Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2008 Oracle. All rights reserved. |
| 3 | * |
| 4 | * This program is free software; you can redistribute it and/or |
| 5 | * modify it under the terms of the GNU General Public |
| 6 | * License v2 as published by the Free Software Foundation. |
| 7 | * |
| 8 | * This program is distributed in the hope that it will be useful, |
| 9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 11 | * General Public License for more details. |
| 12 | * |
| 13 | * You should have received a copy of the GNU General Public |
| 14 | * License along with this program; if not, write to the |
| 15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, |
| 16 | * Boston, MA 021110-1307, USA. |
| 17 | */ |
| 18 | |
| 19 | #include <linux/kernel.h> |
| 20 | #include <linux/bio.h> |
| 21 | #include <linux/buffer_head.h> |
| 22 | #include <linux/file.h> |
| 23 | #include <linux/fs.h> |
| 24 | #include <linux/pagemap.h> |
| 25 | #include <linux/highmem.h> |
| 26 | #include <linux/time.h> |
| 27 | #include <linux/init.h> |
| 28 | #include <linux/string.h> |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 29 | #include <linux/backing-dev.h> |
| 30 | #include <linux/mpage.h> |
| 31 | #include <linux/swap.h> |
| 32 | #include <linux/writeback.h> |
| 33 | #include <linux/bit_spinlock.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 34 | #include <linux/slab.h> |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 35 | #include "ctree.h" |
| 36 | #include "disk-io.h" |
| 37 | #include "transaction.h" |
| 38 | #include "btrfs_inode.h" |
| 39 | #include "volumes.h" |
| 40 | #include "ordered-data.h" |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 41 | #include "compression.h" |
| 42 | #include "extent_io.h" |
| 43 | #include "extent_map.h" |
| 44 | |
| 45 | struct compressed_bio { |
| 46 | /* number of bios pending for this compressed extent */ |
| 47 | atomic_t pending_bios; |
| 48 | |
| 49 | /* the pages with the compressed data on them */ |
| 50 | struct page **compressed_pages; |
| 51 | |
| 52 | /* inode that owns this data */ |
| 53 | struct inode *inode; |
| 54 | |
| 55 | /* starting offset in the inode for our pages */ |
| 56 | u64 start; |
| 57 | |
| 58 | /* number of bytes in the inode we're working on */ |
| 59 | unsigned long len; |
| 60 | |
| 61 | /* number of bytes on disk */ |
| 62 | unsigned long compressed_len; |
| 63 | |
Li Zefan | 261507a0 | 2010-12-17 14:21:50 +0800 | [diff] [blame] | 64 | /* the compression algorithm for this bio */ |
| 65 | int compress_type; |
| 66 | |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 67 | /* number of compressed pages in the array */ |
| 68 | unsigned long nr_pages; |
| 69 | |
| 70 | /* IO errors */ |
| 71 | int errors; |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 72 | int mirror_num; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 73 | |
| 74 | /* for reads, this is the bio we are copying the data into */ |
| 75 | struct bio *orig_bio; |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 76 | |
| 77 | /* |
| 78 | * the start of a variable length array of checksums only |
| 79 | * used by reads |
| 80 | */ |
| 81 | u32 sums; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 82 | }; |
| 83 | |
Eric Sandeen | 48a3b63 | 2013-04-25 20:41:01 +0000 | [diff] [blame] | 84 | static int btrfs_decompress_biovec(int type, struct page **pages_in, |
| 85 | u64 disk_start, struct bio_vec *bvec, |
| 86 | int vcnt, size_t srclen); |
| 87 | |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 88 | static inline int compressed_bio_size(struct btrfs_root *root, |
| 89 | unsigned long disk_size) |
| 90 | { |
David Sterba | 6c41761 | 2011-04-13 15:41:04 +0200 | [diff] [blame] | 91 | u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy); |
| 92 | |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 93 | return sizeof(struct compressed_bio) + |
| 94 | ((disk_size + root->sectorsize - 1) / root->sectorsize) * |
| 95 | csum_size; |
| 96 | } |
| 97 | |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 98 | static struct bio *compressed_bio_alloc(struct block_device *bdev, |
| 99 | u64 first_byte, gfp_t gfp_flags) |
| 100 | { |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 101 | int nr_vecs; |
| 102 | |
| 103 | nr_vecs = bio_get_nr_vecs(bdev); |
Miao Xie | 88f794e | 2010-11-22 03:02:55 +0000 | [diff] [blame] | 104 | return btrfs_bio_alloc(bdev, first_byte >> 9, nr_vecs, gfp_flags); |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 105 | } |
| 106 | |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 107 | static int check_compressed_csum(struct inode *inode, |
| 108 | struct compressed_bio *cb, |
| 109 | u64 disk_start) |
| 110 | { |
| 111 | int ret; |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 112 | struct page *page; |
| 113 | unsigned long i; |
| 114 | char *kaddr; |
| 115 | u32 csum; |
| 116 | u32 *cb_sum = &cb->sums; |
| 117 | |
Christoph Hellwig | 6cbff00 | 2009-04-17 10:37:41 +0200 | [diff] [blame] | 118 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 119 | return 0; |
| 120 | |
| 121 | for (i = 0; i < cb->nr_pages; i++) { |
| 122 | page = cb->compressed_pages[i]; |
| 123 | csum = ~(u32)0; |
| 124 | |
Cong Wang | 7ac687d | 2011-11-25 23:14:28 +0800 | [diff] [blame] | 125 | kaddr = kmap_atomic(page); |
Liu Bo | b049668 | 2013-03-14 14:57:45 +0000 | [diff] [blame] | 126 | csum = btrfs_csum_data(kaddr, csum, PAGE_CACHE_SIZE); |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 127 | btrfs_csum_final(csum, (char *)&csum); |
Cong Wang | 7ac687d | 2011-11-25 23:14:28 +0800 | [diff] [blame] | 128 | kunmap_atomic(kaddr); |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 129 | |
| 130 | if (csum != *cb_sum) { |
Frank Holton | efe120a | 2013-12-20 11:37:06 -0500 | [diff] [blame] | 131 | btrfs_info(BTRFS_I(inode)->root->fs_info, |
| 132 | "csum failed ino %llu extent %llu csum %u wanted %u mirror %d", |
| 133 | btrfs_ino(inode), disk_start, csum, *cb_sum, |
| 134 | cb->mirror_num); |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 135 | ret = -EIO; |
| 136 | goto fail; |
| 137 | } |
| 138 | cb_sum++; |
| 139 | |
| 140 | } |
| 141 | ret = 0; |
| 142 | fail: |
| 143 | return ret; |
| 144 | } |
| 145 | |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 146 | /* when we finish reading compressed pages from the disk, we |
