldstein return 0; 670dae1e52cSAmir Goldstein } 671dae1e52cSAmir Goldstein ei->i_da_metadata_calc_last_lblock = lblock & dind_mask; 672dae1e52cSAmir Goldstein ei->i_da_metadata_calc_len = 1; 673dae1e52cSAmir Goldstein blk_bits = order_base_2(lblock); 674dae1e52cSAmir Goldstein return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1; 675dae1e52cSAmir Goldstein } 676dae1e52cSAmir Goldstein 677fa55a0edSJan Kara /* 678fa55a0edSJan Kara * Calculate number of indirect blocks touched by mapping @nrblocks logically 679fa55a0edSJan Kara * contiguous blocks 680fa55a0edSJan Kara */ 681fa55a0edSJan Kara int ext4_ind_trans_blocks(struct inode *inode, int nrblocks) 682dae1e52cSAmir Goldstein { 683dae1e52cSAmir Goldstein /* 684dae1e52cSAmir Goldstein * With N contiguous data blocks, we need at most 685dae1e52cSAmir Goldstein * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks, 686dae1e52cSAmir Goldstein * 2 dindirect blocks, and 1 tindirect block 687dae1e52cSAmir Goldstein */ 688fa55a0edSJan Kara return DIV_ROUND_UP(nrblocks, EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4; 689dae1e52cSAmir Goldstein } 690dae1e52cSAmir Goldstein 691dae1e52cSAmir Goldstein /* 692dae1e52cSAmir Goldstein * Truncate transactions can be complex and absolutely huge. So we need to 693dae1e52cSAmir Goldstein * be able to restart the transaction at a conventient checkpoint to make 694dae1e52cSAmir Goldstein * sure we don't overflow the journal. 695dae1e52cSAmir Goldstein * 696819c4920STheodore Ts'o * Try to extend this transaction for the purposes of truncation. If 697dae1e52cSAmir Goldstein * extend fails, we need to propagate the failure up and restart the 698dae1e52cSAmir Goldstein * transaction in the top-level truncate loop. --sct 699dae1e52cSAmir Goldstein * 700dae1e52cSAmir Goldstein * Returns 0 if we managed to create more room. If we can't create more 701dae1e52cSAmir Goldstein * room, and the transaction must be restarted we return 1. 702dae1e52cSAmir Goldstein */ 703dae1e52cSAmir Goldstein static int try_to_extend_transaction(handle_t *handle, struct inode *inode) 704dae1e52cSAmir Goldstein { 705dae1e52cSAmir Goldstein if (!ext4_handle_valid(handle)) 706dae1e52cSAmir Goldstein return 0; 707dae1e52cSAmir Goldstein if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1)) 708dae1e52cSAmir Goldstein return 0; 709dae1e52cSAmir Goldstein if (!ext4_journal_extend(handle, ext4_blocks_for_truncate(inode))) 710dae1e52cSAmir Goldstein return 0; 711dae1e52cSAmir Goldstein return 1; 712dae1e52cSAmir Goldstein } 713dae1e52cSAmir Goldstein 714dae1e52cSAmir Goldstein /* 715dae1e52cSAmir Goldstein * Probably it should be a library function... search for first non-zero word 716dae1e52cSAmir Goldstein * or memcmp with zero_page, whatever is better for particular architecture. 717dae1e52cSAmir Goldstein * Linus? 718dae1e52cSAmir Goldstein */ 719dae1e52cSAmir Goldstein static inline int all_zeroes(__le32 *p, __le32 *q) 720dae1e52cSAmir Goldstein { 721dae1e52cSAmir Goldstein while (p < q) 722dae1e52cSAmir Goldstein if (*p++) 723dae1e52cSAmir Goldstein return 0; 724dae1e52cSAmir Goldstein return 1; 725dae1e52cSAmir Goldstein } 726dae1e52cSAmir Goldstein 727dae1e52cSAmir Goldstein /** 728dae1e52cSAmir Goldstein * ext4_find_shared - find the indirect blocks for partial truncation. 729dae1e52cSAmir Goldstein * @inode: inode in question 730dae1e52cSAmir Goldstein * @depth: depth of the affected branch 731dae1e52cSAmir Goldstein * @offsets: offsets of pointers in that branch (see ext4_block_to_path) 732dae1e52cSAmir Goldstein * @chain: place to store the pointers to partial indirect blocks 733dae1e52cSAmir Goldstein * @top: place to the (detached) top of branch 734dae1e52cSAmir Goldstein * 735dae1e52cSAmir Goldstein * This is a helper function used by ext4_truncate(). 736dae1e52cSAmir Goldstein * 737dae1e52cSAmir Goldstein * When we do truncate() we may have to clean the ends of several 738dae1e52cSAmir Goldstein * indirect blocks but leave the blocks themselves alive. Block is 739dae1e52cSAmir Goldstein * partially truncated if some data below the new i_size is referred 740dae1e52cSAmir Goldstein * from it (and it is on the path to the first completely truncated 741dae1e52cSAmir Goldstein * data block, indeed). We have to free the top of that path along 742dae1e52cSAmir Goldstein * with everything to the right of the path. Since no allocation 743dae1e52cSAmir Goldstein * past the truncation point is possible until ext4_truncate() 744dae1e52cSAmir Goldstein * finishes, we may safely do the latter, but top of branch may 745dae1e52cSAmir Goldstein * require special attention - pageout below the truncation point 746dae1e52cSAmir Goldstein * might try to populate it. 747dae1e52cSAmir Goldstein * 748dae1e52cSAmir Goldstein * We atomically detach the top of branch from the tree, store the 749dae1e52cSAmir Goldstein * block number of its root in *@top, pointers to buffer_heads of 750dae1e52cSAmir Goldstein * partially truncated blocks - in @chain[].bh and pointers to 751dae1e52cSAmir Goldstein * their last elements that should not be removed - in 752dae1e52cSAmir Goldstein * @chain[].p. Return value is the pointer to last filled element 753dae1e52cSAmir Goldstein * of @chain. 