Back to home page

OSCL-LXR

 
 

    


0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  * linux/fs/ext2/xattr.c
0004  *
0005  * Copyright (C) 2001-2003 Andreas Gruenbacher <agruen@suse.de>
0006  *
0007  * Fix by Harrison Xing <harrison@mountainviewdata.com>.
0008  * Extended attributes for symlinks and special files added per
0009  *  suggestion of Luka Renko <luka.renko@hermes.si>.
0010  * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
0011  *  Red Hat Inc.
0012  *
0013  */
0014 
0015 /*
0016  * Extended attributes are stored on disk blocks allocated outside of
0017  * any inode. The i_file_acl field is then made to point to this allocated
0018  * block. If all extended attributes of an inode are identical, these
0019  * inodes may share the same extended attribute block. Such situations
0020  * are automatically detected by keeping a cache of recent attribute block
0021  * numbers and hashes over the block's contents in memory.
0022  *
0023  *
0024  * Extended attribute block layout:
0025  *
0026  *   +------------------+
0027  *   | header           |
0028  *   | entry 1          | |
0029  *   | entry 2          | | growing downwards
0030  *   | entry 3          | v
0031  *   | four null bytes  |
0032  *   | . . .            |
0033  *   | value 1          | ^
0034  *   | value 3          | | growing upwards
0035  *   | value 2          | |
0036  *   +------------------+
0037  *
0038  * The block header is followed by multiple entry descriptors. These entry
0039  * descriptors are variable in size, and aligned to EXT2_XATTR_PAD
0040  * byte boundaries. The entry descriptors are sorted by attribute name,
0041  * so that two extended attribute blocks can be compared efficiently.
0042  *
0043  * Attribute values are aligned to the end of the block, stored in
0044  * no specific order. They are also padded to EXT2_XATTR_PAD byte
0045  * boundaries. No additional gaps are left between them.
0046  *
0047  * Locking strategy
0048  * ----------------
0049  * EXT2_I(inode)->i_file_acl is protected by EXT2_I(inode)->xattr_sem.
0050  * EA blocks are only changed if they are exclusive to an inode, so
0051  * holding xattr_sem also means that nothing but the EA block's reference
0052  * count will change. Multiple writers to an EA block are synchronized
0053  * by the bh lock. No more than a single bh lock is held at any time
0054  * to avoid deadlocks.
0055  */
0056 
0057 #include <linux/buffer_head.h>
0058 #include <linux/init.h>
0059 #include <linux/printk.h>
0060 #include <linux/slab.h>
0061 #include <linux/mbcache.h>
0062 #include <linux/quotaops.h>
0063 #include <linux/rwsem.h>
0064 #include <linux/security.h>
0065 #include "ext2.h"
0066 #include "xattr.h"
0067 #include "acl.h"
0068 
0069 #define HDR(bh) ((struct ext2_xattr_header *)((bh)->b_data))
0070 #define ENTRY(ptr) ((struct ext2_xattr_entry *)(ptr))
0071 #define FIRST_ENTRY(bh) ENTRY(HDR(bh)+1)
0072 #define IS_LAST_ENTRY(entry) (*(__u32 *)(entry) == 0)
0073 
0074 #ifdef EXT2_XATTR_DEBUG
0075 # define ea_idebug(inode, f...) do { \
0076         printk(KERN_DEBUG "inode %s:%ld: ", \
0077             inode->i_sb->s_id, inode->i_ino); \
0078         printk(f); \
0079         printk("\n"); \
0080     } while (0)
0081 # define ea_bdebug(bh, f...) do { \
0082         printk(KERN_DEBUG "block %pg:%lu: ", \
0083             bh->b_bdev, (unsigned long) bh->b_blocknr); \
0084         printk(f); \
0085         printk("\n"); \
0086     } while (0)
0087 #else
0088 # define ea_idebug(inode, f...) no_printk(f)
0089 # define ea_bdebug(bh, f...)    no_printk(f)
0090 #endif
0091 
0092 static int ext2_xattr_set2(struct inode *, struct buffer_head *,
0093                struct ext2_xattr_header *);
0094 
0095 static int ext2_xattr_cache_insert(struct mb_cache *, struct buffer_head *);
0096 static struct buffer_head *ext2_xattr_cache_find(struct inode *,
0097                          struct ext2_xattr_header *);
0098 static void ext2_xattr_rehash(struct ext2_xattr_header *,
0099                   struct ext2_xattr_entry *);
0100 
0101 static const struct xattr_handler *ext2_xattr_handler_map[] = {
0102     [EXT2_XATTR_INDEX_USER]          = &ext2_xattr_user_handler,
0103 #ifdef CONFIG_EXT2_FS_POSIX_ACL
0104     [EXT2_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
0105     [EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
0106 #endif
0107     [EXT2_XATTR_INDEX_TRUSTED]       = &ext2_xattr_trusted_handler,
0108 #ifdef CONFIG_EXT2_FS_SECURITY
0109     [EXT2_XATTR_INDEX_SECURITY]      = &ext2_xattr_security_handler,
0110 #endif
0111 };
0112 
0113 const struct xattr_handler *ext2_xattr_handlers[] = {
0114     &ext2_xattr_user_handler,
0115     &ext2_xattr_trusted_handler,
0116 #ifdef CONFIG_EXT2_FS_POSIX_ACL
0117     &posix_acl_access_xattr_handler,
0118     &posix_acl_default_xattr_handler,
0119 #endif
0120 #ifdef CONFIG_EXT2_FS_SECURITY
0121     &ext2_xattr_security_handler,
0122 #endif
0123     NULL
