1 // SPDX-License-Identifier: GPL-2.0-or-later 1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 2 /* 3 * Copyright (C) 2002, 2004 Oracle. All right 3 * Copyright (C) 2002, 2004 Oracle. All rights reserved. 4 */ 4 */ 5 5 6 #include <linux/fs.h> 6 #include <linux/fs.h> 7 #include <linux/slab.h> 7 #include <linux/slab.h> 8 #include <linux/highmem.h> 8 #include <linux/highmem.h> 9 #include <linux/pagemap.h> 9 #include <linux/pagemap.h> 10 #include <asm/byteorder.h> 10 #include <asm/byteorder.h> 11 #include <linux/swap.h> 11 #include <linux/swap.h> 12 #include <linux/mpage.h> 12 #include <linux/mpage.h> 13 #include <linux/quotaops.h> 13 #include <linux/quotaops.h> 14 #include <linux/blkdev.h> 14 #include <linux/blkdev.h> 15 #include <linux/uio.h> 15 #include <linux/uio.h> 16 #include <linux/mm.h> 16 #include <linux/mm.h> 17 17 18 #include <cluster/masklog.h> 18 #include <cluster/masklog.h> 19 19 20 #include "ocfs2.h" 20 #include "ocfs2.h" 21 21 22 #include "alloc.h" 22 #include "alloc.h" 23 #include "aops.h" 23 #include "aops.h" 24 #include "dlmglue.h" 24 #include "dlmglue.h" 25 #include "extent_map.h" 25 #include "extent_map.h" 26 #include "file.h" 26 #include "file.h" 27 #include "inode.h" 27 #include "inode.h" 28 #include "journal.h" 28 #include "journal.h" 29 #include "suballoc.h" 29 #include "suballoc.h" 30 #include "super.h" 30 #include "super.h" 31 #include "symlink.h" 31 #include "symlink.h" 32 #include "refcounttree.h" 32 #include "refcounttree.h" 33 #include "ocfs2_trace.h" 33 #include "ocfs2_trace.h" 34 34 35 #include "buffer_head_io.h" 35 #include "buffer_head_io.h" 36 #include "dir.h" 36 #include "dir.h" 37 #include "namei.h" 37 #include "namei.h" 38 #include "sysfile.h" 38 #include "sysfile.h" 39 39 40 static int ocfs2_symlink_get_block(struct inod 40 static int ocfs2_symlink_get_block(struct inode *inode, sector_t iblock, 41 struct buff 41 struct buffer_head *bh_result, int create) 42 { 42 { 43 int err = -EIO; 43 int err = -EIO; 44 int status; 44 int status; 45 struct ocfs2_dinode *fe = NULL; 45 struct ocfs2_dinode *fe = NULL; 46 struct buffer_head *bh = NULL; 46 struct buffer_head *bh = NULL; 47 struct buffer_head *buffer_cache_bh = 47 struct buffer_head *buffer_cache_bh = NULL; 48 struct ocfs2_super *osb = OCFS2_SB(ino 48 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 49 void *kaddr; 49 void *kaddr; 50 50 51 trace_ocfs2_symlink_get_block( 51 trace_ocfs2_symlink_get_block( 52 (unsigned long long)OC 52 (unsigned long long)OCFS2_I(inode)->ip_blkno, 53 (unsigned long long)ib 53 (unsigned long long)iblock, bh_result, create); 54 54 55 BUG_ON(ocfs2_inode_is_fast_symlink(ino 55 BUG_ON(ocfs2_inode_is_fast_symlink(inode)); 56 56 57 if ((iblock << inode->i_sb->s_blocksiz 57 if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) { 58 mlog(ML_ERROR, "block offset > 58 mlog(ML_ERROR, "block offset > PATH_MAX: %llu", 59 (unsigned long long)ibloc 59 (unsigned long long)iblock); 60 goto bail; 60 goto bail; 61 } 61 } 62 62 63 status = ocfs2_read_inode_block(inode, 63 status = ocfs2_read_inode_block(inode, &bh); 64 if (status < 0) { 64 if (status < 0) { 65 mlog_errno(status); 65 mlog_errno(status); 66 goto bail; 66 goto bail; 67 } 67 } 68 fe = (struct ocfs2_dinode *) bh->b_dat 68 fe = (struct ocfs2_dinode *) bh->b_data; 69 69 70 if ((u64)iblock >= ocfs2_clusters_to_b 70 if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb, 71 71 le32_to_cpu(fe->i_clusters))) { 72 err = -ENOMEM; 72 err = -ENOMEM; 73 mlog(ML_ERROR, "block offset i 73 mlog(ML_ERROR, "block offset is outside the allocated size: " 74 "%llu\n", (unsigned long 74 "%llu\n", (unsigned long long)iblock); 75 goto bail; 75 goto bail; 76 } 76 } 77 77 78 /* We don't use the page cache to crea 78 /* We don't use the page cache to create symlink data, so if 79 * need be, copy it over from the buff 79 * need be, copy it over from the buffer cache. */ 80 if (!buffer_uptodate(bh_result) && ocf 80 if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) { 81 u64 blkno = le64_to_cpu(fe->id 81 u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + 82 iblock; 82 iblock; 83 buffer_cache_bh = sb_getblk(os 83 buffer_cache_bh = sb_getblk(osb->sb, blkno); 84 if (!buffer_cache_bh) { 84 if (!buffer_cache_bh) { 85 err = -ENOMEM; 85 err = -ENOMEM; 86 mlog(ML_ERROR, "couldn 86 mlog(ML_ERROR, "couldn't getblock for symlink!\n"); 87 goto bail; 87 goto bail; 88 } 88 } 89 89 90 /* we haven't locked out trans 90 /* we haven't locked out transactions, so a commit 91 * could've happened. Since we 91 * could've happened. Since we've got a reference on 92 * the bh, even if it commits 92 * the bh, even if it commits while we're doing the 93 * copy, the data is still goo 93 * copy, the data is still good. */ 94 if (buffer_jbd(buffer_cache_bh 94 if (buffer_jbd(buffer_cache_bh) 95 && ocfs2_inode_is_new(inod 95 && ocfs2_inode_is_new(inode)) { 96 kaddr = kmap_atomic(bh 96 kaddr = kmap_atomic(bh_result->b_page); 97 if (!kaddr) { 97 if (!kaddr) { 98 mlog(ML_ERROR, 98 mlog(ML_ERROR, "couldn't kmap!\n"); 99 goto bail; 99 goto bail; 100 } 100 } 101 memcpy(kaddr + (bh_res 101 memcpy(kaddr + (bh_result->b_size * iblock), 102 buffer_cache_bh 102 buffer_cache_bh->b_data, 103 bh_result->b_si 103 bh_result->b_size); 104 kunmap_atomic(kaddr); 104 kunmap_atomic(kaddr); 105 set_buffer_uptodate(bh 105 set_buffer_uptodate(bh_result); 106 } 106 } 107 brelse(buffer_cache_bh); 107 brelse(buffer_cache_bh); 108 } 108 } 109 109 110 map_bh(bh_result, inode->i_sb, 110 map_bh(bh_result, inode->i_sb, 111 le64_to_cpu(fe->id2.i_list.l_re 111 le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock); 112 112 113 err = 0; 113 err = 0; 114 114 115 bail: 115 bail: 116 brelse(bh); 116 brelse(bh); 117 117 118 return err; 118 return err; 119 } 119 } 120 120 121 static int ocfs2_lock_get_block(struct inode * 121 static int ocfs2_lock_get_block(struct inode *inode, sector_t iblock, 122 struct buffer_head *bh_res 122 struct buffer_head *bh_result, int create) 123 { 123 { 124 int ret = 0; 124 int ret = 0; 125 struct ocfs2_inode_info *oi = OCFS2_I( 125 struct ocfs2_inode_info *oi = OCFS2_I(inode); 126 126 127 down_read(&oi->ip_alloc_sem); 127 down_read(&oi->ip_alloc_sem); 128 ret = ocfs2_get_block(inode, iblock, b 128 ret = ocfs2_get_block(inode, iblock, bh_result, create); 129 up_read(&oi->ip_alloc_sem); 129 up_read(&oi->ip_alloc_sem); 130 130 131 return ret; 131 return ret; 132 } 132 } 133 133 134 int ocfs2_get_block(struct inode *inode, secto 134 int ocfs2_get_block(struct inode *inode, sector_t iblock, 135 struct buffer_head *bh_res 135 struct buffer_head *bh_result, int create) 136 { 136 { 137 int err = 0; 137 int err = 0; 138 unsigned int ext_flags; 138 unsigned int ext_flags; 139 u64 max_blocks = bh_result->b_size >> 139 u64 max_blocks = bh_result->b_size >> inode->i_blkbits; 140 u64 p_blkno, count, past_eof; 140 u64 p_blkno, count, past_eof; 141 struct ocfs2_super *osb = OCFS2_SB(ino 141 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 142 142 143 trace_ocfs2_get_block((unsigned long l 143 trace_ocfs2_get_block((unsigned long long)OCFS2_I(inode)->ip_blkno, 144 (unsigned long l 144 (unsigned long long)iblock, bh_result, create); 145 145 146 if (OCFS2_I(inode)->ip_flags & OCFS2_I 146 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) 147 mlog(ML_NOTICE, "get_block on 147 mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n", 148 inode, inode->i_ino); 148 inode, inode->i_ino); 149 149 150 if (S_ISLNK(inode->i_mode)) { 150 if (S_ISLNK(inode->i_mode)) { 151 /* this always does I/O for so 151 /* this always does I/O for some reason. */ 152 err = ocfs2_symlink_get_block( 152 err = ocfs2_symlink_get_block(inode, iblock, bh_result, create); 153 goto bail; 153 goto bail; 154 } 154 } 155 155 156 err = ocfs2_extent_map_get_blocks(inod 156 err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count, 157 &ext 157 &ext_flags); 158 if (err) { 158 if (err) { 159 mlog(ML_ERROR, "Error %d from 159 mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, " 160 "%llu, NULL)\n", err, ino 160 "%llu, NULL)\n", err, inode, (unsigned long long)iblock, 161 (unsigned long long)p_blk 161 (unsigned long long)p_blkno); 162 goto bail; 162 goto bail; 163 } 163 } 164 164 165 if (max_blocks < count) 165 if (max_blocks < count) 166 count = max_blocks; 166 count = max_blocks; 167 167 168 /* 168 /* 169 * ocfs2 never allocates in this funct 169 * ocfs2 never allocates in this function - the only time we 170 * need to use BH_New is when we're ex 170 * need to use BH_New is when we're extending i_size on a file 171 * system which doesn't support holes, 171 * system which doesn't support holes, in which case BH_New 172 * allows __block_write_begin() to zer 172 * allows __block_write_begin() to zero. 173 * 173 * 174 * If we see this on a sparse file sys 174 * If we see this on a sparse file system, then a truncate has 175 * raced us and removed the cluster. I 175 * raced us and removed the cluster. In this case, we clear 176 * the buffers dirty and uptodate bits 176 * the buffers dirty and uptodate bits and let the buffer code 177 * ignore it as a hole. 177 * ignore it as a hole. 178 */ 178 */ 179 if (create && p_blkno == 0 && ocfs2_sp 179 if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) { 180 clear_buffer_dirty(bh_result); 180 clear_buffer_dirty(bh_result); 181 clear_buffer_uptodate(bh_resul 181 clear_buffer_uptodate(bh_result); 182 goto bail; 182 goto bail; 183 } 183 } 184 184 185 /* Treat the unwritten extent as a hol 185 /* Treat the unwritten extent as a hole for zeroing purposes. */ 186 if (p_blkno && !(ext_flags & OCFS2_EXT 186 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN)) 187 map_bh(bh_result, inode->i_sb, 187 map_bh(bh_result, inode->i_sb, p_blkno); 188 188 189 bh_result->b_size = count << inode->i_ 189 bh_result->b_size = count << inode->i_blkbits; 190 190 191 if (!ocfs2_sparse_alloc(osb)) { 191 if (!ocfs2_sparse_alloc(osb)) { 192 if (p_blkno == 0) { 192 if (p_blkno == 0) { 193 err = -EIO; 193 err = -EIO; 194 mlog(ML_ERROR, 194 mlog(ML_ERROR, 195 "iblock = %llu p_ 195 "iblock = %llu p_blkno = %llu blkno=(%llu)\n", 196 (unsigned long lo 196 (unsigned long long)iblock, 197 (unsigned long lo 197 (unsigned long long)p_blkno, 198 (unsigned long lo 198 (unsigned long long)OCFS2_I(inode)->ip_blkno); 199 mlog(ML_ERROR, "Size % 199 mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters); 200 dump_stack(); 200 dump_stack(); 201 goto bail; 201 goto bail; 202 } 202 } 203 } 203 } 204 204 205 past_eof = ocfs2_blocks_for_bytes(inod 205 past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); 206 206 207 trace_ocfs2_get_block_end((unsigned lo 207 trace_ocfs2_get_block_end((unsigned long long)OCFS2_I(inode)->ip_blkno, 208 (unsigned lo 208 (unsigned long long)past_eof); 209 if (create && (iblock >= past_eof)) 209 if (create && (iblock >= past_eof)) 210 set_buffer_new(bh_result); 210 set_buffer_new(bh_result); 211 211 212 bail: 212 bail: 213 if (err < 0) 213 if (err < 0) 214 err = -EIO; 214 err = -EIO; 215 215 216 return err; 216 return err; 217 } 217 } 218 218 219 int ocfs2_read_inline_data(struct inode *inode 219 int ocfs2_read_inline_data(struct inode *inode, struct page *page, 220 struct buffer_head 220 struct buffer_head *di_bh) 221 { 221 { 222 void *kaddr; 222 void *kaddr; 223 loff_t size; 223 loff_t size; 224 struct ocfs2_dinode *di = (struct ocfs 224 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; 225 225 226 if (!(le16_to_cpu(di->i_dyn_features) 226 if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) { 227 ocfs2_error(inode->i_sb, "Inod 227 ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag\n", 228 (unsigned long lon 228 (unsigned long long)OCFS2_I(inode)->ip_blkno); 229 return -EROFS; 229 return -EROFS; 230 } 230 } 231 231 232 size = i_size_read(inode); 232 size = i_size_read(inode); 233 233 234 if (size > PAGE_SIZE || 234 if (size > PAGE_SIZE || 235 size > ocfs2_max_inline_data_with_ 235 size > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) { 236 ocfs2_error(inode->i_sb, 236 ocfs2_error(inode->i_sb, 237 "Inode %llu has wi 237 "Inode %llu has with inline data has bad size: %Lu\n", 238 (unsigned long lon 238 (unsigned long long)OCFS2_I(inode)->ip_blkno, 239 (unsigned long lon 239 (unsigned long long)size); 240 return -EROFS; 240 return -EROFS; 241 } 241 } 242 242 243 kaddr = kmap_atomic(page); 243 kaddr = kmap_atomic(page); 244 if (size) 244 if (size) 245 memcpy(kaddr, di->id2.i_data.i 245 memcpy(kaddr, di->id2.i_data.id_data, size); 246 /* Clear the remaining part of the pag 246 /* Clear the remaining part of the page */ 247 memset(kaddr + size, 0, PAGE_SIZE - si 247 memset(kaddr + size, 0, PAGE_SIZE - size); 248 flush_dcache_page(page); 248 flush_dcache_page(page); 249 kunmap_atomic(kaddr); 249 kunmap_atomic(kaddr); 250 250 251 SetPageUptodate(page); 251 SetPageUptodate(page); 252 252 253 return 0; 253 return 0; 254 } 254 } 255 255 256 static int ocfs2_readpage_inline(struct inode 256 static int ocfs2_readpage_inline(struct inode *inode, struct page *page) 257 { 257 { 258 int ret; 258 int ret; 259 struct buffer_head *di_bh = NULL; 259 struct buffer_head *di_bh = NULL; 260 260 261 BUG_ON(!PageLocked(page)); 261 BUG_ON(!PageLocked(page)); 262 BUG_ON(!(OCFS2_I(inode)->ip_dyn_featur 262 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)); 263 263 264 ret = ocfs2_read_inode_block(inode, &d 264 ret = ocfs2_read_inode_block(inode, &di_bh); 265 if (ret) { 265 if (ret) { 266 mlog_errno(ret); 266 mlog_errno(ret); 267 goto out; 267 goto out; 268 } 268 } 269 269 270 ret = ocfs2_read_inline_data(inode, pa 270 ret = ocfs2_read_inline_data(inode, page, di_bh); 271 out: 271 out: 272 unlock_page(page); 272 unlock_page(page); 273 273 274 brelse(di_bh); 274 brelse(di_bh); 275 return ret; 275 return ret; 276 } 276 } 277 277 278 static int ocfs2_read_folio(struct file *file, 278 static int ocfs2_read_folio(struct file *file, struct folio *folio) 279 { 279 { 280 struct inode *inode = folio->mapping-> 280 struct inode *inode = folio->mapping->host; 281 struct ocfs2_inode_info *oi = OCFS2_I( 281 struct ocfs2_inode_info *oi = OCFS2_I(inode); 282 loff_t start = folio_pos(folio); 282 loff_t start = folio_pos(folio); 283 int ret, unlock = 1; 283 int ret, unlock = 1; 284 284 285 trace_ocfs2_readpage((unsigned long lo 285 trace_ocfs2_readpage((unsigned long long)oi->ip_blkno, folio->index); 286 286 287 ret = ocfs2_inode_lock_with_page(inode 287 ret = ocfs2_inode_lock_with_page(inode, NULL, 0, &folio->page); 288 if (ret != 0) { 288 if (ret != 0) { 289 if (ret == AOP_TRUNCATED_PAGE) 289 if (ret == AOP_TRUNCATED_PAGE) 290 unlock = 0; 290 unlock = 0; 291 mlog_errno(ret); 291 mlog_errno(ret); 292 goto out; 292 goto out; 293 } 293 } 294 294 295 if (down_read_trylock(&oi->ip_alloc_se 295 if (down_read_trylock(&oi->ip_alloc_sem) == 0) { 296 /* 296 /* 297 * Unlock the folio and cycle 297 * Unlock the folio and cycle ip_alloc_sem so that we don't 298 * busyloop waiting for ip_all 298 * busyloop waiting for ip_alloc_sem to unlock 299 */ 299 */ 300 ret = AOP_TRUNCATED_PAGE; 300 ret = AOP_TRUNCATED_PAGE; 301 folio_unlock(folio); 301 folio_unlock(folio); 302 unlock = 0; 302 unlock = 0; 303 down_read(&oi->ip_alloc_sem); 303 down_read(&oi->ip_alloc_sem); 304 up_read(&oi->ip_alloc_sem); 304 up_read(&oi->ip_alloc_sem); 305 goto out_inode_unlock; 305 goto out_inode_unlock; 306 } 306 } 307 307 308 /* 308 /* 309 * i_size might have just been updated 309 * i_size might have just been updated as we grabed the meta lock. We 310 * might now be discovering a truncate 310 * might now be discovering a truncate that hit on another node. 311 * block_read_full_folio->get_block fr 311 * block_read_full_folio->get_block freaks out if it is asked to read 312 * beyond the end of a file, so we che 312 * beyond the end of a file, so we check here. Callers 313 * (generic_file_read, vm_ops->fault) 313 * (generic_file_read, vm_ops->fault) are clever enough to check i_size 314 * and notice that the folio they just 314 * and notice that the folio they just read isn't needed. 