fs/ext2/inode.c

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
ext2_inode_is_fast_symlink
ext2_write_failed
ext2_evict_inode
add_chain
verify_chain
ext2_block_to_path
ext2_get_branch
ext2_find_near
ext2_find_goal
ext2_blks_to_allocate
ext2_alloc_blocks
ext2_alloc_branch
ext2_splice_branch
ext2_get_blocks
ext2_get_block
ext2_iomap_begin
ext2_iomap_end
ext2_fiemap
ext2_writepage
ext2_readpage
ext2_readpages
ext2_write_begin
ext2_write_end
ext2_nobh_write_begin
ext2_nobh_writepage
ext2_bmap
ext2_direct_IO
ext2_writepages
ext2_dax_writepages
all_zeroes
ext2_find_shared
ext2_free_data
ext2_free_branches
__ext2_truncate_blocks
ext2_truncate_blocks
ext2_setsize
ext2_get_inode
ext2_set_inode_flags
ext2_set_file_ops
ext2_iget
__ext2_write_inode
ext2_write_inode
ext2_getattr
ext2_setattr
   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext2/inode.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card (card@masi.ibp.fr)
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  from
  11  *
  12  *  linux/fs/minix/inode.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  Goal-directed block allocation by Stephen Tweedie
  17  *      (sct@dcs.ed.ac.uk), 1993, 1998
  18  *  Big-endian to little-endian byte-swapping/bitmaps by
  19  *        David S. Miller (davem@caip.rutgers.edu), 1995
  20  *  64-bit file support on 64-bit platforms by Jakub Jelinek
  21  *      (jj@sunsite.ms.mff.cuni.cz)
  22  *
  23  *  Assorted race fixes, rewrite of ext2_get_block() by Al Viro, 2000
  24  */
  25 
  26 #include <linux/time.h>
  27 #include <linux/highuid.h>
  28 #include <linux/pagemap.h>
  29 #include <linux/dax.h>
  30 #include <linux/blkdev.h>
  31 #include <linux/quotaops.h>
  32 #include <linux/writeback.h>
  33 #include <linux/buffer_head.h>
  34 #include <linux/mpage.h>
  35 #include <linux/fiemap.h>
  36 #include <linux/iomap.h>
  37 #include <linux/namei.h>
  38 #include <linux/uio.h>
  39 #include "ext2.h"
  40 #include "acl.h"
  41 #include "xattr.h"
  42 
  43 static int __ext2_write_inode(struct inode *inode, int do_sync);
  44 
  45 /*
  46  * Test whether an inode is a fast symlink.
  47  */
  48 static inline int ext2_inode_is_fast_symlink(struct inode *inode)
  49 {
  50         int ea_blocks = EXT2_I(inode)->i_file_acl ?
  51                 (inode->i_sb->s_blocksize >> 9) : 0;
  52 
  53         return (S_ISLNK(inode->i_mode) &&
  54                 inode->i_blocks - ea_blocks == 0);
  55 }
  56 
  57 static void ext2_truncate_blocks(struct inode *inode, loff_t offset);
  58 
  59 static void ext2_write_failed(struct address_space *mapping, loff_t to)
  60 {
  61         struct inode *inode = mapping->host;
  62 
  63         if (to > inode->i_size) {
  64                 truncate_pagecache(inode, inode->i_size);
  65                 ext2_truncate_blocks(inode, inode->i_size);
  66         }
  67 }
  68 
  69 /*
  70  * Called at the last iput() if i_nlink is zero.
  71  */
  72 void ext2_evict_inode(struct inode * inode)
  73 {
  74         struct ext2_block_alloc_info *rsv;
  75         int want_delete = 0;
  76 
  77         if (!inode->i_nlink && !is_bad_inode(inode)) {
  78                 want_delete = 1;
  79                 dquot_initialize(inode);
  80         } else {
  81                 dquot_drop(inode);
  82         }
  83 
  84         truncate_inode_pages_final(&inode->i_data);
  85 
  86         if (want_delete) {
  87                 sb_start_intwrite(inode->i_sb);
  88                 /* set dtime */
  89                 EXT2_I(inode)->i_dtime  = ktime_get_real_seconds();
  90                 mark_inode_dirty(inode);
  91                 __ext2_write_inode(inode, inode_needs_sync(inode));
  92                 /* truncate to 0 */
  93                 inode->i_size = 0;
  94                 if (inode->i_blocks)
  95                         ext2_truncate_blocks(inode, 0);
  96                 ext2_xattr_delete_inode(inode);
  97         }
  98 
  99         invalidate_inode_buffers(inode);
 100         clear_inode(inode);
 101 
 102         ext2_discard_reservation(inode);
 103         rsv = EXT2_I(inode)->i_block_alloc_info;
 104         EXT2_I(inode)->i_block_alloc_info = NULL;
 105         if (unlikely(rsv))
 106                 kfree(rsv);
 107 
 108         if (want_delete) {
 109                 ext2_free_inode(inode);
 110                 sb_end_intwrite(inode->i_sb);
 111         }
 112 }
 113 
 114 typedef struct {
 115         __le32  *p;
 116         __le32  key;
 117         struct buffer_head *bh;
 118 } Indirect;
 119 
 120 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
 121 {
 122         p->key = *(p->p = v);
 123         p->bh = bh;
 124 }
 125 
 126 static inline int verify_chain(Indirect *from, Indirect *to)
 127 {
 128         while (from <= to && from->key == *from->p)
 129                 from++;
 130         return (from > to);
 131 }
 132 
 133 /**
 134  *      ext2_block_to_path - parse the block number into array of offsets
 135  *      @inode: inode in question (we are only interested in its superblock)
 136  *      @i_block: block number to be parsed
 137  *      @offsets: array to store the offsets in
 138  *      @boundary: set this non-zero if the referred-to block is likely to be
 139  *             followed (on disk) by an indirect block.
 140  *      To store the locations of file's data ext2 uses a data structure common
 141  *      for UNIX filesystems - tree of pointers anchored in the inode, with
 142  *      data blocks at leaves and indirect blocks in intermediate nodes.
 143  *      This function translates the block number into path in that tree -
 144  *      return value is the path length and @offsets[n] is the offset of
 145  *      pointer to (n+1)th node in the nth one. If @block is out of range
 146  *      (negative or too large) warning is printed and zero returned.
 147  *
 148  *      Note: function doesn't find node addresses, so no IO is needed. All
 149  *      we need to know is the capacity of indirect blocks (taken from the
 150  *      inode->i_sb).
 151  */
 152 
 153 /*
 154  * Portability note: the last comparison (check that we fit into triple
 155  * indirect block) is spelled differently, because otherwise on an
 156  * architecture with 32-bit longs and 8Kb pages we might get into trouble
 157  * if our filesystem had 8Kb blocks. We might use long long, but that would
 158  * kill us on x86. Oh, well, at least the sign propagation does not matter -
 159  * i_block would have to be negative in the very beginning, so we would not
 160  * get there at all.
 161  */
 162 
 163 static int ext2_block_to_path(struct inode *inode,
 164                         long i_block, int offsets[4], int *boundary)
 165 {
 166         int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb);
 167         int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb);
 168         const long direct_blocks = EXT2_NDIR_BLOCKS,
 169                 indirect_blocks = ptrs,
 170                 double_blocks = (1 << (ptrs_bits * 2));
 171         int n = 0;
 172         int final = 0;
 173 
 174         if (i_block < 0) {
 175                 ext2_msg(inode->i_sb, KERN_WARNING,
 176                         "warning: %s: block < 0", __func__);
 177         } else if (i_block < direct_blocks) {
 178                 offsets[n++] = i_block;
 179                 final = direct_blocks;
 180         } else if ( (i_block -= direct_blocks) < indirect_blocks) {
 181                 offsets[n++] = EXT2_IND_BLOCK;
 182                 offsets[n++] = i_block;
 183                 final = ptrs;
 184         } else if ((i_block -= indirect_blocks) < double_blocks) {
 185                 offsets[n++] = EXT2_DIND_BLOCK;
 186                 offsets[n++] = i_block >> ptrs_bits;
 187                 offsets[n++] = i_block & (ptrs - 1);
 188                 final = ptrs;
 189         } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
 190                 offsets[n++] = EXT2_TIND_BLOCK;
 191                 offsets[n++] = i_block >> (ptrs_bits * 2);
 192                 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
 193                 offsets[n++] = i_block & (ptrs - 1);
 194                 final = ptrs;
 195         } else {
 196                 ext2_msg(inode->i_sb, KERN_WARNING,
 197                         "warning: %s: block is too big", __func__);
 198         }
 199         if (boundary)
 200                 *boundary = final - 1 - (i_block & (ptrs - 1));
 201 
 202         return n;
 203 }
 204 
 205 /**
 206  *      ext2_get_branch - read the chain of indirect blocks leading to data
 207  *      @inode: inode in question
 208  *      @depth: depth of the chain (1 - direct pointer, etc.)
