1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (c) 2014 Christoph Hellwig. 4 */ 5 #include "xfs.h" 6 #include "xfs_shared.h" 7 #include "xfs_format.h" 8 #include "xfs_log_format.h" 9 #include "xfs_trans_resv.h" 10 #include "xfs_mount.h" 11 #include "xfs_inode.h" 12 #include "xfs_trans.h" 13 #include "xfs_bmap.h" 14 #include "xfs_iomap.h" 15 #include "xfs_pnfs.h" 16 17 /* 18 * Ensure that we do not have any outstanding pNFS layouts that can be used by 19 * clients to directly read from or write to this inode. This must be called 20 * before every operation that can remove blocks from the extent map. 21 * Additionally we call it during the write operation, where aren't concerned 22 * about exposing unallocated blocks but just want to provide basic 23 * synchronization between a local writer and pNFS clients. mmap writes would 24 * also benefit from this sort of synchronization, but due to the tricky locking 25 * rules in the page fault path we don't bother. 26 */ 27 int 28 xfs_break_leased_layouts( 29 struct inode *inode, 30 uint *iolock, 31 bool *did_unlock) 32 { 33 struct xfs_inode *ip = XFS_I(inode); 34 int error; 35 36 while ((error = break_layout(inode, false)) == -EWOULDBLOCK) { 37 xfs_iunlock(ip, *iolock); 38 *did_unlock = true; 39 error = break_layout(inode, true); 40 *iolock &= ~XFS_IOLOCK_SHARED; 41 *iolock |= XFS_IOLOCK_EXCL; 42 xfs_ilock(ip, *iolock); 43 } 44 45 return error; 46 } 47 48 /* 49 * Get a unique ID including its location so that the client can identify 50 * the exported device. 51 */ 52 int 53 xfs_fs_get_uuid( 54 struct super_block *sb, 55 u8 *buf, 56 u32 *len, 57 u64 *offset) 58 { 59 struct xfs_mount *mp = XFS_M(sb); 60 61 xfs_notice_once(mp, 62 "Using experimental pNFS feature, use at your own risk!"); 63 64 if (*len < sizeof(uuid_t)) 65 return -EINVAL; 66 67 memcpy(buf, &mp->m_sb.sb_uuid, sizeof(uuid_t)); 68 *len = sizeof(uuid_t); 69 *offset = offsetof(struct xfs_dsb, sb_uuid); 70 return 0; 71 } 72 73 /* 74 * We cannot use file based VFS helpers such as file_modified() to update 75 * inode state as we modify the data/metadata in the inode here. Hence we have 76 * to open code the timestamp updates and SUID/SGID stripping. We also need 77 * to set the inode prealloc flag to ensure that the extents we allocate are not 78 * removed if the inode is reclaimed from memory before xfs_fs_block_commit() 79 * is from the client to indicate that data has been written and the file size 80 * can be extended. 81 */ 82 static int 83 xfs_fs_map_update_inode( 84 struct xfs_inode *ip) 85 { 86 struct xfs_trans *tp; 87 int error; 88 89 error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_writeid, 90 0, 0, 0, &tp); 91 if (error) 92 return error; 93 94 xfs_ilock(ip, XFS_ILOCK_EXCL); 95 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 96 97 VFS_I(ip)->i_mode &= ~S_ISUID; 98 if (VFS_I(ip)->i_mode & S_IXGRP) 99 VFS_I(ip)->i_mode &= ~S_ISGID; 100 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 101 ip->i_diflags |= XFS_DIFLAG_PREALLOC; 102 103 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 104 return xfs_trans_commit(tp); 105 } 106 107 /* 108 * Get a layout for the pNFS client. 109 */ 110 int 111 xfs_fs_map_blocks( 112 struct inode *inode, 113 loff_t offset, 114 u64 length, 115 struct iomap *iomap, 116 bool write, 117 u32 *device_generation) 118 { 119 struct xfs_inode *ip = XFS_I(inode); 120 struct xfs_mount *mp = ip->i_mount; 121 struct xfs_bmbt_irec imap; 122 xfs_fileoff_t offset_fsb, end_fsb; 123 loff_t limit; 124 int bmapi_flags = XFS_BMAPI_ENTIRE; 125 int nimaps = 1; 126 uint lock_flags; 127 int error = 0; 128 129 if (xfs_is_shutdown(mp)) 130 return -EIO; 131 132 /* 133 * We can't export inodes residing on the realtime device. The realtime 134 * device doesn't have a UUID to identify it, so the client has no way 135 * to find it. 136 */ 137 if (XFS_IS_REALTIME_INODE(ip)) 138 return -ENXIO; 139 140 /* 141 * The pNFS block layout spec actually supports reflink like 142 * functionality, but the Linux pNFS server doesn't implement it yet. 143 */ 144 if (xfs_is_reflink_inode(ip)) 145 return -ENXIO; 146 147 /* 148 * Lock out any other I/O before we flush and invalidate the pagecache, 149 * and then hand out a layout to the remote system. This is very 150 * similar to direct I/O, except that the synchronization is much more 151 * complicated. See the comment near xfs_break_leased_layouts 152 * for a detailed explanation. 