| 147 | * decompress them and then run the bio end_io routines on the |
| 148 | * decompressed pages (in the inode address space). |
| 149 | * |
| 150 | * This allows the checksumming and other IO error handling routines |
| 151 | * to work normally |
| 152 | * |
| 153 | * The compressed pages are freed here, and it must be run |
| 154 | * in process context |
| 155 | */ |
| 156 | static void end_compressed_bio_read(struct bio *bio, int err) |
| 157 | { |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 158 | struct compressed_bio *cb = bio->bi_private; |
| 159 | struct inode *inode; |
| 160 | struct page *page; |
| 161 | unsigned long index; |
| 162 | int ret; |
| 163 | |
| 164 | if (err) |
| 165 | cb->errors = 1; |
| 166 | |
| 167 | /* if there are more bios still pending for this compressed |
| 168 | * extent, just exit |
| 169 | */ |
| 170 | if (!atomic_dec_and_test(&cb->pending_bios)) |
| 171 | goto out; |
| 172 | |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 173 | inode = cb->inode; |
| 174 | ret = check_compressed_csum(inode, cb, (u64)bio->bi_sector << 9); |
| 175 | if (ret) |
| 176 | goto csum_failed; |
| 177 | |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 178 | /* ok, we're the last bio for this extent, lets start |
| 179 | * the decompression. |
| 180 | */ |
Li Zefan | 261507a0 | 2010-12-17 14:21:50 +0800 | [diff] [blame] | 181 | ret = btrfs_decompress_biovec(cb->compress_type, |
| 182 | cb->compressed_pages, |
| 183 | cb->start, |
| 184 | cb->orig_bio->bi_io_vec, |
| 185 | cb->orig_bio->bi_vcnt, |
| 186 | cb->compressed_len); |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 187 | csum_failed: |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 188 | if (ret) |
| 189 | cb->errors = 1; |
| 190 | |
| 191 | /* release the compressed pages */ |
| 192 | index = 0; |
| 193 | for (index = 0; index < cb->nr_pages; index++) { |
| 194 | page = cb->compressed_pages[index]; |
| 195 | page->mapping = NULL; |
| 196 | page_cache_release(page); |
| 197 | } |
| 198 | |
| 199 | /* do io completion on the original bio */ |
Chris Mason | 771ed68 | 2008-11-06 22:02:51 -0500 | [diff] [blame] | 200 | if (cb->errors) { |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 201 | bio_io_error(cb->orig_bio); |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 202 | } else { |
| 203 | int bio_index = 0; |
| 204 | struct bio_vec *bvec = cb->orig_bio->bi_io_vec; |
| 205 | |
| 206 | /* |
| 207 | * we have verified the checksum already, set page |
| 208 | * checked so the end_io handlers know about it |
| 209 | */ |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 210 | while (bio_index < cb->orig_bio->bi_vcnt) { |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 211 | SetPageChecked(bvec->bv_page); |
| 212 | bvec++; |
| 213 | bio_index++; |
| 214 | } |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 215 | bio_endio(cb->orig_bio, 0); |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 216 | } |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 217 | |
| 218 | /* finally free the cb struct */ |
| 219 | kfree(cb->compressed_pages); |
| 220 | kfree(cb); |
| 221 | out: |
| 222 | bio_put(bio); |
| 223 | } |
| 224 | |
| 225 | /* |
| 226 | * Clear the writeback bits on all of the file |
| 227 | * pages for a compressed write |
| 228 | */ |
Jeff Mahoney | 143bede | 2012-03-01 14:56:26 +0100 | [diff] [blame] | 229 | static noinline void end_compressed_writeback(struct inode *inode, u64 start, |
| 230 | unsigned long ram_size) |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 231 | { |
| 232 | unsigned long index = start >> PAGE_CACHE_SHIFT; |
| 233 | unsigned long end_index = (start + ram_size - 1) >> PAGE_CACHE_SHIFT; |
| 234 | struct page *pages[16]; |
| 235 | unsigned long nr_pages = end_index - index + 1; |
| 236 | int i; |
| 237 | int ret; |
| 238 | |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 239 | while (nr_pages > 0) { |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 240 | ret = find_get_pages_contig(inode->i_mapping, index, |
Chris Mason | 5b050f0 | 2008-11-11 09:34:41 -0500 | [diff] [blame] | 241 | min_t(unsigned long, |
| 242 | nr_pages, ARRAY_SIZE(pages)), pages); |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 243 | if (ret == 0) { |
| 244 | nr_pages -= 1; |
| 245 | index += 1; |
| 246 | continue; |
| 247 | } |
| 248 | for (i = 0; i < ret; i++) { |
| 249 | end_page_writeback(pages[i]); |
| 250 | page_cache_release(pages[i]); |
| 251 | } |
| 252 | nr_pages -= ret; |
| 253 | index += ret; |
| 254 | } |
| 255 | /* the inode may be gone now */ |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 256 | } |
| 257 | |
| 258 | /* |
| 259 | * do the cleanup once all the compressed pages hit the disk. |
| 260 | * This will clear writeback on the file pages and free the compressed |
| 261 | * pages. |
| 262 | * |
| 263 | * This also calls the writeback end hooks for the file pages so that |
| 264 | * metadata and checksums can be updated in the file. |
| 265 | */ |
| 266 | static void end_compressed_bio_write(struct bio *bio, int err) |
| 267 | { |
| 268 | struct extent_io_tree *tree; |
| 269 | struct compressed_bio *cb = bio->bi_private; |
| 270 | struct inode *inode; |
| 271 | struct page *page; |
| 272 | unsigned long index; |
| 273 | |
| 274 | if (err) |
| 275 | cb->errors = 1; |
| 276 | |
| 277 | /* if there are more bios still pending for this compressed |
| 278 | * extent, just exit |
| 279 | */ |
| 280 | if (!atomic_dec_and_test(&cb->pending_bios)) |
| 281 | goto out; |
| 282 | |
| 283 | /* ok, we're the last bio for this extent, step one is to |
| 284 | * call back into the FS and do all the end_io operations |
| 285 | */ |
| 286 | inode = cb->inode; |
| 287 | tree = &BTRFS_I(inode)->io_tree; |
Chris Mason | 70b99e6 | 2008-10-31 12:46:39 -0400 | [diff] [blame] | 288 | cb->compressed_pages[0]->mapping = cb->inode->i_mapping; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 289 | tree->ops->writepage_end_io_hook(cb->compressed_pages[0], |
| 290 | cb->start, |
| 291 | cb->start + cb->len - 1, |
| 292 | NULL, 1); |
Chris Mason | 70b99e6 | 2008-10-31 12:46:39 -0400 | [diff] [blame] | 293 | cb->compressed_pages[0]->mapping = NULL; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 294 | |