 209  *      @offsets: offsets of pointers in inode/indirect blocks
 210  *      @chain: place to store the result
 211  *      @err: here we store the error value
 212  *
 213  *      Function fills the array of triples <key, p, bh> and returns %NULL
 214  *      if everything went OK or the pointer to the last filled triple
 215  *      (incomplete one) otherwise. Upon the return chain[i].key contains
 216  *      the number of (i+1)-th block in the chain (as it is stored in memory,
 217  *      i.e. little-endian 32-bit), chain[i].p contains the address of that
 218  *      number (it points into struct inode for i==0 and into the bh->b_data
 219  *      for i>0) and chain[i].bh points to the buffer_head of i-th indirect
 220  *      block for i>0 and NULL for i==0. In other words, it holds the block
 221  *      numbers of the chain, addresses they were taken from (and where we can
 222  *      verify that chain did not change) and buffer_heads hosting these
 223  *      numbers.
 224  *
 225  *      Function stops when it stumbles upon zero pointer (absent block)
 226  *              (pointer to last triple returned, *@err == 0)
 227  *      or when it gets an IO error reading an indirect block
 228  *              (ditto, *@err == -EIO)
 229  *      or when it notices that chain had been changed while it was reading
 230  *              (ditto, *@err == -EAGAIN)
 231  *      or when it reads all @depth-1 indirect blocks successfully and finds
 232  *      the whole chain, all way to the data (returns %NULL, *err == 0).
 233  */
 234 static Indirect *ext2_get_branch(struct inode *inode,
 235                                  int depth,
 236                                  int *offsets,
 237                                  Indirect chain[4],
 238                                  int *err)
 239 {
 240         struct super_block *sb = inode->i_sb;
 241         Indirect *p = chain;
 242         struct buffer_head *bh;
 243 
 244         *err = 0;
 245         /* i_data is not going away, no lock needed */
 246         add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets);
 247         if (!p->key)
 248                 goto no_block;
 249         while (--depth) {
 250                 bh = sb_bread(sb, le32_to_cpu(p->key));
 251                 if (!bh)
 252                         goto failure;
 253                 read_lock(&EXT2_I(inode)->i_meta_lock);
 254                 if (!verify_chain(chain, p))
 255                         goto changed;
 256                 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
 257                 read_unlock(&EXT2_I(inode)->i_meta_lock);
 258                 if (!p->key)
 259                         goto no_block;
 260         }
 261         return NULL;
 262 
 263 changed:
 264         read_unlock(&EXT2_I(inode)->i_meta_lock);
 265         brelse(bh);
 266         *err = -EAGAIN;
 267         goto no_block;
 268 failure:
 269         *err = -EIO;
 270 no_block:
 271         return p;
 272 }
 273 
 274 /**
 275  *      ext2_find_near - find a place for allocation with sufficient locality
 276  *      @inode: owner
 277  *      @ind: descriptor of indirect block.
 278  *
 279  *      This function returns the preferred place for block allocation.
 280  *      It is used when heuristic for sequential allocation fails.
 281  *      Rules are:
 282  *        + if there is a block to the left of our position - allocate near it.
 283  *        + if pointer will live in indirect block - allocate near that block.
 284  *        + if pointer will live in inode - allocate in the same cylinder group.
 285  *
 286  * In the latter case we colour the starting block by the callers PID to
 287  * prevent it from clashing with concurrent allocations for a different inode
 288  * in the same block group.   The PID is used here so that functionally related
 289  * files will be close-by on-disk.
 290  *
 291  *      Caller must make sure that @ind is valid and will stay that way.
 292  */
 293 
 294 static ext2_fsblk_t ext2_find_near(struct inode *inode, Indirect *ind)
 295 {
 296         struct ext2_inode_info *ei = EXT2_I(inode);
 297         __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
 298         __le32 *p;
 299         ext2_fsblk_t bg_start;
 300         ext2_fsblk_t colour;
 301 
 302         /* Try to find previous block */
 303         for (p = ind->p - 1; p >= start; p--)
 304                 if (*p)
 305                         return le32_to_cpu(*p);
 306 
 307         /* No such thing, so let's try location of indirect block */
 308         if (ind->bh)
 309                 return ind->bh->b_blocknr;
 310 
 311         /*
 312          * It is going to be referred from inode itself? OK, just put it into
 313          * the same cylinder group then.
 314          */
 315         bg_start = ext2_group_first_block_no(inode->i_sb, ei->i_block_group);
 316         colour = (current->pid % 16) *
 317                         (EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 318         return bg_start + colour;
 319 }
 320 
 321 /**
 322  *      ext2_find_goal - find a preferred place for allocation.
 323  *      @inode: owner
 324  *      @block:  block we want
 325  *      @partial: pointer to the last triple within a chain
 326  *
 327  *      Returns preferred place for a block (the goal).
 328  */
 329 
 330 static inline ext2_fsblk_t ext2_find_goal(struct inode *inode, long block,
 331                                           Indirect *partial)
 332 {
 333         struct ext2_block_alloc_info *block_i;
 334 
 335         block_i = EXT2_I(inode)->i_block_alloc_info;
 336 
 337         /*
 338          * try the heuristic for sequential allocation,
 339          * failing that at least try to get decent locality.
 340          */
 341         if (block_i && (block == block_i->last_alloc_logical_block + 1)
 342                 && (block_i->last_alloc_physical_block != 0)) {
 343                 return block_i->last_alloc_physical_block + 1;
 344         }
 345 
 346         return ext2_find_near(inode, partial);
 347 }
 348 
 349 /**
 350  *      ext2_blks_to_allocate: Look up the block map and count the number
 351  *      of direct blocks need to be allocated for the given branch.
 352  *
 353  *      @branch: chain of indirect blocks
 354  *      @k: number of blocks need for indirect blocks
 355  *      @blks: number of data blocks to be mapped.
 356  *      @blocks_to_boundary:  the offset in the indirect block
 357  *
 358  *      return the total number of blocks to be allocate, including the
 359  *      direct and indirect blocks.
 360  */
 361 static int
 362 ext2_blks_to_allocate(Indirect * branch, int k, unsigned long blks,
 363                 int blocks_to_boundary)
 364 {
 365         unsigned long count = 0;
 366 
 367         /*
 368          * Simple case, [t,d]Indirect block(s) has not allocated yet
 369          * then it's clear blocks on that path have not allocated
 370          */
 371         if (k > 0) {
 372                 /* right now don't hanel cross boundary allocation */
 373                 if (blks < blocks_to_boundary + 1)
 374                         count += blks;
 375                 else
 376                         count += blocks_to_boundary + 1;
 377                 return count;
 378         }
 379 
 380         count++;
 381         while (count < blks && count <= blocks_to_boundary
 382                 && le32_to_cpu(*(branch[0].p + count)) == 0) {
 383                 count++;
 384         }
 385         return count;
 386 }
 387 
 388 /**
 389  *      ext2_alloc_blocks: multiple allocate blocks needed for a branch
 390  *      @indirect_blks: the number of blocks need to allocate for indirect
 391  *                      blocks
 392  *
 393  *      @new_blocks: on return it will store the new block numbers for
 394  *      the indirect blocks(if needed) and the first direct block,
 395  *      @blks:  on return it will store the total number of allocated
 396  *              direct blocks
 397  */
 398 static int ext2_alloc_blocks(struct inode *inode,
 399                         ext2_fsblk_t goal, int indirect_blks, int blks,
 400                         ext2_fsblk_t new_blocks[4], int *err)
 401 {
 402         int target, i;
 403         unsigned long count = 0;
 404         int index = 0;
 405         ext2_fsblk_t current_block = 0;
 406         int ret = 0;
 407 
 408         /*
 409          * Here we try to allocate the requested multiple blocks at once,
 410          * on a best-effort basis.