754dae1e52cSAmir Goldstein * 755dae1e52cSAmir Goldstein * The work left to caller to do the actual freeing of subtrees: 756dae1e52cSAmir Goldstein * a) free the subtree starting from *@top 757dae1e52cSAmir Goldstein * b) free the subtrees whose roots are stored in 758dae1e52cSAmir Goldstein * (@chain[i].p+1 .. end of @chain[i].bh->b_data) 759dae1e52cSAmir Goldstein * c) free the subtrees growing from the inode past the @chain[0]. 760dae1e52cSAmir Goldstein * (no partially truncated stuff there). */ 761dae1e52cSAmir Goldstein 762dae1e52cSAmir Goldstein static Indirect *ext4_find_shared(struct inode *inode, int depth, 763dae1e52cSAmir Goldstein ext4_lblk_t offsets[4], Indirect chain[4], 764dae1e52cSAmir Goldstein __le32 *top) 765dae1e52cSAmir Goldstein { 766dae1e52cSAmir Goldstein Indirect *partial, *p; 767dae1e52cSAmir Goldstein int k, err; 768dae1e52cSAmir Goldstein 769dae1e52cSAmir Goldstein *top = 0; 770dae1e52cSAmir Goldstein /* Make k index the deepest non-null offset + 1 */ 771dae1e52cSAmir Goldstein for (k = depth; k > 1 && !offsets[k-1]; k--) 772dae1e52cSAmir Goldstein ; 773dae1e52cSAmir Goldstein partial = ext4_get_branch(inode, k, offsets, chain, &err); 774dae1e52cSAmir Goldstein /* Writer: pointers */ 775dae1e52cSAmir Goldstein if (!partial) 776dae1e52cSAmir Goldstein partial = chain + k-1; 777dae1e52cSAmir Goldstein /* 778dae1e52cSAmir Goldstein * If the branch acquired continuation since we've looked at it - 779dae1e52cSAmir Goldstein * fine, it should all survive and (new) top doesn't belong to us. 780dae1e52cSAmir Goldstein */ 781dae1e52cSAmir Goldstein if (!partial->key && *partial->p) 782dae1e52cSAmir Goldstein /* Writer: end */ 783dae1e52cSAmir Goldstein goto no_top; 784dae1e52cSAmir Goldstein for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--) 785dae1e52cSAmir Goldstein ; 786dae1e52cSAmir Goldstein /* 787dae1e52cSAmir Goldstein * OK, we've found the last block that must survive. The rest of our 788dae1e52cSAmir Goldstein * branch should be detached before unlocking. However, if that rest 789dae1e52cSAmir Goldstein * of branch is all ours and does not grow immediately from the inode 790dae1e52cSAmir Goldstein * it's easier to cheat and just decrement partial->p. 791dae1e52cSAmir Goldstein */ 792dae1e52cSAmir Goldstein if (p == chain + k - 1 && p > chain) { 793dae1e52cSAmir Goldstein p->p--; 794dae1e52cSAmir Goldstein } else { 795dae1e52cSAmir Goldstein *top = *p->p; 796dae1e52cSAmir Goldstein /* Nope, don't do this in ext4. Must leave the tree intact */ 797dae1e52cSAmir Goldstein #if 0 798dae1e52cSAmir Goldstein *p->p = 0; 799dae1e52cSAmir Goldstein #endif 800dae1e52cSAmir Goldstein } 801dae1e52cSAmir Goldstein /* Writer: end */ 802dae1e52cSAmir Goldstein 803dae1e52cSAmir Goldstein while (partial > p) { 804dae1e52cSAmir Goldstein brelse(partial->bh); 805dae1e52cSAmir Goldstein partial--; 806dae1e52cSAmir Goldstein } 807dae1e52cSAmir Goldstein no_top: 808dae1e52cSAmir Goldstein return partial; 809dae1e52cSAmir Goldstein } 810dae1e52cSAmir Goldstein 811dae1e52cSAmir Goldstein /* 812dae1e52cSAmir Goldstein * Zero a number of block pointers in either an inode or an indirect block. 813dae1e52cSAmir Goldstein * If we restart the transaction we must again get write access to the 814dae1e52cSAmir Goldstein * indirect block for further modification. 815dae1e52cSAmir Goldstein * 816dae1e52cSAmir Goldstein * We release `count' blocks on disk, but (last - first) may be greater 817dae1e52cSAmir Goldstein * than `count' because there can be holes in there. 818dae1e52cSAmir Goldstein * 819dae1e52cSAmir Goldstein * Return 0 on success, 1 on invalid block range 820dae1e52cSAmir Goldstein * and < 0 on fatal error. 821dae1e52cSAmir Goldstein */ 822dae1e52cSAmir Goldstein static int ext4_clear_blocks(handle_t *handle, struct inode *inode, 823dae1e52cSAmir Goldstein struct buffer_head *bh, 824dae1e52cSAmir Goldstein ext4_fsblk_t block_to_free, 825dae1e52cSAmir Goldstein unsigned long count, __le32 *first, 826dae1e52cSAmir Goldstein __le32 *last) 827dae1e52cSAmir Goldstein { 828dae1e52cSAmir Goldstein __le32 *p; 829981250caSTheodore Ts'o int flags = EXT4_FREE_BLOCKS_VALIDATED; 830dae1e52cSAmir Goldstein int err; 831dae1e52cSAmir Goldstein 832*ddfa17e4STahsin Erdogan if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) || 833*ddfa17e4STahsin Erdogan ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE)) 834981250caSTheodore Ts'o flags |= EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_METADATA; 835981250caSTheodore Ts'o else if (ext4_should_journal_data(inode)) 836981250caSTheodore Ts'o flags |= EXT4_FREE_BLOCKS_FORGET; 837dae1e52cSAmir Goldstein 838dae1e52cSAmir Goldstein if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free, 839dae1e52cSAmir Goldstein count)) { 840dae1e52cSAmir Goldstein EXT4_ERROR_INODE(inode, "attempt to clear invalid " 841dae1e52cSAmir Goldstein "blocks %llu len %lu", 842dae1e52cSAmir Goldstein (unsigned long long) block_to_free, count); 843dae1e52cSAmir Goldstein return 1; 844dae1e52cSAmir Goldstein } 845dae1e52cSAmir Goldstein 846dae1e52cSAmir Goldstein if (try_to_extend_transaction(handle, inode)) { 847dae1e52cSAmir Goldstein if (bh) { 848dae1e52cSAmir Goldstein BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); 849dae1e52cSAmir Goldstein err = ext4_handle_dirty_metadata(handle, inode, bh); 850dae1e52cSAmir Goldstein