| 295 | end_compressed_writeback(inode, cb->start, cb->len); |
| 296 | /* note, our inode could be gone now */ |
| 297 | |
| 298 | /* |
| 299 | * release the compressed pages, these came from alloc_page and |
| 300 | * are not attached to the inode at all |
| 301 | */ |
| 302 | index = 0; |
| 303 | for (index = 0; index < cb->nr_pages; index++) { |
| 304 | page = cb->compressed_pages[index]; |
| 305 | page->mapping = NULL; |
| 306 | page_cache_release(page); |
| 307 | } |
| 308 | |
| 309 | /* finally free the cb struct */ |
| 310 | kfree(cb->compressed_pages); |
| 311 | kfree(cb); |
| 312 | out: |
| 313 | bio_put(bio); |
| 314 | } |
| 315 | |
| 316 | /* |
| 317 | * worker function to build and submit bios for previously compressed pages. |
| 318 | * The corresponding pages in the inode should be marked for writeback |
| 319 | * and the compressed pages should have a reference on them for dropping |
| 320 | * when the IO is complete. |
| 321 | * |
| 322 | * This also checksums the file bytes and gets things ready for |
| 323 | * the end io hooks. |
| 324 | */ |
| 325 | int btrfs_submit_compressed_write(struct inode *inode, u64 start, |
| 326 | unsigned long len, u64 disk_start, |
| 327 | unsigned long compressed_len, |
| 328 | struct page **compressed_pages, |
| 329 | unsigned long nr_pages) |
| 330 | { |
| 331 | struct bio *bio = NULL; |
| 332 | struct btrfs_root *root = BTRFS_I(inode)->root; |
| 333 | struct compressed_bio *cb; |
| 334 | unsigned long bytes_left; |
| 335 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
David Sterba | 306e16c | 2011-04-19 14:29:38 +0200 | [diff] [blame] | 336 | int pg_index = 0; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 337 | struct page *page; |
| 338 | u64 first_byte = disk_start; |
| 339 | struct block_device *bdev; |
| 340 | int ret; |
Li Zefan | e55179b | 2011-07-14 03:16:47 +0000 | [diff] [blame] | 341 | int skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 342 | |
| 343 | WARN_ON(start & ((u64)PAGE_CACHE_SIZE - 1)); |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 344 | cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS); |
Yoshinori Sano | dac97e51 | 2011-02-15 12:01:42 +0000 | [diff] [blame] | 345 | if (!cb) |
| 346 | return -ENOMEM; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 347 | atomic_set(&cb->pending_bios, 0); |
| 348 | cb->errors = 0; |
| 349 | cb->inode = inode; |
| 350 | cb->start = start; |
| 351 | cb->len = len; |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 352 | cb->mirror_num = 0; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 353 | cb->compressed_pages = compressed_pages; |
| 354 | cb->compressed_len = compressed_len; |
| 355 | cb->orig_bio = NULL; |
| 356 | cb->nr_pages = nr_pages; |
| 357 | |
| 358 | bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; |
| 359 | |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 360 | bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS); |
Dulshani Gunawardhana | 6787125 | 2013-10-31 10:33:04 +0530 | [diff] [blame] | 361 | if (!bio) { |
Yoshinori Sano | dac97e51 | 2011-02-15 12:01:42 +0000 | [diff] [blame] | 362 | kfree(cb); |
| 363 | return -ENOMEM; |
| 364 | } |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 365 | bio->bi_private = cb; |
| 366 | bio->bi_end_io = end_compressed_bio_write; |
| 367 | atomic_inc(&cb->pending_bios); |
| 368 | |
| 369 | /* create and submit bios for the compressed pages */ |
| 370 | bytes_left = compressed_len; |
David Sterba | 306e16c | 2011-04-19 14:29:38 +0200 | [diff] [blame] | 371 | for (pg_index = 0; pg_index < cb->nr_pages; pg_index++) { |
| 372 | page = compressed_pages[pg_index]; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 373 | page->mapping = inode->i_mapping; |
| 374 | if (bio->bi_size) |
David Woodhouse | 64a1670 | 2009-07-15 23:29:37 +0100 | [diff] [blame] | 375 | ret = io_tree->ops->merge_bio_hook(WRITE, page, 0, |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 376 | PAGE_CACHE_SIZE, |
| 377 | bio, 0); |
| 378 | else |
| 379 | ret = 0; |
| 380 | |
Chris Mason | 70b99e6 | 2008-10-31 12:46:39 -0400 | [diff] [blame] | 381 | page->mapping = NULL; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 382 | if (ret || bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < |
| 383 | PAGE_CACHE_SIZE) { |
| 384 | bio_get(bio); |
| 385 | |
Chris Mason | af09abf | 2008-11-07 12:35:44 -0500 | [diff] [blame] | 386 | /* |
| 387 | * inc the count before we submit the bio so |
| 388 | * we know the end IO handler won't happen before |
| 389 | * we inc the count. Otherwise, the cb might get |
| 390 | * freed before we're done setting it up |
| 391 | */ |
| 392 | atomic_inc(&cb->pending_bios); |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 393 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); |
Jeff Mahoney | 79787ea | 2012-03-12 16:03:00 +0100 | [diff] [blame] | 394 | BUG_ON(ret); /* -ENOMEM */ |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 395 | |
Li Zefan | e55179b | 2011-07-14 03:16:47 +0000 | [diff] [blame] | 396 | if (!skip_sum) { |
| 397 | ret = btrfs_csum_one_bio(root, inode, bio, |
| 398 | start, 1); |
Jeff Mahoney | 79787ea | 2012-03-12 16:03:00 +0100 | [diff] [blame] | 399 | BUG_ON(ret); /* -ENOMEM */ |
Li Zefan | e55179b | 2011-07-14 03:16:47 +0000 | [diff] [blame] | 400 | } |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 401 | |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 402 | ret = btrfs_map_bio(root, WRITE, bio, 0, 1); |
Jeff Mahoney | 79787ea | 2012-03-12 16:03:00 +0100 | [diff] [blame] | 403 | BUG_ON(ret); /* -ENOMEM */ |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 404 | |
| 405 | bio_put(bio); |
| 406 | |
| 407 | bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS); |
Tsutomu Itoh | e627ee7 | 2012-04-12 16:03:56 -0400 | [diff] [blame] | 408 | BUG_ON(!bio); |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 409 | bio->bi_private = cb; |
| 410 | bio->bi_end_io = end_compressed_bio_write; |
| 411 | bio_add_page(bio, page, PAGE_CACHE_SIZE, 0); |
| 412 | } |
Chris Mason | cfbc246 | 2008-10-30 13:22:14 -0400 | [diff] [blame] | 413 | if (bytes_left < PAGE_CACHE_SIZE) { |
Frank Holton | efe120a | 2013-12-20 11:37:06 -0500 | [diff] [blame] | 414 | btrfs_info(BTRFS_I(inode)->root->fs_info, |
| 415 | "bytes left %lu compress len %lu nr %lu", |