315 * 315 * 316 * XXX sys_readahead() seems to get th 316 * XXX sys_readahead() seems to get that wrong? 317 */ 317 */ 318 if (start >= i_size_read(inode)) { 318 if (start >= i_size_read(inode)) { 319 folio_zero_segment(folio, 0, f 319 folio_zero_segment(folio, 0, folio_size(folio)); 320 folio_mark_uptodate(folio); 320 folio_mark_uptodate(folio); 321 ret = 0; 321 ret = 0; 322 goto out_alloc; 322 goto out_alloc; 323 } 323 } 324 324 325 if (oi->ip_dyn_features & OCFS2_INLINE 325 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) 326 ret = ocfs2_readpage_inline(in 326 ret = ocfs2_readpage_inline(inode, &folio->page); 327 else 327 else 328 ret = block_read_full_folio(fo 328 ret = block_read_full_folio(folio, ocfs2_get_block); 329 unlock = 0; 329 unlock = 0; 330 330 331 out_alloc: 331 out_alloc: 332 up_read(&oi->ip_alloc_sem); 332 up_read(&oi->ip_alloc_sem); 333 out_inode_unlock: 333 out_inode_unlock: 334 ocfs2_inode_unlock(inode, 0); 334 ocfs2_inode_unlock(inode, 0); 335 out: 335 out: 336 if (unlock) 336 if (unlock) 337 folio_unlock(folio); 337 folio_unlock(folio); 338 return ret; 338 return ret; 339 } 339 } 340 340 341 /* 341 /* 342 * This is used only for read-ahead. Failures 342 * This is used only for read-ahead. Failures or difficult to handle 343 * situations are safe to ignore. 343 * situations are safe to ignore. 344 * 344 * 345 * Right now, we don't bother with BH_Boundary 345 * Right now, we don't bother with BH_Boundary - in-inode extent lists 346 * are quite large (243 extents on 4k blocks), 346 * are quite large (243 extents on 4k blocks), so most inodes don't 347 * grow out to a tree. If need be, detecting b 347 * grow out to a tree. If need be, detecting boundary extents could 348 * trivially be added in a future version of o 348 * trivially be added in a future version of ocfs2_get_block(). 349 */ 349 */ 350 static void ocfs2_readahead(struct readahead_c 350 static void ocfs2_readahead(struct readahead_control *rac) 351 { 351 { 352 int ret; 352 int ret; 353 struct inode *inode = rac->mapping->ho 353 struct inode *inode = rac->mapping->host; 354 struct ocfs2_inode_info *oi = OCFS2_I( 354 struct ocfs2_inode_info *oi = OCFS2_I(inode); 355 355 356 /* 356 /* 357 * Use the nonblocking flag for the dl 357 * Use the nonblocking flag for the dlm code to avoid page 358 * lock inversion, but don't bother wi 358 * lock inversion, but don't bother with retrying. 359 */ 359 */ 360 ret = ocfs2_inode_lock_full(inode, NUL 360 ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK); 361 if (ret) 361 if (ret) 362 return; 362 return; 363 363 364 if (down_read_trylock(&oi->ip_alloc_se 364 if (down_read_trylock(&oi->ip_alloc_sem) == 0) 365 goto out_unlock; 365 goto out_unlock; 366 366 367 /* 367 /* 368 * Don't bother with inline-data. Ther 368 * Don't bother with inline-data. There isn't anything 369 * to read-ahead in that case anyway.. 369 * to read-ahead in that case anyway... 370 */ 370 */ 371 if (oi->ip_dyn_features & OCFS2_INLINE 371 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) 372 goto out_up; 372 goto out_up; 373 373 374 /* 374 /* 375 * Check whether a remote node truncat 375 * Check whether a remote node truncated this file - we just 376 * drop out in that case as it's not w 376 * drop out in that case as it's not worth handling here. 377 */ 377 */ 378 if (readahead_pos(rac) >= i_size_read( 378 if (readahead_pos(rac) >= i_size_read(inode)) 379 goto out_up; 379 goto out_up; 380 380 381 mpage_readahead(rac, ocfs2_get_block); 381 mpage_readahead(rac, ocfs2_get_block); 382 382 383 out_up: 383 out_up: 384 up_read(&oi->ip_alloc_sem); 384 up_read(&oi->ip_alloc_sem); 385 out_unlock: 385 out_unlock: 386 ocfs2_inode_unlock(inode, 0); 386 ocfs2_inode_unlock(inode, 0); 387 } 387 } 388 388 389 /* Note: Because we don't support holes, our a 389 /* Note: Because we don't support holes, our allocation has 390 * already happened (allocation writes zeros t 390 * already happened (allocation writes zeros to the file data) 391 * so we don't have to worry about ordered wri 391 * so we don't have to worry about ordered writes in 392 * ocfs2_writepages. !! 392 * ocfs2_writepage. 393 * 393 * 394 * ->writepages is called during the process o !! 394 * ->writepage is called during the process of invalidating the page cache 395 * during blocked lock processing. It can't b 395 * during blocked lock processing. It can't block on any cluster locks 396 * to during block mapping. It's relying on t 396 * to during block mapping. It's relying on the fact that the block 397 * mapping can't have disappeared under the di 397 * mapping can't have disappeared under the dirty pages that it is 398 * being asked to write back. 398 * being asked to write back. 399 */ 399 */ 400 static int ocfs2_writepages(struct address_spa !! 400 static int ocfs2_writepage(struct page *page, struct writeback_control *wbc) 401 struct writeback_control *wbc) << 402 { 401 { 403 return mpage_writepages(mapping, wbc, !! 402 trace_ocfs2_writepage( >> 403 (unsigned long long)OCFS2_I(page->mapping->host)->ip_blkno, >> 404 page->index); >> 405 >> 406 return block_write_full_page(page, ocfs2_get_block, wbc); 404 } 407 } 405 408 406 /* Taken from ext3. We don't necessarily need 409 /* Taken from ext3. We don't necessarily need the full blown 407 * functionality yet, but IMHO it's better to 410 * functionality yet, but IMHO it's better to cut and paste the whole 408 * thing so we can avoid introducing our own b 411 * thing so we can avoid introducing our own bugs (and easily pick up 409 * their fixes when they happen) --Mark */ 412 * their fixes when they happen) --Mark */ 410 int walk_page_buffers( handle_t *handle, 413 int walk_page_buffers( handle_t *handle, 411 struct buffer_head *he 414 struct buffer_head *head, 412 unsigned from, 415 unsigned from, 413 unsigned to, 416 unsigned to, 414 int *partial, 417 int *partial, 415 int (*fn)( handle 418 int (*fn)( handle_t *handle, 416 struct 419 struct buffer_head *bh)) 417 { 420 { 418 struct buffer_head *bh; 421 struct buffer_head *bh; 419 unsigned block_start, block_end; 422 unsigned block_start, block_end; 420 unsigned blocksize = head->b_size; 423 unsigned blocksize = head->b_size; 421 int err, ret = 0; 424 int err, ret = 0; 422 struct buffer_head *next; 425 struct buffer_head *next; 423 426 424 for ( bh = head, block_start = 0; 427 for ( bh = head, block_start = 0; 425 ret == 0 && (bh != head || !bl 428 ret == 0 && (bh != head || !block_start); 426 block_start = block_end, bh = 429 block_start = block_end, bh = next) 427 { 430 { 428 next = bh->b_this_page; 431 next = bh->b_this_page; 429 block_end = block_start + bloc 432 block_end = block_start + blocksize; 430 if (block_end <= from || block 433 if (block_end <= from || block_start >= to) { 431 if (partial && !buffer 434 if (partial && !buffer_uptodate(bh)) 432 *partial = 1; 435 *partial = 1; 433 continue; 436 continue; 434 } 437 } 435 err = (*fn)(handle, bh); 438 err = (*fn)(handle, bh); 436 if (!ret) 439 if (!ret) 437 ret = err; 440 ret = err; 438 } 441 } 439 return ret; 442 return ret; 440 } 443 } 441 444 442 static sector_t ocfs2_bmap(struct address_spac 445 static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block) 443 { 446 { 444 sector_t status; 447 sector_t status; 445 u64 p_blkno = 0; 448 u64 p_blkno = 0; 446 int err = 0; 449 int err = 0; 447 struct inode *inode = mapping->host; 450 struct inode *inode = mapping->host; 448 451 449 trace_ocfs2_bmap((unsigned long long)O 452 trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode)->ip_blkno, 450 (unsigned long long)b 453 (unsigned long long)block); 451 454 452 /* 455 /* 453 * The swap code (ab-)uses ->bmap to g 456 * The swap code (ab-)uses ->bmap to get a block mapping and then 454 * bypasseѕ the file system for actua 457 * bypasseѕ the file system for actual I/O. We really can't allow 455 * that on refcounted inodes, so we ha 458 * that on refcounted inodes, so we have to skip out here. And yes, 456 * 0 is the magic code for a bmap erro 459 * 0 is the magic code for a bmap error.. 457 */ 460 */ 458 if (ocfs2_is_refcount_inode(inode)) 461 if (ocfs2_is_refcount_inode(inode)) 459 return 0; 462 return 0; 460 463 461 /* We don't need to lock journal syste 464 /* We don't need to lock journal system files, since they aren't 462 * accessed concurrently from multiple 465 * accessed concurrently from multiple nodes. 463 */ 466 */ 464 if (!INODE_JOURNAL(inode)) { 467 if (!INODE_JOURNAL(inode)) { 465 err = ocfs2_inode_lock(inode, 468 err = ocfs2_inode_lock(inode, NULL, 0); 466 if (err) { 469 if (err) { 467 if (err != -ENOENT) 470 if (err != -ENOENT) 468 mlog_errno(err 471 mlog_errno(err); 469 goto bail; 472 goto bail; 470 } 473 } 471 down_read(&OCFS2_I(inode)->ip_ 474 down_read(&OCFS2_I(inode)->ip_alloc_sem); 472 } 475 } 473 476 474 if (!(OCFS2_I(inode)->ip_dyn_features 477 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)) 475 err = ocfs2_extent_map_get_blo 478 err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL, 476 479 NULL); 477 480 478 if (!INODE_JOURNAL(inode)) { 481 if (!INODE_JOURNAL(inode)) { 479 up_read(&OCFS2_I(inode)->ip_al 482 up_read(&OCFS2_I(inode)->ip_alloc_sem); 480 ocfs2_inode_unlock(inode, 0); 483 ocfs2_inode_unlock(inode, 0); 481 } 484 } 482 485 483 if (err) { 486 if (err) { 484 mlog(ML_ERROR, "get_blocks() f 487 mlog(ML_ERROR, "get_blocks() failed, block = %llu\n", 485 (unsigned long long)block 488 (unsigned long long)block); 486 mlog_errno(err); 489 mlog_errno(err); 487 goto bail; 490 goto bail; 488 } 491 } 489 492 490 bail: 493 bail: 491 status = err ? 0 : p_blkno; 494 status = err ? 0 : p_blkno; 492 495 493 return status; 496 return status; 494 } 497 } 495 498 496 static bool ocfs2_release_folio(struct folio * 499 static bool ocfs2_release_folio(struct folio *folio, gfp_t wait) 497 { 500 { 498 if (!folio_buffers(folio)) 501 if (!folio_buffers(folio)) 499 return false; 502 return false; 500 return try_to_free_buffers(folio); 503 return try_to_free_buffers(folio); 501 } 504 } 502 505 503 static void ocfs2_figure_cluster_boundaries(st 506 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb, 504 u3 507 u32 cpos, 505 un 508 unsigned int *start, 506 un 509 unsigned int *end) 507 { 510 { 508 unsigned int cluster_start = 0, cluste 511 unsigned int cluster_start = 0, cluster_end = PAGE_SIZE; 509 512 510 if (unlikely(PAGE_SHIFT > osb->s_clust 513 if (unlikely(PAGE_SHIFT > osb->s_clustersize_bits)) { 511 unsigned int cpp; 514 unsigned int cpp; 512 515 513 cpp = 1 << (PAGE_SHIFT - osb-> 516 cpp = 1 << (PAGE_SHIFT - osb->s_clustersize_bits); 514 517 515 cluster_start = cpos % cpp; 518 cluster_start = cpos % cpp; 516 cluster_start = cluster_start 519 cluster_start = cluster_start << osb->s_clustersize_bits; 517 520 518 cluster_end = cluster_start + 521 cluster_end = cluster_start + osb->s_clustersize; 519 } 522 } 520 523 521 BUG_ON(cluster_start > PAGE_SIZE); 524 BUG_ON(cluster_start > PAGE_SIZE); 522 BUG_ON(cluster_end > PAGE_SIZE); 525 BUG_ON(cluster_end > PAGE_SIZE); 523 526 524 if (start) 527 if (start) 525 *start = cluster_start; 528 *start = cluster_start; 526 if (end) 529 if (end) 527 *end = cluster_end; 530 *end = cluster_end; 528 } 531 } 529 532 530 /* 533 /* 531 * 'from' and 'to' are the region in the page 534 * 'from' and 'to' are the region in the page to avoid zeroing. 532 * 535 * 533 * If pagesize > clustersize, this function wi 536 * If pagesize > clustersize, this function will avoid zeroing outside 534 * of the cluster boundary. 537 * of the cluster boundary. 535 * 538 * 536 * from == to == 0 is code for "zero the entir 539 * from == to == 0 is code for "zero the entire cluster region" 537 */ 540 */ 538 static void ocfs2_clear_page_regions(struct pa 541 static void ocfs2_clear_page_regions(struct page *page, 539 struct oc 542 struct ocfs2_super *osb, u32 cpos, 540 unsigned 543 unsigned from, unsigned to) 541 { 544 { 542 void *kaddr; 545 void *kaddr; 543 unsigned int cluster_start, cluster_en 546 unsigned int cluster_start, cluster_end; 544 547 545 ocfs2_figure_cluster_boundaries(osb, c 548 ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end); 546 549 547 kaddr = kmap_atomic(page); 550 kaddr = kmap_atomic(page); 548 551 549 if (from || to) { 552 if (from || to) { 550 if (from > cluster_start) 553 if (from > cluster_start) 551 memset(kaddr + cluster 554 memset(kaddr + cluster_start, 0, from - cluster_start); 552 if (to < cluster_end) 555 if (to < cluster_end) 553 memset(kaddr + to, 0, 556 memset(kaddr + to, 0, cluster_end - to); 554 } else { 557 } else { 555 memset(kaddr + cluster_start, 558 memset(kaddr + cluster_start, 0, cluster_end - cluster_start); 556 } 559 } 557 560 558 kunmap_atomic(kaddr); 561 kunmap_atomic(kaddr); 559 } 562 } 560 563 561 /* 564 /* 562 * Nonsparse file systems fully allocate befor 565 * Nonsparse file systems fully allocate before we get to the write 563 * code. This prevents ocfs2_write() from tagg 566 * code. This prevents ocfs2_write() from tagging the write as an 564 * allocating one, which means ocfs2_map_page_ 567 * allocating one, which means ocfs2_map_page_blocks() might try to 565 * read-in the blocks at the tail of our file. 568 * read-in the blocks at the tail of our file. Avoid reading them by 566 * testing i_size against each block offset. 569 * testing i_size against each block offset. 567 */ 570 */ 568 static int ocfs2_should_read_blk(struct inode !! 571 static int ocfs2_should_read_blk(struct inode *inode, struct page *page, 569 unsigned int 572 unsigned int block_start) 570 { 573 { 571 u64 offset = folio_pos(folio) + block_ !! 574 u64 offset = page_offset(page) + block_start; 572 575 573 if (ocfs2_sparse_alloc(OCFS2_SB(inode- 576 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) 574 return 1; 577 return 1; 575 578 576 if (i_size_read(inode) > offset) 579 if (i_size_read(inode) > offset) 577 return 1; 580 return 1; 578 581 579 return 0; 582 return 0; 580 } 583 } 581 584 582 /* 585 /* 583 * Some of this taken from __block_write_begin 586 * Some of this taken from __block_write_begin(). We already have our 584 * mapping by now though, and the entire write 587 * mapping by now though, and the entire write will be allocating or 585 * it won't, so not much need to use BH_New. 588 * it won't, so not much need to use BH_New. 586 * 589 * 587 * This will also skip zeroing, which is handl 590 * This will also skip zeroing, which is handled externally. 588 */ 591 */ 589 int ocfs2_map_page_blocks(struct page *page, u 592 int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno, 590 struct inode *inode, 593 struct inode *inode, unsigned int from, 591 unsigned int to, int 594 unsigned int to, int new) 592 { 595 { 593 struct folio *folio = page_folio(page) << 594 int ret = 0; 596 int ret = 0; 595 struct buffer_head *head, *bh, *wait[2 597 struct buffer_head *head, *bh, *wait[2], **wait_bh = wait; 596 unsigned int block_end, block_start; 598 unsigned int block_end, block_start; 597 unsigned int bsize = i_blocksize(inode 599 unsigned int bsize = i_blocksize(inode); 598 600 599 head = folio_buffers(folio); !! 