 411          * To build a branch, we should allocate blocks for
 412          * the indirect blocks(if not allocated yet), and at least
 413          * the first direct block of this branch.  That's the
 414          * minimum number of blocks need to allocate(required)
 415          */
 416         target = blks + indirect_blks;
 417 
 418         while (1) {
 419                 count = target;
 420                 /* allocating blocks for indirect blocks and direct blocks */
 421                 current_block = ext2_new_blocks(inode,goal,&count,err);
 422                 if (*err)
 423                         goto failed_out;
 424 
 425                 target -= count;
 426                 /* allocate blocks for indirect blocks */
 427                 while (index < indirect_blks && count) {
 428                         new_blocks[index++] = current_block++;
 429                         count--;
 430                 }
 431 
 432                 if (count > 0)
 433                         break;
 434         }
 435 
 436         /* save the new block number for the first direct block */
 437         new_blocks[index] = current_block;
 438 
 439         /* total number of blocks allocated for direct blocks */
 440         ret = count;
 441         *err = 0;
 442         return ret;
 443 failed_out:
 444         for (i = 0; i <index; i++)
 445                 ext2_free_blocks(inode, new_blocks[i], 1);
 446         if (index)
 447                 mark_inode_dirty(inode);
 448         return ret;
 449 }
 450 
 451 /**
 452  *      ext2_alloc_branch - allocate and set up a chain of blocks.
 453  *      @inode: owner
 454  *      @indirect_blks: depth of the chain (number of blocks to allocate)
 455  *      @blks: number of allocated direct blocks
 456  *      @goal: preferred place for allocation
 457  *      @offsets: offsets (in the blocks) to store the pointers to next.
 458  *      @branch: place to store the chain in.
 459  *
 460  *      This function allocates @num blocks, zeroes out all but the last one,
 461  *      links them into chain and (if we are synchronous) writes them to disk.
 462  *      In other words, it prepares a branch that can be spliced onto the
 463  *      inode. It stores the information about that chain in the branch[], in
 464  *      the same format as ext2_get_branch() would do. We are calling it after
 465  *      we had read the existing part of chain and partial points to the last
 466  *      triple of that (one with zero ->key). Upon the exit we have the same
 467  *      picture as after the successful ext2_get_block(), except that in one
 468  *      place chain is disconnected - *branch->p is still zero (we did not
 469  *      set the last link), but branch->key contains the number that should
 470  *      be placed into *branch->p to fill that gap.
 471  *
 472  *      If allocation fails we free all blocks we've allocated (and forget
 473  *      their buffer_heads) and return the error value the from failed
 474  *      ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain
 475  *      as described above and return 0.
 476  */
 477 
 478 static int ext2_alloc_branch(struct inode *inode,
 479                         int indirect_blks, int *blks, ext2_fsblk_t goal,
 480                         int *offsets, Indirect *branch)
 481 {
 482         int blocksize = inode->i_sb->s_blocksize;
 483         int i, n = 0;
 484         int err = 0;
 485         struct buffer_head *bh;
 486         int num;
 487         ext2_fsblk_t new_blocks[4];
 488         ext2_fsblk_t current_block;
 489 
 490         num = ext2_alloc_blocks(inode, goal, indirect_blks,
 491                                 *blks, new_blocks, &err);
 492         if (err)
 493                 return err;
 494 
 495         branch[0].key = cpu_to_le32(new_blocks[0]);
 496         /*
 497          * metadata blocks and data blocks are allocated.
 498          */
 499         for (n = 1; n <= indirect_blks;  n++) {
 500                 /*
 501                  * Get buffer_head for parent block, zero it out
 502                  * and set the pointer to new one, then send
 503                  * parent to disk.
 504                  */
 505                 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
 506                 if (unlikely(!bh)) {
 507                         err = -ENOMEM;
 508                         goto failed;
 509                 }
 510                 branch[n].bh = bh;
 511                 lock_buffer(bh);
 512                 memset(bh->b_data, 0, blocksize);
 513                 branch[n].p = (__le32 *) bh->b_data + offsets[n];
 514                 branch[n].key = cpu_to_le32(new_blocks[n]);
 515                 *branch[n].p = branch[n].key;
 516                 if ( n == indirect_blks) {
 517                         current_block = new_blocks[n];
 518                         /*
 519                          * End of chain, update the last new metablock of
 520                          * the chain to point to the new allocated
 521                          * data blocks numbers
 522                          */
 523                         for (i=1; i < num; i++)
 524                                 *(branch[n].p + i) = cpu_to_le32(++current_block);
 525                 }
 526                 set_buffer_uptodate(bh);
 527                 unlock_buffer(bh);
 528                 mark_buffer_dirty_inode(bh, inode);
 529                 /* We used to sync bh here if IS_SYNC(inode).
 530                  * But we now rely upon generic_write_sync()
 531                  * and b_inode_buffers.  But not for directories.
 532                  */
 533                 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
 534                         sync_dirty_buffer(bh);
 535         }
 536         *blks = num;
 537         return err;
 538 
 539 failed:
 540         for (i = 1; i < n; i++)
 541                 bforget(branch[i].bh);
 542         for (i = 0; i < indirect_blks; i++)
 543                 ext2_free_blocks(inode, new_blocks[i], 1);
 544         ext2_free_blocks(inode, new_blocks[i], num);
 545         return err;
 546 }
 547 
 548 /**
 549  * ext2_splice_branch - splice the allocated branch onto inode.
 550  * @inode: owner
 551  * @block: (logical) number of block we are adding
 552  * @where: location of missing link
 553  * @num:   number of indirect blocks we are adding
 554  * @blks:  number of direct blocks we are adding
 555  *
 556  * This function fills the missing link and does all housekeeping needed in
 557  * inode (->i_blocks, etc.). In case of success we end up with the full
 558  * chain to new block and return 0.
 559  */
 560 static void ext2_splice_branch(struct inode *inode,
 561                         long block, Indirect *where, int num, int blks)
 562 {
 563         int i;
 564         struct ext2_block_alloc_info *block_i;
 565         ext2_fsblk_t current_block;
 566 
 567         block_i = EXT2_I(inode)->i_block_alloc_info;
 568 
 569         /* XXX LOCKING probably should have i_meta_lock ?*/
 570         /* That's it */
 571 
 572         *where->p = where->key;
 573 
 574         /*
 575          * Update the host buffer_head or inode to point to more just allocated
 576          * direct blocks blocks
 577          */
 578         if (num == 0 && blks > 1) {
 579                 current_block = le32_to_cpu(where->key) + 1;
 580                 for (i = 1; i < blks; i++)
 581                         *(where->p + i ) = cpu_to_le32(current_block++);
 582         }
 583 
 584         /*
 585          * update the most recently allocated logical & physical block
 586          * in i_block_alloc_info, to assist find the proper goal block for next
 587          * allocation
 588          */
 589         if (block_i) {
 590                 block_i->last_alloc_logical_block = block + blks - 1;
 591                 block_i->last_alloc_physical_block =
 592                                 le32_to_cpu(where[num].key) + blks - 1;
 593         }
 594 
 595         /* We are done with atomic stuff, now do the rest of housekeeping */
 596 
 597         /* had we spliced it onto indirect block? */
 598         if (where->bh)
 599                 mark_buffer_dirty_inode(where->bh, inode);
 600 
 601         inode->i_ctime = current_time(inode);
 602         mark_inode_dirty(inode);
 603 }
 604 
 605 /*
 606  * Allocation strategy is simple: if we have to allocate something, we will
 607  * have to go the whole way to leaf. So let's do it before attaching anything
 608  * to tree, set linkage between the newborn blocks, write them if sync is
 609  * required, recheck the path, free and repeat if check fails, otherwise
 610  * set the last missing link (that will protect us from any truncate-generated
 611  * removals - all blocks on the path are immune now) and possibly force the
 612  * write on the parent block.