if (unlikely(err)) 851dae1e52cSAmir Goldstein goto out_err; 852dae1e52cSAmir Goldstein } 853dae1e52cSAmir Goldstein err = ext4_mark_inode_dirty(handle, inode); 854dae1e52cSAmir Goldstein if (unlikely(err)) 855dae1e52cSAmir Goldstein goto out_err; 856dae1e52cSAmir Goldstein err = ext4_truncate_restart_trans(handle, inode, 857dae1e52cSAmir Goldstein ext4_blocks_for_truncate(inode)); 858dae1e52cSAmir Goldstein if (unlikely(err)) 859dae1e52cSAmir Goldstein goto out_err; 860dae1e52cSAmir Goldstein if (bh) { 861dae1e52cSAmir Goldstein BUFFER_TRACE(bh, "retaking write access"); 862dae1e52cSAmir Goldstein err = ext4_journal_get_write_access(handle, bh); 863dae1e52cSAmir Goldstein if (unlikely(err)) 864dae1e52cSAmir Goldstein goto out_err; 865dae1e52cSAmir Goldstein } 866dae1e52cSAmir Goldstein } 867dae1e52cSAmir Goldstein 868dae1e52cSAmir Goldstein for (p = first; p < last; p++) 869dae1e52cSAmir Goldstein *p = 0; 870dae1e52cSAmir Goldstein 871dae1e52cSAmir Goldstein ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags); 872dae1e52cSAmir Goldstein return 0; 873dae1e52cSAmir Goldstein out_err: 874dae1e52cSAmir Goldstein ext4_std_error(inode->i_sb, err); 875dae1e52cSAmir Goldstein return err; 876dae1e52cSAmir Goldstein } 877dae1e52cSAmir Goldstein 878dae1e52cSAmir Goldstein /** 879dae1e52cSAmir Goldstein * ext4_free_data - free a list of data blocks 880dae1e52cSAmir Goldstein * @handle: handle for this transaction 881dae1e52cSAmir Goldstein * @inode: inode we are dealing with 882dae1e52cSAmir Goldstein * @this_bh: indirect buffer_head which contains *@first and *@last 883dae1e52cSAmir Goldstein * @first: array of block numbers 884dae1e52cSAmir Goldstein * @last: points immediately past the end of array 885dae1e52cSAmir Goldstein * 886dae1e52cSAmir Goldstein * We are freeing all blocks referred from that array (numbers are stored as 887dae1e52cSAmir Goldstein * little-endian 32-bit) and updating @inode->i_blocks appropriately. 888dae1e52cSAmir Goldstein * 889dae1e52cSAmir Goldstein * We accumulate contiguous runs of blocks to free. Conveniently, if these 890dae1e52cSAmir Goldstein * blocks are contiguous then releasing them at one time will only affect one 891dae1e52cSAmir Goldstein * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't 892dae1e52cSAmir Goldstein * actually use a lot of journal space. 893dae1e52cSAmir Goldstein * 894dae1e52cSAmir Goldstein * @this_bh will be %NULL if @first and @last point into the inode's direct 895dae1e52cSAmir Goldstein * block pointers. 896dae1e52cSAmir Goldstein */ 897dae1e52cSAmir Goldstein static void ext4_free_data(handle_t *handle, struct inode *inode, 898dae1e52cSAmir Goldstein struct buffer_head *this_bh, 899dae1e52cSAmir Goldstein __le32 *first, __le32 *last) 900dae1e52cSAmir Goldstein { 901dae1e52cSAmir Goldstein ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */ 902dae1e52cSAmir Goldstein unsigned long count = 0; /* Number of blocks in the run */ 903dae1e52cSAmir Goldstein __le32 *block_to_free_p = NULL; /* Pointer into inode/ind 904dae1e52cSAmir Goldstein corresponding to 905dae1e52cSAmir Goldstein block_to_free */ 906dae1e52cSAmir Goldstein ext4_fsblk_t nr; /* Current block # */ 907dae1e52cSAmir Goldstein __le32 *p; /* Pointer into inode/ind 908dae1e52cSAmir Goldstein for current block */ 909dae1e52cSAmir Goldstein int err = 0; 910dae1e52cSAmir Goldstein 911dae1e52cSAmir Goldstein if (this_bh) { /* For indirect block */ 912dae1e52cSAmir Goldstein BUFFER_TRACE(this_bh, "get_write_access"); 913dae1e52cSAmir Goldstein err = ext4_journal_get_write_access(handle, this_bh); 914dae1e52cSAmir Goldstein /* Important: if we can't update the indirect pointers 915dae1e52cSAmir Goldstein * to the blocks, we can't free them. */ 916dae1e52cSAmir Goldstein if (err) 917dae1e52cSAmir Goldstein return; 918dae1e52cSAmir Goldstein } 919dae1e52cSAmir Goldstein 920dae1e52cSAmir Goldstein for (p = first; p < last; p++) { 921dae1e52cSAmir Goldstein nr = le32_to_cpu(*p); 922dae1e52cSAmir Goldstein if (nr) { 923dae1e52cSAmir Goldstein /* accumulate blocks to free if they're contiguous */ 924dae1e52cSAmir Goldstein if (count == 0) { 925dae1e52cSAmir Goldstein block_to_free = nr; 926dae1e52cSAmir Goldstein block_to_free_p = p; 927dae1e52cSAmir Goldstein count = 1; 928dae1e52cSAmir Goldstein } else if (nr == block_to_free + count) { 929dae1e52cSAmir Goldstein count++; 930dae1e52cSAmir Goldstein } else { 931dae1e52cSAmir Goldstein err = ext4_clear_blocks(handle, inode, this_bh, 932dae1e52cSAmir Goldstein block_to_free, count, 933dae1e52cSAmir Goldstein block_to_free_p, p); 934dae1e52cSAmir Goldstein if (err) 935dae1e52cSAmir Goldstein break; 936dae1e52cSAmir Goldstein block_to_free = nr; 937dae1e52cSAmir Goldstein block_to_free_p = p; 938dae1e52cSAmir Goldstein count = 1; 939dae1e52cSAmir Goldstein } 940dae1e52cSAmir Goldstein } 941dae1e52cSAmir Goldstein } 942dae1e52cSAmir Goldstein 943dae1e52cSAmir Goldstein if (!err && count > 0) 944dae1e52cSAmir Goldstein err = ext4_clear_blocks(handle, inode, this_bh, block_to_free, 945dae1e52cSAmir Goldstein count, block_to_free_p, p); 946dae1e52cSAmir Goldstein if (err < 0) 947dae1e52cSAmir Goldstein /* fatal error */ 948dae1e52cSAmir Goldstein return; 949dae1e52cSAmir Goldstein 950dae1e52cSAmir Goldstein if (this_bh) { 951dae1e52cSAmir Goldstein BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata"); 952dae1e52cSAmir Goldstein 953dae1e52cSAmir Goldstein /* 954dae1e52cSAmir Goldstein * The buffer head should have an attached journal head at this 955dae1e52cSAmir Goldstein * point. However, if the data is corrupted and an indirect 956dae1e52cSAmir Goldstein * block pointed to itself, it would have been detached when 957dae1e52cSAmir Goldstein * the block was cleared. Check for this instead of OOPSing. 