Chris Mason | cfbc246 | 2008-10-30 13:22:14 -0400 | [diff] [blame] | 416 | bytes_left, cb->compressed_len, cb->nr_pages); |
| 417 | } |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 418 | bytes_left -= PAGE_CACHE_SIZE; |
| 419 | first_byte += PAGE_CACHE_SIZE; |
Chris Mason | 771ed68 | 2008-11-06 22:02:51 -0500 | [diff] [blame] | 420 | cond_resched(); |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 421 | } |
| 422 | bio_get(bio); |
| 423 | |
| 424 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); |
Jeff Mahoney | 79787ea | 2012-03-12 16:03:00 +0100 | [diff] [blame] | 425 | BUG_ON(ret); /* -ENOMEM */ |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 426 | |
Li Zefan | e55179b | 2011-07-14 03:16:47 +0000 | [diff] [blame] | 427 | if (!skip_sum) { |
| 428 | ret = btrfs_csum_one_bio(root, inode, bio, start, 1); |
Jeff Mahoney | 79787ea | 2012-03-12 16:03:00 +0100 | [diff] [blame] | 429 | BUG_ON(ret); /* -ENOMEM */ |
Li Zefan | e55179b | 2011-07-14 03:16:47 +0000 | [diff] [blame] | 430 | } |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 431 | |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 432 | ret = btrfs_map_bio(root, WRITE, bio, 0, 1); |
Jeff Mahoney | 79787ea | 2012-03-12 16:03:00 +0100 | [diff] [blame] | 433 | BUG_ON(ret); /* -ENOMEM */ |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 434 | |
| 435 | bio_put(bio); |
| 436 | return 0; |
| 437 | } |
| 438 | |
Chris Mason | 771ed68 | 2008-11-06 22:02:51 -0500 | [diff] [blame] | 439 | static noinline int add_ra_bio_pages(struct inode *inode, |
| 440 | u64 compressed_end, |
| 441 | struct compressed_bio *cb) |
| 442 | { |
| 443 | unsigned long end_index; |
David Sterba | 306e16c | 2011-04-19 14:29:38 +0200 | [diff] [blame] | 444 | unsigned long pg_index; |
Chris Mason | 771ed68 | 2008-11-06 22:02:51 -0500 | [diff] [blame] | 445 | u64 last_offset; |
| 446 | u64 isize = i_size_read(inode); |
| 447 | int ret; |
| 448 | struct page *page; |
| 449 | unsigned long nr_pages = 0; |
| 450 | struct extent_map *em; |
| 451 | struct address_space *mapping = inode->i_mapping; |
Chris Mason | 771ed68 | 2008-11-06 22:02:51 -0500 | [diff] [blame] | 452 | struct extent_map_tree *em_tree; |
| 453 | struct extent_io_tree *tree; |
| 454 | u64 end; |
| 455 | int misses = 0; |
| 456 | |
| 457 | page = cb->orig_bio->bi_io_vec[cb->orig_bio->bi_vcnt - 1].bv_page; |
| 458 | last_offset = (page_offset(page) + PAGE_CACHE_SIZE); |
| 459 | em_tree = &BTRFS_I(inode)->extent_tree; |
| 460 | tree = &BTRFS_I(inode)->io_tree; |
| 461 | |
| 462 | if (isize == 0) |
| 463 | return 0; |
| 464 | |
| 465 | end_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT; |
| 466 | |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 467 | while (last_offset < compressed_end) { |
David Sterba | 306e16c | 2011-04-19 14:29:38 +0200 | [diff] [blame] | 468 | pg_index = last_offset >> PAGE_CACHE_SHIFT; |
Chris Mason | 771ed68 | 2008-11-06 22:02:51 -0500 | [diff] [blame] | 469 | |
David Sterba | 306e16c | 2011-04-19 14:29:38 +0200 | [diff] [blame] | 470 | if (pg_index > end_index) |
Chris Mason | 771ed68 | 2008-11-06 22:02:51 -0500 | [diff] [blame] | 471 | break; |
| 472 | |
| 473 | rcu_read_lock(); |
David Sterba | 306e16c | 2011-04-19 14:29:38 +0200 | [diff] [blame] | 474 | page = radix_tree_lookup(&mapping->page_tree, pg_index); |
Chris Mason | 771ed68 | 2008-11-06 22:02:51 -0500 | [diff] [blame] | 475 | rcu_read_unlock(); |
| 476 | if (page) { |
| 477 | misses++; |
| 478 | if (misses > 4) |
| 479 | break; |
| 480 | goto next; |
| 481 | } |
| 482 | |
Nick Piggin | 28ecb609 | 2010-03-17 13:31:04 +0000 | [diff] [blame] | 483 | page = __page_cache_alloc(mapping_gfp_mask(mapping) & |
| 484 | ~__GFP_FS); |
Chris Mason | 771ed68 | 2008-11-06 22:02:51 -0500 | [diff] [blame] | 485 | if (!page) |
| 486 | break; |
| 487 | |
David Sterba | 306e16c | 2011-04-19 14:29:38 +0200 | [diff] [blame] | 488 | if (add_to_page_cache_lru(page, mapping, pg_index, |
Nick Piggin | 28ecb609 | 2010-03-17 13:31:04 +0000 | [diff] [blame] | 489 | GFP_NOFS)) { |
Chris Mason | 771ed68 | 2008-11-06 22:02:51 -0500 | [diff] [blame] | 490 | page_cache_release(page); |
| 491 | goto next; |
| 492 | } |
| 493 | |
Chris Mason | 771ed68 | 2008-11-06 22:02:51 -0500 | [diff] [blame] | 494 | end = last_offset + PAGE_CACHE_SIZE - 1; |
| 495 | /* |
| 496 | * at this point, we have a locked page in the page cache |
| 497 | * for these bytes in the file. But, we have to make |
| 498 | * sure they map to this compressed extent on disk. |
| 499 | */ |
| 500 | set_page_extent_mapped(page); |
Jeff Mahoney | d008237 | 2012-03-01 14:57:19 +0100 | [diff] [blame] | 501 | lock_extent(tree, last_offset, end); |
Chris Mason | 890871b | 2009-09-02 16:24:52 -0400 | [diff] [blame] | 502 | read_lock(&em_tree->lock); |
Chris Mason | 771ed68 | 2008-11-06 22:02:51 -0500 | [diff] [blame] | 503 | em = lookup_extent_mapping(em_tree, last_offset, |
| 504 | PAGE_CACHE_SIZE); |
Chris Mason | 890871b | 2009-09-02 16:24:52 -0400 | [diff] [blame] | 505 | read_unlock(&em_tree->lock); |
Chris Mason | 771ed68 | 2008-11-06 22:02:51 -0500 | [diff] [blame] | 506 | |
| 507 | if (!em || last_offset < em->start || |
| 508 | (last_offset + PAGE_CACHE_SIZE > extent_map_end(em)) || |
| 509 | (em->block_start >> 9) != cb->orig_bio->bi_sector) { |
| 510 | free_extent_map(em); |
Jeff Mahoney | d008237 | 2012-03-01 14:57:19 +0100 | [diff] [blame] | 511 | unlock_extent(tree, last_offset, end); |
Chris Mason | 771ed68 | 2008-11-06 22:02:51 -0500 | [diff] [blame] | 512 | unlock_page(page); |
| 513 | page_cache_release(page); |
| 514 | break; |
| 515 | } |
| 516 | free_extent_map(em); |
| 517 | |
| 518 | if (page->index == end_index) { |
| 519 | char *userpage; |
| 520 | size_t zero_offset = isize & (PAGE_CACHE_SIZE - 1); |
| 521 | |
| 522 | if (zero_offset) { |
| 523 | int zeros; |
| 524 | zeros = PAGE_CACHE_SIZE - zero_offset; |
Cong Wang | 7ac687d | 2011-11-25 23:14:28 +0800 | [diff] [blame] | 525 | userpage = kmap_atomic(page); |
Chris Mason | 771ed68 | 2008-11-06 22:02:51 -0500 | [diff] [blame] | 526 | memset(userpage + zero_offset, 0, zeros); |
| 527 | flush_dcache_page(page); |
Cong Wang | 7ac687d | 2011-11-25 23:14:28 +0800 | [diff] [blame] | 528 | kunmap_atomic(userpage); |
Chris Mason | 771ed68 | 2008-11-06 22:02:51 -0500 | [diff] [blame] | 529 | } |