 613  * That has a nice additional property: no special recovery from the failed
 614  * allocations is needed - we simply release blocks and do not touch anything
 615  * reachable from inode.
 616  *
 617  * `handle' can be NULL if create == 0.
 618  *
 619  * return > 0, # of blocks mapped or allocated.
 620  * return = 0, if plain lookup failed.
 621  * return < 0, error case.
 622  */
 623 static int ext2_get_blocks(struct inode *inode,
 624                            sector_t iblock, unsigned long maxblocks,
 625                            u32 *bno, bool *new, bool *boundary,
 626                            int create)
 627 {
 628         int err;
 629         int offsets[4];
 630         Indirect chain[4];
 631         Indirect *partial;
 632         ext2_fsblk_t goal;
 633         int indirect_blks;
 634         int blocks_to_boundary = 0;
 635         int depth;
 636         struct ext2_inode_info *ei = EXT2_I(inode);
 637         int count = 0;
 638         ext2_fsblk_t first_block = 0;
 639 
 640         BUG_ON(maxblocks == 0);
 641 
 642         depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
 643 
 644         if (depth == 0)
 645                 return -EIO;
 646 
 647         partial = ext2_get_branch(inode, depth, offsets, chain, &err);
 648         /* Simplest case - block found, no allocation needed */
 649         if (!partial) {
 650                 first_block = le32_to_cpu(chain[depth - 1].key);
 651                 count++;
 652                 /*map more blocks*/
 653                 while (count < maxblocks && count <= blocks_to_boundary) {
 654                         ext2_fsblk_t blk;
 655 
 656                         if (!verify_chain(chain, chain + depth - 1)) {
 657                                 /*
 658                                  * Indirect block might be removed by
 659                                  * truncate while we were reading it.
 660                                  * Handling of that case: forget what we've
 661                                  * got now, go to reread.
 662                                  */
 663                                 err = -EAGAIN;
 664                                 count = 0;
 665                                 partial = chain + depth - 1;
 666                                 break;
 667                         }
 668                         blk = le32_to_cpu(*(chain[depth-1].p + count));
 669                         if (blk == first_block + count)
 670                                 count++;
 671                         else
 672                                 break;
 673                 }
 674                 if (err != -EAGAIN)
 675                         goto got_it;
 676         }
 677 
 678         /* Next simple case - plain lookup or failed read of indirect block */
 679         if (!create || err == -EIO)
 680                 goto cleanup;
 681 
 682         mutex_lock(&ei->truncate_mutex);
 683         /*
 684          * If the indirect block is missing while we are reading
 685          * the chain(ext2_get_branch() returns -EAGAIN err), or
 686          * if the chain has been changed after we grab the semaphore,
 687          * (either because another process truncated this branch, or
 688          * another get_block allocated this branch) re-grab the chain to see if
 689          * the request block has been allocated or not.
 690          *
 691          * Since we already block the truncate/other get_block
 692          * at this point, we will have the current copy of the chain when we
 693          * splice the branch into the tree.
 694          */
 695         if (err == -EAGAIN || !verify_chain(chain, partial)) {
 696                 while (partial > chain) {
 697                         brelse(partial->bh);
 698                         partial--;
 699                 }
 700                 partial = ext2_get_branch(inode, depth, offsets, chain, &err);
 701                 if (!partial) {
 702                         count++;
 703                         mutex_unlock(&ei->truncate_mutex);
 704                         goto got_it;
 705                 }
 706 
 707                 if (err) {
 708                         mutex_unlock(&ei->truncate_mutex);
 709                         goto cleanup;
 710                 }
 711         }
 712 
 713         /*
 714          * Okay, we need to do block allocation.  Lazily initialize the block
 715          * allocation info here if necessary
 716         */
 717         if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
 718                 ext2_init_block_alloc_info(inode);
 719 
 720         goal = ext2_find_goal(inode, iblock, partial);
 721 
 722         /* the number of blocks need to allocate for [d,t]indirect blocks */
 723         indirect_blks = (chain + depth) - partial - 1;
 724         /*
 725          * Next look up the indirect map to count the total number of
 726          * direct blocks to allocate for this branch.
 727          */
 728         count = ext2_blks_to_allocate(partial, indirect_blks,
 729                                         maxblocks, blocks_to_boundary);
 730         /*
 731          * XXX ???? Block out ext2_truncate while we alter the tree
 732          */
 733         err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
 734                                 offsets + (partial - chain), partial);
 735 
 736         if (err) {
 737                 mutex_unlock(&ei->truncate_mutex);
 738                 goto cleanup;
 739         }
 740 
 741         if (IS_DAX(inode)) {
 742                 /*
 743                  * We must unmap blocks before zeroing so that writeback cannot
 744                  * overwrite zeros with stale data from block device page cache.
 745                  */
 746                 clean_bdev_aliases(inode->i_sb->s_bdev,
 747                                    le32_to_cpu(chain[depth-1].key),
 748                                    count);
 749                 /*
 750                  * block must be initialised before we put it in the tree
 751                  * so that it's not found by another thread before it's
 752                  * initialised
 753                  */
 754                 err = sb_issue_zeroout(inode->i_sb,
 755                                 le32_to_cpu(chain[depth-1].key), count,
 756                                 GFP_NOFS);
 757                 if (err) {
 758                         mutex_unlock(&ei->truncate_mutex);
 759                         goto cleanup;
 760                 }
 761         }
 762         *new = true;
 763 
 764         ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
 765         mutex_unlock(&ei->truncate_mutex);
 766 got_it:
 767         if (count > blocks_to_boundary)
 768                 *boundary = true;
 769         err = count;
 770         /* Clean up and exit */
 771         partial = chain + depth - 1;    /* the whole chain */
 772 cleanup:
 773         while (partial > chain) {
 774                 brelse(partial->bh);
 775                 partial--;
 776         }
 777         if (err > 0)
 778                 *bno = le32_to_cpu(chain[depth-1].key);
 779         return err;
 780 }
 781 
 782 int ext2_get_block(struct inode *inode, sector_t iblock,
 783                 struct buffer_head *bh_result, int create)
 784 {
 785         unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
 786         bool new = false, boundary = false;
 787         u32 bno;
 788         int ret;
 789 
 790         ret = ext2_get_blocks(inode, iblock, max_blocks, &bno, &new, &boundary,
 791                         create);
 792         if (ret <= 0)
 793                 return ret;
 794 
 795         map_bh(bh_result, inode->i_sb, bno);
 796         bh_result->b_size = (ret << inode->i_blkbits);
 797         if (new)
 798                 set_buffer_new(bh_result);
 799         if (boundary)
 800                 set_buffer_boundary(bh_result);
 801         return 0;
 802 
 803 }
 804 
 805 #ifdef CONFIG_FS_DAX
 806 static int ext2_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 807                 unsigned flags, struct iomap *iomap)
 808 {
 809         unsigned int blkbits = inode->i_blkbits;
 810         unsigned long first_block = offset >> blkbits;
 811         unsigned long max_blocks = (length + (1 << blkbits) - 1) >> blkbits;
 812         struct ext2_sb_info *sbi = EXT2_SB(inode->i_sb);
 813         bool new = false, boundary = false;
 814         u32 bno;
 815         int ret;
 816 
 817         ret = ext2_get_blocks(inode, first_block, max_blocks,
 818                         &bno, &new, &boundary, flags & IOMAP_WRITE);
 819         if (ret < 0)
 820                 return ret;