958dae1e52cSAmir Goldstein */ 959dae1e52cSAmir Goldstein if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh)) 960dae1e52cSAmir Goldstein ext4_handle_dirty_metadata(handle, inode, this_bh); 961dae1e52cSAmir Goldstein else 962dae1e52cSAmir Goldstein EXT4_ERROR_INODE(inode, 963dae1e52cSAmir Goldstein "circular indirect block detected at " 964dae1e52cSAmir Goldstein "block %llu", 965dae1e52cSAmir Goldstein (unsigned long long) this_bh->b_blocknr); 966dae1e52cSAmir Goldstein } 967dae1e52cSAmir Goldstein } 968dae1e52cSAmir Goldstein 969dae1e52cSAmir Goldstein /** 970dae1e52cSAmir Goldstein * ext4_free_branches - free an array of branches 971dae1e52cSAmir Goldstein * @handle: JBD handle for this transaction 972dae1e52cSAmir Goldstein * @inode: inode we are dealing with 973dae1e52cSAmir Goldstein * @parent_bh: the buffer_head which contains *@first and *@last 974dae1e52cSAmir Goldstein * @first: array of block numbers 975dae1e52cSAmir Goldstein * @last: pointer immediately past the end of array 976dae1e52cSAmir Goldstein * @depth: depth of the branches to free 977dae1e52cSAmir Goldstein * 978dae1e52cSAmir Goldstein * We are freeing all blocks referred from these branches (numbers are 979dae1e52cSAmir Goldstein * stored as little-endian 32-bit) and updating @inode->i_blocks 980dae1e52cSAmir Goldstein * appropriately. 981dae1e52cSAmir Goldstein */ 982dae1e52cSAmir Goldstein static void ext4_free_branches(handle_t *handle, struct inode *inode, 983dae1e52cSAmir Goldstein struct buffer_head *parent_bh, 984dae1e52cSAmir Goldstein __le32 *first, __le32 *last, int depth) 985dae1e52cSAmir Goldstein { 986dae1e52cSAmir Goldstein ext4_fsblk_t nr; 987dae1e52cSAmir Goldstein __le32 *p; 988dae1e52cSAmir Goldstein 989dae1e52cSAmir Goldstein if (ext4_handle_is_aborted(handle)) 990dae1e52cSAmir Goldstein return; 991dae1e52cSAmir Goldstein 992dae1e52cSAmir Goldstein if (depth--) { 993dae1e52cSAmir Goldstein struct buffer_head *bh; 994dae1e52cSAmir Goldstein int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); 995dae1e52cSAmir Goldstein p = last; 996dae1e52cSAmir Goldstein while (--p >= first) { 997dae1e52cSAmir Goldstein nr = le32_to_cpu(*p); 998dae1e52cSAmir Goldstein if (!nr) 999dae1e52cSAmir Goldstein continue; /* A hole */ 1000dae1e52cSAmir Goldstein 1001dae1e52cSAmir Goldstein if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), 1002dae1e52cSAmir Goldstein nr, 1)) { 1003dae1e52cSAmir Goldstein EXT4_ERROR_INODE(inode, 1004dae1e52cSAmir Goldstein "invalid indirect mapped " 1005dae1e52cSAmir Goldstein "block %lu (level %d)", 1006dae1e52cSAmir Goldstein (unsigned long) nr, depth); 1007dae1e52cSAmir Goldstein break; 1008dae1e52cSAmir Goldstein } 1009dae1e52cSAmir Goldstein 1010dae1e52cSAmir Goldstein /* Go read the buffer for the next level down */ 1011dae1e52cSAmir Goldstein bh = sb_bread(inode->i_sb, nr); 1012dae1e52cSAmir Goldstein 1013dae1e52cSAmir Goldstein /* 1014dae1e52cSAmir Goldstein * A read failure? Report error and clear slot 1015dae1e52cSAmir Goldstein * (should be rare). 1016dae1e52cSAmir Goldstein */ 1017dae1e52cSAmir Goldstein if (!bh) { 1018dae1e52cSAmir Goldstein EXT4_ERROR_INODE_BLOCK(inode, nr, 1019dae1e52cSAmir Goldstein "Read failure"); 1020dae1e52cSAmir Goldstein continue; 1021dae1e52cSAmir Goldstein } 1022dae1e52cSAmir Goldstein 1023dae1e52cSAmir Goldstein /* This zaps the entire block. Bottom up. */ 1024dae1e52cSAmir Goldstein BUFFER_TRACE(bh, "free child branches"); 1025dae1e52cSAmir Goldstein ext4_free_branches(handle, inode, bh, 1026dae1e52cSAmir Goldstein (__le32 *) bh->b_data, 1027dae1e52cSAmir Goldstein (__le32 *) bh->b_data + addr_per_block, 1028dae1e52cSAmir Goldstein depth); 1029dae1e52cSAmir Goldstein brelse(bh); 1030dae1e52cSAmir Goldstein 1031dae1e52cSAmir Goldstein /* 1032dae1e52cSAmir Goldstein * Everything below this this pointer has been 1033dae1e52cSAmir Goldstein * released. Now let this top-of-subtree go. 1034dae1e52cSAmir Goldstein * 1035dae1e52cSAmir Goldstein * We want the freeing of this indirect block to be 1036dae1e52cSAmir Goldstein * atomic in the journal with the updating of the 1037dae1e52cSAmir Goldstein * bitmap block which owns it. So make some room in 1038dae1e52cSAmir Goldstein * the journal. 1039dae1e52cSAmir Goldstein * 1040dae1e52cSAmir Goldstein * We zero the parent pointer *after* freeing its 1041dae1e52cSAmir Goldstein * pointee in the bitmaps, so if extend_transaction() 1042dae1e52cSAmir Goldstein * for some reason fails to put the bitmap changes and 1043dae1e52cSAmir Goldstein * the release into the same transaction, recovery 1044dae1e52cSAmir Goldstein * will merely complain about releasing a free block, 1045dae1e52cSAmir Goldstein * rather than leaking blocks. 