601 if (!page_has_buffers(page)) 600 if (!head) !! 602 create_empty_buffers(page, bsize, 0); 601 head = create_empty_buffers(fo << 602 603 >> 604 head = page_buffers(page); 603 for (bh = head, block_start = 0; bh != 605 for (bh = head, block_start = 0; bh != head || !block_start; 604 bh = bh->b_this_page, block_start 606 bh = bh->b_this_page, block_start += bsize) { 605 block_end = block_start + bsiz 607 block_end = block_start + bsize; 606 608 607 clear_buffer_new(bh); 609 clear_buffer_new(bh); 608 610 609 /* 611 /* 610 * Ignore blocks outside of ou 612 * Ignore blocks outside of our i/o range - 611 * they may belong to unalloca 613 * they may belong to unallocated clusters. 612 */ 614 */ 613 if (block_start >= to || block 615 if (block_start >= to || block_end <= from) { 614 if (folio_test_uptodat !! 616 if (PageUptodate(page)) 615 set_buffer_upt 617 set_buffer_uptodate(bh); 616 continue; 618 continue; 617 } 619 } 618 620 619 /* 621 /* 620 * For an allocating write wit 622 * For an allocating write with cluster size >= page 621 * size, we always write the e 623 * size, we always write the entire page. 622 */ 624 */ 623 if (new) 625 if (new) 624 set_buffer_new(bh); 626 set_buffer_new(bh); 625 627 626 if (!buffer_mapped(bh)) { 628 if (!buffer_mapped(bh)) { 627 map_bh(bh, inode->i_sb 629 map_bh(bh, inode->i_sb, *p_blkno); 628 clean_bdev_bh_alias(bh 630 clean_bdev_bh_alias(bh); 629 } 631 } 630 632 631 if (folio_test_uptodate(folio) !! 633 if (PageUptodate(page)) { 632 set_buffer_uptodate(bh 634 set_buffer_uptodate(bh); 633 } else if (!buffer_uptodate(bh 635 } else if (!buffer_uptodate(bh) && !buffer_delay(bh) && 634 !buffer_new(bh) && 636 !buffer_new(bh) && 635 ocfs2_should_read_b !! 637 ocfs2_should_read_blk(inode, page, block_start) && 636 (block_start < from 638 (block_start < from || block_end > to)) { 637 bh_read_nowait(bh, 0); !! 639 ll_rw_block(REQ_OP_READ, 1, &bh); 638 *wait_bh++=bh; 640 *wait_bh++=bh; 639 } 641 } 640 642 641 *p_blkno = *p_blkno + 1; 643 *p_blkno = *p_blkno + 1; 642 } 644 } 643 645 644 /* 646 /* 645 * If we issued read requests - let th 647 * If we issued read requests - let them complete. 646 */ 648 */ 647 while(wait_bh > wait) { 649 while(wait_bh > wait) { 648 wait_on_buffer(*--wait_bh); 650 wait_on_buffer(*--wait_bh); 649 if (!buffer_uptodate(*wait_bh) 651 if (!buffer_uptodate(*wait_bh)) 650 ret = -EIO; 652 ret = -EIO; 651 } 653 } 652 654 653 if (ret == 0 || !new) 655 if (ret == 0 || !new) 654 return ret; 656 return ret; 655 657 656 /* 658 /* 657 * If we get -EIO above, zero out any 659 * If we get -EIO above, zero out any newly allocated blocks 658 * to avoid exposing stale data. 660 * to avoid exposing stale data. 659 */ 661 */ 660 bh = head; 662 bh = head; 661 block_start = 0; 663 block_start = 0; 662 do { 664 do { 663 block_end = block_start + bsiz 665 block_end = block_start + bsize; 664 if (block_end <= from) 666 if (block_end <= from) 665 goto next_bh; 667 goto next_bh; 666 if (block_start >= to) 668 if (block_start >= to) 667 break; 669 break; 668 670 669 folio_zero_range(folio, block_ !! 671 zero_user(page, block_start, bh->b_size); 670 set_buffer_uptodate(bh); 672 set_buffer_uptodate(bh); 671 mark_buffer_dirty(bh); 673 mark_buffer_dirty(bh); 672 674 673 next_bh: 675 next_bh: 674 block_start = block_end; 676 block_start = block_end; 675 bh = bh->b_this_page; 677 bh = bh->b_this_page; 676 } while (bh != head); 678 } while (bh != head); 677 679 678 return ret; 680 return ret; 679 } 681 } 680 682 681 #if (PAGE_SIZE >= OCFS2_MAX_CLUSTERSIZE) 683 #if (PAGE_SIZE >= OCFS2_MAX_CLUSTERSIZE) 682 #define OCFS2_MAX_CTXT_PAGES 1 684 #define OCFS2_MAX_CTXT_PAGES 1 683 #else 685 #else 684 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLU 686 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_SIZE) 685 #endif 687 #endif 686 688 687 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_ 689 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_SIZE / OCFS2_MIN_CLUSTERSIZE) 688 690 689 struct ocfs2_unwritten_extent { 691 struct ocfs2_unwritten_extent { 690 struct list_head ue_node; 692 struct list_head ue_node; 691 struct list_head ue_ip_node; 693 struct list_head ue_ip_node; 692 u32 ue_cpos; 694 u32 ue_cpos; 693 u32 ue_phys; 695 u32 ue_phys; 694 }; 696 }; 695 697 696 /* 698 /* 697 * Describe the state of a single cluster to b 699 * Describe the state of a single cluster to be written to. 698 */ 700 */ 699 struct ocfs2_write_cluster_desc { 701 struct ocfs2_write_cluster_desc { 700 u32 c_cpos; 702 u32 c_cpos; 701 u32 c_phys; 703 u32 c_phys; 702 /* 704 /* 703 * Give this a unique field because c_ 705 * Give this a unique field because c_phys eventually gets 704 * filled. 706 * filled. 705 */ 707 */ 706 unsigned c_new; 708 unsigned c_new; 707 unsigned c_clear_unwritten; 709 unsigned c_clear_unwritten; 708 unsigned c_needs_zero; 710 unsigned c_needs_zero; 709 }; 711 }; 710 712 711 struct ocfs2_write_ctxt { 713 struct ocfs2_write_ctxt { 712 /* Logical cluster position / len of w 714 /* Logical cluster position / len of write */ 713 u32 w_cpos 715 u32 w_cpos; 714 u32 w_clen 716 u32 w_clen; 715 717 716 /* First cluster allocated in a nonspa 718 /* First cluster allocated in a nonsparse extend */ 717 u32 w_firs 719 u32 w_first_new_cpos; 718 720 719 /* Type of caller. Must be one of buff 721 /* Type of caller. Must be one of buffer, mmap, direct. */ 720 ocfs2_write_type_t w_type 722 ocfs2_write_type_t w_type; 721 723 722 struct ocfs2_write_cluster_desc w_desc 724 struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE]; 723 725 724 /* 726 /* 725 * This is true if page_size > cluster 727 * This is true if page_size > cluster_size. 726 * 728 * 727 * It triggers a set of special cases 729 * It triggers a set of special cases during write which might 728 * have to deal with allocating writes 730 * have to deal with allocating writes to partial pages. 729 */ 731 */ 730 unsigned int w_larg 732 unsigned int w_large_pages; 731 733 732 /* 734 /* 733 * Pages involved in this write. 735 * Pages involved in this write. 734 * 736 * 735 * w_target_page is the page being wri 737 * w_target_page is the page being written to by the user. 736 * 738 * 737 * w_pages is an array of pages which 739 * w_pages is an array of pages which always contains 738 * w_target_page, and in the case of a 740 * w_target_page, and in the case of an allocating write with 739 * page_size < cluster size, it will c 741 * page_size < cluster size, it will contain zero'd and mapped 740 * pages adjacent to w_target_page whi 742 * pages adjacent to w_target_page which need to be written 741 * out in so that future reads from th 743 * out in so that future reads from that region will get 742 * zero's. 744 * zero's. 743 */ 745 */ 744 unsigned int w_num_ 746 unsigned int w_num_pages; 745 struct page *w_pag 747 struct page *w_pages[OCFS2_MAX_CTXT_PAGES]; 746 struct page *w_tar 748 struct page *w_target_page; 747 749 748 /* 750 /* 749 * w_target_locked is used for page_mk 751 * w_target_locked is used for page_mkwrite path indicating no unlocking 750 * against w_target_page in ocfs2_writ 752 * against w_target_page in ocfs2_write_end_nolock. 751 */ 753 */ 752 unsigned int w_targ 754 unsigned int w_target_locked:1; 753 755 754 /* 756 /* 755 * ocfs2_write_end() uses this to know 757 * ocfs2_write_end() uses this to know what the real range to 756 * write in the target should be. 758 * write in the target should be. 757 */ 759 */ 758 unsigned int w_targ 760 unsigned int w_target_from; 759 unsigned int w_targ 761 unsigned int w_target_to; 760 762 761 /* 763 /* 762 * We could use journal_current_handle 764 * We could use journal_current_handle() but this is cleaner, 763 * IMHO -Mark 765 * IMHO -Mark 764 */ 766 */ 765 handle_t *w_han 767 handle_t *w_handle; 766 768 767 struct buffer_head *w_di_ 769 struct buffer_head *w_di_bh; 768 770 769 struct ocfs2_cached_dealloc_ctxt w_dea 771 struct ocfs2_cached_dealloc_ctxt w_dealloc; 770 772 771 struct list_head w_unwr 773 struct list_head w_unwritten_list; 772 unsigned int w_unwr 774 unsigned int w_unwritten_count; 773 }; 775 }; 774 776 775 void ocfs2_unlock_and_free_pages(struct page * 777 void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages) 776 { 778 { 777 int i; 779 int i; 778 780 779 for(i = 0; i < num_pages; i++) { 781 for(i = 0; i < num_pages; i++) { 780 if (pages[i]) { 782 if (pages[i]) { 781 unlock_page(pages[i]); 783 unlock_page(pages[i]); 782 mark_page_accessed(pag 784 mark_page_accessed(pages[i]); 783 put_page(pages[i]); 785 put_page(pages[i]); 784 } 786 } 785 } 787 } 786 } 788 } 787 789 788 static void ocfs2_unlock_pages(struct ocfs2_wr 790 static void ocfs2_unlock_pages(struct ocfs2_write_ctxt *wc) 789 { 791 { 790 int i; 792 int i; 791 793 792 /* 794 /* 793 * w_target_locked is only set to true 795 * w_target_locked is only set to true in the page_mkwrite() case. 794 * The intent is to allow us to lock t 796 * The intent is to allow us to lock the target page from write_begin() 795 * to write_end(). The caller must hol 797 * to write_end(). The caller must hold a ref on w_target_page. 796 */ 798 */ 797 if (wc->w_target_locked) { 799 if (wc->w_target_locked) { 798 BUG_ON(!wc->w_target_page); 800 BUG_ON(!wc->w_target_page); 799 for (i = 0; i < wc->w_num_page 801 for (i = 0; i < wc->w_num_pages; i++) { 800 if (wc->w_target_page 802 if (wc->w_target_page == wc->w_pages[i]) { 801 wc->w_pages[i] 803 wc->w_pages[i] = NULL; 802 break; 804 break; 803 } 805 } 804 } 806 } 805 mark_page_accessed(wc->w_targe 807 mark_page_accessed(wc->w_target_page); 806 put_page(wc->w_target_page); 808 put_page(wc->w_target_page); 807 } 809 } 808 ocfs2_unlock_and_free_pages(wc->w_page 810 ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages); 809 } 811 } 810 812 811 static void ocfs2_free_unwritten_list(struct i 813 static void ocfs2_free_unwritten_list(struct inode *inode, 812 struct list_h 814 struct list_head *head) 813 { 815 { 814 struct ocfs2_inode_info *oi = OCFS2_I( 816 struct ocfs2_inode_info *oi = OCFS2_I(inode); 815 struct ocfs2_unwritten_extent *ue = NU 817 struct ocfs2_unwritten_extent *ue = NULL, *tmp = NULL; 816 818 817 list_for_each_entry_safe(ue, tmp, head 819 list_for_each_entry_safe(ue, tmp, head, ue_node) { 818 list_del(&ue->ue_node); 820 list_del(&ue->ue_node); 819 spin_lock(&oi->ip_lock); 821 spin_lock(&oi->ip_lock); 820 list_del(&ue->ue_ip_node); 822 list_del(&ue->ue_ip_node); 821 spin_unlock(&oi->ip_lock); 823 spin_unlock(&oi->ip_lock); 822 kfree(ue); 824 kfree(ue); 823 } 825 } 824 } 826 } 825 827 826 static void ocfs2_free_write_ctxt(struct inode 828 static void ocfs2_free_write_ctxt(struct inode *inode, 827 struct ocfs2 829 struct ocfs2_write_ctxt *wc) 828 { 830 { 829 ocfs2_free_unwritten_list(inode, &wc-> 831 ocfs2_free_unwritten_list(inode, &wc->w_unwritten_list); 830 ocfs2_unlock_pages(wc); 832 ocfs2_unlock_pages(wc); 831 brelse(wc->w_di_bh); 833 brelse(wc->w_di_bh); 832 kfree(wc); 834 kfree(wc); 833 } 835 } 834 836 835 static int ocfs2_alloc_write_ctxt(struct ocfs2 837 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp, 836 struct ocfs2 838 struct ocfs2_super *osb, loff_t pos, 837 unsigned len 839 unsigned len, ocfs2_write_type_t type, 838 struct buffe 840 struct buffer_head *di_bh) 839 { 841 { 840 u32 cend; 842 u32 cend; 841 struct ocfs2_write_ctxt *wc; 843 struct ocfs2_write_ctxt *wc; 842 844 843 wc = kzalloc(sizeof(struct ocfs2_write 845 wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS); 844 if (!wc) 846 if (!wc) 845 return -ENOMEM; 847 return -ENOMEM; 846 848 847 wc->w_cpos = pos >> osb->s_clustersize 849 wc->w_cpos = pos >> osb->s_clustersize_bits; 848 wc->w_first_new_cpos = UINT_MAX; 850 wc->w_first_new_cpos = UINT_MAX; 849 cend = (pos + len - 1) >> osb->s_clust 851 cend = (pos + len - 1) >> osb->s_clustersize_bits; 850 wc->w_clen = cend - wc->w_cpos + 1; 852 wc->w_clen = cend - wc->w_cpos + 1; 851 get_bh(di_bh); 853 get_bh(di_bh); 852 wc->w_di_bh = di_bh; 854 wc->w_di_bh = di_bh; 853 wc->w_type = type; 855 wc->w_type = type; 854 856 855 if (unlikely(PAGE_SHIFT > osb->s_clust 857 if (unlikely(PAGE_SHIFT > osb->s_clustersize_bits)) 856 wc->w_large_pages = 1; 858 wc->w_large_pages = 1; 857 else 859 else 858 wc->w_large_pages = 0; 860 wc->w_large_pages = 0; 859 861 860 ocfs2_init_dealloc_ctxt(&wc->w_dealloc 862 ocfs2_init_dealloc_ctxt(&wc->w_dealloc); 861 INIT_LIST_HEAD(&wc->w_unwritten_list); 863 INIT_LIST_HEAD(&wc->w_unwritten_list); 862 864 863 *wcp = wc; 865 *wcp = wc; 864 866 865 return 0; 867 return 0; 866 } 868 } 867 869 868 /* 870 /* 869 * If a page has any new buffers, zero them ou 871 * If a page has any new buffers, zero them out here, and mark them uptodate 870 * and dirty so they'll be written out (in ord 872 * and dirty so they'll be written out (in order to prevent uninitialised 871 * block data from leaking). And clear the new 873 * block data from leaking). And clear the new bit. 872 */ 874 */ 873 static void ocfs2_zero_new_buffers(struct page 875 static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to) 874 { 876 { 875 unsigned int block_start, block_end; 877 unsigned int block_start, block_end; 876 struct buffer_head *head, *bh; 878 struct buffer_head *head, *bh; 877 879 878 BUG_ON(!PageLocked(page)); 880 BUG_ON(!PageLocked(page)); 879 if (!page_has_buffers(page)) 881 if (!page_has_buffers(page)) 880 return; 882 return; 881 883 882 bh = head = page_buffers(page); 884 bh = head = page_buffers(page); 883 block_start = 0; 885 block_start = 0; 884 do { 886 do { 885 block_end = block_start + bh-> 887 block_end = block_start + bh->b_size; 886 888 887 if (buffer_new(bh)) { 889 if (buffer_new(bh)) { 888 if (block_end > from & 890 if (block_end > from && block_start < to) { 889 if (!PageUptod 891 if (!PageUptodate(page)) { 890 unsign 892 unsigned start, end; 891 893 892 start 894 start = max(from, block_start); 893 end = 895 end = min(to, block_end); 894 896 895 zero_u 897 zero_user_segment(page, start, end); 896 set_bu 898 set_buffer_uptodate(bh); 897 } 899 } 898 900 899 clear_buffer_n 901 clear_buffer_new(bh); 900 mark_buffer_di 902 mark_buffer_dirty(bh); 901 } 903 } 902 } 904 } 903 905 904 block_start = block_end; 906 block_start = block_end; 905 bh = bh->b_this_page; 907 bh = bh->b_this_page; 906 } while (bh != head); 908 } while (bh != head); 907 } 909 } 908 910 909 /* 911 /* 910 * Only called when we have a failure during a 912 * Only called when we have a failure during allocating write to write 911 * zero's to the newly allocated region. 913 * zero's to the newly allocated region. 