 821 
 822         iomap->flags = 0;
 823         iomap->bdev = inode->i_sb->s_bdev;
 824         iomap->offset = (u64)first_block << blkbits;
 825         iomap->dax_dev = sbi->s_daxdev;
 826 
 827         if (ret == 0) {
 828                 iomap->type = IOMAP_HOLE;
 829                 iomap->addr = IOMAP_NULL_ADDR;
 830                 iomap->length = 1 << blkbits;
 831         } else {
 832                 iomap->type = IOMAP_MAPPED;
 833                 iomap->addr = (u64)bno << blkbits;
 834                 iomap->length = (u64)ret << blkbits;
 835                 iomap->flags |= IOMAP_F_MERGED;
 836         }
 837 
 838         if (new)
 839                 iomap->flags |= IOMAP_F_NEW;
 840         return 0;
 841 }
 842 
 843 static int
 844 ext2_iomap_end(struct inode *inode, loff_t offset, loff_t length,
 845                 ssize_t written, unsigned flags, struct iomap *iomap)
 846 {
 847         if (iomap->type == IOMAP_MAPPED &&
 848             written < length &&
 849             (flags & IOMAP_WRITE))
 850                 ext2_write_failed(inode->i_mapping, offset + length);
 851         return 0;
 852 }
 853 
 854 const struct iomap_ops ext2_iomap_ops = {
 855         .iomap_begin            = ext2_iomap_begin,
 856         .iomap_end              = ext2_iomap_end,
 857 };
 858 #else
 859 /* Define empty ops for !CONFIG_FS_DAX case to avoid ugly ifdefs */
 860 const struct iomap_ops ext2_iomap_ops;
 861 #endif /* CONFIG_FS_DAX */
 862 
 863 int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 864                 u64 start, u64 len)
 865 {
 866         return generic_block_fiemap(inode, fieinfo, start, len,
 867                                     ext2_get_block);
 868 }
 869 
 870 static int ext2_writepage(struct page *page, struct writeback_control *wbc)
 871 {
 872         return block_write_full_page(page, ext2_get_block, wbc);
 873 }
 874 
 875 static int ext2_readpage(struct file *file, struct page *page)
 876 {
 877         return mpage_readpage(page, ext2_get_block);
 878 }
 879 
 880 static int
 881 ext2_readpages(struct file *file, struct address_space *mapping,
 882                 struct list_head *pages, unsigned nr_pages)
 883 {
 884         return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
 885 }
 886 
 887 static int
 888 ext2_write_begin(struct file *file, struct address_space *mapping,
 889                 loff_t pos, unsigned len, unsigned flags,
 890                 struct page **pagep, void **fsdata)
 891 {
 892         int ret;
 893 
 894         ret = block_write_begin(mapping, pos, len, flags, pagep,
 895                                 ext2_get_block);
 896         if (ret < 0)
 897                 ext2_write_failed(mapping, pos + len);
 898         return ret;
 899 }
 900 
 901 static int ext2_write_end(struct file *file, struct address_space *mapping,
 902                         loff_t pos, unsigned len, unsigned copied,
 903                         struct page *page, void *fsdata)
 904 {
 905         int ret;
 906 
 907         ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
 908         if (ret < len)
 909                 ext2_write_failed(mapping, pos + len);
 910         return ret;
 911 }
 912 
 913 static int
 914 ext2_nobh_write_begin(struct file *file, struct address_space *mapping,
 915                 loff_t pos, unsigned len, unsigned flags,
 916                 struct page **pagep, void **fsdata)
 917 {
 918         int ret;
 919 
 920         ret = nobh_write_begin(mapping, pos, len, flags, pagep, fsdata,
 921                                ext2_get_block);
 922         if (ret < 0)
 923                 ext2_write_failed(mapping, pos + len);
 924         return ret;
 925 }
 926 
 927 static int ext2_nobh_writepage(struct page *page,
 928                         struct writeback_control *wbc)
 929 {
 930         return nobh_writepage(page, ext2_get_block, wbc);
 931 }
 932 
 933 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
 934 {
 935         return generic_block_bmap(mapping,block,ext2_get_block);
 936 }
 937 
 938 static ssize_t
 939 ext2_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 940 {
 941         struct file *file = iocb->ki_filp;
 942         struct address_space *mapping = file->f_mapping;
 943         struct inode *inode = mapping->host;
 944         size_t count = iov_iter_count(iter);
 945         loff_t offset = iocb->ki_pos;
 946         ssize_t ret;
 947 
 948         ret = blockdev_direct_IO(iocb, inode, iter, ext2_get_block);
 949         if (ret < 0 && iov_iter_rw(iter) == WRITE)
 950                 ext2_write_failed(mapping, offset + count);
 951         return ret;
 952 }
 953 
 954 static int
 955 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
 956 {
 957         return mpage_writepages(mapping, wbc, ext2_get_block);
 958 }
 959 
 960 static int
 961 ext2_dax_writepages(struct address_space *mapping, struct writeback_control *wbc)
 962 {
 963         return dax_writeback_mapping_range(mapping,
 964                         mapping->host->i_sb->s_bdev, wbc);
 965 }
 966 
 967 const struct address_space_operations ext2_aops = {
 968         .readpage               = ext2_readpage,
 969         .readpages              = ext2_readpages,
 970         .writepage              = ext2_writepage,
 971         .write_begin            = ext2_write_begin,
 972         .write_end              = ext2_write_end,
 973         .bmap                   = ext2_bmap,
 974         .direct_IO              = ext2_direct_IO,
 975         .writepages             = ext2_writepages,
 976         .migratepage            = buffer_migrate_page,
 977         .is_partially_uptodate  = block_is_partially_uptodate,
 978         .error_remove_page      = generic_error_remove_page,
 979 };
 980 
 981 const struct address_space_operations ext2_nobh_aops = {
 982         .readpage               = ext2_readpage,
 983         .readpages              = ext2_readpages,
 984         .writepage              = ext2_nobh_writepage,
 985         .write_begin            = ext2_nobh_write_begin,
 986         .write_end              = nobh_write_end,
 987         .bmap                   = ext2_bmap,
 988         .direct_IO              = ext2_direct_IO,
 989         .writepages             = ext2_writepages,
 990         .migratepage            = buffer_migrate_page,
 991         .error_remove_page      = generic_error_remove_page,
 992 };
 993 
 994 static const struct address_space_operations ext2_dax_aops = {
 995         .writepages             = ext2_dax_writepages,
 996         .direct_IO              = noop_direct_IO,
 997         .set_page_dirty         = noop_set_page_dirty,
 998         .invalidatepage         = noop_invalidatepage,
 999 };
1000 
1001 /*
1002  * Probably it should be a library function... search for first non-zero word
1003  * or memcmp with zero_page, whatever is better for particular architecture.
1004  * Linus?
1005  */
1006 static inline int all_zeroes(__le32 *p, __le32 *q)
1007 {
1008         while (p < q)
1009                 if (*p++)
1010                         return 0;
1011         return 1;
1012 }
1013 
1014 /**
1015  *      ext2_find_shared - find the indirect blocks for partial truncation.
1016  *      @inode:   inode in question
1017  *      @depth:   depth of the affected branch
1018  *      @offsets: offsets of pointers in that branch (see ext2_block_to_path)
1019  *      @chain:   place to store the pointers to partial indirect blocks
1020  *      @top:     place to the (detached) top of branch
1021  *
1022  *      This is a helper function used by ext2_truncate().
1023  *
1024  *      When we do truncate() we may have to clean the ends of several indirect
1025  *      blocks but leave the blocks themselves alive. Block is partially
1026  *      truncated if some data below the new i_size is referred from it (and
1027  *      it is on the path to the first completely truncated data block, indeed).
1028  *      We have to free the top of that path along with everything to the right
1029  *      of the path. Since no allocation past the truncation point is possible
1030  *      until ext2_truncate() finishes, we may safely do the latter, but top
1031  *      of branch may require special attention - pageout below the truncation
1032  *      point might try to populate it.
1033  *
1034  *      We atomically detach the top of branch from the tree, store the block
1035  *      number of its root in *@top, pointers to buffer_heads of partially
1036  *      truncated blocks - in @chain[].bh and pointers to their last elements
1037  *      that should not be removed - in @chain[].p. Return value is the pointer
1038  *      to last filled element of @chain.