1046dae1e52cSAmir Goldstein */ 1047dae1e52cSAmir Goldstein if (ext4_handle_is_aborted(handle)) 1048dae1e52cSAmir Goldstein return; 1049dae1e52cSAmir Goldstein if (try_to_extend_transaction(handle, inode)) { 1050dae1e52cSAmir Goldstein ext4_mark_inode_dirty(handle, inode); 1051dae1e52cSAmir Goldstein ext4_truncate_restart_trans(handle, inode, 1052dae1e52cSAmir Goldstein ext4_blocks_for_truncate(inode)); 1053dae1e52cSAmir Goldstein } 1054dae1e52cSAmir Goldstein 1055dae1e52cSAmir Goldstein /* 1056dae1e52cSAmir Goldstein * The forget flag here is critical because if 1057dae1e52cSAmir Goldstein * we are journaling (and not doing data 1058dae1e52cSAmir Goldstein * journaling), we have to make sure a revoke 1059dae1e52cSAmir Goldstein * record is written to prevent the journal 1060dae1e52cSAmir Goldstein * replay from overwriting the (former) 1061dae1e52cSAmir Goldstein * indirect block if it gets reallocated as a 1062dae1e52cSAmir Goldstein * data block. This must happen in the same 1063dae1e52cSAmir Goldstein * transaction where the data blocks are 1064dae1e52cSAmir Goldstein * actually freed. 1065dae1e52cSAmir Goldstein */ 1066dae1e52cSAmir Goldstein ext4_free_blocks(handle, inode, NULL, nr, 1, 1067dae1e52cSAmir Goldstein EXT4_FREE_BLOCKS_METADATA| 1068dae1e52cSAmir Goldstein EXT4_FREE_BLOCKS_FORGET); 1069dae1e52cSAmir Goldstein 1070dae1e52cSAmir Goldstein if (parent_bh) { 1071dae1e52cSAmir Goldstein /* 1072dae1e52cSAmir Goldstein * The block which we have just freed is 1073dae1e52cSAmir Goldstein * pointed to by an indirect block: journal it 1074dae1e52cSAmir Goldstein */ 1075dae1e52cSAmir Goldstein BUFFER_TRACE(parent_bh, "get_write_access"); 1076dae1e52cSAmir Goldstein if (!ext4_journal_get_write_access(handle, 1077dae1e52cSAmir Goldstein parent_bh)){ 1078dae1e52cSAmir Goldstein *p = 0; 1079dae1e52cSAmir Goldstein BUFFER_TRACE(parent_bh, 1080dae1e52cSAmir Goldstein "call ext4_handle_dirty_metadata"); 1081dae1e52cSAmir Goldstein ext4_handle_dirty_metadata(handle, 1082dae1e52cSAmir Goldstein inode, 1083dae1e52cSAmir Goldstein parent_bh); 1084dae1e52cSAmir Goldstein } 1085dae1e52cSAmir Goldstein } 1086dae1e52cSAmir Goldstein } 1087dae1e52cSAmir Goldstein } else { 1088dae1e52cSAmir Goldstein /* We have reached the bottom of the tree. */ 1089dae1e52cSAmir Goldstein BUFFER_TRACE(parent_bh, "free data blocks"); 1090dae1e52cSAmir Goldstein ext4_free_data(handle, inode, parent_bh, first, last); 1091dae1e52cSAmir Goldstein } 1092dae1e52cSAmir Goldstein } 1093dae1e52cSAmir Goldstein 1094819c4920STheodore Ts'o void ext4_ind_truncate(handle_t *handle, struct inode *inode) 1095dae1e52cSAmir Goldstein { 1096dae1e52cSAmir Goldstein struct ext4_inode_info *ei = EXT4_I(inode); 1097dae1e52cSAmir Goldstein __le32 *i_data = ei->i_data; 1098dae1e52cSAmir Goldstein int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); 1099dae1e52cSAmir Goldstein ext4_lblk_t offsets[4]; 1100dae1e52cSAmir Goldstein Indirect chain[4]; 1101dae1e52cSAmir Goldstein Indirect *partial; 1102dae1e52cSAmir Goldstein __le32 nr = 0; 1103dae1e52cSAmir Goldstein int n = 0; 1104dae1e52cSAmir Goldstein ext4_lblk_t last_block, max_block; 1105dae1e52cSAmir Goldstein unsigned blocksize = inode->i_sb->s_blocksize; 1106dae1e52cSAmir Goldstein 1107dae1e52cSAmir Goldstein last_block = (inode->i_size + blocksize-1) 1108dae1e52cSAmir Goldstein >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); 1109dae1e52cSAmir Goldstein max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1) 1110dae1e52cSAmir Goldstein >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); 1111dae1e52cSAmir Goldstein 1112dae1e52cSAmir Goldstein if (last_block != max_block) { 1113dae1e52cSAmir Goldstein n = ext4_block_to_path(inode, last_block, offsets, NULL); 1114dae1e52cSAmir Goldstein if (n == 0) 1115819c4920STheodore Ts'o return; 1116dae1e52cSAmir Goldstein } 1117dae1e52cSAmir Goldstein 111851865fdaSZheng Liu ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block); 1119dae1e52cSAmir Goldstein 1120dae1e52cSAmir Goldstein /* 1121dae1e52cSAmir Goldstein * The orphan list entry will now protect us from any crash which 1122dae1e52cSAmir Goldstein * occurs before the truncate completes, so it is now safe to propagate 1123dae1e52cSAmir Goldstein * the new, shorter inode size (held for now in i_size) into the 1124dae1e52cSAmir Goldstein * on-disk inode. We do this via i_disksize, which is the value which 1125dae1e52cSAmir Goldstein * ext4 *really* writes onto the disk inode. 1126dae1e52cSAmir Goldstein */ 1127dae1e52cSAmir Goldstein ei->i_disksize = inode->i_size; 1128dae1e52cSAmir Goldstein 1129dae1e52cSAmir Goldstein if (last_block == max_block) { 1130dae1e52cSAmir Goldstein /* 1131dae1e52cSAmir Goldstein * It is unnecessary to free any data blocks if last_block is 1132dae1e52cSAmir Goldstein * equal to the indirect block limit. 