0124 };
0125 
0126 #define EA_BLOCK_CACHE(inode)   (EXT2_SB(inode->i_sb)->s_ea_block_cache)
0127 
0128 static inline const struct xattr_handler *
0129 ext2_xattr_handler(int name_index)
0130 {
0131     const struct xattr_handler *handler = NULL;
0132 
0133     if (name_index > 0 && name_index < ARRAY_SIZE(ext2_xattr_handler_map))
0134         handler = ext2_xattr_handler_map[name_index];
0135     return handler;
0136 }
0137 
0138 static bool
0139 ext2_xattr_header_valid(struct ext2_xattr_header *header)
0140 {
0141     if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
0142         header->h_blocks != cpu_to_le32(1))
0143         return false;
0144 
0145     return true;
0146 }
0147 
0148 static bool
0149 ext2_xattr_entry_valid(struct ext2_xattr_entry *entry,
0150                char *end, size_t end_offs)
0151 {
0152     struct ext2_xattr_entry *next;
0153     size_t size;
0154 
0155     next = EXT2_XATTR_NEXT(entry);
0156     if ((char *)next >= end)
0157         return false;
0158 
0159     if (entry->e_value_block != 0)
0160         return false;
0161 
0162     size = le32_to_cpu(entry->e_value_size);
0163     if (size > end_offs ||
0164         le16_to_cpu(entry->e_value_offs) + size > end_offs)
0165         return false;
0166 
0167     return true;
0168 }
0169 
0170 static int
0171 ext2_xattr_cmp_entry(int name_index, size_t name_len, const char *name,
0172              struct ext2_xattr_entry *entry)
0173 {
0174     int cmp;
0175 
0176     cmp = name_index - entry->e_name_index;
0177     if (!cmp)
0178         cmp = name_len - entry->e_name_len;
0179     if (!cmp)
0180         cmp = memcmp(name, entry->e_name, name_len);
0181 
0182     return cmp;
0183 }
0184 
0185 /*
0186  * ext2_xattr_get()
0187  *
0188  * Copy an extended attribute into the buffer
0189  * provided, or compute the buffer size required.
0190  * Buffer is NULL to compute the size of the buffer required.
0191  *
0192  * Returns a negative error number on failure, or the number of bytes
0193  * used / required on success.
0194  */
0195 int
0196 ext2_xattr_get(struct inode *inode, int name_index, const char *name,
0197            void *buffer, size_t buffer_size)
0198 {
0199     struct buffer_head *bh = NULL;
0200     struct ext2_xattr_entry *entry;
0201     size_t name_len, size;
0202     char *end;
0203     int error, not_found;
0204     struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
0205 
0206     ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
0207           name_index, name, buffer, (long)buffer_size);
0208 
0209     if (name == NULL)
0210         return -EINVAL;
0211     name_len = strlen(name);
0212     if (name_len > 255)
0213         return -ERANGE;
0214 
0215     down_read(&EXT2_I(inode)->xattr_sem);
0216     error = -ENODATA;
0217     if (!EXT2_I(inode)->i_file_acl)
0218         goto cleanup;
0219     ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
0220     bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
0221     error = -EIO;
0222     if (!bh)
0223         goto cleanup;
0224     ea_bdebug(bh, "b_count=%d, refcount=%d",
0225         atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
0226     end = bh->b_data + bh->b_size;
0227     if (!ext2_xattr_header_valid(HDR(bh))) {
0228 bad_block:
0229         ext2_error(inode->i_sb, "ext2_xattr_get",
0230             "inode %ld: bad block %d", inode->i_ino,
0231             EXT2_I(inode)->i_file_acl);
0232         error = -EIO;
0233         goto cleanup;
0234     }
0235 
0236     /* find named attribute */
0237     entry = FIRST_ENTRY(bh);
0238     while (!IS_LAST_ENTRY(entry)) {
0239         if (!ext2_xattr_entry_valid(entry, end,
0240             inode->i_sb->s_blocksize))
0241             goto bad_block;
0242 
0243         not_found = ext2_xattr_cmp_entry(name_index, name_len, name,
0244                          entry);
0245         if (!not_found)
0246             goto found;
0247         if (not_found < 0)
0248             break;
0249 
0250         entry = EXT2_XATTR_NEXT(entry);
0251     }
0252     if (ext2_xattr_cache_insert(ea_block_cache, bh))
0253         ea_idebug(inode, "cache insert failed");
0254     error = -ENODATA;
0255     goto cleanup;
0256 found:
0257     size = le32_to_cpu(entry->e_value_size);
0258     if (ext2_xattr_cache_insert(ea_block_cache, bh))
0259         ea_idebug(inode, "cache insert failed");
0260     if (buffer) {
0261         error = -ERANGE;
0262         if (size > buffer_size)
0263             goto cleanup;
0264         /* return value of attribute */
0265         memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
0266             size);
0267     }
0268     error = size;
0269 
0270 cleanup:
0271     brelse(bh);
0272     up_read(&EXT2_I(inode)->xattr_sem);
0273 
0274     return error;
0275 }
0276 
0277 /*
0278  * ext2_xattr_list()
0279  *
0280  * Copy a list of attribute names into the buffer
0281  * provided, or compute the buffer size required.