1039  *
1040  *      The work left to caller to do the actual freeing of subtrees:
1041  *              a) free the subtree starting from *@top
1042  *              b) free the subtrees whose roots are stored in
1043  *                      (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1044  *              c) free the subtrees growing from the inode past the @chain[0].p
1045  *                      (no partially truncated stuff there).
1046  */
1047 
1048 static Indirect *ext2_find_shared(struct inode *inode,
1049                                 int depth,
1050                                 int offsets[4],
1051                                 Indirect chain[4],
1052                                 __le32 *top)
1053 {
1054         Indirect *partial, *p;
1055         int k, err;
1056 
1057         *top = 0;
1058         for (k = depth; k > 1 && !offsets[k-1]; k--)
1059                 ;
1060         partial = ext2_get_branch(inode, k, offsets, chain, &err);
1061         if (!partial)
1062                 partial = chain + k-1;
1063         /*
1064          * If the branch acquired continuation since we've looked at it -
1065          * fine, it should all survive and (new) top doesn't belong to us.
1066          */
1067         write_lock(&EXT2_I(inode)->i_meta_lock);
1068         if (!partial->key && *partial->p) {
1069                 write_unlock(&EXT2_I(inode)->i_meta_lock);
1070                 goto no_top;
1071         }
1072         for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
1073                 ;
1074         /*
1075          * OK, we've found the last block that must survive. The rest of our
1076          * branch should be detached before unlocking. However, if that rest
1077          * of branch is all ours and does not grow immediately from the inode
1078          * it's easier to cheat and just decrement partial->p.
1079          */
1080         if (p == chain + k - 1 && p > chain) {
1081                 p->p--;
1082         } else {
1083                 *top = *p->p;
1084                 *p->p = 0;
1085         }
1086         write_unlock(&EXT2_I(inode)->i_meta_lock);
1087 
1088         while(partial > p)
1089         {
1090                 brelse(partial->bh);
1091                 partial--;
1092         }
1093 no_top:
1094         return partial;
1095 }
1096 
1097 /**
1098  *      ext2_free_data - free a list of data blocks
1099  *      @inode: inode we are dealing with
1100  *      @p:     array of block numbers
1101  *      @q:     points immediately past the end of array
1102  *
1103  *      We are freeing all blocks referred from that array (numbers are
1104  *      stored as little-endian 32-bit) and updating @inode->i_blocks
1105  *      appropriately.
1106  */
1107 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
1108 {
1109         unsigned long block_to_free = 0, count = 0;
1110         unsigned long nr;
1111 
1112         for ( ; p < q ; p++) {
1113                 nr = le32_to_cpu(*p);
1114                 if (nr) {
1115                         *p = 0;
1116                         /* accumulate blocks to free if they're contiguous */
1117                         if (count == 0)
1118                                 goto free_this;
1119                         else if (block_to_free == nr - count)
1120                                 count++;
1121                         else {
1122                                 ext2_free_blocks (inode, block_to_free, count);
1123                                 mark_inode_dirty(inode);
1124                         free_this:
1125                                 block_to_free = nr;
1126                                 count = 1;
1127                         }
1128                 }
1129         }
1130         if (count > 0) {
1131                 ext2_free_blocks (inode, block_to_free, count);
1132                 mark_inode_dirty(inode);
1133         }
1134 }
1135 
1136 /**
1137  *      ext2_free_branches - free an array of branches
1138  *      @inode: inode we are dealing with
1139  *      @p:     array of block numbers
1140  *      @q:     pointer immediately past the end of array
1141  *      @depth: depth of the branches to free
1142  *
1143  *      We are freeing all blocks referred from these branches (numbers are
1144  *      stored as little-endian 32-bit) and updating @inode->i_blocks
1145  *      appropriately.
1146  */
1147 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
1148 {
1149         struct buffer_head * bh;
1150         unsigned long nr;
1151 
1152         if (depth--) {
1153                 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1154                 for ( ; p < q ; p++) {
1155                         nr = le32_to_cpu(*p);
1156                         if (!nr)
1157                                 continue;
1158                         *p = 0;
1159                         bh = sb_bread(inode->i_sb, nr);
1160                         /*
1161                          * A read failure? Report error and clear slot
1162                          * (should be rare).
1163                          */ 
1164                         if (!bh) {
1165                                 ext2_error(inode->i_sb, "ext2_free_branches",
1166                                         "Read failure, inode=%ld, block=%ld",
1167                                         inode->i_ino, nr);
1168                                 continue;
1169                         }
1170                         ext2_free_branches(inode,
1171                                            (__le32*)bh->b_data,
1172                                            (__le32*)bh->b_data + addr_per_block,
1173                                            depth);
1174                         bforget(bh);
1175                         ext2_free_blocks(inode, nr, 1);
1176                         mark_inode_dirty(inode);
1177                 }
1178         } else
1179                 ext2_free_data(inode, p, q);
1180 }
1181 
1182 /* dax_sem must be held when calling this function */
1183 static void __ext2_truncate_blocks(struct inode *inode, loff_t offset)
1184 {
1185         __le32 *i_data = EXT2_I(inode)->i_data;
1186         struct ext2_inode_info *ei = EXT2_I(inode);
1187         int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1188         int offsets[4];
1189         Indirect chain[4];
1190         Indirect *partial;
1191         __le32 nr = 0;
1192         int n;
1193         long iblock;
1194         unsigned blocksize;
1195         blocksize = inode->i_sb->s_blocksize;
1196         iblock = (offset + blocksize-1) >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1197 
1198 #ifdef CONFIG_FS_DAX
1199         WARN_ON(!rwsem_is_locked(&ei->dax_sem));
1200 #endif
1201 
1202         n = ext2_block_to_path(inode, iblock, offsets, NULL);
1203         if (n == 0)
1204                 return;
1205 
1206         /*
1207          * From here we block out all ext2_get_block() callers who want to
1208          * modify the block allocation tree.