912 */ 914 */ 913 static void ocfs2_write_failure(struct inode * 915 static void ocfs2_write_failure(struct inode *inode, 914 struct ocfs2_w 916 struct ocfs2_write_ctxt *wc, 915 loff_t user_po 917 loff_t user_pos, unsigned user_len) 916 { 918 { 917 int i; 919 int i; 918 unsigned from = user_pos & (PAGE_SIZE 920 unsigned from = user_pos & (PAGE_SIZE - 1), 919 to = user_pos + user_len; 921 to = user_pos + user_len; 920 struct page *tmppage; 922 struct page *tmppage; 921 923 922 if (wc->w_target_page) 924 if (wc->w_target_page) 923 ocfs2_zero_new_buffers(wc->w_t 925 ocfs2_zero_new_buffers(wc->w_target_page, from, to); 924 926 925 for(i = 0; i < wc->w_num_pages; i++) { 927 for(i = 0; i < wc->w_num_pages; i++) { 926 tmppage = wc->w_pages[i]; 928 tmppage = wc->w_pages[i]; 927 929 928 if (tmppage && page_has_buffer 930 if (tmppage && page_has_buffers(tmppage)) { 929 if (ocfs2_should_order 931 if (ocfs2_should_order_data(inode)) 930 ocfs2_jbd2_ino 932 ocfs2_jbd2_inode_add_write(wc->w_handle, inode, 931 933 user_pos, user_len); 932 934 933 block_commit_write(tmp 935 block_commit_write(tmppage, from, to); 934 } 936 } 935 } 937 } 936 } 938 } 937 939 938 static int ocfs2_prepare_page_for_write(struct 940 static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno, 939 struct 941 struct ocfs2_write_ctxt *wc, 940 struct 942 struct page *page, u32 cpos, 941 loff_t 943 loff_t user_pos, unsigned user_len, 942 int ne 944 int new) 943 { 945 { 944 int ret; 946 int ret; 945 unsigned int map_from = 0, map_to = 0; 947 unsigned int map_from = 0, map_to = 0; 946 unsigned int cluster_start, cluster_en 948 unsigned int cluster_start, cluster_end; 947 unsigned int user_data_from = 0, user_ 949 unsigned int user_data_from = 0, user_data_to = 0; 948 950 949 ocfs2_figure_cluster_boundaries(OCFS2_ 951 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos, 950 &clust 952 &cluster_start, &cluster_end); 951 953 952 /* treat the write as new if the a hol 954 /* treat the write as new if the a hole/lseek spanned across 953 * the page boundary. 955 * the page boundary. 954 */ 956 */ 955 new = new | ((i_size_read(inode) <= pa 957 new = new | ((i_size_read(inode) <= page_offset(page)) && 956 (page_offset(page) <= 958 (page_offset(page) <= user_pos)); 957 959 958 if (page == wc->w_target_page) { 960 if (page == wc->w_target_page) { 959 map_from = user_pos & (PAGE_SI 961 map_from = user_pos & (PAGE_SIZE - 1); 960 map_to = map_from + user_len; 962 map_to = map_from + user_len; 961 963 962 if (new) 964 if (new) 963 ret = ocfs2_map_page_b 965 ret = ocfs2_map_page_blocks(page, p_blkno, inode, 964 966 cluster_start, cluster_end, 965 967 new); 966 else 968 else 967 ret = ocfs2_map_page_b 969 ret = ocfs2_map_page_blocks(page, p_blkno, inode, 968 970 map_from, map_to, new); 969 if (ret) { 971 if (ret) { 970 mlog_errno(ret); 972 mlog_errno(ret); 971 goto out; 973 goto out; 972 } 974 } 973 975 974 user_data_from = map_from; 976 user_data_from = map_from; 975 user_data_to = map_to; 977 user_data_to = map_to; 976 if (new) { 978 if (new) { 977 map_from = cluster_sta 979 map_from = cluster_start; 978 map_to = cluster_end; 980 map_to = cluster_end; 979 } 981 } 980 } else { 982 } else { 981 /* 983 /* 982 * If we haven't allocated the 984 * If we haven't allocated the new page yet, we 983 * shouldn't be writing it out 985 * shouldn't be writing it out without copying user 984 * data. This is likely a math 986 * data. This is likely a math error from the caller. 985 */ 987 */ 986 BUG_ON(!new); 988 BUG_ON(!new); 987 989 988 map_from = cluster_start; 990 map_from = cluster_start; 989 map_to = cluster_end; 991 map_to = cluster_end; 990 992 991 ret = ocfs2_map_page_blocks(pa 993 ret = ocfs2_map_page_blocks(page, p_blkno, inode, 992 cl 994 cluster_start, cluster_end, new); 993 if (ret) { 995 if (ret) { 994 mlog_errno(ret); 996 mlog_errno(ret); 995 goto out; 997 goto out; 996 } 998 } 997 } 999 } 998 1000 999 /* 1001 /* 1000 * Parts of newly allocated pages nee 1002 * Parts of newly allocated pages need to be zero'd. 1001 * 1003 * 1002 * Above, we have also rewritten 'to' 1004 * Above, we have also rewritten 'to' and 'from' - as far as 1003 * the rest of the function is concer 1005 * the rest of the function is concerned, the entire cluster 1004 * range inside of a page needs to be 1006 * range inside of a page needs to be written. 1005 * 1007 * 1006 * We can skip this if the page is up 1008 * We can skip this if the page is up to date - it's already 1007 * been zero'd from being read in as 1009 * been zero'd from being read in as a hole. 1008 */ 1010 */ 1009 if (new && !PageUptodate(page)) 1011 if (new && !PageUptodate(page)) 1010 ocfs2_clear_page_regions(page 1012 ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb), 1011 cpos 1013 cpos, user_data_from, user_data_to); 1012 1014 1013 flush_dcache_page(page); 1015 flush_dcache_page(page); 1014 1016 1015 out: 1017 out: 1016 return ret; 1018 return ret; 1017 } 1019 } 1018 1020 1019 /* 1021 /* 1020 * This function will only grab one clusters 1022 * This function will only grab one clusters worth of pages. 1021 */ 1023 */ 1022 static int ocfs2_grab_pages_for_write(struct 1024 static int ocfs2_grab_pages_for_write(struct address_space *mapping, 1023 struct 1025 struct ocfs2_write_ctxt *wc, 1024 u32 cpo 1026 u32 cpos, loff_t user_pos, 1025 unsigne 1027 unsigned user_len, int new, 1026 struct 1028 struct page *mmap_page) 1027 { 1029 { 1028 int ret = 0, i; 1030 int ret = 0, i; 1029 unsigned long start, target_index, en 1031 unsigned long start, target_index, end_index, index; 1030 struct inode *inode = mapping->host; 1032 struct inode *inode = mapping->host; 1031 loff_t last_byte; 1033 loff_t last_byte; 1032 1034 1033 target_index = user_pos >> PAGE_SHIFT 1035 target_index = user_pos >> PAGE_SHIFT; 1034 1036 1035 /* 1037 /* 1036 * Figure out how many pages we'll be 1038 * Figure out how many pages we'll be manipulating here. For 1037 * non allocating write, we just chan 1039 * non allocating write, we just change the one 1038 * page. Otherwise, we'll need a whol 1040 * page. Otherwise, we'll need a whole clusters worth. If we're 1039 * writing past i_size, we only need 1041 * writing past i_size, we only need enough pages to cover the 1040 * last page of the write. 1042 * last page of the write. 1041 */ 1043 */ 1042 if (new) { 1044 if (new) { 1043 wc->w_num_pages = ocfs2_pages 1045 wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb); 1044 start = ocfs2_align_clusters_ 1046 start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos); 1045 /* 1047 /* 1046 * We need the index *past* t 1048 * We need the index *past* the last page we could possibly 1047 * touch. This is the page p 1049 * touch. This is the page past the end of the write or 1048 * i_size, whichever is great 1050 * i_size, whichever is greater. 1049 */ 1051 */ 1050 last_byte = max(user_pos + us 1052 last_byte = max(user_pos + user_len, i_size_read(inode)); 1051 BUG_ON(last_byte < 1); 1053 BUG_ON(last_byte < 1); 1052 end_index = ((last_byte - 1) 1054 end_index = ((last_byte - 1) >> PAGE_SHIFT) + 1; 1053 if ((start + wc->w_num_pages) 1055 if ((start + wc->w_num_pages) > end_index) 1054 wc->w_num_pages = end 1056 wc->w_num_pages = end_index - start; 1055 } else { 1057 } else { 1056 wc->w_num_pages = 1; 1058 wc->w_num_pages = 1; 1057 start = target_index; 1059 start = target_index; 1058 } 1060 } 1059 end_index = (user_pos + user_len - 1) 1061 end_index = (user_pos + user_len - 1) >> PAGE_SHIFT; 1060 1062 1061 for(i = 0; i < wc->w_num_pages; i++) 1063 for(i = 0; i < wc->w_num_pages; i++) { 1062 index = start + i; 1064 index = start + i; 1063 1065 1064 if (index >= target_index && 1066 if (index >= target_index && index <= end_index && 1065 wc->w_type == OCFS2_WRITE 1067 wc->w_type == OCFS2_WRITE_MMAP) { 1066 /* 1068 /* 1067 * ocfs2_pagemkwrite( 1069 * ocfs2_pagemkwrite() is a little different 1068 * and wants us to di 1070 * and wants us to directly use the page 1069 * passed in. 1071 * passed in. 1070 */ 1072 */ 1071 lock_page(mmap_page); 1073 lock_page(mmap_page); 1072 1074 1073 /* Exit and let the c 1075 /* Exit and let the caller retry */ 1074 if (mmap_page->mappin 1076 if (mmap_page->mapping != mapping) { 1075 WARN_ON(mmap_ 1077 WARN_ON(mmap_page->mapping); 1076 unlock_page(m 1078 unlock_page(mmap_page); 1077 ret = -EAGAIN 1079 ret = -EAGAIN; 1078 goto out; 1080 goto out; 1079 } 1081 } 1080 1082 1081 get_page(mmap_page); 1083 get_page(mmap_page); 1082 wc->w_pages[i] = mmap 1084 wc->w_pages[i] = mmap_page; 1083 wc->w_target_locked = 1085 wc->w_target_locked = true; 1084 } else if (index >= target_in 1086 } else if (index >= target_index && index <= end_index && 1085 wc->w_type == OCFS 1087 wc->w_type == OCFS2_WRITE_DIRECT) { 1086 /* Direct write has n 1088 /* Direct write has no mapping page. */ 1087 wc->w_pages[i] = NULL 1089 wc->w_pages[i] = NULL; 1088 continue; 1090 continue; 1089 } else { 1091 } else { 1090 wc->w_pages[i] = find 1092 wc->w_pages[i] = find_or_create_page(mapping, index, 1091 1093 GFP_NOFS); 1092 if (!wc->w_pages[i]) 1094 if (!wc->w_pages[i]) { 1093 ret = -ENOMEM 1095 ret = -ENOMEM; 1094 mlog_errno(re 1096 mlog_errno(ret); 1095 goto out; 1097 goto out; 1096 } 1098 } 1097 } 1099 } 1098 wait_for_stable_page(wc->w_pa 1100 wait_for_stable_page(wc->w_pages[i]); 1099 1101 1100 if (index == target_index) 1102 if (index == target_index) 1101 wc->w_target_page = w 1103 wc->w_target_page = wc->w_pages[i]; 1102 } 1104 } 1103 out: 1105 out: 1104 if (ret) 1106 if (ret) 1105 wc->w_target_locked = false; 1107 wc->w_target_locked = false; 1106 return ret; 1108 return ret; 1107 } 1109 } 1108 1110 1109 /* 1111 /* 1110 * Prepare a single cluster for write one clu 1112 * Prepare a single cluster for write one cluster into the file. 1111 */ 1113 */ 1112 static int ocfs2_write_cluster(struct address 1114 static int ocfs2_write_cluster(struct address_space *mapping, 1113 u32 *phys, uns 1115 u32 *phys, unsigned int new, 1114 unsigned int c 1116 unsigned int clear_unwritten, 1115 unsigned int s 1117 unsigned int should_zero, 1116 struct ocfs2_a 1118 struct ocfs2_alloc_context *data_ac, 1117 struct ocfs2_a 1119 struct ocfs2_alloc_context *meta_ac, 1118 struct ocfs2_w 1120 struct ocfs2_write_ctxt *wc, u32 cpos, 1119 loff_t user_po 1121 loff_t user_pos, unsigned user_len) 1120 { 1122 { 1121 int ret, i; 1123 int ret, i; 1122 u64 p_blkno; 1124 u64 p_blkno; 1123 struct inode *inode = mapping->host; 1125 struct inode *inode = mapping->host; 1124 struct ocfs2_extent_tree et; 1126 struct ocfs2_extent_tree et; 1125 int bpc = ocfs2_clusters_to_blocks(in 1127 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1); 1126 1128 1127 if (new) { 1129 if (new) { 1128 u32 tmp_pos; 1130 u32 tmp_pos; 1129 1131 1130 /* 1132 /* 1131 * This is safe to call with 1133 * This is safe to call with the page locks - it won't take 1132 * any additional semaphores 1134 * any additional semaphores or cluster locks. 1133 */ 1135 */ 1134 tmp_pos = cpos; 1136 tmp_pos = cpos; 1135 ret = ocfs2_add_inode_data(OC 1137 ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode, 1136 &t 1138 &tmp_pos, 1, !clear_unwritten, 1137 wc 1139 wc->w_di_bh, wc->w_handle, 1138 da 1140 data_ac, meta_ac, NULL); 1139 /* 1141 /* 1140 * This shouldn't happen beca 1142 * This shouldn't happen because we must have already 1141 * calculated the correct met 1143 * calculated the correct meta data allocation required. The 1142 * internal tree allocation c 1144 * internal tree allocation code should know how to increase 1143 * transaction credits itself 1145 * transaction credits itself. 1144 * 1146 * 1145 * If need be, we could handl 1147 * If need be, we could handle -EAGAIN for a 1146 * RESTART_TRANS here. 1148 * RESTART_TRANS here. 1147 */ 1149 */ 1148 mlog_bug_on_msg(ret == -EAGAI 1150 mlog_bug_on_msg(ret == -EAGAIN, 1149 "Inode %llu: 1151 "Inode %llu: EAGAIN return during allocation.\n", 1150 (unsigned lon 1152 (unsigned long long)OCFS2_I(inode)->ip_blkno); 1151 if (ret < 0) { 1153 if (ret < 0) { 1152 mlog_errno(ret); 1154 mlog_errno(ret); 1153 goto out; 1155 goto out; 1154 } 1156 } 1155 } else if (clear_unwritten) { 1157 } else if (clear_unwritten) { 1156 ocfs2_init_dinode_extent_tree 1158 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), 1157 1159 wc->w_di_bh); 1158 ret = ocfs2_mark_extent_writt 1160 ret = ocfs2_mark_extent_written(inode, &et, 1159 1161 wc->w_handle, cpos, 1, *phys, 1160 1162 meta_ac, &wc->w_dealloc); 1161 if (ret < 0) { 1163 if (ret < 0) { 1162 mlog_errno(ret); 1164 mlog_errno(ret); 1163 goto out; 1165 goto out; 1164 } 1166 } 1165 } 1167 } 1166 1168 1167 /* 1169 /* 1168 * The only reason this should fail i 1170 * The only reason this should fail is due to an inability to 1169 * find the extent added. 1171 * find the extent added. 1170 */ 1172 */ 1171 ret = ocfs2_get_clusters(inode, cpos, 1173 ret = ocfs2_get_clusters(inode, cpos, phys, NULL, NULL); 1172 if (ret < 0) { 1174 if (ret < 0) { 1173 mlog(ML_ERROR, "Get physical 1175 mlog(ML_ERROR, "Get physical blkno failed for inode %llu, " 1174 "at logical clust 1176 "at logical cluster %u", 1175 (unsigned long lo 1177 (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos); 1176 goto out; 1178 goto out; 1177 } 1179 } 1178 1180 1179 BUG_ON(*phys == 0); 1181 BUG_ON(*phys == 0); 1180 1182 1181 p_blkno = ocfs2_clusters_to_blocks(in 1183 p_blkno = ocfs2_clusters_to_blocks(inode->i_sb, *phys); 1182 if (!should_zero) 1184 if (!should_zero) 1183 p_blkno += (user_pos >> inode 1185 p_blkno += (user_pos >> inode->i_sb->s_blocksize_bits) & (u64)(bpc - 1); 1184 1186 1185 for(i = 0; i < wc->w_num_pages; i++) 1187 for(i = 0; i < wc->w_num_pages; i++) { 1186 int tmpret; 1188 int tmpret; 1187 1189 1188 /* This is the direct io targ 1190 /* This is the direct io target page. */ 1189 if (wc->w_pages[i] == NULL) { 1191 if (wc->w_pages[i] == NULL) { 1190 p_blkno++; 1192 p_blkno++; 1191 continue; 1193 continue; 1192 } 1194 } 1193 1195 1194 tmpret = ocfs2_prepare_page_f 1196 tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc, 1195 1197 wc->w_pages[i], cpos, 1196 1198 user_pos, user_len, 1197 1199 should_zero); 1198 if (tmpret) { 1200 if (tmpret) { 1199 mlog_errno(tmpret); 1201 mlog_errno(tmpret); 1200 if (ret == 0) 1202 if (ret == 0) 1201 ret = tmpret; 1203 ret = tmpret; 1202 } 1204 } 1203 } 1205 } 1204 1206 1205 /* 1207 /* 1206 * We only have cleanup to do in case 1208 * We only have cleanup to do in case of allocating write. 1207 */ 1209 */ 1208 if (ret && new) 1210 if (ret && new) 1209 ocfs2_write_failure(inode, wc 1211 ocfs2_write_failure(inode, wc, user_pos, user_len); 1210 1212 1211 out: 1213 out: 1212 1214 1213 return ret; 1215 return ret; 1214 } 1216 } 1215 1217 1216 static int ocfs2_write_cluster_by_desc(struct 1218 static int ocfs2_write_cluster_by_desc(struct address_space *mapping, 1217 struct 1219 struct ocfs2_alloc_context *data_ac, 1218 struct 1220 struct ocfs2_alloc_context *meta_ac, 1219 struct 1221 struct ocfs2_write_ctxt *wc, 1220 loff_t 1222 loff_t pos, unsigned len) 1221 { 1223 { 1222 int ret, i; 1224 int ret, i; 1223 loff_t cluster_off; 1225 loff_t cluster_off; 1224 unsigned int local_len = len; 1226 unsigned int local_len = len; 1225 struct ocfs2_write_cluster_desc *desc 1227 struct ocfs2_write_cluster_desc *desc; 1226 struct ocfs2_super *osb = OCFS2_SB(ma 1228 struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb); 1227 1229 1228 for (i = 0; i < wc->w_clen; i++) { 1230 for (i = 0; i < wc->w_clen; i++) { 1229 desc = &wc->w_desc[i]; 1231 desc = &wc->w_desc[i]; 1230 1232 1231 /* 1233 /* 1232 * We have to make sure that 1234 * We have to make sure that the total write passed in 1233 * doesn't extend past a sing 1235 * doesn't extend past a single cluster. 