0282  * Buffer is NULL to compute the size of the buffer required.
0283  *
0284  * Returns a negative error number on failure, or the number of bytes
0285  * used / required on success.
0286  */
0287 static int
0288 ext2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
0289 {
0290     struct inode *inode = d_inode(dentry);
0291     struct buffer_head *bh = NULL;
0292     struct ext2_xattr_entry *entry;
0293     char *end;
0294     size_t rest = buffer_size;
0295     int error;
0296     struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
0297 
0298     ea_idebug(inode, "buffer=%p, buffer_size=%ld",
0299           buffer, (long)buffer_size);
0300 
0301     down_read(&EXT2_I(inode)->xattr_sem);
0302     error = 0;
0303     if (!EXT2_I(inode)->i_file_acl)
0304         goto cleanup;
0305     ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
0306     bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
0307     error = -EIO;
0308     if (!bh)
0309         goto cleanup;
0310     ea_bdebug(bh, "b_count=%d, refcount=%d",
0311         atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
0312     end = bh->b_data + bh->b_size;
0313     if (!ext2_xattr_header_valid(HDR(bh))) {
0314 bad_block:
0315         ext2_error(inode->i_sb, "ext2_xattr_list",
0316             "inode %ld: bad block %d", inode->i_ino,
0317             EXT2_I(inode)->i_file_acl);
0318         error = -EIO;
0319         goto cleanup;
0320     }
0321 
0322     /* check the on-disk data structure */
0323     entry = FIRST_ENTRY(bh);
0324     while (!IS_LAST_ENTRY(entry)) {
0325         if (!ext2_xattr_entry_valid(entry, end,
0326             inode->i_sb->s_blocksize))
0327             goto bad_block;
0328         entry = EXT2_XATTR_NEXT(entry);
0329     }
0330     if (ext2_xattr_cache_insert(ea_block_cache, bh))
0331         ea_idebug(inode, "cache insert failed");
0332 
0333     /* list the attribute names */
0334     for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry);
0335          entry = EXT2_XATTR_NEXT(entry)) {
0336         const struct xattr_handler *handler =
0337             ext2_xattr_handler(entry->e_name_index);
0338 
0339         if (handler && (!handler->list || handler->list(dentry))) {
0340             const char *prefix = handler->prefix ?: handler->name;
0341             size_t prefix_len = strlen(prefix);
0342             size_t size = prefix_len + entry->e_name_len + 1;
0343 
0344             if (buffer) {
0345                 if (size > rest) {
0346                     error = -ERANGE;
0347                     goto cleanup;
0348                 }
0349                 memcpy(buffer, prefix, prefix_len);
0350                 buffer += prefix_len;
0351                 memcpy(buffer, entry->e_name, entry->e_name_len);
0352                 buffer += entry->e_name_len;
0353                 *buffer++ = 0;
0354             }
0355             rest -= size;
0356         }
0357     }
0358     error = buffer_size - rest;  /* total size */
0359 
0360 cleanup:
0361     brelse(bh);
0362     up_read(&EXT2_I(inode)->xattr_sem);
0363 
0364     return error;
0365 }
0366 
0367 /*
0368  * Inode operation listxattr()
0369  *
0370  * d_inode(dentry)->i_mutex: don't care
0371  */
0372 ssize_t
0373 ext2_listxattr(struct dentry *dentry, char *buffer, size_t size)
0374 {
0375     return ext2_xattr_list(dentry, buffer, size);
0376 }
0377 
0378 /*
0379  * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is
0380  * not set, set it.
0381  */
0382 static void ext2_xattr_update_super_block(struct super_block *sb)
0383 {
0384     if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR))
0385         return;
0386 
0387     spin_lock(&EXT2_SB(sb)->s_lock);
0388     ext2_update_dynamic_rev(sb);
0389     EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR);
0390     spin_unlock(&EXT2_SB(sb)->s_lock);
0391     mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
0392 }
0393 
0394 /*
0395  * ext2_xattr_set()
0396  *
0397  * Create, replace or remove an extended attribute for this inode.  Value
0398  * is NULL to remove an existing extended attribute, and non-NULL to
0399  * either replace an existing extended attribute, or create a new extended
0400  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
0401  * specify that an extended attribute must exist and must not exist
0402  * previous to the call, respectively.