| 530 | } |
| 531 | |
| 532 | ret = bio_add_page(cb->orig_bio, page, |
| 533 | PAGE_CACHE_SIZE, 0); |
| 534 | |
| 535 | if (ret == PAGE_CACHE_SIZE) { |
| 536 | nr_pages++; |
| 537 | page_cache_release(page); |
| 538 | } else { |
Jeff Mahoney | d008237 | 2012-03-01 14:57:19 +0100 | [diff] [blame] | 539 | unlock_extent(tree, last_offset, end); |
Chris Mason | 771ed68 | 2008-11-06 22:02:51 -0500 | [diff] [blame] | 540 | unlock_page(page); |
| 541 | page_cache_release(page); |
| 542 | break; |
| 543 | } |
| 544 | next: |
| 545 | last_offset += PAGE_CACHE_SIZE; |
| 546 | } |
Chris Mason | 771ed68 | 2008-11-06 22:02:51 -0500 | [diff] [blame] | 547 | return 0; |
| 548 | } |
| 549 | |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 550 | /* |
| 551 | * for a compressed read, the bio we get passed has all the inode pages |
| 552 | * in it. We don't actually do IO on those pages but allocate new ones |
| 553 | * to hold the compressed pages on disk. |
| 554 | * |
| 555 | * bio->bi_sector points to the compressed extent on disk |
| 556 | * bio->bi_io_vec points to all of the inode pages |
| 557 | * bio->bi_vcnt is a count of pages |
| 558 | * |
| 559 | * After the compressed pages are read, we copy the bytes into the |
| 560 | * bio we were passed and then call the bio end_io calls |
| 561 | */ |
| 562 | int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio, |
| 563 | int mirror_num, unsigned long bio_flags) |
| 564 | { |
| 565 | struct extent_io_tree *tree; |
| 566 | struct extent_map_tree *em_tree; |
| 567 | struct compressed_bio *cb; |
| 568 | struct btrfs_root *root = BTRFS_I(inode)->root; |
| 569 | unsigned long uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE; |
| 570 | unsigned long compressed_len; |
| 571 | unsigned long nr_pages; |
David Sterba | 306e16c | 2011-04-19 14:29:38 +0200 | [diff] [blame] | 572 | unsigned long pg_index; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 573 | struct page *page; |
| 574 | struct block_device *bdev; |
| 575 | struct bio *comp_bio; |
| 576 | u64 cur_disk_byte = (u64)bio->bi_sector << 9; |
Chris Mason | e04ca62 | 2008-11-10 11:44:58 -0500 | [diff] [blame] | 577 | u64 em_len; |
| 578 | u64 em_start; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 579 | struct extent_map *em; |
liubo | 6b82ce8 | 2011-01-26 06:21:39 +0000 | [diff] [blame] | 580 | int ret = -ENOMEM; |
Josef Bacik | 15e3004 | 2012-10-05 13:39:50 -0400 | [diff] [blame] | 581 | int faili = 0; |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 582 | u32 *sums; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 583 | |
| 584 | tree = &BTRFS_I(inode)->io_tree; |
| 585 | em_tree = &BTRFS_I(inode)->extent_tree; |
| 586 | |
| 587 | /* we need the actual starting offset of this extent in the file */ |
Chris Mason | 890871b | 2009-09-02 16:24:52 -0400 | [diff] [blame] | 588 | read_lock(&em_tree->lock); |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 589 | em = lookup_extent_mapping(em_tree, |
| 590 | page_offset(bio->bi_io_vec->bv_page), |
| 591 | PAGE_CACHE_SIZE); |
Chris Mason | 890871b | 2009-09-02 16:24:52 -0400 | [diff] [blame] | 592 | read_unlock(&em_tree->lock); |
Tsutomu Itoh | 285190d | 2012-02-16 16:23:58 +0900 | [diff] [blame] | 593 | if (!em) |
| 594 | return -EIO; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 595 | |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 596 | compressed_len = em->block_len; |
| 597 | cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS); |
liubo | 6b82ce8 | 2011-01-26 06:21:39 +0000 | [diff] [blame] | 598 | if (!cb) |
| 599 | goto out; |
| 600 | |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 601 | atomic_set(&cb->pending_bios, 0); |
| 602 | cb->errors = 0; |
| 603 | cb->inode = inode; |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 604 | cb->mirror_num = mirror_num; |
| 605 | sums = &cb->sums; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 606 | |
Yan Zheng | ff5b7ee | 2008-11-10 07:34:43 -0500 | [diff] [blame] | 607 | cb->start = em->orig_start; |
Chris Mason | e04ca62 | 2008-11-10 11:44:58 -0500 | [diff] [blame] | 608 | em_len = em->len; |
| 609 | em_start = em->start; |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 610 | |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 611 | free_extent_map(em); |
Chris Mason | e04ca62 | 2008-11-10 11:44:58 -0500 | [diff] [blame] | 612 | em = NULL; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 613 | |
| 614 | cb->len = uncompressed_len; |
| 615 | cb->compressed_len = compressed_len; |
Li Zefan | 261507a0 | 2010-12-17 14:21:50 +0800 | [diff] [blame] | 616 | cb->compress_type = extent_compress_type(bio_flags); |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 617 | cb->orig_bio = bio; |
| 618 | |
| 619 | nr_pages = (compressed_len + PAGE_CACHE_SIZE - 1) / |
| 620 | PAGE_CACHE_SIZE; |
liubo | 6b82ce8 | 2011-01-26 06:21:39 +0000 | [diff] [blame] | 621 | cb->compressed_pages = kzalloc(sizeof(struct page *) * nr_pages, |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 622 | GFP_NOFS); |
liubo | 6b82ce8 | 2011-01-26 06:21:39 +0000 | [diff] [blame] | 623 | if (!cb->compressed_pages) |
| 624 | goto fail1; |
| 625 | |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 626 | bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; |
| 627 | |
David Sterba | 306e16c | 2011-04-19 14:29:38 +0200 | [diff] [blame] | 628 | for (pg_index = 0; pg_index < nr_pages; pg_index++) { |
| 629 | cb->compressed_pages[pg_index] = alloc_page(GFP_NOFS | |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 630 | __GFP_HIGHMEM); |
Josef Bacik | 15e3004 | 2012-10-05 13:39:50 -0400 | [diff] [blame] | 631 | if (!cb->compressed_pages[pg_index]) { |
| 632 | faili = pg_index - 1; |
| 633 | ret = -ENOMEM; |
liubo | 6b82ce8 | 2011-01-26 06:21:39 +0000 | [diff] [blame] | 634 | goto fail2; |
Josef Bacik | 15e3004 | 2012-10-05 13:39:50 -0400 | [diff] [blame] | 635 | } |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 636 | } |
Josef Bacik | 15e3004 | 2012-10-05 13:39:50 -0400 | [diff] [blame] | 637 | faili = nr_pages - 1; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 638 | cb->nr_pages = nr_pages; |
| 639 | |
Mark Fasheh | 4b38431 | 2013-08-06 11:42:50 -0700 | [diff] [blame] | 640 | /* In the parent-locked case, we only locked the range we are |