1234 */ 1236 */ 1235 local_len = len; 1237 local_len = len; 1236 cluster_off = pos & (osb->s_c 1238 cluster_off = pos & (osb->s_clustersize - 1); 1237 if ((cluster_off + local_len) 1239 if ((cluster_off + local_len) > osb->s_clustersize) 1238 local_len = osb->s_cl 1240 local_len = osb->s_clustersize - cluster_off; 1239 1241 1240 ret = ocfs2_write_cluster(map 1242 ret = ocfs2_write_cluster(mapping, &desc->c_phys, 1241 des 1243 desc->c_new, 1242 des 1244 desc->c_clear_unwritten, 1243 des 1245 desc->c_needs_zero, 1244 dat 1246 data_ac, meta_ac, 1245 wc, 1247 wc, desc->c_cpos, pos, local_len); 1246 if (ret) { 1248 if (ret) { 1247 mlog_errno(ret); 1249 mlog_errno(ret); 1248 goto out; 1250 goto out; 1249 } 1251 } 1250 1252 1251 len -= local_len; 1253 len -= local_len; 1252 pos += local_len; 1254 pos += local_len; 1253 } 1255 } 1254 1256 1255 ret = 0; 1257 ret = 0; 1256 out: 1258 out: 1257 return ret; 1259 return ret; 1258 } 1260 } 1259 1261 1260 /* 1262 /* 1261 * ocfs2_write_end() wants to know which part 1263 * ocfs2_write_end() wants to know which parts of the target page it 1262 * should complete the write on. It's easiest 1264 * should complete the write on. It's easiest to compute them ahead of 1263 * time when a more complete view of the writ 1265 * time when a more complete view of the write is available. 1264 */ 1266 */ 1265 static void ocfs2_set_target_boundaries(struc 1267 static void ocfs2_set_target_boundaries(struct ocfs2_super *osb, 1266 struc 1268 struct ocfs2_write_ctxt *wc, 1267 loff_ 1269 loff_t pos, unsigned len, int alloc) 1268 { 1270 { 1269 struct ocfs2_write_cluster_desc *desc 1271 struct ocfs2_write_cluster_desc *desc; 1270 1272 1271 wc->w_target_from = pos & (PAGE_SIZE 1273 wc->w_target_from = pos & (PAGE_SIZE - 1); 1272 wc->w_target_to = wc->w_target_from + 1274 wc->w_target_to = wc->w_target_from + len; 1273 1275 1274 if (alloc == 0) 1276 if (alloc == 0) 1275 return; 1277 return; 1276 1278 1277 /* 1279 /* 1278 * Allocating write - we may have dif 1280 * Allocating write - we may have different boundaries based 1279 * on page size and cluster size. 1281 * on page size and cluster size. 1280 * 1282 * 1281 * NOTE: We can no longer compute one 1283 * NOTE: We can no longer compute one value from the other as 1282 * the actual write length and user p 1284 * the actual write length and user provided length may be 1283 * different. 1285 * different. 1284 */ 1286 */ 1285 1287 1286 if (wc->w_large_pages) { 1288 if (wc->w_large_pages) { 1287 /* 1289 /* 1288 * We only care about the 1st 1290 * We only care about the 1st and last cluster within 1289 * our range and whether they 1291 * our range and whether they should be zero'd or not. Either 1290 * value may be extended out 1292 * value may be extended out to the start/end of a 1291 * newly allocated cluster. 1293 * newly allocated cluster. 1292 */ 1294 */ 1293 desc = &wc->w_desc[0]; 1295 desc = &wc->w_desc[0]; 1294 if (desc->c_needs_zero) 1296 if (desc->c_needs_zero) 1295 ocfs2_figure_cluster_ 1297 ocfs2_figure_cluster_boundaries(osb, 1296 1298 desc->c_cpos, 1297 1299 &wc->w_target_from, 1298 1300 NULL); 1299 1301 1300 desc = &wc->w_desc[wc->w_clen 1302 desc = &wc->w_desc[wc->w_clen - 1]; 1301 if (desc->c_needs_zero) 1303 if (desc->c_needs_zero) 1302 ocfs2_figure_cluster_ 1304 ocfs2_figure_cluster_boundaries(osb, 1303 1305 desc->c_cpos, 1304 1306 NULL, 1305 1307 &wc->w_target_to); 1306 } else { 1308 } else { 1307 wc->w_target_from = 0; 1309 wc->w_target_from = 0; 1308 wc->w_target_to = PAGE_SIZE; 1310 wc->w_target_to = PAGE_SIZE; 1309 } 1311 } 1310 } 1312 } 1311 1313 1312 /* 1314 /* 1313 * Check if this extent is marked UNWRITTEN b 1315 * Check if this extent is marked UNWRITTEN by direct io. If so, we need not to 1314 * do the zero work. And should not to clear 1316 * do the zero work. And should not to clear UNWRITTEN since it will be cleared 1315 * by the direct io procedure. 1317 * by the direct io procedure. 1316 * If this is a new extent that allocated by 1318 * If this is a new extent that allocated by direct io, we should mark it in 1317 * the ip_unwritten_list. 1319 * the ip_unwritten_list. 1318 */ 1320 */ 1319 static int ocfs2_unwritten_check(struct inode 1321 static int ocfs2_unwritten_check(struct inode *inode, 1320 struct ocfs2 1322 struct ocfs2_write_ctxt *wc, 1321 struct ocfs2 1323 struct ocfs2_write_cluster_desc *desc) 1322 { 1324 { 1323 struct ocfs2_inode_info *oi = OCFS2_I 1325 struct ocfs2_inode_info *oi = OCFS2_I(inode); 1324 struct ocfs2_unwritten_extent *ue = N 1326 struct ocfs2_unwritten_extent *ue = NULL, *new = NULL; 1325 int ret = 0; 1327 int ret = 0; 1326 1328 1327 if (!desc->c_needs_zero) 1329 if (!desc->c_needs_zero) 1328 return 0; 1330 return 0; 1329 1331 1330 retry: 1332 retry: 1331 spin_lock(&oi->ip_lock); 1333 spin_lock(&oi->ip_lock); 1332 /* Needs not to zero no metter buffer 1334 /* Needs not to zero no metter buffer or direct. The one who is zero 1333 * the cluster is doing zero. And he 1335 * the cluster is doing zero. And he will clear unwritten after all 1334 * cluster io finished. */ 1336 * cluster io finished. */ 1335 list_for_each_entry(ue, &oi->ip_unwri 1337 list_for_each_entry(ue, &oi->ip_unwritten_list, ue_ip_node) { 1336 if (desc->c_cpos == ue->ue_cp 1338 if (desc->c_cpos == ue->ue_cpos) { 1337 BUG_ON(desc->c_new); 1339 BUG_ON(desc->c_new); 1338 desc->c_needs_zero = 1340 desc->c_needs_zero = 0; 1339 desc->c_clear_unwritt 1341 desc->c_clear_unwritten = 0; 1340 goto unlock; 1342 goto unlock; 1341 } 1343 } 1342 } 1344 } 1343 1345 1344 if (wc->w_type != OCFS2_WRITE_DIRECT) 1346 if (wc->w_type != OCFS2_WRITE_DIRECT) 1345 goto unlock; 1347 goto unlock; 1346 1348 1347 if (new == NULL) { 1349 if (new == NULL) { 1348 spin_unlock(&oi->ip_lock); 1350 spin_unlock(&oi->ip_lock); 1349 new = kmalloc(sizeof(struct o 1351 new = kmalloc(sizeof(struct ocfs2_unwritten_extent), 1350 GFP_NOFS); 1352 GFP_NOFS); 1351 if (new == NULL) { 1353 if (new == NULL) { 1352 ret = -ENOMEM; 1354 ret = -ENOMEM; 1353 goto out; 1355 goto out; 1354 } 1356 } 1355 goto retry; 1357 goto retry; 1356 } 1358 } 1357 /* This direct write will doing zero. 1359 /* This direct write will doing zero. */ 1358 new->ue_cpos = desc->c_cpos; 1360 new->ue_cpos = desc->c_cpos; 1359 new->ue_phys = desc->c_phys; 1361 new->ue_phys = desc->c_phys; 1360 desc->c_clear_unwritten = 0; 1362 desc->c_clear_unwritten = 0; 1361 list_add_tail(&new->ue_ip_node, &oi-> 1363 list_add_tail(&new->ue_ip_node, &oi->ip_unwritten_list); 1362 list_add_tail(&new->ue_node, &wc->w_u 1364 list_add_tail(&new->ue_node, &wc->w_unwritten_list); 1363 wc->w_unwritten_count++; 1365 wc->w_unwritten_count++; 1364 new = NULL; 1366 new = NULL; 1365 unlock: 1367 unlock: 1366 spin_unlock(&oi->ip_lock); 1368 spin_unlock(&oi->ip_lock); 1367 out: 1369 out: 1368 kfree(new); 1370 kfree(new); 1369 return ret; 1371 return ret; 1370 } 1372 } 1371 1373 1372 /* 1374 /* 1373 * Populate each single-cluster write descrip 1375 * Populate each single-cluster write descriptor in the write context 1374 * with information about the i/o to be done. 1376 * with information about the i/o to be done. 1375 * 1377 * 1376 * Returns the number of clusters that will h 1378 * Returns the number of clusters that will have to be allocated, as 1377 * well as a worst case estimate of the numbe 1379 * well as a worst case estimate of the number of extent records that 1378 * would have to be created during a write to 1380 * would have to be created during a write to an unwritten region. 1379 */ 1381 */ 1380 static int ocfs2_populate_write_desc(struct i 1382 static int ocfs2_populate_write_desc(struct inode *inode, 1381 struct o 1383 struct ocfs2_write_ctxt *wc, 1382 unsigned 1384 unsigned int *clusters_to_alloc, 1383 unsigned 1385 unsigned int *extents_to_split) 1384 { 1386 { 1385 int ret; 1387 int ret; 1386 struct ocfs2_write_cluster_desc *desc 1388 struct ocfs2_write_cluster_desc *desc; 1387 unsigned int num_clusters = 0; 1389 unsigned int num_clusters = 0; 1388 unsigned int ext_flags = 0; 1390 unsigned int ext_flags = 0; 1389 u32 phys = 0; 1391 u32 phys = 0; 1390 int i; 1392 int i; 1391 1393 1392 *clusters_to_alloc = 0; 1394 *clusters_to_alloc = 0; 1393 *extents_to_split = 0; 1395 *extents_to_split = 0; 1394 1396 1395 for (i = 0; i < wc->w_clen; i++) { 1397 for (i = 0; i < wc->w_clen; i++) { 1396 desc = &wc->w_desc[i]; 1398 desc = &wc->w_desc[i]; 1397 desc->c_cpos = wc->w_cpos + i 1399 desc->c_cpos = wc->w_cpos + i; 1398 1400 1399 if (num_clusters == 0) { 1401 if (num_clusters == 0) { 1400 /* 1402 /* 1401 * Need to look up th 1403 * Need to look up the next extent record. 1402 */ 1404 */ 1403 ret = ocfs2_get_clust 1405 ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys, 1404 1406 &num_clusters, &ext_flags); 1405 if (ret) { 1407 if (ret) { 1406 mlog_errno(re 1408 mlog_errno(ret); 1407 goto out; 1409 goto out; 1408 } 1410 } 1409 1411 1410 /* We should already 1412 /* We should already CoW the refcountd extent. */ 1411 BUG_ON(ext_flags & OC 1413 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED); 1412 1414 1413 /* 1415 /* 1414 * Assume worst case 1416 * Assume worst case - that we're writing in 1415 * the middle of the 1417 * the middle of the extent. 1416 * 1418 * 1417 * We can assume that 1419 * We can assume that the write proceeds from 1418 * left to right, in 1420 * left to right, in which case the extent 1419 * insert code is sma 1421 * insert code is smart enough to coalesce the 1420 * next splits into t 1422 * next splits into the previous records created. 1421 */ 1423 */ 1422 if (ext_flags & OCFS2 1424 if (ext_flags & OCFS2_EXT_UNWRITTEN) 1423 *extents_to_s 1425 *extents_to_split = *extents_to_split + 2; 1424 } else if (phys) { 1426 } else if (phys) { 1425 /* 1427 /* 1426 * Only increment phy 1428 * Only increment phys if it doesn't describe 1427 * a hole. 1429 * a hole. 1428 */ 1430 */ 1429 phys++; 1431 phys++; 1430 } 1432 } 1431 1433 1432 /* 1434 /* 1433 * If w_first_new_cpos is < U 1435 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse 1434 * file that got extended. w 1436 * file that got extended. w_first_new_cpos tells us 1435 * where the newly allocated 1437 * where the newly allocated clusters are so we can 1436 * zero them. 1438 * zero them. 1437 */ 1439 */ 1438 if (desc->c_cpos >= wc->w_fir 1440 if (desc->c_cpos >= wc->w_first_new_cpos) { 1439 BUG_ON(phys == 0); 1441 BUG_ON(phys == 0); 1440 desc->c_needs_zero = 1442 desc->c_needs_zero = 1; 1441 } 1443 } 1442 1444 1443 desc->c_phys = phys; 1445 desc->c_phys = phys; 1444 if (phys == 0) { 1446 if (phys == 0) { 1445 desc->c_new = 1; 1447 desc->c_new = 1; 1446 desc->c_needs_zero = 1448 desc->c_needs_zero = 1; 1447 desc->c_clear_unwritt 1449 desc->c_clear_unwritten = 1; 1448 *clusters_to_alloc = 1450 *clusters_to_alloc = *clusters_to_alloc + 1; 1449 } 1451 } 1450 1452 1451 if (ext_flags & OCFS2_EXT_UNW 1453 if (ext_flags & OCFS2_EXT_UNWRITTEN) { 1452 desc->c_clear_unwritt 1454 desc->c_clear_unwritten = 1; 1453 desc->c_needs_zero = 1455 desc->c_needs_zero = 1; 1454 } 1456 } 1455 1457 1456 ret = ocfs2_unwritten_check(i 1458 ret = ocfs2_unwritten_check(inode, wc, desc); 1457 if (ret) { 1459 if (ret) { 1458 mlog_errno(ret); 1460 mlog_errno(ret); 1459 goto out; 1461 goto out; 1460 } 1462 } 1461 1463 1462 num_clusters--; 1464 num_clusters--; 1463 } 1465 } 1464 1466 1465 ret = 0; 1467 ret = 0; 1466 out: 1468 out: 1467 return ret; 1469 return ret; 1468 } 1470 } 1469 1471 1470 static int ocfs2_write_begin_inline(struct ad 1472 static int ocfs2_write_begin_inline(struct address_space *mapping, 1471 struct in 1473 struct inode *inode, 1472 struct oc 1474 struct ocfs2_write_ctxt *wc) 1473 { 1475 { 1474 int ret; 1476 int ret; 1475 struct ocfs2_super *osb = OCFS2_SB(in 1477 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 1476 struct page *page; 1478 struct page *page; 1477 handle_t *handle; 1479 handle_t *handle; 1478 struct ocfs2_dinode *di = (struct ocf 1480 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; 1479 1481 1480 handle = ocfs2_start_trans(osb, OCFS2 1482 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); 1481 if (IS_ERR(handle)) { 1483 if (IS_ERR(handle)) { 1482 ret = PTR_ERR(handle); 1484 ret = PTR_ERR(handle); 1483 mlog_errno(ret); 1485 mlog_errno(ret); 1484 goto out; 1486 goto out; 1485 } 1487 } 1486 1488 1487 page = find_or_create_page(mapping, 0 1489 page = find_or_create_page(mapping, 0, GFP_NOFS); 1488 if (!page) { 1490 if (!page) { 1489 ocfs2_commit_trans(osb, handl 1491 ocfs2_commit_trans(osb, handle); 1490 ret = -ENOMEM; 1492 ret = -ENOMEM; 1491 mlog_errno(ret); 1493 mlog_errno(ret); 1492 goto out; 1494 goto out; 1493 } 1495 } 1494 /* 1496 /* 1495 * If we don't set w_num_pages then t 1497 * If we don't set w_num_pages then this page won't get unlocked 1496 * and freed on cleanup of the write 1498 * and freed on cleanup of the write context. 1497 */ 1499 */ 1498 wc->w_pages[0] = wc->w_target_page = 1500 wc->w_pages[0] = wc->w_target_page = page; 1499 wc->w_num_pages = 1; 1501 wc->w_num_pages = 1; 1500 1502 1501 ret = ocfs2_journal_access_di(handle, 1503 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh, 1502 OCFS2_J 1504 OCFS2_JOURNAL_ACCESS_WRITE); 1503 if (ret) { 1505 if (ret) { 1504 ocfs2_commit_trans(osb, handl 1506 ocfs2_commit_trans(osb, handle); 1505 1507 1506 mlog_errno(ret); 1508 mlog_errno(ret); 1507 goto out; 1509 goto out; 1508 } 1510 } 1509 1511 1510 if (!(OCFS2_I(inode)->ip_dyn_features 1512 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)) 1511 ocfs2_set_inode_data_inline(i 1513 ocfs2_set_inode_data_inline(inode, di); 1512 1514 1513 if (!PageUptodate(page)) { 1515 if (!PageUptodate(page)) { 1514 ret = ocfs2_read_inline_data( 1516 ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh); 1515 if (ret) { 1517 if (ret) { 1516 ocfs2_commit_trans(os 1518 ocfs2_commit_trans(osb, handle); 1517 1519 1518 goto out; 1520 goto out; 1519 } 1521 } 1520 } 1522 } 1521 1523 1522 wc->w_handle = handle; 1524 wc->w_handle = handle; 1523 out: 1525 out: 1524 return ret; 1526 return ret; 1525 } 1527 } 1526 1528 1527 int ocfs2_size_fits_inline_data(struct buffer 1529 int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size) 1528 { 1530 { 1529 struct ocfs2_dinode *di = (struct ocf 1531 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; 1530 1532 1531 if (new_size <= le16_to_cpu(di->id2.i 1533 if (new_size <= le16_to_cpu(di->id2.i_data.id_count)) 1532 return 1; 1534 return 1; 1533 return 0; 1535 return 0; 1534 } 1536 } 1535 1537 1536 static int ocfs2_try_to_write_inline_data(str 1538 static int ocfs2_try_to_write_inline_data(struct address_space *mapping, 1537 str 1539 struct inode *inode, loff_t pos, 1538 uns 1540 unsigned len, struct page *mmap_page, 1539 str 1541 struct ocfs2_write_ctxt *wc) 1540 { 1542 { 1541 int ret, written = 0; 1543 int ret, written = 0; 1542 loff_t end = pos + len; 1544 loff_t end = pos + len; 1543 struct ocfs2_inode_info *oi = OCFS2_I 1545 struct ocfs2_inode_info *oi = OCFS2_I(inode); 1544 struct ocfs2_dinode *di = NULL; 1546 struct ocfs2_dinode *di = NULL; 1545 1547 1546 trace_ocfs2_try_to_write_inline_data( 1548 trace_ocfs2_try_to_write_inline_data((unsigned long long)oi->ip_blkno, 1547 1549 len, (unsigned long long)pos, 1548 1550 oi->ip_dyn_features); 1549 1551 1550 /* 1552 /* 1551 * Handle inodes which already have i 1553 * Handle inodes which already have inline data 1st. 1552 */ 1554 */ 1553 if (oi->ip_dyn_features & OCFS2_INLIN 1555 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) { 1554 if (mmap_page == NULL && 1556 if (mmap_page == NULL && 1555 ocfs2_size_fits_inline_da 1557 ocfs2_size_fits_inline_data(wc->w_di_bh, end)) 1556 goto do_inline_write; 1558 goto do_inline_write; 1557 1559 1558 /* 1560 /* 1559 * The write won't fit - we h 1561 * The write won't fit - we have to give this inode an 1560 * inline extent list now. 1562 * inline extent list now. 1561 */ 1563 */ 1562 ret = ocfs2_convert_inline_da 1564 ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh); 1563 if (ret) 1565 if (ret) 1564 mlog_errno(ret); 1566 mlog_errno(ret); 1565 goto out; 1567 goto out; 1566 } 1568 } 1567 1569 1568 /* 1570 /* 1569 * Check whether the inode can accept 1571 * Check whether the inode can accept inline data. 1570 */ 1572 */ 1571 if (oi->ip_clusters != 0 || i_size_re 1573 if (oi->ip_clusters != 0 || i_size_read(inode) != 0) 1572 return 0; 1574 return 0; 1573 1575 1574 /* 1576 /* 1575 * Check whether the write can fit. 1577 * Check whether the write can fit. 1576 */ 1578 */ 1577 di = (struct ocfs2_dinode *)wc->w_di_ 1579 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; 1578 if (mmap_page || 1580 if (mmap_page || 1579 end > ocfs2_max_inline_data_with_ 1581 end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) 1580 return 0; 1582 return 0; 1581 1583 1582 do_inline_write: 1584 do_inline_write: 1583 ret = ocfs2_write_begin_inline(mappin 1585 ret = ocfs2_write_begin_inline(mapping, inode, wc); 1584 if (ret) { 1586 if (ret) { 1585 mlog_errno(ret); 1587 mlog_errno(ret); 1586 goto out; 1588 goto out; 1587 } 1589 } 1588 1590 1589 /* 1591 /* 1590 * This signals to the caller that th 1592 * This signals to the caller that the data can be written 1591 * inline. 