0403  *
0404  * Returns 0, or a negative error number on failure.
0405  */
0406 int
0407 ext2_xattr_set(struct inode *inode, int name_index, const char *name,
0408            const void *value, size_t value_len, int flags)
0409 {
0410     struct super_block *sb = inode->i_sb;
0411     struct buffer_head *bh = NULL;
0412     struct ext2_xattr_header *header = NULL;
0413     struct ext2_xattr_entry *here = NULL, *last = NULL;
0414     size_t name_len, free, min_offs = sb->s_blocksize;
0415     int not_found = 1, error;
0416     char *end;
0417     
0418     /*
0419      * header -- Points either into bh, or to a temporarily
0420      *           allocated buffer.
0421      * here -- The named entry found, or the place for inserting, within
0422      *         the block pointed to by header.
0423      * last -- Points right after the last named entry within the block
0424      *         pointed to by header.
0425      * min_offs -- The offset of the first value (values are aligned
0426      *             towards the end of the block).
0427      * end -- Points right after the block pointed to by header.
0428      */
0429     
0430     ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
0431           name_index, name, value, (long)value_len);
0432 
0433     if (value == NULL)
0434         value_len = 0;
0435     if (name == NULL)
0436         return -EINVAL;
0437     name_len = strlen(name);
0438     if (name_len > 255 || value_len > sb->s_blocksize)
0439         return -ERANGE;
0440     error = dquot_initialize(inode);
0441     if (error)
0442         return error;
0443     down_write(&EXT2_I(inode)->xattr_sem);
0444     if (EXT2_I(inode)->i_file_acl) {
0445         /* The inode already has an extended attribute block. */
0446         bh = sb_bread(sb, EXT2_I(inode)->i_file_acl);
0447         error = -EIO;
0448         if (!bh)
0449             goto cleanup;
0450         ea_bdebug(bh, "b_count=%d, refcount=%d",
0451             atomic_read(&(bh->b_count)),
0452             le32_to_cpu(HDR(bh)->h_refcount));
0453         header = HDR(bh);
0454         end = bh->b_data + bh->b_size;
0455         if (!ext2_xattr_header_valid(header)) {
0456 bad_block:
0457             ext2_error(sb, "ext2_xattr_set",
0458                 "inode %ld: bad block %d", inode->i_ino, 
0459                    EXT2_I(inode)->i_file_acl);
0460             error = -EIO;
0461             goto cleanup;
0462         }
0463         /*
0464          * Find the named attribute. If not found, 'here' will point
0465          * to entry where the new attribute should be inserted to
0466          * maintain sorting.
0467          */
0468         last = FIRST_ENTRY(bh);
0469         while (!IS_LAST_ENTRY(last)) {
0470             if (!ext2_xattr_entry_valid(last, end, sb->s_blocksize))
0471                 goto bad_block;
0472             if (last->e_value_size) {
0473                 size_t offs = le16_to_cpu(last->e_value_offs);
0474                 if (offs < min_offs)
0475                     min_offs = offs;
0476             }
0477             if (not_found > 0) {
0478                 not_found = ext2_xattr_cmp_entry(name_index,
0479                                  name_len,
0480                                  name, last);
0481                 if (not_found <= 0)
0482                     here = last;
0483             }
0484             last = EXT2_XATTR_NEXT(last);
0485         }
0486         if (not_found > 0)
0487             here = last;
0488 
0489         /* Check whether we have enough space left. */
0490         free = min_offs - ((char*)last - (char*)header) - sizeof(__u32);
0491     } else {
0492         /* We will use a new extended attribute block. */
0493         free = sb->s_blocksize -
0494             sizeof(struct ext2_xattr_header) - sizeof(__u32);
0495     }
0496 
0497     if (not_found) {
0498         /* Request to remove a nonexistent attribute? */
0499         error = -ENODATA;
0500         if (flags & XATTR_REPLACE)
0501             goto cleanup;
0502         error = 0;
0503         if (value == NULL)
0504             goto cleanup;
0505     } else {
0506         /* Request to create an existing attribute? */
0507         error = -EEXIST;
0508         if (flags & XATTR_CREATE)
0509             goto cleanup;
0510         free += EXT2_XATTR_SIZE(le32_to_cpu(here->e_value_size));
0511         free += EXT2_XATTR_LEN(name_len);
0512     }
0513     error = -ENOSPC;
0514     if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len))
0515         goto cleanup;
0516 
0517     /* Here we know that we can set the new attribute. */
0518 
0519     if (header) {
0520         int offset;
0521 
0522         lock_buffer(bh);
0523         if (header->h_refcount == cpu_to_le32(1)) {
0524             __u32 hash = le32_to_cpu(header->h_hash);
0525             struct mb_cache_entry *oe;
0526 
0527             oe = mb_cache_entry_delete_or_get(EA_BLOCK_CACHE(inode),
0528                     hash, bh->b_blocknr);
0529             if (!oe) {
0530                 ea_bdebug(bh, "modifying in-place");
0531                 goto update_block;
0532             }
0533             /*
0534              * Someone is trying to reuse the block, leave it alone
0535              */