1209          */
1210         mutex_lock(&ei->truncate_mutex);
1211 
1212         if (n == 1) {
1213                 ext2_free_data(inode, i_data+offsets[0],
1214                                         i_data + EXT2_NDIR_BLOCKS);
1215                 goto do_indirects;
1216         }
1217 
1218         partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1219         /* Kill the top of shared branch (already detached) */
1220         if (nr) {
1221                 if (partial == chain)
1222                         mark_inode_dirty(inode);
1223                 else
1224                         mark_buffer_dirty_inode(partial->bh, inode);
1225                 ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1226         }
1227         /* Clear the ends of indirect blocks on the shared branch */
1228         while (partial > chain) {
1229                 ext2_free_branches(inode,
1230                                    partial->p + 1,
1231                                    (__le32*)partial->bh->b_data+addr_per_block,
1232                                    (chain+n-1) - partial);
1233                 mark_buffer_dirty_inode(partial->bh, inode);
1234                 brelse (partial->bh);
1235                 partial--;
1236         }
1237 do_indirects:
1238         /* Kill the remaining (whole) subtrees */
1239         switch (offsets[0]) {
1240                 default:
1241                         nr = i_data[EXT2_IND_BLOCK];
1242                         if (nr) {
1243                                 i_data[EXT2_IND_BLOCK] = 0;
1244                                 mark_inode_dirty(inode);
1245                                 ext2_free_branches(inode, &nr, &nr+1, 1);
1246                         }
1247                         /* fall through */
1248                 case EXT2_IND_BLOCK:
1249                         nr = i_data[EXT2_DIND_BLOCK];
1250                         if (nr) {
1251                                 i_data[EXT2_DIND_BLOCK] = 0;
1252                                 mark_inode_dirty(inode);
1253                                 ext2_free_branches(inode, &nr, &nr+1, 2);
1254                         }
1255                         /* fall through */
1256                 case EXT2_DIND_BLOCK:
1257                         nr = i_data[EXT2_TIND_BLOCK];
1258                         if (nr) {
1259                                 i_data[EXT2_TIND_BLOCK] = 0;
1260                                 mark_inode_dirty(inode);
1261                                 ext2_free_branches(inode, &nr, &nr+1, 3);
1262                         }
1263                 case EXT2_TIND_BLOCK:
1264                         ;
1265         }
1266 
1267         ext2_discard_reservation(inode);
1268 
1269         mutex_unlock(&ei->truncate_mutex);
1270 }
1271 
1272 static void ext2_truncate_blocks(struct inode *inode, loff_t offset)
1273 {
1274         if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1275             S_ISLNK(inode->i_mode)))
1276                 return;
1277         if (ext2_inode_is_fast_symlink(inode))
1278                 return;
1279 
1280         dax_sem_down_write(EXT2_I(inode));
1281         __ext2_truncate_blocks(inode, offset);
1282         dax_sem_up_write(EXT2_I(inode));
1283 }
1284 
1285 static int ext2_setsize(struct inode *inode, loff_t newsize)
1286 {
1287         int error;
1288 
1289         if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1290             S_ISLNK(inode->i_mode)))
1291                 return -EINVAL;
1292         if (ext2_inode_is_fast_symlink(inode))
1293                 return -EINVAL;
1294         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1295                 return -EPERM;
1296 
1297         inode_dio_wait(inode);
1298 
1299         if (IS_DAX(inode)) {
1300                 error = iomap_zero_range(inode, newsize,
1301                                          PAGE_ALIGN(newsize) - newsize, NULL,
1302                                          &ext2_iomap_ops);
1303         } else if (test_opt(inode->i_sb, NOBH))
1304                 error = nobh_truncate_page(inode->i_mapping,
1305                                 newsize, ext2_get_block);
1306         else
1307                 error = block_truncate_page(inode->i_mapping,
1308                                 newsize, ext2_get_block);
1309         if (error)
1310                 return error;
1311 
1312         dax_sem_down_write(EXT2_I(inode));
1313         truncate_setsize(inode, newsize);
1314         __ext2_truncate_blocks(inode, newsize);
1315         dax_sem_up_write(EXT2_I(inode));
1316 
1317         inode->i_mtime = inode->i_ctime = current_time(inode);
1318         if (inode_needs_sync(inode)) {
1319                 sync_mapping_buffers(inode->i_mapping);
1320                 sync_inode_metadata(inode, 1);
1321         } else {
1322                 mark_inode_dirty(inode);
1323         }
1324 
1325         return 0;
1326 }
1327 
1328 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1329                                         struct buffer_head **p)
1330 {
1331         struct buffer_head * bh;
1332         unsigned long block_group;
1333         unsigned long block;
1334         unsigned long offset;
1335         struct ext2_group_desc * gdp;
1336 
1337         *p = NULL;
1338         if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1339             ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1340                 goto Einval;
1341 
1342         block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1343         gdp = ext2_get_group_desc(sb, block_group, NULL);
1344         if (!gdp)
1345                 goto Egdp;
1346         /*
1347          * Figure out the offset within the block group inode table
1348          */
1349         offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1350         block = le32_to_cpu(gdp->bg_inode_table) +
1351                 (offset >> EXT2_BLOCK_SIZE_BITS(sb));
1352         if (!(bh = sb_bread(sb, block)))
1353                 goto Eio;
1354 
1355         *p = bh;
1356         offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1357         return (struct ext2_inode *) (bh->b_data + offset);
1358 
1359 Einval:
1360         ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1361                    (unsigned long) ino);
1362         return ERR_PTR(-EINVAL);
1363 Eio:
1364         ext2_error(sb, "ext2_get_inode",
1365                    "unable to read inode block - inode=%lu, block=%lu",
1366                    (unsigned long) ino, block);
1367 Egdp:
1368         return ERR_PTR(-EIO);
1369 }
1370 
1371 void ext2_set_inode_flags(struct inode *inode)
1372 {
1373         unsigned int flags = EXT2_I(inode)->i_flags;
1374 
1375         inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
1376                                 S_DIRSYNC | S_DAX);
1377         if (flags & EXT2_SYNC_FL)
1378                 inode->i_flags |= S_SYNC;
1379         if (flags & EXT2_APPEND_FL)
1380                 inode->i_flags |= S_APPEND;
1381         if (flags & EXT2_IMMUTABLE_FL)
1382                 inode->i_flags |= S_IMMUTABLE;
1383         if (flags & EXT2_NOATIME_FL)
1384                 inode->i_flags |= S_NOATIME;
1385         if (flags & EXT2_DIRSYNC_FL)
1386                 inode->i_flags |= S_DIRSYNC;
1387         if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode))
1388                 inode->i_flags |= S_DAX;
1389 }
1390 
1391 void ext2_set_file_ops(struct inode *inode)
1392 {
1393         inode->i_op = &ext2_file_inode_operations;
1394         inode->i_fop = &ext2_file_operations;
1395         if (IS_DAX(inode))
1396                 inode->i_mapping->a_ops = &ext2_dax_aops;
1397         else if (test_opt(inode->i_sb, NOBH))
1398                 inode->i_mapping->a_ops = &ext2_nobh_aops;
1399         else
1400                 inode->i_mapping->a_ops = &ext2_aops;
1401 }
1402 
1403 struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
1404 {
1405         struct ext2_inode_info *ei;
1406         struct buffer_head * bh = NULL;
1407         struct ext2_inode *raw_inode;
1408         struct inode *inode;
1409         long ret = -EIO;
1410         int n;
1411         uid_t i_uid;
1412         gid_t i_gid;
1413 
1414         inode = iget_locked(sb, ino);
1415         if (!inode)
1416                 return ERR_PTR(-ENOMEM);
1417         if (!(inode->i_state & I_NEW))
1418                 return inode;
1419 
1420         ei = EXT2_I(inode);
1421         ei->i_block_alloc_info = NULL;
1422 
1423         raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1424         if (IS_ERR(raw_inode)) {
1425                 ret = PTR_ERR(raw_inode);
1426                 goto bad_inode;
1427         }
1428 
1429         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1430         i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1431         i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1432         if (!(test_opt (inode->i_sb, NO_UID32))) {
1433                 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1434                 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1435         }
1436         i_uid_write(inode, i_uid);
1437         i_gid_write(inode, i_gid);
1438         set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
1439         inode->i_size = le32_to_cpu(raw_inode->i_size);
1440         inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
1441         inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
1442         inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
1443         inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1444         ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1445         /* We now have enough fields to check if the inode was active or not.
1446          * This is needed because nfsd might try to access dead inodes
1447          * the test is that same one that e2fsck uses
1448          * NeilBrown 1999oct15
1449          */
1450         if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1451                 /* this inode is deleted */
1452                 ret = -ESTALE;
1453                 goto bad_inode;
1454         }
1455         inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1456         ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1457         ext2_set_inode_flags(inode);
1458         ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1459         ei->i_frag_no = raw_inode->i_frag;
1460         ei->i_frag_size = raw_inode->i_fsize;
1461         ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1462         ei->i_dir_acl = 0;
1463 
1464         if (ei->i_file_acl &&
1465             !ext2_data_block_valid(EXT2_SB(sb), ei->i_file_acl, 1)) {
1466                 ext2_error(sb, "ext2_iget", "bad extended attribute block %u",
1467                            ei->i_file_acl);
1468                 ret = -EFSCORRUPTED;
1469                 goto bad_inode;
1470         }
1471 
1472         if (S_ISREG(inode->i_mode))
1473                 inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1474         else
1475                 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1476         if (i_size_read(inode) < 0) {
1477                 ret = -EFSCORRUPTED;
1478                 goto bad_inode;
1479         }
1480         ei->i_dtime = 0;
1481         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1482         ei->i_state = 0;
1483         ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1484         ei->i_dir_start_lookup = 0;
1485 
1486         /*
1487          * NOTE! The in-memory inode i_data array is in little-endian order
1488          * even on big-endian machines: we do NOT byteswap the block numbers!