1593 * inline. 1592 */ 1594 */ 1593 written = 1; 1595 written = 1; 1594 out: 1596 out: 1595 return written ? written : ret; 1597 return written ? written : ret; 1596 } 1598 } 1597 1599 1598 /* 1600 /* 1599 * This function only does anything for file 1601 * This function only does anything for file systems which can't 1600 * handle sparse files. 1602 * handle sparse files. 1601 * 1603 * 1602 * What we want to do here is fill in any hol 1604 * What we want to do here is fill in any hole between the current end 1603 * of allocation and the end of our write. Th 1605 * of allocation and the end of our write. That way the rest of the 1604 * write path can treat it as an non-allocati 1606 * write path can treat it as an non-allocating write, which has no 1605 * special case code for sparse/nonsparse fil 1607 * special case code for sparse/nonsparse files. 1606 */ 1608 */ 1607 static int ocfs2_expand_nonsparse_inode(struc 1609 static int ocfs2_expand_nonsparse_inode(struct inode *inode, 1608 struc 1610 struct buffer_head *di_bh, 1609 loff_ 1611 loff_t pos, unsigned len, 1610 struc 1612 struct ocfs2_write_ctxt *wc) 1611 { 1613 { 1612 int ret; 1614 int ret; 1613 loff_t newsize = pos + len; 1615 loff_t newsize = pos + len; 1614 1616 1615 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(in 1617 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))); 1616 1618 1617 if (newsize <= i_size_read(inode)) 1619 if (newsize <= i_size_read(inode)) 1618 return 0; 1620 return 0; 1619 1621 1620 ret = ocfs2_extend_no_holes(inode, di 1622 ret = ocfs2_extend_no_holes(inode, di_bh, newsize, pos); 1621 if (ret) 1623 if (ret) 1622 mlog_errno(ret); 1624 mlog_errno(ret); 1623 1625 1624 /* There is no wc if this is call fro 1626 /* There is no wc if this is call from direct. */ 1625 if (wc) 1627 if (wc) 1626 wc->w_first_new_cpos = 1628 wc->w_first_new_cpos = 1627 ocfs2_clusters_for_by 1629 ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode)); 1628 1630 1629 return ret; 1631 return ret; 1630 } 1632 } 1631 1633 1632 static int ocfs2_zero_tail(struct inode *inod 1634 static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh, 1633 loff_t pos) 1635 loff_t pos) 1634 { 1636 { 1635 int ret = 0; 1637 int ret = 0; 1636 1638 1637 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(i 1639 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))); 1638 if (pos > i_size_read(inode)) 1640 if (pos > i_size_read(inode)) 1639 ret = ocfs2_zero_extend(inode 1641 ret = ocfs2_zero_extend(inode, di_bh, pos); 1640 1642 1641 return ret; 1643 return ret; 1642 } 1644 } 1643 1645 1644 int ocfs2_write_begin_nolock(struct address_s 1646 int ocfs2_write_begin_nolock(struct address_space *mapping, 1645 loff_t pos, unsi 1647 loff_t pos, unsigned len, ocfs2_write_type_t type, 1646 struct page **pa 1648 struct page **pagep, void **fsdata, 1647 struct buffer_he 1649 struct buffer_head *di_bh, struct page *mmap_page) 1648 { 1650 { 1649 int ret, cluster_of_pages, credits = 1651 int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS; 1650 unsigned int clusters_to_alloc, exten 1652 unsigned int clusters_to_alloc, extents_to_split, clusters_need = 0; 1651 struct ocfs2_write_ctxt *wc; 1653 struct ocfs2_write_ctxt *wc; 1652 struct inode *inode = mapping->host; 1654 struct inode *inode = mapping->host; 1653 struct ocfs2_super *osb = OCFS2_SB(in 1655 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 1654 struct ocfs2_dinode *di; 1656 struct ocfs2_dinode *di; 1655 struct ocfs2_alloc_context *data_ac = 1657 struct ocfs2_alloc_context *data_ac = NULL; 1656 struct ocfs2_alloc_context *meta_ac = 1658 struct ocfs2_alloc_context *meta_ac = NULL; 1657 handle_t *handle; 1659 handle_t *handle; 1658 struct ocfs2_extent_tree et; 1660 struct ocfs2_extent_tree et; 1659 int try_free = 1, ret1; 1661 int try_free = 1, ret1; 1660 1662 1661 try_again: 1663 try_again: 1662 ret = ocfs2_alloc_write_ctxt(&wc, osb 1664 ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, type, di_bh); 1663 if (ret) { 1665 if (ret) { 1664 mlog_errno(ret); 1666 mlog_errno(ret); 1665 return ret; 1667 return ret; 1666 } 1668 } 1667 1669 1668 if (ocfs2_supports_inline_data(osb)) 1670 if (ocfs2_supports_inline_data(osb)) { 1669 ret = ocfs2_try_to_write_inli 1671 ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len, 1670 1672 mmap_page, wc); 1671 if (ret == 1) { 1673 if (ret == 1) { 1672 ret = 0; 1674 ret = 0; 1673 goto success; 1675 goto success; 1674 } 1676 } 1675 if (ret < 0) { 1677 if (ret < 0) { 1676 mlog_errno(ret); 1678 mlog_errno(ret); 1677 goto out; 1679 goto out; 1678 } 1680 } 1679 } 1681 } 1680 1682 1681 /* Direct io change i_size late, shou 1683 /* Direct io change i_size late, should not zero tail here. */ 1682 if (type != OCFS2_WRITE_DIRECT) { 1684 if (type != OCFS2_WRITE_DIRECT) { 1683 if (ocfs2_sparse_alloc(osb)) 1685 if (ocfs2_sparse_alloc(osb)) 1684 ret = ocfs2_zero_tail 1686 ret = ocfs2_zero_tail(inode, di_bh, pos); 1685 else 1687 else 1686 ret = ocfs2_expand_no 1688 ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos, 1687 1689 len, wc); 1688 if (ret) { 1690 if (ret) { 1689 mlog_errno(ret); 1691 mlog_errno(ret); 1690 goto out; 1692 goto out; 1691 } 1693 } 1692 } 1694 } 1693 1695 1694 ret = ocfs2_check_range_for_refcount( 1696 ret = ocfs2_check_range_for_refcount(inode, pos, len); 1695 if (ret < 0) { 1697 if (ret < 0) { 1696 mlog_errno(ret); 1698 mlog_errno(ret); 1697 goto out; 1699 goto out; 1698 } else if (ret == 1) { 1700 } else if (ret == 1) { 1699 clusters_need = wc->w_clen; 1701 clusters_need = wc->w_clen; 1700 ret = ocfs2_refcount_cow(inod 1702 ret = ocfs2_refcount_cow(inode, di_bh, 1701 wc-> 1703 wc->w_cpos, wc->w_clen, UINT_MAX); 1702 if (ret) { 1704 if (ret) { 1703 mlog_errno(ret); 1705 mlog_errno(ret); 1704 goto out; 1706 goto out; 1705 } 1707 } 1706 } 1708 } 1707 1709 1708 ret = ocfs2_populate_write_desc(inode 1710 ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc, 1709 &exte 1711 &extents_to_split); 1710 if (ret) { 1712 if (ret) { 1711 mlog_errno(ret); 1713 mlog_errno(ret); 1712 goto out; 1714 goto out; 1713 } 1715 } 1714 clusters_need += clusters_to_alloc; 1716 clusters_need += clusters_to_alloc; 1715 1717 1716 di = (struct ocfs2_dinode *)wc->w_di_ 1718 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; 1717 1719 1718 trace_ocfs2_write_begin_nolock( 1720 trace_ocfs2_write_begin_nolock( 1719 (unsigned long long)O 1721 (unsigned long long)OCFS2_I(inode)->ip_blkno, 1720 (long long)i_size_rea 1722 (long long)i_size_read(inode), 1721 le32_to_cpu(di->i_clu 1723 le32_to_cpu(di->i_clusters), 1722 pos, len, type, mmap_ 1724 pos, len, type, mmap_page, 1723 clusters_to_alloc, ex 1725 clusters_to_alloc, extents_to_split); 1724 1726 1725 /* 1727 /* 1726 * We set w_target_from, w_target_to 1728 * We set w_target_from, w_target_to here so that 1727 * ocfs2_write_end() knows which rang 1729 * ocfs2_write_end() knows which range in the target page to 1728 * write out. An allocation requires 1730 * write out. An allocation requires that we write the entire 1729 * cluster range. 1731 * cluster range. 1730 */ 1732 */ 1731 if (clusters_to_alloc || extents_to_s 1733 if (clusters_to_alloc || extents_to_split) { 1732 /* 1734 /* 1733 * XXX: We are stretching the 1735 * XXX: We are stretching the limits of 1734 * ocfs2_lock_allocators(). I 1736 * ocfs2_lock_allocators(). It greatly over-estimates 1735 * the work to be done. 1737 * the work to be done. 1736 */ 1738 */ 1737 ocfs2_init_dinode_extent_tree 1739 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), 1738 1740 wc->w_di_bh); 1739 ret = ocfs2_lock_allocators(i 1741 ret = ocfs2_lock_allocators(inode, &et, 1740 c 1742 clusters_to_alloc, extents_to_split, 1741 & 1743 &data_ac, &meta_ac); 1742 if (ret) { 1744 if (ret) { 1743 mlog_errno(ret); 1745 mlog_errno(ret); 1744 goto out; 1746 goto out; 1745 } 1747 } 1746 1748 1747 if (data_ac) 1749 if (data_ac) 1748 data_ac->ac_resv = &O 1750 data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv; 1749 1751 1750 credits = ocfs2_calc_extend_c 1752 credits = ocfs2_calc_extend_credits(inode->i_sb, 1751 1753 &di->id2.i_list); 1752 } else if (type == OCFS2_WRITE_DIRECT 1754 } else if (type == OCFS2_WRITE_DIRECT) 1753 /* direct write needs not to 1755 /* direct write needs not to start trans if no extents alloc. */ 1754 goto success; 1756 goto success; 1755 1757 1756 /* 1758 /* 1757 * We have to zero sparse allocated c 1759 * We have to zero sparse allocated clusters, unwritten extent clusters, 1758 * and non-sparse clusters we just ex 1760 * and non-sparse clusters we just extended. For non-sparse writes, 1759 * we know zeros will only be needed 1761 * we know zeros will only be needed in the first and/or last cluster. 1760 */ 1762 */ 1761 if (wc->w_clen && (wc->w_desc[0].c_ne 1763 if (wc->w_clen && (wc->w_desc[0].c_needs_zero || 1762 wc->w_desc[wc->w_c 1764 wc->w_desc[wc->w_clen - 1].c_needs_zero)) 1763 cluster_of_pages = 1; 1765 cluster_of_pages = 1; 1764 else 1766 else 1765 cluster_of_pages = 0; 1767 cluster_of_pages = 0; 1766 1768 1767 ocfs2_set_target_boundaries(osb, wc, 1769 ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages); 1768 1770 1769 handle = ocfs2_start_trans(osb, credi 1771 handle = ocfs2_start_trans(osb, credits); 1770 if (IS_ERR(handle)) { 1772 if (IS_ERR(handle)) { 1771 ret = PTR_ERR(handle); 1773 ret = PTR_ERR(handle); 1772 mlog_errno(ret); 1774 mlog_errno(ret); 1773 goto out; 1775 goto out; 1774 } 1776 } 1775 1777 1776 wc->w_handle = handle; 1778 wc->w_handle = handle; 1777 1779 1778 if (clusters_to_alloc) { 1780 if (clusters_to_alloc) { 1779 ret = dquot_alloc_space_nodir 1781 ret = dquot_alloc_space_nodirty(inode, 1780 ocfs2_clusters_to_byt 1782 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc)); 1781 if (ret) 1783 if (ret) 1782 goto out_commit; 1784 goto out_commit; 1783 } 1785 } 1784 1786 1785 ret = ocfs2_journal_access_di(handle, 1787 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh, 1786 OCFS2_J 1788 OCFS2_JOURNAL_ACCESS_WRITE); 1787 if (ret) { 1789 if (ret) { 1788 mlog_errno(ret); 1790 mlog_errno(ret); 1789 goto out_quota; 1791 goto out_quota; 1790 } 1792 } 1791 1793 1792 /* 1794 /* 1793 * Fill our page array first. That wa 1795 * Fill our page array first. That way we've grabbed enough so 1794 * that we can zero and flush if we e 1796 * that we can zero and flush if we error after adding the 1795 * extent. 1797 * extent. 1796 */ 1798 */ 1797 ret = ocfs2_grab_pages_for_write(mapp 1799 ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, len, 1798 clus 1800 cluster_of_pages, mmap_page); 1799 if (ret) { 1801 if (ret) { 1800 /* 1802 /* 1801 * ocfs2_grab_pages_for_write 1803 * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock 1802 * the target page. In this c 1804 * the target page. In this case, we exit with no error and no target 1803 * page. This will trigger th 1805 * page. This will trigger the caller, page_mkwrite(), to re-try 1804 * the operation. 1806 * the operation. 1805 */ 1807 */ 1806 if (type == OCFS2_WRITE_MMAP 1808 if (type == OCFS2_WRITE_MMAP && ret == -EAGAIN) { 1807 BUG_ON(wc->w_target_p 1809 BUG_ON(wc->w_target_page); 1808 ret = 0; 1810 ret = 0; 1809 goto out_quota; 1811 goto out_quota; 1810 } 1812 } 1811 1813 1812 mlog_errno(ret); 1814 mlog_errno(ret); 1813 goto out_quota; 1815 goto out_quota; 1814 } 1816 } 1815 1817 1816 ret = ocfs2_write_cluster_by_desc(map 1818 ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos, 1817 len 1819 len); 1818 if (ret) { 1820 if (ret) { 1819 mlog_errno(ret); 1821 mlog_errno(ret); 1820 goto out_quota; 1822 goto out_quota; 1821 } 1823 } 1822 1824 1823 if (data_ac) 1825 if (data_ac) 1824 ocfs2_free_alloc_context(data 1826 ocfs2_free_alloc_context(data_ac); 1825 if (meta_ac) 1827 if (meta_ac) 1826 ocfs2_free_alloc_context(meta 1828 ocfs2_free_alloc_context(meta_ac); 1827 1829 1828 success: 1830 success: 1829 if (pagep) 1831 if (pagep) 1830 *pagep = wc->w_target_page; 1832 *pagep = wc->w_target_page; 1831 *fsdata = wc; 1833 *fsdata = wc; 1832 return 0; 1834 return 0; 1833 out_quota: 1835 out_quota: 1834 if (clusters_to_alloc) 1836 if (clusters_to_alloc) 1835 dquot_free_space(inode, 1837 dquot_free_space(inode, 1836 ocfs2_clusters_to_b 1838 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc)); 1837 out_commit: 1839 out_commit: 1838 ocfs2_commit_trans(osb, handle); 1840 ocfs2_commit_trans(osb, handle); 1839 1841 1840 out: 1842 out: 1841 /* 1843 /* 1842 * The mmapped page won't be unlocked 1844 * The mmapped page won't be unlocked in ocfs2_free_write_ctxt(), 1843 * even in case of error here like EN 1845 * even in case of error here like ENOSPC and ENOMEM. So, we need 1844 * to unlock the target page manually 1846 * to unlock the target page manually to prevent deadlocks when 1845 * retrying again on ENOSPC, or when 1847 * retrying again on ENOSPC, or when returning non-VM_FAULT_LOCKED 1846 * to VM code. 1848 * to VM code. 1847 */ 1849 */ 1848 if (wc->w_target_locked) 1850 if (wc->w_target_locked) 1849 unlock_page(mmap_page); 1851 unlock_page(mmap_page); 1850 1852 1851 ocfs2_free_write_ctxt(inode, wc); 1853 ocfs2_free_write_ctxt(inode, wc); 1852 1854 1853 if (data_ac) { 1855 if (data_ac) { 1854 ocfs2_free_alloc_context(data 1856 ocfs2_free_alloc_context(data_ac); 1855 data_ac = NULL; 1857 data_ac = NULL; 1856 } 1858 } 1857 if (meta_ac) { 1859 if (meta_ac) { 1858 ocfs2_free_alloc_context(meta 1860 ocfs2_free_alloc_context(meta_ac); 1859 meta_ac = NULL; 1861 meta_ac = NULL; 1860 } 1862 } 1861 1863 1862 if (ret == -ENOSPC && try_free) { 1864 if (ret == -ENOSPC && try_free) { 1863 /* 1865 /* 1864 * Try to free some truncate 1866 * Try to free some truncate log so that we can have enough 1865 * clusters to allocate. 1867 * clusters to allocate. 