0536             mb_cache_entry_put(EA_BLOCK_CACHE(inode), oe);
0537         }
0538         unlock_buffer(bh);
0539         ea_bdebug(bh, "cloning");
0540         header = kmemdup(HDR(bh), bh->b_size, GFP_KERNEL);
0541         error = -ENOMEM;
0542         if (header == NULL)
0543             goto cleanup;
0544         header->h_refcount = cpu_to_le32(1);
0545 
0546         offset = (char *)here - bh->b_data;
0547         here = ENTRY((char *)header + offset);
0548         offset = (char *)last - bh->b_data;
0549         last = ENTRY((char *)header + offset);
0550     } else {
0551         /* Allocate a buffer where we construct the new block. */
0552         header = kzalloc(sb->s_blocksize, GFP_KERNEL);
0553         error = -ENOMEM;
0554         if (header == NULL)
0555             goto cleanup;
0556         end = (char *)header + sb->s_blocksize;
0557         header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC);
0558         header->h_blocks = header->h_refcount = cpu_to_le32(1);
0559         last = here = ENTRY(header+1);
0560     }
0561 
0562 update_block:
0563     /* Iff we are modifying the block in-place, bh is locked here. */
0564 
0565     if (not_found) {
0566         /* Insert the new name. */
0567         size_t size = EXT2_XATTR_LEN(name_len);
0568         size_t rest = (char *)last - (char *)here;
0569         memmove((char *)here + size, here, rest);
0570         memset(here, 0, size);
0571         here->e_name_index = name_index;
0572         here->e_name_len = name_len;
0573         memcpy(here->e_name, name, name_len);
0574     } else {
0575         if (here->e_value_size) {
0576             char *first_val = (char *)header + min_offs;
0577             size_t offs = le16_to_cpu(here->e_value_offs);
0578             char *val = (char *)header + offs;
0579             size_t size = EXT2_XATTR_SIZE(
0580                 le32_to_cpu(here->e_value_size));
0581 
0582             if (size == EXT2_XATTR_SIZE(value_len)) {
0583                 /* The old and the new value have the same
0584                    size. Just replace. */
0585                 here->e_value_size = cpu_to_le32(value_len);
0586                 memset(val + size - EXT2_XATTR_PAD, 0,
0587                        EXT2_XATTR_PAD); /* Clear pad bytes. */
0588                 memcpy(val, value, value_len);
0589                 goto skip_replace;
0590             }
0591 
0592             /* Remove the old value. */
0593             memmove(first_val + size, first_val, val - first_val);
0594             memset(first_val, 0, size);
0595             min_offs += size;
0596 
0597             /* Adjust all value offsets. */
0598             last = ENTRY(header+1);
0599             while (!IS_LAST_ENTRY(last)) {
0600                 size_t o = le16_to_cpu(last->e_value_offs);
0601                 if (o < offs)
0602                     last->e_value_offs =
0603                         cpu_to_le16(o + size);
0604                 last = EXT2_XATTR_NEXT(last);
0605             }
0606 
0607             here->e_value_offs = 0;
0608         }
0609         if (value == NULL) {
0610             /* Remove the old name. */
0611             size_t size = EXT2_XATTR_LEN(name_len);
0612             last = ENTRY((char *)last - size);
0613             memmove(here, (char*)here + size,
0614                 (char*)last - (char*)here);
0615             memset(last, 0, size);
0616         }
0617     }
0618 
0619     if (value != NULL) {
0620         /* Insert the new value. */
0621         here->e_value_size = cpu_to_le32(value_len);
0622         if (value_len) {
0623             size_t size = EXT2_XATTR_SIZE(value_len);
0624             char *val = (char *)header + min_offs - size;
0625             here->e_value_offs =
0626                 cpu_to_le16((char *)val - (char *)header);
0627             memset(val + size - EXT2_XATTR_PAD, 0,
0628                    EXT2_XATTR_PAD); /* Clear the pad bytes. */
0629             memcpy(val, value, value_len);
0630         }
0631     }
0632 
0633 skip_replace:
0634     if (IS_LAST_ENTRY(ENTRY(header+1))) {
0635         /* This block is now empty. */
0636         if (bh && header == HDR(bh))
0637             unlock_buffer(bh);  /* we were modifying in-place. */
0638         error = ext2_xattr_set2(inode, bh, NULL);
0639     } else {
0640         ext2_xattr_rehash(header, here);
0641         if (bh && header == HDR(bh))
0642             unlock_buffer(bh);  /* we were modifying in-place. */
0643         error = ext2_xattr_set2(inode, bh, header);
0644     }
0645 
0646 cleanup:
0647     if (!(bh && header == HDR(bh)))
0648         kfree(header);
0649     brelse(bh);
0650     up_write(&EXT2_I(inode)->xattr_sem);
0651 
0652     return error;
0653 }
0654 
0655 static void ext2_xattr_release_block(struct inode *inode,
0656                      struct buffer_head *bh)
0657 {
0658     struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
0659 
0660 retry_ref:
0661     lock_buffer(bh);
0662     if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
0663         __u32 hash = le32_to_cpu(HDR(bh)->h_hash);
0664         struct mb_cache_entry *oe;
0665 
0666         /*
0667          * This must happen under buffer lock to properly
0668          * serialize with ext2_xattr_set() reusing the block.