1489          */
1490         for (n = 0; n < EXT2_N_BLOCKS; n++)
1491                 ei->i_data[n] = raw_inode->i_block[n];
1492 
1493         if (S_ISREG(inode->i_mode)) {
1494                 ext2_set_file_ops(inode);
1495         } else if (S_ISDIR(inode->i_mode)) {
1496                 inode->i_op = &ext2_dir_inode_operations;
1497                 inode->i_fop = &ext2_dir_operations;
1498                 if (test_opt(inode->i_sb, NOBH))
1499                         inode->i_mapping->a_ops = &ext2_nobh_aops;
1500                 else
1501                         inode->i_mapping->a_ops = &ext2_aops;
1502         } else if (S_ISLNK(inode->i_mode)) {
1503                 if (ext2_inode_is_fast_symlink(inode)) {
1504                         inode->i_link = (char *)ei->i_data;
1505                         inode->i_op = &ext2_fast_symlink_inode_operations;
1506                         nd_terminate_link(ei->i_data, inode->i_size,
1507                                 sizeof(ei->i_data) - 1);
1508                 } else {
1509                         inode->i_op = &ext2_symlink_inode_operations;
1510                         inode_nohighmem(inode);
1511                         if (test_opt(inode->i_sb, NOBH))
1512                                 inode->i_mapping->a_ops = &ext2_nobh_aops;
1513                         else
1514                                 inode->i_mapping->a_ops = &ext2_aops;
1515                 }
1516         } else {
1517                 inode->i_op = &ext2_special_inode_operations;
1518                 if (raw_inode->i_block[0])
1519                         init_special_inode(inode, inode->i_mode,
1520                            old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1521                 else 
1522                         init_special_inode(inode, inode->i_mode,
1523                            new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1524         }
1525         brelse (bh);
1526         unlock_new_inode(inode);
1527         return inode;
1528         
1529 bad_inode:
1530         brelse(bh);
1531         iget_failed(inode);
1532         return ERR_PTR(ret);
1533 }
1534 
1535 static int __ext2_write_inode(struct inode *inode, int do_sync)
1536 {
1537         struct ext2_inode_info *ei = EXT2_I(inode);
1538         struct super_block *sb = inode->i_sb;
1539         ino_t ino = inode->i_ino;
1540         uid_t uid = i_uid_read(inode);
1541         gid_t gid = i_gid_read(inode);
1542         struct buffer_head * bh;
1543         struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1544         int n;
1545         int err = 0;
1546 
1547         if (IS_ERR(raw_inode))
1548                 return -EIO;
1549 
1550         /* For fields not not tracking in the in-memory inode,
1551          * initialise them to zero for new inodes. */
1552         if (ei->i_state & EXT2_STATE_NEW)
1553                 memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1554 
1555         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1556         if (!(test_opt(sb, NO_UID32))) {
1557                 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1558                 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1559 /*
1560  * Fix up interoperability with old kernels. Otherwise, old inodes get
1561  * re-used with the upper 16 bits of the uid/gid intact
1562  */
1563                 if (!ei->i_dtime) {
1564                         raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1565                         raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1566                 } else {
1567                         raw_inode->i_uid_high = 0;
1568                         raw_inode->i_gid_high = 0;
1569                 }
1570         } else {
1571                 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1572                 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1573                 raw_inode->i_uid_high = 0;
1574                 raw_inode->i_gid_high = 0;
1575         }
1576         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1577         raw_inode->i_size = cpu_to_le32(inode->i_size);
1578         raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1579         raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1580         raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1581 
1582         raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1583         raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1584         raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1585         raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1586         raw_inode->i_frag = ei->i_frag_no;
1587         raw_inode->i_fsize = ei->i_frag_size;
1588         raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1589         if (!S_ISREG(inode->i_mode))
1590                 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1591         else {
1592                 raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1593                 if (inode->i_size > 0x7fffffffULL) {
1594                         if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1595                                         EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1596                             EXT2_SB(sb)->s_es->s_rev_level ==
1597                                         cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1598                                /* If this is the first large file
1599                                 * created, add a flag to the superblock.
1600                                 */
1601                                 spin_lock(&EXT2_SB(sb)->s_lock);
1602                                 ext2_update_dynamic_rev(sb);
1603                                 EXT2_SET_RO_COMPAT_FEATURE(sb,
1604                                         EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1605                                 spin_unlock(&EXT2_SB(sb)->s_lock);
1606                                 ext2_sync_super(sb, EXT2_SB(sb)->s_es, 1);
1607                         }
1608                 }
1609         }
1610         
1611         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1612         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1613                 if (old_valid_dev(inode->i_rdev)) {
1614                         raw_inode->i_block[0] =
1615                                 cpu_to_le32(old_encode_dev(inode->i_rdev));
1616                         raw_inode->i_block[1] = 0;
1617                 } else {
1618                         raw_inode->i_block[0] = 0;
1619                         raw_inode->i_block[1] =
1620                                 cpu_to_le32(new_encode_dev(inode->i_rdev));
1621                         raw_inode->i_block[2] = 0;
1622                 }
1623         } else for (n = 0; n < EXT2_N_BLOCKS; n++)
1624                 raw_inode->i_block[n] = ei->i_data[n];
1625         mark_buffer_dirty(bh);
1626         if (do_sync) {
1627                 sync_dirty_buffer(bh);
1628                 if (buffer_req(bh) && !buffer_uptodate(bh)) {
1629                         printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1630                                 sb->s_id, (unsigned long) ino);
1631                         err = -EIO;
1632                 }
1633         }
1634         ei->i_state &= ~EXT2_STATE_NEW;
1635         brelse (bh);
1636         return err;
1637 }
1638 
1639 int ext2_write_inode(struct inode *inode, struct writeback_control *wbc)
1640 {
1641         return __ext2_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1642 }
1643 
1644 int ext2_getattr(const struct path *path, struct kstat *stat,
1645                 u32 request_mask, unsigned int query_flags)
1646 {
1647         struct inode *inode = d_inode(path->dentry);
1648         struct ext2_inode_info *ei = EXT2_I(inode);
1649         unsigned int flags;
1650 
1651         flags = ei->i_flags & EXT2_FL_USER_VISIBLE;
1652         if (flags & EXT2_APPEND_FL)
1653                 stat->attributes |= STATX_ATTR_APPEND;
1654         if (flags & EXT2_COMPR_FL)
1655                 stat->attributes |= STATX_ATTR_COMPRESSED;
1656         if (flags & EXT2_IMMUTABLE_FL)
1657                 stat->attributes |= STATX_ATTR_IMMUTABLE;
1658         if (flags & EXT2_NODUMP_FL)
1659                 stat->attributes |= STATX_ATTR_NODUMP;
1660         stat->attributes_mask |= (STATX_ATTR_APPEND |
1661                         STATX_ATTR_COMPRESSED |
1662                         STATX_ATTR_ENCRYPTED |
1663                         STATX_ATTR_IMMUTABLE |
1664                         STATX_ATTR_NODUMP);
1665 
1666         generic_fillattr(inode, stat);
1667         return 0;
1668 }
1669 
1670 int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1671 {
1672         struct inode *inode = d_inode(dentry);
1673         int error;
1674 
1675         error = setattr_prepare(dentry, iattr);
1676         if (error)
1677                 return error;
1678 
1679         if (is_quota_modification(inode, iattr)) {
1680                 error = dquot_initialize(inode);
1681                 if (error)
1682                         return error;
1683         }
1684         if ((iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)) ||
1685             (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid))) {
1686                 error = dquot_transfer(inode, iattr);
1687                 if (error)
1688                         return error;
1689         }
1690         if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) {
1691                 error = ext2_setsize(inode, iattr->ia_size);
1692                 if (error)
1693                         return error;
1694         }
1695         setattr_copy(inode, iattr);
1696         if (iattr->ia_valid & ATTR_MODE)
1697                 error = posix_acl_chmod(inode, inode->i_mode);
1698         mark_inode_dirty(inode);
1699 
1700         return error;
1701 }
/* [<][>][^][v][top][bottom][index][help] */
root/fs/ext2/inode.c

DEFINITIONS