| 641 | * interested in. In all other cases, we can opportunistically |
| 642 | * cache decompressed data that goes beyond the requested range. */ |
| 643 | if (!(bio_flags & EXTENT_BIO_PARENT_LOCKED)) |
| 644 | add_ra_bio_pages(inode, em_start + em_len, cb); |
Chris Mason | 771ed68 | 2008-11-06 22:02:51 -0500 | [diff] [blame] | 645 | |
Chris Mason | 771ed68 | 2008-11-06 22:02:51 -0500 | [diff] [blame] | 646 | /* include any pages we added in add_ra-bio_pages */ |
| 647 | uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE; |
| 648 | cb->len = uncompressed_len; |
| 649 | |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 650 | comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, GFP_NOFS); |
liubo | 6b82ce8 | 2011-01-26 06:21:39 +0000 | [diff] [blame] | 651 | if (!comp_bio) |
| 652 | goto fail2; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 653 | comp_bio->bi_private = cb; |
| 654 | comp_bio->bi_end_io = end_compressed_bio_read; |
| 655 | atomic_inc(&cb->pending_bios); |
| 656 | |
David Sterba | 306e16c | 2011-04-19 14:29:38 +0200 | [diff] [blame] | 657 | for (pg_index = 0; pg_index < nr_pages; pg_index++) { |
| 658 | page = cb->compressed_pages[pg_index]; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 659 | page->mapping = inode->i_mapping; |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 660 | page->index = em_start >> PAGE_CACHE_SHIFT; |
| 661 | |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 662 | if (comp_bio->bi_size) |
David Woodhouse | 64a1670 | 2009-07-15 23:29:37 +0100 | [diff] [blame] | 663 | ret = tree->ops->merge_bio_hook(READ, page, 0, |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 664 | PAGE_CACHE_SIZE, |
| 665 | comp_bio, 0); |
| 666 | else |
| 667 | ret = 0; |
| 668 | |
Chris Mason | 70b99e6 | 2008-10-31 12:46:39 -0400 | [diff] [blame] | 669 | page->mapping = NULL; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 670 | if (ret || bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0) < |
| 671 | PAGE_CACHE_SIZE) { |
| 672 | bio_get(comp_bio); |
| 673 | |
| 674 | ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0); |
Jeff Mahoney | 79787ea | 2012-03-12 16:03:00 +0100 | [diff] [blame] | 675 | BUG_ON(ret); /* -ENOMEM */ |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 676 | |
Chris Mason | af09abf | 2008-11-07 12:35:44 -0500 | [diff] [blame] | 677 | /* |
| 678 | * inc the count before we submit the bio so |
| 679 | * we know the end IO handler won't happen before |
| 680 | * we inc the count. Otherwise, the cb might get |
| 681 | * freed before we're done setting it up |
| 682 | */ |
| 683 | atomic_inc(&cb->pending_bios); |
| 684 | |
Christoph Hellwig | 6cbff00 | 2009-04-17 10:37:41 +0200 | [diff] [blame] | 685 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { |
Tsutomu Itoh | c2db107 | 2011-03-01 06:48:31 +0000 | [diff] [blame] | 686 | ret = btrfs_lookup_bio_sums(root, inode, |
| 687 | comp_bio, sums); |
Jeff Mahoney | 79787ea | 2012-03-12 16:03:00 +0100 | [diff] [blame] | 688 | BUG_ON(ret); /* -ENOMEM */ |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 689 | } |
| 690 | sums += (comp_bio->bi_size + root->sectorsize - 1) / |
| 691 | root->sectorsize; |
| 692 | |
| 693 | ret = btrfs_map_bio(root, READ, comp_bio, |
| 694 | mirror_num, 0); |
Stefan Behrens | 6189192 | 2012-11-05 18:51:52 +0100 | [diff] [blame] | 695 | if (ret) |
| 696 | bio_endio(comp_bio, ret); |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 697 | |
| 698 | bio_put(comp_bio); |
| 699 | |
| 700 | comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, |
| 701 | GFP_NOFS); |
Tsutomu Itoh | e627ee7 | 2012-04-12 16:03:56 -0400 | [diff] [blame] | 702 | BUG_ON(!comp_bio); |
Chris Mason | 771ed68 | 2008-11-06 22:02:51 -0500 | [diff] [blame] | 703 | comp_bio->bi_private = cb; |
| 704 | comp_bio->bi_end_io = end_compressed_bio_read; |
| 705 | |
| 706 | bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0); |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 707 | } |
| 708 | cur_disk_byte += PAGE_CACHE_SIZE; |
| 709 | } |
| 710 | bio_get(comp_bio); |
| 711 | |
| 712 | ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0); |
Jeff Mahoney | 79787ea | 2012-03-12 16:03:00 +0100 | [diff] [blame] | 713 | BUG_ON(ret); /* -ENOMEM */ |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 714 | |
Tsutomu Itoh | c2db107 | 2011-03-01 06:48:31 +0000 | [diff] [blame] | 715 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { |
| 716 | ret = btrfs_lookup_bio_sums(root, inode, comp_bio, sums); |
Jeff Mahoney | 79787ea | 2012-03-12 16:03:00 +0100 | [diff] [blame] | 717 | BUG_ON(ret); /* -ENOMEM */ |
Tsutomu Itoh | c2db107 | 2011-03-01 06:48:31 +0000 | [diff] [blame] | 718 | } |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 719 | |
| 720 | ret = btrfs_map_bio(root, READ, comp_bio, mirror_num, 0); |
Stefan Behrens | 6189192 | 2012-11-05 18:51:52 +0100 | [diff] [blame] | 721 | if (ret) |
| 722 | bio_endio(comp_bio, ret); |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 723 | |
| 724 | bio_put(comp_bio); |
| 725 | return 0; |
liubo | 6b82ce8 | 2011-01-26 06:21:39 +0000 | [diff] [blame] | 726 | |
| 727 | fail2: |
Josef Bacik | 15e3004 | 2012-10-05 13:39:50 -0400 | [diff] [blame] | 728 | while (faili >= 0) { |
| 729 | __free_page(cb->compressed_pages[faili]); |
| 730 | faili--; |
| 731 | } |
liubo | 6b82ce8 | 2011-01-26 06:21:39 +0000 | [diff] [blame] | 732 | |
| 733 | kfree(cb->compressed_pages); |
| 734 | fail1: |
| 735 | kfree(cb); |
| 736 | out: |
| 737 | free_extent_map(em); |
| 738 | return ret; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 739 | } |
Li Zefan | 261507a0 | 2010-12-17 14:21:50 +0800 | [diff] [blame] | 740 | |
| 741 | static struct list_head comp_idle_workspace[BTRFS_COMPRESS_TYPES]; |
| 742 | static spinlock_t comp_workspace_lock[BTRFS_COMPRESS_TYPES]; |
| 743 | static int comp_num_workspace[BTRFS_COMPRESS_TYPES]; |
| 744 | static atomic_t comp_alloc_workspace[BTRFS_COMPRESS_TYPES]; |
| 745 | static wait_queue_head_t comp_workspace_wait[BTRFS_COMPRESS_TYPES]; |
| 746 | |
Eric Sandeen | 48a3b63 | 2013-04-25 20:41:01 +0000 | [diff] [blame] | 747 | static struct btrfs_compress_op *btrfs_compress_op[] = { |
Li Zefan | 261507a0 | 2010-12-17 14:21:50 +0800 | [diff] [blame] | 748 | &btrfs_zlib_compress, |