1133dae1e52cSAmir Goldstein */ 1134819c4920STheodore Ts'o return; 1135dae1e52cSAmir Goldstein } else if (n == 1) { /* direct blocks */ 1136dae1e52cSAmir Goldstein ext4_free_data(handle, inode, NULL, i_data+offsets[0], 1137dae1e52cSAmir Goldstein i_data + EXT4_NDIR_BLOCKS); 1138dae1e52cSAmir Goldstein goto do_indirects; 1139dae1e52cSAmir Goldstein } 1140dae1e52cSAmir Goldstein 1141dae1e52cSAmir Goldstein partial = ext4_find_shared(inode, n, offsets, chain, &nr); 1142dae1e52cSAmir Goldstein /* Kill the top of shared branch (not detached) */ 1143dae1e52cSAmir Goldstein if (nr) { 1144dae1e52cSAmir Goldstein if (partial == chain) { 1145dae1e52cSAmir Goldstein /* Shared branch grows from the inode */ 1146dae1e52cSAmir Goldstein ext4_free_branches(handle, inode, NULL, 1147dae1e52cSAmir Goldstein &nr, &nr+1, (chain+n-1) - partial); 1148dae1e52cSAmir Goldstein *partial->p = 0; 1149dae1e52cSAmir Goldstein /* 1150dae1e52cSAmir Goldstein * We mark the inode dirty prior to restart, 1151dae1e52cSAmir Goldstein * and prior to stop. No need for it here. 1152dae1e52cSAmir Goldstein */ 1153dae1e52cSAmir Goldstein } else { 1154dae1e52cSAmir Goldstein /* Shared branch grows from an indirect block */ 1155dae1e52cSAmir Goldstein BUFFER_TRACE(partial->bh, "get_write_access"); 1156dae1e52cSAmir Goldstein ext4_free_branches(handle, inode, partial->bh, 1157dae1e52cSAmir Goldstein partial->p, 1158dae1e52cSAmir Goldstein partial->p+1, (chain+n-1) - partial); 1159dae1e52cSAmir Goldstein } 1160dae1e52cSAmir Goldstein } 1161dae1e52cSAmir Goldstein /* Clear the ends of indirect blocks on the shared branch */ 1162dae1e52cSAmir Goldstein while (partial > chain) { 1163dae1e52cSAmir Goldstein ext4_free_branches(handle, inode, partial->bh, partial->p + 1, 1164dae1e52cSAmir Goldstein (__le32*)partial->bh->b_data+addr_per_block, 1165dae1e52cSAmir Goldstein (chain+n-1) - partial); 1166dae1e52cSAmir Goldstein BUFFER_TRACE(partial->bh, "call brelse"); 1167dae1e52cSAmir Goldstein brelse(partial->bh); 1168dae1e52cSAmir Goldstein partial--; 1169dae1e52cSAmir Goldstein } 1170dae1e52cSAmir Goldstein do_indirects: 1171dae1e52cSAmir Goldstein /* Kill the remaining (whole) subtrees */ 1172dae1e52cSAmir Goldstein switch (offsets[0]) { 1173dae1e52cSAmir Goldstein default: 1174dae1e52cSAmir Goldstein nr = i_data[EXT4_IND_BLOCK]; 1175dae1e52cSAmir Goldstein if (nr) { 1176dae1e52cSAmir Goldstein ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1); 1177dae1e52cSAmir Goldstein i_data[EXT4_IND_BLOCK] = 0; 1178dae1e52cSAmir Goldstein } 1179dae1e52cSAmir Goldstein case EXT4_IND_BLOCK: 1180dae1e52cSAmir Goldstein nr = i_data[EXT4_DIND_BLOCK]; 1181dae1e52cSAmir Goldstein if (nr) { 1182dae1e52cSAmir Goldstein ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2); 1183dae1e52cSAmir Goldstein i_data[EXT4_DIND_BLOCK] = 0; 1184dae1e52cSAmir Goldstein } 1185dae1e52cSAmir Goldstein case EXT4_DIND_BLOCK: 1186dae1e52cSAmir Goldstein nr = i_data[EXT4_TIND_BLOCK]; 1187dae1e52cSAmir Goldstein if (nr) { 1188dae1e52cSAmir Goldstein ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3); 1189dae1e52cSAmir Goldstein i_data[EXT4_TIND_BLOCK] = 0; 1190dae1e52cSAmir Goldstein } 1191dae1e52cSAmir Goldstein case EXT4_TIND_BLOCK: 1192dae1e52cSAmir Goldstein ; 1193dae1e52cSAmir Goldstein } 1194dae1e52cSAmir Goldstein } 1195dae1e52cSAmir Goldstein 11964f579ae7SLukas Czerner /** 11974f579ae7SLukas Czerner * ext4_ind_remove_space - remove space from the range 11984f579ae7SLukas Czerner * @handle: JBD handle for this transaction 11994f579ae7SLukas Czerner * @inode: inode we are dealing with 12004f579ae7SLukas Czerner * @start: First block to remove 12014f579ae7SLukas Czerner * @end: One block after the last block to remove (exclusive) 12024f579ae7SLukas Czerner * 12034f579ae7SLukas Czerner * Free the blocks in the defined range (end is exclusive endpoint of 12044f579ae7SLukas Czerner * range). This is used by ext4_punch_hole(). 12054f579ae7SLukas Czerner */ 12064f579ae7SLukas Czerner int ext4_ind_remove_space(handle_t *handle, struct inode *inode, 12074f579ae7SLukas Czerner ext4_lblk_t start, ext4_lblk_t end) 12088bad6fc8SZheng Liu { 12094f579ae7SLukas Czerner struct ext4_inode_info *ei = EXT4_I(inode); 12104f579ae7SLukas Czerner __le32 *i_data = ei->i_data; 12118bad6fc8SZheng Liu int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); 12124f579ae7SLukas Czerner ext4_lblk_t offsets[4], offsets2[4]; 12134f579ae7SLukas Czerner Indirect chain[4], chain2[4]; 12144f579ae7SLukas Czerner Indirect *partial, *partial2; 12154f579ae7SLukas Czerner ext4_lblk_t max_block; 12164f579ae7SLukas Czerner __le32 nr = 0, nr2 = 0; 12174f579ae7SLukas Czerner int n = 0, n2 = 0; 12184f579ae7SLukas Czerner unsigned blocksize = inode->i_sb->s_blocksize; 12198bad6fc8SZheng Liu 12204f579ae7SLukas Czerner max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1) 12214f579ae7SLukas Czerner >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); 12224f579ae7SLukas Czerner if (end >= max_block) 12234f579ae7SLukas Czerner end = max_block; 12244f579ae7SLukas Czerner if ((start >= end) || (start > max_block)) 12254f579ae7SLukas Czerner return 0; 1226a93cd4cfSJan Kara 12274f579ae7SLukas Czerner n = ext4_block_to_path(inode, start, offsets, NULL); 12284f579ae7SLukas Czerner n2 = ext4_block_to_path(inode, end, offsets2, NULL); 12294f579ae7SLukas Czerner 12304f579ae7SLukas Czerner BUG_ON(n > n2); 12314f579ae7SLukas Czerner 12324f579ae7SLukas Czerner if ((n == 1) && (n == n2)) { 12334f579ae7SLukas Czerner /* We're punching only within direct block range */ 12344f579ae7SLukas Czerner ext4_free_data(handle, inode, NULL, i_data + offsets[0], 12354f579ae7SLukas Czerner i_data + offsets2[0]); 12364f579ae7SLukas Czerner return 0; 12374f579ae7SLukas Czerner } else if (n2 > n) { 12384f579ae7SLukas Czerner /* 12394f579ae7SLukas Czerner * Start and end are on a different levels so we're going to 12404f579ae7SLukas Czerner * free partial block at start, and partial block at end of 12414f579ae7SLukas Czerner * the range. If there are some levels in between then 12424f579ae7SLukas Czerner * do_indirects label will take care of that. 12434f579ae7SLukas Czerner */ 12444f579ae7SLukas Czerner 12454f579ae7SLukas Czerner if (n == 1) { 12464f579ae7SLukas Czerner /* 12474f579ae7SLukas Czerner * Start is at the direct block level, free 12484f579ae7SLukas Czerner * everything to the end of the level. 