0669          */
0670         oe = mb_cache_entry_delete_or_get(ea_block_cache, hash,
0671                           bh->b_blocknr);
0672         if (oe) {
0673             /*
0674              * Someone is trying to reuse the block. Wait
0675              * and retry.
0676              */
0677             unlock_buffer(bh);
0678             mb_cache_entry_wait_unused(oe);
0679             mb_cache_entry_put(ea_block_cache, oe);
0680             goto retry_ref;
0681         }
0682 
0683         /* Free the old block. */
0684         ea_bdebug(bh, "freeing");
0685         ext2_free_blocks(inode, bh->b_blocknr, 1);
0686         /* We let our caller release bh, so we
0687          * need to duplicate the buffer before. */
0688         get_bh(bh);
0689         bforget(bh);
0690         unlock_buffer(bh);
0691     } else {
0692         /* Decrement the refcount only. */
0693         le32_add_cpu(&HDR(bh)->h_refcount, -1);
0694         dquot_free_block(inode, 1);
0695         mark_buffer_dirty(bh);
0696         unlock_buffer(bh);
0697         ea_bdebug(bh, "refcount now=%d",
0698             le32_to_cpu(HDR(bh)->h_refcount));
0699         if (IS_SYNC(inode))
0700             sync_dirty_buffer(bh);
0701     }
0702 }
0703 
0704 /*
0705  * Second half of ext2_xattr_set(): Update the file system.
0706  */
0707 static int
0708 ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh,
0709         struct ext2_xattr_header *header)
0710 {
0711     struct super_block *sb = inode->i_sb;
0712     struct buffer_head *new_bh = NULL;
0713     int error;
0714     struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
0715 
0716     if (header) {
0717         new_bh = ext2_xattr_cache_find(inode, header);
0718         if (new_bh) {
0719             /* We found an identical block in the cache. */
0720             if (new_bh == old_bh) {
0721                 ea_bdebug(new_bh, "keeping this block");
0722             } else {
0723                 /* The old block is released after updating
0724                    the inode.  */
0725                 ea_bdebug(new_bh, "reusing block");
0726 
0727                 error = dquot_alloc_block(inode, 1);
0728                 if (error) {
0729                     unlock_buffer(new_bh);
0730                     goto cleanup;
0731                 }
0732                 le32_add_cpu(&HDR(new_bh)->h_refcount, 1);
0733                 ea_bdebug(new_bh, "refcount now=%d",
0734                     le32_to_cpu(HDR(new_bh)->h_refcount));
0735             }
0736             unlock_buffer(new_bh);
0737         } else if (old_bh && header == HDR(old_bh)) {
0738             /* Keep this block. No need to lock the block as we
0739                don't need to change the reference count. */
0740             new_bh = old_bh;
0741             get_bh(new_bh);
0742             ext2_xattr_cache_insert(ea_block_cache, new_bh);
0743         } else {
0744             /* We need to allocate a new block */
0745             ext2_fsblk_t goal = ext2_group_first_block_no(sb,
0746                         EXT2_I(inode)->i_block_group);
0747             int block = ext2_new_block(inode, goal, &error);
0748             if (error)
0749                 goto cleanup;
0750             ea_idebug(inode, "creating block %d", block);
0751 
0752             new_bh = sb_getblk(sb, block);
0753             if (unlikely(!new_bh)) {
0754                 ext2_free_blocks(inode, block, 1);
0755                 mark_inode_dirty(inode);
0756                 error = -ENOMEM;
0757                 goto cleanup;
0758             }
0759             lock_buffer(new_bh);
0760             memcpy(new_bh->b_data, header, new_bh->b_size);
0761             set_buffer_uptodate(new_bh);
0762             unlock_buffer(new_bh);
0763             ext2_xattr_cache_insert(ea_block_cache, new_bh);
0764             
0765             ext2_xattr_update_super_block(sb);
0766         }
0767         mark_buffer_dirty(new_bh);
0768         if (IS_SYNC(inode)) {
0769             sync_dirty_buffer(new_bh);
0770             error = -EIO;
0771             if (buffer_req(new_bh) && !buffer_uptodate(new_bh))
0772                 goto cleanup;
0773         }
0774     }
0775 
0776     /* Update the inode. */
0777     EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
0778     inode->i_ctime = current_time(inode);
0779     if (IS_SYNC(inode)) {
0780         error = sync_inode_metadata(inode, 1);
0781         /* In case sync failed due to ENOSPC the inode was actually
0782          * written (only some dirty data were not) so we just proceed
0783          * as if nothing happened and cleanup the unused block */
0784         if (error && error != -ENOSPC) {
0785             if (new_bh && new_bh != old_bh) {
0786                 dquot_free_block_nodirty(inode, 1);
0787                 mark_inode_dirty(inode);
0788             }
0789             goto cleanup;
0790         }
0791     } else
0792         mark_inode_dirty(inode);
0793 
0794     error = 0;
0795     if (old_bh && old_bh != new_bh) {
0796         /*
0797          * If there was an old block and we are no longer using it,
0798          * release the old block.