Li Zefan | a6fa6fa | 2010-10-25 15:12:26 +0800 | [diff] [blame] | 749 | &btrfs_lzo_compress, |
Li Zefan | 261507a0 | 2010-12-17 14:21:50 +0800 | [diff] [blame] | 750 | }; |
| 751 | |
Jeff Mahoney | 143bede | 2012-03-01 14:56:26 +0100 | [diff] [blame] | 752 | void __init btrfs_init_compress(void) |
Li Zefan | 261507a0 | 2010-12-17 14:21:50 +0800 | [diff] [blame] | 753 | { |
| 754 | int i; |
| 755 | |
| 756 | for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) { |
| 757 | INIT_LIST_HEAD(&comp_idle_workspace[i]); |
| 758 | spin_lock_init(&comp_workspace_lock[i]); |
| 759 | atomic_set(&comp_alloc_workspace[i], 0); |
| 760 | init_waitqueue_head(&comp_workspace_wait[i]); |
| 761 | } |
Li Zefan | 261507a0 | 2010-12-17 14:21:50 +0800 | [diff] [blame] | 762 | } |
| 763 | |
| 764 | /* |
| 765 | * this finds an available workspace or allocates a new one |
| 766 | * ERR_PTR is returned if things go bad. |
| 767 | */ |
| 768 | static struct list_head *find_workspace(int type) |
| 769 | { |
| 770 | struct list_head *workspace; |
| 771 | int cpus = num_online_cpus(); |
| 772 | int idx = type - 1; |
| 773 | |
| 774 | struct list_head *idle_workspace = &comp_idle_workspace[idx]; |
| 775 | spinlock_t *workspace_lock = &comp_workspace_lock[idx]; |
| 776 | atomic_t *alloc_workspace = &comp_alloc_workspace[idx]; |
| 777 | wait_queue_head_t *workspace_wait = &comp_workspace_wait[idx]; |
| 778 | int *num_workspace = &comp_num_workspace[idx]; |
| 779 | again: |
| 780 | spin_lock(workspace_lock); |
| 781 | if (!list_empty(idle_workspace)) { |
| 782 | workspace = idle_workspace->next; |
| 783 | list_del(workspace); |
| 784 | (*num_workspace)--; |
| 785 | spin_unlock(workspace_lock); |
| 786 | return workspace; |
| 787 | |
| 788 | } |
| 789 | if (atomic_read(alloc_workspace) > cpus) { |
| 790 | DEFINE_WAIT(wait); |
| 791 | |
| 792 | spin_unlock(workspace_lock); |
| 793 | prepare_to_wait(workspace_wait, &wait, TASK_UNINTERRUPTIBLE); |
| 794 | if (atomic_read(alloc_workspace) > cpus && !*num_workspace) |
| 795 | schedule(); |
| 796 | finish_wait(workspace_wait, &wait); |
| 797 | goto again; |
| 798 | } |
| 799 | atomic_inc(alloc_workspace); |
| 800 | spin_unlock(workspace_lock); |
| 801 | |
| 802 | workspace = btrfs_compress_op[idx]->alloc_workspace(); |
| 803 | if (IS_ERR(workspace)) { |
| 804 | atomic_dec(alloc_workspace); |
| 805 | wake_up(workspace_wait); |
| 806 | } |
| 807 | return workspace; |
| 808 | } |
| 809 | |
| 810 | /* |
| 811 | * put a workspace struct back on the list or free it if we have enough |
| 812 | * idle ones sitting around |
| 813 | */ |
| 814 | static void free_workspace(int type, struct list_head *workspace) |
| 815 | { |
| 816 | int idx = type - 1; |
| 817 | struct list_head *idle_workspace = &comp_idle_workspace[idx]; |
| 818 | spinlock_t *workspace_lock = &comp_workspace_lock[idx]; |
| 819 | atomic_t *alloc_workspace = &comp_alloc_workspace[idx]; |
| 820 | wait_queue_head_t *workspace_wait = &comp_workspace_wait[idx]; |
| 821 | int *num_workspace = &comp_num_workspace[idx]; |
| 822 | |
| 823 | spin_lock(workspace_lock); |
| 824 | if (*num_workspace < num_online_cpus()) { |
| 825 | list_add_tail(workspace, idle_workspace); |
| 826 | (*num_workspace)++; |
| 827 | spin_unlock(workspace_lock); |
| 828 | goto wake; |
| 829 | } |
| 830 | spin_unlock(workspace_lock); |
| 831 | |
| 832 | btrfs_compress_op[idx]->free_workspace(workspace); |
| 833 | atomic_dec(alloc_workspace); |
| 834 | wake: |
Josef Bacik | 66657b3 | 2012-08-01 15:36:24 -0400 | [diff] [blame] | 835 | smp_mb(); |
Li Zefan | 261507a0 | 2010-12-17 14:21:50 +0800 | [diff] [blame] | 836 | if (waitqueue_active(workspace_wait)) |
| 837 | wake_up(workspace_wait); |
| 838 | } |
| 839 | |
| 840 | /* |
| 841 | * cleanup function for module exit |
| 842 | */ |
| 843 | static void free_workspaces(void) |
| 844 | { |
| 845 | struct list_head *workspace; |
| 846 | int i; |
| 847 | |
| 848 | for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) { |
| 849 | while (!list_empty(&comp_idle_workspace[i])) { |
| 850 | workspace = comp_idle_workspace[i].next; |
| 851 | list_del(workspace); |
| 852 | btrfs_compress_op[i]->free_workspace(workspace); |
| 853 | atomic_dec(&comp_alloc_workspace[i]); |
| 854 | } |
| 855 | } |
| 856 | } |
| 857 | |
| 858 | /* |
| 859 | * given an address space and start/len, compress the bytes. |
| 860 | * |
| 861 | * pages are allocated to hold the compressed result and stored |
| 862 | * in 'pages' |
| 863 | * |
| 864 | * out_pages is used to return the number of pages allocated. There |
| 865 | * may be pages allocated even if we return an error |
| 866 | * |
| 867 | * total_in is used to return the number of bytes actually read. It |
| 868 | * may be smaller then len if we had to exit early because we |
| 869 | * ran out of room in the pages array or because we cross the |
| 870 | * max_out threshold. |
| 871 | * |
| 872 | * total_out is used to return the total number of compressed bytes |
| 873 | * |
| 874 | * max_out tells us the max number of bytes that we're allowed to |
| 875 | * stuff into pages |
| 876 | */ |
| 877 | int btrfs_compress_pages(int type, struct address_space *mapping, |
| 878 | u64 start, unsigned long len, |
| 879 | struct page **pages, |
| 880 | unsigned long nr_dest_pages, |
| 881 | unsigned long *out_pages, |
| 882 | unsigned long *total_in, |
| 883 | unsigned long *total_out, |
| 884 | unsigned long max_out) |
| 885 | { |
| 886 | struct list_head *workspace; |
| 887 | int ret; |
| 888 | |
| 889 | workspace = find_workspace(type); |
| 890 | if (IS_ERR(workspace)) |
| 891 | return -1; |
| 892 | |
| 893 | ret = btrfs_compress_op[type-1]->compress_pages(workspace, mapping, |
| 894 | start, len, pages, |
| 895 | nr_dest_pages, out_pages, |
| 896 | total_in, total_out, |
| 897 | max_out); |
| 898 | free_workspace(type, workspace); |
| 899 | return ret; |
| 900 | } |
| 901 | |
| 902 | /* |
| 903 | * pages_in is an array of pages with compressed data. |
| 904 | * |
| 905 | * disk_start is the starting logical offset of this array in the file |
| 906 | * |
| 907 | * bvec is a bio_vec of pages from the file that we want to decompress into |
| 908 | * |
| 909 | * vcnt is the count of pages in the biovec |
| 910 | * |
| 911 | * srclen is the number of bytes in pages_in |
| 912 | * |
| 913 | * The basic idea is that we have a bio that was created by readpages. |