12494f579ae7SLukas Czerner */ 12504f579ae7SLukas Czerner ext4_free_data(handle, inode, NULL, i_data + offsets[0], 12514f579ae7SLukas Czerner i_data + EXT4_NDIR_BLOCKS); 12524f579ae7SLukas Czerner goto end_range; 12538bad6fc8SZheng Liu } 12544f579ae7SLukas Czerner 12554f579ae7SLukas Czerner 12564f579ae7SLukas Czerner partial = ext4_find_shared(inode, n, offsets, chain, &nr); 12574f579ae7SLukas Czerner if (nr) { 12584f579ae7SLukas Czerner if (partial == chain) { 12594f579ae7SLukas Czerner /* Shared branch grows from the inode */ 12604f579ae7SLukas Czerner ext4_free_branches(handle, inode, NULL, 12614f579ae7SLukas Czerner &nr, &nr+1, (chain+n-1) - partial); 12624f579ae7SLukas Czerner *partial->p = 0; 1263a93cd4cfSJan Kara } else { 12644f579ae7SLukas Czerner /* Shared branch grows from an indirect block */ 12654f579ae7SLukas Czerner BUFFER_TRACE(partial->bh, "get_write_access"); 12664f579ae7SLukas Czerner ext4_free_branches(handle, inode, partial->bh, 12674f579ae7SLukas Czerner partial->p, 12684f579ae7SLukas Czerner partial->p+1, (chain+n-1) - partial); 1269a93cd4cfSJan Kara } 12708bad6fc8SZheng Liu } 12718bad6fc8SZheng Liu 12724f579ae7SLukas Czerner /* 12734f579ae7SLukas Czerner * Clear the ends of indirect blocks on the shared branch 12744f579ae7SLukas Czerner * at the start of the range 12754f579ae7SLukas Czerner */ 12764f579ae7SLukas Czerner while (partial > chain) { 12774f579ae7SLukas Czerner ext4_free_branches(handle, inode, partial->bh, 12784f579ae7SLukas Czerner partial->p + 1, 12794f579ae7SLukas Czerner (__le32 *)partial->bh->b_data+addr_per_block, 12804f579ae7SLukas Czerner (chain+n-1) - partial); 12814f579ae7SLukas Czerner BUFFER_TRACE(partial->bh, "call brelse"); 12824f579ae7SLukas Czerner brelse(partial->bh); 12834f579ae7SLukas Czerner partial--; 12848bad6fc8SZheng Liu } 12858bad6fc8SZheng Liu 12864f579ae7SLukas Czerner end_range: 12874f579ae7SLukas Czerner partial2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2); 12884f579ae7SLukas Czerner if (nr2) { 12894f579ae7SLukas Czerner if (partial2 == chain2) { 12904f579ae7SLukas Czerner /* 12914f579ae7SLukas Czerner * Remember, end is exclusive so here we're at 12924f579ae7SLukas Czerner * the start of the next level we're not going 12934f579ae7SLukas Czerner * to free. Everything was covered by the start 12944f579ae7SLukas Czerner * of the range. 12954f579ae7SLukas Czerner */ 12966f30b7e3SOmar Sandoval goto do_indirects; 12978bad6fc8SZheng Liu } 12984f579ae7SLukas Czerner } else { 12994f579ae7SLukas Czerner /* 13004f579ae7SLukas Czerner * ext4_find_shared returns Indirect structure which 13014f579ae7SLukas Czerner * points to the last element which should not be 13024f579ae7SLukas Czerner * removed by truncate. But this is end of the range 13034f579ae7SLukas Czerner * in punch_hole so we need to point to the next element 13044f579ae7SLukas Czerner */ 13054f579ae7SLukas Czerner partial2->p++; 13064f579ae7SLukas Czerner } 13074f579ae7SLukas Czerner 13084f579ae7SLukas Czerner /* 13094f579ae7SLukas Czerner * Clear the ends of indirect blocks on the shared branch 13104f579ae7SLukas Czerner * at the end of the range 13114f579ae7SLukas Czerner */ 13124f579ae7SLukas Czerner while (partial2 > chain2) { 13134f579ae7SLukas Czerner ext4_free_branches(handle, inode, partial2->bh, 13144f579ae7SLukas Czerner (__le32 *)partial2->bh->b_data, 13154f579ae7SLukas Czerner partial2->p, 13164f579ae7SLukas Czerner (chain2+n2-1) - partial2); 13174f579ae7SLukas Czerner BUFFER_TRACE(partial2->bh, "call brelse"); 13184f579ae7SLukas Czerner brelse(partial2->bh); 13194f579ae7SLukas Czerner partial2--; 13204f579ae7SLukas Czerner } 13214f579ae7SLukas Czerner goto do_indirects; 13224f579ae7SLukas Czerner } 13234f579ae7SLukas Czerner 13244f579ae7SLukas Czerner /* Punch happened within the same level (n == n2) */ 13254f579ae7SLukas Czerner partial = ext4_find_shared(inode, n, offsets, chain, &nr); 13264f579ae7SLukas Czerner partial2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2); 13276f30b7e3SOmar Sandoval 13286f30b7e3SOmar Sandoval /* Free top, but only if partial2 isn't its subtree. */ 13296f30b7e3SOmar Sandoval if (nr) { 13306f30b7e3SOmar Sandoval int level = min(partial - chain, partial2 - chain2); 13316f30b7e3SOmar Sandoval int i; 13326f30b7e3SOmar Sandoval int subtree = 1; 13336f30b7e3SOmar Sandoval 13346f30b7e3SOmar Sandoval for (i = 0; i <= level; i++) { 13356f30b7e3SOmar Sandoval if (offsets[i] != offsets2[i]) { 13366f30b7e3SOmar Sandoval subtree = 0; 13376f30b7e3SOmar Sandoval break; 13386f30b7e3SOmar Sandoval } 13396f30b7e3SOmar Sandoval } 13406f30b7e3SOmar Sandoval 13416f30b7e3SOmar Sandoval if (!subtree) { 13426f30b7e3SOmar Sandoval if (partial == chain) { 13436f30b7e3SOmar Sandoval /* Shared branch grows from the inode */ 13446f30b7e3SOmar Sandoval