153 */ 154 xfs_ilock(ip, XFS_IOLOCK_EXCL); 155 156 error = -EINVAL; 157 limit = mp->m_super->s_maxbytes; 158 if (!write) 159 limit = max(limit, round_up(i_size_read(inode), 160 inode->i_sb->s_blocksize)); 161 if (offset > limit) 162 goto out_unlock; 163 if (offset > limit - length) 164 length = limit - offset; 165 166 error = filemap_write_and_wait(inode->i_mapping); 167 if (error) 168 goto out_unlock; 169 error = invalidate_inode_pages2(inode->i_mapping); 170 if (WARN_ON_ONCE(error)) 171 goto out_unlock; 172 173 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + length); 174 offset_fsb = XFS_B_TO_FSBT(mp, offset); 175 176 lock_flags = xfs_ilock_data_map_shared(ip); 177 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, 178 &imap, &nimaps, bmapi_flags); 179 180 ASSERT(!nimaps || imap.br_startblock != DELAYSTARTBLOCK); 181 182 if (!error && write && 183 (!nimaps || imap.br_startblock == HOLESTARTBLOCK)) { 184 if (offset + length > XFS_ISIZE(ip)) 185 end_fsb = xfs_iomap_eof_align_last_fsb(ip, end_fsb); 186 else if (nimaps && imap.br_startblock == HOLESTARTBLOCK) 187 end_fsb = min(end_fsb, imap.br_startoff + 188 imap.br_blockcount); 189 xfs_iunlock(ip, lock_flags); 190 191 error = xfs_iomap_write_direct(ip, offset_fsb, 192 end_fsb - offset_fsb, 0, &imap); 193 if (error) 194 goto out_unlock; 195 196 /* 197 * Ensure the next transaction is committed synchronously so 198 * that the blocks allocated and handed out to the client are 199 * guaranteed to be present even after a server crash. 200 */ 201 error = xfs_fs_map_update_inode(ip); 202 if (!error) 203 error = xfs_log_force_inode(ip); 204 if (error) 205 goto out_unlock; 206 207 } else { 208 xfs_iunlock(ip, lock_flags); 209 } 210 xfs_iunlock(ip, XFS_IOLOCK_EXCL); 211 212 error = xfs_bmbt_to_iomap(ip, iomap, &imap, 0, 0); 213 *device_generation = mp->m_generation; 214 return error; 215 out_unlock: 216 xfs_iunlock(ip, XFS_IOLOCK_EXCL); 217 return error; 218 } 219 220 /* 221 * Ensure the size update falls into a valid allocated block. 222 */ 223 static int 224 xfs_pnfs_validate_isize( 225 struct xfs_inode *ip, 226 xfs_off_t isize) 227 { 228 struct xfs_bmbt_irec imap; 229 int nimaps = 1; 230 int error = 0; 231 232 xfs_ilock(ip, XFS_ILOCK_SHARED); 233 error = xfs_bmapi_read(ip, XFS_B_TO_FSBT(ip->i_mount, isize - 1), 1, 234 &imap, &nimaps, 0); 235 xfs_iunlock(ip, XFS_ILOCK_SHARED); 236 if (error) 237 return error; 238 239 if (imap.br_startblock == HOLESTARTBLOCK || 240 imap.br_startblock == DELAYSTARTBLOCK || 241 imap.br_state == XFS_EXT_UNWRITTEN) 242 return -EIO; 243 return 0; 244 } 245 246 /* 247 * Make sure the blocks described by maps are stable on disk. This includes 248 * converting any unwritten extents, flushing the disk cache and updating the 249 * time stamps. 250 * 251 * Note that we rely on the caller to always send us a timestamp update so that 252 * we always commit a transaction here. If that stops being true we will have 253 * to manually flush the cache here similar to what the fsync code path does 254 * for datasyncs on files that have no dirty metadata. 255 */ 256 int 257 xfs_fs_commit_blocks( 258 struct inode *inode, 259 struct iomap *maps, 260 int nr_maps, 261 struct iattr *iattr) 262 { 263 struct xfs_inode *ip = XFS_I(inode); 264 struct xfs_mount *mp = ip->i_mount; 265 struct xfs_trans *tp; 266 bool update_isize = false; 267 int error, i; 268 loff_t size; 269 270 ASSERT(iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME)); 271 272 xfs_ilock(ip, XFS_IOLOCK_EXCL); 273 274 size = i_size_read(inode); 275 if ((iattr->ia_valid & ATTR_SIZE) && iattr->ia_size > size) { 276 update_isize = true; 277 size = iattr->ia_size; 278 } 279 280 for (i = 0; i < nr_maps; i++) { 281 u64 start, length, end; 282 283 start = maps[i].offset; 284 if (start > size) 285 continue; 286 287 end = start + maps[i].length; 288 if (end > size) 289 end = size; 290 291 length = end - start; 292 if (!length) 293 continue; 294 295 /* 296 * Make sure reads through the pagecache see the new data. 297 */ 298 error = invalidate_inode_pages2_range(inode->i_mapping, 299 start >> PAGE_SHIFT, 300 (end - 1) >> PAGE_SHIFT); 301 WARN_ON_ONCE(error); 302 303 error = xfs_iomap_write_unwritten(ip, start, length, false); 304 if (error) 305 goto out_drop_iolock; 306 } 307 308 if (update_isize) { 309 error = xfs_pnfs_validate_isize(ip, size); 310 if (error) 311 goto out_drop_iolock; 312 } 313 314 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp); 315 if (error) 316 goto out_drop_iolock; 317 318 xfs_ilock(ip, XFS_ILOCK_EXCL); 319 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 320 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 321 322 xfs_setattr_time(ip, iattr); 323 if (update_isize) { 324 i_size_write(inode, iattr->ia_size); 325 ip->i_disk_size = iattr->ia_size; 326 } 327 328 xfs_trans_set_sync(tp); 329 error = xfs_trans_commit(tp); 330 331 out_drop_iolock: 332 xfs_iunlock(ip, XFS_IOLOCK_EXCL); 333 return error; 334 } 335