| 914 | * The pages in the bio are for the uncompressed data, and they may not |
| 915 | * be contiguous. They all correspond to the range of bytes covered by |
| 916 | * the compressed extent. |
| 917 | */ |
Eric Sandeen | 48a3b63 | 2013-04-25 20:41:01 +0000 | [diff] [blame] | 918 | static int btrfs_decompress_biovec(int type, struct page **pages_in, |
| 919 | u64 disk_start, struct bio_vec *bvec, |
| 920 | int vcnt, size_t srclen) |
Li Zefan | 261507a0 | 2010-12-17 14:21:50 +0800 | [diff] [blame] | 921 | { |
| 922 | struct list_head *workspace; |
| 923 | int ret; |
| 924 | |
| 925 | workspace = find_workspace(type); |
| 926 | if (IS_ERR(workspace)) |
| 927 | return -ENOMEM; |
| 928 | |
| 929 | ret = btrfs_compress_op[type-1]->decompress_biovec(workspace, pages_in, |
| 930 | disk_start, |
| 931 | bvec, vcnt, srclen); |
| 932 | free_workspace(type, workspace); |
| 933 | return ret; |
| 934 | } |
| 935 | |
| 936 | /* |
| 937 | * a less complex decompression routine. Our compressed data fits in a |
| 938 | * single page, and we want to read a single page out of it. |
| 939 | * start_byte tells us the offset into the compressed data we're interested in |
| 940 | */ |
| 941 | int btrfs_decompress(int type, unsigned char *data_in, struct page *dest_page, |
| 942 | unsigned long start_byte, size_t srclen, size_t destlen) |
| 943 | { |
| 944 | struct list_head *workspace; |
| 945 | int ret; |
| 946 | |
| 947 | workspace = find_workspace(type); |
| 948 | if (IS_ERR(workspace)) |
| 949 | return -ENOMEM; |
| 950 | |
| 951 | ret = btrfs_compress_op[type-1]->decompress(workspace, data_in, |
| 952 | dest_page, start_byte, |
| 953 | srclen, destlen); |
| 954 | |
| 955 | free_workspace(type, workspace); |
| 956 | return ret; |
| 957 | } |
| 958 | |
Alexey Charkov | 8e4eef7 | 2011-02-02 21:15:35 +0000 | [diff] [blame] | 959 | void btrfs_exit_compress(void) |
Li Zefan | 261507a0 | 2010-12-17 14:21:50 +0800 | [diff] [blame] | 960 | { |
| 961 | free_workspaces(); |
| 962 | } |
Li Zefan | 3a39c18 | 2010-11-08 15:22:19 +0800 | [diff] [blame] | 963 | |
| 964 | /* |
| 965 | * Copy uncompressed data from working buffer to pages. |
| 966 | * |
| 967 | * buf_start is the byte offset we're of the start of our workspace buffer. |
| 968 | * |
| 969 | * total_out is the last byte of the buffer |
| 970 | */ |
| 971 | int btrfs_decompress_buf2page(char *buf, unsigned long buf_start, |
| 972 | unsigned long total_out, u64 disk_start, |
| 973 | struct bio_vec *bvec, int vcnt, |
David Sterba | 306e16c | 2011-04-19 14:29:38 +0200 | [diff] [blame] | 974 | unsigned long *pg_index, |
Li Zefan | 3a39c18 | 2010-11-08 15:22:19 +0800 | [diff] [blame] | 975 | unsigned long *pg_offset) |
| 976 | { |
| 977 | unsigned long buf_offset; |
| 978 | unsigned long current_buf_start; |
| 979 | unsigned long start_byte; |
| 980 | unsigned long working_bytes = total_out - buf_start; |
| 981 | unsigned long bytes; |
| 982 | char *kaddr; |
David Sterba | 306e16c | 2011-04-19 14:29:38 +0200 | [diff] [blame] | 983 | struct page *page_out = bvec[*pg_index].bv_page; |
Li Zefan | 3a39c18 | 2010-11-08 15:22:19 +0800 | [diff] [blame] | 984 | |
| 985 | /* |
| 986 | * start byte is the first byte of the page we're currently |
| 987 | * copying into relative to the start of the compressed data. |
| 988 | */ |
| 989 | start_byte = page_offset(page_out) - disk_start; |
| 990 | |
| 991 | /* we haven't yet hit data corresponding to this page */ |
| 992 | if (total_out <= start_byte) |
| 993 | return 1; |
| 994 | |
| 995 | /* |
| 996 | * the start of the data we care about is offset into |
| 997 | * the middle of our working buffer |
| 998 | */ |
| 999 | if (total_out > start_byte && buf_start < start_byte) { |
| 1000 | buf_offset = start_byte - buf_start; |
| 1001 | working_bytes -= buf_offset; |
| 1002 | } else { |
| 1003 | buf_offset = 0; |
| 1004 | } |
| 1005 | current_buf_start = buf_start; |
| 1006 | |
| 1007 | /* copy bytes from the working buffer into the pages */ |
| 1008 | while (working_bytes > 0) { |
| 1009 | bytes = min(PAGE_CACHE_SIZE - *pg_offset, |
| 1010 | PAGE_CACHE_SIZE - buf_offset); |
| 1011 | bytes = min(bytes, working_bytes); |
Cong Wang | 7ac687d | 2011-11-25 23:14:28 +0800 | [diff] [blame] | 1012 | kaddr = kmap_atomic(page_out); |
Li Zefan | 3a39c18 | 2010-11-08 15:22:19 +0800 | [diff] [blame] | 1013 | memcpy(kaddr + *pg_offset, buf + buf_offset, bytes); |
Filipe David Borba Manana | a2aa75e | 2014-02-08 15:47:46 +0000 | [diff] [blame^] | 1014 | if (*pg_index == (vcnt - 1) && *pg_offset == 0) |
| 1015 | memset(kaddr + bytes, 0, PAGE_CACHE_SIZE - bytes); |
Cong Wang | 7ac687d | 2011-11-25 23:14:28 +0800 | [diff] [blame] | 1016 | kunmap_atomic(kaddr); |
Li Zefan | 3a39c18 | 2010-11-08 15:22:19 +0800 | [diff] [blame] | 1017 | flush_dcache_page(page_out); |
| 1018 | |
| 1019 | *pg_offset += bytes; |
| 1020 | buf_offset += bytes; |
| 1021 | working_bytes -= bytes; |
| 1022 | current_buf_start += bytes; |
| 1023 | |
| 1024 | /* check if we need to pick another page */ |
| 1025 | if (*pg_offset == PAGE_CACHE_SIZE) { |
David Sterba | 306e16c | 2011-04-19 14:29:38 +0200 | [diff] [blame] | 1026 | (*pg_index)++; |
| 1027 | if (*pg_index >= vcnt) |
Li Zefan | 3a39c18 | 2010-11-08 15:22:19 +0800 | [diff] [blame] | 1028 | return 0; |
| 1029 | |
David Sterba | 306e16c | 2011-04-19 14:29:38 +0200 | [diff] [blame] | 1030 | page_out = bvec[*pg_index].bv_page; |
Li Zefan | 3a39c18 | 2010-11-08 15:22:19 +0800 | [diff] [blame] | 1031 | *pg_offset = 0; |
| 1032 | start_byte = page_offset(page_out) - disk_start; |
| 1033 | |
| 1034 | /* |
| 1035 | * make sure our new page is covered by this |
| 1036 | * working buffer |
| 1037 | */ |
| 1038 | if (total_out <= start_byte) |
| 1039 | return 1; |
| 1040 | |
| 1041 | /* |
| 1042 | * the next page in the biovec might not be adjacent |
| 1043 | * to the last page, but it might still be found |
| 1044 | * inside this working buffer. bump our offset pointer |
| 1045 | */ |
| 1046 | if (total_out > start_byte && |
| 1047 | current_buf_start < start_byte) { |
| 1048 | buf_offset = start_byte - buf_start; |
| 1049 | working_bytes = total_out - start_byte; |
| 1050 | current_buf_start = buf_start + buf_offset; |
| 1051 | } |
| 1052 | } |
| 1053 | } |
| 1054 | |
| 1055 | return 1; |
| 1056 | } |