ext4_free_branches(handle, inode, NULL, 13456f30b7e3SOmar Sandoval &nr, &nr+1, 13466f30b7e3SOmar Sandoval (chain+n-1) - partial); 13476f30b7e3SOmar Sandoval *partial->p = 0; 13486f30b7e3SOmar Sandoval } else { 13496f30b7e3SOmar Sandoval /* Shared branch grows from an indirect block */ 13506f30b7e3SOmar Sandoval BUFFER_TRACE(partial->bh, "get_write_access"); 13516f30b7e3SOmar Sandoval ext4_free_branches(handle, inode, partial->bh, 13526f30b7e3SOmar Sandoval partial->p, 13536f30b7e3SOmar Sandoval partial->p+1, 13546f30b7e3SOmar Sandoval (chain+n-1) - partial); 13556f30b7e3SOmar Sandoval } 13566f30b7e3SOmar Sandoval } 13576f30b7e3SOmar Sandoval } 13586f30b7e3SOmar Sandoval 13596f30b7e3SOmar Sandoval if (!nr2) { 13604f579ae7SLukas Czerner /* 13614f579ae7SLukas Czerner * ext4_find_shared returns Indirect structure which 13624f579ae7SLukas Czerner * points to the last element which should not be 13634f579ae7SLukas Czerner * removed by truncate. But this is end of the range 13644f579ae7SLukas Czerner * in punch_hole so we need to point to the next element 13654f579ae7SLukas Czerner */ 13664f579ae7SLukas Czerner partial2->p++; 13676f30b7e3SOmar Sandoval } 13686f30b7e3SOmar Sandoval 13696f30b7e3SOmar Sandoval while (partial > chain || partial2 > chain2) { 13706f30b7e3SOmar Sandoval int depth = (chain+n-1) - partial; 13716f30b7e3SOmar Sandoval int depth2 = (chain2+n2-1) - partial2; 13726f30b7e3SOmar Sandoval 13736f30b7e3SOmar Sandoval if (partial > chain && partial2 > chain2 && 13746f30b7e3SOmar Sandoval partial->bh->b_blocknr == partial2->bh->b_blocknr) { 13756f30b7e3SOmar Sandoval /* 13766f30b7e3SOmar Sandoval * We've converged on the same block. Clear the range, 13776f30b7e3SOmar Sandoval * then we're done. 13786f30b7e3SOmar Sandoval */ 13794f579ae7SLukas Czerner ext4_free_branches(handle, inode, partial->bh, 13804f579ae7SLukas Czerner partial->p + 1, 13814f579ae7SLukas Czerner partial2->p, 13824f579ae7SLukas Czerner (chain+n-1) - partial); 13834f579ae7SLukas Czerner BUFFER_TRACE(partial->bh, "call brelse"); 13844f579ae7SLukas Czerner brelse(partial->bh); 13854f579ae7SLukas Czerner BUFFER_TRACE(partial2->bh, "call brelse"); 13864f579ae7SLukas Czerner brelse(partial2->bh); 13874f579ae7SLukas Czerner return 0; 13884f579ae7SLukas Czerner } 13896f30b7e3SOmar Sandoval 13904f579ae7SLukas Czerner /* 13916f30b7e3SOmar Sandoval * The start and end partial branches may not be at the same 13926f30b7e3SOmar Sandoval * level even though the punch happened within one level. So, we 13936f30b7e3SOmar Sandoval * give them a chance to arrive at the same level, then walk 13946f30b7e3SOmar Sandoval * them in step with each other until we converge on the same 13956f30b7e3SOmar Sandoval * block. 13964f579ae7SLukas Czerner */ 13976f30b7e3SOmar Sandoval if (partial > chain && depth <= depth2) { 13984f579ae7SLukas Czerner ext4_free_branches(handle, inode, partial->bh, 13994f579ae7SLukas Czerner partial->p + 1, 14004f579ae7SLukas Czerner (__le32 *)partial->bh->b_data+addr_per_block, 14014f579ae7SLukas Czerner (chain+n-1) - partial); 14024f579ae7SLukas Czerner BUFFER_TRACE(partial->bh, "call brelse"); 14034f579ae7SLukas Czerner brelse(partial->bh); 14044f579ae7SLukas Czerner partial--; 14054f579ae7SLukas Czerner } 14066f30b7e3SOmar Sandoval if (partial2 > chain2 && depth2 <= depth) { 14074f579ae7SLukas Czerner ext4_free_branches(handle, inode, partial2->bh, 14084f579ae7SLukas Czerner (__le32 *)partial2->bh->b_data, 14094f579ae7SLukas Czerner partial2->p, 14106f30b7e3SOmar Sandoval (chain2+n2-1) - partial2); 14114f579ae7SLukas Czerner BUFFER_TRACE(partial2->bh, "call brelse"); 14124f579ae7SLukas Czerner brelse(partial2->bh); 14134f579ae7SLukas Czerner partial2--; 14148bad6fc8SZheng Liu } 14158bad6fc8SZheng Liu } 14166f30b7e3SOmar Sandoval return 0; 14178bad6fc8SZheng Liu 14184f579ae7SLukas Czerner do_indirects: 14194f579ae7SLukas Czerner /* Kill the remaining (whole) subtrees */ 14204f579ae7SLukas Czerner switch (offsets[0]) { 14214f579ae7SLukas Czerner default: 14224f579ae7SLukas Czerner if (++n >= n2) 14234f579ae7SLukas Czerner return 0; 14244f579ae7SLukas Czerner nr = i_data[EXT4_IND_BLOCK]; 14254f579ae7SLukas Czerner if (nr) { 14264f579ae7SLukas Czerner ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1); 14274f579ae7SLukas Czerner i_data[EXT4_IND_BLOCK] = 0; 14288bad6fc8SZheng Liu } 14294f579ae7SLukas Czerner case EXT4_IND_BLOCK: 14304f579ae7SLukas Czerner if (++n >= n2) 14314f579ae7SLukas Czerner return 0; 14324f579ae7SLukas Czerner nr = i_data[EXT4_DIND_BLOCK]; 14334f579ae7SLukas Czerner if (nr) { 14344f579ae7SLukas Czerner ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2); 14354f579ae7SLukas Czerner i_data[EXT4_DIND_BLOCK] = 0; 14364f579ae7SLukas Czerner } 14374f579ae7SLukas Czerner case EXT4_DIND_BLOCK: 14384f579ae7SLukas Czerner if (++n >= n2) 14394f579ae7SLukas Czerner return 0; 14404f579ae7SLukas Czerner nr = i_data[EXT4_TIND_BLOCK]; 14414f579ae7SLukas Czerner if (nr) { 14424f579ae7SLukas Czerner ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3); 14434f579ae7SLukas Czerner i_data[EXT4_TIND_BLOCK] = 0; 14444f579ae7SLukas Czerner } 14454f579ae7SLukas Czerner case EXT4_TIND_BLOCK: 14464f579ae7SLukas Czerner ; 14474f579ae7SLukas Czerner } 14484f579ae7SLukas Czerner return 0; 14494f579ae7SLukas Czerner } 1450