1866 */ 1868 */ 1867 try_free = 0; 1869 try_free = 0; 1868 1870 1869 ret1 = ocfs2_try_to_free_trun 1871 ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need); 1870 if (ret1 == 1) 1872 if (ret1 == 1) 1871 goto try_again; 1873 goto try_again; 1872 1874 1873 if (ret1 < 0) 1875 if (ret1 < 0) 1874 mlog_errno(ret1); 1876 mlog_errno(ret1); 1875 } 1877 } 1876 1878 1877 return ret; 1879 return ret; 1878 } 1880 } 1879 1881 1880 static int ocfs2_write_begin(struct file *fil 1882 static int ocfs2_write_begin(struct file *file, struct address_space *mapping, 1881 loff_t pos, unsi 1883 loff_t pos, unsigned len, 1882 struct page **pa 1884 struct page **pagep, void **fsdata) 1883 { 1885 { 1884 int ret; 1886 int ret; 1885 struct buffer_head *di_bh = NULL; 1887 struct buffer_head *di_bh = NULL; 1886 struct inode *inode = mapping->host; 1888 struct inode *inode = mapping->host; 1887 1889 1888 ret = ocfs2_inode_lock(inode, &di_bh, 1890 ret = ocfs2_inode_lock(inode, &di_bh, 1); 1889 if (ret) { 1891 if (ret) { 1890 mlog_errno(ret); 1892 mlog_errno(ret); 1891 return ret; 1893 return ret; 1892 } 1894 } 1893 1895 1894 /* 1896 /* 1895 * Take alloc sem here to prevent con 1897 * Take alloc sem here to prevent concurrent lookups. That way 1896 * the mapping, zeroing and tree mani 1898 * the mapping, zeroing and tree manipulation within 1897 * ocfs2_write() will be safe against 1899 * ocfs2_write() will be safe against ->read_folio(). This 1898 * should also serve to lock out allo 1900 * should also serve to lock out allocation from a shared 1899 * writeable region. 1901 * writeable region. 1900 */ 1902 */ 1901 down_write(&OCFS2_I(inode)->ip_alloc_ 1903 down_write(&OCFS2_I(inode)->ip_alloc_sem); 1902 1904 1903 ret = ocfs2_write_begin_nolock(mappin 1905 ret = ocfs2_write_begin_nolock(mapping, pos, len, OCFS2_WRITE_BUFFER, 1904 pagep, 1906 pagep, fsdata, di_bh, NULL); 1905 if (ret) { 1907 if (ret) { 1906 mlog_errno(ret); 1908 mlog_errno(ret); 1907 goto out_fail; 1909 goto out_fail; 1908 } 1910 } 1909 1911 1910 brelse(di_bh); 1912 brelse(di_bh); 1911 1913 1912 return 0; 1914 return 0; 1913 1915 1914 out_fail: 1916 out_fail: 1915 up_write(&OCFS2_I(inode)->ip_alloc_se 1917 up_write(&OCFS2_I(inode)->ip_alloc_sem); 1916 1918 1917 brelse(di_bh); 1919 brelse(di_bh); 1918 ocfs2_inode_unlock(inode, 1); 1920 ocfs2_inode_unlock(inode, 1); 1919 1921 1920 return ret; 1922 return ret; 1921 } 1923 } 1922 1924 1923 static void ocfs2_write_end_inline(struct ino 1925 static void ocfs2_write_end_inline(struct inode *inode, loff_t pos, 1924 unsigned l 1926 unsigned len, unsigned *copied, 1925 struct ocf 1927 struct ocfs2_dinode *di, 1926 struct ocf 1928 struct ocfs2_write_ctxt *wc) 1927 { 1929 { 1928 void *kaddr; 1930 void *kaddr; 1929 1931 1930 if (unlikely(*copied < len)) { 1932 if (unlikely(*copied < len)) { 1931 if (!PageUptodate(wc->w_targe 1933 if (!PageUptodate(wc->w_target_page)) { 1932 *copied = 0; 1934 *copied = 0; 1933 return; 1935 return; 1934 } 1936 } 1935 } 1937 } 1936 1938 1937 kaddr = kmap_atomic(wc->w_target_page 1939 kaddr = kmap_atomic(wc->w_target_page); 1938 memcpy(di->id2.i_data.id_data + pos, 1940 memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied); 1939 kunmap_atomic(kaddr); 1941 kunmap_atomic(kaddr); 1940 1942 1941 trace_ocfs2_write_end_inline( 1943 trace_ocfs2_write_end_inline( 1942 (unsigned long long)OCFS2_I(inod 1944 (unsigned long long)OCFS2_I(inode)->ip_blkno, 1943 (unsigned long long)pos, *copied 1945 (unsigned long long)pos, *copied, 1944 le16_to_cpu(di->id2.i_data.id_co 1946 le16_to_cpu(di->id2.i_data.id_count), 1945 le16_to_cpu(di->i_dyn_features)) 1947 le16_to_cpu(di->i_dyn_features)); 1946 } 1948 } 1947 1949 1948 int ocfs2_write_end_nolock(struct address_spa 1950 int ocfs2_write_end_nolock(struct address_space *mapping, 1949 loff_t pos, unsign 1951 loff_t pos, unsigned len, unsigned copied, void *fsdata) 1950 { 1952 { 1951 int i, ret; 1953 int i, ret; 1952 unsigned from, to, start = pos & (PAG 1954 unsigned from, to, start = pos & (PAGE_SIZE - 1); 1953 struct inode *inode = mapping->host; 1955 struct inode *inode = mapping->host; 1954 struct ocfs2_super *osb = OCFS2_SB(in 1956 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 1955 struct ocfs2_write_ctxt *wc = fsdata; 1957 struct ocfs2_write_ctxt *wc = fsdata; 1956 struct ocfs2_dinode *di = (struct ocf 1958 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; 1957 handle_t *handle = wc->w_handle; 1959 handle_t *handle = wc->w_handle; 1958 struct page *tmppage; 1960 struct page *tmppage; 1959 1961 1960 BUG_ON(!list_empty(&wc->w_unwritten_l 1962 BUG_ON(!list_empty(&wc->w_unwritten_list)); 1961 1963 1962 if (handle) { 1964 if (handle) { 1963 ret = ocfs2_journal_access_di 1965 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), 1964 wc->w_di_bh, 1966 wc->w_di_bh, OCFS2_JOURNAL_ACCESS_WRITE); 1965 if (ret) { 1967 if (ret) { 1966 copied = ret; 1968 copied = ret; 1967 mlog_errno(ret); 1969 mlog_errno(ret); 1968 goto out; 1970 goto out; 1969 } 1971 } 1970 } 1972 } 1971 1973 1972 if (OCFS2_I(inode)->ip_dyn_features & 1974 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { 1973 ocfs2_write_end_inline(inode, 1975 ocfs2_write_end_inline(inode, pos, len, &copied, di, wc); 1974 goto out_write_size; 1976 goto out_write_size; 1975 } 1977 } 1976 1978 1977 if (unlikely(copied < len) && wc->w_t 1979 if (unlikely(copied < len) && wc->w_target_page) { 1978 loff_t new_isize; << 1979 << 1980 if (!PageUptodate(wc->w_targe 1980 if (!PageUptodate(wc->w_target_page)) 1981 copied = 0; 1981 copied = 0; 1982 1982 1983 new_isize = max_t(loff_t, i_s !! 1983 ocfs2_zero_new_buffers(wc->w_target_page, start+copied, 1984 if (new_isize > page_offset(w !! 1984 start+len); 1985 ocfs2_zero_new_buffer << 1986 << 1987 else { << 1988 /* << 1989 * When page is fully << 1990 * failed), do not bo << 1991 * it instead so that << 1992 * put page & buffer << 1993 * state. << 1994 */ << 1995 block_invalidate_foli << 1996 << 1997 } << 1998 } 1985 } 1999 if (wc->w_target_page) 1986 if (wc->w_target_page) 2000 flush_dcache_page(wc->w_targe 1987 flush_dcache_page(wc->w_target_page); 2001 1988 2002 for(i = 0; i < wc->w_num_pages; i++) 1989 for(i = 0; i < wc->w_num_pages; i++) { 2003 tmppage = wc->w_pages[i]; 1990 tmppage = wc->w_pages[i]; 2004 1991 2005 /* This is the direct io targ 1992 /* This is the direct io target page. */ 2006 if (tmppage == NULL) 1993 if (tmppage == NULL) 2007 continue; 1994 continue; 2008 1995 2009 if (tmppage == wc->w_target_p 1996 if (tmppage == wc->w_target_page) { 2010 from = wc->w_target_f 1997 from = wc->w_target_from; 2011 to = wc->w_target_to; 1998 to = wc->w_target_to; 2012 1999 2013 BUG_ON(from > PAGE_SI 2000 BUG_ON(from > PAGE_SIZE || 2014 to > PAGE_SIZE 2001 to > PAGE_SIZE || 2015 to < from); 2002 to < from); 2016 } else { 2003 } else { 2017 /* 2004 /* 2018 * Pages adjacent to 2005 * Pages adjacent to the target (if any) imply 2019 * a hole-filling wri 2006 * a hole-filling write in which case we want 2020 * to flush their ent 2007 * to flush their entire range. 2021 */ 2008 */ 2022 from = 0; 2009 from = 0; 2023 to = PAGE_SIZE; 2010 to = PAGE_SIZE; 2024 } 2011 } 2025 2012 2026 if (page_has_buffers(tmppage) 2013 if (page_has_buffers(tmppage)) { 2027 if (handle && ocfs2_s 2014 if (handle && ocfs2_should_order_data(inode)) { 2028 loff_t start_ 2015 loff_t start_byte = 2029 ((lof 2016 ((loff_t)tmppage->index << PAGE_SHIFT) + 2030 from; 2017 from; 2031 loff_t length 2018 loff_t length = to - from; 2032 ocfs2_jbd2_in 2019 ocfs2_jbd2_inode_add_write(handle, inode, 2033 2020 start_byte, length); 2034 } 2021 } 2035 block_commit_write(tm 2022 block_commit_write(tmppage, from, to); 2036 } 2023 } 2037 } 2024 } 2038 2025 2039 out_write_size: 2026 out_write_size: 2040 /* Direct io do not update i_size her 2027 /* Direct io do not update i_size here. */ 2041 if (wc->w_type != OCFS2_WRITE_DIRECT) 2028 if (wc->w_type != OCFS2_WRITE_DIRECT) { 2042 pos += copied; 2029 pos += copied; 2043 if (pos > i_size_read(inode)) 2030 if (pos > i_size_read(inode)) { 2044 i_size_write(inode, p 2031 i_size_write(inode, pos); 2045 mark_inode_dirty(inod 2032 mark_inode_dirty(inode); 2046 } 2033 } 2047 inode->i_blocks = ocfs2_inode 2034 inode->i_blocks = ocfs2_inode_sector_count(inode); 2048 di->i_size = cpu_to_le64((u64 2035 di->i_size = cpu_to_le64((u64)i_size_read(inode)); 2049 inode_set_mtime_to_ts(inode, !! 2036 inode->i_mtime = inode->i_ctime = current_time(inode); 2050 di->i_mtime = di->i_ctime = c !! 2037 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec); 2051 di->i_mtime_nsec = di->i_ctim !! 2038 di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); 2052 if (handle) 2039 if (handle) 2053 ocfs2_update_inode_fs 2040 ocfs2_update_inode_fsync_trans(handle, inode, 1); 2054 } 2041 } 2055 if (handle) 2042 if (handle) 2056 ocfs2_journal_dirty(handle, w 2043 ocfs2_journal_dirty(handle, wc->w_di_bh); 2057 2044 2058 out: 2045 out: 2059 /* unlock pages before dealloc since 2046 /* unlock pages before dealloc since it needs acquiring j_trans_barrier 2060 * lock, or it will cause a deadlock 2047 * lock, or it will cause a deadlock since journal commit threads holds 2061 * this lock and will ask for the pag 2048 * this lock and will ask for the page lock when flushing the data. 2062 * put it here to preserve the unlock 2049 * put it here to preserve the unlock order. 2063 */ 2050 */ 2064 ocfs2_unlock_pages(wc); 2051 ocfs2_unlock_pages(wc); 2065 2052 2066 if (handle) 2053 if (handle) 2067 ocfs2_commit_trans(osb, handl 2054 ocfs2_commit_trans(osb, handle); 2068 2055 2069 ocfs2_run_deallocs(osb, &wc->w_deallo 2056 ocfs2_run_deallocs(osb, &wc->w_dealloc); 2070 2057 2071 brelse(wc->w_di_bh); 2058 brelse(wc->w_di_bh); 2072 kfree(wc); 2059 kfree(wc); 2073 2060 2074 return copied; 2061 return copied; 2075 } 2062 } 2076 2063 2077 static int ocfs2_write_end(struct file *file, 2064 static int ocfs2_write_end(struct file *file, struct address_space *mapping, 2078 loff_t pos, unsign 2065 loff_t pos, unsigned len, unsigned copied, 2079 struct page *page, 2066 struct page *page, void *fsdata) 2080 { 2067 { 2081 int ret; 2068 int ret; 2082 struct inode *inode = mapping->host; 2069 struct inode *inode = mapping->host; 2083 2070 2084 ret = ocfs2_write_end_nolock(mapping, 2071 ret = ocfs2_write_end_nolock(mapping, pos, len, copied, fsdata); 2085 2072 2086 up_write(&OCFS2_I(inode)->ip_alloc_se 2073 up_write(&OCFS2_I(inode)->ip_alloc_sem); 2087 ocfs2_inode_unlock(inode, 1); 2074 ocfs2_inode_unlock(inode, 1); 2088 2075 2089 return ret; 2076 return ret; 2090 } 2077 } 2091 2078 2092 struct ocfs2_dio_write_ctxt { 2079 struct ocfs2_dio_write_ctxt { 2093 struct list_head dw_zero_list; 2080 struct list_head dw_zero_list; 2094 unsigned dw_zero_count 2081 unsigned dw_zero_count; 2095 int dw_orphaned; 2082 int dw_orphaned; 2096 pid_t dw_writer_pid 2083 pid_t dw_writer_pid; 2097 }; 2084 }; 2098 2085 2099 static struct ocfs2_dio_write_ctxt * 2086 static struct ocfs2_dio_write_ctxt * 2100 ocfs2_dio_alloc_write_ctx(struct buffer_head 2087 ocfs2_dio_alloc_write_ctx(struct buffer_head *bh, int *alloc) 2101 { 2088 { 2102 struct ocfs2_dio_write_ctxt *dwc = NU 2089 struct ocfs2_dio_write_ctxt *dwc = NULL; 2103 2090 2104 if (bh->b_private) 2091 if (bh->b_private) 2105 return bh->b_private; 2092 return bh->b_private; 2106 2093 2107 dwc = kmalloc(sizeof(struct ocfs2_dio 2094 dwc = kmalloc(sizeof(struct ocfs2_dio_write_ctxt), GFP_NOFS); 2108 if (dwc == NULL) 2095 if (dwc == NULL) 2109 return NULL; 2096 return NULL; 2110 INIT_LIST_HEAD(&dwc->dw_zero_list); 2097 INIT_LIST_HEAD(&dwc->dw_zero_list); 2111 dwc->dw_zero_count = 0; 2098 dwc->dw_zero_count = 0; 2112 dwc->dw_orphaned = 0; 2099 dwc->dw_orphaned = 0; 2113 dwc->dw_writer_pid = task_pid_nr(curr 2100 dwc->dw_writer_pid = task_pid_nr(current); 2114 bh->b_private = dwc; 2101 bh->b_private = dwc; 2115 *alloc = 1; 2102 *alloc = 1; 2116 2103 2117 return dwc; 2104 return dwc; 2118 } 2105 } 2119 2106 2120 static void ocfs2_dio_free_write_ctx(struct i 2107 static void ocfs2_dio_free_write_ctx(struct inode *inode, 2121 struct o 2108 struct ocfs2_dio_write_ctxt *dwc) 2122 { 2109 { 2123 ocfs2_free_unwritten_list(inode, &dwc 2110 ocfs2_free_unwritten_list(inode, &dwc->dw_zero_list); 2124 kfree(dwc); 2111 kfree(dwc); 2125 } 2112 } 2126 2113 2127 /* 2114 /* 2128 * TODO: Make this into a generic get_blocks 2115 * TODO: Make this into a generic get_blocks function. 2129 * 2116 * 2130 * From do_direct_io in direct-io.c: 2117 * From do_direct_io in direct-io.c: 2131 * "So what we do is to permit the ->get_blo 2118 * "So what we do is to permit the ->get_blocks function to populate 2132 * bh.b_size with the size of IO which is p 2119 * bh.b_size with the size of IO which is permitted at this offset and 2133 * this i_blkbits." 2120 * this i_blkbits." 2134 * 2121 * 2135 * This function is called directly from get_ 2122 * This function is called directly from get_more_blocks in direct-io.c. 2136 * 2123 * 2137 * called like this: dio->get_blocks(dio->ino 2124 * called like this: dio->get_blocks(dio->inode, fs_startblk, 2138 * fs_co 2125 * fs_count, map_bh, dio->rw == WRITE); 2139 */ 2126 */ 2140 static int ocfs2_dio_wr_get_block(struct inod 2127 static int ocfs2_dio_wr_get_block(struct inode *inode, sector_t iblock, 2141 struct buffer_ 2128 struct buffer_head *bh_result, int create) 2142 { 2129 { 2143 struct ocfs2_super *osb = OCFS2_SB(in 2130 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 2144 struct ocfs2_inode_info *oi = OCFS2_I 2131 struct ocfs2_inode_info *oi = OCFS2_I(inode); 2145 struct ocfs2_write_ctxt *wc; 2132 struct ocfs2_write_ctxt *wc; 2146 struct ocfs2_write_cluster_desc *desc 2133 struct ocfs2_write_cluster_desc *desc = NULL; 2147 struct ocfs2_dio_write_ctxt *dwc = NU 2134 struct ocfs2_dio_write_ctxt *dwc = NULL; 2148 struct buffer_head *di_bh = NULL; 2135 struct buffer_head *di_bh = NULL; 2149 u64 p_blkno; 2136 u64 p_blkno; 2150 unsigned int i_blkbits = inode->i_sb- 2137 unsigned int i_blkbits = inode->i_sb->s_blocksize_bits; 2151 loff_t pos = iblock << i_blkbits; 2138 loff_t pos = iblock << i_blkbits; 2152 sector_t endblk = (i_size_read(inode) 2139 sector_t endblk = (i_size_read(inode) - 1) >> i_blkbits; 2153 unsigned len, total_len = bh_result-> 2140 unsigned len, total_len = bh_result->b_size; 2154 int ret = 0, first_get_block = 0; 2141 int ret = 0, first_get_block = 0; 2155 2142 2156 len = osb->s_clustersize - (pos & (os 2143 len = osb->s_clustersize - (pos & (osb->s_clustersize - 1)); 2157 len = min(total_len, len); 2144 len = min(total_len, len); 2158 2145 2159 /* 2146 /* 2160 * bh_result->b_size is count in get_ 2147 * bh_result->b_size is count in get_more_blocks according to write 2161 * "pos" and "end", we need map twice 2148 * "pos" and "end", we need map twice to return different buffer state: 2162 * 1. area in file size, not set NEW; 2149 * 1. area in file size, not set NEW; 2163 * 2. area out file size, set NEW. 2150 * 2. area out file size, set NEW. 2164 * 2151 * 2165 * iblock endblk 2152 * iblock endblk 2166 * |--------|---------|---------|---- 2153 * |--------|---------|---------|--------- 2167 * |<-------area in file------->| 2154 * |<-------area in file------->| 2168 */ 2155 */ 2169 2156 2170 if ((iblock <= endblk) && 2157 if ((iblock <= endblk) && 2171 ((iblock + ((len - 1) >> i_blkbit 2158 ((iblock + ((len - 1) >> i_blkbits)) > endblk)) 2172 len = (endblk - iblock + 1) < 2159 len = (endblk - iblock + 1) << i_blkbits; 2173 2160 2174 mlog(0, "get block of %lu at %llu:%u 2161 mlog(0, "get block of %lu at %llu:%u req %u\n", 2175 inode->i_ino, pos, le 2162 inode->i_ino, pos, len, total_len); 2176 2163 2177 /* 2164 /* 2178 * Because we need to change file siz 2165 * Because we need to change file size in ocfs2_dio_end_io_write(), or 2179 * we may need to add it to orphan di 2166 * we may need to add it to orphan dir. So can not fall to fast path 2180 * while file size will be changed. 2167 * while file size will be changed. 