0799          */
0800         ext2_xattr_release_block(inode, old_bh);
0801     }
0802 
0803 cleanup:
0804     brelse(new_bh);
0805 
0806     return error;
0807 }
0808 
0809 /*
0810  * ext2_xattr_delete_inode()
0811  *
0812  * Free extended attribute resources associated with this inode. This
0813  * is called immediately before an inode is freed.
0814  */
0815 void
0816 ext2_xattr_delete_inode(struct inode *inode)
0817 {
0818     struct buffer_head *bh = NULL;
0819     struct ext2_sb_info *sbi = EXT2_SB(inode->i_sb);
0820 
0821     /*
0822      * We are the only ones holding inode reference. The xattr_sem should
0823      * better be unlocked! We could as well just not acquire xattr_sem at
0824      * all but this makes the code more futureproof. OTOH we need trylock
0825      * here to avoid false-positive warning from lockdep about reclaim
0826      * circular dependency.
0827      */
0828     if (WARN_ON_ONCE(!down_write_trylock(&EXT2_I(inode)->xattr_sem)))
0829         return;
0830     if (!EXT2_I(inode)->i_file_acl)
0831         goto cleanup;
0832 
0833     if (!ext2_data_block_valid(sbi, EXT2_I(inode)->i_file_acl, 1)) {
0834         ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
0835             "inode %ld: xattr block %d is out of data blocks range",
0836             inode->i_ino, EXT2_I(inode)->i_file_acl);
0837         goto cleanup;
0838     }
0839 
0840     bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
0841     if (!bh) {
0842         ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
0843             "inode %ld: block %d read error", inode->i_ino,
0844             EXT2_I(inode)->i_file_acl);
0845         goto cleanup;
0846     }
0847     ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
0848     if (!ext2_xattr_header_valid(HDR(bh))) {
0849         ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
0850             "inode %ld: bad block %d", inode->i_ino,
0851             EXT2_I(inode)->i_file_acl);
0852         goto cleanup;
0853     }
0854     ext2_xattr_release_block(inode, bh);
0855     EXT2_I(inode)->i_file_acl = 0;
0856 
0857 cleanup:
0858     brelse(bh);
0859     up_write(&EXT2_I(inode)->xattr_sem);
0860 }
0861 
0862 /*
0863  * ext2_xattr_cache_insert()
0864  *
0865  * Create a new entry in the extended attribute cache, and insert
0866  * it unless such an entry is already in the cache.
0867  *
0868  * Returns 0, or a negative error number on failure.
0869  */
0870 static int
0871 ext2_xattr_cache_insert(struct mb_cache *cache, struct buffer_head *bh)
0872 {
0873     __u32 hash = le32_to_cpu(HDR(bh)->h_hash);
0874     int error;
0875 
0876     error = mb_cache_entry_create(cache, GFP_NOFS, hash, bh->b_blocknr,
0877                       true);
0878     if (error) {
0879         if (error == -EBUSY) {
0880             ea_bdebug(bh, "already in cache");
0881             error = 0;
0882         }
0883     } else
0884         ea_bdebug(bh, "inserting [%x]", (int)hash);
0885     return error;
0886 }
0887 
0888 /*
0889  * ext2_xattr_cmp()
0890  *
0891  * Compare two extended attribute blocks for equality.
0892  *
0893  * Returns 0 if the blocks are equal, 1 if they differ, and
0894  * a negative error number on errors.
0895  */
0896 static int
0897 ext2_xattr_cmp(struct ext2_xattr_header *header1,
0898            struct ext2_xattr_header *header2)
0899 {
0900     struct ext2_xattr_entry *entry1, *entry2;
0901 
0902     entry1 = ENTRY(header1+1);
0903     entry2 = ENTRY(header2+1);
0904     while (!IS_LAST_ENTRY(entry1)) {
0905         if (IS_LAST_ENTRY(entry2))
0906             return 1;
0907         if (entry1->e_hash != entry2->e_hash ||
0908             entry1->e_name_index != entry2->e_name_index ||
0909             entry1->e_name_len != entry2->e_name_len ||
0910             entry1->e_value_size != entry2->e_value_size ||
0911             memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
0912             return 1;
0913         if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
0914             return -EIO;
0915         if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
0916                (char *)header2 + le16_to_cpu(entry2->e_value_offs),
0917                le32_to_cpu(entry1->e_value_size)))
0918             return 1;
0919 
0920         entry1 = EXT2_XATTR_NEXT(entry1);
0921         entry2 = EXT2_XATTR_NEXT(entry2);
0922     }
0923     if (!IS_LAST_ENTRY(entry2))
0924         return 1;
0925     return 0;
0926 }
0927 
0928 /*
0929  * ext2_xattr_cache_find()
0930  *
0931  * Find an identical extended attribute block.