2181 */ 2168 */ 2182 if (pos + total_len <= i_size_read(in 2169 if (pos + total_len <= i_size_read(inode)) { 2183 2170 2184 /* This is the fast path for 2171 /* This is the fast path for re-write. */ 2185 ret = ocfs2_lock_get_block(in 2172 ret = ocfs2_lock_get_block(inode, iblock, bh_result, create); 2186 if (buffer_mapped(bh_result) 2173 if (buffer_mapped(bh_result) && 2187 !buffer_new(bh_result) && 2174 !buffer_new(bh_result) && 2188 ret == 0) 2175 ret == 0) 2189 goto out; 2176 goto out; 2190 2177 2191 /* Clear state set by ocfs2_g 2178 /* Clear state set by ocfs2_get_block. */ 2192 bh_result->b_state = 0; 2179 bh_result->b_state = 0; 2193 } 2180 } 2194 2181 2195 dwc = ocfs2_dio_alloc_write_ctx(bh_re 2182 dwc = ocfs2_dio_alloc_write_ctx(bh_result, &first_get_block); 2196 if (unlikely(dwc == NULL)) { 2183 if (unlikely(dwc == NULL)) { 2197 ret = -ENOMEM; 2184 ret = -ENOMEM; 2198 mlog_errno(ret); 2185 mlog_errno(ret); 2199 goto out; 2186 goto out; 2200 } 2187 } 2201 2188 2202 if (ocfs2_clusters_for_bytes(inode->i 2189 if (ocfs2_clusters_for_bytes(inode->i_sb, pos + total_len) > 2203 ocfs2_clusters_for_bytes(inode->i 2190 ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode)) && 2204 !dwc->dw_orphaned) { 2191 !dwc->dw_orphaned) { 2205 /* 2192 /* 2206 * when we are going to alloc 2193 * when we are going to alloc extents beyond file size, add the 2207 * inode to orphan dir, so we 2194 * inode to orphan dir, so we can recall those spaces when 2208 * system crashed during writ 2195 * system crashed during write. 2209 */ 2196 */ 2210 ret = ocfs2_add_inode_to_orph 2197 ret = ocfs2_add_inode_to_orphan(osb, inode); 2211 if (ret < 0) { 2198 if (ret < 0) { 2212 mlog_errno(ret); 2199 mlog_errno(ret); 2213 goto out; 2200 goto out; 2214 } 2201 } 2215 dwc->dw_orphaned = 1; 2202 dwc->dw_orphaned = 1; 2216 } 2203 } 2217 2204 2218 ret = ocfs2_inode_lock(inode, &di_bh, 2205 ret = ocfs2_inode_lock(inode, &di_bh, 1); 2219 if (ret) { 2206 if (ret) { 2220 mlog_errno(ret); 2207 mlog_errno(ret); 2221 goto out; 2208 goto out; 2222 } 2209 } 2223 2210 2224 down_write(&oi->ip_alloc_sem); 2211 down_write(&oi->ip_alloc_sem); 2225 2212 2226 if (first_get_block) { 2213 if (first_get_block) { 2227 if (ocfs2_sparse_alloc(osb)) 2214 if (ocfs2_sparse_alloc(osb)) 2228 ret = ocfs2_zero_tail 2215 ret = ocfs2_zero_tail(inode, di_bh, pos); 2229 else 2216 else 2230 ret = ocfs2_expand_no 2217 ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos, 2231 2218 total_len, NULL); 2232 if (ret < 0) { 2219 if (ret < 0) { 2233 mlog_errno(ret); 2220 mlog_errno(ret); 2234 goto unlock; 2221 goto unlock; 2235 } 2222 } 2236 } 2223 } 2237 2224 2238 ret = ocfs2_write_begin_nolock(inode- 2225 ret = ocfs2_write_begin_nolock(inode->i_mapping, pos, len, 2239 OCFS2_ 2226 OCFS2_WRITE_DIRECT, NULL, 2240 (void 2227 (void **)&wc, di_bh, NULL); 2241 if (ret) { 2228 if (ret) { 2242 mlog_errno(ret); 2229 mlog_errno(ret); 2243 goto unlock; 2230 goto unlock; 2244 } 2231 } 2245 2232 2246 desc = &wc->w_desc[0]; 2233 desc = &wc->w_desc[0]; 2247 2234 2248 p_blkno = ocfs2_clusters_to_blocks(in 2235 p_blkno = ocfs2_clusters_to_blocks(inode->i_sb, desc->c_phys); 2249 BUG_ON(p_blkno == 0); 2236 BUG_ON(p_blkno == 0); 2250 p_blkno += iblock & (u64)(ocfs2_clust 2237 p_blkno += iblock & (u64)(ocfs2_clusters_to_blocks(inode->i_sb, 1) - 1); 2251 2238 2252 map_bh(bh_result, inode->i_sb, p_blkn 2239 map_bh(bh_result, inode->i_sb, p_blkno); 2253 bh_result->b_size = len; 2240 bh_result->b_size = len; 2254 if (desc->c_needs_zero) 2241 if (desc->c_needs_zero) 2255 set_buffer_new(bh_result); 2242 set_buffer_new(bh_result); 2256 2243 2257 if (iblock > endblk) 2244 if (iblock > endblk) 2258 set_buffer_new(bh_result); 2245 set_buffer_new(bh_result); 2259 2246 2260 /* May sleep in end_io. It should not 2247 /* May sleep in end_io. It should not happen in a irq context. So defer 2261 * it to dio work queue. */ 2248 * it to dio work queue. */ 2262 set_buffer_defer_completion(bh_result 2249 set_buffer_defer_completion(bh_result); 2263 2250 2264 if (!list_empty(&wc->w_unwritten_list 2251 if (!list_empty(&wc->w_unwritten_list)) { 2265 struct ocfs2_unwritten_extent 2252 struct ocfs2_unwritten_extent *ue = NULL; 2266 2253 2267 ue = list_first_entry(&wc->w_ 2254 ue = list_first_entry(&wc->w_unwritten_list, 2268 struct 2255 struct ocfs2_unwritten_extent, 2269 ue_node 2256 ue_node); 2270 BUG_ON(ue->ue_cpos != desc->c 2257 BUG_ON(ue->ue_cpos != desc->c_cpos); 2271 /* The physical address may b 2258 /* The physical address may be 0, fill it. */ 2272 ue->ue_phys = desc->c_phys; 2259 ue->ue_phys = desc->c_phys; 2273 2260 2274 list_splice_tail_init(&wc->w_ 2261 list_splice_tail_init(&wc->w_unwritten_list, &dwc->dw_zero_list); 2275 dwc->dw_zero_count += wc->w_u 2262 dwc->dw_zero_count += wc->w_unwritten_count; 2276 } 2263 } 2277 2264 2278 ret = ocfs2_write_end_nolock(inode->i 2265 ret = ocfs2_write_end_nolock(inode->i_mapping, pos, len, len, wc); 2279 BUG_ON(ret != len); 2266 BUG_ON(ret != len); 2280 ret = 0; 2267 ret = 0; 2281 unlock: 2268 unlock: 2282 up_write(&oi->ip_alloc_sem); 2269 up_write(&oi->ip_alloc_sem); 2283 ocfs2_inode_unlock(inode, 1); 2270 ocfs2_inode_unlock(inode, 1); 2284 brelse(di_bh); 2271 brelse(di_bh); 2285 out: 2272 out: >> 2273 if (ret < 0) >> 2274 ret = -EIO; 2286 return ret; 2275 return ret; 2287 } 2276 } 2288 2277 2289 static int ocfs2_dio_end_io_write(struct inod 2278 static int ocfs2_dio_end_io_write(struct inode *inode, 2290 struct ocfs 2279 struct ocfs2_dio_write_ctxt *dwc, 2291 loff_t offs 2280 loff_t offset, 2292 ssize_t byt 2281 ssize_t bytes) 2293 { 2282 { 2294 struct ocfs2_cached_dealloc_ctxt deal 2283 struct ocfs2_cached_dealloc_ctxt dealloc; 2295 struct ocfs2_extent_tree et; 2284 struct ocfs2_extent_tree et; 2296 struct ocfs2_super *osb = OCFS2_SB(in 2285 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 2297 struct ocfs2_inode_info *oi = OCFS2_I 2286 struct ocfs2_inode_info *oi = OCFS2_I(inode); 2298 struct ocfs2_unwritten_extent *ue = N 2287 struct ocfs2_unwritten_extent *ue = NULL; 2299 struct buffer_head *di_bh = NULL; 2288 struct buffer_head *di_bh = NULL; 2300 struct ocfs2_dinode *di; 2289 struct ocfs2_dinode *di; 2301 struct ocfs2_alloc_context *data_ac = 2290 struct ocfs2_alloc_context *data_ac = NULL; 2302 struct ocfs2_alloc_context *meta_ac = 2291 struct ocfs2_alloc_context *meta_ac = NULL; 2303 handle_t *handle = NULL; 2292 handle_t *handle = NULL; 2304 loff_t end = offset + bytes; 2293 loff_t end = offset + bytes; 2305 int ret = 0, credits = 0; 2294 int ret = 0, credits = 0; 2306 2295 2307 ocfs2_init_dealloc_ctxt(&dealloc); 2296 ocfs2_init_dealloc_ctxt(&dealloc); 2308 2297 2309 /* We do clear unwritten, delete orph 2298 /* We do clear unwritten, delete orphan, change i_size here. If neither 2310 * of these happen, we can skip all t 2299 * of these happen, we can skip all this. */ 2311 if (list_empty(&dwc->dw_zero_list) && 2300 if (list_empty(&dwc->dw_zero_list) && 2312 end <= i_size_read(inode) && 2301 end <= i_size_read(inode) && 2313 !dwc->dw_orphaned) 2302 !dwc->dw_orphaned) 2314 goto out; 2303 goto out; 2315 2304 2316 ret = ocfs2_inode_lock(inode, &di_bh, 2305 ret = ocfs2_inode_lock(inode, &di_bh, 1); 2317 if (ret < 0) { 2306 if (ret < 0) { 2318 mlog_errno(ret); 2307 mlog_errno(ret); 2319 goto out; 2308 goto out; 2320 } 2309 } 2321 2310 2322 down_write(&oi->ip_alloc_sem); 2311 down_write(&oi->ip_alloc_sem); 2323 2312 2324 /* Delete orphan before acquire i_rws 2313 /* Delete orphan before acquire i_rwsem. */ 2325 if (dwc->dw_orphaned) { 2314 if (dwc->dw_orphaned) { 2326 BUG_ON(dwc->dw_writer_pid != 2315 BUG_ON(dwc->dw_writer_pid != task_pid_nr(current)); 2327 2316 2328 end = end > i_size_read(inode 2317 end = end > i_size_read(inode) ? end : 0; 2329 2318 2330 ret = ocfs2_del_inode_from_or 2319 ret = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 2331 !!end, end); 2320 !!end, end); 2332 if (ret < 0) 2321 if (ret < 0) 2333 mlog_errno(ret); 2322 mlog_errno(ret); 2334 } 2323 } 2335 2324 2336 di = (struct ocfs2_dinode *)di_bh->b_ 2325 di = (struct ocfs2_dinode *)di_bh->b_data; 2337 2326 2338 ocfs2_init_dinode_extent_tree(&et, IN 2327 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh); 2339 2328 2340 /* Attach dealloc with extent tree in 2329 /* Attach dealloc with extent tree in case that we may reuse extents 2341 * which are already unlinked from cu 2330 * which are already unlinked from current extent tree due to extent 2342 * rotation and merging. 2331 * rotation and merging. 2343 */ 2332 */ 2344 et.et_dealloc = &dealloc; 2333 et.et_dealloc = &dealloc; 2345 2334 2346 ret = ocfs2_lock_allocators(inode, &e 2335 ret = ocfs2_lock_allocators(inode, &et, 0, dwc->dw_zero_count*2, 2347 &data_ac, 2336 &data_ac, &meta_ac); 2348 if (ret) { 2337 if (ret) { 2349 mlog_errno(ret); 2338 mlog_errno(ret); 2350 goto unlock; 2339 goto unlock; 2351 } 2340 } 2352 2341 2353 credits = ocfs2_calc_extend_credits(i 2342 credits = ocfs2_calc_extend_credits(inode->i_sb, &di->id2.i_list); 2354 2343 2355 handle = ocfs2_start_trans(osb, credi 2344 handle = ocfs2_start_trans(osb, credits); 2356 if (IS_ERR(handle)) { 2345 if (IS_ERR(handle)) { 2357 ret = PTR_ERR(handle); 2346 ret = PTR_ERR(handle); 2358 mlog_errno(ret); 2347 mlog_errno(ret); 2359 goto unlock; 2348 goto unlock; 2360 } 2349 } 2361 ret = ocfs2_journal_access_di(handle, 2350 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, 2362 OCFS2_J 2351 OCFS2_JOURNAL_ACCESS_WRITE); 2363 if (ret) { 2352 if (ret) { 2364 mlog_errno(ret); 2353 mlog_errno(ret); 2365 goto commit; 2354 goto commit; 2366 } 2355 } 2367 2356 2368 list_for_each_entry(ue, &dwc->dw_zero 2357 list_for_each_entry(ue, &dwc->dw_zero_list, ue_node) { 2369 ret = ocfs2_assure_trans_cred << 2370 if (ret < 0) { << 2371 mlog_errno(ret); << 2372 break; << 2373 } << 2374 ret = ocfs2_mark_extent_writt 2358 ret = ocfs2_mark_extent_written(inode, &et, handle, 2375 2359 ue->ue_cpos, 1, 2376 2360 ue->ue_phys, 2377 2361 meta_ac, &dealloc); 2378 if (ret < 0) { 2362 if (ret < 0) { 2379 mlog_errno(ret); 2363 mlog_errno(ret); 2380 break; 2364 break; 2381 } 2365 } 2382 } 2366 } 2383 2367 2384 if (end > i_size_read(inode)) { 2368 if (end > i_size_read(inode)) { 2385 ret = ocfs2_set_inode_size(ha 2369 ret = ocfs2_set_inode_size(handle, inode, di_bh, end); 2386 if (ret < 0) 2370 if (ret < 0) 2387 mlog_errno(ret); 2371 mlog_errno(ret); 2388 } 2372 } 2389 commit: 2373 commit: 2390 ocfs2_commit_trans(osb, handle); 2374 ocfs2_commit_trans(osb, handle); 2391 unlock: 2375 unlock: 2392 up_write(&oi->ip_alloc_sem); 2376 up_write(&oi->ip_alloc_sem); 2393 ocfs2_inode_unlock(inode, 1); 2377 ocfs2_inode_unlock(inode, 1); 2394 brelse(di_bh); 2378 brelse(di_bh); 2395 out: 2379 out: 2396 if (data_ac) 2380 if (data_ac) 2397 ocfs2_free_alloc_context(data 2381 ocfs2_free_alloc_context(data_ac); 2398 if (meta_ac) 2382 if (meta_ac) 2399 ocfs2_free_alloc_context(meta 2383 ocfs2_free_alloc_context(meta_ac); 2400 ocfs2_run_deallocs(osb, &dealloc); 2384 ocfs2_run_deallocs(osb, &dealloc); 2401 ocfs2_dio_free_write_ctx(inode, dwc); 2385 ocfs2_dio_free_write_ctx(inode, dwc); 2402 2386 2403 return ret; 2387 return ret; 2404 } 2388 } 2405 2389 2406 /* 2390 /* 2407 * ocfs2_dio_end_io is called by the dio core 2391 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're 2408 * particularly interested in the aio/dio cas 2392 * particularly interested in the aio/dio case. We use the rw_lock DLM lock 2409 * to protect io on one node from truncation 2393 * to protect io on one node from truncation on another. 2410 */ 2394 */ 2411 static int ocfs2_dio_end_io(struct kiocb *ioc 2395 static int ocfs2_dio_end_io(struct kiocb *iocb, 2412 loff_t offset, 2396 loff_t offset, 2413 ssize_t bytes, 2397 ssize_t bytes, 2414 void *private) 2398 void *private) 2415 { 2399 { 2416 struct inode *inode = file_inode(iocb 2400 struct inode *inode = file_inode(iocb->ki_filp); 2417 int level; 2401 int level; 2418 int ret = 0; 2402 int ret = 0; 2419 2403 2420 /* this io's submitter should not hav 2404 /* this io's submitter should not have unlocked this before we could */ 2421 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb) 2405 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb)); 2422 2406 2423 if (bytes <= 0) 2407 if (bytes <= 0) 2424 mlog_ratelimited(ML_ERROR, "D 2408 mlog_ratelimited(ML_ERROR, "Direct IO failed, bytes = %lld", 2425 (long long)b 2409 (long long)bytes); 2426 if (private) { 2410 if (private) { 2427 if (bytes > 0) 2411 if (bytes > 0) 2428 ret = ocfs2_dio_end_i 2412 ret = ocfs2_dio_end_io_write(inode, private, offset, 2429 2413 bytes); 2430 else 2414 else 2431 ocfs2_dio_free_write_ 2415 ocfs2_dio_free_write_ctx(inode, private); 2432 } 2416 } 2433 2417 2434 ocfs2_iocb_clear_rw_locked(iocb); 2418 ocfs2_iocb_clear_rw_locked(iocb); 2435 2419 2436 level = ocfs2_iocb_rw_locked_level(io 2420 level = ocfs2_iocb_rw_locked_level(iocb); 2437 ocfs2_rw_unlock(inode, level); 2421 ocfs2_rw_unlock(inode, level); 2438 return ret; 2422 return ret; 2439 } 2423 } 2440 2424 2441 static ssize_t ocfs2_direct_IO(struct kiocb * 2425 static ssize_t ocfs2_direct_IO(struct kiocb *iocb, struct iov_iter *iter) 2442 { 2426 { 2443 struct file *file = iocb->ki_filp; 2427 struct file *file = iocb->ki_filp; 2444 struct inode *inode = file->f_mapping 2428 struct inode *inode = file->f_mapping->host; 2445 struct ocfs2_super *osb = OCFS2_SB(in 2429 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); 2446 get_block_t *get_block; 2430 get_block_t *get_block; 2447 2431 2448 /* 2432 /* 2449 * Fallback to buffered I/O if we see 2433 * Fallback to buffered I/O if we see an inode without 2450 * extents. 2434 * extents. 2451 */ 2435 */ 2452 if (OCFS2_I(inode)->ip_dyn_features & 2436 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) 2453 return 0; 2437 return 0; 2454 2438 2455 /* Fallback to buffered I/O if we do 2439 /* Fallback to buffered I/O if we do not support append dio. */ 2456 if (iocb->ki_pos + iter->count > i_si 2440 if (iocb->ki_pos + iter->count > i_size_read(inode) && 2457 !ocfs2_supports_append_dio(osb)) 2441 !ocfs2_supports_append_dio(osb)) 2458 return 0; 2442 return 0; 2459 2443 2460 if (iov_iter_rw(iter) == READ) 2444 if (iov_iter_rw(iter) == READ) 2461 get_block = ocfs2_lock_get_bl 2445 get_block = ocfs2_lock_get_block; 2462 else 2446 else 2463 get_block = ocfs2_dio_wr_get_ 2447 get_block = ocfs2_dio_wr_get_block; 2464 2448 2465 return __blockdev_direct_IO(iocb, ino 2449 return __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, 2466 iter, get 2450 iter, get_block, 2467 ocfs2_dio !! 2451 ocfs2_dio_end_io, NULL, 0); 2468 } 2452 } 2469 2453 2470 const struct address_space_operations ocfs2_a 2454 const struct address_space_operations ocfs2_aops = { 2471 .dirty_folio = block_dirty 2455 .dirty_folio = block_dirty_folio, 2472 .read_folio = ocfs2_read_ 2456 .read_folio = ocfs2_read_folio, 2473 .readahead = ocfs2_reada 2457 .readahead = ocfs2_readahead, 2474 .writepages = ocfs2_write !! 2458 .writepage = ocfs2_writepage, 2475 .write_begin = ocfs2_write 2459 .write_begin = ocfs2_write_begin, 2476 .write_end = ocfs2_write 2460 .write_end = ocfs2_write_end, 2477 .bmap = ocfs2_bmap, 2461 .bmap = ocfs2_bmap, 2478 .direct_IO = ocfs2_direc 2462 .direct_IO = ocfs2_direct_IO, 2479 .invalidate_folio = block_inval 2463 .invalidate_folio = block_invalidate_folio, 2480 .release_folio = ocfs2_relea 2464 .release_folio = ocfs2_release_folio, 2481 .migrate_folio = buffer_migr 2465 .migrate_folio = buffer_migrate_folio, 2482 .is_partially_uptodate = block_is_pa 2466 .is_partially_uptodate = block_is_partially_uptodate, 2483 .error_remove_folio = generic_err !! 2467 .error_remove_page = generic_error_remove_page, 2484 }; 2468 }; 2485 2469
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.