0932  *
0933  * Returns a locked buffer head to the block found, or NULL if such
0934  * a block was not found or an error occurred.
0935  */
0936 static struct buffer_head *
0937 ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
0938 {
0939     __u32 hash = le32_to_cpu(header->h_hash);
0940     struct mb_cache_entry *ce;
0941     struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
0942 
0943     if (!header->h_hash)
0944         return NULL;  /* never share */
0945     ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
0946 
0947     ce = mb_cache_entry_find_first(ea_block_cache, hash);
0948     while (ce) {
0949         struct buffer_head *bh;
0950 
0951         bh = sb_bread(inode->i_sb, ce->e_value);
0952         if (!bh) {
0953             ext2_error(inode->i_sb, "ext2_xattr_cache_find",
0954                 "inode %ld: block %ld read error",
0955                 inode->i_ino, (unsigned long) ce->e_value);
0956         } else {
0957             lock_buffer(bh);
0958             if (le32_to_cpu(HDR(bh)->h_refcount) >
0959                 EXT2_XATTR_REFCOUNT_MAX) {
0960                 ea_idebug(inode, "block %ld refcount %d>%d",
0961                       (unsigned long) ce->e_value,
0962                       le32_to_cpu(HDR(bh)->h_refcount),
0963                       EXT2_XATTR_REFCOUNT_MAX);
0964             } else if (!ext2_xattr_cmp(header, HDR(bh))) {
0965                 ea_bdebug(bh, "b_count=%d",
0966                       atomic_read(&(bh->b_count)));
0967                 mb_cache_entry_touch(ea_block_cache, ce);
0968                 mb_cache_entry_put(ea_block_cache, ce);
0969                 return bh;
0970             }
0971             unlock_buffer(bh);
0972             brelse(bh);
0973         }
0974         ce = mb_cache_entry_find_next(ea_block_cache, ce);
0975     }
0976     return NULL;
0977 }
0978 
0979 #define NAME_HASH_SHIFT 5
0980 #define VALUE_HASH_SHIFT 16
0981 
0982 /*
0983  * ext2_xattr_hash_entry()
0984  *
0985  * Compute the hash of an extended attribute.
0986  */
0987 static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
0988                      struct ext2_xattr_entry *entry)
0989 {
0990     __u32 hash = 0;
0991     char *name = entry->e_name;
0992     int n;
0993 
0994     for (n=0; n < entry->e_name_len; n++) {
0995         hash = (hash << NAME_HASH_SHIFT) ^
0996                (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
0997                *name++;
0998     }
0999 
1000     if (entry->e_value_block == 0 && entry->e_value_size != 0) {
1001         __le32 *value = (__le32 *)((char *)header +
1002             le16_to_cpu(entry->e_value_offs));
1003         for (n = (le32_to_cpu(entry->e_value_size) +
1004              EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
1005             hash = (hash << VALUE_HASH_SHIFT) ^
1006                    (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1007                    le32_to_cpu(*value++);
1008         }
1009     }
1010     entry->e_hash = cpu_to_le32(hash);
1011 }
1012 
1013 #undef NAME_HASH_SHIFT
1014 #undef VALUE_HASH_SHIFT
1015 
1016 #define BLOCK_HASH_SHIFT 16
1017 
1018 /*
1019  * ext2_xattr_rehash()
1020  *
1021  * Re-compute the extended attribute hash value after an entry has changed.
1022  */
1023 static void ext2_xattr_rehash(struct ext2_xattr_header *header,
1024                   struct ext2_xattr_entry *entry)
1025 {
1026     struct ext2_xattr_entry *here;
1027     __u32 hash = 0;
1028     
1029     ext2_xattr_hash_entry(header, entry);
1030     here = ENTRY(header+1);
1031     while (!IS_LAST_ENTRY(here)) {
1032         if (!here->e_hash) {
1033             /* Block is not shared if an entry's hash value == 0 */
1034             hash = 0;
1035             break;
1036         }
1037         hash = (hash << BLOCK_HASH_SHIFT) ^
1038                (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1039                le32_to_cpu(here->e_hash);
1040         here = EXT2_XATTR_NEXT(here);
1041     }
1042     header->h_hash = cpu_to_le32(hash);
1043 }
1044 
1045 #undef BLOCK_HASH_SHIFT
1046 
1047 #define HASH_BUCKET_BITS 10
1048 
1049 struct mb_cache *ext2_xattr_create_cache(void)
1050 {
1051     return mb_cache_create(HASH_BUCKET_BITS);
1052 }
1053 
1054 void ext2_xattr_destroy_cache(struct mb_cache *cache)
1055 {
1056     if (cache)
1057         mb_cache_destroy(cache);
1058 }